fs/btrfs/volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/slab.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/iocontext.h>
  24 #include <linux/capability.h>
  25 #include <linux/ratelimit.h>
  26 #include <linux/kthread.h>
  27 #include <linux/raid/pq.h>
  28 #include <linux/semaphore.h>
  29 #include <linux/uuid.h>
  30 #include <asm/div64.h>
  31 #include "ctree.h"
  32 #include "extent_map.h"
  33 #include "disk-io.h"
  34 #include "transaction.h"
  35 #include "print-tree.h"
  36 #include "volumes.h"
  37 #include "raid56.h"
  38 #include "async-thread.h"
  39 #include "check-integrity.h"
  40 #include "rcu-string.h"
  41 #include "math.h"
  42 #include "dev-replace.h"
  43 #include "sysfs.h"
  44
  45 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
  46         [BTRFS_RAID_RAID10] = {
  47                 .sub_stripes    = 2,
  48                 .dev_stripes    = 1,
  49                 .devs_max       = 0,    /* 0 == as many as possible */
  50                 .devs_min       = 4,
  51                 .tolerated_failures = 1,
  52                 .devs_increment = 2,
  53                 .ncopies        = 2,
  54         },
  55         [BTRFS_RAID_RAID1] = {
  56                 .sub_stripes    = 1,
  57                 .dev_stripes    = 1,
  58                 .devs_max       = 2,
  59                 .devs_min       = 2,
  60                 .tolerated_failures = 1,
  61                 .devs_increment = 2,
  62                 .ncopies        = 2,
  63         },
  64         [BTRFS_RAID_DUP] = {
  65                 .sub_stripes    = 1,
  66                 .dev_stripes    = 2,
  67                 .devs_max       = 1,
  68                 .devs_min       = 1,
  69                 .tolerated_failures = 0,
  70                 .devs_increment = 1,
  71                 .ncopies        = 2,
  72         },
  73         [BTRFS_RAID_RAID0] = {
  74                 .sub_stripes    = 1,
  75                 .dev_stripes    = 1,
  76                 .devs_max       = 0,
  77                 .devs_min       = 2,
  78                 .tolerated_failures = 0,
  79                 .devs_increment = 1,
  80                 .ncopies        = 1,
  81         },
  82         [BTRFS_RAID_SINGLE] = {
  83                 .sub_stripes    = 1,
  84                 .dev_stripes    = 1,
  85                 .devs_max       = 1,
  86                 .devs_min       = 1,
  87                 .tolerated_failures = 0,
  88                 .devs_increment = 1,
  89                 .ncopies        = 1,
  90         },
  91         [BTRFS_RAID_RAID5] = {
  92                 .sub_stripes    = 1,
  93                 .dev_stripes    = 1,
  94                 .devs_max       = 0,
  95                 .devs_min       = 2,
  96                 .tolerated_failures = 1,
  97                 .devs_increment = 1,
  98                 .ncopies        = 2,
  99         },
 100         [BTRFS_RAID_RAID6] = {
 101                 .sub_stripes    = 1,
 102                 .dev_stripes    = 1,
 103                 .devs_max       = 0,
 104                 .devs_min       = 3,
 105                 .tolerated_failures = 2,
 106                 .devs_increment = 1,
 107                 .ncopies        = 3,
 108         },
 109 };
 110
 111 const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
 112         [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
 113         [BTRFS_RAID_RAID1]  = BTRFS_BLOCK_GROUP_RAID1,
 114         [BTRFS_RAID_DUP]    = BTRFS_BLOCK_GROUP_DUP,
 115         [BTRFS_RAID_RAID0]  = BTRFS_BLOCK_GROUP_RAID0,
 116         [BTRFS_RAID_SINGLE] = 0,
 117         [BTRFS_RAID_RAID5]  = BTRFS_BLOCK_GROUP_RAID5,
 118         [BTRFS_RAID_RAID6]  = BTRFS_BLOCK_GROUP_RAID6,
 119 };
 120
 121 /*
 122  * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
 123  * condition is not met. Zero means there's no corresponding
 124  * BTRFS_ERROR_DEV_*_NOT_MET value.
 125  */
 126 const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
 127         [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
 128         [BTRFS_RAID_RAID1]  = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
 129         [BTRFS_RAID_DUP]    = 0,
 130         [BTRFS_RAID_RAID0]  = 0,
 131         [BTRFS_RAID_SINGLE] = 0,
 132         [BTRFS_RAID_RAID5]  = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
 133         [BTRFS_RAID_RAID6]  = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
 134 };
 135
 136 static int init_first_rw_device(struct btrfs_trans_handle *trans,
 137                                 struct btrfs_fs_info *fs_info);
 138 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
 139 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
 140 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
 141 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
 142 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
 143                              enum btrfs_map_op op,
 144                              u64 logical, u64 *length,
 145                              struct btrfs_bio **bbio_ret,
 146                              int mirror_num, int need_raid_map);
 147
 148 DEFINE_MUTEX(uuid_mutex);
 149 static LIST_HEAD(fs_uuids);
 150 struct list_head *btrfs_get_fs_uuids(void)
 151 {
 152         return &fs_uuids;
 153 }
 154
 155 /*
 156  * alloc_fs_devices - allocate struct btrfs_fs_devices
 157  * @fsid:       if not NULL, copy the uuid to fs_devices::fsid
 158  *
 159  * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
 160  * The returned struct is not linked onto any lists and can be destroyed with
 161  * kfree() right away.
 162  */
 163 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
 164 {
 165         struct btrfs_fs_devices *fs_devs;
 166
 167         fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
 168         if (!fs_devs)
 169                 return ERR_PTR(-ENOMEM);
 170
 171         mutex_init(&fs_devs->device_list_mutex);
 172
 173         INIT_LIST_HEAD(&fs_devs->devices);
 174         INIT_LIST_HEAD(&fs_devs->resized_devices);
 175         INIT_LIST_HEAD(&fs_devs->alloc_list);
 176         INIT_LIST_HEAD(&fs_devs->list);
 177         if (fsid)
 178                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 179
 180         return fs_devs;
 181 }
 182
 183 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 184 {
 185         struct btrfs_device *device;
 186         WARN_ON(fs_devices->opened);
 187         while (!list_empty(&fs_devices->devices)) {
 188                 device = list_entry(fs_devices->devices.next,
 189                                     struct btrfs_device, dev_list);
 190                 list_del(&device->dev_list);
 191                 rcu_string_free(device->name);
 192                 bio_put(device->flush_bio);
 193                 kfree(device);
 194         }
 195         kfree(fs_devices);
 196 }
 197
 198 static void btrfs_kobject_uevent(struct block_device *bdev,
 199                                  enum kobject_action action)
 200 {
 201         int ret;
 202
 203         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 204         if (ret)
 205                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 206                         action,
 207                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 208                         &disk_to_dev(bdev->bd_disk)->kobj);
 209 }
 210
 211 void btrfs_cleanup_fs_uuids(void)
 212 {
 213         struct btrfs_fs_devices *fs_devices;
 214
 215         while (!list_empty(&fs_uuids)) {
 216                 fs_devices = list_entry(fs_uuids.next,
 217                                         struct btrfs_fs_devices, list);
 218                 list_del(&fs_devices->list);
 219                 free_fs_devices(fs_devices);
 220         }
 221 }
 222
 223 static struct btrfs_device *__alloc_device(void)
 224 {
 225         struct btrfs_device *dev;
 226
 227         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 228         if (!dev)
 229                 return ERR_PTR(-ENOMEM);
 230
 231         /*
 232          * Preallocate a bio that's always going to be used for flushing device
 233          * barriers and matches the device lifespan
 234          */
 235         dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
 236         if (!dev->flush_bio) {
 237                 kfree(dev);
 238                 return ERR_PTR(-ENOMEM);
 239         }
 240
 241         INIT_LIST_HEAD(&dev->dev_list);
 242         INIT_LIST_HEAD(&dev->dev_alloc_list);
 243         INIT_LIST_HEAD(&dev->resized_list);
 244
 245         spin_lock_init(&dev->io_lock);
 246
 247         spin_lock_init(&dev->reada_lock);
 248         atomic_set(&dev->reada_in_flight, 0);
 249         atomic_set(&dev->dev_stats_ccnt, 0);
 250         btrfs_device_data_ordered_init(dev);
 251         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 252         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
 253
 254         return dev;
 255 }
 256
 257 /*
 258  * Find a device specified by @devid or @uuid in the list of @fs_devices, or
 259  * return NULL.
 260  *
 261  * If devid and uuid are both specified, the match must be exact, otherwise
 262  * only devid is used.
 263  */
 264 static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices,
 265                 u64 devid, const u8 *uuid)
 266 {
 267         struct list_head *head = &fs_devices->devices;
 268         struct btrfs_device *dev;
 269
 270         list_for_each_entry(dev, head, dev_list) {
 271                 if (dev->devid == devid &&
 272                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 273                         return dev;
 274                 }
 275         }
 276         return NULL;
 277 }
 278
 279 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 280 {
 281         struct btrfs_fs_devices *fs_devices;
 282
 283         list_for_each_entry(fs_devices, &fs_uuids, list) {
 284                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 285                         return fs_devices;
 286         }
 287         return NULL;
 288 }
 289
 290 static int
 291 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 292                       int flush, struct block_device **bdev,
 293                       struct buffer_head **bh)
 294 {
 295         int ret;
 296
 297         *bdev = blkdev_get_by_path(device_path, flags, holder);
 298
 299         if (IS_ERR(*bdev)) {
 300                 ret = PTR_ERR(*bdev);
 301                 goto error;
 302         }
 303
 304         if (flush)
 305                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 306         ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
 307         if (ret) {
 308                 blkdev_put(*bdev, flags);
 309                 goto error;
 310         }
 311         invalidate_bdev(*bdev);
 312         *bh = btrfs_read_dev_super(*bdev);
 313         if (IS_ERR(*bh)) {
 314                 ret = PTR_ERR(*bh);
 315                 blkdev_put(*bdev, flags);
 316                 goto error;
 317         }
 318
 319         return 0;
 320
 321 error:
 322         *bdev = NULL;
 323         *bh = NULL;
 324         return ret;
 325 }
 326
 327 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 328                         struct bio *head, struct bio *tail)
 329 {
 330
 331         struct bio *old_head;
 332
 333         old_head = pending_bios->head;
 334         pending_bios->head = head;
 335         if (pending_bios->tail)
 336                 tail->bi_next = old_head;
 337         else
 338                 pending_bios->tail = tail;
 339 }
 340
 341 /*
 342  * we try to collect pending bios for a device so we don't get a large
 343  * number of procs sending bios down to the same device.  This greatly
 344  * improves the schedulers ability to collect and merge the bios.
 345  *
 346  * But, it also turns into a long list of bios to process and that is sure
 347  * to eventually make the worker thread block.  The solution here is to
 348  * make some progress and then put this work struct back at the end of
 349  * the list if the block device is congested.  This way, multiple devices
 350  * can make progress from a single worker thread.
 351  */
 352 static noinline void run_scheduled_bios(struct btrfs_device *device)
 353 {
 354         struct btrfs_fs_info *fs_info = device->fs_info;
 355         struct bio *pending;
 356         struct backing_dev_info *bdi;
 357         struct btrfs_pending_bios *pending_bios;
 358         struct bio *tail;
 359         struct bio *cur;
 360         int again = 0;
 361         unsigned long num_run;
 362         unsigned long batch_run = 0;
 363         unsigned long last_waited = 0;
 364         int force_reg = 0;
 365         int sync_pending = 0;
 366         struct blk_plug plug;
 367
 368         /*
 369          * this function runs all the bios we've collected for
 370          * a particular device.  We don't want to wander off to
 371          * another device without first sending all of these down.
 372          * So, setup a plug here and finish it off before we return
 373          */
 374         blk_start_plug(&plug);
 375
 376         bdi = device->bdev->bd_bdi;
 377
 378 loop:
 379         spin_lock(&device->io_lock);
 380
 381 loop_lock:
 382         num_run = 0;
 383
 384         /* take all the bios off the list at once and process them
 385          * later on (without the lock held).  But, remember the
 386          * tail and other pointers so the bios can be properly reinserted
 387          * into the list if we hit congestion
 388          */
 389         if (!force_reg && device->pending_sync_bios.head) {
 390                 pending_bios = &device->pending_sync_bios;
 391                 force_reg = 1;
 392         } else {
 393                 pending_bios = &device->pending_bios;
 394                 force_reg = 0;
 395         }
 396
 397         pending = pending_bios->head;
 398         tail = pending_bios->tail;
 399         WARN_ON(pending && !tail);
 400
 401         /*
 402          * if pending was null this time around, no bios need processing
 403          * at all and we can stop.  Otherwise it'll loop back up again
 404          * and do an additional check so no bios are missed.
 405          *
 406          * device->running_pending is used to synchronize with the
 407          * schedule_bio code.
 408          */
 409         if (device->pending_sync_bios.head == NULL &&
 410             device->pending_bios.head == NULL) {
 411                 again = 0;
 412                 device->running_pending = 0;
 413         } else {
 414                 again = 1;
 415                 device->running_pending = 1;
 416         }
 417
 418         pending_bios->head = NULL;
 419         pending_bios->tail = NULL;
 420
 421         spin_unlock(&device->io_lock);
 422
 423         while (pending) {
 424
 425                 rmb();
 426                 /* we want to work on both lists, but do more bios on the
 427                  * sync list than the regular list
 428                  */
 429                 if ((num_run > 32 &&
 430                     pending_bios != &device->pending_sync_bios &&
 431                     device->pending_sync_bios.head) ||
 432                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 433                     device->pending_bios.head)) {
 434                         spin_lock(&device->io_lock);
 435                         requeue_list(pending_bios, pending, tail);
 436                         goto loop_lock;
 437                 }
 438
 439                 cur = pending;
 440                 pending = pending->bi_next;
 441                 cur->bi_next = NULL;
 442
 443                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
 444
 445                 /*
 446                  * if we're doing the sync list, record that our
 447                  * plug has some sync requests on it
 448                  *
 449                  * If we're doing the regular list and there are
 450                  * sync requests sitting around, unplug before
 451                  * we add more
 452                  */
 453                 if (pending_bios == &device->pending_sync_bios) {
 454                         sync_pending = 1;
 455                 } else if (sync_pending) {
 456                         blk_finish_plug(&plug);
 457                         blk_start_plug(&plug);
 458                         sync_pending = 0;
 459                 }
 460
 461                 btrfsic_submit_bio(cur);
 462                 num_run++;
 463                 batch_run++;
 464
 465                 cond_resched();
 466
 467                 /*
 468                  * we made progress, there is more work to do and the bdi
 469                  * is now congested.  Back off and let other work structs
 470                  * run instead
 471                  */
 472                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 473                     fs_info->fs_devices->open_devices > 1) {
 474                         struct io_context *ioc;
 475
 476                         ioc = current->io_context;
 477
 478                         /*
 479                          * the main goal here is that we don't want to
 480                          * block if we're going to be able to submit
 481                          * more requests without blocking.
 482                          *
 483                          * This code does two great things, it pokes into
 484                          * the elevator code from a filesystem _and_
 485                          * it makes assumptions about how batching works.
 486                          */
 487                         if (ioc && ioc->nr_batch_requests > 0 &&
 488                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 489                             (last_waited == 0 ||
 490                              ioc->last_waited == last_waited)) {
 491                                 /*
 492                                  * we want to go through our batch of
 493                                  * requests and stop.  So, we copy out
 494                                  * the ioc->last_waited time and test
 495                                  * against it before looping
 496                                  */
 497                                 last_waited = ioc->last_waited;
 498                                 cond_resched();
 499                                 continue;
 500                         }
 501                         spin_lock(&device->io_lock);
 502                         requeue_list(pending_bios, pending, tail);
 503                         device->running_pending = 1;
 504
 505                         spin_unlock(&device->io_lock);
 506                         btrfs_queue_work(fs_info->submit_workers,
 507                                          &device->work);
 508                         goto done;
 509                 }
 510         }
 511
 512         cond_resched();
 513         if (again)
 514                 goto loop;
 515
 516         spin_lock(&device->io_lock);
 517         if (device->pending_bios.head || device->pending_sync_bios.head)
 518                 goto loop_lock;
 519         spin_unlock(&device->io_lock);
 520
 521 done:
 522         blk_finish_plug(&plug);
 523 }
 524
 525 static void pending_bios_fn(struct btrfs_work *work)
 526 {
 527         struct btrfs_device *device;
 528
 529         device = container_of(work, struct btrfs_device, work);
 530         run_scheduled_bios(device);
 531 }
 532
 533
 534 static void btrfs_free_stale_device(struct btrfs_device *cur_dev)
 535 {
 536         struct btrfs_fs_devices *fs_devs;
 537         struct btrfs_device *dev;
 538
 539         if (!cur_dev->name)
 540                 return;
 541
 542         list_for_each_entry(fs_devs, &fs_uuids, list) {
 543                 int del = 1;
 544
 545                 if (fs_devs->opened)
 546                         continue;
 547                 if (fs_devs->seeding)
 548                         continue;
 549
 550                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
 551
 552                         if (dev == cur_dev)
 553                                 continue;
 554                         if (!dev->name)
 555                                 continue;
 556
 557                         /*
 558                          * Todo: This won't be enough. What if the same device
 559                          * comes back (with new uuid and) with its mapper path?
 560                          * But for now, this does help as mostly an admin will
 561                          * either use mapper or non mapper path throughout.
 562                          */
 563                         rcu_read_lock();
 564                         del = strcmp(rcu_str_deref(dev->name),
 565                                                 rcu_str_deref(cur_dev->name));
 566                         rcu_read_unlock();
 567                         if (!del)
 568                                 break;
 569                 }
 570
 571                 if (!del) {
 572                         /* delete the stale device */
 573                         if (fs_devs->num_devices == 1) {
 574                                 btrfs_sysfs_remove_fsid(fs_devs);
 575                                 list_del(&fs_devs->list);
 576                                 free_fs_devices(fs_devs);
 577                         } else {
 578                                 fs_devs->num_devices--;
 579                                 list_del(&dev->dev_list);
 580                                 rcu_string_free(dev->name);
 581                                 bio_put(dev->flush_bio);
 582                                 kfree(dev);
 583                         }
 584                         break;
 585                 }
 586         }
 587 }
 588
 589 /*
 590  * Add new device to list of registered devices
 591  *
 592  * Returns:
 593  * 1   - first time device is seen
 594  * 0   - device already known
 595  * < 0 - error
 596  */
 597 static noinline int device_list_add(const char *path,
 598                            struct btrfs_super_block *disk_super,
 599                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 600 {
 601         struct btrfs_device *device;
 602         struct btrfs_fs_devices *fs_devices;
 603         struct rcu_string *name;
 604         int ret = 0;
 605         u64 found_transid = btrfs_super_generation(disk_super);
 606
 607         fs_devices = find_fsid(disk_super->fsid);
 608         if (!fs_devices) {
 609                 fs_devices = alloc_fs_devices(disk_super->fsid);
 610                 if (IS_ERR(fs_devices))
 611                         return PTR_ERR(fs_devices);
 612
 613                 list_add(&fs_devices->list, &fs_uuids);
 614
 615                 device = NULL;
 616         } else {
 617                 device = find_device(fs_devices, devid,
 618                                 disk_super->dev_item.uuid);
 619         }
 620
 621         if (!device) {
 622                 if (fs_devices->opened)
 623                         return -EBUSY;
 624
 625                 device = btrfs_alloc_device(NULL, &devid,
 626                                             disk_super->dev_item.uuid);
 627                 if (IS_ERR(device)) {
 628                         /* we can safely leave the fs_devices entry around */
 629                         return PTR_ERR(device);
 630                 }
 631
 632                 name = rcu_string_strdup(path, GFP_NOFS);
 633                 if (!name) {
 634                         bio_put(device->flush_bio);
 635                         kfree(device);
 636                         return -ENOMEM;
 637                 }
 638                 rcu_assign_pointer(device->name, name);
 639
 640                 mutex_lock(&fs_devices->device_list_mutex);
 641                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 642                 fs_devices->num_devices++;
 643                 mutex_unlock(&fs_devices->device_list_mutex);
 644
 645                 ret = 1;
 646                 device->fs_devices = fs_devices;
 647         } else if (!device->name || strcmp(device->name->str, path)) {
 648                 /*
 649                  * When FS is already mounted.
 650                  * 1. If you are here and if the device->name is NULL that
 651                  *    means this device was missing at time of FS mount.
 652                  * 2. If you are here and if the device->name is different
 653                  *    from 'path' that means either
 654                  *      a. The same device disappeared and reappeared with
 655                  *         different name. or
 656                  *      b. The missing-disk-which-was-replaced, has
 657                  *         reappeared now.
 658                  *
 659                  * We must allow 1 and 2a above. But 2b would be a spurious
 660                  * and unintentional.
 661                  *
 662                  * Further in case of 1 and 2a above, the disk at 'path'
 663                  * would have missed some transaction when it was away and
 664                  * in case of 2a the stale bdev has to be updated as well.
 665                  * 2b must not be allowed at all time.
 666                  */
 667
 668                 /*
 669                  * For now, we do allow update to btrfs_fs_device through the
 670                  * btrfs dev scan cli after FS has been mounted.  We're still
 671                  * tracking a problem where systems fail mount by subvolume id
 672                  * when we reject replacement on a mounted FS.
 673                  */
 674                 if (!fs_devices->opened && found_transid < device->generation) {
 675                         /*
 676                          * That is if the FS is _not_ mounted and if you
 677                          * are here, that means there is more than one
 678                          * disk with same uuid and devid.We keep the one
 679                          * with larger generation number or the last-in if
 680                          * generation are equal.
 681                          */
 682                         return -EEXIST;
 683                 }
 684
 685                 name = rcu_string_strdup(path, GFP_NOFS);
 686                 if (!name)
 687                         return -ENOMEM;
 688                 rcu_string_free(device->name);
 689                 rcu_assign_pointer(device->name, name);
 690                 if (device->missing) {
 691                         fs_devices->missing_devices--;
 692                         device->missing = 0;
 693                 }
 694         }
 695
 696         /*
 697          * Unmount does not free the btrfs_device struct but would zero
 698          * generation along with most of the other members. So just update
 699          * it back. We need it to pick the disk with largest generation
 700          * (as above).
 701          */
 702         if (!fs_devices->opened)
 703                 device->generation = found_transid;
 704
 705         /*
 706          * if there is new btrfs on an already registered device,
 707          * then remove the stale device entry.
 708          */
 709         if (ret > 0)
 710                 btrfs_free_stale_device(device);
 711
 712         *fs_devices_ret = fs_devices;
 713
 714         return ret;
 715 }
 716
 717 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 718 {
 719         struct btrfs_fs_devices *fs_devices;
 720         struct btrfs_device *device;
 721         struct btrfs_device *orig_dev;
 722
 723         fs_devices = alloc_fs_devices(orig->fsid);
 724         if (IS_ERR(fs_devices))
 725                 return fs_devices;
 726
 727         mutex_lock(&orig->device_list_mutex);
 728         fs_devices->total_devices = orig->total_devices;
 729
 730         /* We have held the volume lock, it is safe to get the devices. */
 731         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 732                 struct rcu_string *name;
 733
 734                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 735                                             orig_dev->uuid);
 736                 if (IS_ERR(device))
 737                         goto error;
 738
 739                 /*
 740                  * This is ok to do without rcu read locked because we hold the
 741                  * uuid mutex so nothing we touch in here is going to disappear.
 742                  */
 743                 if (orig_dev->name) {
 744                         name = rcu_string_strdup(orig_dev->name->str,
 745                                         GFP_KERNEL);
 746                         if (!name) {
 747                                 bio_put(device->flush_bio);
 748                                 kfree(device);
 749                                 goto error;
 750                         }
 751                         rcu_assign_pointer(device->name, name);
 752                 }
 753
 754                 list_add(&device->dev_list, &fs_devices->devices);
 755                 device->fs_devices = fs_devices;
 756                 fs_devices->num_devices++;
 757         }
 758         mutex_unlock(&orig->device_list_mutex);
 759         return fs_devices;
 760 error:
 761         mutex_unlock(&orig->device_list_mutex);
 762         free_fs_devices(fs_devices);
 763         return ERR_PTR(-ENOMEM);
 764 }
 765
 766 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
 767 {
 768         struct btrfs_device *device, *next;
 769         struct btrfs_device *latest_dev = NULL;
 770
 771         mutex_lock(&uuid_mutex);
 772 again:
 773         /* This is the initialized path, it is safe to release the devices. */
 774         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 775                 if (device->in_fs_metadata) {
 776                         if (!device->is_tgtdev_for_dev_replace &&
 777                             (!latest_dev ||
 778                              device->generation > latest_dev->generation)) {
 779                                 latest_dev = device;
 780                         }
 781                         continue;
 782                 }
 783
 784                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 785                         /*
 786                          * In the first step, keep the device which has
 787                          * the correct fsid and the devid that is used
 788                          * for the dev_replace procedure.
 789                          * In the second step, the dev_replace state is
 790                          * read from the device tree and it is known
 791                          * whether the procedure is really active or
 792                          * not, which means whether this device is
 793                          * used or whether it should be removed.
 794                          */
 795                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 796                                 continue;
 797                         }
 798                 }
 799                 if (device->bdev) {
 800                         blkdev_put(device->bdev, device->mode);
 801                         device->bdev = NULL;
 802                         fs_devices->open_devices--;
 803                 }
 804                 if (device->writeable) {
 805                         list_del_init(&device->dev_alloc_list);
 806                         device->writeable = 0;
 807                         if (!device->is_tgtdev_for_dev_replace)
 808                                 fs_devices->rw_devices--;
 809                 }
 810                 list_del_init(&device->dev_list);
 811                 fs_devices->num_devices--;
 812                 rcu_string_free(device->name);
 813                 bio_put(device->flush_bio);
 814                 kfree(device);
 815         }
 816
 817         if (fs_devices->seed) {
 818                 fs_devices = fs_devices->seed;
 819                 goto again;
 820         }
 821
 822         fs_devices->latest_bdev = latest_dev->bdev;
 823
 824         mutex_unlock(&uuid_mutex);
 825 }
 826
 827 static void __free_device(struct work_struct *work)
 828 {
 829         struct btrfs_device *device;
 830
 831         device = container_of(work, struct btrfs_device, rcu_work);
 832         rcu_string_free(device->name);
 833         bio_put(device->flush_bio);
 834         kfree(device);
 835 }
 836
 837 static void free_device(struct rcu_head *head)
 838 {
 839         struct btrfs_device *device;
 840
 841         device = container_of(head, struct btrfs_device, rcu);
 842
 843         INIT_WORK(&device->rcu_work, __free_device);
 844         schedule_work(&device->rcu_work);
 845 }
 846
 847 static void btrfs_close_bdev(struct btrfs_device *device)
 848 {
 849         if (device->bdev && device->writeable) {
 850                 sync_blockdev(device->bdev);
 851                 invalidate_bdev(device->bdev);
 852         }
 853
 854         if (device->bdev)
 855                 blkdev_put(device->bdev, device->mode);
 856 }
 857
 858 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
 859 {
 860         struct btrfs_fs_devices *fs_devices = device->fs_devices;
 861         struct btrfs_device *new_device;
 862         struct rcu_string *name;
 863
 864         if (device->bdev)
 865                 fs_devices->open_devices--;
 866
 867         if (device->writeable &&
 868             device->devid != BTRFS_DEV_REPLACE_DEVID) {
 869                 list_del_init(&device->dev_alloc_list);
 870                 fs_devices->rw_devices--;
 871         }
 872
 873         if (device->missing)
 874                 fs_devices->missing_devices--;
 875
 876         new_device = btrfs_alloc_device(NULL, &device->devid,
 877                                         device->uuid);
 878         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
 879
 880         /* Safe because we are under uuid_mutex */
 881         if (device->name) {
 882                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
 883                 BUG_ON(!name); /* -ENOMEM */
 884                 rcu_assign_pointer(new_device->name, name);
 885         }
 886
 887         list_replace_rcu(&device->dev_list, &new_device->dev_list);
 888         new_device->fs_devices = device->fs_devices;
 889 }
 890
 891 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 892 {
 893         struct btrfs_device *device, *tmp;
 894         struct list_head pending_put;
 895
 896         INIT_LIST_HEAD(&pending_put);
 897
 898         if (--fs_devices->opened > 0)
 899                 return 0;
 900
 901         mutex_lock(&fs_devices->device_list_mutex);
 902         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
 903                 btrfs_prepare_close_one_device(device);
 904                 list_add(&device->dev_list, &pending_put);
 905         }
 906         mutex_unlock(&fs_devices->device_list_mutex);
 907
 908         /*
 909          * btrfs_show_devname() is using the device_list_mutex,
 910          * sometimes call to blkdev_put() leads vfs calling
 911          * into this func. So do put outside of device_list_mutex,
 912          * as of now.
 913          */
 914         while (!list_empty(&pending_put)) {
 915                 device = list_first_entry(&pending_put,
 916                                 struct btrfs_device, dev_list);
 917                 list_del(&device->dev_list);
 918                 btrfs_close_bdev(device);
 919                 call_rcu(&device->rcu, free_device);
 920         }
 921
 922         WARN_ON(fs_devices->open_devices);
 923         WARN_ON(fs_devices->rw_devices);
 924         fs_devices->opened = 0;
 925         fs_devices->seeding = 0;
 926
 927         return 0;
 928 }
 929
 930 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 931 {
 932         struct btrfs_fs_devices *seed_devices = NULL;
 933         int ret;
 934
 935         mutex_lock(&uuid_mutex);
 936         ret = __btrfs_close_devices(fs_devices);
 937         if (!fs_devices->opened) {
 938                 seed_devices = fs_devices->seed;
 939                 fs_devices->seed = NULL;
 940         }
 941         mutex_unlock(&uuid_mutex);
 942
 943         while (seed_devices) {
 944                 fs_devices = seed_devices;
 945                 seed_devices = fs_devices->seed;
 946                 __btrfs_close_devices(fs_devices);
 947                 free_fs_devices(fs_devices);
 948         }
 949         /*
 950          * Wait for rcu kworkers under __btrfs_close_devices
 951          * to finish all blkdev_puts so device is really
 952          * free when umount is done.
 953          */
 954         rcu_barrier();
 955         return ret;
 956 }
 957
 958 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 959                                 fmode_t flags, void *holder)
 960 {
 961         struct request_queue *q;
 962         struct block_device *bdev;
 963         struct list_head *head = &fs_devices->devices;
 964         struct btrfs_device *device;
 965         struct btrfs_device *latest_dev = NULL;
 966         struct buffer_head *bh;
 967         struct btrfs_super_block *disk_super;
 968         u64 devid;
 969         int seeding = 1;
 970         int ret = 0;
 971
 972         flags |= FMODE_EXCL;
 973
 974         list_for_each_entry(device, head, dev_list) {
 975                 if (device->bdev)
 976                         continue;
 977                 if (!device->name)
 978                         continue;
 979
 980                 /* Just open everything we can; ignore failures here */
 981                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 982                                             &bdev, &bh))
 983                         continue;
 984
 985                 disk_super = (struct btrfs_super_block *)bh->b_data;
 986                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 987                 if (devid != device->devid)
 988                         goto error_brelse;
 989
 990                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 991                            BTRFS_UUID_SIZE))
 992                         goto error_brelse;
 993
 994                 device->generation = btrfs_super_generation(disk_super);
 995                 if (!latest_dev ||
 996                     device->generation > latest_dev->generation)
 997                         latest_dev = device;
 998
 999                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
1000                         device->writeable = 0;
1001                 } else {
1002                         device->writeable = !bdev_read_only(bdev);
1003                         seeding = 0;
1004                 }
1005
1006                 q = bdev_get_queue(bdev);
1007                 if (blk_queue_discard(q))
1008                         device->can_discard = 1;
1009                 if (!blk_queue_nonrot(q))
1010                         fs_devices->rotating = 1;
1011
1012                 device->bdev = bdev;
1013                 device->in_fs_metadata = 0;
1014                 device->mode = flags;
1015
1016                 fs_devices->open_devices++;
1017                 if (device->writeable &&
1018                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
1019                         fs_devices->rw_devices++;
1020                         list_add(&device->dev_alloc_list,
1021                                  &fs_devices->alloc_list);
1022                 }
1023                 brelse(bh);
1024                 continue;
1025
1026 error_brelse:
1027                 brelse(bh);
1028                 blkdev_put(bdev, flags);
1029                 continue;
1030         }
1031         if (fs_devices->open_devices == 0) {
1032                 ret = -EINVAL;
1033                 goto out;
1034         }
1035         fs_devices->seeding = seeding;
1036         fs_devices->opened = 1;
1037         fs_devices->latest_bdev = latest_dev->bdev;
1038         fs_devices->total_rw_bytes = 0;
1039 out:
1040         return ret;
1041 }
1042
1043 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1044                        fmode_t flags, void *holder)
1045 {
1046         int ret;
1047
1048         mutex_lock(&uuid_mutex);
1049         if (fs_devices->opened) {
1050                 fs_devices->opened++;
1051                 ret = 0;
1052         } else {
1053                 ret = __btrfs_open_devices(fs_devices, flags, holder);
1054         }
1055         mutex_unlock(&uuid_mutex);
1056         return ret;
1057 }
1058
1059 static void btrfs_release_disk_super(struct page *page)
1060 {
1061         kunmap(page);
1062         put_page(page);
1063 }
1064
1065 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1066                                  struct page **page,
1067                                  struct btrfs_super_block **disk_super)
1068 {
1069         void *p;
1070         pgoff_t index;
1071
1072         /* make sure our super fits in the device */
1073         if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1074                 return 1;
1075
1076         /* make sure our super fits in the page */
1077         if (sizeof(**disk_super) > PAGE_SIZE)
1078                 return 1;
1079
1080         /* make sure our super doesn't straddle pages on disk */
1081         index = bytenr >> PAGE_SHIFT;
1082         if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1083                 return 1;
1084
1085         /* pull in the page with our super */
1086         *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1087                                    index, GFP_KERNEL);
1088
1089         if (IS_ERR_OR_NULL(*page))
1090                 return 1;
1091
1092         p = kmap(*page);
1093
1094         /* align our pointer to the offset of the super block */
1095         *disk_super = p + (bytenr & ~PAGE_MASK);
1096
1097         if (btrfs_super_bytenr(*disk_super) != bytenr ||
1098             btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1099                 btrfs_release_disk_super(*page);
1100                 return 1;
1101         }
1102
1103         if ((*disk_super)->label[0] &&
1104                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1105                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1106
1107         return 0;
1108 }
1109
1110 /*
1111  * Look for a btrfs signature on a device. This may be called out of the mount path
1112  * and we are not allowed to call set_blocksize during the scan. The superblock
1113  * is read via pagecache
1114  */
1115 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1116                           struct btrfs_fs_devices **fs_devices_ret)
1117 {
1118         struct btrfs_super_block *disk_super;
1119         struct block_device *bdev;
1120         struct page *page;
1121         int ret = -EINVAL;
1122         u64 devid;
1123         u64 transid;
1124         u64 total_devices;
1125         u64 bytenr;
1126
1127         /*
1128          * we would like to check all the supers, but that would make
1129          * a btrfs mount succeed after a mkfs from a different FS.
1130          * So, we need to add a special mount option to scan for
1131          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1132          */
1133         bytenr = btrfs_sb_offset(0);
1134         flags |= FMODE_EXCL;
1135         mutex_lock(&uuid_mutex);
1136
1137         bdev = blkdev_get_by_path(path, flags, holder);
1138         if (IS_ERR(bdev)) {
1139                 ret = PTR_ERR(bdev);
1140                 goto error;
1141         }
1142
1143         if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1144                 goto error_bdev_put;
1145
1146         devid = btrfs_stack_device_id(&disk_super->dev_item);
1147         transid = btrfs_super_generation(disk_super);
1148         total_devices = btrfs_super_num_devices(disk_super);
1149
1150         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1151         if (ret > 0) {
1152                 if (disk_super->label[0]) {
1153                         pr_info("BTRFS: device label %s ", disk_super->label);
1154                 } else {
1155                         pr_info("BTRFS: device fsid %pU ", disk_super->fsid);
1156                 }
1157
1158                 pr_cont("devid %llu transid %llu %s\n", devid, transid, path);
1159                 ret = 0;
1160         }
1161         if (!ret && fs_devices_ret)
1162                 (*fs_devices_ret)->total_devices = total_devices;
1163
1164         btrfs_release_disk_super(page);
1165
1166 error_bdev_put:
1167         blkdev_put(bdev, flags);
1168 error:
1169         mutex_unlock(&uuid_mutex);
1170         return ret;
1171 }
1172
1173 /* helper to account the used device space in the range */
1174 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1175                                    u64 end, u64 *length)
1176 {
1177         struct btrfs_key key;
1178         struct btrfs_root *root = device->fs_info->dev_root;
1179         struct btrfs_dev_extent *dev_extent;
1180         struct btrfs_path *path;
1181         u64 extent_end;
1182         int ret;
1183         int slot;
1184         struct extent_buffer *l;
1185
1186         *length = 0;
1187
1188         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1189                 return 0;
1190
1191         path = btrfs_alloc_path();
1192         if (!path)
1193                 return -ENOMEM;
1194         path->reada = READA_FORWARD;
1195
1196         key.objectid = device->devid;
1197         key.offset = start;
1198         key.type = BTRFS_DEV_EXTENT_KEY;
1199
1200         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1201         if (ret < 0)
1202                 goto out;
1203         if (ret > 0) {
1204                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1205                 if (ret < 0)
1206                         goto out;
1207         }
1208
1209         while (1) {
1210                 l = path->nodes[0];
1211                 slot = path->slots[0];
1212                 if (slot >= btrfs_header_nritems(l)) {
1213                         ret = btrfs_next_leaf(root, path);
1214                         if (ret == 0)
1215                                 continue;
1216                         if (ret < 0)
1217                                 goto out;
1218
1219                         break;
1220                 }
1221                 btrfs_item_key_to_cpu(l, &key, slot);
1222
1223                 if (key.objectid < device->devid)
1224                         goto next;
1225
1226                 if (key.objectid > device->devid)
1227                         break;
1228
1229                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1230                         goto next;
1231
1232                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1233                 extent_end = key.offset + btrfs_dev_extent_length(l,
1234                                                                   dev_extent);
1235                 if (key.offset <= start && extent_end > end) {
1236                         *length = end - start + 1;
1237                         break;
1238                 } else if (key.offset <= start && extent_end > start)
1239                         *length += extent_end - start;
1240                 else if (key.offset > start && extent_end <= end)
1241                         *length += extent_end - key.offset;
1242                 else if (key.offset > start && key.offset <= end) {
1243                         *length += end - key.offset + 1;
1244                         break;
1245                 } else if (key.offset > end)
1246                         break;
1247
1248 next:
1249                 path->slots[0]++;
1250         }
1251         ret = 0;
1252 out:
1253         btrfs_free_path(path);
1254         return ret;
1255 }
1256
1257 static int contains_pending_extent(struct btrfs_transaction *transaction,
1258                                    struct btrfs_device *device,
1259                                    u64 *start, u64 len)
1260 {
1261         struct btrfs_fs_info *fs_info = device->fs_info;
1262         struct extent_map *em;
1263         struct list_head *search_list = &fs_info->pinned_chunks;
1264         int ret = 0;
1265         u64 physical_start = *start;
1266
1267         if (transaction)
1268                 search_list = &transaction->pending_chunks;
1269 again:
1270         list_for_each_entry(em, search_list, list) {
1271                 struct map_lookup *map;
1272                 int i;
1273
1274                 map = em->map_lookup;
1275                 for (i = 0; i < map->num_stripes; i++) {
1276                         u64 end;
1277
1278                         if (map->stripes[i].dev != device)
1279                                 continue;
1280                         if (map->stripes[i].physical >= physical_start + len ||
1281                             map->stripes[i].physical + em->orig_block_len <=
1282                             physical_start)
1283                                 continue;
1284                         /*
1285                          * Make sure that while processing the pinned list we do
1286                          * not override our *start with a lower value, because
1287                          * we can have pinned chunks that fall within this
1288                          * device hole and that have lower physical addresses
1289                          * than the pending chunks we processed before. If we
1290                          * do not take this special care we can end up getting
1291                          * 2 pending chunks that start at the same physical
1292                          * device offsets because the end offset of a pinned
1293                          * chunk can be equal to the start offset of some
1294                          * pending chunk.
1295                          */
1296                         end = map->stripes[i].physical + em->orig_block_len;
1297                         if (end > *start) {
1298                                 *start = end;
1299                                 ret = 1;
1300                         }
1301                 }
1302         }
1303         if (search_list != &fs_info->pinned_chunks) {
1304                 search_list = &fs_info->pinned_chunks;
1305                 goto again;
1306         }
1307
1308         return ret;
1309 }
1310
1311
1312 /*
1313  * find_free_dev_extent_start - find free space in the specified device
1314  * @device:       the device which we search the free space in
1315  * @num_bytes:    the size of the free space that we need
1316  * @search_start: the position from which to begin the search
1317  * @start:        store the start of the free space.
1318  * @len:          the size of the free space. that we find, or the size
1319  *                of the max free space if we don't find suitable free space
1320  *
1321  * this uses a pretty simple search, the expectation is that it is
1322  * called very infrequently and that a given device has a small number
1323  * of extents
1324  *
1325  * @start is used to store the start of the free space if we find. But if we
1326  * don't find suitable free space, it will be used to store the start position
1327  * of the max free space.
1328  *
1329  * @len is used to store the size of the free space that we find.
1330  * But if we don't find suitable free space, it is used to store the size of
1331  * the max free space.
1332  */
1333 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1334                                struct btrfs_device *device, u64 num_bytes,
1335                                u64 search_start, u64 *start, u64 *len)
1336 {
1337         struct btrfs_fs_info *fs_info = device->fs_info;
1338         struct btrfs_root *root = fs_info->dev_root;
1339         struct btrfs_key key;
1340         struct btrfs_dev_extent *dev_extent;
1341         struct btrfs_path *path;
1342         u64 hole_size;
1343         u64 max_hole_start;
1344         u64 max_hole_size;
1345         u64 extent_end;
1346         u64 search_end = device->total_bytes;
1347         int ret;
1348         int slot;
1349         struct extent_buffer *l;
1350
1351         /*
1352          * We don't want to overwrite the superblock on the drive nor any area
1353          * used by the boot loader (grub for example), so we make sure to start
1354          * at an offset of at least 1MB.
1355          */
1356         search_start = max_t(u64, search_start, SZ_1M);
1357
1358         path = btrfs_alloc_path();
1359         if (!path)
1360                 return -ENOMEM;
1361
1362         max_hole_start = search_start;
1363         max_hole_size = 0;
1364
1365 again:
1366         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1367                 ret = -ENOSPC;
1368                 goto out;
1369         }
1370
1371         path->reada = READA_FORWARD;
1372         path->search_commit_root = 1;
1373         path->skip_locking = 1;
1374
1375         key.objectid = device->devid;
1376         key.offset = search_start;
1377         key.type = BTRFS_DEV_EXTENT_KEY;
1378
1379         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1380         if (ret < 0)
1381                 goto out;
1382         if (ret > 0) {
1383                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1384                 if (ret < 0)
1385                         goto out;
1386         }
1387
1388         while (1) {
1389                 l = path->nodes[0];
1390                 slot = path->slots[0];
1391                 if (slot >= btrfs_header_nritems(l)) {
1392                         ret = btrfs_next_leaf(root, path);
1393                         if (ret == 0)
1394                                 continue;
1395                         if (ret < 0)
1396                                 goto out;
1397
1398                         break;
1399                 }
1400                 btrfs_item_key_to_cpu(l, &key, slot);
1401
1402                 if (key.objectid < device->devid)
1403                         goto next;
1404
1405                 if (key.objectid > device->devid)
1406                         break;
1407
1408                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1409                         goto next;
1410
1411                 if (key.offset > search_start) {
1412                         hole_size = key.offset - search_start;
1413
1414                         /*
1415                          * Have to check before we set max_hole_start, otherwise
1416                          * we could end up sending back this offset anyway.
1417                          */
1418                         if (contains_pending_extent(transaction, device,
1419                                                     &search_start,
1420                                                     hole_size)) {
1421                                 if (key.offset >= search_start) {
1422                                         hole_size = key.offset - search_start;
1423                                 } else {
1424                                         WARN_ON_ONCE(1);
1425                                         hole_size = 0;
1426                                 }
1427                         }
1428
1429                         if (hole_size > max_hole_size) {
1430                                 max_hole_start = search_start;
1431                                 max_hole_size = hole_size;
1432                         }
1433
1434                         /*
1435                          * If this free space is greater than which we need,
1436                          * it must be the max free space that we have found
1437                          * until now, so max_hole_start must point to the start
1438                          * of this free space and the length of this free space
1439                          * is stored in max_hole_size. Thus, we return
1440                          * max_hole_start and max_hole_size and go back to the
1441                          * caller.
1442                          */
1443                         if (hole_size >= num_bytes) {
1444                                 ret = 0;
1445                                 goto out;
1446                         }
1447                 }
1448
1449                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1450                 extent_end = key.offset + btrfs_dev_extent_length(l,
1451                                                                   dev_extent);
1452                 if (extent_end > search_start)
1453                         search_start = extent_end;
1454 next:
1455                 path->slots[0]++;
1456                 cond_resched();
1457         }
1458
1459         /*
1460          * At this point, search_start should be the end of
1461          * allocated dev extents, and when shrinking the device,
1462          * search_end may be smaller than search_start.
1463          */
1464         if (search_end > search_start) {
1465                 hole_size = search_end - search_start;
1466
1467                 if (contains_pending_extent(transaction, device, &search_start,
1468                                             hole_size)) {
1469                         btrfs_release_path(path);
1470                         goto again;
1471                 }
1472
1473                 if (hole_size > max_hole_size) {
1474                         max_hole_start = search_start;
1475                         max_hole_size = hole_size;
1476                 }
1477         }
1478
1479         /* See above. */
1480         if (max_hole_size < num_bytes)
1481                 ret = -ENOSPC;
1482         else
1483                 ret = 0;
1484
1485 out:
1486         btrfs_free_path(path);
1487         *start = max_hole_start;
1488         if (len)
1489                 *len = max_hole_size;
1490         return ret;
1491 }
1492
1493 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1494                          struct btrfs_device *device, u64 num_bytes,
1495                          u64 *start, u64 *len)
1496 {
1497         /* FIXME use last free of some kind */
1498         return find_free_dev_extent_start(trans->transaction, device,
1499                                           num_bytes, 0, start, len);
1500 }
1501
1502 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1503                           struct btrfs_device *device,
1504                           u64 start, u64 *dev_extent_len)
1505 {
1506         struct btrfs_fs_info *fs_info = device->fs_info;
1507         struct btrfs_root *root = fs_info->dev_root;
1508         int ret;
1509         struct btrfs_path *path;
1510         struct btrfs_key key;
1511         struct btrfs_key found_key;
1512         struct extent_buffer *leaf = NULL;
1513         struct btrfs_dev_extent *extent = NULL;
1514
1515         path = btrfs_alloc_path();
1516         if (!path)
1517                 return -ENOMEM;
1518
1519         key.objectid = device->devid;
1520         key.offset = start;
1521         key.type = BTRFS_DEV_EXTENT_KEY;
1522 again:
1523         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1524         if (ret > 0) {
1525                 ret = btrfs_previous_item(root, path, key.objectid,
1526                                           BTRFS_DEV_EXTENT_KEY);
1527                 if (ret)
1528                         goto out;
1529                 leaf = path->nodes[0];
1530                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1531                 extent = btrfs_item_ptr(leaf, path->slots[0],
1532                                         struct btrfs_dev_extent);
1533                 BUG_ON(found_key.offset > start || found_key.offset +
1534                        btrfs_dev_extent_length(leaf, extent) < start);
1535                 key = found_key;
1536                 btrfs_release_path(path);
1537                 goto again;
1538         } else if (ret == 0) {
1539                 leaf = path->nodes[0];
1540                 extent = btrfs_item_ptr(leaf, path->slots[0],
1541                                         struct btrfs_dev_extent);
1542         } else {
1543                 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1544                 goto out;
1545         }
1546
1547         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1548
1549         ret = btrfs_del_item(trans, root, path);
1550         if (ret) {
1551                 btrfs_handle_fs_error(fs_info, ret,
1552                                       "Failed to remove dev extent item");
1553         } else {
1554                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1555         }
1556 out:
1557         btrfs_free_path(path);
1558         return ret;
1559 }
1560
1561 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1562                                   struct btrfs_device *device,
1563                                   u64 chunk_offset, u64 start, u64 num_bytes)
1564 {
1565         int ret;
1566         struct btrfs_path *path;
1567         struct btrfs_fs_info *fs_info = device->fs_info;
1568         struct btrfs_root *root = fs_info->dev_root;
1569         struct btrfs_dev_extent *extent;
1570         struct extent_buffer *leaf;
1571         struct btrfs_key key;
1572
1573         WARN_ON(!device->in_fs_metadata);
1574         WARN_ON(device->is_tgtdev_for_dev_replace);
1575         path = btrfs_alloc_path();
1576         if (!path)
1577                 return -ENOMEM;
1578
1579         key.objectid = device->devid;
1580         key.offset = start;
1581         key.type = BTRFS_DEV_EXTENT_KEY;
1582         ret = btrfs_insert_empty_item(trans, root, path, &key,
1583                                       sizeof(*extent));
1584         if (ret)
1585                 goto out;
1586
1587         leaf = path->nodes[0];
1588         extent = btrfs_item_ptr(leaf, path->slots[0],
1589                                 struct btrfs_dev_extent);
1590         btrfs_set_dev_extent_chunk_tree(leaf, extent,
1591                                         BTRFS_CHUNK_TREE_OBJECTID);
1592         btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1593                                             BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1594         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1595
1596         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1597         btrfs_mark_buffer_dirty(leaf);
1598 out:
1599         btrfs_free_path(path);
1600         return ret;
1601 }
1602
1603 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1604 {
1605         struct extent_map_tree *em_tree;
1606         struct extent_map *em;
1607         struct rb_node *n;
1608         u64 ret = 0;
1609
1610         em_tree = &fs_info->mapping_tree.map_tree;
1611         read_lock(&em_tree->lock);
1612         n = rb_last(&em_tree->map);
1613         if (n) {
1614                 em = rb_entry(n, struct extent_map, rb_node);
1615                 ret = em->start + em->len;
1616         }
1617         read_unlock(&em_tree->lock);
1618
1619         return ret;
1620 }
1621
1622 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1623                                     u64 *devid_ret)
1624 {
1625         int ret;
1626         struct btrfs_key key;
1627         struct btrfs_key found_key;
1628         struct btrfs_path *path;
1629
1630         path = btrfs_alloc_path();
1631         if (!path)
1632                 return -ENOMEM;
1633
1634         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1635         key.type = BTRFS_DEV_ITEM_KEY;
1636         key.offset = (u64)-1;
1637
1638         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1639         if (ret < 0)
1640                 goto error;
1641
1642         BUG_ON(ret == 0); /* Corruption */
1643
1644         ret = btrfs_previous_item(fs_info->chunk_root, path,
1645                                   BTRFS_DEV_ITEMS_OBJECTID,
1646                                   BTRFS_DEV_ITEM_KEY);
1647         if (ret) {
1648                 *devid_ret = 1;
1649         } else {
1650                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1651                                       path->slots[0]);
1652                 *devid_ret = found_key.offset + 1;
1653         }
1654         ret = 0;
1655 error:
1656         btrfs_free_path(path);
1657         return ret;
1658 }
1659
1660 /*
1661  * the device information is stored in the chunk root
1662  * the btrfs_device struct should be fully filled in
1663  */
1664 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1665                             struct btrfs_fs_info *fs_info,
1666                             struct btrfs_device *device)
1667 {
1668         struct btrfs_root *root = fs_info->chunk_root;
1669         int ret;
1670         struct btrfs_path *path;
1671         struct btrfs_dev_item *dev_item;
1672         struct extent_buffer *leaf;
1673         struct btrfs_key key;
1674         unsigned long ptr;
1675
1676         path = btrfs_alloc_path();
1677         if (!path)
1678                 return -ENOMEM;
1679
1680         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1681         key.type = BTRFS_DEV_ITEM_KEY;
1682         key.offset = device->devid;
1683
1684         ret = btrfs_insert_empty_item(trans, root, path, &key,
1685                                       sizeof(*dev_item));
1686         if (ret)
1687                 goto out;
1688
1689         leaf = path->nodes[0];
1690         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1691
1692         btrfs_set_device_id(leaf, dev_item, device->devid);
1693         btrfs_set_device_generation(leaf, dev_item, 0);
1694         btrfs_set_device_type(leaf, dev_item, device->type);
1695         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1696         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1697         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1698         btrfs_set_device_total_bytes(leaf, dev_item,
1699                                      btrfs_device_get_disk_total_bytes(device));
1700         btrfs_set_device_bytes_used(leaf, dev_item,
1701                                     btrfs_device_get_bytes_used(device));
1702         btrfs_set_device_group(leaf, dev_item, 0);
1703         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1704         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1705         btrfs_set_device_start_offset(leaf, dev_item, 0);
1706
1707         ptr = btrfs_device_uuid(dev_item);
1708         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1709         ptr = btrfs_device_fsid(dev_item);
1710         write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1711         btrfs_mark_buffer_dirty(leaf);
1712
1713         ret = 0;
1714 out:
1715         btrfs_free_path(path);
1716         return ret;
1717 }
1718
1719 /*
1720  * Function to update ctime/mtime for a given device path.
1721  * Mainly used for ctime/mtime based probe like libblkid.
1722  */
1723 static void update_dev_time(const char *path_name)
1724 {
1725         struct file *filp;
1726
1727         filp = filp_open(path_name, O_RDWR, 0);
1728         if (IS_ERR(filp))
1729                 return;
1730         file_update_time(filp);
1731         filp_close(filp, NULL);
1732 }
1733
1734 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1735                              struct btrfs_device *device)
1736 {
1737         struct btrfs_root *root = fs_info->chunk_root;
1738         int ret;
1739         struct btrfs_path *path;
1740         struct btrfs_key key;
1741         struct btrfs_trans_handle *trans;
1742
1743         path = btrfs_alloc_path();
1744         if (!path)
1745                 return -ENOMEM;
1746
1747         trans = btrfs_start_transaction(root, 0);
1748         if (IS_ERR(trans)) {
1749                 btrfs_free_path(path);
1750                 return PTR_ERR(trans);
1751         }
1752         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1753         key.type = BTRFS_DEV_ITEM_KEY;
1754         key.offset = device->devid;
1755
1756         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1757         if (ret) {
1758                 if (ret > 0)
1759                         ret = -ENOENT;
1760                 btrfs_abort_transaction(trans, ret);
1761                 btrfs_end_transaction(trans);
1762                 goto out;
1763         }
1764
1765         ret = btrfs_del_item(trans, root, path);
1766         if (ret) {
1767                 btrfs_abort_transaction(trans, ret);
1768                 btrfs_end_transaction(trans);
1769         }
1770
1771 out:
1772         btrfs_free_path(path);
1773         if (!ret)
1774                 ret = btrfs_commit_transaction(trans);
1775         return ret;
1776 }
1777
1778 /*
1779  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1780  * filesystem. It's up to the caller to adjust that number regarding eg. device
1781  * replace.
1782  */
1783 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1784                 u64 num_devices)
1785 {
1786         u64 all_avail;
1787         unsigned seq;
1788         int i;
1789
1790         do {
1791                 seq = read_seqbegin(&fs_info->profiles_lock);
1792
1793                 all_avail = fs_info->avail_data_alloc_bits |
1794                             fs_info->avail_system_alloc_bits |
1795                             fs_info->avail_metadata_alloc_bits;
1796         } while (read_seqretry(&fs_info->profiles_lock, seq));
1797
1798         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1799                 if (!(all_avail & btrfs_raid_group[i]))
1800                         continue;
1801
1802                 if (num_devices < btrfs_raid_array[i].devs_min) {
1803                         int ret = btrfs_raid_mindev_error[i];
1804
1805                         if (ret)
1806                                 return ret;
1807                 }
1808         }
1809
1810         return 0;
1811 }
1812
1813 static struct btrfs_device * btrfs_find_next_active_device(
1814                 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1815 {
1816         struct btrfs_device *next_device;
1817
1818         list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1819                 if (next_device != device &&
1820                         !next_device->missing && next_device->bdev)
1821                         return next_device;
1822         }
1823
1824         return NULL;
1825 }
1826
1827 /*
1828  * Helper function to check if the given device is part of s_bdev / latest_bdev
1829  * and replace it with the provided or the next active device, in the context
1830  * where this function called, there should be always be another device (or
1831  * this_dev) which is active.
1832  */
1833 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1834                 struct btrfs_device *device, struct btrfs_device *this_dev)
1835 {
1836         struct btrfs_device *next_device;
1837
1838         if (this_dev)
1839                 next_device = this_dev;
1840         else
1841                 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1842                                                                 device);
1843         ASSERT(next_device);
1844
1845         if (fs_info->sb->s_bdev &&
1846                         (fs_info->sb->s_bdev == device->bdev))
1847                 fs_info->sb->s_bdev = next_device->bdev;
1848
1849         if (fs_info->fs_devices->latest_bdev == device->bdev)
1850                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1851 }
1852
1853 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1854                 u64 devid)
1855 {
1856         struct btrfs_device *device;
1857         struct btrfs_fs_devices *cur_devices;
1858         u64 num_devices;
1859         int ret = 0;
1860
1861         mutex_lock(&uuid_mutex);
1862
1863         num_devices = fs_info->fs_devices->num_devices;
1864         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1865         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1866                 WARN_ON(num_devices < 1);
1867                 num_devices--;
1868         }
1869         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1870
1871         ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1872         if (ret)
1873                 goto out;
1874
1875         ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1876                                            &device);
1877         if (ret)
1878                 goto out;
1879
1880         if (device->is_tgtdev_for_dev_replace) {
1881                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1882                 goto out;
1883         }
1884
1885         if (device->writeable && fs_info->fs_devices->rw_devices == 1) {
1886                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1887                 goto out;
1888         }
1889
1890         if (device->writeable) {
1891                 mutex_lock(&fs_info->chunk_mutex);
1892                 list_del_init(&device->dev_alloc_list);
1893                 device->fs_devices->rw_devices--;
1894                 mutex_unlock(&fs_info->chunk_mutex);
1895         }
1896
1897         mutex_unlock(&uuid_mutex);
1898         ret = btrfs_shrink_device(device, 0);
1899         mutex_lock(&uuid_mutex);
1900         if (ret)
1901                 goto error_undo;
1902
1903         /*
1904          * TODO: the superblock still includes this device in its num_devices
1905          * counter although write_all_supers() is not locked out. This
1906          * could give a filesystem state which requires a degraded mount.
1907          */
1908         ret = btrfs_rm_dev_item(fs_info, device);
1909         if (ret)
1910                 goto error_undo;
1911
1912         device->in_fs_metadata = 0;
1913         btrfs_scrub_cancel_dev(fs_info, device);
1914
1915         /*
1916          * the device list mutex makes sure that we don't change
1917          * the device list while someone else is writing out all
1918          * the device supers. Whoever is writing all supers, should
1919          * lock the device list mutex before getting the number of
1920          * devices in the super block (super_copy). Conversely,
1921          * whoever updates the number of devices in the super block
1922          * (super_copy) should hold the device list mutex.
1923          */
1924
1925         cur_devices = device->fs_devices;
1926         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1927         list_del_rcu(&device->dev_list);
1928
1929         device->fs_devices->num_devices--;
1930         device->fs_devices->total_devices--;
1931
1932         if (device->missing)
1933                 device->fs_devices->missing_devices--;
1934
1935         btrfs_assign_next_active_device(fs_info, device, NULL);
1936
1937         if (device->bdev) {
1938                 device->fs_devices->open_devices--;
1939                 /* remove sysfs entry */
1940                 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
1941         }
1942
1943         num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
1944         btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
1945         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1946
1947         /*
1948          * at this point, the device is zero sized and detached from
1949          * the devices list.  All that's left is to zero out the old
1950          * supers and free the device.
1951          */
1952         if (device->writeable)
1953                 btrfs_scratch_superblocks(device->bdev, device->name->str);
1954
1955         btrfs_close_bdev(device);
1956         call_rcu(&device->rcu, free_device);
1957
1958         if (cur_devices->open_devices == 0) {
1959                 struct btrfs_fs_devices *fs_devices;
1960                 fs_devices = fs_info->fs_devices;
1961                 while (fs_devices) {
1962                         if (fs_devices->seed == cur_devices) {
1963                                 fs_devices->seed = cur_devices->seed;
1964                                 break;
1965                         }
1966                         fs_devices = fs_devices->seed;
1967                 }
1968                 cur_devices->seed = NULL;
1969                 __btrfs_close_devices(cur_devices);
1970                 free_fs_devices(cur_devices);
1971         }
1972
1973 out:
1974         mutex_unlock(&uuid_mutex);
1975         return ret;
1976
1977 error_undo:
1978         if (device->writeable) {
1979                 mutex_lock(&fs_info->chunk_mutex);
1980                 list_add(&device->dev_alloc_list,
1981                          &fs_info->fs_devices->alloc_list);
1982                 device->fs_devices->rw_devices++;
1983                 mutex_unlock(&fs_info->chunk_mutex);
1984         }
1985         goto out;
1986 }
1987
1988 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1989                                         struct btrfs_device *srcdev)
1990 {
1991         struct btrfs_fs_devices *fs_devices;
1992
1993         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1994
1995         /*
1996          * in case of fs with no seed, srcdev->fs_devices will point
1997          * to fs_devices of fs_info. However when the dev being replaced is
1998          * a seed dev it will point to the seed's local fs_devices. In short
1999          * srcdev will have its correct fs_devices in both the cases.
2000          */
2001         fs_devices = srcdev->fs_devices;
2002
2003         list_del_rcu(&srcdev->dev_list);
2004         list_del(&srcdev->dev_alloc_list);
2005         fs_devices->num_devices--;
2006         if (srcdev->missing)
2007                 fs_devices->missing_devices--;
2008
2009         if (srcdev->writeable)
2010                 fs_devices->rw_devices--;
2011
2012         if (srcdev->bdev)
2013                 fs_devices->open_devices--;
2014 }
2015
2016 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2017                                       struct btrfs_device *srcdev)
2018 {
2019         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2020
2021         if (srcdev->writeable) {
2022                 /* zero out the old super if it is writable */
2023                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2024         }
2025
2026         btrfs_close_bdev(srcdev);
2027         call_rcu(&srcdev->rcu, free_device);
2028
2029         /* if this is no devs we rather delete the fs_devices */
2030         if (!fs_devices->num_devices) {
2031                 struct btrfs_fs_devices *tmp_fs_devices;
2032
2033                 /*
2034                  * On a mounted FS, num_devices can't be zero unless it's a
2035                  * seed. In case of a seed device being replaced, the replace
2036                  * target added to the sprout FS, so there will be no more
2037                  * device left under the seed FS.
2038                  */
2039                 ASSERT(fs_devices->seeding);
2040
2041                 tmp_fs_devices = fs_info->fs_devices;
2042                 while (tmp_fs_devices) {
2043                         if (tmp_fs_devices->seed == fs_devices) {
2044                                 tmp_fs_devices->seed = fs_devices->seed;
2045                                 break;
2046                         }
2047                         tmp_fs_devices = tmp_fs_devices->seed;
2048                 }
2049                 fs_devices->seed = NULL;
2050                 __btrfs_close_devices(fs_devices);
2051                 free_fs_devices(fs_devices);
2052         }
2053 }
2054
2055 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2056                                       struct btrfs_device *tgtdev)
2057 {
2058         mutex_lock(&uuid_mutex);
2059         WARN_ON(!tgtdev);
2060         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2061
2062         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2063
2064         if (tgtdev->bdev)
2065                 fs_info->fs_devices->open_devices--;
2066
2067         fs_info->fs_devices->num_devices--;
2068
2069         btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2070
2071         list_del_rcu(&tgtdev->dev_list);
2072
2073         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2074         mutex_unlock(&uuid_mutex);
2075
2076         /*
2077          * The update_dev_time() with in btrfs_scratch_superblocks()
2078          * may lead to a call to btrfs_show_devname() which will try
2079          * to hold device_list_mutex. And here this device
2080          * is already out of device list, so we don't have to hold
2081          * the device_list_mutex lock.
2082          */
2083         btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2084
2085         btrfs_close_bdev(tgtdev);
2086         call_rcu(&tgtdev->rcu, free_device);
2087 }
2088
2089 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2090                                      const char *device_path,
2091                                      struct btrfs_device **device)
2092 {
2093         int ret = 0;
2094         struct btrfs_super_block *disk_super;
2095         u64 devid;
2096         u8 *dev_uuid;
2097         struct block_device *bdev;
2098         struct buffer_head *bh;
2099
2100         *device = NULL;
2101         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2102                                     fs_info->bdev_holder, 0, &bdev, &bh);
2103         if (ret)
2104                 return ret;
2105         disk_super = (struct btrfs_super_block *)bh->b_data;
2106         devid = btrfs_stack_device_id(&disk_super->dev_item);
2107         dev_uuid = disk_super->dev_item.uuid;
2108         *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2109         brelse(bh);
2110         if (!*device)
2111                 ret = -ENOENT;
2112         blkdev_put(bdev, FMODE_READ);
2113         return ret;
2114 }
2115
2116 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2117                                          const char *device_path,
2118                                          struct btrfs_device **device)
2119 {
2120         *device = NULL;
2121         if (strcmp(device_path, "missing") == 0) {
2122                 struct list_head *devices;
2123                 struct btrfs_device *tmp;
2124
2125                 devices = &fs_info->fs_devices->devices;
2126                 /*
2127                  * It is safe to read the devices since the volume_mutex
2128                  * is held by the caller.
2129                  */
2130                 list_for_each_entry(tmp, devices, dev_list) {
2131                         if (tmp->in_fs_metadata && !tmp->bdev) {
2132                                 *device = tmp;
2133                                 break;
2134                         }
2135                 }
2136
2137                 if (!*device)
2138                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2139
2140                 return 0;
2141         } else {
2142                 return btrfs_find_device_by_path(fs_info, device_path, device);
2143         }
2144 }
2145
2146 /*
2147  * Lookup a device given by device id, or the path if the id is 0.
2148  */
2149 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2150                                  const char *devpath,
2151                                  struct btrfs_device **device)
2152 {
2153         int ret;
2154
2155         if (devid) {
2156                 ret = 0;
2157                 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2158                 if (!*device)
2159                         ret = -ENOENT;
2160         } else {
2161                 if (!devpath || !devpath[0])
2162                         return -EINVAL;
2163
2164                 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2165                                                            device);
2166         }
2167         return ret;
2168 }
2169
2170 /*
2171  * does all the dirty work required for changing file system's UUID.
2172  */
2173 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2174 {
2175         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2176         struct btrfs_fs_devices *old_devices;
2177         struct btrfs_fs_devices *seed_devices;
2178         struct btrfs_super_block *disk_super = fs_info->super_copy;
2179         struct btrfs_device *device;
2180         u64 super_flags;
2181
2182         BUG_ON(!mutex_is_locked(&uuid_mutex));
2183         if (!fs_devices->seeding)
2184                 return -EINVAL;
2185
2186         seed_devices = alloc_fs_devices(NULL);
2187         if (IS_ERR(seed_devices))
2188                 return PTR_ERR(seed_devices);
2189
2190         old_devices = clone_fs_devices(fs_devices);
2191         if (IS_ERR(old_devices)) {
2192                 kfree(seed_devices);
2193                 return PTR_ERR(old_devices);
2194         }
2195
2196         list_add(&old_devices->list, &fs_uuids);
2197
2198         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2199         seed_devices->opened = 1;
2200         INIT_LIST_HEAD(&seed_devices->devices);
2201         INIT_LIST_HEAD(&seed_devices->alloc_list);
2202         mutex_init(&seed_devices->device_list_mutex);
2203
2204         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2205         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2206                               synchronize_rcu);
2207         list_for_each_entry(device, &seed_devices->devices, dev_list)
2208                 device->fs_devices = seed_devices;
2209
2210         mutex_lock(&fs_info->chunk_mutex);
2211         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2212         mutex_unlock(&fs_info->chunk_mutex);
2213
2214         fs_devices->seeding = 0;
2215         fs_devices->num_devices = 0;
2216         fs_devices->open_devices = 0;
2217         fs_devices->missing_devices = 0;
2218         fs_devices->rotating = 0;
2219         fs_devices->seed = seed_devices;
2220
2221         generate_random_uuid(fs_devices->fsid);
2222         memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2223         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2224         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2225
2226         super_flags = btrfs_super_flags(disk_super) &
2227                       ~BTRFS_SUPER_FLAG_SEEDING;
2228         btrfs_set_super_flags(disk_super, super_flags);
2229
2230         return 0;
2231 }
2232
2233 /*
2234  * Store the expected generation for seed devices in device items.
2235  */
2236 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2237                                struct btrfs_fs_info *fs_info)
2238 {
2239         struct btrfs_root *root = fs_info->chunk_root;
2240         struct btrfs_path *path;
2241         struct extent_buffer *leaf;
2242         struct btrfs_dev_item *dev_item;
2243         struct btrfs_device *device;
2244         struct btrfs_key key;
2245         u8 fs_uuid[BTRFS_FSID_SIZE];
2246         u8 dev_uuid[BTRFS_UUID_SIZE];
2247         u64 devid;
2248         int ret;
2249
2250         path = btrfs_alloc_path();
2251         if (!path)
2252                 return -ENOMEM;
2253
2254         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2255         key.offset = 0;
2256         key.type = BTRFS_DEV_ITEM_KEY;
2257
2258         while (1) {
2259                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2260                 if (ret < 0)
2261                         goto error;
2262
2263                 leaf = path->nodes[0];
2264 next_slot:
2265                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2266                         ret = btrfs_next_leaf(root, path);
2267                         if (ret > 0)
2268                                 break;
2269                         if (ret < 0)
2270                                 goto error;
2271                         leaf = path->nodes[0];
2272                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2273                         btrfs_release_path(path);
2274                         continue;
2275                 }
2276
2277                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2278                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2279                     key.type != BTRFS_DEV_ITEM_KEY)
2280                         break;
2281
2282                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2283                                           struct btrfs_dev_item);
2284                 devid = btrfs_device_id(leaf, dev_item);
2285                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2286                                    BTRFS_UUID_SIZE);
2287                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2288                                    BTRFS_FSID_SIZE);
2289                 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2290                 BUG_ON(!device); /* Logic error */
2291
2292                 if (device->fs_devices->seeding) {
2293                         btrfs_set_device_generation(leaf, dev_item,
2294                                                     device->generation);
2295                         btrfs_mark_buffer_dirty(leaf);
2296                 }
2297
2298                 path->slots[0]++;
2299                 goto next_slot;
2300         }
2301         ret = 0;
2302 error:
2303         btrfs_free_path(path);
2304         return ret;
2305 }
2306
2307 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2308 {
2309         struct btrfs_root *root = fs_info->dev_root;
2310         struct request_queue *q;
2311         struct btrfs_trans_handle *trans;
2312         struct btrfs_device *device;
2313         struct block_device *bdev;
2314         struct list_head *devices;
2315         struct super_block *sb = fs_info->sb;
2316         struct rcu_string *name;
2317         u64 tmp;
2318         int seeding_dev = 0;
2319         int ret = 0;
2320         bool unlocked = false;
2321
2322         if (sb_rdonly(sb) && !fs_info->fs_devices->seeding)
2323                 return -EROFS;
2324
2325         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2326                                   fs_info->bdev_holder);
2327         if (IS_ERR(bdev))
2328                 return PTR_ERR(bdev);
2329
2330         if (fs_info->fs_devices->seeding) {
2331                 seeding_dev = 1;
2332                 down_write(&sb->s_umount);
2333                 mutex_lock(&uuid_mutex);
2334         }
2335
2336         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2337
2338         devices = &fs_info->fs_devices->devices;
2339
2340         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2341         list_for_each_entry(device, devices, dev_list) {
2342                 if (device->bdev == bdev) {
2343                         ret = -EEXIST;
2344                         mutex_unlock(
2345                                 &fs_info->fs_devices->device_list_mutex);
2346                         goto error;
2347                 }
2348         }
2349         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2350
2351         device = btrfs_alloc_device(fs_info, NULL, NULL);
2352         if (IS_ERR(device)) {
2353                 /* we can safely leave the fs_devices entry around */
2354                 ret = PTR_ERR(device);
2355                 goto error;
2356         }
2357
2358         name = rcu_string_strdup(device_path, GFP_KERNEL);
2359         if (!name) {
2360                 bio_put(device->flush_bio);
2361                 kfree(device);
2362                 ret = -ENOMEM;
2363                 goto error;
2364         }
2365         rcu_assign_pointer(device->name, name);
2366
2367         trans = btrfs_start_transaction(root, 0);
2368         if (IS_ERR(trans)) {
2369                 rcu_string_free(device->name);
2370                 bio_put(device->flush_bio);
2371                 kfree(device);
2372                 ret = PTR_ERR(trans);
2373                 goto error;
2374         }
2375
2376         q = bdev_get_queue(bdev);
2377         if (blk_queue_discard(q))
2378                 device->can_discard = 1;
2379         device->writeable = 1;
2380         device->generation = trans->transid;
2381         device->io_width = fs_info->sectorsize;
2382         device->io_align = fs_info->sectorsize;
2383         device->sector_size = fs_info->sectorsize;
2384         device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2385                                          fs_info->sectorsize);
2386         device->disk_total_bytes = device->total_bytes;
2387         device->commit_total_bytes = device->total_bytes;
2388         device->fs_info = fs_info;
2389         device->bdev = bdev;
2390         device->in_fs_metadata = 1;
2391         device->is_tgtdev_for_dev_replace = 0;
2392         device->mode = FMODE_EXCL;
2393         device->dev_stats_valid = 1;
2394         set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2395
2396         if (seeding_dev) {
2397                 sb->s_flags &= ~SB_RDONLY;
2398                 ret = btrfs_prepare_sprout(fs_info);
2399                 if (ret) {
2400                         btrfs_abort_transaction(trans, ret);
2401                         goto error_trans;
2402                 }
2403         }
2404
2405         device->fs_devices = fs_info->fs_devices;
2406
2407         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2408         mutex_lock(&fs_info->chunk_mutex);
2409         list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2410         list_add(&device->dev_alloc_list,
2411                  &fs_info->fs_devices->alloc_list);
2412         fs_info->fs_devices->num_devices++;
2413         fs_info->fs_devices->open_devices++;
2414         fs_info->fs_devices->rw_devices++;
2415         fs_info->fs_devices->total_devices++;
2416         fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2417
2418         atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2419
2420         if (!blk_queue_nonrot(q))
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                 round_down(tmp + device->total_bytes, fs_info->sectorsize));
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_sysfs;
2449                 }
2450         }
2451
2452         ret = btrfs_add_device(trans, fs_info, device);
2453         if (ret) {
2454                 btrfs_abort_transaction(trans, ret);
2455                 goto error_sysfs;
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_sysfs;
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         ret = btrfs_commit_transaction(trans);
2478
2479         if (seeding_dev) {
2480                 mutex_unlock(&uuid_mutex);
2481                 up_write(&sb->s_umount);
2482                 unlocked = true;
2483
2484                 if (ret) /* transaction commit */
2485                         return ret;
2486
2487                 ret = btrfs_relocate_sys_chunks(fs_info);
2488                 if (ret < 0)
2489                         btrfs_handle_fs_error(fs_info, ret,
2490                                     "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2491                 trans = btrfs_attach_transaction(root);
2492                 if (IS_ERR(trans)) {
2493                         if (PTR_ERR(trans) == -ENOENT)
2494                                 return 0;
2495                         ret = PTR_ERR(trans);
2496                         trans = NULL;
2497                         goto error_sysfs;
2498                 }
2499                 ret = btrfs_commit_transaction(trans);
2500         }
2501
2502         /* Update ctime/mtime for libblkid */
2503         update_dev_time(device_path);
2504         return ret;
2505
2506 error_sysfs:
2507         btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2508 error_trans:
2509         if (seeding_dev)
2510                 sb->s_flags |= SB_RDONLY;
2511         if (trans)
2512                 btrfs_end_transaction(trans);
2513         rcu_string_free(device->name);
2514         bio_put(device->flush_bio);
2515         kfree(device);
2516 error:
2517         blkdev_put(bdev, FMODE_EXCL);
2518         if (seeding_dev && !unlocked) {
2519                 mutex_unlock(&uuid_mutex);
2520                 up_write(&sb->s_umount);
2521         }
2522         return ret;
2523 }
2524
2525 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2526                                   const char *device_path,
2527                                   struct btrfs_device *srcdev,
2528                                   struct btrfs_device **device_out)
2529 {
2530         struct request_queue *q;
2531         struct btrfs_device *device;
2532         struct block_device *bdev;
2533         struct list_head *devices;
2534         struct rcu_string *name;
2535         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2536         int ret = 0;
2537
2538         *device_out = NULL;
2539         if (fs_info->fs_devices->seeding) {
2540                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2541                 return -EINVAL;
2542         }
2543
2544         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2545                                   fs_info->bdev_holder);
2546         if (IS_ERR(bdev)) {
2547                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2548                 return PTR_ERR(bdev);
2549         }
2550
2551         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2552
2553         devices = &fs_info->fs_devices->devices;
2554         list_for_each_entry(device, devices, dev_list) {
2555                 if (device->bdev == bdev) {
2556                         btrfs_err(fs_info,
2557                                   "target device is in the filesystem!");
2558                         ret = -EEXIST;
2559                         goto error;
2560                 }
2561         }
2562
2563
2564         if (i_size_read(bdev->bd_inode) <
2565             btrfs_device_get_total_bytes(srcdev)) {
2566                 btrfs_err(fs_info,
2567                           "target device is smaller than source device!");
2568                 ret = -EINVAL;
2569                 goto error;
2570         }
2571
2572
2573         device = btrfs_alloc_device(NULL, &devid, NULL);
2574         if (IS_ERR(device)) {
2575                 ret = PTR_ERR(device);
2576                 goto error;
2577         }
2578
2579         name = rcu_string_strdup(device_path, GFP_KERNEL);
2580         if (!name) {
2581                 bio_put(device->flush_bio);
2582                 kfree(device);
2583                 ret = -ENOMEM;
2584                 goto error;
2585         }
2586         rcu_assign_pointer(device->name, name);
2587
2588         q = bdev_get_queue(bdev);
2589         if (blk_queue_discard(q))
2590                 device->can_discard = 1;
2591         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2592         device->writeable = 1;
2593         device->generation = 0;
2594         device->io_width = fs_info->sectorsize;
2595         device->io_align = fs_info->sectorsize;
2596         device->sector_size = fs_info->sectorsize;
2597         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2598         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2599         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2600         ASSERT(list_empty(&srcdev->resized_list));
2601         device->commit_total_bytes = srcdev->commit_total_bytes;
2602         device->commit_bytes_used = device->bytes_used;
2603         device->fs_info = fs_info;
2604         device->bdev = bdev;
2605         device->in_fs_metadata = 1;
2606         device->is_tgtdev_for_dev_replace = 1;
2607         device->mode = FMODE_EXCL;
2608         device->dev_stats_valid = 1;
2609         set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2610         device->fs_devices = fs_info->fs_devices;
2611         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2612         fs_info->fs_devices->num_devices++;
2613         fs_info->fs_devices->open_devices++;
2614         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2615
2616         *device_out = device;
2617         return ret;
2618
2619 error:
2620         blkdev_put(bdev, FMODE_EXCL);
2621         return ret;
2622 }
2623
2624 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2625                                               struct btrfs_device *tgtdev)
2626 {
2627         u32 sectorsize = fs_info->sectorsize;
2628
2629         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2630         tgtdev->io_width = sectorsize;
2631         tgtdev->io_align = sectorsize;
2632         tgtdev->sector_size = sectorsize;
2633         tgtdev->fs_info = fs_info;
2634         tgtdev->in_fs_metadata = 1;
2635 }
2636
2637 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2638                                         struct btrfs_device *device)
2639 {
2640         int ret;
2641         struct btrfs_path *path;
2642         struct btrfs_root *root = device->fs_info->chunk_root;
2643         struct btrfs_dev_item *dev_item;
2644         struct extent_buffer *leaf;
2645         struct btrfs_key key;
2646
2647         path = btrfs_alloc_path();
2648         if (!path)
2649                 return -ENOMEM;
2650
2651         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2652         key.type = BTRFS_DEV_ITEM_KEY;
2653         key.offset = device->devid;
2654
2655         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2656         if (ret < 0)
2657                 goto out;
2658
2659         if (ret > 0) {
2660                 ret = -ENOENT;
2661                 goto out;
2662         }
2663
2664         leaf = path->nodes[0];
2665         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2666
2667         btrfs_set_device_id(leaf, dev_item, device->devid);
2668         btrfs_set_device_type(leaf, dev_item, device->type);
2669         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2670         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2671         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2672         btrfs_set_device_total_bytes(leaf, dev_item,
2673                                      btrfs_device_get_disk_total_bytes(device));
2674         btrfs_set_device_bytes_used(leaf, dev_item,
2675                                     btrfs_device_get_bytes_used(device));
2676         btrfs_mark_buffer_dirty(leaf);
2677
2678 out:
2679         btrfs_free_path(path);
2680         return ret;
2681 }
2682
2683 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2684                       struct btrfs_device *device, u64 new_size)
2685 {
2686         struct btrfs_fs_info *fs_info = device->fs_info;
2687         struct btrfs_super_block *super_copy = fs_info->super_copy;
2688         struct btrfs_fs_devices *fs_devices;
2689         u64 old_total;
2690         u64 diff;
2691
2692         if (!device->writeable)
2693                 return -EACCES;
2694
2695         new_size = round_down(new_size, fs_info->sectorsize);
2696
2697         mutex_lock(&fs_info->chunk_mutex);
2698         old_total = btrfs_super_total_bytes(super_copy);
2699         diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2700
2701         if (new_size <= device->total_bytes ||
2702             device->is_tgtdev_for_dev_replace) {
2703                 mutex_unlock(&fs_info->chunk_mutex);
2704                 return -EINVAL;
2705         }
2706
2707         fs_devices = fs_info->fs_devices;
2708
2709         btrfs_set_super_total_bytes(super_copy,
2710                         round_down(old_total + diff, fs_info->sectorsize));
2711         device->fs_devices->total_rw_bytes += diff;
2712
2713         btrfs_device_set_total_bytes(device, new_size);
2714         btrfs_device_set_disk_total_bytes(device, new_size);
2715         btrfs_clear_space_info_full(device->fs_info);
2716         if (list_empty(&device->resized_list))
2717                 list_add_tail(&device->resized_list,
2718                               &fs_devices->resized_devices);
2719         mutex_unlock(&fs_info->chunk_mutex);
2720
2721         return btrfs_update_device(trans, device);
2722 }
2723
2724 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2725                             struct btrfs_fs_info *fs_info, u64 chunk_offset)
2726 {
2727         struct btrfs_root *root = fs_info->chunk_root;
2728         int ret;
2729         struct btrfs_path *path;
2730         struct btrfs_key key;
2731
2732         path = btrfs_alloc_path();
2733         if (!path)
2734                 return -ENOMEM;
2735
2736         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2737         key.offset = chunk_offset;
2738         key.type = BTRFS_CHUNK_ITEM_KEY;
2739
2740         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2741         if (ret < 0)
2742                 goto out;
2743         else if (ret > 0) { /* Logic error or corruption */
2744                 btrfs_handle_fs_error(fs_info, -ENOENT,
2745                                       "Failed lookup while freeing chunk.");
2746                 ret = -ENOENT;
2747                 goto out;
2748         }
2749
2750         ret = btrfs_del_item(trans, root, path);
2751         if (ret < 0)
2752                 btrfs_handle_fs_error(fs_info, ret,
2753                                       "Failed to delete chunk item.");
2754 out:
2755         btrfs_free_path(path);
2756         return ret;
2757 }
2758
2759 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2760 {
2761         struct btrfs_super_block *super_copy = fs_info->super_copy;
2762         struct btrfs_disk_key *disk_key;
2763         struct btrfs_chunk *chunk;
2764         u8 *ptr;
2765         int ret = 0;
2766         u32 num_stripes;
2767         u32 array_size;
2768         u32 len = 0;
2769         u32 cur;
2770         struct btrfs_key key;
2771
2772         mutex_lock(&fs_info->chunk_mutex);
2773         array_size = btrfs_super_sys_array_size(super_copy);
2774
2775         ptr = super_copy->sys_chunk_array;
2776         cur = 0;
2777
2778         while (cur < array_size) {
2779                 disk_key = (struct btrfs_disk_key *)ptr;
2780                 btrfs_disk_key_to_cpu(&key, disk_key);
2781
2782                 len = sizeof(*disk_key);
2783
2784                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2785                         chunk = (struct btrfs_chunk *)(ptr + len);
2786                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2787                         len += btrfs_chunk_item_size(num_stripes);
2788                 } else {
2789                         ret = -EIO;
2790                         break;
2791                 }
2792                 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2793                     key.offset == chunk_offset) {
2794                         memmove(ptr, ptr + len, array_size - (cur + len));
2795                         array_size -= len;
2796                         btrfs_set_super_sys_array_size(super_copy, array_size);
2797                 } else {
2798                         ptr += len;
2799                         cur += len;
2800                 }
2801         }
2802         mutex_unlock(&fs_info->chunk_mutex);
2803         return ret;
2804 }
2805
2806 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2807                                         u64 logical, u64 length)
2808 {
2809         struct extent_map_tree *em_tree;
2810         struct extent_map *em;
2811
2812         em_tree = &fs_info->mapping_tree.map_tree;
2813         read_lock(&em_tree->lock);
2814         em = lookup_extent_mapping(em_tree, logical, length);
2815         read_unlock(&em_tree->lock);
2816
2817         if (!em) {
2818                 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2819                            logical, length);
2820                 return ERR_PTR(-EINVAL);
2821         }
2822
2823         if (em->start > logical || em->start + em->len < logical) {
2824                 btrfs_crit(fs_info,
2825                            "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2826                            logical, length, em->start, em->start + em->len);
2827                 free_extent_map(em);
2828                 return ERR_PTR(-EINVAL);
2829         }
2830
2831         /* callers are responsible for dropping em's ref. */
2832         return em;
2833 }
2834
2835 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2836                        struct btrfs_fs_info *fs_info, u64 chunk_offset)
2837 {
2838         struct extent_map *em;
2839         struct map_lookup *map;
2840         u64 dev_extent_len = 0;
2841         int i, ret = 0;
2842         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2843
2844         em = get_chunk_map(fs_info, chunk_offset, 1);
2845         if (IS_ERR(em)) {
2846                 /*
2847                  * This is a logic error, but we don't want to just rely on the
2848                  * user having built with ASSERT enabled, so if ASSERT doesn't
2849                  * do anything we still error out.
2850                  */
2851                 ASSERT(0);
2852                 return PTR_ERR(em);
2853         }
2854         map = em->map_lookup;
2855         mutex_lock(&fs_info->chunk_mutex);
2856         check_system_chunk(trans, fs_info, map->type);
2857         mutex_unlock(&fs_info->chunk_mutex);
2858
2859         /*
2860          * Take the device list mutex to prevent races with the final phase of
2861          * a device replace operation that replaces the device object associated
2862          * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2863          */
2864         mutex_lock(&fs_devices->device_list_mutex);
2865         for (i = 0; i < map->num_stripes; i++) {
2866                 struct btrfs_device *device = map->stripes[i].dev;
2867                 ret = btrfs_free_dev_extent(trans, device,
2868                                             map->stripes[i].physical,
2869                                             &dev_extent_len);
2870                 if (ret) {
2871                         mutex_unlock(&fs_devices->device_list_mutex);
2872                         btrfs_abort_transaction(trans, ret);
2873                         goto out;
2874                 }
2875
2876                 if (device->bytes_used > 0) {
2877                         mutex_lock(&fs_info->chunk_mutex);
2878                         btrfs_device_set_bytes_used(device,
2879                                         device->bytes_used - dev_extent_len);
2880                         atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2881                         btrfs_clear_space_info_full(fs_info);
2882                         mutex_unlock(&fs_info->chunk_mutex);
2883                 }
2884
2885                 if (map->stripes[i].dev) {
2886                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2887                         if (ret) {
2888                                 mutex_unlock(&fs_devices->device_list_mutex);
2889                                 btrfs_abort_transaction(trans, ret);
2890                                 goto out;
2891                         }
2892                 }
2893         }
2894         mutex_unlock(&fs_devices->device_list_mutex);
2895
2896         ret = btrfs_free_chunk(trans, fs_info, chunk_offset);
2897         if (ret) {
2898                 btrfs_abort_transaction(trans, ret);
2899                 goto out;
2900         }
2901
2902         trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2903
2904         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2905                 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2906                 if (ret) {
2907                         btrfs_abort_transaction(trans, ret);
2908                         goto out;
2909                 }
2910         }
2911
2912         ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2913         if (ret) {
2914                 btrfs_abort_transaction(trans, ret);
2915                 goto out;
2916         }
2917
2918 out:
2919         /* once for us */
2920         free_extent_map(em);
2921         return ret;
2922 }
2923
2924 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2925 {
2926         struct btrfs_root *root = fs_info->chunk_root;
2927         struct btrfs_trans_handle *trans;
2928         int ret;
2929
2930         /*
2931          * Prevent races with automatic removal of unused block groups.
2932          * After we relocate and before we remove the chunk with offset
2933          * chunk_offset, automatic removal of the block group can kick in,
2934          * resulting in a failure when calling btrfs_remove_chunk() below.
2935          *
2936          * Make sure to acquire this mutex before doing a tree search (dev
2937          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2938          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2939          * we release the path used to search the chunk/dev tree and before
2940          * the current task acquires this mutex and calls us.
2941          */
2942         ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
2943
2944         ret = btrfs_can_relocate(fs_info, chunk_offset);
2945         if (ret)
2946                 return -ENOSPC;
2947
2948         /* step one, relocate all the extents inside this chunk */
2949         btrfs_scrub_pause(fs_info);
2950         ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2951         btrfs_scrub_continue(fs_info);
2952         if (ret)
2953                 return ret;
2954
2955         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2956                                                      chunk_offset);
2957         if (IS_ERR(trans)) {
2958                 ret = PTR_ERR(trans);
2959                 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2960                 return ret;
2961         }
2962
2963         /*
2964          * step two, delete the device extents and the
2965          * chunk tree entries
2966          */
2967         ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2968         btrfs_end_transaction(trans);
2969         return ret;
2970 }
2971
2972 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2973 {
2974         struct btrfs_root *chunk_root = fs_info->chunk_root;
2975         struct btrfs_path *path;
2976         struct extent_buffer *leaf;
2977         struct btrfs_chunk *chunk;
2978         struct btrfs_key key;
2979         struct btrfs_key found_key;
2980         u64 chunk_type;
2981         bool retried = false;
2982         int failed = 0;
2983         int ret;
2984
2985         path = btrfs_alloc_path();
2986         if (!path)
2987                 return -ENOMEM;
2988
2989 again:
2990         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2991         key.offset = (u64)-1;
2992         key.type = BTRFS_CHUNK_ITEM_KEY;
2993
2994         while (1) {
2995                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2996                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2997                 if (ret < 0) {
2998                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2999                         goto error;
3000                 }
3001                 BUG_ON(ret == 0); /* Corruption */
3002
3003                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3004                                           key.type);
3005                 if (ret)
3006                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3007                 if (ret < 0)
3008                         goto error;
3009                 if (ret > 0)
3010                         break;
3011
3012                 leaf = path->nodes[0];
3013                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3014
3015                 chunk = btrfs_item_ptr(leaf, path->slots[0],
3016                                        struct btrfs_chunk);
3017                 chunk_type = btrfs_chunk_type(leaf, chunk);
3018                 btrfs_release_path(path);
3019
3020                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3021                         ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3022                         if (ret == -ENOSPC)
3023                                 failed++;
3024                         else
3025                                 BUG_ON(ret);
3026                 }
3027                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3028
3029                 if (found_key.offset == 0)
3030                         break;
3031                 key.offset = found_key.offset - 1;
3032         }
3033         ret = 0;
3034         if (failed && !retried) {
3035                 failed = 0;
3036                 retried = true;
3037                 goto again;
3038         } else if (WARN_ON(failed && retried)) {
3039                 ret = -ENOSPC;
3040         }
3041 error:
3042         btrfs_free_path(path);
3043         return ret;
3044 }
3045
3046 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3047                                struct btrfs_balance_control *bctl)
3048 {
3049         struct btrfs_root *root = fs_info->tree_root;
3050         struct btrfs_trans_handle *trans;
3051         struct btrfs_balance_item *item;
3052         struct btrfs_disk_balance_args disk_bargs;
3053         struct btrfs_path *path;
3054         struct extent_buffer *leaf;
3055         struct btrfs_key key;
3056         int ret, err;
3057
3058         path = btrfs_alloc_path();
3059         if (!path)
3060                 return -ENOMEM;
3061
3062         trans = btrfs_start_transaction(root, 0);
3063         if (IS_ERR(trans)) {
3064                 btrfs_free_path(path);
3065                 return PTR_ERR(trans);
3066         }
3067
3068         key.objectid = BTRFS_BALANCE_OBJECTID;
3069         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3070         key.offset = 0;
3071
3072         ret = btrfs_insert_empty_item(trans, root, path, &key,
3073                                       sizeof(*item));
3074         if (ret)
3075                 goto out;
3076
3077         leaf = path->nodes[0];
3078         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3079
3080         memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3081
3082         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3083         btrfs_set_balance_data(leaf, item, &disk_bargs);
3084         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3085         btrfs_set_balance_meta(leaf, item, &disk_bargs);
3086         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3087         btrfs_set_balance_sys(leaf, item, &disk_bargs);
3088
3089         btrfs_set_balance_flags(leaf, item, bctl->flags);
3090
3091         btrfs_mark_buffer_dirty(leaf);
3092 out:
3093         btrfs_free_path(path);
3094         err = btrfs_commit_transaction(trans);
3095         if (err && !ret)
3096                 ret = err;
3097         return ret;
3098 }
3099
3100 static int del_balance_item(struct btrfs_fs_info *fs_info)
3101 {
3102         struct btrfs_root *root = fs_info->tree_root;
3103         struct btrfs_trans_handle *trans;
3104         struct btrfs_path *path;
3105         struct btrfs_key key;
3106         int ret, err;
3107
3108         path = btrfs_alloc_path();
3109         if (!path)
3110                 return -ENOMEM;
3111
3112         trans = btrfs_start_transaction(root, 0);
3113         if (IS_ERR(trans)) {
3114                 btrfs_free_path(path);
3115                 return PTR_ERR(trans);
3116         }
3117
3118         key.objectid = BTRFS_BALANCE_OBJECTID;
3119         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3120         key.offset = 0;
3121
3122         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3123         if (ret < 0)
3124                 goto out;
3125         if (ret > 0) {
3126                 ret = -ENOENT;
3127                 goto out;
3128         }
3129
3130         ret = btrfs_del_item(trans, root, path);
3131 out:
3132         btrfs_free_path(path);
3133         err = btrfs_commit_transaction(trans);
3134         if (err && !ret)
3135                 ret = err;
3136         return ret;
3137 }
3138
3139 /*
3140  * This is a heuristic used to reduce the number of chunks balanced on
3141  * resume after balance was interrupted.
3142  */
3143 static void update_balance_args(struct btrfs_balance_control *bctl)
3144 {
3145         /*
3146          * Turn on soft mode for chunk types that were being converted.
3147          */
3148         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3149                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3150         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3151                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3152         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3153                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3154
3155         /*
3156          * Turn on usage filter if is not already used.  The idea is
3157          * that chunks that we have already balanced should be
3158          * reasonably full.  Don't do it for chunks that are being
3159          * converted - that will keep us from relocating unconverted
3160          * (albeit full) chunks.
3161          */
3162         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3163             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3164             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3165                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3166                 bctl->data.usage = 90;
3167         }
3168         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3169             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3170             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3171                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3172                 bctl->sys.usage = 90;
3173         }
3174         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3175             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3176             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3177                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3178                 bctl->meta.usage = 90;
3179         }
3180 }
3181
3182 /*
3183  * Should be called with both balance and volume mutexes held to
3184  * serialize other volume operations (add_dev/rm_dev/resize) with
3185  * restriper.  Same goes for unset_balance_control.
3186  */
3187 static void set_balance_control(struct btrfs_balance_control *bctl)
3188 {
3189         struct btrfs_fs_info *fs_info = bctl->fs_info;
3190
3191         BUG_ON(fs_info->balance_ctl);
3192
3193         spin_lock(&fs_info->balance_lock);
3194         fs_info->balance_ctl = bctl;
3195         spin_unlock(&fs_info->balance_lock);
3196 }
3197
3198 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3199 {
3200         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3201
3202         BUG_ON(!fs_info->balance_ctl);
3203
3204         spin_lock(&fs_info->balance_lock);
3205         fs_info->balance_ctl = NULL;
3206         spin_unlock(&fs_info->balance_lock);
3207
3208         kfree(bctl);
3209 }
3210
3211 /*
3212  * Balance filters.  Return 1 if chunk should be filtered out
3213  * (should not be balanced).
3214  */
3215 static int chunk_profiles_filter(u64 chunk_type,
3216                                  struct btrfs_balance_args *bargs)
3217 {
3218         chunk_type = chunk_to_extended(chunk_type) &
3219                                 BTRFS_EXTENDED_PROFILE_MASK;
3220
3221         if (bargs->profiles & chunk_type)
3222                 return 0;
3223
3224         return 1;
3225 }
3226
3227 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3228                               struct btrfs_balance_args *bargs)
3229 {
3230         struct btrfs_block_group_cache *cache;
3231         u64 chunk_used;
3232         u64 user_thresh_min;
3233         u64 user_thresh_max;
3234         int ret = 1;
3235
3236         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3237         chunk_used = btrfs_block_group_used(&cache->item);
3238
3239         if (bargs->usage_min == 0)
3240                 user_thresh_min = 0;
3241         else
3242                 user_thresh_min = div_factor_fine(cache->key.offset,
3243                                         bargs->usage_min);
3244
3245         if (bargs->usage_max == 0)
3246                 user_thresh_max = 1;
3247         else if (bargs->usage_max > 100)
3248                 user_thresh_max = cache->key.offset;
3249         else
3250                 user_thresh_max = div_factor_fine(cache->key.offset,
3251                                         bargs->usage_max);
3252
3253         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3254                 ret = 0;
3255
3256         btrfs_put_block_group(cache);
3257         return ret;
3258 }
3259
3260 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3261                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3262 {
3263         struct btrfs_block_group_cache *cache;
3264         u64 chunk_used, user_thresh;
3265         int ret = 1;
3266
3267         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3268         chunk_used = btrfs_block_group_used(&cache->item);
3269
3270         if (bargs->usage_min == 0)
3271                 user_thresh = 1;
3272         else if (bargs->usage > 100)
3273                 user_thresh = cache->key.offset;
3274         else
3275                 user_thresh = div_factor_fine(cache->key.offset,
3276                                               bargs->usage);
3277
3278         if (chunk_used < user_thresh)
3279                 ret = 0;
3280
3281         btrfs_put_block_group(cache);
3282         return ret;
3283 }
3284
3285 static int chunk_devid_filter(struct extent_buffer *leaf,
3286                               struct btrfs_chunk *chunk,
3287                               struct btrfs_balance_args *bargs)
3288 {
3289         struct btrfs_stripe *stripe;
3290         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3291         int i;
3292
3293         for (i = 0; i < num_stripes; i++) {
3294                 stripe = btrfs_stripe_nr(chunk, i);
3295                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3296                         return 0;
3297         }
3298
3299         return 1;
3300 }
3301
3302 /* [pstart, pend) */
3303 static int chunk_drange_filter(struct extent_buffer *leaf,
3304                                struct btrfs_chunk *chunk,
3305                                struct btrfs_balance_args *bargs)
3306 {
3307         struct btrfs_stripe *stripe;
3308         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3309         u64 stripe_offset;
3310         u64 stripe_length;
3311         int factor;
3312         int i;
3313
3314         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3315                 return 0;
3316
3317         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3318              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3319                 factor = num_stripes / 2;
3320         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3321                 factor = num_stripes - 1;
3322         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3323                 factor = num_stripes - 2;
3324         } else {
3325                 factor = num_stripes;
3326         }
3327
3328         for (i = 0; i < num_stripes; i++) {
3329                 stripe = btrfs_stripe_nr(chunk, i);
3330                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3331                         continue;
3332
3333                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3334                 stripe_length = btrfs_chunk_length(leaf, chunk);
3335                 stripe_length = div_u64(stripe_length, factor);
3336
3337                 if (stripe_offset < bargs->pend &&
3338                     stripe_offset + stripe_length > bargs->pstart)
3339                         return 0;
3340         }
3341
3342         return 1;
3343 }
3344
3345 /* [vstart, vend) */
3346 static int chunk_vrange_filter(struct extent_buffer *leaf,
3347                                struct btrfs_chunk *chunk,
3348                                u64 chunk_offset,
3349                                struct btrfs_balance_args *bargs)
3350 {
3351         if (chunk_offset < bargs->vend &&
3352             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3353                 /* at least part of the chunk is inside this vrange */
3354                 return 0;
3355
3356         return 1;
3357 }
3358
3359 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3360                                struct btrfs_chunk *chunk,
3361                                struct btrfs_balance_args *bargs)
3362 {
3363         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3364
3365         if (bargs->stripes_min <= num_stripes
3366                         && num_stripes <= bargs->stripes_max)
3367                 return 0;
3368
3369         return 1;
3370 }
3371
3372 static int chunk_soft_convert_filter(u64 chunk_type,
3373                                      struct btrfs_balance_args *bargs)
3374 {
3375         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3376                 return 0;
3377
3378         chunk_type = chunk_to_extended(chunk_type) &
3379                                 BTRFS_EXTENDED_PROFILE_MASK;
3380
3381         if (bargs->target == chunk_type)
3382                 return 1;
3383
3384         return 0;
3385 }
3386
3387 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3388                                 struct extent_buffer *leaf,
3389                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3390 {
3391         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3392         struct btrfs_balance_args *bargs = NULL;
3393         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3394
3395         /* type filter */
3396         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3397               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3398                 return 0;
3399         }
3400
3401         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3402                 bargs = &bctl->data;
3403         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3404                 bargs = &bctl->sys;
3405         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3406                 bargs = &bctl->meta;
3407
3408         /* profiles filter */
3409         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3410             chunk_profiles_filter(chunk_type, bargs)) {
3411                 return 0;
3412         }
3413
3414         /* usage filter */
3415         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3416             chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3417                 return 0;
3418         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3419             chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3420                 return 0;
3421         }
3422
3423         /* devid filter */
3424         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3425             chunk_devid_filter(leaf, chunk, bargs)) {
3426                 return 0;
3427         }
3428
3429         /* drange filter, makes sense only with devid filter */
3430         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3431             chunk_drange_filter(leaf, chunk, bargs)) {
3432                 return 0;
3433         }
3434
3435         /* vrange filter */
3436         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3437             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3438                 return 0;
3439         }
3440
3441         /* stripes filter */
3442         if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3443             chunk_stripes_range_filter(leaf, chunk, bargs)) {
3444                 return 0;
3445         }
3446
3447         /* soft profile changing mode */
3448         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3449             chunk_soft_convert_filter(chunk_type, bargs)) {
3450                 return 0;
3451         }
3452
3453         /*
3454          * limited by count, must be the last filter
3455          */
3456         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3457                 if (bargs->limit == 0)
3458                         return 0;
3459                 else
3460                         bargs->limit--;
3461         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3462                 /*
3463                  * Same logic as the 'limit' filter; the minimum cannot be
3464                  * determined here because we do not have the global information
3465                  * about the count of all chunks that satisfy the filters.
3466                  */
3467                 if (bargs->limit_max == 0)
3468                         return 0;
3469                 else
3470                         bargs->limit_max--;
3471         }
3472
3473         return 1;
3474 }
3475
3476 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3477 {
3478         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3479         struct btrfs_root *chunk_root = fs_info->chunk_root;
3480         struct btrfs_root *dev_root = fs_info->dev_root;
3481         struct list_head *devices;
3482         struct btrfs_device *device;
3483         u64 old_size;
3484         u64 size_to_free;
3485         u64 chunk_type;
3486         struct btrfs_chunk *chunk;
3487         struct btrfs_path *path = NULL;
3488         struct btrfs_key key;
3489         struct btrfs_key found_key;
3490         struct btrfs_trans_handle *trans;
3491         struct extent_buffer *leaf;
3492         int slot;
3493         int ret;
3494         int enospc_errors = 0;
3495         bool counting = true;
3496         /* The single value limit and min/max limits use the same bytes in the */
3497         u64 limit_data = bctl->data.limit;
3498         u64 limit_meta = bctl->meta.limit;
3499         u64 limit_sys = bctl->sys.limit;
3500         u32 count_data = 0;
3501         u32 count_meta = 0;
3502         u32 count_sys = 0;
3503         int chunk_reserved = 0;
3504         u64 bytes_used = 0;
3505
3506         /* step one make some room on all the devices */
3507         devices = &fs_info->fs_devices->devices;
3508         list_for_each_entry(device, devices, dev_list) {
3509                 old_size = btrfs_device_get_total_bytes(device);
3510                 size_to_free = div_factor(old_size, 1);
3511                 size_to_free = min_t(u64, size_to_free, SZ_1M);
3512                 if (!device->writeable ||
3513                     btrfs_device_get_total_bytes(device) -
3514                     btrfs_device_get_bytes_used(device) > size_to_free ||
3515                     device->is_tgtdev_for_dev_replace)
3516                         continue;
3517
3518                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3519                 if (ret == -ENOSPC)
3520                         break;
3521                 if (ret) {
3522                         /* btrfs_shrink_device never returns ret > 0 */
3523                         WARN_ON(ret > 0);
3524                         goto error;
3525                 }
3526
3527                 trans = btrfs_start_transaction(dev_root, 0);
3528                 if (IS_ERR(trans)) {
3529                         ret = PTR_ERR(trans);
3530                         btrfs_info_in_rcu(fs_info,
3531                  "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3532                                           rcu_str_deref(device->name), ret,
3533                                           old_size, old_size - size_to_free);
3534                         goto error;
3535                 }
3536
3537                 ret = btrfs_grow_device(trans, device, old_size);
3538                 if (ret) {
3539                         btrfs_end_transaction(trans);
3540                         /* btrfs_grow_device never returns ret > 0 */
3541                         WARN_ON(ret > 0);
3542                         btrfs_info_in_rcu(fs_info,
3543                  "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3544                                           rcu_str_deref(device->name), ret,
3545                                           old_size, old_size - size_to_free);
3546                         goto error;
3547                 }
3548
3549                 btrfs_end_transaction(trans);
3550         }
3551
3552         /* step two, relocate all the chunks */
3553         path = btrfs_alloc_path();
3554         if (!path) {
3555                 ret = -ENOMEM;
3556                 goto error;
3557         }
3558
3559         /* zero out stat counters */
3560         spin_lock(&fs_info->balance_lock);
3561         memset(&bctl->stat, 0, sizeof(bctl->stat));
3562         spin_unlock(&fs_info->balance_lock);
3563 again:
3564         if (!counting) {
3565                 /*
3566                  * The single value limit and min/max limits use the same bytes
3567                  * in the
3568                  */
3569                 bctl->data.limit = limit_data;
3570                 bctl->meta.limit = limit_meta;
3571                 bctl->sys.limit = limit_sys;
3572         }
3573         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3574         key.offset = (u64)-1;
3575         key.type = BTRFS_CHUNK_ITEM_KEY;
3576
3577         while (1) {
3578                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3579                     atomic_read(&fs_info->balance_cancel_req)) {
3580                         ret = -ECANCELED;
3581                         goto error;
3582                 }
3583
3584                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3585                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3586                 if (ret < 0) {
3587                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3588                         goto error;
3589                 }
3590
3591                 /*
3592                  * this shouldn't happen, it means the last relocate
3593                  * failed
3594                  */
3595                 if (ret == 0)
3596                         BUG(); /* FIXME break ? */
3597
3598                 ret = btrfs_previous_item(chunk_root, path, 0,
3599                                           BTRFS_CHUNK_ITEM_KEY);
3600                 if (ret) {
3601                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3602                         ret = 0;
3603                         break;
3604                 }
3605
3606                 leaf = path->nodes[0];
3607                 slot = path->slots[0];
3608                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3609
3610                 if (found_key.objectid != key.objectid) {
3611                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3612                         break;
3613                 }
3614
3615                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3616                 chunk_type = btrfs_chunk_type(leaf, chunk);
3617
3618                 if (!counting) {
3619                         spin_lock(&fs_info->balance_lock);
3620                         bctl->stat.considered++;
3621                         spin_unlock(&fs_info->balance_lock);
3622                 }
3623
3624                 ret = should_balance_chunk(fs_info, leaf, chunk,
3625                                            found_key.offset);
3626
3627                 btrfs_release_path(path);
3628                 if (!ret) {
3629                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3630                         goto loop;
3631                 }
3632
3633                 if (counting) {
3634                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3635                         spin_lock(&fs_info->balance_lock);
3636                         bctl->stat.expected++;
3637                         spin_unlock(&fs_info->balance_lock);
3638
3639                         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3640                                 count_data++;
3641                         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3642                                 count_sys++;
3643                         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3644                                 count_meta++;
3645
3646                         goto loop;
3647                 }
3648
3649                 /*
3650                  * Apply limit_min filter, no need to check if the LIMITS
3651                  * filter is used, limit_min is 0 by default
3652                  */
3653                 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3654                                         count_data < bctl->data.limit_min)
3655                                 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3656                                         count_meta < bctl->meta.limit_min)
3657                                 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3658                                         count_sys < bctl->sys.limit_min)) {
3659                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3660                         goto loop;
3661                 }
3662
3663                 ASSERT(fs_info->data_sinfo);
3664                 spin_lock(&fs_info->data_sinfo->lock);
3665                 bytes_used = fs_info->data_sinfo->bytes_used;
3666                 spin_unlock(&fs_info->data_sinfo->lock);
3667
3668                 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3669                     !chunk_reserved && !bytes_used) {
3670                         trans = btrfs_start_transaction(chunk_root, 0);
3671                         if (IS_ERR(trans)) {
3672                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3673                                 ret = PTR_ERR(trans);
3674                                 goto error;
3675                         }
3676
3677                         ret = btrfs_force_chunk_alloc(trans, fs_info,
3678                                                       BTRFS_BLOCK_GROUP_DATA);
3679                         btrfs_end_transaction(trans);
3680                         if (ret < 0) {
3681                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3682                                 goto error;
3683                         }
3684                         chunk_reserved = 1;
3685                 }
3686
3687                 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3688                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3689                 if (ret && ret != -ENOSPC)
3690                         goto error;
3691                 if (ret == -ENOSPC) {
3692                         enospc_errors++;
3693                 } else {
3694                         spin_lock(&fs_info->balance_lock);
3695                         bctl->stat.completed++;
3696                         spin_unlock(&fs_info->balance_lock);
3697                 }
3698 loop:
3699                 if (found_key.offset == 0)
3700                         break;
3701                 key.offset = found_key.offset - 1;
3702         }
3703
3704         if (counting) {
3705                 btrfs_release_path(path);
3706                 counting = false;
3707                 goto again;
3708         }
3709 error:
3710         btrfs_free_path(path);
3711         if (enospc_errors) {
3712                 btrfs_info(fs_info, "%d enospc errors during balance",
3713                            enospc_errors);
3714                 if (!ret)
3715                         ret = -ENOSPC;
3716         }
3717
3718         return ret;
3719 }
3720
3721 /**
3722  * alloc_profile_is_valid - see if a given profile is valid and reduced
3723  * @flags: profile to validate
3724  * @extended: if true @flags is treated as an extended profile
3725  */
3726 static int alloc_profile_is_valid(u64 flags, int extended)
3727 {
3728         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3729                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3730
3731         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3732
3733         /* 1) check that all other bits are zeroed */
3734         if (flags & ~mask)
3735                 return 0;
3736
3737         /* 2) see if profile is reduced */
3738         if (flags == 0)
3739                 return !extended; /* "0" is valid for usual profiles */
3740
3741         /* true if exactly one bit set */
3742         return (flags & (flags - 1)) == 0;
3743 }
3744
3745 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3746 {
3747         /* cancel requested || normal exit path */
3748         return atomic_read(&fs_info->balance_cancel_req) ||
3749                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3750                  atomic_read(&fs_info->balance_cancel_req) == 0);
3751 }
3752
3753 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3754 {
3755         int ret;
3756
3757         unset_balance_control(fs_info);
3758         ret = del_balance_item(fs_info);
3759         if (ret)
3760                 btrfs_handle_fs_error(fs_info, ret, NULL);
3761
3762         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3763 }
3764
3765 /* Non-zero return value signifies invalidity */
3766 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3767                 u64 allowed)
3768 {
3769         return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3770                 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3771                  (bctl_arg->target & ~allowed)));
3772 }
3773
3774 /*
3775  * Should be called with both balance and volume mutexes held
3776  */
3777 int btrfs_balance(struct btrfs_balance_control *bctl,
3778                   struct btrfs_ioctl_balance_args *bargs)
3779 {
3780         struct btrfs_fs_info *fs_info = bctl->fs_info;
3781         u64 meta_target, data_target;
3782         u64 allowed;
3783         int mixed = 0;
3784         int ret;
3785         u64 num_devices;
3786         unsigned seq;
3787
3788         if (btrfs_fs_closing(fs_info) ||
3789             atomic_read(&fs_info->balance_pause_req) ||
3790             atomic_read(&fs_info->balance_cancel_req)) {
3791                 ret = -EINVAL;
3792                 goto out;
3793         }
3794
3795         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3796         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3797                 mixed = 1;
3798
3799         /*
3800          * In case of mixed groups both data and meta should be picked,
3801          * and identical options should be given for both of them.
3802          */
3803         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3804         if (mixed && (bctl->flags & allowed)) {
3805                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3806                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3807                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3808                         btrfs_err(fs_info,
3809                                   "with mixed groups data and metadata balance options must be the same");
3810                         ret = -EINVAL;
3811                         goto out;
3812                 }
3813         }
3814
3815         num_devices = fs_info->fs_devices->num_devices;
3816         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3817         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3818                 BUG_ON(num_devices < 1);
3819                 num_devices--;
3820         }
3821         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3822         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3823         if (num_devices > 1)
3824                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3825         if (num_devices > 2)
3826                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3827         if (num_devices > 3)
3828                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3829                             BTRFS_BLOCK_GROUP_RAID6);
3830         if (validate_convert_profile(&bctl->data, allowed)) {
3831                 btrfs_err(fs_info,
3832                           "unable to start balance with target data profile %llu",
3833                           bctl->data.target);
3834                 ret = -EINVAL;
3835                 goto out;
3836         }
3837         if (validate_convert_profile(&bctl->meta, allowed)) {
3838                 btrfs_err(fs_info,
3839                           "unable to start balance with target metadata profile %llu",
3840                           bctl->meta.target);
3841                 ret = -EINVAL;
3842                 goto out;
3843         }
3844         if (validate_convert_profile(&bctl->sys, allowed)) {
3845                 btrfs_err(fs_info,
3846                           "unable to start balance with target system profile %llu",
3847                           bctl->sys.target);
3848                 ret = -EINVAL;
3849                 goto out;
3850         }
3851
3852         /* allow to reduce meta or sys integrity only if force set */
3853         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3854                         BTRFS_BLOCK_GROUP_RAID10 |
3855                         BTRFS_BLOCK_GROUP_RAID5 |
3856                         BTRFS_BLOCK_GROUP_RAID6;
3857         do {
3858                 seq = read_seqbegin(&fs_info->profiles_lock);
3859
3860                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3861                      (fs_info->avail_system_alloc_bits & allowed) &&
3862                      !(bctl->sys.target & allowed)) ||
3863                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3864                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3865                      !(bctl->meta.target & allowed))) {
3866                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3867                                 btrfs_info(fs_info,
3868                                            "force reducing metadata integrity");
3869                         } else {
3870                                 btrfs_err(fs_info,
3871                                           "balance will reduce metadata integrity, use force if you want this");
3872                                 ret = -EINVAL;
3873                                 goto out;
3874                         }
3875                 }
3876         } while (read_seqretry(&fs_info->profiles_lock, seq));
3877
3878         /* if we're not converting, the target field is uninitialized */
3879         meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3880                 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3881         data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3882                 bctl->data.target : fs_info->avail_data_alloc_bits;
3883         if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3884                 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3885                 btrfs_warn(fs_info,
3886                            "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3887                            meta_target, data_target);
3888         }
3889
3890         ret = insert_balance_item(fs_info, bctl);
3891         if (ret && ret != -EEXIST)
3892                 goto out;
3893
3894         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3895                 BUG_ON(ret == -EEXIST);
3896                 set_balance_control(bctl);
3897         } else {
3898                 BUG_ON(ret != -EEXIST);
3899                 spin_lock(&fs_info->balance_lock);
3900                 update_balance_args(bctl);
3901                 spin_unlock(&fs_info->balance_lock);
3902         }
3903
3904         atomic_inc(&fs_info->balance_running);
3905         mutex_unlock(&fs_info->balance_mutex);
3906
3907         ret = __btrfs_balance(fs_info);
3908
3909         mutex_lock(&fs_info->balance_mutex);
3910         atomic_dec(&fs_info->balance_running);
3911
3912         if (bargs) {
3913                 memset(bargs, 0, sizeof(*bargs));
3914                 update_ioctl_balance_args(fs_info, 0, bargs);
3915         }
3916
3917         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3918             balance_need_close(fs_info)) {
3919                 __cancel_balance(fs_info);
3920         }
3921
3922         wake_up(&fs_info->balance_wait_q);
3923
3924         return ret;
3925 out:
3926         if (bctl->flags & BTRFS_BALANCE_RESUME)
3927                 __cancel_balance(fs_info);
3928         else {
3929                 kfree(bctl);
3930                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3931         }
3932         return ret;
3933 }
3934
3935 static int balance_kthread(void *data)
3936 {
3937         struct btrfs_fs_info *fs_info = data;
3938         int ret = 0;
3939
3940         mutex_lock(&fs_info->volume_mutex);
3941         mutex_lock(&fs_info->balance_mutex);
3942
3943         if (fs_info->balance_ctl) {
3944                 btrfs_info(fs_info, "continuing balance");
3945                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3946         }
3947
3948         mutex_unlock(&fs_info->balance_mutex);
3949         mutex_unlock(&fs_info->volume_mutex);
3950
3951         return ret;
3952 }
3953
3954 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3955 {
3956         struct task_struct *tsk;
3957
3958         spin_lock(&fs_info->balance_lock);
3959         if (!fs_info->balance_ctl) {
3960                 spin_unlock(&fs_info->balance_lock);
3961                 return 0;
3962         }
3963         spin_unlock(&fs_info->balance_lock);
3964
3965         if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3966                 btrfs_info(fs_info, "force skipping balance");
3967                 return 0;
3968         }
3969
3970         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3971         return PTR_ERR_OR_ZERO(tsk);
3972 }
3973
3974 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3975 {
3976         struct btrfs_balance_control *bctl;
3977         struct btrfs_balance_item *item;
3978         struct btrfs_disk_balance_args disk_bargs;
3979         struct btrfs_path *path;
3980         struct extent_buffer *leaf;
3981         struct btrfs_key key;
3982         int ret;
3983
3984         path = btrfs_alloc_path();
3985         if (!path)
3986                 return -ENOMEM;
3987
3988         key.objectid = BTRFS_BALANCE_OBJECTID;
3989         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3990         key.offset = 0;
3991
3992         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3993         if (ret < 0)
3994                 goto out;
3995         if (ret > 0) { /* ret = -ENOENT; */
3996                 ret = 0;
3997                 goto out;
3998         }
3999
4000         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4001         if (!bctl) {
4002                 ret = -ENOMEM;
4003                 goto out;
4004         }
4005
4006         leaf = path->nodes[0];
4007         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4008
4009         bctl->fs_info = fs_info;
4010         bctl->flags = btrfs_balance_flags(leaf, item);
4011         bctl->flags |= BTRFS_BALANCE_RESUME;
4012
4013         btrfs_balance_data(leaf, item, &disk_bargs);
4014         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4015         btrfs_balance_meta(leaf, item, &disk_bargs);
4016         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4017         btrfs_balance_sys(leaf, item, &disk_bargs);
4018         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4019
4020         WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4021
4022         mutex_lock(&fs_info->volume_mutex);
4023         mutex_lock(&fs_info->balance_mutex);
4024
4025         set_balance_control(bctl);
4026
4027         mutex_unlock(&fs_info->balance_mutex);
4028         mutex_unlock(&fs_info->volume_mutex);
4029 out:
4030         btrfs_free_path(path);
4031         return ret;
4032 }
4033
4034 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4035 {
4036         int ret = 0;
4037
4038         mutex_lock(&fs_info->balance_mutex);
4039         if (!fs_info->balance_ctl) {
4040                 mutex_unlock(&fs_info->balance_mutex);
4041                 return -ENOTCONN;
4042         }
4043
4044         if (atomic_read(&fs_info->balance_running)) {
4045                 atomic_inc(&fs_info->balance_pause_req);
4046                 mutex_unlock(&fs_info->balance_mutex);
4047
4048                 wait_event(fs_info->balance_wait_q,
4049                            atomic_read(&fs_info->balance_running) == 0);
4050
4051                 mutex_lock(&fs_info->balance_mutex);
4052                 /* we are good with balance_ctl ripped off from under us */
4053                 BUG_ON(atomic_read(&fs_info->balance_running));
4054                 atomic_dec(&fs_info->balance_pause_req);
4055         } else {
4056                 ret = -ENOTCONN;
4057         }
4058
4059         mutex_unlock(&fs_info->balance_mutex);
4060         return ret;
4061 }
4062
4063 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4064 {
4065         if (sb_rdonly(fs_info->sb))
4066                 return -EROFS;
4067
4068         mutex_lock(&fs_info->balance_mutex);
4069         if (!fs_info->balance_ctl) {
4070                 mutex_unlock(&fs_info->balance_mutex);
4071                 return -ENOTCONN;
4072         }
4073
4074         atomic_inc(&fs_info->balance_cancel_req);
4075         /*
4076          * if we are running just wait and return, balance item is
4077          * deleted in btrfs_balance in this case
4078          */
4079         if (atomic_read(&fs_info->balance_running)) {
4080                 mutex_unlock(&fs_info->balance_mutex);
4081                 wait_event(fs_info->balance_wait_q,
4082                            atomic_read(&fs_info->balance_running) == 0);
4083                 mutex_lock(&fs_info->balance_mutex);
4084         } else {
4085                 /* __cancel_balance needs volume_mutex */
4086                 mutex_unlock(&fs_info->balance_mutex);
4087                 mutex_lock(&fs_info->volume_mutex);
4088                 mutex_lock(&fs_info->balance_mutex);
4089
4090                 if (fs_info->balance_ctl)
4091                         __cancel_balance(fs_info);
4092
4093                 mutex_unlock(&fs_info->volume_mutex);
4094         }
4095
4096         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4097         atomic_dec(&fs_info->balance_cancel_req);
4098         mutex_unlock(&fs_info->balance_mutex);
4099         return 0;
4100 }
4101
4102 static int btrfs_uuid_scan_kthread(void *data)
4103 {
4104         struct btrfs_fs_info *fs_info = data;
4105         struct btrfs_root *root = fs_info->tree_root;
4106         struct btrfs_key key;
4107         struct btrfs_path *path = NULL;
4108         int ret = 0;
4109         struct extent_buffer *eb;
4110         int slot;
4111         struct btrfs_root_item root_item;
4112         u32 item_size;
4113         struct btrfs_trans_handle *trans = NULL;
4114
4115         path = btrfs_alloc_path();
4116         if (!path) {
4117                 ret = -ENOMEM;
4118                 goto out;
4119         }
4120
4121         key.objectid = 0;
4122         key.type = BTRFS_ROOT_ITEM_KEY;
4123         key.offset = 0;
4124
4125         while (1) {
4126                 ret = btrfs_search_forward(root, &key, path, 0);
4127                 if (ret) {
4128                         if (ret > 0)
4129                                 ret = 0;
4130                         break;
4131                 }
4132
4133                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4134                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4135                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4136                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
4137                         goto skip;
4138
4139                 eb = path->nodes[0];
4140                 slot = path->slots[0];
4141                 item_size = btrfs_item_size_nr(eb, slot);
4142                 if (item_size < sizeof(root_item))
4143                         goto skip;
4144
4145                 read_extent_buffer(eb, &root_item,
4146                                    btrfs_item_ptr_offset(eb, slot),
4147                                    (int)sizeof(root_item));
4148                 if (btrfs_root_refs(&root_item) == 0)
4149                         goto skip;
4150
4151                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4152                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
4153                         if (trans)
4154                                 goto update_tree;
4155
4156                         btrfs_release_path(path);
4157                         /*
4158                          * 1 - subvol uuid item
4159                          * 1 - received_subvol uuid item
4160                          */
4161                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4162                         if (IS_ERR(trans)) {
4163                                 ret = PTR_ERR(trans);
4164                                 break;
4165                         }
4166                         continue;
4167                 } else {
4168                         goto skip;
4169                 }
4170 update_tree:
4171                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4172                         ret = btrfs_uuid_tree_add(trans, fs_info,
4173                                                   root_item.uuid,
4174                                                   BTRFS_UUID_KEY_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                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4184                         ret = btrfs_uuid_tree_add(trans, fs_info,
4185                                                   root_item.received_uuid,
4186                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4187                                                   key.objectid);
4188                         if (ret < 0) {
4189                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4190                                         ret);
4191                                 break;
4192                         }
4193                 }
4194
4195 skip:
4196                 if (trans) {
4197                         ret = btrfs_end_transaction(trans);
4198                         trans = NULL;
4199                         if (ret)
4200                                 break;
4201                 }
4202
4203                 btrfs_release_path(path);
4204                 if (key.offset < (u64)-1) {
4205                         key.offset++;
4206                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4207                         key.offset = 0;
4208                         key.type = BTRFS_ROOT_ITEM_KEY;
4209                 } else if (key.objectid < (u64)-1) {
4210                         key.offset = 0;
4211                         key.type = BTRFS_ROOT_ITEM_KEY;
4212                         key.objectid++;
4213                 } else {
4214                         break;
4215                 }
4216                 cond_resched();
4217         }
4218
4219 out:
4220         btrfs_free_path(path);
4221         if (trans && !IS_ERR(trans))
4222                 btrfs_end_transaction(trans);
4223         if (ret)
4224                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4225         else
4226                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4227         up(&fs_info->uuid_tree_rescan_sem);
4228         return 0;
4229 }
4230
4231 /*
4232  * Callback for btrfs_uuid_tree_iterate().
4233  * returns:
4234  * 0    check succeeded, the entry is not outdated.
4235  * < 0  if an error occurred.
4236  * > 0  if the check failed, which means the caller shall remove the entry.
4237  */
4238 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4239                                        u8 *uuid, u8 type, u64 subid)
4240 {
4241         struct btrfs_key key;
4242         int ret = 0;
4243         struct btrfs_root *subvol_root;
4244
4245         if (type != BTRFS_UUID_KEY_SUBVOL &&
4246             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4247                 goto out;
4248
4249         key.objectid = subid;
4250         key.type = BTRFS_ROOT_ITEM_KEY;
4251         key.offset = (u64)-1;
4252         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4253         if (IS_ERR(subvol_root)) {
4254                 ret = PTR_ERR(subvol_root);
4255                 if (ret == -ENOENT)
4256                         ret = 1;
4257                 goto out;
4258         }
4259
4260         switch (type) {
4261         case BTRFS_UUID_KEY_SUBVOL:
4262                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4263                         ret = 1;
4264                 break;
4265         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4266                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4267                            BTRFS_UUID_SIZE))
4268                         ret = 1;
4269                 break;
4270         }
4271
4272 out:
4273         return ret;
4274 }
4275
4276 static int btrfs_uuid_rescan_kthread(void *data)
4277 {
4278         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4279         int ret;
4280
4281         /*
4282          * 1st step is to iterate through the existing UUID tree and
4283          * to delete all entries that contain outdated data.
4284          * 2nd step is to add all missing entries to the UUID tree.
4285          */
4286         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4287         if (ret < 0) {
4288                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4289                 up(&fs_info->uuid_tree_rescan_sem);
4290                 return ret;
4291         }
4292         return btrfs_uuid_scan_kthread(data);
4293 }
4294
4295 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4296 {
4297         struct btrfs_trans_handle *trans;
4298         struct btrfs_root *tree_root = fs_info->tree_root;
4299         struct btrfs_root *uuid_root;
4300         struct task_struct *task;
4301         int ret;
4302
4303         /*
4304          * 1 - root node
4305          * 1 - root item
4306          */
4307         trans = btrfs_start_transaction(tree_root, 2);
4308         if (IS_ERR(trans))
4309                 return PTR_ERR(trans);
4310
4311         uuid_root = btrfs_create_tree(trans, fs_info,
4312                                       BTRFS_UUID_TREE_OBJECTID);
4313         if (IS_ERR(uuid_root)) {
4314                 ret = PTR_ERR(uuid_root);
4315                 btrfs_abort_transaction(trans, ret);
4316                 btrfs_end_transaction(trans);
4317                 return ret;
4318         }
4319
4320         fs_info->uuid_root = uuid_root;
4321
4322         ret = btrfs_commit_transaction(trans);
4323         if (ret)
4324                 return ret;
4325
4326         down(&fs_info->uuid_tree_rescan_sem);
4327         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4328         if (IS_ERR(task)) {
4329                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4330                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4331                 up(&fs_info->uuid_tree_rescan_sem);
4332                 return PTR_ERR(task);
4333         }
4334
4335         return 0;
4336 }
4337
4338 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4339 {
4340         struct task_struct *task;
4341
4342         down(&fs_info->uuid_tree_rescan_sem);
4343         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4344         if (IS_ERR(task)) {
4345                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4346                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4347                 up(&fs_info->uuid_tree_rescan_sem);
4348                 return PTR_ERR(task);
4349         }
4350
4351         return 0;
4352 }
4353
4354 /*
4355  * shrinking a device means finding all of the device extents past
4356  * the new size, and then following the back refs to the chunks.
4357  * The chunk relocation code actually frees the device extent
4358  */
4359 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4360 {
4361         struct btrfs_fs_info *fs_info = device->fs_info;
4362         struct btrfs_root *root = fs_info->dev_root;
4363         struct btrfs_trans_handle *trans;
4364         struct btrfs_dev_extent *dev_extent = NULL;
4365         struct btrfs_path *path;
4366         u64 length;
4367         u64 chunk_offset;
4368         int ret;
4369         int slot;
4370         int failed = 0;
4371         bool retried = false;
4372         bool checked_pending_chunks = false;
4373         struct extent_buffer *l;
4374         struct btrfs_key key;
4375         struct btrfs_super_block *super_copy = fs_info->super_copy;
4376         u64 old_total = btrfs_super_total_bytes(super_copy);
4377         u64 old_size = btrfs_device_get_total_bytes(device);
4378         u64 diff;
4379
4380         new_size = round_down(new_size, fs_info->sectorsize);
4381         diff = round_down(old_size - new_size, fs_info->sectorsize);
4382
4383         if (device->is_tgtdev_for_dev_replace)
4384                 return -EINVAL;
4385
4386         path = btrfs_alloc_path();
4387         if (!path)
4388                 return -ENOMEM;
4389
4390         path->reada = READA_FORWARD;
4391
4392         mutex_lock(&fs_info->chunk_mutex);
4393
4394         btrfs_device_set_total_bytes(device, new_size);
4395         if (device->writeable) {
4396                 device->fs_devices->total_rw_bytes -= diff;
4397                 atomic64_sub(diff, &fs_info->free_chunk_space);
4398         }
4399         mutex_unlock(&fs_info->chunk_mutex);
4400
4401 again:
4402         key.objectid = device->devid;
4403         key.offset = (u64)-1;
4404         key.type = BTRFS_DEV_EXTENT_KEY;
4405
4406         do {
4407                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4408                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4409                 if (ret < 0) {
4410                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4411                         goto done;
4412                 }
4413
4414                 ret = btrfs_previous_item(root, path, 0, key.type);
4415                 if (ret)
4416                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4417                 if (ret < 0)
4418                         goto done;
4419                 if (ret) {
4420                         ret = 0;
4421                         btrfs_release_path(path);
4422                         break;
4423                 }
4424
4425                 l = path->nodes[0];
4426                 slot = path->slots[0];
4427                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4428
4429                 if (key.objectid != device->devid) {
4430                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4431                         btrfs_release_path(path);
4432                         break;
4433                 }
4434
4435                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4436                 length = btrfs_dev_extent_length(l, dev_extent);
4437
4438                 if (key.offset + length <= new_size) {
4439                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4440                         btrfs_release_path(path);
4441                         break;
4442                 }
4443
4444                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4445                 btrfs_release_path(path);
4446
4447                 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4448                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4449                 if (ret && ret != -ENOSPC)
4450                         goto done;
4451                 if (ret == -ENOSPC)
4452                         failed++;
4453         } while (key.offset-- > 0);
4454
4455         if (failed && !retried) {
4456                 failed = 0;
4457                 retried = true;
4458                 goto again;
4459         } else if (failed && retried) {
4460                 ret = -ENOSPC;
4461                 goto done;
4462         }
4463
4464         /* Shrinking succeeded, else we would be at "done". */
4465         trans = btrfs_start_transaction(root, 0);
4466         if (IS_ERR(trans)) {
4467                 ret = PTR_ERR(trans);
4468                 goto done;
4469         }
4470
4471         mutex_lock(&fs_info->chunk_mutex);
4472
4473         /*
4474          * We checked in the above loop all device extents that were already in
4475          * the device tree. However before we have updated the device's
4476          * total_bytes to the new size, we might have had chunk allocations that
4477          * have not complete yet (new block groups attached to transaction
4478          * handles), and therefore their device extents were not yet in the
4479          * device tree and we missed them in the loop above. So if we have any
4480          * pending chunk using a device extent that overlaps the device range
4481          * that we can not use anymore, commit the current transaction and
4482          * repeat the search on the device tree - this way we guarantee we will
4483          * not have chunks using device extents that end beyond 'new_size'.
4484          */
4485         if (!checked_pending_chunks) {
4486                 u64 start = new_size;
4487                 u64 len = old_size - new_size;
4488
4489                 if (contains_pending_extent(trans->transaction, device,
4490                                             &start, len)) {
4491                         mutex_unlock(&fs_info->chunk_mutex);
4492                         checked_pending_chunks = true;
4493                         failed = 0;
4494                         retried = false;
4495                         ret = btrfs_commit_transaction(trans);
4496                         if (ret)
4497                                 goto done;
4498                         goto again;
4499                 }
4500         }
4501
4502         btrfs_device_set_disk_total_bytes(device, new_size);
4503         if (list_empty(&device->resized_list))
4504                 list_add_tail(&device->resized_list,
4505                               &fs_info->fs_devices->resized_devices);
4506
4507         WARN_ON(diff > old_total);
4508         btrfs_set_super_total_bytes(super_copy,
4509                         round_down(old_total - diff, fs_info->sectorsize));
4510         mutex_unlock(&fs_info->chunk_mutex);
4511
4512         /* Now btrfs_update_device() will change the on-disk size. */
4513         ret = btrfs_update_device(trans, device);
4514         btrfs_end_transaction(trans);
4515 done:
4516         btrfs_free_path(path);
4517         if (ret) {
4518                 mutex_lock(&fs_info->chunk_mutex);
4519                 btrfs_device_set_total_bytes(device, old_size);
4520                 if (device->writeable)
4521                         device->fs_devices->total_rw_bytes += diff;
4522                 atomic64_add(diff, &fs_info->free_chunk_space);
4523                 mutex_unlock(&fs_info->chunk_mutex);
4524         }
4525         return ret;
4526 }
4527
4528 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4529                            struct btrfs_key *key,
4530                            struct btrfs_chunk *chunk, int item_size)
4531 {
4532         struct btrfs_super_block *super_copy = fs_info->super_copy;
4533         struct btrfs_disk_key disk_key;
4534         u32 array_size;
4535         u8 *ptr;
4536
4537         mutex_lock(&fs_info->chunk_mutex);
4538         array_size = btrfs_super_sys_array_size(super_copy);
4539         if (array_size + item_size + sizeof(disk_key)
4540                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4541                 mutex_unlock(&fs_info->chunk_mutex);
4542                 return -EFBIG;
4543         }
4544
4545         ptr = super_copy->sys_chunk_array + array_size;
4546         btrfs_cpu_key_to_disk(&disk_key, key);
4547         memcpy(ptr, &disk_key, sizeof(disk_key));
4548         ptr += sizeof(disk_key);
4549         memcpy(ptr, chunk, item_size);
4550         item_size += sizeof(disk_key);
4551         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4552         mutex_unlock(&fs_info->chunk_mutex);
4553
4554         return 0;
4555 }
4556
4557 /*
4558  * sort the devices in descending order by max_avail, total_avail
4559  */
4560 static int btrfs_cmp_device_info(const void *a, const void *b)
4561 {
4562         const struct btrfs_device_info *di_a = a;
4563         const struct btrfs_device_info *di_b = b;
4564
4565         if (di_a->max_avail > di_b->max_avail)
4566                 return -1;
4567         if (di_a->max_avail < di_b->max_avail)
4568                 return 1;
4569         if (di_a->total_avail > di_b->total_avail)
4570                 return -1;
4571         if (di_a->total_avail < di_b->total_avail)
4572                 return 1;
4573         return 0;
4574 }
4575
4576 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4577 {
4578         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4579                 return;
4580
4581         btrfs_set_fs_incompat(info, RAID56);
4582 }
4583
4584 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info)             \
4585                         - sizeof(struct btrfs_chunk))           \
4586                         / sizeof(struct btrfs_stripe) + 1)
4587
4588 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4589                                 - 2 * sizeof(struct btrfs_disk_key)     \
4590                                 - 2 * sizeof(struct btrfs_chunk))       \
4591                                 / sizeof(struct btrfs_stripe) + 1)
4592
4593 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4594                                u64 start, u64 type)
4595 {
4596         struct btrfs_fs_info *info = trans->fs_info;
4597         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4598         struct btrfs_device *device;
4599         struct map_lookup *map = NULL;
4600         struct extent_map_tree *em_tree;
4601         struct extent_map *em;
4602         struct btrfs_device_info *devices_info = NULL;
4603         u64 total_avail;
4604         int num_stripes;        /* total number of stripes to allocate */
4605         int data_stripes;       /* number of stripes that count for
4606                                    block group size */
4607         int sub_stripes;        /* sub_stripes info for map */
4608         int dev_stripes;        /* stripes per dev */
4609         int devs_max;           /* max devs to use */
4610         int devs_min;           /* min devs needed */
4611         int devs_increment;     /* ndevs has to be a multiple of this */
4612         int ncopies;            /* how many copies to data has */
4613         int ret;
4614         u64 max_stripe_size;
4615         u64 max_chunk_size;
4616         u64 stripe_size;
4617         u64 num_bytes;
4618         int ndevs;
4619         int i;
4620         int j;
4621         int index;
4622
4623         BUG_ON(!alloc_profile_is_valid(type, 0));
4624
4625         if (list_empty(&fs_devices->alloc_list))
4626                 return -ENOSPC;
4627
4628         index = __get_raid_index(type);
4629
4630         sub_stripes = btrfs_raid_array[index].sub_stripes;
4631         dev_stripes = btrfs_raid_array[index].dev_stripes;
4632         devs_max = btrfs_raid_array[index].devs_max;
4633         devs_min = btrfs_raid_array[index].devs_min;
4634         devs_increment = btrfs_raid_array[index].devs_increment;
4635         ncopies = btrfs_raid_array[index].ncopies;
4636
4637         if (type & BTRFS_BLOCK_GROUP_DATA) {
4638                 max_stripe_size = SZ_1G;
4639                 max_chunk_size = 10 * max_stripe_size;
4640                 if (!devs_max)
4641                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4642         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4643                 /* for larger filesystems, use larger metadata chunks */
4644                 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4645                         max_stripe_size = SZ_1G;
4646                 else
4647                         max_stripe_size = SZ_256M;
4648                 max_chunk_size = max_stripe_size;
4649                 if (!devs_max)
4650                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4651         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4652                 max_stripe_size = SZ_32M;
4653                 max_chunk_size = 2 * max_stripe_size;
4654                 if (!devs_max)
4655                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4656         } else {
4657                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4658                        type);
4659                 BUG_ON(1);
4660         }
4661
4662         /* we don't want a chunk larger than 10% of writeable space */
4663         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4664                              max_chunk_size);
4665
4666         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4667                                GFP_NOFS);
4668         if (!devices_info)
4669                 return -ENOMEM;
4670
4671         /*
4672          * in the first pass through the devices list, we gather information
4673          * about the available holes on each device.
4674          */
4675         ndevs = 0;
4676         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4677                 u64 max_avail;
4678                 u64 dev_offset;
4679
4680                 if (!device->writeable) {
4681                         WARN(1, KERN_ERR
4682                                "BTRFS: read-only device in alloc_list\n");
4683                         continue;
4684                 }
4685
4686                 if (!device->in_fs_metadata ||
4687                     device->is_tgtdev_for_dev_replace)
4688                         continue;
4689
4690                 if (device->total_bytes > device->bytes_used)
4691                         total_avail = device->total_bytes - device->bytes_used;
4692                 else
4693                         total_avail = 0;
4694
4695                 /* If there is no space on this device, skip it. */
4696                 if (total_avail == 0)
4697                         continue;
4698
4699                 ret = find_free_dev_extent(trans, device,
4700                                            max_stripe_size * dev_stripes,
4701                                            &dev_offset, &max_avail);
4702                 if (ret && ret != -ENOSPC)
4703                         goto error;
4704
4705                 if (ret == 0)
4706                         max_avail = max_stripe_size * dev_stripes;
4707
4708                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4709                         continue;
4710
4711                 if (ndevs == fs_devices->rw_devices) {
4712                         WARN(1, "%s: found more than %llu devices\n",
4713                              __func__, fs_devices->rw_devices);
4714                         break;
4715                 }
4716                 devices_info[ndevs].dev_offset = dev_offset;
4717                 devices_info[ndevs].max_avail = max_avail;
4718                 devices_info[ndevs].total_avail = total_avail;
4719                 devices_info[ndevs].dev = device;
4720                 ++ndevs;
4721         }
4722
4723         /*
4724          * now sort the devices by hole size / available space
4725          */
4726         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4727              btrfs_cmp_device_info, NULL);
4728
4729         /* round down to number of usable stripes */
4730         ndevs = round_down(ndevs, devs_increment);
4731
4732         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4733                 ret = -ENOSPC;
4734                 goto error;
4735         }
4736
4737         ndevs = min(ndevs, devs_max);
4738
4739         /*
4740          * the primary goal is to maximize the number of stripes, so use as many
4741          * devices as possible, even if the stripes are not maximum sized.
4742          */
4743         stripe_size = devices_info[ndevs-1].max_avail;
4744         num_stripes = ndevs * dev_stripes;
4745
4746         /*
4747          * this will have to be fixed for RAID1 and RAID10 over
4748          * more drives
4749          */
4750         data_stripes = num_stripes / ncopies;
4751
4752         if (type & BTRFS_BLOCK_GROUP_RAID5)
4753                 data_stripes = num_stripes - 1;
4754
4755         if (type & BTRFS_BLOCK_GROUP_RAID6)
4756                 data_stripes = num_stripes - 2;
4757
4758         /*
4759          * Use the number of data stripes to figure out how big this chunk
4760          * is really going to be in terms of logical address space,
4761          * and compare that answer with the max chunk size
4762          */
4763         if (stripe_size * data_stripes > max_chunk_size) {
4764                 u64 mask = (1ULL << 24) - 1;
4765
4766                 stripe_size = div_u64(max_chunk_size, data_stripes);
4767
4768                 /* bump the answer up to a 16MB boundary */
4769                 stripe_size = (stripe_size + mask) & ~mask;
4770
4771                 /* but don't go higher than the limits we found
4772                  * while searching for free extents
4773                  */
4774                 if (stripe_size > devices_info[ndevs-1].max_avail)
4775                         stripe_size = devices_info[ndevs-1].max_avail;
4776         }
4777
4778         stripe_size = div_u64(stripe_size, dev_stripes);
4779
4780         /* align to BTRFS_STRIPE_LEN */
4781         stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
4782
4783         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4784         if (!map) {
4785                 ret = -ENOMEM;
4786                 goto error;
4787         }
4788         map->num_stripes = num_stripes;
4789
4790         for (i = 0; i < ndevs; ++i) {
4791                 for (j = 0; j < dev_stripes; ++j) {
4792                         int s = i * dev_stripes + j;
4793                         map->stripes[s].dev = devices_info[i].dev;
4794                         map->stripes[s].physical = devices_info[i].dev_offset +
4795                                                    j * stripe_size;
4796                 }
4797         }
4798         map->stripe_len = BTRFS_STRIPE_LEN;
4799         map->io_align = BTRFS_STRIPE_LEN;
4800         map->io_width = BTRFS_STRIPE_LEN;
4801         map->type = type;
4802         map->sub_stripes = sub_stripes;
4803
4804         num_bytes = stripe_size * data_stripes;
4805
4806         trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4807
4808         em = alloc_extent_map();
4809         if (!em) {
4810                 kfree(map);
4811                 ret = -ENOMEM;
4812                 goto error;
4813         }
4814         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4815         em->map_lookup = map;
4816         em->start = start;
4817         em->len = num_bytes;
4818         em->block_start = 0;
4819         em->block_len = em->len;
4820         em->orig_block_len = stripe_size;
4821
4822         em_tree = &info->mapping_tree.map_tree;
4823         write_lock(&em_tree->lock);
4824         ret = add_extent_mapping(em_tree, em, 0);
4825         if (ret) {
4826                 write_unlock(&em_tree->lock);
4827                 free_extent_map(em);
4828                 goto error;
4829         }
4830
4831         list_add_tail(&em->list, &trans->transaction->pending_chunks);
4832         refcount_inc(&em->refs);
4833         write_unlock(&em_tree->lock);
4834
4835         ret = btrfs_make_block_group(trans, info, 0, type, start, num_bytes);
4836         if (ret)
4837                 goto error_del_extent;
4838
4839         for (i = 0; i < map->num_stripes; i++) {
4840                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4841                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4842         }
4843
4844         atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4845
4846         free_extent_map(em);
4847         check_raid56_incompat_flag(info, type);
4848
4849         kfree(devices_info);
4850         return 0;
4851
4852 error_del_extent:
4853         write_lock(&em_tree->lock);
4854         remove_extent_mapping(em_tree, em);
4855         write_unlock(&em_tree->lock);
4856
4857         /* One for our allocation */
4858         free_extent_map(em);
4859         /* One for the tree reference */
4860         free_extent_map(em);
4861         /* One for the pending_chunks list reference */
4862         free_extent_map(em);
4863 error:
4864         kfree(devices_info);
4865         return ret;
4866 }
4867
4868 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4869                                 struct btrfs_fs_info *fs_info,
4870                                 u64 chunk_offset, u64 chunk_size)
4871 {
4872         struct btrfs_root *extent_root = fs_info->extent_root;
4873         struct btrfs_root *chunk_root = fs_info->chunk_root;
4874         struct btrfs_key key;
4875         struct btrfs_device *device;
4876         struct btrfs_chunk *chunk;
4877         struct btrfs_stripe *stripe;
4878         struct extent_map *em;
4879         struct map_lookup *map;
4880         size_t item_size;
4881         u64 dev_offset;
4882         u64 stripe_size;
4883         int i = 0;
4884         int ret = 0;
4885
4886         em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4887         if (IS_ERR(em))
4888                 return PTR_ERR(em);
4889
4890         map = em->map_lookup;
4891         item_size = btrfs_chunk_item_size(map->num_stripes);
4892         stripe_size = em->orig_block_len;
4893
4894         chunk = kzalloc(item_size, GFP_NOFS);
4895         if (!chunk) {
4896                 ret = -ENOMEM;
4897                 goto out;
4898         }
4899
4900         /*
4901          * Take the device list mutex to prevent races with the final phase of
4902          * a device replace operation that replaces the device object associated
4903          * with the map's stripes, because the device object's id can change
4904          * at any time during that final phase of the device replace operation
4905          * (dev-replace.c:btrfs_dev_replace_finishing()).
4906          */
4907         mutex_lock(&fs_info->fs_devices->device_list_mutex);
4908         for (i = 0; i < map->num_stripes; i++) {
4909                 device = map->stripes[i].dev;
4910                 dev_offset = map->stripes[i].physical;
4911
4912                 ret = btrfs_update_device(trans, device);
4913                 if (ret)
4914                         break;
4915                 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
4916                                              dev_offset, stripe_size);
4917                 if (ret)
4918                         break;
4919         }
4920         if (ret) {
4921                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4922                 goto out;
4923         }
4924
4925         stripe = &chunk->stripe;
4926         for (i = 0; i < map->num_stripes; i++) {
4927                 device = map->stripes[i].dev;
4928                 dev_offset = map->stripes[i].physical;
4929
4930                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4931                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4932                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4933                 stripe++;
4934         }
4935         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4936
4937         btrfs_set_stack_chunk_length(chunk, chunk_size);
4938         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4939         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4940         btrfs_set_stack_chunk_type(chunk, map->type);
4941         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4942         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4943         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4944         btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
4945         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4946
4947         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4948         key.type = BTRFS_CHUNK_ITEM_KEY;
4949         key.offset = chunk_offset;
4950
4951         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4952         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4953                 /*
4954                  * TODO: Cleanup of inserted chunk root in case of
4955                  * failure.
4956                  */
4957                 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
4958         }
4959
4960 out:
4961         kfree(chunk);
4962         free_extent_map(em);
4963         return ret;
4964 }
4965
4966 /*
4967  * Chunk allocation falls into two parts. The first part does works
4968  * that make the new allocated chunk useable, but not do any operation
4969  * that modifies the chunk tree. The second part does the works that
4970  * require modifying the chunk tree. This division is important for the
4971  * bootstrap process of adding storage to a seed btrfs.
4972  */
4973 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4974                       struct btrfs_fs_info *fs_info, u64 type)
4975 {
4976         u64 chunk_offset;
4977
4978         ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
4979         chunk_offset = find_next_chunk(fs_info);
4980         return __btrfs_alloc_chunk(trans, chunk_offset, type);
4981 }
4982
4983 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4984                                          struct btrfs_fs_info *fs_info)
4985 {
4986         u64 chunk_offset;
4987         u64 sys_chunk_offset;
4988         u64 alloc_profile;
4989         int ret;
4990
4991         chunk_offset = find_next_chunk(fs_info);
4992         alloc_profile = btrfs_metadata_alloc_profile(fs_info);
4993         ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
4994         if (ret)
4995                 return ret;
4996
4997         sys_chunk_offset = find_next_chunk(fs_info);
4998         alloc_profile = btrfs_system_alloc_profile(fs_info);
4999         ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5000         return ret;
5001 }
5002
5003 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5004 {
5005         int max_errors;
5006
5007         if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5008                          BTRFS_BLOCK_GROUP_RAID10 |
5009                          BTRFS_BLOCK_GROUP_RAID5 |
5010                          BTRFS_BLOCK_GROUP_DUP)) {
5011                 max_errors = 1;
5012         } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5013                 max_errors = 2;
5014         } else {
5015                 max_errors = 0;
5016         }
5017
5018         return max_errors;
5019 }
5020
5021 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5022 {
5023         struct extent_map *em;
5024         struct map_lookup *map;
5025         int readonly = 0;
5026         int miss_ndevs = 0;
5027         int i;
5028
5029         em = get_chunk_map(fs_info, chunk_offset, 1);
5030         if (IS_ERR(em))
5031                 return 1;
5032
5033         map = em->map_lookup;
5034         for (i = 0; i < map->num_stripes; i++) {
5035                 if (map->stripes[i].dev->missing) {
5036                         miss_ndevs++;
5037                         continue;
5038                 }
5039
5040                 if (!map->stripes[i].dev->writeable) {
5041                         readonly = 1;
5042                         goto end;
5043                 }
5044         }
5045
5046         /*
5047          * If the number of missing devices is larger than max errors,
5048          * we can not write the data into that chunk successfully, so
5049          * set it readonly.
5050          */
5051         if (miss_ndevs > btrfs_chunk_max_errors(map))
5052                 readonly = 1;
5053 end:
5054         free_extent_map(em);
5055         return readonly;
5056 }
5057
5058 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5059 {
5060         extent_map_tree_init(&tree->map_tree);
5061 }
5062
5063 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5064 {
5065         struct extent_map *em;
5066
5067         while (1) {
5068                 write_lock(&tree->map_tree.lock);
5069                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5070                 if (em)
5071                         remove_extent_mapping(&tree->map_tree, em);
5072                 write_unlock(&tree->map_tree.lock);
5073                 if (!em)
5074                         break;
5075                 /* once for us */
5076                 free_extent_map(em);
5077                 /* once for the tree */
5078                 free_extent_map(em);
5079         }
5080 }
5081
5082 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5083 {
5084         struct extent_map *em;
5085         struct map_lookup *map;
5086         int ret;
5087
5088         em = get_chunk_map(fs_info, logical, len);
5089         if (IS_ERR(em))
5090                 /*
5091                  * We could return errors for these cases, but that could get
5092                  * ugly and we'd probably do the same thing which is just not do
5093                  * anything else and exit, so return 1 so the callers don't try
5094                  * to use other copies.
5095                  */
5096                 return 1;
5097
5098         map = em->map_lookup;
5099         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5100                 ret = map->num_stripes;
5101         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5102                 ret = map->sub_stripes;
5103         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5104                 ret = 2;
5105         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5106                 ret = 3;
5107         else
5108                 ret = 1;
5109         free_extent_map(em);
5110
5111         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5112         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5113             fs_info->dev_replace.tgtdev)
5114                 ret++;
5115         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5116
5117         return ret;
5118 }
5119
5120 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5121                                     u64 logical)
5122 {
5123         struct extent_map *em;
5124         struct map_lookup *map;
5125         unsigned long len = fs_info->sectorsize;
5126
5127         em = get_chunk_map(fs_info, logical, len);
5128
5129         if (!WARN_ON(IS_ERR(em))) {
5130                 map = em->map_lookup;
5131                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5132                         len = map->stripe_len * nr_data_stripes(map);
5133                 free_extent_map(em);
5134         }
5135         return len;
5136 }
5137
5138 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5139 {
5140         struct extent_map *em;
5141         struct map_lookup *map;
5142         int ret = 0;
5143
5144         em = get_chunk_map(fs_info, logical, len);
5145
5146         if(!WARN_ON(IS_ERR(em))) {
5147                 map = em->map_lookup;
5148                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5149                         ret = 1;
5150                 free_extent_map(em);
5151         }
5152         return ret;
5153 }
5154
5155 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5156                             struct map_lookup *map, int first, int num,
5157                             int optimal, int dev_replace_is_ongoing)
5158 {
5159         int i;
5160         int tolerance;
5161         struct btrfs_device *srcdev;
5162
5163         if (dev_replace_is_ongoing &&
5164             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5165              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5166                 srcdev = fs_info->dev_replace.srcdev;
5167         else
5168                 srcdev = NULL;
5169
5170         /*
5171          * try to avoid the drive that is the source drive for a
5172          * dev-replace procedure, only choose it if no other non-missing
5173          * mirror is available
5174          */
5175         for (tolerance = 0; tolerance < 2; tolerance++) {
5176                 if (map->stripes[optimal].dev->bdev &&
5177                     (tolerance || map->stripes[optimal].dev != srcdev))
5178                         return optimal;
5179                 for (i = first; i < first + num; i++) {
5180                         if (map->stripes[i].dev->bdev &&
5181                             (tolerance || map->stripes[i].dev != srcdev))
5182                                 return i;
5183                 }
5184         }
5185
5186         /* we couldn't find one that doesn't fail.  Just return something
5187          * and the io error handling code will clean up eventually
5188          */
5189         return optimal;
5190 }
5191
5192 static inline int parity_smaller(u64 a, u64 b)
5193 {
5194         return a > b;
5195 }
5196
5197 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5198 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5199 {
5200         struct btrfs_bio_stripe s;
5201         int i;
5202         u64 l;
5203         int again = 1;
5204
5205         while (again) {
5206                 again = 0;
5207                 for (i = 0; i < num_stripes - 1; i++) {
5208                         if (parity_smaller(bbio->raid_map[i],
5209                                            bbio->raid_map[i+1])) {
5210                                 s = bbio->stripes[i];
5211                                 l = bbio->raid_map[i];
5212                                 bbio->stripes[i] = bbio->stripes[i+1];
5213                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5214                                 bbio->stripes[i+1] = s;
5215                                 bbio->raid_map[i+1] = l;
5216
5217                                 again = 1;
5218                         }
5219                 }
5220         }
5221 }
5222
5223 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5224 {
5225         struct btrfs_bio *bbio = kzalloc(
5226                  /* the size of the btrfs_bio */
5227                 sizeof(struct btrfs_bio) +
5228                 /* plus the variable array for the stripes */
5229                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5230                 /* plus the variable array for the tgt dev */
5231                 sizeof(int) * (real_stripes) +
5232                 /*
5233                  * plus the raid_map, which includes both the tgt dev
5234                  * and the stripes
5235                  */
5236                 sizeof(u64) * (total_stripes),
5237                 GFP_NOFS|__GFP_NOFAIL);
5238
5239         atomic_set(&bbio->error, 0);
5240         refcount_set(&bbio->refs, 1);
5241
5242         return bbio;
5243 }
5244
5245 void btrfs_get_bbio(struct btrfs_bio *bbio)
5246 {
5247         WARN_ON(!refcount_read(&bbio->refs));
5248         refcount_inc(&bbio->refs);
5249 }
5250
5251 void btrfs_put_bbio(struct btrfs_bio *bbio)
5252 {
5253         if (!bbio)
5254                 return;
5255         if (refcount_dec_and_test(&bbio->refs))
5256                 kfree(bbio);
5257 }
5258
5259 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5260 /*
5261  * Please note that, discard won't be sent to target device of device
5262  * replace.
5263  */
5264 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5265                                          u64 logical, u64 length,
5266                                          struct btrfs_bio **bbio_ret)
5267 {
5268         struct extent_map *em;
5269         struct map_lookup *map;
5270         struct btrfs_bio *bbio;
5271         u64 offset;
5272         u64 stripe_nr;
5273         u64 stripe_nr_end;
5274         u64 stripe_end_offset;
5275         u64 stripe_cnt;
5276         u64 stripe_len;
5277         u64 stripe_offset;
5278         u64 num_stripes;
5279         u32 stripe_index;
5280         u32 factor = 0;
5281         u32 sub_stripes = 0;
5282         u64 stripes_per_dev = 0;
5283         u32 remaining_stripes = 0;
5284         u32 last_stripe = 0;
5285         int ret = 0;
5286         int i;
5287
5288         /* discard always return a bbio */
5289         ASSERT(bbio_ret);
5290
5291         em = get_chunk_map(fs_info, logical, length);
5292         if (IS_ERR(em))
5293                 return PTR_ERR(em);
5294
5295         map = em->map_lookup;
5296         /* we don't discard raid56 yet */
5297         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5298                 ret = -EOPNOTSUPP;
5299                 goto out;
5300         }
5301
5302         offset = logical - em->start;
5303         length = min_t(u64, em->len - offset, length);
5304
5305         stripe_len = map->stripe_len;
5306         /*
5307          * stripe_nr counts the total number of stripes we have to stride
5308          * to get to this block
5309          */
5310         stripe_nr = div64_u64(offset, stripe_len);
5311
5312         /* stripe_offset is the offset of this block in its stripe */
5313         stripe_offset = offset - stripe_nr * stripe_len;
5314
5315         stripe_nr_end = round_up(offset + length, map->stripe_len);
5316         stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5317         stripe_cnt = stripe_nr_end - stripe_nr;
5318         stripe_end_offset = stripe_nr_end * map->stripe_len -
5319                             (offset + length);
5320         /*
5321          * after this, stripe_nr is the number of stripes on this
5322          * device we have to walk to find the data, and stripe_index is
5323          * the number of our device in the stripe array
5324          */
5325         num_stripes = 1;
5326         stripe_index = 0;
5327         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5328                          BTRFS_BLOCK_GROUP_RAID10)) {
5329                 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5330                         sub_stripes = 1;
5331                 else
5332                         sub_stripes = map->sub_stripes;
5333
5334                 factor = map->num_stripes / sub_stripes;
5335                 num_stripes = min_t(u64, map->num_stripes,
5336                                     sub_stripes * stripe_cnt);
5337                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5338                 stripe_index *= sub_stripes;
5339                 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5340                                               &remaining_stripes);
5341                 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5342                 last_stripe *= sub_stripes;
5343         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5344                                 BTRFS_BLOCK_GROUP_DUP)) {
5345                 num_stripes = map->num_stripes;
5346         } else {
5347                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5348                                         &stripe_index);
5349         }
5350
5351         bbio = alloc_btrfs_bio(num_stripes, 0);
5352         if (!bbio) {
5353                 ret = -ENOMEM;
5354                 goto out;
5355         }
5356
5357         for (i = 0; i < num_stripes; i++) {
5358                 bbio->stripes[i].physical =
5359                         map->stripes[stripe_index].physical +
5360                         stripe_offset + stripe_nr * map->stripe_len;
5361                 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5362
5363                 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5364                                  BTRFS_BLOCK_GROUP_RAID10)) {
5365                         bbio->stripes[i].length = stripes_per_dev *
5366                                 map->stripe_len;
5367
5368                         if (i / sub_stripes < remaining_stripes)
5369                                 bbio->stripes[i].length +=
5370                                         map->stripe_len;
5371
5372                         /*
5373                          * Special for the first stripe and
5374                          * the last stripe:
5375                          *
5376                          * |-------|...|-------|
5377                          *     |----------|
5378                          *    off     end_off
5379                          */
5380                         if (i < sub_stripes)
5381                                 bbio->stripes[i].length -=
5382                                         stripe_offset;
5383
5384                         if (stripe_index >= last_stripe &&
5385                             stripe_index <= (last_stripe +
5386                                              sub_stripes - 1))
5387                                 bbio->stripes[i].length -=
5388                                         stripe_end_offset;
5389
5390                         if (i == sub_stripes - 1)
5391                                 stripe_offset = 0;
5392                 } else {
5393                         bbio->stripes[i].length = length;
5394                 }
5395
5396                 stripe_index++;
5397                 if (stripe_index == map->num_stripes) {
5398                         stripe_index = 0;
5399                         stripe_nr++;
5400                 }
5401         }
5402
5403         *bbio_ret = bbio;
5404         bbio->map_type = map->type;
5405         bbio->num_stripes = num_stripes;
5406 out:
5407         free_extent_map(em);
5408         return ret;
5409 }
5410
5411 /*
5412  * In dev-replace case, for repair case (that's the only case where the mirror
5413  * is selected explicitly when calling btrfs_map_block), blocks left of the
5414  * left cursor can also be read from the target drive.
5415  *
5416  * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5417  * array of stripes.
5418  * For READ, it also needs to be supported using the same mirror number.
5419  *
5420  * If the requested block is not left of the left cursor, EIO is returned. This
5421  * can happen because btrfs_num_copies() returns one more in the dev-replace
5422  * case.
5423  */
5424 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5425                                          u64 logical, u64 length,
5426                                          u64 srcdev_devid, int *mirror_num,
5427                                          u64 *physical)
5428 {
5429         struct btrfs_bio *bbio = NULL;
5430         int num_stripes;
5431         int index_srcdev = 0;
5432         int found = 0;
5433         u64 physical_of_found = 0;
5434         int i;
5435         int ret = 0;
5436
5437         ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5438                                 logical, &length, &bbio, 0, 0);
5439         if (ret) {
5440                 ASSERT(bbio == NULL);
5441                 return ret;
5442         }
5443
5444         num_stripes = bbio->num_stripes;
5445         if (*mirror_num > num_stripes) {
5446                 /*
5447                  * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5448                  * that means that the requested area is not left of the left
5449                  * cursor
5450                  */
5451                 btrfs_put_bbio(bbio);
5452                 return -EIO;
5453         }
5454
5455         /*
5456          * process the rest of the function using the mirror_num of the source
5457          * drive. Therefore look it up first.  At the end, patch the device
5458          * pointer to the one of the target drive.
5459          */
5460         for (i = 0; i < num_stripes; i++) {
5461                 if (bbio->stripes[i].dev->devid != srcdev_devid)
5462                         continue;
5463
5464                 /*
5465                  * In case of DUP, in order to keep it simple, only add the
5466                  * mirror with the lowest physical address
5467                  */
5468                 if (found &&
5469                     physical_of_found <= bbio->stripes[i].physical)
5470                         continue;
5471
5472                 index_srcdev = i;
5473                 found = 1;
5474                 physical_of_found = bbio->stripes[i].physical;
5475         }
5476
5477         btrfs_put_bbio(bbio);
5478
5479         ASSERT(found);
5480         if (!found)
5481                 return -EIO;
5482
5483         *mirror_num = index_srcdev + 1;
5484         *physical = physical_of_found;
5485         return ret;
5486 }
5487
5488 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5489                                       struct btrfs_bio **bbio_ret,
5490                                       struct btrfs_dev_replace *dev_replace,
5491                                       int *num_stripes_ret, int *max_errors_ret)
5492 {
5493         struct btrfs_bio *bbio = *bbio_ret;
5494         u64 srcdev_devid = dev_replace->srcdev->devid;
5495         int tgtdev_indexes = 0;
5496         int num_stripes = *num_stripes_ret;
5497         int max_errors = *max_errors_ret;
5498         int i;
5499
5500         if (op == BTRFS_MAP_WRITE) {
5501                 int index_where_to_add;
5502
5503                 /*
5504                  * duplicate the write operations while the dev replace
5505                  * procedure is running. Since the copying of the old disk to
5506                  * the new disk takes place at run time while the filesystem is
5507                  * mounted writable, the regular write operations to the old
5508                  * disk have to be duplicated to go to the new disk as well.
5509                  *
5510                  * Note that device->missing is handled by the caller, and that
5511                  * the write to the old disk is already set up in the stripes
5512                  * array.
5513                  */
5514                 index_where_to_add = num_stripes;
5515                 for (i = 0; i < num_stripes; i++) {
5516                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5517                                 /* write to new disk, too */
5518                                 struct btrfs_bio_stripe *new =
5519                                         bbio->stripes + index_where_to_add;
5520                                 struct btrfs_bio_stripe *old =
5521                                         bbio->stripes + i;
5522
5523                                 new->physical = old->physical;
5524                                 new->length = old->length;
5525                                 new->dev = dev_replace->tgtdev;
5526                                 bbio->tgtdev_map[i] = index_where_to_add;
5527                                 index_where_to_add++;
5528                                 max_errors++;
5529                                 tgtdev_indexes++;
5530                         }
5531                 }
5532                 num_stripes = index_where_to_add;
5533         } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5534                 int index_srcdev = 0;
5535                 int found = 0;
5536                 u64 physical_of_found = 0;
5537
5538                 /*
5539                  * During the dev-replace procedure, the target drive can also
5540                  * be used to read data in case it is needed to repair a corrupt
5541                  * block elsewhere. This is possible if the requested area is
5542                  * left of the left cursor. In this area, the target drive is a
5543                  * full copy of the source drive.
5544                  */
5545                 for (i = 0; i < num_stripes; i++) {
5546                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5547                                 /*
5548                                  * In case of DUP, in order to keep it simple,
5549                                  * only add the mirror with the lowest physical
5550                                  * address
5551                                  */
5552                                 if (found &&
5553                                     physical_of_found <=
5554                                      bbio->stripes[i].physical)
5555                                         continue;
5556                                 index_srcdev = i;
5557                                 found = 1;
5558                                 physical_of_found = bbio->stripes[i].physical;
5559                         }
5560                 }
5561                 if (found) {
5562                         struct btrfs_bio_stripe *tgtdev_stripe =
5563                                 bbio->stripes + num_stripes;
5564
5565                         tgtdev_stripe->physical = physical_of_found;
5566                         tgtdev_stripe->length =
5567                                 bbio->stripes[index_srcdev].length;
5568                         tgtdev_stripe->dev = dev_replace->tgtdev;
5569                         bbio->tgtdev_map[index_srcdev] = num_stripes;
5570
5571                         tgtdev_indexes++;
5572                         num_stripes++;
5573                 }
5574         }
5575
5576         *num_stripes_ret = num_stripes;
5577         *max_errors_ret = max_errors;
5578         bbio->num_tgtdevs = tgtdev_indexes;
5579         *bbio_ret = bbio;
5580 }
5581
5582 static bool need_full_stripe(enum btrfs_map_op op)
5583 {
5584         return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5585 }
5586
5587 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5588                              enum btrfs_map_op op,
5589                              u64 logical, u64 *length,
5590                              struct btrfs_bio **bbio_ret,
5591                              int mirror_num, int need_raid_map)
5592 {
5593         struct extent_map *em;
5594         struct map_lookup *map;
5595         u64 offset;
5596         u64 stripe_offset;
5597         u64 stripe_nr;
5598         u64 stripe_len;
5599         u32 stripe_index;
5600         int i;
5601         int ret = 0;
5602         int num_stripes;
5603         int max_errors = 0;
5604         int tgtdev_indexes = 0;
5605         struct btrfs_bio *bbio = NULL;
5606         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5607         int dev_replace_is_ongoing = 0;
5608         int num_alloc_stripes;
5609         int patch_the_first_stripe_for_dev_replace = 0;
5610         u64 physical_to_patch_in_first_stripe = 0;
5611         u64 raid56_full_stripe_start = (u64)-1;
5612
5613         if (op == BTRFS_MAP_DISCARD)
5614                 return __btrfs_map_block_for_discard(fs_info, logical,
5615                                                      *length, bbio_ret);
5616
5617         em = get_chunk_map(fs_info, logical, *length);
5618         if (IS_ERR(em))
5619                 return PTR_ERR(em);
5620
5621         map = em->map_lookup;
5622         offset = logical - em->start;
5623
5624         stripe_len = map->stripe_len;
5625         stripe_nr = offset;
5626         /*
5627          * stripe_nr counts the total number of stripes we have to stride
5628          * to get to this block
5629          */
5630         stripe_nr = div64_u64(stripe_nr, stripe_len);
5631
5632         stripe_offset = stripe_nr * stripe_len;
5633         if (offset < stripe_offset) {
5634                 btrfs_crit(fs_info,
5635                            "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5636                            stripe_offset, offset, em->start, logical,
5637                            stripe_len);
5638                 free_extent_map(em);
5639                 return -EINVAL;
5640         }
5641
5642         /* stripe_offset is the offset of this block in its stripe*/
5643         stripe_offset = offset - stripe_offset;
5644
5645         /* if we're here for raid56, we need to know the stripe aligned start */
5646         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5647                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5648                 raid56_full_stripe_start = offset;
5649
5650                 /* allow a write of a full stripe, but make sure we don't
5651                  * allow straddling of stripes
5652                  */
5653                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5654                                 full_stripe_len);
5655                 raid56_full_stripe_start *= full_stripe_len;
5656         }
5657
5658         if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5659                 u64 max_len;
5660                 /* For writes to RAID[56], allow a full stripeset across all disks.
5661                    For other RAID types and for RAID[56] reads, just allow a single
5662                    stripe (on a single disk). */
5663                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5664                     (op == BTRFS_MAP_WRITE)) {
5665                         max_len = stripe_len * nr_data_stripes(map) -
5666                                 (offset - raid56_full_stripe_start);
5667                 } else {
5668                         /* we limit the length of each bio to what fits in a stripe */
5669                         max_len = stripe_len - stripe_offset;
5670                 }
5671                 *length = min_t(u64, em->len - offset, max_len);
5672         } else {
5673                 *length = em->len - offset;
5674         }
5675
5676         /* This is for when we're called from btrfs_merge_bio_hook() and all
5677            it cares about is the length */
5678         if (!bbio_ret)
5679                 goto out;
5680
5681         btrfs_dev_replace_lock(dev_replace, 0);
5682         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5683         if (!dev_replace_is_ongoing)
5684                 btrfs_dev_replace_unlock(dev_replace, 0);
5685         else
5686                 btrfs_dev_replace_set_lock_blocking(dev_replace);
5687
5688         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5689             !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5690                 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5691                                                     dev_replace->srcdev->devid,
5692                                                     &mirror_num,
5693                                             &physical_to_patch_in_first_stripe);
5694                 if (ret)
5695                         goto out;
5696                 else
5697                         patch_the_first_stripe_for_dev_replace = 1;
5698         } else if (mirror_num > map->num_stripes) {
5699                 mirror_num = 0;
5700         }
5701
5702         num_stripes = 1;
5703         stripe_index = 0;
5704         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5705                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5706                                 &stripe_index);
5707                 if (!need_full_stripe(op))
5708                         mirror_num = 1;
5709         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5710                 if (need_full_stripe(op))
5711                         num_stripes = map->num_stripes;
5712                 else if (mirror_num)
5713                         stripe_index = mirror_num - 1;
5714                 else {
5715                         stripe_index = find_live_mirror(fs_info, map, 0,
5716                                             map->num_stripes,
5717                                             current->pid % map->num_stripes,
5718                                             dev_replace_is_ongoing);
5719                         mirror_num = stripe_index + 1;
5720                 }
5721
5722         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5723                 if (need_full_stripe(op)) {
5724                         num_stripes = map->num_stripes;
5725                 } else if (mirror_num) {
5726                         stripe_index = mirror_num - 1;
5727                 } else {
5728                         mirror_num = 1;
5729                 }
5730
5731         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5732                 u32 factor = map->num_stripes / map->sub_stripes;
5733
5734                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5735                 stripe_index *= map->sub_stripes;
5736
5737                 if (need_full_stripe(op))
5738                         num_stripes = map->sub_stripes;
5739                 else if (mirror_num)
5740                         stripe_index += mirror_num - 1;
5741                 else {
5742                         int old_stripe_index = stripe_index;
5743                         stripe_index = find_live_mirror(fs_info, map,
5744                                               stripe_index,
5745                                               map->sub_stripes, stripe_index +
5746                                               current->pid % map->sub_stripes,
5747                                               dev_replace_is_ongoing);
5748                         mirror_num = stripe_index - old_stripe_index + 1;
5749                 }
5750
5751         } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5752                 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
5753                         /* push stripe_nr back to the start of the full stripe */
5754                         stripe_nr = div64_u64(raid56_full_stripe_start,
5755                                         stripe_len * nr_data_stripes(map));
5756
5757                         /* RAID[56] write or recovery. Return all stripes */
5758                         num_stripes = map->num_stripes;
5759                         max_errors = nr_parity_stripes(map);
5760
5761                         *length = map->stripe_len;
5762                         stripe_index = 0;
5763                         stripe_offset = 0;
5764                 } else {
5765                         /*
5766                          * Mirror #0 or #1 means the original data block.
5767                          * Mirror #2 is RAID5 parity block.
5768                          * Mirror #3 is RAID6 Q block.
5769                          */
5770                         stripe_nr = div_u64_rem(stripe_nr,
5771                                         nr_data_stripes(map), &stripe_index);
5772                         if (mirror_num > 1)
5773                                 stripe_index = nr_data_stripes(map) +
5774                                                 mirror_num - 2;
5775
5776                         /* We distribute the parity blocks across stripes */
5777                         div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5778                                         &stripe_index);
5779                         if (!need_full_stripe(op) && mirror_num <= 1)
5780                                 mirror_num = 1;
5781                 }
5782         } else {
5783                 /*
5784                  * after this, stripe_nr is the number of stripes on this
5785                  * device we have to walk to find the data, and stripe_index is
5786                  * the number of our device in the stripe array
5787                  */
5788                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5789                                 &stripe_index);
5790                 mirror_num = stripe_index + 1;
5791         }
5792         if (stripe_index >= map->num_stripes) {
5793                 btrfs_crit(fs_info,
5794                            "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5795                            stripe_index, map->num_stripes);
5796                 ret = -EINVAL;
5797                 goto out;
5798         }
5799
5800         num_alloc_stripes = num_stripes;
5801         if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5802                 if (op == BTRFS_MAP_WRITE)
5803                         num_alloc_stripes <<= 1;
5804                 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5805                         num_alloc_stripes++;
5806                 tgtdev_indexes = num_stripes;
5807         }
5808
5809         bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5810         if (!bbio) {
5811                 ret = -ENOMEM;
5812                 goto out;
5813         }
5814         if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5815                 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5816
5817         /* build raid_map */
5818         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5819             (need_full_stripe(op) || mirror_num > 1)) {
5820                 u64 tmp;
5821                 unsigned rot;
5822
5823                 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5824                                  sizeof(struct btrfs_bio_stripe) *
5825                                  num_alloc_stripes +
5826                                  sizeof(int) * tgtdev_indexes);
5827
5828                 /* Work out the disk rotation on this stripe-set */
5829                 div_u64_rem(stripe_nr, num_stripes, &rot);
5830
5831                 /* Fill in the logical address of each stripe */
5832                 tmp = stripe_nr * nr_data_stripes(map);
5833                 for (i = 0; i < nr_data_stripes(map); i++)
5834                         bbio->raid_map[(i+rot) % num_stripes] =
5835                                 em->start + (tmp + i) * map->stripe_len;
5836
5837                 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5838                 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5839                         bbio->raid_map[(i+rot+1) % num_stripes] =
5840                                 RAID6_Q_STRIPE;
5841         }
5842
5843
5844         for (i = 0; i < num_stripes; i++) {
5845                 bbio->stripes[i].physical =
5846                         map->stripes[stripe_index].physical +
5847                         stripe_offset +
5848                         stripe_nr * map->stripe_len;
5849                 bbio->stripes[i].dev =
5850                         map->stripes[stripe_index].dev;
5851                 stripe_index++;
5852         }
5853
5854         if (need_full_stripe(op))
5855                 max_errors = btrfs_chunk_max_errors(map);
5856
5857         if (bbio->raid_map)
5858                 sort_parity_stripes(bbio, num_stripes);
5859
5860         if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
5861             need_full_stripe(op)) {
5862                 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
5863                                           &max_errors);
5864         }
5865
5866         *bbio_ret = bbio;
5867         bbio->map_type = map->type;
5868         bbio->num_stripes = num_stripes;
5869         bbio->max_errors = max_errors;
5870         bbio->mirror_num = mirror_num;
5871
5872         /*
5873          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5874          * mirror_num == num_stripes + 1 && dev_replace target drive is
5875          * available as a mirror
5876          */
5877         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5878                 WARN_ON(num_stripes > 1);
5879                 bbio->stripes[0].dev = dev_replace->tgtdev;
5880                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5881                 bbio->mirror_num = map->num_stripes + 1;
5882         }
5883 out:
5884         if (dev_replace_is_ongoing) {
5885                 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5886                 btrfs_dev_replace_unlock(dev_replace, 0);
5887         }
5888         free_extent_map(em);
5889         return ret;
5890 }
5891
5892 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5893                       u64 logical, u64 *length,
5894                       struct btrfs_bio **bbio_ret, int mirror_num)
5895 {
5896         return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5897                                  mirror_num, 0);
5898 }
5899
5900 /* For Scrub/replace */
5901 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5902                      u64 logical, u64 *length,
5903                      struct btrfs_bio **bbio_ret)
5904 {
5905         return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
5906 }
5907
5908 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
5909                      u64 chunk_start, u64 physical, u64 devid,
5910                      u64 **logical, int *naddrs, int *stripe_len)
5911 {
5912         struct extent_map *em;
5913         struct map_lookup *map;
5914         u64 *buf;
5915         u64 bytenr;
5916         u64 length;
5917         u64 stripe_nr;
5918         u64 rmap_len;
5919         int i, j, nr = 0;
5920
5921         em = get_chunk_map(fs_info, chunk_start, 1);
5922         if (IS_ERR(em))
5923                 return -EIO;
5924
5925         map = em->map_lookup;
5926         length = em->len;
5927         rmap_len = map->stripe_len;
5928
5929         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5930                 length = div_u64(length, map->num_stripes / map->sub_stripes);
5931         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5932                 length = div_u64(length, map->num_stripes);
5933         else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5934                 length = div_u64(length, nr_data_stripes(map));
5935                 rmap_len = map->stripe_len * nr_data_stripes(map);
5936         }
5937
5938         buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5939         BUG_ON(!buf); /* -ENOMEM */
5940
5941         for (i = 0; i < map->num_stripes; i++) {
5942                 if (devid && map->stripes[i].dev->devid != devid)
5943                         continue;
5944                 if (map->stripes[i].physical > physical ||
5945                     map->stripes[i].physical + length <= physical)
5946                         continue;
5947
5948                 stripe_nr = physical - map->stripes[i].physical;
5949                 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
5950
5951                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5952                         stripe_nr = stripe_nr * map->num_stripes + i;
5953                         stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5954                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5955                         stripe_nr = stripe_nr * map->num_stripes + i;
5956                 } /* else if RAID[56], multiply by nr_data_stripes().
5957                    * Alternatively, just use rmap_len below instead of
5958                    * map->stripe_len */
5959
5960                 bytenr = chunk_start + stripe_nr * rmap_len;
5961                 WARN_ON(nr >= map->num_stripes);
5962                 for (j = 0; j < nr; j++) {
5963                         if (buf[j] == bytenr)
5964                                 break;
5965                 }
5966                 if (j == nr) {
5967                         WARN_ON(nr >= map->num_stripes);
5968                         buf[nr++] = bytenr;
5969                 }
5970         }
5971
5972         *logical = buf;
5973         *naddrs = nr;
5974         *stripe_len = rmap_len;
5975
5976         free_extent_map(em);
5977         return 0;
5978 }
5979
5980 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5981 {
5982         bio->bi_private = bbio->private;
5983         bio->bi_end_io = bbio->end_io;
5984         bio_endio(bio);
5985
5986         btrfs_put_bbio(bbio);
5987 }
5988
5989 static void btrfs_end_bio(struct bio *bio)
5990 {
5991         struct btrfs_bio *bbio = bio->bi_private;
5992         int is_orig_bio = 0;
5993
5994         if (bio->bi_status) {
5995                 atomic_inc(&bbio->error);
5996                 if (bio->bi_status == BLK_STS_IOERR ||
5997                     bio->bi_status == BLK_STS_TARGET) {
5998                         unsigned int stripe_index =
5999                                 btrfs_io_bio(bio)->stripe_index;
6000                         struct btrfs_device *dev;
6001
6002                         BUG_ON(stripe_index >= bbio->num_stripes);
6003                         dev = bbio->stripes[stripe_index].dev;
6004                         if (dev->bdev) {
6005                                 if (bio_op(bio) == REQ_OP_WRITE)
6006                                         btrfs_dev_stat_inc(dev,
6007                                                 BTRFS_DEV_STAT_WRITE_ERRS);
6008                                 else
6009                                         btrfs_dev_stat_inc(dev,
6010                                                 BTRFS_DEV_STAT_READ_ERRS);
6011                                 if (bio->bi_opf & REQ_PREFLUSH)
6012                                         btrfs_dev_stat_inc(dev,
6013                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
6014                                 btrfs_dev_stat_print_on_error(dev);
6015                         }
6016                 }
6017         }
6018
6019         if (bio == bbio->orig_bio)
6020                 is_orig_bio = 1;
6021
6022         btrfs_bio_counter_dec(bbio->fs_info);
6023
6024         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6025                 if (!is_orig_bio) {
6026                         bio_put(bio);
6027                         bio = bbio->orig_bio;
6028                 }
6029
6030                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6031                 /* only send an error to the higher layers if it is
6032                  * beyond the tolerance of the btrfs bio
6033                  */
6034                 if (atomic_read(&bbio->error) > bbio->max_errors) {
6035                         bio->bi_status = BLK_STS_IOERR;
6036                 } else {
6037                         /*
6038                          * this bio is actually up to date, we didn't
6039                          * go over the max number of errors
6040                          */
6041                         bio->bi_status = BLK_STS_OK;
6042                 }
6043
6044                 btrfs_end_bbio(bbio, bio);
6045         } else if (!is_orig_bio) {
6046                 bio_put(bio);
6047         }
6048 }
6049
6050 /*
6051  * see run_scheduled_bios for a description of why bios are collected for
6052  * async submit.
6053  *
6054  * This will add one bio to the pending list for a device and make sure
6055  * the work struct is scheduled.
6056  */
6057 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6058                                         struct bio *bio)
6059 {
6060         struct btrfs_fs_info *fs_info = device->fs_info;
6061         int should_queue = 1;
6062         struct btrfs_pending_bios *pending_bios;
6063
6064         if (device->missing || !device->bdev) {
6065                 bio_io_error(bio);
6066                 return;
6067         }
6068
6069         /* don't bother with additional async steps for reads, right now */
6070         if (bio_op(bio) == REQ_OP_READ) {
6071                 bio_get(bio);
6072                 btrfsic_submit_bio(bio);
6073                 bio_put(bio);
6074                 return;
6075         }
6076
6077         WARN_ON(bio->bi_next);
6078         bio->bi_next = NULL;
6079
6080         spin_lock(&device->io_lock);
6081         if (op_is_sync(bio->bi_opf))
6082                 pending_bios = &device->pending_sync_bios;
6083         else
6084                 pending_bios = &device->pending_bios;
6085
6086         if (pending_bios->tail)
6087                 pending_bios->tail->bi_next = bio;
6088
6089         pending_bios->tail = bio;
6090         if (!pending_bios->head)
6091                 pending_bios->head = bio;
6092         if (device->running_pending)
6093                 should_queue = 0;
6094
6095         spin_unlock(&device->io_lock);
6096
6097         if (should_queue)
6098                 btrfs_queue_work(fs_info->submit_workers, &device->work);
6099 }
6100
6101 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6102                               u64 physical, int dev_nr, int async)
6103 {
6104         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6105         struct btrfs_fs_info *fs_info = bbio->fs_info;
6106
6107         bio->bi_private = bbio;
6108         btrfs_io_bio(bio)->stripe_index = dev_nr;
6109         bio->bi_end_io = btrfs_end_bio;
6110         bio->bi_iter.bi_sector = physical >> 9;
6111 #ifdef DEBUG
6112         {
6113                 struct rcu_string *name;
6114
6115                 rcu_read_lock();
6116                 name = rcu_dereference(dev->name);
6117                 btrfs_debug(fs_info,
6118                         "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6119                         bio_op(bio), bio->bi_opf,
6120                         (u64)bio->bi_iter.bi_sector,
6121                         (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6122                         bio->bi_iter.bi_size);
6123                 rcu_read_unlock();
6124         }
6125 #endif
6126         bio_set_dev(bio, dev->bdev);
6127
6128         btrfs_bio_counter_inc_noblocked(fs_info);
6129
6130         if (async)
6131                 btrfs_schedule_bio(dev, bio);
6132         else
6133                 btrfsic_submit_bio(bio);
6134 }
6135
6136 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6137 {
6138         atomic_inc(&bbio->error);
6139         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6140                 /* Should be the original bio. */
6141                 WARN_ON(bio != bbio->orig_bio);
6142
6143                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6144                 bio->bi_iter.bi_sector = logical >> 9;
6145                 if (atomic_read(&bbio->error) > bbio->max_errors)
6146                         bio->bi_status = BLK_STS_IOERR;
6147                 else
6148                         bio->bi_status = BLK_STS_OK;
6149                 btrfs_end_bbio(bbio, bio);
6150         }
6151 }
6152
6153 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6154                            int mirror_num, int async_submit)
6155 {
6156         struct btrfs_device *dev;
6157         struct bio *first_bio = bio;
6158         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6159         u64 length = 0;
6160         u64 map_length;
6161         int ret;
6162         int dev_nr;
6163         int total_devs;
6164         struct btrfs_bio *bbio = NULL;
6165
6166         length = bio->bi_iter.bi_size;
6167         map_length = length;
6168
6169         btrfs_bio_counter_inc_blocked(fs_info);
6170         ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6171                                 &map_length, &bbio, mirror_num, 1);
6172         if (ret) {
6173                 btrfs_bio_counter_dec(fs_info);
6174                 return errno_to_blk_status(ret);
6175         }
6176
6177         total_devs = bbio->num_stripes;
6178         bbio->orig_bio = first_bio;
6179         bbio->private = first_bio->bi_private;
6180         bbio->end_io = first_bio->bi_end_io;
6181         bbio->fs_info = fs_info;
6182         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6183
6184         if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6185             ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6186                 /* In this case, map_length has been set to the length of
6187                    a single stripe; not the whole write */
6188                 if (bio_op(bio) == REQ_OP_WRITE) {
6189                         ret = raid56_parity_write(fs_info, bio, bbio,
6190                                                   map_length);
6191                 } else {
6192                         ret = raid56_parity_recover(fs_info, bio, bbio,
6193                                                     map_length, mirror_num, 1);
6194                 }
6195
6196                 btrfs_bio_counter_dec(fs_info);
6197                 return errno_to_blk_status(ret);
6198         }
6199
6200         if (map_length < length) {
6201                 btrfs_crit(fs_info,
6202                            "mapping failed logical %llu bio len %llu len %llu",
6203                            logical, length, map_length);
6204                 BUG();
6205         }
6206
6207         for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6208                 dev = bbio->stripes[dev_nr].dev;
6209                 if (!dev || !dev->bdev ||
6210                     (bio_op(first_bio) == REQ_OP_WRITE && !dev->writeable)) {
6211                         bbio_error(bbio, first_bio, logical);
6212                         continue;
6213                 }
6214
6215                 if (dev_nr < total_devs - 1)
6216                         bio = btrfs_bio_clone(first_bio);
6217                 else
6218                         bio = first_bio;
6219
6220                 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6221                                   dev_nr, async_submit);
6222         }
6223         btrfs_bio_counter_dec(fs_info);
6224         return BLK_STS_OK;
6225 }
6226
6227 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6228                                        u8 *uuid, u8 *fsid)
6229 {
6230         struct btrfs_device *device;
6231         struct btrfs_fs_devices *cur_devices;
6232
6233         cur_devices = fs_info->fs_devices;
6234         while (cur_devices) {
6235                 if (!fsid ||
6236                     !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) {
6237                         device = find_device(cur_devices, devid, uuid);
6238                         if (device)
6239                                 return device;
6240                 }
6241                 cur_devices = cur_devices->seed;
6242         }
6243         return NULL;
6244 }
6245
6246 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6247                                             u64 devid, u8 *dev_uuid)
6248 {
6249         struct btrfs_device *device;
6250
6251         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6252         if (IS_ERR(device))
6253                 return device;
6254
6255         list_add(&device->dev_list, &fs_devices->devices);
6256         device->fs_devices = fs_devices;
6257         fs_devices->num_devices++;
6258
6259         device->missing = 1;
6260         fs_devices->missing_devices++;
6261
6262         return device;
6263 }
6264
6265 /**
6266  * btrfs_alloc_device - allocate struct btrfs_device
6267  * @fs_info:    used only for generating a new devid, can be NULL if
6268  *              devid is provided (i.e. @devid != NULL).
6269  * @devid:      a pointer to devid for this device.  If NULL a new devid
6270  *              is generated.
6271  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
6272  *              is generated.
6273  *
6274  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6275  * on error.  Returned struct is not linked onto any lists and can be
6276  * destroyed with kfree() right away.
6277  */
6278 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6279                                         const u64 *devid,
6280                                         const u8 *uuid)
6281 {
6282         struct btrfs_device *dev;
6283         u64 tmp;
6284
6285         if (WARN_ON(!devid && !fs_info))
6286                 return ERR_PTR(-EINVAL);
6287
6288         dev = __alloc_device();
6289         if (IS_ERR(dev))
6290                 return dev;
6291
6292         if (devid)
6293                 tmp = *devid;
6294         else {
6295                 int ret;
6296
6297                 ret = find_next_devid(fs_info, &tmp);
6298                 if (ret) {
6299                         bio_put(dev->flush_bio);
6300                         kfree(dev);
6301                         return ERR_PTR(ret);
6302                 }
6303         }
6304         dev->devid = tmp;
6305
6306         if (uuid)
6307                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6308         else
6309                 generate_random_uuid(dev->uuid);
6310
6311         btrfs_init_work(&dev->work, btrfs_submit_helper,
6312                         pending_bios_fn, NULL, NULL);
6313
6314         return dev;
6315 }
6316
6317 /* Return -EIO if any error, otherwise return 0. */
6318 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6319                                    struct extent_buffer *leaf,
6320                                    struct btrfs_chunk *chunk, u64 logical)
6321 {
6322         u64 length;
6323         u64 stripe_len;
6324         u16 num_stripes;
6325         u16 sub_stripes;
6326         u64 type;
6327
6328         length = btrfs_chunk_length(leaf, chunk);
6329         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6330         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6331         sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6332         type = btrfs_chunk_type(leaf, chunk);
6333
6334         if (!num_stripes) {
6335                 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6336                           num_stripes);
6337                 return -EIO;
6338         }
6339         if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6340                 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6341                 return -EIO;
6342         }
6343         if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6344                 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6345                           btrfs_chunk_sector_size(leaf, chunk));
6346                 return -EIO;
6347         }
6348         if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6349                 btrfs_err(fs_info, "invalid chunk length %llu", length);
6350                 return -EIO;
6351         }
6352         if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6353                 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6354                           stripe_len);
6355                 return -EIO;
6356         }
6357         if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6358             type) {
6359                 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6360                           ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6361                             BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6362                           btrfs_chunk_type(leaf, chunk));
6363                 return -EIO;
6364         }
6365         if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6366             (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6367             (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6368             (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6369             (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6370             ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6371              num_stripes != 1)) {
6372                 btrfs_err(fs_info,
6373                         "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6374                         num_stripes, sub_stripes,
6375                         type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6376                 return -EIO;
6377         }
6378
6379         return 0;
6380 }
6381
6382 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6383                                         u64 devid, u8 *uuid, bool error)
6384 {
6385         if (error)
6386                 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6387                               devid, uuid);
6388         else
6389                 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6390                               devid, uuid);
6391 }
6392
6393 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6394                           struct extent_buffer *leaf,
6395                           struct btrfs_chunk *chunk)
6396 {
6397         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6398         struct map_lookup *map;
6399         struct extent_map *em;
6400         u64 logical;
6401         u64 length;
6402         u64 devid;
6403         u8 uuid[BTRFS_UUID_SIZE];
6404         int num_stripes;
6405         int ret;
6406         int i;
6407
6408         logical = key->offset;
6409         length = btrfs_chunk_length(leaf, chunk);
6410         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6411
6412         ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6413         if (ret)
6414                 return ret;
6415
6416         read_lock(&map_tree->map_tree.lock);
6417         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6418         read_unlock(&map_tree->map_tree.lock);
6419
6420         /* already mapped? */
6421         if (em && em->start <= logical && em->start + em->len > logical) {
6422                 free_extent_map(em);
6423                 return 0;
6424         } else if (em) {
6425                 free_extent_map(em);
6426         }
6427
6428         em = alloc_extent_map();
6429         if (!em)
6430                 return -ENOMEM;
6431         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6432         if (!map) {
6433                 free_extent_map(em);
6434                 return -ENOMEM;
6435         }
6436
6437         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6438         em->map_lookup = map;
6439         em->start = logical;
6440         em->len = length;
6441         em->orig_start = 0;
6442         em->block_start = 0;
6443         em->block_len = em->len;
6444
6445         map->num_stripes = num_stripes;
6446         map->io_width = btrfs_chunk_io_width(leaf, chunk);
6447         map->io_align = btrfs_chunk_io_align(leaf, chunk);
6448         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6449         map->type = btrfs_chunk_type(leaf, chunk);
6450         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6451         for (i = 0; i < num_stripes; i++) {
6452                 map->stripes[i].physical =
6453                         btrfs_stripe_offset_nr(leaf, chunk, i);
6454                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6455                 read_extent_buffer(leaf, uuid, (unsigned long)
6456                                    btrfs_stripe_dev_uuid_nr(chunk, i),
6457                                    BTRFS_UUID_SIZE);
6458                 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6459                                                         uuid, NULL);
6460                 if (!map->stripes[i].dev &&
6461                     !btrfs_test_opt(fs_info, DEGRADED)) {
6462                         free_extent_map(em);
6463                         btrfs_report_missing_device(fs_info, devid, uuid, true);
6464                         return -ENOENT;
6465                 }
6466                 if (!map->stripes[i].dev) {
6467                         map->stripes[i].dev =
6468                                 add_missing_dev(fs_info->fs_devices, devid,
6469                                                 uuid);
6470                         if (IS_ERR(map->stripes[i].dev)) {
6471                                 free_extent_map(em);
6472                                 btrfs_err(fs_info,
6473                                         "failed to init missing dev %llu: %ld",
6474                                         devid, PTR_ERR(map->stripes[i].dev));
6475                                 return PTR_ERR(map->stripes[i].dev);
6476                         }
6477                         btrfs_report_missing_device(fs_info, devid, uuid, false);
6478                 }
6479                 map->stripes[i].dev->in_fs_metadata = 1;
6480         }
6481
6482         write_lock(&map_tree->map_tree.lock);
6483         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6484         write_unlock(&map_tree->map_tree.lock);
6485         BUG_ON(ret); /* Tree corruption */
6486         free_extent_map(em);
6487
6488         return 0;
6489 }
6490
6491 static void fill_device_from_item(struct extent_buffer *leaf,
6492                                  struct btrfs_dev_item *dev_item,
6493                                  struct btrfs_device *device)
6494 {
6495         unsigned long ptr;
6496
6497         device->devid = btrfs_device_id(leaf, dev_item);
6498         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6499         device->total_bytes = device->disk_total_bytes;
6500         device->commit_total_bytes = device->disk_total_bytes;
6501         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6502         device->commit_bytes_used = device->bytes_used;
6503         device->type = btrfs_device_type(leaf, dev_item);
6504         device->io_align = btrfs_device_io_align(leaf, dev_item);
6505         device->io_width = btrfs_device_io_width(leaf, dev_item);
6506         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6507         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6508         device->is_tgtdev_for_dev_replace = 0;
6509
6510         ptr = btrfs_device_uuid(dev_item);
6511         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6512 }
6513
6514 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6515                                                   u8 *fsid)
6516 {
6517         struct btrfs_fs_devices *fs_devices;
6518         int ret;
6519
6520         BUG_ON(!mutex_is_locked(&uuid_mutex));
6521         ASSERT(fsid);
6522
6523         fs_devices = fs_info->fs_devices->seed;
6524         while (fs_devices) {
6525                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6526                         return fs_devices;
6527
6528                 fs_devices = fs_devices->seed;
6529         }
6530
6531         fs_devices = find_fsid(fsid);
6532         if (!fs_devices) {
6533                 if (!btrfs_test_opt(fs_info, DEGRADED))
6534                         return ERR_PTR(-ENOENT);
6535
6536                 fs_devices = alloc_fs_devices(fsid);
6537                 if (IS_ERR(fs_devices))
6538                         return fs_devices;
6539
6540                 fs_devices->seeding = 1;
6541                 fs_devices->opened = 1;
6542                 return fs_devices;
6543         }
6544
6545         fs_devices = clone_fs_devices(fs_devices);
6546         if (IS_ERR(fs_devices))
6547                 return fs_devices;
6548
6549         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6550                                    fs_info->bdev_holder);
6551         if (ret) {
6552                 free_fs_devices(fs_devices);
6553                 fs_devices = ERR_PTR(ret);
6554                 goto out;
6555         }
6556
6557         if (!fs_devices->seeding) {
6558                 __btrfs_close_devices(fs_devices);
6559                 free_fs_devices(fs_devices);
6560                 fs_devices = ERR_PTR(-EINVAL);
6561                 goto out;
6562         }
6563
6564         fs_devices->seed = fs_info->fs_devices->seed;
6565         fs_info->fs_devices->seed = fs_devices;
6566 out:
6567         return fs_devices;
6568 }
6569
6570 static int read_one_dev(struct btrfs_fs_info *fs_info,
6571                         struct extent_buffer *leaf,
6572                         struct btrfs_dev_item *dev_item)
6573 {
6574         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6575         struct btrfs_device *device;
6576         u64 devid;
6577         int ret;
6578         u8 fs_uuid[BTRFS_FSID_SIZE];
6579         u8 dev_uuid[BTRFS_UUID_SIZE];
6580
6581         devid = btrfs_device_id(leaf, dev_item);
6582         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6583                            BTRFS_UUID_SIZE);
6584         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6585                            BTRFS_FSID_SIZE);
6586
6587         if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) {
6588                 fs_devices = open_seed_devices(fs_info, fs_uuid);
6589                 if (IS_ERR(fs_devices))
6590                         return PTR_ERR(fs_devices);
6591         }
6592
6593         device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6594         if (!device) {
6595                 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6596                         btrfs_report_missing_device(fs_info, devid,
6597                                                         dev_uuid, true);
6598                         return -ENOENT;
6599                 }
6600
6601                 device = add_missing_dev(fs_devices, devid, dev_uuid);
6602                 if (IS_ERR(device)) {
6603                         btrfs_err(fs_info,
6604                                 "failed to add missing dev %llu: %ld",
6605                                 devid, PTR_ERR(device));
6606                         return PTR_ERR(device);
6607                 }
6608                 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6609         } else {
6610                 if (!device->bdev) {
6611                         if (!btrfs_test_opt(fs_info, DEGRADED)) {
6612                                 btrfs_report_missing_device(fs_info,
6613                                                 devid, dev_uuid, true);
6614                                 return -ENOENT;
6615                         }
6616                         btrfs_report_missing_device(fs_info, devid,
6617                                                         dev_uuid, false);
6618                 }
6619
6620                 if(!device->bdev && !device->missing) {
6621                         /*
6622                          * this happens when a device that was properly setup
6623                          * in the device info lists suddenly goes bad.
6624                          * device->bdev is NULL, and so we have to set
6625                          * device->missing to one here
6626                          */
6627                         device->fs_devices->missing_devices++;
6628                         device->missing = 1;
6629                 }
6630
6631                 /* Move the device to its own fs_devices */
6632                 if (device->fs_devices != fs_devices) {
6633                         ASSERT(device->missing);
6634
6635                         list_move(&device->dev_list, &fs_devices->devices);
6636                         device->fs_devices->num_devices--;
6637                         fs_devices->num_devices++;
6638
6639                         device->fs_devices->missing_devices--;
6640                         fs_devices->missing_devices++;
6641
6642                         device->fs_devices = fs_devices;
6643                 }
6644         }
6645
6646         if (device->fs_devices != fs_info->fs_devices) {
6647                 BUG_ON(device->writeable);
6648                 if (device->generation !=
6649                     btrfs_device_generation(leaf, dev_item))
6650                         return -EINVAL;
6651         }
6652
6653         fill_device_from_item(leaf, dev_item, device);
6654         device->in_fs_metadata = 1;
6655         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6656                 device->fs_devices->total_rw_bytes += device->total_bytes;
6657                 atomic64_add(device->total_bytes - device->bytes_used,
6658                                 &fs_info->free_chunk_space);
6659         }
6660         ret = 0;
6661         return ret;
6662 }
6663
6664 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6665 {
6666         struct btrfs_root *root = fs_info->tree_root;
6667         struct btrfs_super_block *super_copy = fs_info->super_copy;
6668         struct extent_buffer *sb;
6669         struct btrfs_disk_key *disk_key;
6670         struct btrfs_chunk *chunk;
6671         u8 *array_ptr;
6672         unsigned long sb_array_offset;
6673         int ret = 0;
6674         u32 num_stripes;
6675         u32 array_size;
6676         u32 len = 0;
6677         u32 cur_offset;
6678         u64 type;
6679         struct btrfs_key key;
6680
6681         ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6682         /*
6683          * This will create extent buffer of nodesize, superblock size is
6684          * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6685          * overallocate but we can keep it as-is, only the first page is used.
6686          */
6687         sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6688         if (IS_ERR(sb))
6689                 return PTR_ERR(sb);
6690         set_extent_buffer_uptodate(sb);
6691         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6692         /*
6693          * The sb extent buffer is artificial and just used to read the system array.
6694          * set_extent_buffer_uptodate() call does not properly mark all it's
6695          * pages up-to-date when the page is larger: extent does not cover the
6696          * whole page and consequently check_page_uptodate does not find all
6697          * the page's extents up-to-date (the hole beyond sb),
6698          * write_extent_buffer then triggers a WARN_ON.
6699          *
6700          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6701          * but sb spans only this function. Add an explicit SetPageUptodate call
6702          * to silence the warning eg. on PowerPC 64.
6703          */
6704         if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6705                 SetPageUptodate(sb->pages[0]);
6706
6707         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6708         array_size = btrfs_super_sys_array_size(super_copy);
6709
6710         array_ptr = super_copy->sys_chunk_array;
6711         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6712         cur_offset = 0;
6713
6714         while (cur_offset < array_size) {
6715                 disk_key = (struct btrfs_disk_key *)array_ptr;
6716                 len = sizeof(*disk_key);
6717                 if (cur_offset + len > array_size)
6718                         goto out_short_read;
6719
6720                 btrfs_disk_key_to_cpu(&key, disk_key);
6721
6722                 array_ptr += len;
6723                 sb_array_offset += len;
6724                 cur_offset += len;
6725
6726                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6727                         chunk = (struct btrfs_chunk *)sb_array_offset;
6728                         /*
6729                          * At least one btrfs_chunk with one stripe must be
6730                          * present, exact stripe count check comes afterwards
6731                          */
6732                         len = btrfs_chunk_item_size(1);
6733                         if (cur_offset + len > array_size)
6734                                 goto out_short_read;
6735
6736                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6737                         if (!num_stripes) {
6738                                 btrfs_err(fs_info,
6739                                         "invalid number of stripes %u in sys_array at offset %u",
6740                                         num_stripes, cur_offset);
6741                                 ret = -EIO;
6742                                 break;
6743                         }
6744
6745                         type = btrfs_chunk_type(sb, chunk);
6746                         if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6747                                 btrfs_err(fs_info,
6748                             "invalid chunk type %llu in sys_array at offset %u",
6749                                         type, cur_offset);
6750                                 ret = -EIO;
6751                                 break;
6752                         }
6753
6754                         len = btrfs_chunk_item_size(num_stripes);
6755                         if (cur_offset + len > array_size)
6756                                 goto out_short_read;
6757
6758                         ret = read_one_chunk(fs_info, &key, sb, chunk);
6759                         if (ret)
6760                                 break;
6761                 } else {
6762                         btrfs_err(fs_info,
6763                             "unexpected item type %u in sys_array at offset %u",
6764                                   (u32)key.type, cur_offset);
6765                         ret = -EIO;
6766                         break;
6767                 }
6768                 array_ptr += len;
6769                 sb_array_offset += len;
6770                 cur_offset += len;
6771         }
6772         clear_extent_buffer_uptodate(sb);
6773         free_extent_buffer_stale(sb);
6774         return ret;
6775
6776 out_short_read:
6777         btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6778                         len, cur_offset);
6779         clear_extent_buffer_uptodate(sb);
6780         free_extent_buffer_stale(sb);
6781         return -EIO;
6782 }
6783
6784 /*
6785  * Check if all chunks in the fs are OK for read-write degraded mount
6786  *
6787  * Return true if all chunks meet the minimal RW mount requirements.
6788  * Return false if any chunk doesn't meet the minimal RW mount requirements.
6789  */
6790 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info)
6791 {
6792         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6793         struct extent_map *em;
6794         u64 next_start = 0;
6795         bool ret = true;
6796
6797         read_lock(&map_tree->map_tree.lock);
6798         em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1);
6799         read_unlock(&map_tree->map_tree.lock);
6800         /* No chunk at all? Return false anyway */
6801         if (!em) {
6802                 ret = false;
6803                 goto out;
6804         }
6805         while (em) {
6806                 struct map_lookup *map;
6807                 int missing = 0;
6808                 int max_tolerated;
6809                 int i;
6810
6811                 map = em->map_lookup;
6812                 max_tolerated =
6813                         btrfs_get_num_tolerated_disk_barrier_failures(
6814                                         map->type);
6815                 for (i = 0; i < map->num_stripes; i++) {
6816                         struct btrfs_device *dev = map->stripes[i].dev;
6817
6818                         if (!dev || !dev->bdev || dev->missing ||
6819                             dev->last_flush_error)
6820                                 missing++;
6821                 }
6822                 if (missing > max_tolerated) {
6823                         btrfs_warn(fs_info,
6824         "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6825                                    em->start, missing, max_tolerated);
6826                         free_extent_map(em);
6827                         ret = false;
6828                         goto out;
6829                 }
6830                 next_start = extent_map_end(em);
6831                 free_extent_map(em);
6832
6833                 read_lock(&map_tree->map_tree.lock);
6834                 em = lookup_extent_mapping(&map_tree->map_tree, next_start,
6835                                            (u64)(-1) - next_start);
6836                 read_unlock(&map_tree->map_tree.lock);
6837         }
6838 out:
6839         return ret;
6840 }
6841
6842 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6843 {
6844         struct btrfs_root *root = fs_info->chunk_root;
6845         struct btrfs_path *path;
6846         struct extent_buffer *leaf;
6847         struct btrfs_key key;
6848         struct btrfs_key found_key;
6849         int ret;
6850         int slot;
6851         u64 total_dev = 0;
6852
6853         path = btrfs_alloc_path();
6854         if (!path)
6855                 return -ENOMEM;
6856
6857         mutex_lock(&uuid_mutex);
6858         mutex_lock(&fs_info->chunk_mutex);
6859
6860         /*
6861          * Read all device items, and then all the chunk items. All
6862          * device items are found before any chunk item (their object id
6863          * is smaller than the lowest possible object id for a chunk
6864          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6865          */
6866         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6867         key.offset = 0;
6868         key.type = 0;
6869         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6870         if (ret < 0)
6871                 goto error;
6872         while (1) {
6873                 leaf = path->nodes[0];
6874                 slot = path->slots[0];
6875                 if (slot >= btrfs_header_nritems(leaf)) {
6876                         ret = btrfs_next_leaf(root, path);
6877                         if (ret == 0)
6878                                 continue;
6879                         if (ret < 0)
6880                                 goto error;
6881                         break;
6882                 }
6883                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6884                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6885                         struct btrfs_dev_item *dev_item;
6886                         dev_item = btrfs_item_ptr(leaf, slot,
6887                                                   struct btrfs_dev_item);
6888                         ret = read_one_dev(fs_info, leaf, dev_item);
6889                         if (ret)
6890                                 goto error;
6891                         total_dev++;
6892                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6893                         struct btrfs_chunk *chunk;
6894                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6895                         ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
6896                         if (ret)
6897                                 goto error;
6898                 }
6899                 path->slots[0]++;
6900         }
6901
6902         /*
6903          * After loading chunk tree, we've got all device information,
6904          * do another round of validation checks.
6905          */
6906         if (total_dev != fs_info->fs_devices->total_devices) {
6907                 btrfs_err(fs_info,
6908            "super_num_devices %llu mismatch with num_devices %llu found here",
6909                           btrfs_super_num_devices(fs_info->super_copy),
6910                           total_dev);
6911                 ret = -EINVAL;
6912                 goto error;
6913         }
6914         if (btrfs_super_total_bytes(fs_info->super_copy) <
6915             fs_info->fs_devices->total_rw_bytes) {
6916                 btrfs_err(fs_info,
6917         "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6918                           btrfs_super_total_bytes(fs_info->super_copy),
6919                           fs_info->fs_devices->total_rw_bytes);
6920                 ret = -EINVAL;
6921                 goto error;
6922         }
6923         ret = 0;
6924 error:
6925         mutex_unlock(&fs_info->chunk_mutex);
6926         mutex_unlock(&uuid_mutex);
6927
6928         btrfs_free_path(path);
6929         return ret;
6930 }
6931
6932 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6933 {
6934         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6935         struct btrfs_device *device;
6936
6937         while (fs_devices) {
6938                 mutex_lock(&fs_devices->device_list_mutex);
6939                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6940                         device->fs_info = fs_info;
6941                 mutex_unlock(&fs_devices->device_list_mutex);
6942
6943                 fs_devices = fs_devices->seed;
6944         }
6945 }
6946
6947 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6948 {
6949         int i;
6950
6951         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6952                 btrfs_dev_stat_reset(dev, i);
6953 }
6954
6955 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6956 {
6957         struct btrfs_key key;
6958         struct btrfs_key found_key;
6959         struct btrfs_root *dev_root = fs_info->dev_root;
6960         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6961         struct extent_buffer *eb;
6962         int slot;
6963         int ret = 0;
6964         struct btrfs_device *device;
6965         struct btrfs_path *path = NULL;
6966         int i;
6967
6968         path = btrfs_alloc_path();
6969         if (!path) {
6970                 ret = -ENOMEM;
6971                 goto out;
6972         }
6973
6974         mutex_lock(&fs_devices->device_list_mutex);
6975         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6976                 int item_size;
6977                 struct btrfs_dev_stats_item *ptr;
6978
6979                 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6980                 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6981                 key.offset = device->devid;
6982                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6983                 if (ret) {
6984                         __btrfs_reset_dev_stats(device);
6985                         device->dev_stats_valid = 1;
6986                         btrfs_release_path(path);
6987                         continue;
6988                 }
6989                 slot = path->slots[0];
6990                 eb = path->nodes[0];
6991                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6992                 item_size = btrfs_item_size_nr(eb, slot);
6993
6994                 ptr = btrfs_item_ptr(eb, slot,
6995                                      struct btrfs_dev_stats_item);
6996
6997                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6998                         if (item_size >= (1 + i) * sizeof(__le64))
6999                                 btrfs_dev_stat_set(device, i,
7000                                         btrfs_dev_stats_value(eb, ptr, i));
7001                         else
7002                                 btrfs_dev_stat_reset(device, i);
7003                 }
7004
7005                 device->dev_stats_valid = 1;
7006                 btrfs_dev_stat_print_on_load(device);
7007                 btrfs_release_path(path);
7008         }
7009         mutex_unlock(&fs_devices->device_list_mutex);
7010
7011 out:
7012         btrfs_free_path(path);
7013         return ret < 0 ? ret : 0;
7014 }
7015
7016 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7017                                 struct btrfs_fs_info *fs_info,
7018                                 struct btrfs_device *device)
7019 {
7020         struct btrfs_root *dev_root = fs_info->dev_root;
7021         struct btrfs_path *path;
7022         struct btrfs_key key;
7023         struct extent_buffer *eb;
7024         struct btrfs_dev_stats_item *ptr;
7025         int ret;
7026         int i;
7027
7028         key.objectid = BTRFS_DEV_STATS_OBJECTID;
7029         key.type = BTRFS_PERSISTENT_ITEM_KEY;
7030         key.offset = device->devid;
7031
7032         path = btrfs_alloc_path();
7033         if (!path)
7034                 return -ENOMEM;
7035         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7036         if (ret < 0) {
7037                 btrfs_warn_in_rcu(fs_info,
7038                         "error %d while searching for dev_stats item for device %s",
7039                               ret, rcu_str_deref(device->name));
7040                 goto out;
7041         }
7042
7043         if (ret == 0 &&
7044             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7045                 /* need to delete old one and insert a new one */
7046                 ret = btrfs_del_item(trans, dev_root, path);
7047                 if (ret != 0) {
7048                         btrfs_warn_in_rcu(fs_info,
7049                                 "delete too small dev_stats item for device %s failed %d",
7050                                       rcu_str_deref(device->name), ret);
7051                         goto out;
7052                 }
7053                 ret = 1;
7054         }
7055
7056         if (ret == 1) {
7057                 /* need to insert a new item */
7058                 btrfs_release_path(path);
7059                 ret = btrfs_insert_empty_item(trans, dev_root, path,
7060                                               &key, sizeof(*ptr));
7061                 if (ret < 0) {
7062                         btrfs_warn_in_rcu(fs_info,
7063                                 "insert dev_stats item for device %s failed %d",
7064                                 rcu_str_deref(device->name), ret);
7065                         goto out;
7066                 }
7067         }
7068
7069         eb = path->nodes[0];
7070         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7071         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7072                 btrfs_set_dev_stats_value(eb, ptr, i,
7073                                           btrfs_dev_stat_read(device, i));
7074         btrfs_mark_buffer_dirty(eb);
7075
7076 out:
7077         btrfs_free_path(path);
7078         return ret;
7079 }
7080
7081 /*
7082  * called from commit_transaction. Writes all changed device stats to disk.
7083  */
7084 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7085                         struct btrfs_fs_info *fs_info)
7086 {
7087         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7088         struct btrfs_device *device;
7089         int stats_cnt;
7090         int ret = 0;
7091
7092         mutex_lock(&fs_devices->device_list_mutex);
7093         list_for_each_entry(device, &fs_devices->devices, dev_list) {
7094                 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
7095                         continue;
7096
7097                 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7098                 ret = update_dev_stat_item(trans, fs_info, device);
7099                 if (!ret)
7100                         atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7101         }
7102         mutex_unlock(&fs_devices->device_list_mutex);
7103
7104         return ret;
7105 }
7106
7107 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7108 {
7109         btrfs_dev_stat_inc(dev, index);
7110         btrfs_dev_stat_print_on_error(dev);
7111 }
7112
7113 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7114 {
7115         if (!dev->dev_stats_valid)
7116                 return;
7117         btrfs_err_rl_in_rcu(dev->fs_info,
7118                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7119                            rcu_str_deref(dev->name),
7120                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7121                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7122                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7123                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7124                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7125 }
7126
7127 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7128 {
7129         int i;
7130
7131         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7132                 if (btrfs_dev_stat_read(dev, i) != 0)
7133                         break;
7134         if (i == BTRFS_DEV_STAT_VALUES_MAX)
7135                 return; /* all values == 0, suppress message */
7136
7137         btrfs_info_in_rcu(dev->fs_info,
7138                 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7139                rcu_str_deref(dev->name),
7140                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7141                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7142                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7143                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7144                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7145 }
7146
7147 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7148                         struct btrfs_ioctl_get_dev_stats *stats)
7149 {
7150         struct btrfs_device *dev;
7151         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7152         int i;
7153
7154         mutex_lock(&fs_devices->device_list_mutex);
7155         dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7156         mutex_unlock(&fs_devices->device_list_mutex);
7157
7158         if (!dev) {
7159                 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7160                 return -ENODEV;
7161         } else if (!dev->dev_stats_valid) {
7162                 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7163                 return -ENODEV;
7164         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7165                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7166                         if (stats->nr_items > i)
7167                                 stats->values[i] =
7168                                         btrfs_dev_stat_read_and_reset(dev, i);
7169                         else
7170                                 btrfs_dev_stat_reset(dev, i);
7171                 }
7172         } else {
7173                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7174                         if (stats->nr_items > i)
7175                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
7176         }
7177         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7178                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7179         return 0;
7180 }
7181
7182 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7183 {
7184         struct buffer_head *bh;
7185         struct btrfs_super_block *disk_super;
7186         int copy_num;
7187
7188         if (!bdev)
7189                 return;
7190
7191         for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7192                 copy_num++) {
7193
7194                 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7195                         continue;
7196
7197                 disk_super = (struct btrfs_super_block *)bh->b_data;
7198
7199                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7200                 set_buffer_dirty(bh);
7201                 sync_dirty_buffer(bh);
7202                 brelse(bh);
7203         }
7204
7205         /* Notify udev that device has changed */
7206         btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7207
7208         /* Update ctime/mtime for device path for libblkid */
7209         update_dev_time(device_path);
7210 }
7211
7212 /*
7213  * Update the size of all devices, which is used for writing out the
7214  * super blocks.
7215  */
7216 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7217 {
7218         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7219         struct btrfs_device *curr, *next;
7220
7221         if (list_empty(&fs_devices->resized_devices))
7222                 return;
7223
7224         mutex_lock(&fs_devices->device_list_mutex);
7225         mutex_lock(&fs_info->chunk_mutex);
7226         list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7227                                  resized_list) {
7228                 list_del_init(&curr->resized_list);
7229                 curr->commit_total_bytes = curr->disk_total_bytes;
7230         }
7231         mutex_unlock(&fs_info->chunk_mutex);
7232         mutex_unlock(&fs_devices->device_list_mutex);
7233 }
7234
7235 /* Must be invoked during the transaction commit */
7236 void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info *fs_info,
7237                                         struct btrfs_transaction *transaction)
7238 {
7239         struct extent_map *em;
7240         struct map_lookup *map;
7241         struct btrfs_device *dev;
7242         int i;
7243
7244         if (list_empty(&transaction->pending_chunks))
7245                 return;
7246
7247         /* In order to kick the device replace finish process */
7248         mutex_lock(&fs_info->chunk_mutex);
7249         list_for_each_entry(em, &transaction->pending_chunks, list) {
7250                 map = em->map_lookup;
7251
7252                 for (i = 0; i < map->num_stripes; i++) {
7253                         dev = map->stripes[i].dev;
7254                         dev->commit_bytes_used = dev->bytes_used;
7255                 }
7256         }
7257         mutex_unlock(&fs_info->chunk_mutex);
7258 }
7259
7260 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7261 {
7262         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7263         while (fs_devices) {
7264                 fs_devices->fs_info = fs_info;
7265                 fs_devices = fs_devices->seed;
7266         }
7267 }
7268
7269 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7270 {
7271         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7272         while (fs_devices) {
7273                 fs_devices->fs_info = NULL;
7274                 fs_devices = fs_devices->seed;
7275         }
7276 }