fs/btrfs/free-space-cache.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2008 Red Hat.  All rights reserved.
   4  */
   5
   6 #include <linux/pagemap.h>
   7 #include <linux/sched.h>
   8 #include <linux/sched/signal.h>
   9 #include <linux/slab.h>
  10 #include <linux/math64.h>
  11 #include <linux/ratelimit.h>
  12 #include <linux/error-injection.h>
  13 #include <linux/sched/mm.h>
  14 #include "misc.h"
  15 #include "ctree.h"
  16 #include "free-space-cache.h"
  17 #include "transaction.h"
  18 #include "disk-io.h"
  19 #include "extent_io.h"
  20 #include "volumes.h"
  21 #include "space-info.h"
  22 #include "delalloc-space.h"
  23 #include "block-group.h"
  24 #include "discard.h"
  25
  26 #define BITS_PER_BITMAP         (PAGE_SIZE * 8UL)
  27 #define MAX_CACHE_BYTES_PER_GIG SZ_64K
  28 #define FORCE_EXTENT_THRESHOLD  SZ_1M
  29
  30 struct btrfs_trim_range {
  31         u64 start;
  32         u64 bytes;
  33         struct list_head list;
  34 };
  35
  36 static int link_free_space(struct btrfs_free_space_ctl *ctl,
  37                            struct btrfs_free_space *info);
  38 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
  39                               struct btrfs_free_space *info);
  40 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
  41                          struct btrfs_free_space *bitmap_info, u64 *offset,
  42                          u64 *bytes, bool for_alloc);
  43 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
  44                         struct btrfs_free_space *bitmap_info);
  45 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
  46                               struct btrfs_free_space *info, u64 offset,
  47                               u64 bytes);
  48
  49 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
  50                                                struct btrfs_path *path,
  51                                                u64 offset)
  52 {
  53         struct btrfs_fs_info *fs_info = root->fs_info;
  54         struct btrfs_key key;
  55         struct btrfs_key location;
  56         struct btrfs_disk_key disk_key;
  57         struct btrfs_free_space_header *header;
  58         struct extent_buffer *leaf;
  59         struct inode *inode = NULL;
  60         unsigned nofs_flag;
  61         int ret;
  62
  63         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
  64         key.offset = offset;
  65         key.type = 0;
  66
  67         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
  68         if (ret < 0)
  69                 return ERR_PTR(ret);
  70         if (ret > 0) {
  71                 btrfs_release_path(path);
  72                 return ERR_PTR(-ENOENT);
  73         }
  74
  75         leaf = path->nodes[0];
  76         header = btrfs_item_ptr(leaf, path->slots[0],
  77                                 struct btrfs_free_space_header);
  78         btrfs_free_space_key(leaf, header, &disk_key);
  79         btrfs_disk_key_to_cpu(&location, &disk_key);
  80         btrfs_release_path(path);
  81
  82         /*
  83          * We are often under a trans handle at this point, so we need to make
  84          * sure NOFS is set to keep us from deadlocking.
  85          */
  86         nofs_flag = memalloc_nofs_save();
  87         inode = btrfs_iget_path(fs_info->sb, location.objectid, root, path);
  88         btrfs_release_path(path);
  89         memalloc_nofs_restore(nofs_flag);
  90         if (IS_ERR(inode))
  91                 return inode;
  92
  93         mapping_set_gfp_mask(inode->i_mapping,
  94                         mapping_gfp_constraint(inode->i_mapping,
  95                         ~(__GFP_FS | __GFP_HIGHMEM)));
  96
  97         return inode;
  98 }
  99
 100 struct inode *lookup_free_space_inode(struct btrfs_block_group *block_group,
 101                 struct btrfs_path *path)
 102 {
 103         struct btrfs_fs_info *fs_info = block_group->fs_info;
 104         struct inode *inode = NULL;
 105         u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
 106
 107         spin_lock(&block_group->lock);
 108         if (block_group->inode)
 109                 inode = igrab(block_group->inode);
 110         spin_unlock(&block_group->lock);
 111         if (inode)
 112                 return inode;
 113
 114         inode = __lookup_free_space_inode(fs_info->tree_root, path,
 115                                           block_group->start);
 116         if (IS_ERR(inode))
 117                 return inode;
 118
 119         spin_lock(&block_group->lock);
 120         if (!((BTRFS_I(inode)->flags & flags) == flags)) {
 121                 btrfs_info(fs_info, "Old style space inode found, converting.");
 122                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
 123                         BTRFS_INODE_NODATACOW;
 124                 block_group->disk_cache_state = BTRFS_DC_CLEAR;
 125         }
 126
 127         if (!block_group->iref) {
 128                 block_group->inode = igrab(inode);
 129                 block_group->iref = 1;
 130         }
 131         spin_unlock(&block_group->lock);
 132
 133         return inode;
 134 }
 135
 136 static int __create_free_space_inode(struct btrfs_root *root,
 137                                      struct btrfs_trans_handle *trans,
 138                                      struct btrfs_path *path,
 139                                      u64 ino, u64 offset)
 140 {
 141         struct btrfs_key key;
 142         struct btrfs_disk_key disk_key;
 143         struct btrfs_free_space_header *header;
 144         struct btrfs_inode_item *inode_item;
 145         struct extent_buffer *leaf;
 146         /* We inline CRCs for the free disk space cache */
 147         const u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC |
 148                           BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
 149         int ret;
 150
 151         ret = btrfs_insert_empty_inode(trans, root, path, ino);
 152         if (ret)
 153                 return ret;
 154
 155         leaf = path->nodes[0];
 156         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 157                                     struct btrfs_inode_item);
 158         btrfs_item_key(leaf, &disk_key, path->slots[0]);
 159         memzero_extent_buffer(leaf, (unsigned long)inode_item,
 160                              sizeof(*inode_item));
 161         btrfs_set_inode_generation(leaf, inode_item, trans->transid);
 162         btrfs_set_inode_size(leaf, inode_item, 0);
 163         btrfs_set_inode_nbytes(leaf, inode_item, 0);
 164         btrfs_set_inode_uid(leaf, inode_item, 0);
 165         btrfs_set_inode_gid(leaf, inode_item, 0);
 166         btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
 167         btrfs_set_inode_flags(leaf, inode_item, flags);
 168         btrfs_set_inode_nlink(leaf, inode_item, 1);
 169         btrfs_set_inode_transid(leaf, inode_item, trans->transid);
 170         btrfs_set_inode_block_group(leaf, inode_item, offset);
 171         btrfs_mark_buffer_dirty(leaf);
 172         btrfs_release_path(path);
 173
 174         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
 175         key.offset = offset;
 176         key.type = 0;
 177         ret = btrfs_insert_empty_item(trans, root, path, &key,
 178                                       sizeof(struct btrfs_free_space_header));
 179         if (ret < 0) {
 180                 btrfs_release_path(path);
 181                 return ret;
 182         }
 183
 184         leaf = path->nodes[0];
 185         header = btrfs_item_ptr(leaf, path->slots[0],
 186                                 struct btrfs_free_space_header);
 187         memzero_extent_buffer(leaf, (unsigned long)header, sizeof(*header));
 188         btrfs_set_free_space_key(leaf, header, &disk_key);
 189         btrfs_mark_buffer_dirty(leaf);
 190         btrfs_release_path(path);
 191
 192         return 0;
 193 }
 194
 195 int create_free_space_inode(struct btrfs_trans_handle *trans,
 196                             struct btrfs_block_group *block_group,
 197                             struct btrfs_path *path)
 198 {
 199         int ret;
 200         u64 ino;
 201
 202         ret = btrfs_get_free_objectid(trans->fs_info->tree_root, &ino);
 203         if (ret < 0)
 204                 return ret;
 205
 206         return __create_free_space_inode(trans->fs_info->tree_root, trans, path,
 207                                          ino, block_group->start);
 208 }
 209
 210 /*
 211  * inode is an optional sink: if it is NULL, btrfs_remove_free_space_inode
 212  * handles lookup, otherwise it takes ownership and iputs the inode.
 213  * Don't reuse an inode pointer after passing it into this function.
 214  */
 215 int btrfs_remove_free_space_inode(struct btrfs_trans_handle *trans,
 216                                   struct inode *inode,
 217                                   struct btrfs_block_group *block_group)
 218 {
 219         struct btrfs_path *path;
 220         struct btrfs_key key;
 221         int ret = 0;
 222
 223         path = btrfs_alloc_path();
 224         if (!path)
 225                 return -ENOMEM;
 226
 227         if (!inode)
 228                 inode = lookup_free_space_inode(block_group, path);
 229         if (IS_ERR(inode)) {
 230                 if (PTR_ERR(inode) != -ENOENT)
 231                         ret = PTR_ERR(inode);
 232                 goto out;
 233         }
 234         ret = btrfs_orphan_add(trans, BTRFS_I(inode));
 235         if (ret) {
 236                 btrfs_add_delayed_iput(inode);
 237                 goto out;
 238         }
 239         clear_nlink(inode);
 240         /* One for the block groups ref */
 241         spin_lock(&block_group->lock);
 242         if (block_group->iref) {
 243                 block_group->iref = 0;
 244                 block_group->inode = NULL;
 245                 spin_unlock(&block_group->lock);
 246                 iput(inode);
 247         } else {
 248                 spin_unlock(&block_group->lock);
 249         }
 250         /* One for the lookup ref */
 251         btrfs_add_delayed_iput(inode);
 252
 253         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
 254         key.type = 0;
 255         key.offset = block_group->start;
 256         ret = btrfs_search_slot(trans, trans->fs_info->tree_root, &key, path,
 257                                 -1, 1);
 258         if (ret) {
 259                 if (ret > 0)
 260                         ret = 0;
 261                 goto out;
 262         }
 263         ret = btrfs_del_item(trans, trans->fs_info->tree_root, path);
 264 out:
 265         btrfs_free_path(path);
 266         return ret;
 267 }
 268
 269 int btrfs_check_trunc_cache_free_space(struct btrfs_fs_info *fs_info,
 270                                        struct btrfs_block_rsv *rsv)
 271 {
 272         u64 needed_bytes;
 273         int ret;
 274
 275         /* 1 for slack space, 1 for updating the inode */
 276         needed_bytes = btrfs_calc_insert_metadata_size(fs_info, 1) +
 277                 btrfs_calc_metadata_size(fs_info, 1);
 278
 279         spin_lock(&rsv->lock);
 280         if (rsv->reserved < needed_bytes)
 281                 ret = -ENOSPC;
 282         else
 283                 ret = 0;
 284         spin_unlock(&rsv->lock);
 285         return ret;
 286 }
 287
 288 int btrfs_truncate_free_space_cache(struct btrfs_trans_handle *trans,
 289                                     struct btrfs_block_group *block_group,
 290                                     struct inode *inode)
 291 {
 292         struct btrfs_root *root = BTRFS_I(inode)->root;
 293         int ret = 0;
 294         bool locked = false;
 295
 296         if (block_group) {
 297                 struct btrfs_path *path = btrfs_alloc_path();
 298
 299                 if (!path) {
 300                         ret = -ENOMEM;
 301                         goto fail;
 302                 }
 303                 locked = true;
 304                 mutex_lock(&trans->transaction->cache_write_mutex);
 305                 if (!list_empty(&block_group->io_list)) {
 306                         list_del_init(&block_group->io_list);
 307
 308                         btrfs_wait_cache_io(trans, block_group, path);
 309                         btrfs_put_block_group(block_group);
 310                 }
 311
 312                 /*
 313                  * now that we've truncated the cache away, its no longer
 314                  * setup or written
 315                  */
 316                 spin_lock(&block_group->lock);
 317                 block_group->disk_cache_state = BTRFS_DC_CLEAR;
 318                 spin_unlock(&block_group->lock);
 319                 btrfs_free_path(path);
 320         }
 321
 322         btrfs_i_size_write(BTRFS_I(inode), 0);
 323         truncate_pagecache(inode, 0);
 324
 325         /*
 326          * We skip the throttling logic for free space cache inodes, so we don't
 327          * need to check for -EAGAIN.
 328          */
 329         ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
 330                                          0, BTRFS_EXTENT_DATA_KEY, NULL);
 331         if (ret)
 332                 goto fail;
 333
 334         ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
 335
 336 fail:
 337         if (locked)
 338                 mutex_unlock(&trans->transaction->cache_write_mutex);
 339         if (ret)
 340                 btrfs_abort_transaction(trans, ret);
 341
 342         return ret;
 343 }
 344
 345 static void readahead_cache(struct inode *inode)
 346 {
 347         struct file_ra_state ra;
 348         unsigned long last_index;
 349
 350         file_ra_state_init(&ra, inode->i_mapping);
 351         last_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
 352
 353         page_cache_sync_readahead(inode->i_mapping, &ra, NULL, 0, last_index);
 354 }
 355
 356 static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode,
 357                        int write)
 358 {
 359         int num_pages;
 360
 361         num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
 362
 363         /* Make sure we can fit our crcs and generation into the first page */
 364         if (write && (num_pages * sizeof(u32) + sizeof(u64)) > PAGE_SIZE)
 365                 return -ENOSPC;
 366
 367         memset(io_ctl, 0, sizeof(struct btrfs_io_ctl));
 368
 369         io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS);
 370         if (!io_ctl->pages)
 371                 return -ENOMEM;
 372
 373         io_ctl->num_pages = num_pages;
 374         io_ctl->fs_info = btrfs_sb(inode->i_sb);
 375         io_ctl->inode = inode;
 376
 377         return 0;
 378 }
 379 ALLOW_ERROR_INJECTION(io_ctl_init, ERRNO);
 380
 381 static void io_ctl_free(struct btrfs_io_ctl *io_ctl)
 382 {
 383         kfree(io_ctl->pages);
 384         io_ctl->pages = NULL;
 385 }
 386
 387 static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl)
 388 {
 389         if (io_ctl->cur) {
 390                 io_ctl->cur = NULL;
 391                 io_ctl->orig = NULL;
 392         }
 393 }
 394
 395 static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear)
 396 {
 397         ASSERT(io_ctl->index < io_ctl->num_pages);
 398         io_ctl->page = io_ctl->pages[io_ctl->index++];
 399         io_ctl->cur = page_address(io_ctl->page);
 400         io_ctl->orig = io_ctl->cur;
 401         io_ctl->size = PAGE_SIZE;
 402         if (clear)
 403                 clear_page(io_ctl->cur);
 404 }
 405
 406 static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl)
 407 {
 408         int i;
 409
 410         io_ctl_unmap_page(io_ctl);
 411
 412         for (i = 0; i < io_ctl->num_pages; i++) {
 413                 if (io_ctl->pages[i]) {
 414                         ClearPageChecked(io_ctl->pages[i]);
 415                         unlock_page(io_ctl->pages[i]);
 416                         put_page(io_ctl->pages[i]);
 417                 }
 418         }
 419 }
 420
 421 static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, bool uptodate)
 422 {
 423         struct page *page;
 424         struct inode *inode = io_ctl->inode;
 425         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
 426         int i;
 427
 428         for (i = 0; i < io_ctl->num_pages; i++) {
 429                 int ret;
 430
 431                 page = find_or_create_page(inode->i_mapping, i, mask);
 432                 if (!page) {
 433                         io_ctl_drop_pages(io_ctl);
 434                         return -ENOMEM;
 435                 }
 436
 437                 ret = set_page_extent_mapped(page);
 438                 if (ret < 0) {
 439                         unlock_page(page);
 440                         put_page(page);
 441                         io_ctl_drop_pages(io_ctl);
 442                         return ret;
 443                 }
 444
 445                 io_ctl->pages[i] = page;
 446                 if (uptodate && !PageUptodate(page)) {
 447                         btrfs_readpage(NULL, page);
 448                         lock_page(page);
 449                         if (page->mapping != inode->i_mapping) {
 450                                 btrfs_err(BTRFS_I(inode)->root->fs_info,
 451                                           "free space cache page truncated");
 452                                 io_ctl_drop_pages(io_ctl);
 453                                 return -EIO;
 454                         }
 455                         if (!PageUptodate(page)) {
 456                                 btrfs_err(BTRFS_I(inode)->root->fs_info,
 457                                            "error reading free space cache");
 458                                 io_ctl_drop_pages(io_ctl);
 459                                 return -EIO;
 460                         }
 461                 }
 462         }
 463
 464         for (i = 0; i < io_ctl->num_pages; i++)
 465                 clear_page_dirty_for_io(io_ctl->pages[i]);
 466
 467         return 0;
 468 }
 469
 470 static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
 471 {
 472         io_ctl_map_page(io_ctl, 1);
 473
 474         /*
 475          * Skip the csum areas.  If we don't check crcs then we just have a
 476          * 64bit chunk at the front of the first page.
 477          */
 478         io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
 479         io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
 480
 481         put_unaligned_le64(generation, io_ctl->cur);
 482         io_ctl->cur += sizeof(u64);
 483 }
 484
 485 static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
 486 {
 487         u64 cache_gen;
 488
 489         /*
 490          * Skip the crc area.  If we don't check crcs then we just have a 64bit
 491          * chunk at the front of the first page.
 492          */
 493         io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
 494         io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
 495
 496         cache_gen = get_unaligned_le64(io_ctl->cur);
 497         if (cache_gen != generation) {
 498                 btrfs_err_rl(io_ctl->fs_info,
 499                         "space cache generation (%llu) does not match inode (%llu)",
 500                                 cache_gen, generation);
 501                 io_ctl_unmap_page(io_ctl);
 502                 return -EIO;
 503         }
 504         io_ctl->cur += sizeof(u64);
 505         return 0;
 506 }
 507
 508 static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index)
 509 {
 510         u32 *tmp;
 511         u32 crc = ~(u32)0;
 512         unsigned offset = 0;
 513
 514         if (index == 0)
 515                 offset = sizeof(u32) * io_ctl->num_pages;
 516
 517         crc = btrfs_crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
 518         btrfs_crc32c_final(crc, (u8 *)&crc);
 519         io_ctl_unmap_page(io_ctl);
 520         tmp = page_address(io_ctl->pages[0]);
 521         tmp += index;
 522         *tmp = crc;
 523 }
 524
 525 static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index)
 526 {
 527         u32 *tmp, val;
 528         u32 crc = ~(u32)0;
 529         unsigned offset = 0;
 530
 531         if (index == 0)
 532                 offset = sizeof(u32) * io_ctl->num_pages;
 533
 534         tmp = page_address(io_ctl->pages[0]);
 535         tmp += index;
 536         val = *tmp;
 537
 538         io_ctl_map_page(io_ctl, 0);
 539         crc = btrfs_crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
 540         btrfs_crc32c_final(crc, (u8 *)&crc);
 541         if (val != crc) {
 542                 btrfs_err_rl(io_ctl->fs_info,
 543                         "csum mismatch on free space cache");
 544                 io_ctl_unmap_page(io_ctl);
 545                 return -EIO;
 546         }
 547
 548         return 0;
 549 }
 550
 551 static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes,
 552                             void *bitmap)
 553 {
 554         struct btrfs_free_space_entry *entry;
 555
 556         if (!io_ctl->cur)
 557                 return -ENOSPC;
 558
 559         entry = io_ctl->cur;
 560         put_unaligned_le64(offset, &entry->offset);
 561         put_unaligned_le64(bytes, &entry->bytes);
 562         entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
 563                 BTRFS_FREE_SPACE_EXTENT;
 564         io_ctl->cur += sizeof(struct btrfs_free_space_entry);
 565         io_ctl->size -= sizeof(struct btrfs_free_space_entry);
 566
 567         if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
 568                 return 0;
 569
 570         io_ctl_set_crc(io_ctl, io_ctl->index - 1);
 571
 572         /* No more pages to map */
 573         if (io_ctl->index >= io_ctl->num_pages)
 574                 return 0;
 575
 576         /* map the next page */
 577         io_ctl_map_page(io_ctl, 1);
 578         return 0;
 579 }
 580
 581 static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap)
 582 {
 583         if (!io_ctl->cur)
 584                 return -ENOSPC;
 585
 586         /*
 587          * If we aren't at the start of the current page, unmap this one and
 588          * map the next one if there is any left.
 589          */
 590         if (io_ctl->cur != io_ctl->orig) {
 591                 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
 592                 if (io_ctl->index >= io_ctl->num_pages)
 593                         return -ENOSPC;
 594                 io_ctl_map_page(io_ctl, 0);
 595         }
 596
 597         copy_page(io_ctl->cur, bitmap);
 598         io_ctl_set_crc(io_ctl, io_ctl->index - 1);
 599         if (io_ctl->index < io_ctl->num_pages)
 600                 io_ctl_map_page(io_ctl, 0);
 601         return 0;
 602 }
 603
 604 static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl)
 605 {
 606         /*
 607          * If we're not on the boundary we know we've modified the page and we
 608          * need to crc the page.
 609          */
 610         if (io_ctl->cur != io_ctl->orig)
 611                 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
 612         else
 613                 io_ctl_unmap_page(io_ctl);
 614
 615         while (io_ctl->index < io_ctl->num_pages) {
 616                 io_ctl_map_page(io_ctl, 1);
 617                 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
 618         }
 619 }
 620
 621 static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl,
 622                             struct btrfs_free_space *entry, u8 *type)
 623 {
 624         struct btrfs_free_space_entry *e;
 625         int ret;
 626
 627         if (!io_ctl->cur) {
 628                 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
 629                 if (ret)
 630                         return ret;
 631         }
 632
 633         e = io_ctl->cur;
 634         entry->offset = get_unaligned_le64(&e->offset);
 635         entry->bytes = get_unaligned_le64(&e->bytes);
 636         *type = e->type;
 637         io_ctl->cur += sizeof(struct btrfs_free_space_entry);
 638         io_ctl->size -= sizeof(struct btrfs_free_space_entry);
 639
 640         if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
 641                 return 0;
 642
 643         io_ctl_unmap_page(io_ctl);
 644
 645         return 0;
 646 }
 647
 648 static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl,
 649                               struct btrfs_free_space *entry)
 650 {
 651         int ret;
 652
 653         ret = io_ctl_check_crc(io_ctl, io_ctl->index);
 654         if (ret)
 655                 return ret;
 656
 657         copy_page(entry->bitmap, io_ctl->cur);
 658         io_ctl_unmap_page(io_ctl);
 659
 660         return 0;
 661 }
 662
 663 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
 664 {
 665         struct btrfs_block_group *block_group = ctl->private;
 666         u64 max_bytes;
 667         u64 bitmap_bytes;
 668         u64 extent_bytes;
 669         u64 size = block_group->length;
 670         u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
 671         u64 max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
 672
 673         max_bitmaps = max_t(u64, max_bitmaps, 1);
 674
 675         if (ctl->total_bitmaps > max_bitmaps)
 676                 btrfs_err(block_group->fs_info,
 677 "invalid free space control: bg start=%llu len=%llu total_bitmaps=%u unit=%u max_bitmaps=%llu bytes_per_bg=%llu",
 678                           block_group->start, block_group->length,
 679                           ctl->total_bitmaps, ctl->unit, max_bitmaps,
 680                           bytes_per_bg);
 681         ASSERT(ctl->total_bitmaps <= max_bitmaps);
 682
 683         /*
 684          * We are trying to keep the total amount of memory used per 1GiB of
 685          * space to be MAX_CACHE_BYTES_PER_GIG.  However, with a reclamation
 686          * mechanism of pulling extents >= FORCE_EXTENT_THRESHOLD out of
 687          * bitmaps, we may end up using more memory than this.
 688          */
 689         if (size < SZ_1G)
 690                 max_bytes = MAX_CACHE_BYTES_PER_GIG;
 691         else
 692                 max_bytes = MAX_CACHE_BYTES_PER_GIG * div_u64(size, SZ_1G);
 693
 694         bitmap_bytes = ctl->total_bitmaps * ctl->unit;
 695
 696         /*
 697          * we want the extent entry threshold to always be at most 1/2 the max
 698          * bytes we can have, or whatever is less than that.
 699          */
 700         extent_bytes = max_bytes - bitmap_bytes;
 701         extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1);
 702
 703         ctl->extents_thresh =
 704                 div_u64(extent_bytes, sizeof(struct btrfs_free_space));
 705 }
 706
 707 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
 708                                    struct btrfs_free_space_ctl *ctl,
 709                                    struct btrfs_path *path, u64 offset)
 710 {
 711         struct btrfs_fs_info *fs_info = root->fs_info;
 712         struct btrfs_free_space_header *header;
 713         struct extent_buffer *leaf;
 714         struct btrfs_io_ctl io_ctl;
 715         struct btrfs_key key;
 716         struct btrfs_free_space *e, *n;
 717         LIST_HEAD(bitmaps);
 718         u64 num_entries;
 719         u64 num_bitmaps;
 720         u64 generation;
 721         u8 type;
 722         int ret = 0;
 723
 724         /* Nothing in the space cache, goodbye */
 725         if (!i_size_read(inode))
 726                 return 0;
 727
 728         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
 729         key.offset = offset;
 730         key.type = 0;
 731
 732         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 733         if (ret < 0)
 734                 return 0;
 735         else if (ret > 0) {
 736                 btrfs_release_path(path);
 737                 return 0;
 738         }
 739
 740         ret = -1;
 741
 742         leaf = path->nodes[0];
 743         header = btrfs_item_ptr(leaf, path->slots[0],
 744                                 struct btrfs_free_space_header);
 745         num_entries = btrfs_free_space_entries(leaf, header);
 746         num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
 747         generation = btrfs_free_space_generation(leaf, header);
 748         btrfs_release_path(path);
 749
 750         if (!BTRFS_I(inode)->generation) {
 751                 btrfs_info(fs_info,
 752                            "the free space cache file (%llu) is invalid, skip it",
 753                            offset);
 754                 return 0;
 755         }
 756
 757         if (BTRFS_I(inode)->generation != generation) {
 758                 btrfs_err(fs_info,
 759                           "free space inode generation (%llu) did not match free space cache generation (%llu)",
 760                           BTRFS_I(inode)->generation, generation);
 761                 return 0;
 762         }
 763
 764         if (!num_entries)
 765                 return 0;
 766
 767         ret = io_ctl_init(&io_ctl, inode, 0);
 768         if (ret)
 769                 return ret;
 770
 771         readahead_cache(inode);
 772
 773         ret = io_ctl_prepare_pages(&io_ctl, true);
 774         if (ret)
 775                 goto out;
 776
 777         ret = io_ctl_check_crc(&io_ctl, 0);
 778         if (ret)
 779                 goto free_cache;
 780
 781         ret = io_ctl_check_generation(&io_ctl, generation);
 782         if (ret)
 783                 goto free_cache;
 784
 785         while (num_entries) {
 786                 e = kmem_cache_zalloc(btrfs_free_space_cachep,
 787                                       GFP_NOFS);
 788                 if (!e) {
 789                         ret = -ENOMEM;
 790                         goto free_cache;
 791                 }
 792
 793                 ret = io_ctl_read_entry(&io_ctl, e, &type);
 794                 if (ret) {
 795                         kmem_cache_free(btrfs_free_space_cachep, e);
 796                         goto free_cache;
 797                 }
 798
 799                 if (!e->bytes) {
 800                         ret = -1;
 801                         kmem_cache_free(btrfs_free_space_cachep, e);
 802                         goto free_cache;
 803                 }
 804
 805                 if (type == BTRFS_FREE_SPACE_EXTENT) {
 806                         spin_lock(&ctl->tree_lock);
 807                         ret = link_free_space(ctl, e);
 808                         spin_unlock(&ctl->tree_lock);
 809                         if (ret) {
 810                                 btrfs_err(fs_info,
 811                                         "Duplicate entries in free space cache, dumping");
 812                                 kmem_cache_free(btrfs_free_space_cachep, e);
 813                                 goto free_cache;
 814                         }
 815                 } else {
 816                         ASSERT(num_bitmaps);
 817                         num_bitmaps--;
 818                         e->bitmap = kmem_cache_zalloc(
 819                                         btrfs_free_space_bitmap_cachep, GFP_NOFS);
 820                         if (!e->bitmap) {
 821                                 ret = -ENOMEM;
 822                                 kmem_cache_free(
 823                                         btrfs_free_space_cachep, e);
 824                                 goto free_cache;
 825                         }
 826                         spin_lock(&ctl->tree_lock);
 827                         ret = link_free_space(ctl, e);
 828                         ctl->total_bitmaps++;
 829                         recalculate_thresholds(ctl);
 830                         spin_unlock(&ctl->tree_lock);
 831                         if (ret) {
 832                                 btrfs_err(fs_info,
 833                                         "Duplicate entries in free space cache, dumping");
 834                                 kmem_cache_free(btrfs_free_space_cachep, e);
 835                                 goto free_cache;
 836                         }
 837                         list_add_tail(&e->list, &bitmaps);
 838                 }
 839
 840                 num_entries--;
 841         }
 842
 843         io_ctl_unmap_page(&io_ctl);
 844
 845         /*
 846          * We add the bitmaps at the end of the entries in order that
 847          * the bitmap entries are added to the cache.
 848          */
 849         list_for_each_entry_safe(e, n, &bitmaps, list) {
 850                 list_del_init(&e->list);
 851                 ret = io_ctl_read_bitmap(&io_ctl, e);
 852                 if (ret)
 853                         goto free_cache;
 854         }
 855
 856         io_ctl_drop_pages(&io_ctl);
 857         ret = 1;
 858 out:
 859         io_ctl_free(&io_ctl);
 860         return ret;
 861 free_cache:
 862         io_ctl_drop_pages(&io_ctl);
 863         __btrfs_remove_free_space_cache(ctl);
 864         goto out;
 865 }
 866
 867 static int copy_free_space_cache(struct btrfs_block_group *block_group,
 868                                  struct btrfs_free_space_ctl *ctl)
 869 {
 870         struct btrfs_free_space *info;
 871         struct rb_node *n;
 872         int ret = 0;
 873
 874         while (!ret && (n = rb_first(&ctl->free_space_offset)) != NULL) {
 875                 info = rb_entry(n, struct btrfs_free_space, offset_index);
 876                 if (!info->bitmap) {
 877                         unlink_free_space(ctl, info);
 878                         ret = btrfs_add_free_space(block_group, info->offset,
 879                                                    info->bytes);
 880                         kmem_cache_free(btrfs_free_space_cachep, info);
 881                 } else {
 882                         u64 offset = info->offset;
 883                         u64 bytes = ctl->unit;
 884
 885                         while (search_bitmap(ctl, info, &offset, &bytes,
 886                                              false) == 0) {
 887                                 ret = btrfs_add_free_space(block_group, offset,
 888                                                            bytes);
 889                                 if (ret)
 890                                         break;
 891                                 bitmap_clear_bits(ctl, info, offset, bytes);
 892                                 offset = info->offset;
 893                                 bytes = ctl->unit;
 894                         }
 895                         free_bitmap(ctl, info);
 896                 }
 897                 cond_resched();
 898         }
 899         return ret;
 900 }
 901
 902 int load_free_space_cache(struct btrfs_block_group *block_group)
 903 {
 904         struct btrfs_fs_info *fs_info = block_group->fs_info;
 905         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
 906         struct btrfs_free_space_ctl tmp_ctl = {};
 907         struct inode *inode;
 908         struct btrfs_path *path;
 909         int ret = 0;
 910         bool matched;
 911         u64 used = block_group->used;
 912
 913         /*
 914          * Because we could potentially discard our loaded free space, we want
 915          * to load everything into a temporary structure first, and then if it's
 916          * valid copy it all into the actual free space ctl.
 917          */
 918         btrfs_init_free_space_ctl(block_group, &tmp_ctl);
 919
 920         /*
 921          * If this block group has been marked to be cleared for one reason or
 922          * another then we can't trust the on disk cache, so just return.
 923          */
 924         spin_lock(&block_group->lock);
 925         if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
 926                 spin_unlock(&block_group->lock);
 927                 return 0;
 928         }
 929         spin_unlock(&block_group->lock);
 930
 931         path = btrfs_alloc_path();
 932         if (!path)
 933                 return 0;
 934         path->search_commit_root = 1;
 935         path->skip_locking = 1;
 936
 937         /*
 938          * We must pass a path with search_commit_root set to btrfs_iget in
 939          * order to avoid a deadlock when allocating extents for the tree root.
 940          *
 941          * When we are COWing an extent buffer from the tree root, when looking
 942          * for a free extent, at extent-tree.c:find_free_extent(), we can find
 943          * block group without its free space cache loaded. When we find one
 944          * we must load its space cache which requires reading its free space
 945          * cache's inode item from the root tree. If this inode item is located
 946          * in the same leaf that we started COWing before, then we end up in
 947          * deadlock on the extent buffer (trying to read lock it when we
 948          * previously write locked it).
 949          *
 950          * It's safe to read the inode item using the commit root because
 951          * block groups, once loaded, stay in memory forever (until they are
 952          * removed) as well as their space caches once loaded. New block groups
 953          * once created get their ->cached field set to BTRFS_CACHE_FINISHED so
 954          * we will never try to read their inode item while the fs is mounted.
 955          */
 956         inode = lookup_free_space_inode(block_group, path);
 957         if (IS_ERR(inode)) {
 958                 btrfs_free_path(path);
 959                 return 0;
 960         }
 961
 962         /* We may have converted the inode and made the cache invalid. */
 963         spin_lock(&block_group->lock);
 964         if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
 965                 spin_unlock(&block_group->lock);
 966                 btrfs_free_path(path);
 967                 goto out;
 968         }
 969         spin_unlock(&block_group->lock);
 970
 971         ret = __load_free_space_cache(fs_info->tree_root, inode, &tmp_ctl,
 972                                       path, block_group->start);
 973         btrfs_free_path(path);
 974         if (ret <= 0)
 975                 goto out;
 976
 977         matched = (tmp_ctl.free_space == (block_group->length - used -
 978                                           block_group->bytes_super));
 979
 980         if (matched) {
 981                 ret = copy_free_space_cache(block_group, &tmp_ctl);
 982                 /*
 983                  * ret == 1 means we successfully loaded the free space cache,
 984                  * so we need to re-set it here.
 985                  */
 986                 if (ret == 0)
 987                         ret = 1;
 988         } else {
 989                 __btrfs_remove_free_space_cache(&tmp_ctl);
 990                 btrfs_warn(fs_info,
 991                            "block group %llu has wrong amount of free space",
 992                            block_group->start);
 993                 ret = -1;
 994         }
 995 out:
 996         if (ret < 0) {
 997                 /* This cache is bogus, make sure it gets cleared */
 998                 spin_lock(&block_group->lock);
 999                 block_group->disk_cache_state = BTRFS_DC_CLEAR;
1000                 spin_unlock(&block_group->lock);
1001                 ret = 0;
1002
1003                 btrfs_warn(fs_info,
1004                            "failed to load free space cache for block group %llu, rebuilding it now",
1005                            block_group->start);
1006         }
1007
1008         spin_lock(&ctl->tree_lock);
1009         btrfs_discard_update_discardable(block_group);
1010         spin_unlock(&ctl->tree_lock);
1011         iput(inode);
1012         return ret;
1013 }
1014
1015 static noinline_for_stack
1016 int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl,
1017                               struct btrfs_free_space_ctl *ctl,
1018                               struct btrfs_block_group *block_group,
1019                               int *entries, int *bitmaps,
1020                               struct list_head *bitmap_list)
1021 {
1022         int ret;
1023         struct btrfs_free_cluster *cluster = NULL;
1024         struct btrfs_free_cluster *cluster_locked = NULL;
1025         struct rb_node *node = rb_first(&ctl->free_space_offset);
1026         struct btrfs_trim_range *trim_entry;
1027
1028         /* Get the cluster for this block_group if it exists */
1029         if (block_group && !list_empty(&block_group->cluster_list)) {
1030                 cluster = list_entry(block_group->cluster_list.next,
1031                                      struct btrfs_free_cluster,
1032                                      block_group_list);
1033         }
1034
1035         if (!node && cluster) {
1036                 cluster_locked = cluster;
1037                 spin_lock(&cluster_locked->lock);
1038                 node = rb_first(&cluster->root);
1039                 cluster = NULL;
1040         }
1041
1042         /* Write out the extent entries */
1043         while (node) {
1044                 struct btrfs_free_space *e;
1045
1046                 e = rb_entry(node, struct btrfs_free_space, offset_index);
1047                 *entries += 1;
1048
1049                 ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
1050                                        e->bitmap);
1051                 if (ret)
1052                         goto fail;
1053
1054                 if (e->bitmap) {
1055                         list_add_tail(&e->list, bitmap_list);
1056                         *bitmaps += 1;
1057                 }
1058                 node = rb_next(node);
1059                 if (!node && cluster) {
1060                         node = rb_first(&cluster->root);
1061                         cluster_locked = cluster;
1062                         spin_lock(&cluster_locked->lock);
1063                         cluster = NULL;
1064                 }
1065         }
1066         if (cluster_locked) {
1067                 spin_unlock(&cluster_locked->lock);
1068                 cluster_locked = NULL;
1069         }
1070
1071         /*
1072          * Make sure we don't miss any range that was removed from our rbtree
1073          * because trimming is running. Otherwise after a umount+mount (or crash
1074          * after committing the transaction) we would leak free space and get
1075          * an inconsistent free space cache report from fsck.
1076          */
1077         list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) {
1078                 ret = io_ctl_add_entry(io_ctl, trim_entry->start,
1079                                        trim_entry->bytes, NULL);
1080                 if (ret)
1081                         goto fail;
1082                 *entries += 1;
1083         }
1084
1085         return 0;
1086 fail:
1087         if (cluster_locked)
1088                 spin_unlock(&cluster_locked->lock);
1089         return -ENOSPC;
1090 }
1091
1092 static noinline_for_stack int
1093 update_cache_item(struct btrfs_trans_handle *trans,
1094                   struct btrfs_root *root,
1095                   struct inode *inode,
1096                   struct btrfs_path *path, u64 offset,
1097                   int entries, int bitmaps)
1098 {
1099         struct btrfs_key key;
1100         struct btrfs_free_space_header *header;
1101         struct extent_buffer *leaf;
1102         int ret;
1103
1104         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1105         key.offset = offset;
1106         key.type = 0;
1107
1108         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1109         if (ret < 0) {
1110                 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1111                                  EXTENT_DELALLOC, 0, 0, NULL);
1112                 goto fail;
1113         }
1114         leaf = path->nodes[0];
1115         if (ret > 0) {
1116                 struct btrfs_key found_key;
1117                 ASSERT(path->slots[0]);
1118                 path->slots[0]--;
1119                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1120                 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1121                     found_key.offset != offset) {
1122                         clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1123                                          inode->i_size - 1, EXTENT_DELALLOC, 0,
1124                                          0, NULL);
1125                         btrfs_release_path(path);
1126                         goto fail;
1127                 }
1128         }
1129
1130         BTRFS_I(inode)->generation = trans->transid;
1131         header = btrfs_item_ptr(leaf, path->slots[0],
1132                                 struct btrfs_free_space_header);
1133         btrfs_set_free_space_entries(leaf, header, entries);
1134         btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1135         btrfs_set_free_space_generation(leaf, header, trans->transid);
1136         btrfs_mark_buffer_dirty(leaf);
1137         btrfs_release_path(path);
1138
1139         return 0;
1140
1141 fail:
1142         return -1;
1143 }
1144
1145 static noinline_for_stack int write_pinned_extent_entries(
1146                             struct btrfs_trans_handle *trans,
1147                             struct btrfs_block_group *block_group,
1148                             struct btrfs_io_ctl *io_ctl,
1149                             int *entries)
1150 {
1151         u64 start, extent_start, extent_end, len;
1152         struct extent_io_tree *unpin = NULL;
1153         int ret;
1154
1155         if (!block_group)
1156                 return 0;
1157
1158         /*
1159          * We want to add any pinned extents to our free space cache
1160          * so we don't leak the space
1161          *
1162          * We shouldn't have switched the pinned extents yet so this is the
1163          * right one
1164          */
1165         unpin = &trans->transaction->pinned_extents;
1166
1167         start = block_group->start;
1168
1169         while (start < block_group->start + block_group->length) {
1170                 ret = find_first_extent_bit(unpin, start,
1171                                             &extent_start, &extent_end,
1172                                             EXTENT_DIRTY, NULL);
1173                 if (ret)
1174                         return 0;
1175
1176                 /* This pinned extent is out of our range */
1177                 if (extent_start >= block_group->start + block_group->length)
1178                         return 0;
1179
1180                 extent_start = max(extent_start, start);
1181                 extent_end = min(block_group->start + block_group->length,
1182                                  extent_end + 1);
1183                 len = extent_end - extent_start;
1184
1185                 *entries += 1;
1186                 ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
1187                 if (ret)
1188                         return -ENOSPC;
1189
1190                 start = extent_end;
1191         }
1192
1193         return 0;
1194 }
1195
1196 static noinline_for_stack int
1197 write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list)
1198 {
1199         struct btrfs_free_space *entry, *next;
1200         int ret;
1201
1202         /* Write out the bitmaps */
1203         list_for_each_entry_safe(entry, next, bitmap_list, list) {
1204                 ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
1205                 if (ret)
1206                         return -ENOSPC;
1207                 list_del_init(&entry->list);
1208         }
1209
1210         return 0;
1211 }
1212
1213 static int flush_dirty_cache(struct inode *inode)
1214 {
1215         int ret;
1216
1217         ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1218         if (ret)
1219                 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1220                                  EXTENT_DELALLOC, 0, 0, NULL);
1221
1222         return ret;
1223 }
1224
1225 static void noinline_for_stack
1226 cleanup_bitmap_list(struct list_head *bitmap_list)
1227 {
1228         struct btrfs_free_space *entry, *next;
1229
1230         list_for_each_entry_safe(entry, next, bitmap_list, list)
1231                 list_del_init(&entry->list);
1232 }
1233
1234 static void noinline_for_stack
1235 cleanup_write_cache_enospc(struct inode *inode,
1236                            struct btrfs_io_ctl *io_ctl,
1237                            struct extent_state **cached_state)
1238 {
1239         io_ctl_drop_pages(io_ctl);
1240         unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1241                              i_size_read(inode) - 1, cached_state);
1242 }
1243
1244 static int __btrfs_wait_cache_io(struct btrfs_root *root,
1245                                  struct btrfs_trans_handle *trans,
1246                                  struct btrfs_block_group *block_group,
1247                                  struct btrfs_io_ctl *io_ctl,
1248                                  struct btrfs_path *path, u64 offset)
1249 {
1250         int ret;
1251         struct inode *inode = io_ctl->inode;
1252
1253         if (!inode)
1254                 return 0;
1255
1256         /* Flush the dirty pages in the cache file. */
1257         ret = flush_dirty_cache(inode);
1258         if (ret)
1259                 goto out;
1260
1261         /* Update the cache item to tell everyone this cache file is valid. */
1262         ret = update_cache_item(trans, root, inode, path, offset,
1263                                 io_ctl->entries, io_ctl->bitmaps);
1264 out:
1265         if (ret) {
1266                 invalidate_inode_pages2(inode->i_mapping);
1267                 BTRFS_I(inode)->generation = 0;
1268                 if (block_group)
1269                         btrfs_debug(root->fs_info,
1270           "failed to write free space cache for block group %llu error %d",
1271                                   block_group->start, ret);
1272         }
1273         btrfs_update_inode(trans, root, BTRFS_I(inode));
1274
1275         if (block_group) {
1276                 /* the dirty list is protected by the dirty_bgs_lock */
1277                 spin_lock(&trans->transaction->dirty_bgs_lock);
1278
1279                 /* the disk_cache_state is protected by the block group lock */
1280                 spin_lock(&block_group->lock);
1281
1282                 /*
1283                  * only mark this as written if we didn't get put back on
1284                  * the dirty list while waiting for IO.   Otherwise our
1285                  * cache state won't be right, and we won't get written again
1286                  */
1287                 if (!ret && list_empty(&block_group->dirty_list))
1288                         block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1289                 else if (ret)
1290                         block_group->disk_cache_state = BTRFS_DC_ERROR;
1291
1292                 spin_unlock(&block_group->lock);
1293                 spin_unlock(&trans->transaction->dirty_bgs_lock);
1294                 io_ctl->inode = NULL;
1295                 iput(inode);
1296         }
1297
1298         return ret;
1299
1300 }
1301
1302 int btrfs_wait_cache_io(struct btrfs_trans_handle *trans,
1303                         struct btrfs_block_group *block_group,
1304                         struct btrfs_path *path)
1305 {
1306         return __btrfs_wait_cache_io(block_group->fs_info->tree_root, trans,
1307                                      block_group, &block_group->io_ctl,
1308                                      path, block_group->start);
1309 }
1310
1311 /**
1312  * Write out cached info to an inode
1313  *
1314  * @root:        root the inode belongs to
1315  * @inode:       freespace inode we are writing out
1316  * @ctl:         free space cache we are going to write out
1317  * @block_group: block_group for this cache if it belongs to a block_group
1318  * @io_ctl:      holds context for the io
1319  * @trans:       the trans handle
1320  *
1321  * This function writes out a free space cache struct to disk for quick recovery
1322  * on mount.  This will return 0 if it was successful in writing the cache out,
1323  * or an errno if it was not.
1324  */
1325 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
1326                                    struct btrfs_free_space_ctl *ctl,
1327                                    struct btrfs_block_group *block_group,
1328                                    struct btrfs_io_ctl *io_ctl,
1329                                    struct btrfs_trans_handle *trans)
1330 {
1331         struct extent_state *cached_state = NULL;
1332         LIST_HEAD(bitmap_list);
1333         int entries = 0;
1334         int bitmaps = 0;
1335         int ret;
1336         int must_iput = 0;
1337
1338         if (!i_size_read(inode))
1339                 return -EIO;
1340
1341         WARN_ON(io_ctl->pages);
1342         ret = io_ctl_init(io_ctl, inode, 1);
1343         if (ret)
1344                 return ret;
1345
1346         if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) {
1347                 down_write(&block_group->data_rwsem);
1348                 spin_lock(&block_group->lock);
1349                 if (block_group->delalloc_bytes) {
1350                         block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1351                         spin_unlock(&block_group->lock);
1352                         up_write(&block_group->data_rwsem);
1353                         BTRFS_I(inode)->generation = 0;
1354                         ret = 0;
1355                         must_iput = 1;
1356                         goto out;
1357                 }
1358                 spin_unlock(&block_group->lock);
1359         }
1360
1361         /* Lock all pages first so we can lock the extent safely. */
1362         ret = io_ctl_prepare_pages(io_ctl, false);
1363         if (ret)
1364                 goto out_unlock;
1365
1366         lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1367                          &cached_state);
1368
1369         io_ctl_set_generation(io_ctl, trans->transid);
1370
1371         mutex_lock(&ctl->cache_writeout_mutex);
1372         /* Write out the extent entries in the free space cache */
1373         spin_lock(&ctl->tree_lock);
1374         ret = write_cache_extent_entries(io_ctl, ctl,
1375                                          block_group, &entries, &bitmaps,
1376                                          &bitmap_list);
1377         if (ret)
1378                 goto out_nospc_locked;
1379
1380         /*
1381          * Some spaces that are freed in the current transaction are pinned,
1382          * they will be added into free space cache after the transaction is
1383          * committed, we shouldn't lose them.
1384          *
1385          * If this changes while we are working we'll get added back to
1386          * the dirty list and redo it.  No locking needed
1387          */
1388         ret = write_pinned_extent_entries(trans, block_group, io_ctl, &entries);
1389         if (ret)
1390                 goto out_nospc_locked;
1391
1392         /*
1393          * At last, we write out all the bitmaps and keep cache_writeout_mutex
1394          * locked while doing it because a concurrent trim can be manipulating
1395          * or freeing the bitmap.
1396          */
1397         ret = write_bitmap_entries(io_ctl, &bitmap_list);
1398         spin_unlock(&ctl->tree_lock);
1399         mutex_unlock(&ctl->cache_writeout_mutex);
1400         if (ret)
1401                 goto out_nospc;
1402
1403         /* Zero out the rest of the pages just to make sure */
1404         io_ctl_zero_remaining_pages(io_ctl);
1405
1406         /* Everything is written out, now we dirty the pages in the file. */
1407         ret = btrfs_dirty_pages(BTRFS_I(inode), io_ctl->pages,
1408                                 io_ctl->num_pages, 0, i_size_read(inode),
1409                                 &cached_state, false);
1410         if (ret)
1411                 goto out_nospc;
1412
1413         if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1414                 up_write(&block_group->data_rwsem);
1415         /*
1416          * Release the pages and unlock the extent, we will flush
1417          * them out later
1418          */
1419         io_ctl_drop_pages(io_ctl);
1420         io_ctl_free(io_ctl);
1421
1422         unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1423                              i_size_read(inode) - 1, &cached_state);
1424
1425         /*
1426          * at this point the pages are under IO and we're happy,
1427          * The caller is responsible for waiting on them and updating
1428          * the cache and the inode
1429          */
1430         io_ctl->entries = entries;
1431         io_ctl->bitmaps = bitmaps;
1432
1433         ret = btrfs_fdatawrite_range(inode, 0, (u64)-1);
1434         if (ret)
1435                 goto out;
1436
1437         return 0;
1438
1439 out_nospc_locked:
1440         cleanup_bitmap_list(&bitmap_list);
1441         spin_unlock(&ctl->tree_lock);
1442         mutex_unlock(&ctl->cache_writeout_mutex);
1443
1444 out_nospc:
1445         cleanup_write_cache_enospc(inode, io_ctl, &cached_state);
1446
1447 out_unlock:
1448         if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1449                 up_write(&block_group->data_rwsem);
1450
1451 out:
1452         io_ctl->inode = NULL;
1453         io_ctl_free(io_ctl);
1454         if (ret) {
1455                 invalidate_inode_pages2(inode->i_mapping);
1456                 BTRFS_I(inode)->generation = 0;
1457         }
1458         btrfs_update_inode(trans, root, BTRFS_I(inode));
1459         if (must_iput)
1460                 iput(inode);
1461         return ret;
1462 }
1463
1464 int btrfs_write_out_cache(struct btrfs_trans_handle *trans,
1465                           struct btrfs_block_group *block_group,
1466                           struct btrfs_path *path)
1467 {
1468         struct btrfs_fs_info *fs_info = trans->fs_info;
1469         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1470         struct inode *inode;
1471         int ret = 0;
1472
1473         spin_lock(&block_group->lock);
1474         if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1475                 spin_unlock(&block_group->lock);
1476                 return 0;
1477         }
1478         spin_unlock(&block_group->lock);
1479
1480         inode = lookup_free_space_inode(block_group, path);
1481         if (IS_ERR(inode))
1482                 return 0;
1483
1484         ret = __btrfs_write_out_cache(fs_info->tree_root, inode, ctl,
1485                                 block_group, &block_group->io_ctl, trans);
1486         if (ret) {
1487                 btrfs_debug(fs_info,
1488           "failed to write free space cache for block group %llu error %d",
1489                           block_group->start, ret);
1490                 spin_lock(&block_group->lock);
1491                 block_group->disk_cache_state = BTRFS_DC_ERROR;
1492                 spin_unlock(&block_group->lock);
1493
1494                 block_group->io_ctl.inode = NULL;
1495                 iput(inode);
1496         }
1497
1498         /*
1499          * if ret == 0 the caller is expected to call btrfs_wait_cache_io
1500          * to wait for IO and put the inode
1501          */
1502
1503         return ret;
1504 }
1505
1506 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1507                                           u64 offset)
1508 {
1509         ASSERT(offset >= bitmap_start);
1510         offset -= bitmap_start;
1511         return (unsigned long)(div_u64(offset, unit));
1512 }
1513
1514 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1515 {
1516         return (unsigned long)(div_u64(bytes, unit));
1517 }
1518
1519 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1520                                    u64 offset)
1521 {
1522         u64 bitmap_start;
1523         u64 bytes_per_bitmap;
1524
1525         bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1526         bitmap_start = offset - ctl->start;
1527         bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1528         bitmap_start *= bytes_per_bitmap;
1529         bitmap_start += ctl->start;
1530
1531         return bitmap_start;
1532 }
1533
1534 static int tree_insert_offset(struct rb_root *root, u64 offset,
1535                               struct rb_node *node, int bitmap)
1536 {
1537         struct rb_node **p = &root->rb_node;
1538         struct rb_node *parent = NULL;
1539         struct btrfs_free_space *info;
1540
1541         while (*p) {
1542                 parent = *p;
1543                 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1544
1545                 if (offset < info->offset) {
1546                         p = &(*p)->rb_left;
1547                 } else if (offset > info->offset) {
1548                         p = &(*p)->rb_right;
1549                 } else {
1550                         /*
1551                          * we could have a bitmap entry and an extent entry
1552                          * share the same offset.  If this is the case, we want
1553                          * the extent entry to always be found first if we do a
1554                          * linear search through the tree, since we want to have
1555                          * the quickest allocation time, and allocating from an
1556                          * extent is faster than allocating from a bitmap.  So
1557                          * if we're inserting a bitmap and we find an entry at
1558                          * this offset, we want to go right, or after this entry
1559                          * logically.  If we are inserting an extent and we've
1560                          * found a bitmap, we want to go left, or before
1561                          * logically.
1562                          */
1563                         if (bitmap) {
1564                                 if (info->bitmap) {
1565                                         WARN_ON_ONCE(1);
1566                                         return -EEXIST;
1567                                 }
1568                                 p = &(*p)->rb_right;
1569                         } else {
1570                                 if (!info->bitmap) {
1571                                         WARN_ON_ONCE(1);
1572                                         return -EEXIST;
1573                                 }
1574                                 p = &(*p)->rb_left;
1575                         }
1576                 }
1577         }
1578
1579         rb_link_node(node, parent, p);
1580         rb_insert_color(node, root);
1581
1582         return 0;
1583 }
1584
1585 /*
1586  * searches the tree for the given offset.
1587  *
1588  * fuzzy - If this is set, then we are trying to make an allocation, and we just
1589  * want a section that has at least bytes size and comes at or after the given
1590  * offset.
1591  */
1592 static struct btrfs_free_space *
1593 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1594                    u64 offset, int bitmap_only, int fuzzy)
1595 {
1596         struct rb_node *n = ctl->free_space_offset.rb_node;
1597         struct btrfs_free_space *entry, *prev = NULL;
1598
1599         /* find entry that is closest to the 'offset' */
1600         while (1) {
1601                 if (!n) {
1602                         entry = NULL;
1603                         break;
1604                 }
1605
1606                 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1607                 prev = entry;
1608
1609                 if (offset < entry->offset)
1610                         n = n->rb_left;
1611                 else if (offset > entry->offset)
1612                         n = n->rb_right;
1613                 else
1614                         break;
1615         }
1616
1617         if (bitmap_only) {
1618                 if (!entry)
1619                         return NULL;
1620                 if (entry->bitmap)
1621                         return entry;
1622
1623                 /*
1624                  * bitmap entry and extent entry may share same offset,
1625                  * in that case, bitmap entry comes after extent entry.
1626                  */
1627                 n = rb_next(n);
1628                 if (!n)
1629                         return NULL;
1630                 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1631                 if (entry->offset != offset)
1632                         return NULL;
1633
1634                 WARN_ON(!entry->bitmap);
1635                 return entry;
1636         } else if (entry) {
1637                 if (entry->bitmap) {
1638                         /*
1639                          * if previous extent entry covers the offset,
1640                          * we should return it instead of the bitmap entry
1641                          */
1642                         n = rb_prev(&entry->offset_index);
1643                         if (n) {
1644                                 prev = rb_entry(n, struct btrfs_free_space,
1645                                                 offset_index);
1646                                 if (!prev->bitmap &&
1647                                     prev->offset + prev->bytes > offset)
1648                                         entry = prev;
1649                         }
1650                 }
1651                 return entry;
1652         }
1653
1654         if (!prev)
1655                 return NULL;
1656
1657         /* find last entry before the 'offset' */
1658         entry = prev;
1659         if (entry->offset > offset) {
1660                 n = rb_prev(&entry->offset_index);
1661                 if (n) {
1662                         entry = rb_entry(n, struct btrfs_free_space,
1663                                         offset_index);
1664                         ASSERT(entry->offset <= offset);
1665                 } else {
1666                         if (fuzzy)
1667                                 return entry;
1668                         else
1669                                 return NULL;
1670                 }
1671         }
1672
1673         if (entry->bitmap) {
1674                 n = rb_prev(&entry->offset_index);
1675                 if (n) {
1676                         prev = rb_entry(n, struct btrfs_free_space,
1677                                         offset_index);
1678                         if (!prev->bitmap &&
1679                             prev->offset + prev->bytes > offset)
1680                                 return prev;
1681                 }
1682                 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1683                         return entry;
1684         } else if (entry->offset + entry->bytes > offset)
1685                 return entry;
1686
1687         if (!fuzzy)
1688                 return NULL;
1689
1690         while (1) {
1691                 if (entry->bitmap) {
1692                         if (entry->offset + BITS_PER_BITMAP *
1693                             ctl->unit > offset)
1694                                 break;
1695                 } else {
1696                         if (entry->offset + entry->bytes > offset)
1697                                 break;
1698                 }
1699
1700                 n = rb_next(&entry->offset_index);
1701                 if (!n)
1702                         return NULL;
1703                 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1704         }
1705         return entry;
1706 }
1707
1708 static inline void
1709 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1710                     struct btrfs_free_space *info)
1711 {
1712         rb_erase(&info->offset_index, &ctl->free_space_offset);
1713         ctl->free_extents--;
1714
1715         if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1716                 ctl->discardable_extents[BTRFS_STAT_CURR]--;
1717                 ctl->discardable_bytes[BTRFS_STAT_CURR] -= info->bytes;
1718         }
1719 }
1720
1721 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1722                               struct btrfs_free_space *info)
1723 {
1724         __unlink_free_space(ctl, info);
1725         ctl->free_space -= info->bytes;
1726 }
1727
1728 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1729                            struct btrfs_free_space *info)
1730 {
1731         int ret = 0;
1732
1733         ASSERT(info->bytes || info->bitmap);
1734         ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1735                                  &info->offset_index, (info->bitmap != NULL));
1736         if (ret)
1737                 return ret;
1738
1739         if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1740                 ctl->discardable_extents[BTRFS_STAT_CURR]++;
1741                 ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
1742         }
1743
1744         ctl->free_space += info->bytes;
1745         ctl->free_extents++;
1746         return ret;
1747 }
1748
1749 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1750                                        struct btrfs_free_space *info,
1751                                        u64 offset, u64 bytes)
1752 {
1753         unsigned long start, count, end;
1754         int extent_delta = -1;
1755
1756         start = offset_to_bit(info->offset, ctl->unit, offset);
1757         count = bytes_to_bits(bytes, ctl->unit);
1758         end = start + count;
1759         ASSERT(end <= BITS_PER_BITMAP);
1760
1761         bitmap_clear(info->bitmap, start, count);
1762
1763         info->bytes -= bytes;
1764         if (info->max_extent_size > ctl->unit)
1765                 info->max_extent_size = 0;
1766
1767         if (start && test_bit(start - 1, info->bitmap))
1768                 extent_delta++;
1769
1770         if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
1771                 extent_delta++;
1772
1773         info->bitmap_extents += extent_delta;
1774         if (!btrfs_free_space_trimmed(info)) {
1775                 ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1776                 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
1777         }
1778 }
1779
1780 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1781                               struct btrfs_free_space *info, u64 offset,
1782                               u64 bytes)
1783 {
1784         __bitmap_clear_bits(ctl, info, offset, bytes);
1785         ctl->free_space -= bytes;
1786 }
1787
1788 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1789                             struct btrfs_free_space *info, u64 offset,
1790                             u64 bytes)
1791 {
1792         unsigned long start, count, end;
1793         int extent_delta = 1;
1794
1795         start = offset_to_bit(info->offset, ctl->unit, offset);
1796         count = bytes_to_bits(bytes, ctl->unit);
1797         end = start + count;
1798         ASSERT(end <= BITS_PER_BITMAP);
1799
1800         bitmap_set(info->bitmap, start, count);
1801
1802         info->bytes += bytes;
1803         ctl->free_space += bytes;
1804
1805         if (start && test_bit(start - 1, info->bitmap))
1806                 extent_delta--;
1807
1808         if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
1809                 extent_delta--;
1810
1811         info->bitmap_extents += extent_delta;
1812         if (!btrfs_free_space_trimmed(info)) {
1813                 ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1814                 ctl->discardable_bytes[BTRFS_STAT_CURR] += bytes;
1815         }
1816 }
1817
1818 /*
1819  * If we can not find suitable extent, we will use bytes to record
1820  * the size of the max extent.
1821  */
1822 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1823                          struct btrfs_free_space *bitmap_info, u64 *offset,
1824                          u64 *bytes, bool for_alloc)
1825 {
1826         unsigned long found_bits = 0;
1827         unsigned long max_bits = 0;
1828         unsigned long bits, i;
1829         unsigned long next_zero;
1830         unsigned long extent_bits;
1831
1832         /*
1833          * Skip searching the bitmap if we don't have a contiguous section that
1834          * is large enough for this allocation.
1835          */
1836         if (for_alloc &&
1837             bitmap_info->max_extent_size &&
1838             bitmap_info->max_extent_size < *bytes) {
1839                 *bytes = bitmap_info->max_extent_size;
1840                 return -1;
1841         }
1842
1843         i = offset_to_bit(bitmap_info->offset, ctl->unit,
1844                           max_t(u64, *offset, bitmap_info->offset));
1845         bits = bytes_to_bits(*bytes, ctl->unit);
1846
1847         for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1848                 if (for_alloc && bits == 1) {
1849                         found_bits = 1;
1850                         break;
1851                 }
1852                 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1853                                                BITS_PER_BITMAP, i);
1854                 extent_bits = next_zero - i;
1855                 if (extent_bits >= bits) {
1856                         found_bits = extent_bits;
1857                         break;
1858                 } else if (extent_bits > max_bits) {
1859                         max_bits = extent_bits;
1860                 }
1861                 i = next_zero;
1862         }
1863
1864         if (found_bits) {
1865                 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1866                 *bytes = (u64)(found_bits) * ctl->unit;
1867                 return 0;
1868         }
1869
1870         *bytes = (u64)(max_bits) * ctl->unit;
1871         bitmap_info->max_extent_size = *bytes;
1872         return -1;
1873 }
1874
1875 static inline u64 get_max_extent_size(struct btrfs_free_space *entry)
1876 {
1877         if (entry->bitmap)
1878                 return entry->max_extent_size;
1879         return entry->bytes;
1880 }
1881
1882 /* Cache the size of the max extent in bytes */
1883 static struct btrfs_free_space *
1884 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1885                 unsigned long align, u64 *max_extent_size)
1886 {
1887         struct btrfs_free_space *entry;
1888         struct rb_node *node;
1889         u64 tmp;
1890         u64 align_off;
1891         int ret;
1892
1893         if (!ctl->free_space_offset.rb_node)
1894                 goto out;
1895
1896         entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1897         if (!entry)
1898                 goto out;
1899
1900         for (node = &entry->offset_index; node; node = rb_next(node)) {
1901                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1902                 if (entry->bytes < *bytes) {
1903                         *max_extent_size = max(get_max_extent_size(entry),
1904                                                *max_extent_size);
1905                         continue;
1906                 }
1907
1908                 /* make sure the space returned is big enough
1909                  * to match our requested alignment
1910                  */
1911                 if (*bytes >= align) {
1912                         tmp = entry->offset - ctl->start + align - 1;
1913                         tmp = div64_u64(tmp, align);
1914                         tmp = tmp * align + ctl->start;
1915                         align_off = tmp - entry->offset;
1916                 } else {
1917                         align_off = 0;
1918                         tmp = entry->offset;
1919                 }
1920
1921                 if (entry->bytes < *bytes + align_off) {
1922                         *max_extent_size = max(get_max_extent_size(entry),
1923                                                *max_extent_size);
1924                         continue;
1925                 }
1926
1927                 if (entry->bitmap) {
1928                         u64 size = *bytes;
1929
1930                         ret = search_bitmap(ctl, entry, &tmp, &size, true);
1931                         if (!ret) {
1932                                 *offset = tmp;
1933                                 *bytes = size;
1934                                 return entry;
1935                         } else {
1936                                 *max_extent_size =
1937                                         max(get_max_extent_size(entry),
1938                                             *max_extent_size);
1939                         }
1940                         continue;
1941                 }
1942
1943                 *offset = tmp;
1944                 *bytes = entry->bytes - align_off;
1945                 return entry;
1946         }
1947 out:
1948         return NULL;
1949 }
1950
1951 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1952                            struct btrfs_free_space *info, u64 offset)
1953 {
1954         info->offset = offset_to_bitmap(ctl, offset);
1955         info->bytes = 0;
1956         info->bitmap_extents = 0;
1957         INIT_LIST_HEAD(&info->list);
1958         link_free_space(ctl, info);
1959         ctl->total_bitmaps++;
1960         recalculate_thresholds(ctl);
1961 }
1962
1963 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1964                         struct btrfs_free_space *bitmap_info)
1965 {
1966         /*
1967          * Normally when this is called, the bitmap is completely empty. However,
1968          * if we are blowing up the free space cache for one reason or another
1969          * via __btrfs_remove_free_space_cache(), then it may not be freed and
1970          * we may leave stats on the table.
1971          */
1972         if (bitmap_info->bytes && !btrfs_free_space_trimmed(bitmap_info)) {
1973                 ctl->discardable_extents[BTRFS_STAT_CURR] -=
1974                         bitmap_info->bitmap_extents;
1975                 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bitmap_info->bytes;
1976
1977         }
1978         unlink_free_space(ctl, bitmap_info);
1979         kmem_cache_free(btrfs_free_space_bitmap_cachep, bitmap_info->bitmap);
1980         kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1981         ctl->total_bitmaps--;
1982         recalculate_thresholds(ctl);
1983 }
1984
1985 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1986                               struct btrfs_free_space *bitmap_info,
1987                               u64 *offset, u64 *bytes)
1988 {
1989         u64 end;
1990         u64 search_start, search_bytes;
1991         int ret;
1992
1993 again:
1994         end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1995
1996         /*
1997          * We need to search for bits in this bitmap.  We could only cover some
1998          * of the extent in this bitmap thanks to how we add space, so we need
1999          * to search for as much as it as we can and clear that amount, and then
2000          * go searching for the next bit.
2001          */
2002         search_start = *offset;
2003         search_bytes = ctl->unit;
2004         search_bytes = min(search_bytes, end - search_start + 1);
2005         ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes,
2006                             false);
2007         if (ret < 0 || search_start != *offset)
2008                 return -EINVAL;
2009
2010         /* We may have found more bits than what we need */
2011         search_bytes = min(search_bytes, *bytes);
2012
2013         /* Cannot clear past the end of the bitmap */
2014         search_bytes = min(search_bytes, end - search_start + 1);
2015
2016         bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
2017         *offset += search_bytes;
2018         *bytes -= search_bytes;
2019
2020         if (*bytes) {
2021                 struct rb_node *next = rb_next(&bitmap_info->offset_index);
2022                 if (!bitmap_info->bytes)
2023                         free_bitmap(ctl, bitmap_info);
2024
2025                 /*
2026                  * no entry after this bitmap, but we still have bytes to
2027                  * remove, so something has gone wrong.
2028                  */
2029                 if (!next)
2030                         return -EINVAL;
2031
2032                 bitmap_info = rb_entry(next, struct btrfs_free_space,
2033                                        offset_index);
2034
2035                 /*
2036                  * if the next entry isn't a bitmap we need to return to let the
2037                  * extent stuff do its work.
2038                  */
2039                 if (!bitmap_info->bitmap)
2040                         return -EAGAIN;
2041
2042                 /*
2043                  * Ok the next item is a bitmap, but it may not actually hold
2044                  * the information for the rest of this free space stuff, so
2045                  * look for it, and if we don't find it return so we can try
2046                  * everything over again.
2047                  */
2048                 search_start = *offset;
2049                 search_bytes = ctl->unit;
2050                 ret = search_bitmap(ctl, bitmap_info, &search_start,
2051                                     &search_bytes, false);
2052                 if (ret < 0 || search_start != *offset)
2053                         return -EAGAIN;
2054
2055                 goto again;
2056         } else if (!bitmap_info->bytes)
2057                 free_bitmap(ctl, bitmap_info);
2058
2059         return 0;
2060 }
2061
2062 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
2063                                struct btrfs_free_space *info, u64 offset,
2064                                u64 bytes, enum btrfs_trim_state trim_state)
2065 {
2066         u64 bytes_to_set = 0;
2067         u64 end;
2068
2069         /*
2070          * This is a tradeoff to make bitmap trim state minimal.  We mark the
2071          * whole bitmap untrimmed if at any point we add untrimmed regions.
2072          */
2073         if (trim_state == BTRFS_TRIM_STATE_UNTRIMMED) {
2074                 if (btrfs_free_space_trimmed(info)) {
2075                         ctl->discardable_extents[BTRFS_STAT_CURR] +=
2076                                 info->bitmap_extents;
2077                         ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
2078                 }
2079                 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2080         }
2081
2082         end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
2083
2084         bytes_to_set = min(end - offset, bytes);
2085
2086         bitmap_set_bits(ctl, info, offset, bytes_to_set);
2087
2088         /*
2089          * We set some bytes, we have no idea what the max extent size is
2090          * anymore.
2091          */
2092         info->max_extent_size = 0;
2093
2094         return bytes_to_set;
2095
2096 }
2097
2098 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
2099                       struct btrfs_free_space *info)
2100 {
2101         struct btrfs_block_group *block_group = ctl->private;
2102         struct btrfs_fs_info *fs_info = block_group->fs_info;
2103         bool forced = false;
2104
2105 #ifdef CONFIG_BTRFS_DEBUG
2106         if (btrfs_should_fragment_free_space(block_group))
2107                 forced = true;
2108 #endif
2109
2110         /* This is a way to reclaim large regions from the bitmaps. */
2111         if (!forced && info->bytes >= FORCE_EXTENT_THRESHOLD)
2112                 return false;
2113
2114         /*
2115          * If we are below the extents threshold then we can add this as an
2116          * extent, and don't have to deal with the bitmap
2117          */
2118         if (!forced && ctl->free_extents < ctl->extents_thresh) {
2119                 /*
2120                  * If this block group has some small extents we don't want to
2121                  * use up all of our free slots in the cache with them, we want
2122                  * to reserve them to larger extents, however if we have plenty
2123                  * of cache left then go ahead an dadd them, no sense in adding
2124                  * the overhead of a bitmap if we don't have to.
2125                  */
2126                 if (info->bytes <= fs_info->sectorsize * 8) {
2127                         if (ctl->free_extents * 3 <= ctl->extents_thresh)
2128                                 return false;
2129                 } else {
2130                         return false;
2131                 }
2132         }
2133
2134         /*
2135          * The original block groups from mkfs can be really small, like 8
2136          * megabytes, so don't bother with a bitmap for those entries.  However
2137          * some block groups can be smaller than what a bitmap would cover but
2138          * are still large enough that they could overflow the 32k memory limit,
2139          * so allow those block groups to still be allowed to have a bitmap
2140          * entry.
2141          */
2142         if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->length)
2143                 return false;
2144
2145         return true;
2146 }
2147
2148 static const struct btrfs_free_space_op free_space_op = {
2149         .use_bitmap             = use_bitmap,
2150 };
2151
2152 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
2153                               struct btrfs_free_space *info)
2154 {
2155         struct btrfs_free_space *bitmap_info;
2156         struct btrfs_block_group *block_group = NULL;
2157         int added = 0;
2158         u64 bytes, offset, bytes_added;
2159         enum btrfs_trim_state trim_state;
2160         int ret;
2161
2162         bytes = info->bytes;
2163         offset = info->offset;
2164         trim_state = info->trim_state;
2165
2166         if (!ctl->op->use_bitmap(ctl, info))
2167                 return 0;
2168
2169         if (ctl->op == &free_space_op)
2170                 block_group = ctl->private;
2171 again:
2172         /*
2173          * Since we link bitmaps right into the cluster we need to see if we
2174          * have a cluster here, and if so and it has our bitmap we need to add
2175          * the free space to that bitmap.
2176          */
2177         if (block_group && !list_empty(&block_group->cluster_list)) {
2178                 struct btrfs_free_cluster *cluster;
2179                 struct rb_node *node;
2180                 struct btrfs_free_space *entry;
2181
2182                 cluster = list_entry(block_group->cluster_list.next,
2183                                      struct btrfs_free_cluster,
2184                                      block_group_list);
2185                 spin_lock(&cluster->lock);
2186                 node = rb_first(&cluster->root);
2187                 if (!node) {
2188                         spin_unlock(&cluster->lock);
2189                         goto no_cluster_bitmap;
2190                 }
2191
2192                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2193                 if (!entry->bitmap) {
2194                         spin_unlock(&cluster->lock);
2195                         goto no_cluster_bitmap;
2196                 }
2197
2198                 if (entry->offset == offset_to_bitmap(ctl, offset)) {
2199                         bytes_added = add_bytes_to_bitmap(ctl, entry, offset,
2200                                                           bytes, trim_state);
2201                         bytes -= bytes_added;
2202                         offset += bytes_added;
2203                 }
2204                 spin_unlock(&cluster->lock);
2205                 if (!bytes) {
2206                         ret = 1;
2207                         goto out;
2208                 }
2209         }
2210
2211 no_cluster_bitmap:
2212         bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2213                                          1, 0);
2214         if (!bitmap_info) {
2215                 ASSERT(added == 0);
2216                 goto new_bitmap;
2217         }
2218
2219         bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes,
2220                                           trim_state);
2221         bytes -= bytes_added;
2222         offset += bytes_added;
2223         added = 0;
2224
2225         if (!bytes) {
2226                 ret = 1;
2227                 goto out;
2228         } else
2229                 goto again;
2230
2231 new_bitmap:
2232         if (info && info->bitmap) {
2233                 add_new_bitmap(ctl, info, offset);
2234                 added = 1;
2235                 info = NULL;
2236                 goto again;
2237         } else {
2238                 spin_unlock(&ctl->tree_lock);
2239
2240                 /* no pre-allocated info, allocate a new one */
2241                 if (!info) {
2242                         info = kmem_cache_zalloc(btrfs_free_space_cachep,
2243                                                  GFP_NOFS);
2244                         if (!info) {
2245                                 spin_lock(&ctl->tree_lock);
2246                                 ret = -ENOMEM;
2247                                 goto out;
2248                         }
2249                 }
2250
2251                 /* allocate the bitmap */
2252                 info->bitmap = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep,
2253                                                  GFP_NOFS);
2254                 info->trim_state = BTRFS_TRIM_STATE_TRIMMED;
2255                 spin_lock(&ctl->tree_lock);
2256                 if (!info->bitmap) {
2257                         ret = -ENOMEM;
2258                         goto out;
2259                 }
2260                 goto again;
2261         }
2262
2263 out:
2264         if (info) {
2265                 if (info->bitmap)
2266                         kmem_cache_free(btrfs_free_space_bitmap_cachep,
2267                                         info->bitmap);
2268                 kmem_cache_free(btrfs_free_space_cachep, info);
2269         }
2270
2271         return ret;
2272 }
2273
2274 /*
2275  * Free space merging rules:
2276  *  1) Merge trimmed areas together
2277  *  2) Let untrimmed areas coalesce with trimmed areas
2278  *  3) Always pull neighboring regions from bitmaps
2279  *
2280  * The above rules are for when we merge free space based on btrfs_trim_state.
2281  * Rules 2 and 3 are subtle because they are suboptimal, but are done for the
2282  * same reason: to promote larger extent regions which makes life easier for
2283  * find_free_extent().  Rule 2 enables coalescing based on the common path
2284  * being returning free space from btrfs_finish_extent_commit().  So when free
2285  * space is trimmed, it will prevent aggregating trimmed new region and
2286  * untrimmed regions in the rb_tree.  Rule 3 is purely to obtain larger extents
2287  * and provide find_free_extent() with the largest extents possible hoping for
2288  * the reuse path.
2289  */
2290 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
2291                           struct btrfs_free_space *info, bool update_stat)
2292 {
2293         struct btrfs_free_space *left_info = NULL;
2294         struct btrfs_free_space *right_info;
2295         bool merged = false;
2296         u64 offset = info->offset;
2297         u64 bytes = info->bytes;
2298         const bool is_trimmed = btrfs_free_space_trimmed(info);
2299
2300         /*
2301          * first we want to see if there is free space adjacent to the range we
2302          * are adding, if there is remove that struct and add a new one to
2303          * cover the entire range
2304          */
2305         right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
2306         if (right_info && rb_prev(&right_info->offset_index))
2307                 left_info = rb_entry(rb_prev(&right_info->offset_index),
2308                                      struct btrfs_free_space, offset_index);
2309         else if (!right_info)
2310                 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
2311
2312         /* See try_merge_free_space() comment. */
2313         if (right_info && !right_info->bitmap &&
2314             (!is_trimmed || btrfs_free_space_trimmed(right_info))) {
2315                 if (update_stat)
2316                         unlink_free_space(ctl, right_info);
2317                 else
2318                         __unlink_free_space(ctl, right_info);
2319                 info->bytes += right_info->bytes;
2320                 kmem_cache_free(btrfs_free_space_cachep, right_info);
2321                 merged = true;
2322         }
2323
2324         /* See try_merge_free_space() comment. */
2325         if (left_info && !left_info->bitmap &&
2326             left_info->offset + left_info->bytes == offset &&
2327             (!is_trimmed || btrfs_free_space_trimmed(left_info))) {
2328                 if (update_stat)
2329                         unlink_free_space(ctl, left_info);
2330                 else
2331                         __unlink_free_space(ctl, left_info);
2332                 info->offset = left_info->offset;
2333                 info->bytes += left_info->bytes;
2334                 kmem_cache_free(btrfs_free_space_cachep, left_info);
2335                 merged = true;
2336         }
2337
2338         return merged;
2339 }
2340
2341 static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl,
2342                                      struct btrfs_free_space *info,
2343                                      bool update_stat)
2344 {
2345         struct btrfs_free_space *bitmap;
2346         unsigned long i;
2347         unsigned long j;
2348         const u64 end = info->offset + info->bytes;
2349         const u64 bitmap_offset = offset_to_bitmap(ctl, end);
2350         u64 bytes;
2351
2352         bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2353         if (!bitmap)
2354                 return false;
2355
2356         i = offset_to_bit(bitmap->offset, ctl->unit, end);
2357         j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i);
2358         if (j == i)
2359                 return false;
2360         bytes = (j - i) * ctl->unit;
2361         info->bytes += bytes;
2362
2363         /* See try_merge_free_space() comment. */
2364         if (!btrfs_free_space_trimmed(bitmap))
2365                 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2366
2367         if (update_stat)
2368                 bitmap_clear_bits(ctl, bitmap, end, bytes);
2369         else
2370                 __bitmap_clear_bits(ctl, bitmap, end, bytes);
2371
2372         if (!bitmap->bytes)
2373                 free_bitmap(ctl, bitmap);
2374
2375         return true;
2376 }
2377
2378 static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl,
2379                                        struct btrfs_free_space *info,
2380                                        bool update_stat)
2381 {
2382         struct btrfs_free_space *bitmap;
2383         u64 bitmap_offset;
2384         unsigned long i;
2385         unsigned long j;
2386         unsigned long prev_j;
2387         u64 bytes;
2388
2389         bitmap_offset = offset_to_bitmap(ctl, info->offset);
2390         /* If we're on a boundary, try the previous logical bitmap. */
2391         if (bitmap_offset == info->offset) {
2392                 if (info->offset == 0)
2393                         return false;
2394                 bitmap_offset = offset_to_bitmap(ctl, info->offset - 1);
2395         }
2396
2397         bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2398         if (!bitmap)
2399                 return false;
2400
2401         i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1;
2402         j = 0;
2403         prev_j = (unsigned long)-1;
2404         for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) {
2405                 if (j > i)
2406                         break;
2407                 prev_j = j;
2408         }
2409         if (prev_j == i)
2410                 return false;
2411
2412         if (prev_j == (unsigned long)-1)
2413                 bytes = (i + 1) * ctl->unit;
2414         else
2415                 bytes = (i - prev_j) * ctl->unit;
2416
2417         info->offset -= bytes;
2418         info->bytes += bytes;
2419
2420         /* See try_merge_free_space() comment. */
2421         if (!btrfs_free_space_trimmed(bitmap))
2422                 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2423
2424         if (update_stat)
2425                 bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2426         else
2427                 __bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2428
2429         if (!bitmap->bytes)
2430                 free_bitmap(ctl, bitmap);
2431
2432         return true;
2433 }
2434
2435 /*
2436  * We prefer always to allocate from extent entries, both for clustered and
2437  * non-clustered allocation requests. So when attempting to add a new extent
2438  * entry, try to see if there's adjacent free space in bitmap entries, and if
2439  * there is, migrate that space from the bitmaps to the extent.
2440  * Like this we get better chances of satisfying space allocation requests
2441  * because we attempt to satisfy them based on a single cache entry, and never
2442  * on 2 or more entries - even if the entries represent a contiguous free space
2443  * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
2444  * ends).
2445  */
2446 static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl,
2447                               struct btrfs_free_space *info,
2448                               bool update_stat)
2449 {
2450         /*
2451          * Only work with disconnected entries, as we can change their offset,
2452          * and must be extent entries.
2453          */
2454         ASSERT(!info->bitmap);
2455         ASSERT(RB_EMPTY_NODE(&info->offset_index));
2456
2457         if (ctl->total_bitmaps > 0) {
2458                 bool stole_end;
2459                 bool stole_front = false;
2460
2461                 stole_end = steal_from_bitmap_to_end(ctl, info, update_stat);
2462                 if (ctl->total_bitmaps > 0)
2463                         stole_front = steal_from_bitmap_to_front(ctl, info,
2464                                                                  update_stat);
2465
2466                 if (stole_end || stole_front)
2467                         try_merge_free_space(ctl, info, update_stat);
2468         }
2469 }
2470
2471 int __btrfs_add_free_space(struct btrfs_fs_info *fs_info,
2472                            struct btrfs_free_space_ctl *ctl,
2473                            u64 offset, u64 bytes,
2474                            enum btrfs_trim_state trim_state)
2475 {
2476         struct btrfs_block_group *block_group = ctl->private;
2477         struct btrfs_free_space *info;
2478         int ret = 0;
2479         u64 filter_bytes = bytes;
2480
2481         ASSERT(!btrfs_is_zoned(fs_info));
2482
2483         info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2484         if (!info)
2485                 return -ENOMEM;
2486
2487         info->offset = offset;
2488         info->bytes = bytes;
2489         info->trim_state = trim_state;
2490         RB_CLEAR_NODE(&info->offset_index);
2491
2492         spin_lock(&ctl->tree_lock);
2493
2494         if (try_merge_free_space(ctl, info, true))
2495                 goto link;
2496
2497         /*
2498          * There was no extent directly to the left or right of this new
2499          * extent then we know we're going to have to allocate a new extent, so
2500          * before we do that see if we need to drop this into a bitmap
2501          */
2502         ret = insert_into_bitmap(ctl, info);
2503         if (ret < 0) {
2504                 goto out;
2505         } else if (ret) {
2506                 ret = 0;
2507                 goto out;
2508         }
2509 link:
2510         /*
2511          * Only steal free space from adjacent bitmaps if we're sure we're not
2512          * going to add the new free space to existing bitmap entries - because
2513          * that would mean unnecessary work that would be reverted. Therefore
2514          * attempt to steal space from bitmaps if we're adding an extent entry.
2515          */
2516         steal_from_bitmap(ctl, info, true);
2517
2518         filter_bytes = max(filter_bytes, info->bytes);
2519
2520         ret = link_free_space(ctl, info);
2521         if (ret)
2522                 kmem_cache_free(btrfs_free_space_cachep, info);
2523 out:
2524         btrfs_discard_update_discardable(block_group);
2525         spin_unlock(&ctl->tree_lock);
2526
2527         if (ret) {
2528                 btrfs_crit(fs_info, "unable to add free space :%d", ret);
2529                 ASSERT(ret != -EEXIST);
2530         }
2531
2532         if (trim_state != BTRFS_TRIM_STATE_TRIMMED) {
2533                 btrfs_discard_check_filter(block_group, filter_bytes);
2534                 btrfs_discard_queue_work(&fs_info->discard_ctl, block_group);
2535         }
2536
2537         return ret;
2538 }
2539
2540 static int __btrfs_add_free_space_zoned(struct btrfs_block_group *block_group,
2541                                         u64 bytenr, u64 size, bool used)
2542 {
2543         struct btrfs_fs_info *fs_info = block_group->fs_info;
2544         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2545         u64 offset = bytenr - block_group->start;
2546         u64 to_free, to_unusable;
2547         const int bg_reclaim_threshold = READ_ONCE(fs_info->bg_reclaim_threshold);
2548
2549         spin_lock(&ctl->tree_lock);
2550         if (!used)
2551                 to_free = size;
2552         else if (offset >= block_group->alloc_offset)
2553                 to_free = size;
2554         else if (offset + size <= block_group->alloc_offset)
2555                 to_free = 0;
2556         else
2557                 to_free = offset + size - block_group->alloc_offset;
2558         to_unusable = size - to_free;
2559
2560         ctl->free_space += to_free;
2561         /*
2562          * If the block group is read-only, we should account freed space into
2563          * bytes_readonly.
2564          */
2565         if (!block_group->ro)
2566                 block_group->zone_unusable += to_unusable;
2567         spin_unlock(&ctl->tree_lock);
2568         if (!used) {
2569                 spin_lock(&block_group->lock);
2570                 block_group->alloc_offset -= size;
2571                 spin_unlock(&block_group->lock);
2572         }
2573
2574         /* All the region is now unusable. Mark it as unused and reclaim */
2575         if (block_group->zone_unusable == block_group->length) {
2576                 btrfs_mark_bg_unused(block_group);
2577         } else if (bg_reclaim_threshold &&
2578                    block_group->zone_unusable >=
2579                    div_factor_fine(block_group->length, bg_reclaim_threshold)) {
2580                 btrfs_mark_bg_to_reclaim(block_group);
2581         }
2582
2583         return 0;
2584 }
2585
2586 int btrfs_add_free_space(struct btrfs_block_group *block_group,
2587                          u64 bytenr, u64 size)
2588 {
2589         enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2590
2591         if (btrfs_is_zoned(block_group->fs_info))
2592                 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2593                                                     true);
2594
2595         if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC))
2596                 trim_state = BTRFS_TRIM_STATE_TRIMMED;
2597
2598         return __btrfs_add_free_space(block_group->fs_info,
2599                                       block_group->free_space_ctl,
2600                                       bytenr, size, trim_state);
2601 }
2602
2603 int btrfs_add_free_space_unused(struct btrfs_block_group *block_group,
2604                                 u64 bytenr, u64 size)
2605 {
2606         if (btrfs_is_zoned(block_group->fs_info))
2607                 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2608                                                     false);
2609
2610         return btrfs_add_free_space(block_group, bytenr, size);
2611 }
2612
2613 /*
2614  * This is a subtle distinction because when adding free space back in general,
2615  * we want it to be added as untrimmed for async. But in the case where we add
2616  * it on loading of a block group, we want to consider it trimmed.
2617  */
2618 int btrfs_add_free_space_async_trimmed(struct btrfs_block_group *block_group,
2619                                        u64 bytenr, u64 size)
2620 {
2621         enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2622
2623         if (btrfs_is_zoned(block_group->fs_info))
2624                 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2625                                                     true);
2626
2627         if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC) ||
2628             btrfs_test_opt(block_group->fs_info, DISCARD_ASYNC))
2629                 trim_state = BTRFS_TRIM_STATE_TRIMMED;
2630
2631         return __btrfs_add_free_space(block_group->fs_info,
2632                                       block_group->free_space_ctl,
2633                                       bytenr, size, trim_state);
2634 }
2635
2636 int btrfs_remove_free_space(struct btrfs_block_group *block_group,
2637                             u64 offset, u64 bytes)
2638 {
2639         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2640         struct btrfs_free_space *info;
2641         int ret;
2642         bool re_search = false;
2643
2644         if (btrfs_is_zoned(block_group->fs_info)) {
2645                 /*
2646                  * This can happen with conventional zones when replaying log.
2647                  * Since the allocation info of tree-log nodes are not recorded
2648                  * to the extent-tree, calculate_alloc_pointer() failed to
2649                  * advance the allocation pointer after last allocated tree log
2650                  * node blocks.
2651                  *
2652                  * This function is called from
2653                  * btrfs_pin_extent_for_log_replay() when replaying the log.
2654                  * Advance the pointer not to overwrite the tree-log nodes.
2655                  */
2656                 if (block_group->start + block_group->alloc_offset <
2657                     offset + bytes) {
2658                         block_group->alloc_offset =
2659                                 offset + bytes - block_group->start;
2660                 }
2661                 return 0;
2662         }
2663
2664         spin_lock(&ctl->tree_lock);
2665
2666 again:
2667         ret = 0;
2668         if (!bytes)
2669                 goto out_lock;
2670
2671         info = tree_search_offset(ctl, offset, 0, 0);
2672         if (!info) {
2673                 /*
2674                  * oops didn't find an extent that matched the space we wanted
2675                  * to remove, look for a bitmap instead
2676                  */
2677                 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2678                                           1, 0);
2679                 if (!info) {
2680                         /*
2681                          * If we found a partial bit of our free space in a
2682                          * bitmap but then couldn't find the other part this may
2683                          * be a problem, so WARN about it.
2684                          */
2685                         WARN_ON(re_search);
2686                         goto out_lock;
2687                 }
2688         }
2689
2690         re_search = false;
2691         if (!info->bitmap) {
2692                 unlink_free_space(ctl, info);
2693                 if (offset == info->offset) {
2694                         u64 to_free = min(bytes, info->bytes);
2695
2696                         info->bytes -= to_free;
2697                         info->offset += to_free;
2698                         if (info->bytes) {
2699                                 ret = link_free_space(ctl, info);
2700                                 WARN_ON(ret);
2701                         } else {
2702                                 kmem_cache_free(btrfs_free_space_cachep, info);
2703                         }
2704
2705                         offset += to_free;
2706                         bytes -= to_free;
2707                         goto again;
2708                 } else {
2709                         u64 old_end = info->bytes + info->offset;
2710
2711                         info->bytes = offset - info->offset;
2712                         ret = link_free_space(ctl, info);
2713                         WARN_ON(ret);
2714                         if (ret)
2715                                 goto out_lock;
2716
2717                         /* Not enough bytes in this entry to satisfy us */
2718                         if (old_end < offset + bytes) {
2719                                 bytes -= old_end - offset;
2720                                 offset = old_end;
2721                                 goto again;
2722                         } else if (old_end == offset + bytes) {
2723                                 /* all done */
2724                                 goto out_lock;
2725                         }
2726                         spin_unlock(&ctl->tree_lock);
2727
2728                         ret = __btrfs_add_free_space(block_group->fs_info, ctl,
2729                                                      offset + bytes,
2730                                                      old_end - (offset + bytes),
2731                                                      info->trim_state);
2732                         WARN_ON(ret);
2733                         goto out;
2734                 }
2735         }
2736
2737         ret = remove_from_bitmap(ctl, info, &offset, &bytes);
2738         if (ret == -EAGAIN) {
2739                 re_search = true;
2740                 goto again;
2741         }
2742 out_lock:
2743         btrfs_discard_update_discardable(block_group);
2744         spin_unlock(&ctl->tree_lock);
2745 out:
2746         return ret;
2747 }
2748
2749 void btrfs_dump_free_space(struct btrfs_block_group *block_group,
2750                            u64 bytes)
2751 {
2752         struct btrfs_fs_info *fs_info = block_group->fs_info;
2753         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2754         struct btrfs_free_space *info;
2755         struct rb_node *n;
2756         int count = 0;
2757
2758         /*
2759          * Zoned btrfs does not use free space tree and cluster. Just print
2760          * out the free space after the allocation offset.
2761          */
2762         if (btrfs_is_zoned(fs_info)) {
2763                 btrfs_info(fs_info, "free space %llu",
2764                            block_group->length - block_group->alloc_offset);
2765                 return;
2766         }
2767
2768         spin_lock(&ctl->tree_lock);
2769         for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2770                 info = rb_entry(n, struct btrfs_free_space, offset_index);
2771                 if (info->bytes >= bytes && !block_group->ro)
2772                         count++;
2773                 btrfs_crit(fs_info, "entry offset %llu, bytes %llu, bitmap %s",
2774                            info->offset, info->bytes,
2775                        (info->bitmap) ? "yes" : "no");
2776         }
2777         spin_unlock(&ctl->tree_lock);
2778         btrfs_info(fs_info, "block group has cluster?: %s",
2779                list_empty(&block_group->cluster_list) ? "no" : "yes");
2780         btrfs_info(fs_info,
2781                    "%d blocks of free space at or bigger than bytes is", count);
2782 }
2783
2784 void btrfs_init_free_space_ctl(struct btrfs_block_group *block_group,
2785                                struct btrfs_free_space_ctl *ctl)
2786 {
2787         struct btrfs_fs_info *fs_info = block_group->fs_info;
2788
2789         spin_lock_init(&ctl->tree_lock);
2790         ctl->unit = fs_info->sectorsize;
2791         ctl->start = block_group->start;
2792         ctl->private = block_group;
2793         ctl->op = &free_space_op;
2794         INIT_LIST_HEAD(&ctl->trimming_ranges);
2795         mutex_init(&ctl->cache_writeout_mutex);
2796
2797         /*
2798          * we only want to have 32k of ram per block group for keeping
2799          * track of free space, and if we pass 1/2 of that we want to
2800          * start converting things over to using bitmaps
2801          */
2802         ctl->extents_thresh = (SZ_32K / 2) / sizeof(struct btrfs_free_space);
2803 }
2804
2805 /*
2806  * for a given cluster, put all of its extents back into the free
2807  * space cache.  If the block group passed doesn't match the block group
2808  * pointed to by the cluster, someone else raced in and freed the
2809  * cluster already.  In that case, we just return without changing anything
2810  */
2811 static void __btrfs_return_cluster_to_free_space(
2812                              struct btrfs_block_group *block_group,
2813                              struct btrfs_free_cluster *cluster)
2814 {
2815         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2816         struct btrfs_free_space *entry;
2817         struct rb_node *node;
2818
2819         spin_lock(&cluster->lock);
2820         if (cluster->block_group != block_group) {
2821                 spin_unlock(&cluster->lock);
2822                 return;
2823         }
2824
2825         cluster->block_group = NULL;
2826         cluster->window_start = 0;
2827         list_del_init(&cluster->block_group_list);
2828
2829         node = rb_first(&cluster->root);
2830         while (node) {
2831                 bool bitmap;
2832
2833                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2834                 node = rb_next(&entry->offset_index);
2835                 rb_erase(&entry->offset_index, &cluster->root);
2836                 RB_CLEAR_NODE(&entry->offset_index);
2837
2838                 bitmap = (entry->bitmap != NULL);
2839                 if (!bitmap) {
2840                         /* Merging treats extents as if they were new */
2841                         if (!btrfs_free_space_trimmed(entry)) {
2842                                 ctl->discardable_extents[BTRFS_STAT_CURR]--;
2843                                 ctl->discardable_bytes[BTRFS_STAT_CURR] -=
2844                                         entry->bytes;
2845                         }
2846
2847                         try_merge_free_space(ctl, entry, false);
2848                         steal_from_bitmap(ctl, entry, false);
2849
2850                         /* As we insert directly, update these statistics */
2851                         if (!btrfs_free_space_trimmed(entry)) {
2852                                 ctl->discardable_extents[BTRFS_STAT_CURR]++;
2853                                 ctl->discardable_bytes[BTRFS_STAT_CURR] +=
2854                                         entry->bytes;
2855                         }
2856                 }
2857                 tree_insert_offset(&ctl->free_space_offset,
2858                                    entry->offset, &entry->offset_index, bitmap);
2859         }
2860         cluster->root = RB_ROOT;
2861         spin_unlock(&cluster->lock);
2862         btrfs_put_block_group(block_group);
2863 }
2864
2865 static void __btrfs_remove_free_space_cache_locked(
2866                                 struct btrfs_free_space_ctl *ctl)
2867 {
2868         struct btrfs_free_space *info;
2869         struct rb_node *node;
2870
2871         while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2872                 info = rb_entry(node, struct btrfs_free_space, offset_index);
2873                 if (!info->bitmap) {
2874                         unlink_free_space(ctl, info);
2875                         kmem_cache_free(btrfs_free_space_cachep, info);
2876                 } else {
2877                         free_bitmap(ctl, info);
2878                 }
2879
2880                 cond_resched_lock(&ctl->tree_lock);
2881         }
2882 }
2883
2884 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2885 {
2886         spin_lock(&ctl->tree_lock);
2887         __btrfs_remove_free_space_cache_locked(ctl);
2888         if (ctl->private)
2889                 btrfs_discard_update_discardable(ctl->private);
2890         spin_unlock(&ctl->tree_lock);
2891 }
2892
2893 void btrfs_remove_free_space_cache(struct btrfs_block_group *block_group)
2894 {
2895         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2896         struct btrfs_free_cluster *cluster;
2897         struct list_head *head;
2898
2899         spin_lock(&ctl->tree_lock);
2900         while ((head = block_group->cluster_list.next) !=
2901                &block_group->cluster_list) {
2902                 cluster = list_entry(head, struct btrfs_free_cluster,
2903                                      block_group_list);
2904
2905                 WARN_ON(cluster->block_group != block_group);
2906                 __btrfs_return_cluster_to_free_space(block_group, cluster);
2907
2908                 cond_resched_lock(&ctl->tree_lock);
2909         }
2910         __btrfs_remove_free_space_cache_locked(ctl);
2911         btrfs_discard_update_discardable(block_group);
2912         spin_unlock(&ctl->tree_lock);
2913
2914 }
2915
2916 /**
2917  * btrfs_is_free_space_trimmed - see if everything is trimmed
2918  * @block_group: block_group of interest
2919  *
2920  * Walk @block_group's free space rb_tree to determine if everything is trimmed.
2921  */
2922 bool btrfs_is_free_space_trimmed(struct btrfs_block_group *block_group)
2923 {
2924         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2925         struct btrfs_free_space *info;
2926         struct rb_node *node;
2927         bool ret = true;
2928
2929         spin_lock(&ctl->tree_lock);
2930         node = rb_first(&ctl->free_space_offset);
2931
2932         while (node) {
2933                 info = rb_entry(node, struct btrfs_free_space, offset_index);
2934
2935                 if (!btrfs_free_space_trimmed(info)) {
2936                         ret = false;
2937                         break;
2938                 }
2939
2940                 node = rb_next(node);
2941         }
2942
2943         spin_unlock(&ctl->tree_lock);
2944         return ret;
2945 }
2946
2947 u64 btrfs_find_space_for_alloc(struct btrfs_block_group *block_group,
2948                                u64 offset, u64 bytes, u64 empty_size,
2949                                u64 *max_extent_size)
2950 {
2951         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2952         struct btrfs_discard_ctl *discard_ctl =
2953                                         &block_group->fs_info->discard_ctl;
2954         struct btrfs_free_space *entry = NULL;
2955         u64 bytes_search = bytes + empty_size;
2956         u64 ret = 0;
2957         u64 align_gap = 0;
2958         u64 align_gap_len = 0;
2959         enum btrfs_trim_state align_gap_trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2960
2961         ASSERT(!btrfs_is_zoned(block_group->fs_info));
2962
2963         spin_lock(&ctl->tree_lock);
2964         entry = find_free_space(ctl, &offset, &bytes_search,
2965                                 block_group->full_stripe_len, max_extent_size);
2966         if (!entry)
2967                 goto out;
2968
2969         ret = offset;
2970         if (entry->bitmap) {
2971                 bitmap_clear_bits(ctl, entry, offset, bytes);
2972
2973                 if (!btrfs_free_space_trimmed(entry))
2974                         atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
2975
2976                 if (!entry->bytes)
2977                         free_bitmap(ctl, entry);
2978         } else {
2979                 unlink_free_space(ctl, entry);
2980                 align_gap_len = offset - entry->offset;
2981                 align_gap = entry->offset;
2982                 align_gap_trim_state = entry->trim_state;
2983
2984                 if (!btrfs_free_space_trimmed(entry))
2985                         atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
2986
2987                 entry->offset = offset + bytes;
2988                 WARN_ON(entry->bytes < bytes + align_gap_len);
2989
2990                 entry->bytes -= bytes + align_gap_len;
2991                 if (!entry->bytes)
2992                         kmem_cache_free(btrfs_free_space_cachep, entry);
2993                 else
2994                         link_free_space(ctl, entry);
2995         }
2996 out:
2997         btrfs_discard_update_discardable(block_group);
2998         spin_unlock(&ctl->tree_lock);
2999
3000         if (align_gap_len)
3001                 __btrfs_add_free_space(block_group->fs_info, ctl,
3002                                        align_gap, align_gap_len,
3003                                        align_gap_trim_state);
3004         return ret;
3005 }
3006
3007 /*
3008  * given a cluster, put all of its extents back into the free space
3009  * cache.  If a block group is passed, this function will only free
3010  * a cluster that belongs to the passed block group.
3011  *
3012  * Otherwise, it'll get a reference on the block group pointed to by the
3013  * cluster and remove the cluster from it.
3014  */
3015 void btrfs_return_cluster_to_free_space(
3016                                struct btrfs_block_group *block_group,
3017                                struct btrfs_free_cluster *cluster)
3018 {
3019         struct btrfs_free_space_ctl *ctl;
3020
3021         /* first, get a safe pointer to the block group */
3022         spin_lock(&cluster->lock);
3023         if (!block_group) {
3024                 block_group = cluster->block_group;
3025                 if (!block_group) {
3026                         spin_unlock(&cluster->lock);
3027                         return;
3028                 }
3029         } else if (cluster->block_group != block_group) {
3030                 /* someone else has already freed it don't redo their work */
3031                 spin_unlock(&cluster->lock);
3032                 return;
3033         }
3034         btrfs_get_block_group(block_group);
3035         spin_unlock(&cluster->lock);
3036
3037         ctl = block_group->free_space_ctl;
3038
3039         /* now return any extents the cluster had on it */
3040         spin_lock(&ctl->tree_lock);
3041         __btrfs_return_cluster_to_free_space(block_group, cluster);
3042         spin_unlock(&ctl->tree_lock);
3043
3044         btrfs_discard_queue_work(&block_group->fs_info->discard_ctl, block_group);
3045
3046         /* finally drop our ref */
3047         btrfs_put_block_group(block_group);
3048 }
3049
3050 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group *block_group,
3051                                    struct btrfs_free_cluster *cluster,
3052                                    struct btrfs_free_space *entry,
3053                                    u64 bytes, u64 min_start,
3054                                    u64 *max_extent_size)
3055 {
3056         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3057         int err;
3058         u64 search_start = cluster->window_start;
3059         u64 search_bytes = bytes;
3060         u64 ret = 0;
3061
3062         search_start = min_start;
3063         search_bytes = bytes;
3064
3065         err = search_bitmap(ctl, entry, &search_start, &search_bytes, true);
3066         if (err) {
3067                 *max_extent_size = max(get_max_extent_size(entry),
3068                                        *max_extent_size);
3069                 return 0;
3070         }
3071
3072         ret = search_start;
3073         __bitmap_clear_bits(ctl, entry, ret, bytes);
3074
3075         return ret;
3076 }
3077
3078 /*
3079  * given a cluster, try to allocate 'bytes' from it, returns 0
3080  * if it couldn't find anything suitably large, or a logical disk offset
3081  * if things worked out
3082  */
3083 u64 btrfs_alloc_from_cluster(struct btrfs_block_group *block_group,
3084                              struct btrfs_free_cluster *cluster, u64 bytes,
3085                              u64 min_start, u64 *max_extent_size)
3086 {
3087         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3088         struct btrfs_discard_ctl *discard_ctl =
3089                                         &block_group->fs_info->discard_ctl;
3090         struct btrfs_free_space *entry = NULL;
3091         struct rb_node *node;
3092         u64 ret = 0;
3093
3094         ASSERT(!btrfs_is_zoned(block_group->fs_info));
3095
3096         spin_lock(&cluster->lock);
3097         if (bytes > cluster->max_size)
3098                 goto out;
3099
3100         if (cluster->block_group != block_group)
3101                 goto out;
3102
3103         node = rb_first(&cluster->root);
3104         if (!node)
3105                 goto out;
3106
3107         entry = rb_entry(node, struct btrfs_free_space, offset_index);
3108         while (1) {
3109                 if (entry->bytes < bytes)
3110                         *max_extent_size = max(get_max_extent_size(entry),
3111                                                *max_extent_size);
3112
3113                 if (entry->bytes < bytes ||
3114                     (!entry->bitmap && entry->offset < min_start)) {
3115                         node = rb_next(&entry->offset_index);
3116                         if (!node)
3117                                 break;
3118                         entry = rb_entry(node, struct btrfs_free_space,
3119                                          offset_index);
3120                         continue;
3121                 }
3122
3123                 if (entry->bitmap) {
3124                         ret = btrfs_alloc_from_bitmap(block_group,
3125                                                       cluster, entry, bytes,
3126                                                       cluster->window_start,
3127                                                       max_extent_size);
3128                         if (ret == 0) {
3129                                 node = rb_next(&entry->offset_index);
3130                                 if (!node)
3131                                         break;
3132                                 entry = rb_entry(node, struct btrfs_free_space,
3133                                                  offset_index);
3134                                 continue;
3135                         }
3136                         cluster->window_start += bytes;
3137                 } else {
3138                         ret = entry->offset;
3139
3140                         entry->offset += bytes;
3141                         entry->bytes -= bytes;
3142                 }
3143
3144                 break;
3145         }
3146 out:
3147         spin_unlock(&cluster->lock);
3148
3149         if (!ret)
3150                 return 0;
3151
3152         spin_lock(&ctl->tree_lock);
3153
3154         if (!btrfs_free_space_trimmed(entry))
3155                 atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3156
3157         ctl->free_space -= bytes;
3158         if (!entry->bitmap && !btrfs_free_space_trimmed(entry))
3159                 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
3160
3161         spin_lock(&cluster->lock);
3162         if (entry->bytes == 0) {
3163                 rb_erase(&entry->offset_index, &cluster->root);
3164                 ctl->free_extents--;
3165                 if (entry->bitmap) {
3166                         kmem_cache_free(btrfs_free_space_bitmap_cachep,
3167                                         entry->bitmap);
3168                         ctl->total_bitmaps--;
3169                         recalculate_thresholds(ctl);
3170                 } else if (!btrfs_free_space_trimmed(entry)) {
3171                         ctl->discardable_extents[BTRFS_STAT_CURR]--;
3172                 }
3173                 kmem_cache_free(btrfs_free_space_cachep, entry);
3174         }
3175
3176         spin_unlock(&cluster->lock);
3177         spin_unlock(&ctl->tree_lock);
3178
3179         return ret;
3180 }
3181
3182 static int btrfs_bitmap_cluster(struct btrfs_block_group *block_group,
3183                                 struct btrfs_free_space *entry,
3184                                 struct btrfs_free_cluster *cluster,
3185                                 u64 offset, u64 bytes,
3186                                 u64 cont1_bytes, u64 min_bytes)
3187 {
3188         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3189         unsigned long next_zero;
3190         unsigned long i;
3191         unsigned long want_bits;
3192         unsigned long min_bits;
3193         unsigned long found_bits;
3194         unsigned long max_bits = 0;
3195         unsigned long start = 0;
3196         unsigned long total_found = 0;
3197         int ret;
3198
3199         i = offset_to_bit(entry->offset, ctl->unit,
3200                           max_t(u64, offset, entry->offset));
3201         want_bits = bytes_to_bits(bytes, ctl->unit);
3202         min_bits = bytes_to_bits(min_bytes, ctl->unit);
3203
3204         /*
3205          * Don't bother looking for a cluster in this bitmap if it's heavily
3206          * fragmented.
3207          */
3208         if (entry->max_extent_size &&
3209             entry->max_extent_size < cont1_bytes)
3210                 return -ENOSPC;
3211 again:
3212         found_bits = 0;
3213         for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
3214                 next_zero = find_next_zero_bit(entry->bitmap,
3215                                                BITS_PER_BITMAP, i);
3216                 if (next_zero - i >= min_bits) {
3217                         found_bits = next_zero - i;
3218                         if (found_bits > max_bits)
3219                                 max_bits = found_bits;
3220                         break;
3221                 }
3222                 if (next_zero - i > max_bits)
3223                         max_bits = next_zero - i;
3224                 i = next_zero;
3225         }
3226
3227         if (!found_bits) {
3228                 entry->max_extent_size = (u64)max_bits * ctl->unit;
3229                 return -ENOSPC;
3230         }
3231
3232         if (!total_found) {
3233                 start = i;
3234                 cluster->max_size = 0;
3235         }
3236
3237         total_found += found_bits;
3238
3239         if (cluster->max_size < found_bits * ctl->unit)
3240                 cluster->max_size = found_bits * ctl->unit;
3241
3242         if (total_found < want_bits || cluster->max_size < cont1_bytes) {
3243                 i = next_zero + 1;
3244                 goto again;
3245         }
3246
3247         cluster->window_start = start * ctl->unit + entry->offset;
3248         rb_erase(&entry->offset_index, &ctl->free_space_offset);
3249         ret = tree_insert_offset(&cluster->root, entry->offset,
3250                                  &entry->offset_index, 1);
3251         ASSERT(!ret); /* -EEXIST; Logic error */
3252
3253         trace_btrfs_setup_cluster(block_group, cluster,
3254                                   total_found * ctl->unit, 1);
3255         return 0;
3256 }
3257
3258 /*
3259  * This searches the block group for just extents to fill the cluster with.
3260  * Try to find a cluster with at least bytes total bytes, at least one
3261  * extent of cont1_bytes, and other clusters of at least min_bytes.
3262  */
3263 static noinline int
3264 setup_cluster_no_bitmap(struct btrfs_block_group *block_group,
3265                         struct btrfs_free_cluster *cluster,
3266                         struct list_head *bitmaps, u64 offset, u64 bytes,
3267                         u64 cont1_bytes, u64 min_bytes)
3268 {
3269         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3270         struct btrfs_free_space *first = NULL;
3271         struct btrfs_free_space *entry = NULL;
3272         struct btrfs_free_space *last;
3273         struct rb_node *node;
3274         u64 window_free;
3275         u64 max_extent;
3276         u64 total_size = 0;
3277
3278         entry = tree_search_offset(ctl, offset, 0, 1);
3279         if (!entry)
3280                 return -ENOSPC;
3281
3282         /*
3283          * We don't want bitmaps, so just move along until we find a normal
3284          * extent entry.
3285          */
3286         while (entry->bitmap || entry->bytes < min_bytes) {
3287                 if (entry->bitmap && list_empty(&entry->list))
3288                         list_add_tail(&entry->list, bitmaps);
3289                 node = rb_next(&entry->offset_index);
3290                 if (!node)
3291                         return -ENOSPC;
3292                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3293         }
3294
3295         window_free = entry->bytes;
3296         max_extent = entry->bytes;
3297         first = entry;
3298         last = entry;
3299
3300         for (node = rb_next(&entry->offset_index); node;
3301              node = rb_next(&entry->offset_index)) {
3302                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3303
3304                 if (entry->bitmap) {
3305                         if (list_empty(&entry->list))
3306                                 list_add_tail(&entry->list, bitmaps);
3307                         continue;
3308                 }
3309
3310                 if (entry->bytes < min_bytes)
3311                         continue;
3312
3313                 last = entry;
3314                 window_free += entry->bytes;
3315                 if (entry->bytes > max_extent)
3316                         max_extent = entry->bytes;
3317         }
3318
3319         if (window_free < bytes || max_extent < cont1_bytes)
3320                 return -ENOSPC;
3321
3322         cluster->window_start = first->offset;
3323
3324         node = &first->offset_index;
3325
3326         /*
3327          * now we've found our entries, pull them out of the free space
3328          * cache and put them into the cluster rbtree
3329          */
3330         do {
3331                 int ret;
3332
3333                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3334                 node = rb_next(&entry->offset_index);
3335                 if (entry->bitmap || entry->bytes < min_bytes)
3336                         continue;
3337
3338                 rb_erase(&entry->offset_index, &ctl->free_space_offset);
3339                 ret = tree_insert_offset(&cluster->root, entry->offset,
3340                                          &entry->offset_index, 0);
3341                 total_size += entry->bytes;
3342                 ASSERT(!ret); /* -EEXIST; Logic error */
3343         } while (node && entry != last);
3344
3345         cluster->max_size = max_extent;
3346         trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
3347         return 0;
3348 }
3349
3350 /*
3351  * This specifically looks for bitmaps that may work in the cluster, we assume
3352  * that we have already failed to find extents that will work.
3353  */
3354 static noinline int
3355 setup_cluster_bitmap(struct btrfs_block_group *block_group,
3356                      struct btrfs_free_cluster *cluster,
3357                      struct list_head *bitmaps, u64 offset, u64 bytes,
3358                      u64 cont1_bytes, u64 min_bytes)
3359 {
3360         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3361         struct btrfs_free_space *entry = NULL;
3362         int ret = -ENOSPC;
3363         u64 bitmap_offset = offset_to_bitmap(ctl, offset);
3364
3365         if (ctl->total_bitmaps == 0)
3366                 return -ENOSPC;
3367
3368         /*
3369          * The bitmap that covers offset won't be in the list unless offset
3370          * is just its start offset.
3371          */
3372         if (!list_empty(bitmaps))
3373                 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
3374
3375         if (!entry || entry->offset != bitmap_offset) {
3376                 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
3377                 if (entry && list_empty(&entry->list))
3378                         list_add(&entry->list, bitmaps);
3379         }
3380
3381         list_for_each_entry(entry, bitmaps, list) {
3382                 if (entry->bytes < bytes)
3383                         continue;
3384                 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
3385                                            bytes, cont1_bytes, min_bytes);
3386                 if (!ret)
3387                         return 0;
3388         }
3389
3390         /*
3391          * The bitmaps list has all the bitmaps that record free space
3392          * starting after offset, so no more search is required.
3393          */
3394         return -ENOSPC;
3395 }
3396
3397 /*
3398  * here we try to find a cluster of blocks in a block group.  The goal
3399  * is to find at least bytes+empty_size.
3400  * We might not find them all in one contiguous area.
3401  *
3402  * returns zero and sets up cluster if things worked out, otherwise
3403  * it returns -enospc
3404  */
3405 int btrfs_find_space_cluster(struct btrfs_block_group *block_group,
3406                              struct btrfs_free_cluster *cluster,
3407                              u64 offset, u64 bytes, u64 empty_size)
3408 {
3409         struct btrfs_fs_info *fs_info = block_group->fs_info;
3410         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3411         struct btrfs_free_space *entry, *tmp;
3412         LIST_HEAD(bitmaps);
3413         u64 min_bytes;
3414         u64 cont1_bytes;
3415         int ret;
3416
3417         /*
3418          * Choose the minimum extent size we'll require for this
3419          * cluster.  For SSD_SPREAD, don't allow any fragmentation.
3420          * For metadata, allow allocates with smaller extents.  For
3421          * data, keep it dense.
3422          */
3423         if (btrfs_test_opt(fs_info, SSD_SPREAD)) {
3424                 cont1_bytes = min_bytes = bytes + empty_size;
3425         } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
3426                 cont1_bytes = bytes;
3427                 min_bytes = fs_info->sectorsize;
3428         } else {
3429                 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
3430                 min_bytes = fs_info->sectorsize;
3431         }
3432
3433         spin_lock(&ctl->tree_lock);
3434
3435         /*
3436          * If we know we don't have enough space to make a cluster don't even
3437          * bother doing all the work to try and find one.
3438          */
3439         if (ctl->free_space < bytes) {
3440                 spin_unlock(&ctl->tree_lock);
3441                 return -ENOSPC;
3442         }
3443
3444         spin_lock(&cluster->lock);
3445
3446         /* someone already found a cluster, hooray */
3447         if (cluster->block_group) {
3448                 ret = 0;
3449                 goto out;
3450         }
3451
3452         trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
3453                                  min_bytes);
3454
3455         ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
3456                                       bytes + empty_size,
3457                                       cont1_bytes, min_bytes);
3458         if (ret)
3459                 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
3460                                            offset, bytes + empty_size,
3461                                            cont1_bytes, min_bytes);
3462
3463         /* Clear our temporary list */
3464         list_for_each_entry_safe(entry, tmp, &bitmaps, list)
3465                 list_del_init(&entry->list);
3466
3467         if (!ret) {
3468                 btrfs_get_block_group(block_group);
3469                 list_add_tail(&cluster->block_group_list,
3470                               &block_group->cluster_list);
3471                 cluster->block_group = block_group;
3472         } else {
3473                 trace_btrfs_failed_cluster_setup(block_group);
3474         }
3475 out:
3476         spin_unlock(&cluster->lock);
3477         spin_unlock(&ctl->tree_lock);
3478
3479         return ret;
3480 }
3481
3482 /*
3483  * simple code to zero out a cluster
3484  */
3485 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
3486 {
3487         spin_lock_init(&cluster->lock);
3488         spin_lock_init(&cluster->refill_lock);
3489         cluster->root = RB_ROOT;
3490         cluster->max_size = 0;
3491         cluster->fragmented = false;
3492         INIT_LIST_HEAD(&cluster->block_group_list);
3493         cluster->block_group = NULL;
3494 }
3495
3496 static int do_trimming(struct btrfs_block_group *block_group,
3497                        u64 *total_trimmed, u64 start, u64 bytes,
3498                        u64 reserved_start, u64 reserved_bytes,
3499                        enum btrfs_trim_state reserved_trim_state,
3500                        struct btrfs_trim_range *trim_entry)
3501 {
3502         struct btrfs_space_info *space_info = block_group->space_info;
3503         struct btrfs_fs_info *fs_info = block_group->fs_info;
3504         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3505         int ret;
3506         int update = 0;
3507         const u64 end = start + bytes;
3508         const u64 reserved_end = reserved_start + reserved_bytes;
3509         enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3510         u64 trimmed = 0;
3511
3512         spin_lock(&space_info->lock);
3513         spin_lock(&block_group->lock);
3514         if (!block_group->ro) {
3515                 block_group->reserved += reserved_bytes;
3516                 space_info->bytes_reserved += reserved_bytes;
3517                 update = 1;
3518         }
3519         spin_unlock(&block_group->lock);
3520         spin_unlock(&space_info->lock);
3521
3522         ret = btrfs_discard_extent(fs_info, start, bytes, &trimmed);
3523         if (!ret) {
3524                 *total_trimmed += trimmed;
3525                 trim_state = BTRFS_TRIM_STATE_TRIMMED;
3526         }
3527
3528         mutex_lock(&ctl->cache_writeout_mutex);
3529         if (reserved_start < start)
3530                 __btrfs_add_free_space(fs_info, ctl, reserved_start,
3531                                        start - reserved_start,
3532                                        reserved_trim_state);
3533         if (start + bytes < reserved_start + reserved_bytes)
3534                 __btrfs_add_free_space(fs_info, ctl, end, reserved_end - end,
3535                                        reserved_trim_state);
3536         __btrfs_add_free_space(fs_info, ctl, start, bytes, trim_state);
3537         list_del(&trim_entry->list);
3538         mutex_unlock(&ctl->cache_writeout_mutex);
3539
3540         if (update) {
3541                 spin_lock(&space_info->lock);
3542                 spin_lock(&block_group->lock);
3543                 if (block_group->ro)
3544                         space_info->bytes_readonly += reserved_bytes;
3545                 block_group->reserved -= reserved_bytes;
3546                 space_info->bytes_reserved -= reserved_bytes;
3547                 spin_unlock(&block_group->lock);
3548                 spin_unlock(&space_info->lock);
3549         }
3550
3551         return ret;
3552 }
3553
3554 /*
3555  * If @async is set, then we will trim 1 region and return.
3556  */
3557 static int trim_no_bitmap(struct btrfs_block_group *block_group,
3558                           u64 *total_trimmed, u64 start, u64 end, u64 minlen,
3559                           bool async)
3560 {
3561         struct btrfs_discard_ctl *discard_ctl =
3562                                         &block_group->fs_info->discard_ctl;
3563         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3564         struct btrfs_free_space *entry;
3565         struct rb_node *node;
3566         int ret = 0;
3567         u64 extent_start;
3568         u64 extent_bytes;
3569         enum btrfs_trim_state extent_trim_state;
3570         u64 bytes;
3571         const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3572
3573         while (start < end) {
3574                 struct btrfs_trim_range trim_entry;
3575
3576                 mutex_lock(&ctl->cache_writeout_mutex);
3577                 spin_lock(&ctl->tree_lock);
3578
3579                 if (ctl->free_space < minlen)
3580                         goto out_unlock;
3581
3582                 entry = tree_search_offset(ctl, start, 0, 1);
3583                 if (!entry)
3584                         goto out_unlock;
3585
3586                 /* Skip bitmaps and if async, already trimmed entries */
3587                 while (entry->bitmap ||
3588                        (async && btrfs_free_space_trimmed(entry))) {
3589                         node = rb_next(&entry->offset_index);
3590                         if (!node)
3591                                 goto out_unlock;
3592                         entry = rb_entry(node, struct btrfs_free_space,
3593                                          offset_index);
3594                 }
3595
3596                 if (entry->offset >= end)
3597                         goto out_unlock;
3598
3599                 extent_start = entry->offset;
3600                 extent_bytes = entry->bytes;
3601                 extent_trim_state = entry->trim_state;
3602                 if (async) {
3603                         start = entry->offset;
3604                         bytes = entry->bytes;
3605                         if (bytes < minlen) {
3606                                 spin_unlock(&ctl->tree_lock);
3607                                 mutex_unlock(&ctl->cache_writeout_mutex);
3608                                 goto next;
3609                         }
3610                         unlink_free_space(ctl, entry);
3611                         /*
3612                          * Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
3613                          * If X < BTRFS_ASYNC_DISCARD_MIN_FILTER, we won't trim
3614                          * X when we come back around.  So trim it now.
3615                          */
3616                         if (max_discard_size &&
3617                             bytes >= (max_discard_size +
3618                                       BTRFS_ASYNC_DISCARD_MIN_FILTER)) {
3619                                 bytes = max_discard_size;
3620                                 extent_bytes = max_discard_size;
3621                                 entry->offset += max_discard_size;
3622                                 entry->bytes -= max_discard_size;
3623                                 link_free_space(ctl, entry);
3624                         } else {
3625                                 kmem_cache_free(btrfs_free_space_cachep, entry);
3626                         }
3627                 } else {
3628                         start = max(start, extent_start);
3629                         bytes = min(extent_start + extent_bytes, end) - start;
3630                         if (bytes < minlen) {
3631                                 spin_unlock(&ctl->tree_lock);
3632                                 mutex_unlock(&ctl->cache_writeout_mutex);
3633                                 goto next;
3634                         }
3635
3636                         unlink_free_space(ctl, entry);
3637                         kmem_cache_free(btrfs_free_space_cachep, entry);
3638                 }
3639
3640                 spin_unlock(&ctl->tree_lock);
3641                 trim_entry.start = extent_start;
3642                 trim_entry.bytes = extent_bytes;
3643                 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3644                 mutex_unlock(&ctl->cache_writeout_mutex);
3645
3646                 ret = do_trimming(block_group, total_trimmed, start, bytes,
3647                                   extent_start, extent_bytes, extent_trim_state,
3648                                   &trim_entry);
3649                 if (ret) {
3650                         block_group->discard_cursor = start + bytes;
3651                         break;
3652                 }
3653 next:
3654                 start += bytes;
3655                 block_group->discard_cursor = start;
3656                 if (async && *total_trimmed)
3657                         break;
3658
3659                 if (fatal_signal_pending(current)) {
3660                         ret = -ERESTARTSYS;
3661                         break;
3662                 }
3663
3664                 cond_resched();
3665         }
3666
3667         return ret;
3668
3669 out_unlock:
3670         block_group->discard_cursor = btrfs_block_group_end(block_group);
3671         spin_unlock(&ctl->tree_lock);
3672         mutex_unlock(&ctl->cache_writeout_mutex);
3673
3674         return ret;
3675 }
3676
3677 /*
3678  * If we break out of trimming a bitmap prematurely, we should reset the
3679  * trimming bit.  In a rather contrieved case, it's possible to race here so
3680  * reset the state to BTRFS_TRIM_STATE_UNTRIMMED.
3681  *
3682  * start = start of bitmap
3683  * end = near end of bitmap
3684  *
3685  * Thread 1:                    Thread 2:
3686  * trim_bitmaps(start)
3687  *                              trim_bitmaps(end)
3688  *                              end_trimming_bitmap()
3689  * reset_trimming_bitmap()
3690  */
3691 static void reset_trimming_bitmap(struct btrfs_free_space_ctl *ctl, u64 offset)
3692 {
3693         struct btrfs_free_space *entry;
3694
3695         spin_lock(&ctl->tree_lock);
3696         entry = tree_search_offset(ctl, offset, 1, 0);
3697         if (entry) {
3698                 if (btrfs_free_space_trimmed(entry)) {
3699                         ctl->discardable_extents[BTRFS_STAT_CURR] +=
3700                                 entry->bitmap_extents;
3701                         ctl->discardable_bytes[BTRFS_STAT_CURR] += entry->bytes;
3702                 }
3703                 entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3704         }
3705
3706         spin_unlock(&ctl->tree_lock);
3707 }
3708
3709 static void end_trimming_bitmap(struct btrfs_free_space_ctl *ctl,
3710                                 struct btrfs_free_space *entry)
3711 {
3712         if (btrfs_free_space_trimming_bitmap(entry)) {
3713                 entry->trim_state = BTRFS_TRIM_STATE_TRIMMED;
3714                 ctl->discardable_extents[BTRFS_STAT_CURR] -=
3715                         entry->bitmap_extents;
3716                 ctl->discardable_bytes[BTRFS_STAT_CURR] -= entry->bytes;
3717         }
3718 }
3719
3720 /*
3721  * If @async is set, then we will trim 1 region and return.
3722  */
3723 static int trim_bitmaps(struct btrfs_block_group *block_group,
3724                         u64 *total_trimmed, u64 start, u64 end, u64 minlen,
3725                         u64 maxlen, bool async)
3726 {
3727         struct btrfs_discard_ctl *discard_ctl =
3728                                         &block_group->fs_info->discard_ctl;
3729         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3730         struct btrfs_free_space *entry;
3731         int ret = 0;
3732         int ret2;
3733         u64 bytes;
3734         u64 offset = offset_to_bitmap(ctl, start);
3735         const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3736
3737         while (offset < end) {
3738                 bool next_bitmap = false;
3739                 struct btrfs_trim_range trim_entry;
3740
3741                 mutex_lock(&ctl->cache_writeout_mutex);
3742                 spin_lock(&ctl->tree_lock);
3743
3744                 if (ctl->free_space < minlen) {
3745                         block_group->discard_cursor =
3746                                 btrfs_block_group_end(block_group);
3747                         spin_unlock(&ctl->tree_lock);
3748                         mutex_unlock(&ctl->cache_writeout_mutex);
3749                         break;
3750                 }
3751
3752                 entry = tree_search_offset(ctl, offset, 1, 0);
3753                 /*
3754                  * Bitmaps are marked trimmed lossily now to prevent constant
3755                  * discarding of the same bitmap (the reason why we are bound
3756                  * by the filters).  So, retrim the block group bitmaps when we
3757                  * are preparing to punt to the unused_bgs list.  This uses
3758                  * @minlen to determine if we are in BTRFS_DISCARD_INDEX_UNUSED
3759                  * which is the only discard index which sets minlen to 0.
3760                  */
3761                 if (!entry || (async && minlen && start == offset &&
3762                                btrfs_free_space_trimmed(entry))) {
3763                         spin_unlock(&ctl->tree_lock);
3764                         mutex_unlock(&ctl->cache_writeout_mutex);
3765                         next_bitmap = true;
3766                         goto next;
3767                 }
3768
3769                 /*
3770                  * Async discard bitmap trimming begins at by setting the start
3771                  * to be key.objectid and the offset_to_bitmap() aligns to the
3772                  * start of the bitmap.  This lets us know we are fully
3773                  * scanning the bitmap rather than only some portion of it.
3774                  */
3775                 if (start == offset)
3776                         entry->trim_state = BTRFS_TRIM_STATE_TRIMMING;
3777
3778                 bytes = minlen;
3779                 ret2 = search_bitmap(ctl, entry, &start, &bytes, false);
3780                 if (ret2 || start >= end) {
3781                         /*
3782                          * We lossily consider a bitmap trimmed if we only skip
3783                          * over regions <= BTRFS_ASYNC_DISCARD_MIN_FILTER.
3784                          */
3785                         if (ret2 && minlen <= BTRFS_ASYNC_DISCARD_MIN_FILTER)
3786                                 end_trimming_bitmap(ctl, entry);
3787                         else
3788                                 entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3789                         spin_unlock(&ctl->tree_lock);
3790                         mutex_unlock(&ctl->cache_writeout_mutex);
3791                         next_bitmap = true;
3792                         goto next;
3793                 }
3794
3795                 /*
3796                  * We already trimmed a region, but are using the locking above
3797                  * to reset the trim_state.
3798                  */
3799                 if (async && *total_trimmed) {
3800                         spin_unlock(&ctl->tree_lock);
3801                         mutex_unlock(&ctl->cache_writeout_mutex);
3802                         goto out;
3803                 }
3804
3805                 bytes = min(bytes, end - start);
3806                 if (bytes < minlen || (async && maxlen && bytes > maxlen)) {
3807                         spin_unlock(&ctl->tree_lock);
3808                         mutex_unlock(&ctl->cache_writeout_mutex);
3809                         goto next;
3810                 }
3811
3812                 /*
3813                  * Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
3814                  * If X < @minlen, we won't trim X when we come back around.
3815                  * So trim it now.  We differ here from trimming extents as we
3816                  * don't keep individual state per bit.
3817                  */
3818                 if (async &&
3819                     max_discard_size &&
3820                     bytes > (max_discard_size + minlen))
3821                         bytes = max_discard_size;
3822
3823                 bitmap_clear_bits(ctl, entry, start, bytes);
3824                 if (entry->bytes == 0)
3825                         free_bitmap(ctl, entry);
3826
3827                 spin_unlock(&ctl->tree_lock);
3828                 trim_entry.start = start;
3829                 trim_entry.bytes = bytes;
3830                 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3831                 mutex_unlock(&ctl->cache_writeout_mutex);
3832
3833                 ret = do_trimming(block_group, total_trimmed, start, bytes,
3834                                   start, bytes, 0, &trim_entry);
3835                 if (ret) {
3836                         reset_trimming_bitmap(ctl, offset);
3837                         block_group->discard_cursor =
3838                                 btrfs_block_group_end(block_group);
3839                         break;
3840                 }
3841 next:
3842                 if (next_bitmap) {
3843                         offset += BITS_PER_BITMAP * ctl->unit;
3844                         start = offset;
3845                 } else {
3846                         start += bytes;
3847                 }
3848                 block_group->discard_cursor = start;
3849
3850                 if (fatal_signal_pending(current)) {
3851                         if (start != offset)
3852                                 reset_trimming_bitmap(ctl, offset);
3853                         ret = -ERESTARTSYS;
3854                         break;
3855                 }
3856
3857                 cond_resched();
3858         }
3859
3860         if (offset >= end)
3861                 block_group->discard_cursor = end;
3862
3863 out:
3864         return ret;
3865 }
3866
3867 int btrfs_trim_block_group(struct btrfs_block_group *block_group,
3868                            u64 *trimmed, u64 start, u64 end, u64 minlen)
3869 {
3870         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3871         int ret;
3872         u64 rem = 0;
3873
3874         ASSERT(!btrfs_is_zoned(block_group->fs_info));
3875
3876         *trimmed = 0;
3877
3878         spin_lock(&block_group->lock);
3879         if (block_group->removed) {
3880                 spin_unlock(&block_group->lock);
3881                 return 0;
3882         }
3883         btrfs_freeze_block_group(block_group);
3884         spin_unlock(&block_group->lock);
3885
3886         ret = trim_no_bitmap(block_group, trimmed, start, end, minlen, false);
3887         if (ret)
3888                 goto out;
3889
3890         ret = trim_bitmaps(block_group, trimmed, start, end, minlen, 0, false);
3891         div64_u64_rem(end, BITS_PER_BITMAP * ctl->unit, &rem);
3892         /* If we ended in the middle of a bitmap, reset the trimming flag */
3893         if (rem)
3894                 reset_trimming_bitmap(ctl, offset_to_bitmap(ctl, end));
3895 out:
3896         btrfs_unfreeze_block_group(block_group);
3897         return ret;
3898 }
3899
3900 int btrfs_trim_block_group_extents(struct btrfs_block_group *block_group,
3901                                    u64 *trimmed, u64 start, u64 end, u64 minlen,
3902                                    bool async)
3903 {
3904         int ret;
3905
3906         *trimmed = 0;
3907
3908         spin_lock(&block_group->lock);
3909         if (block_group->removed) {
3910                 spin_unlock(&block_group->lock);
3911                 return 0;
3912         }
3913         btrfs_freeze_block_group(block_group);
3914         spin_unlock(&block_group->lock);
3915
3916         ret = trim_no_bitmap(block_group, trimmed, start, end, minlen, async);
3917         btrfs_unfreeze_block_group(block_group);
3918
3919         return ret;
3920 }
3921
3922 int btrfs_trim_block_group_bitmaps(struct btrfs_block_group *block_group,
3923                                    u64 *trimmed, u64 start, u64 end, u64 minlen,
3924                                    u64 maxlen, bool async)
3925 {
3926         int ret;
3927
3928         *trimmed = 0;
3929
3930         spin_lock(&block_group->lock);
3931         if (block_group->removed) {
3932                 spin_unlock(&block_group->lock);
3933                 return 0;
3934         }
3935         btrfs_freeze_block_group(block_group);
3936         spin_unlock(&block_group->lock);
3937
3938         ret = trim_bitmaps(block_group, trimmed, start, end, minlen, maxlen,
3939                            async);
3940
3941         btrfs_unfreeze_block_group(block_group);
3942
3943         return ret;
3944 }
3945
3946 bool btrfs_free_space_cache_v1_active(struct btrfs_fs_info *fs_info)
3947 {
3948         return btrfs_super_cache_generation(fs_info->super_copy);
3949 }
3950
3951 static int cleanup_free_space_cache_v1(struct btrfs_fs_info *fs_info,
3952                                        struct btrfs_trans_handle *trans)
3953 {
3954         struct btrfs_block_group *block_group;
3955         struct rb_node *node;
3956         int ret = 0;
3957
3958         btrfs_info(fs_info, "cleaning free space cache v1");
3959
3960         node = rb_first(&fs_info->block_group_cache_tree);
3961         while (node) {
3962                 block_group = rb_entry(node, struct btrfs_block_group, cache_node);
3963                 ret = btrfs_remove_free_space_inode(trans, NULL, block_group);
3964                 if (ret)
3965                         goto out;
3966                 node = rb_next(node);
3967         }
3968 out:
3969         return ret;
3970 }
3971
3972 int btrfs_set_free_space_cache_v1_active(struct btrfs_fs_info *fs_info, bool active)
3973 {
3974         struct btrfs_trans_handle *trans;
3975         int ret;
3976
3977         /*
3978          * update_super_roots will appropriately set or unset
3979          * super_copy->cache_generation based on SPACE_CACHE and
3980          * BTRFS_FS_CLEANUP_SPACE_CACHE_V1. For this reason, we need a
3981          * transaction commit whether we are enabling space cache v1 and don't
3982          * have any other work to do, or are disabling it and removing free
3983          * space inodes.
3984          */
3985         trans = btrfs_start_transaction(fs_info->tree_root, 0);
3986         if (IS_ERR(trans))
3987                 return PTR_ERR(trans);
3988
3989         if (!active) {
3990                 set_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags);
3991                 ret = cleanup_free_space_cache_v1(fs_info, trans);
3992                 if (ret) {
3993                         btrfs_abort_transaction(trans, ret);
3994                         btrfs_end_transaction(trans);
3995                         goto out;
3996                 }
3997         }
3998
3999         ret = btrfs_commit_transaction(trans);
4000 out:
4001         clear_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags);
4002
4003         return ret;
4004 }
4005
4006 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4007 /*
4008  * Use this if you need to make a bitmap or extent entry specifically, it
4009  * doesn't do any of the merging that add_free_space does, this acts a lot like
4010  * how the free space cache loading stuff works, so you can get really weird
4011  * configurations.
4012  */
4013 int test_add_free_space_entry(struct btrfs_block_group *cache,
4014                               u64 offset, u64 bytes, bool bitmap)
4015 {
4016         struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
4017         struct btrfs_free_space *info = NULL, *bitmap_info;
4018         void *map = NULL;
4019         enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_TRIMMED;
4020         u64 bytes_added;
4021         int ret;
4022
4023 again:
4024         if (!info) {
4025                 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
4026                 if (!info)
4027                         return -ENOMEM;
4028         }
4029
4030         if (!bitmap) {
4031                 spin_lock(&ctl->tree_lock);
4032                 info->offset = offset;
4033                 info->bytes = bytes;
4034                 info->max_extent_size = 0;
4035                 ret = link_free_space(ctl, info);
4036                 spin_unlock(&ctl->tree_lock);
4037                 if (ret)
4038                         kmem_cache_free(btrfs_free_space_cachep, info);
4039                 return ret;
4040         }
4041
4042         if (!map) {
4043                 map = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep, GFP_NOFS);
4044                 if (!map) {
4045                         kmem_cache_free(btrfs_free_space_cachep, info);
4046                         return -ENOMEM;
4047                 }
4048         }
4049
4050         spin_lock(&ctl->tree_lock);
4051         bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
4052                                          1, 0);
4053         if (!bitmap_info) {
4054                 info->bitmap = map;
4055                 map = NULL;
4056                 add_new_bitmap(ctl, info, offset);
4057                 bitmap_info = info;
4058                 info = NULL;
4059         }
4060
4061         bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes,
4062                                           trim_state);
4063
4064         bytes -= bytes_added;
4065         offset += bytes_added;
4066         spin_unlock(&ctl->tree_lock);
4067
4068         if (bytes)
4069                 goto again;
4070
4071         if (info)
4072                 kmem_cache_free(btrfs_free_space_cachep, info);
4073         if (map)
4074                 kmem_cache_free(btrfs_free_space_bitmap_cachep, map);
4075         return 0;
4076 }
4077
4078 /*
4079  * Checks to see if the given range is in the free space cache.  This is really
4080  * just used to check the absence of space, so if there is free space in the
4081  * range at all we will return 1.
4082  */
4083 int test_check_exists(struct btrfs_block_group *cache,
4084                       u64 offset, u64 bytes)
4085 {
4086         struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
4087         struct btrfs_free_space *info;
4088         int ret = 0;
4089
4090         spin_lock(&ctl->tree_lock);
4091         info = tree_search_offset(ctl, offset, 0, 0);
4092         if (!info) {
4093                 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
4094                                           1, 0);
4095                 if (!info)
4096                         goto out;
4097         }
4098
4099 have_info:
4100         if (info->bitmap) {
4101                 u64 bit_off, bit_bytes;
4102                 struct rb_node *n;
4103                 struct btrfs_free_space *tmp;
4104
4105                 bit_off = offset;
4106                 bit_bytes = ctl->unit;
4107                 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes, false);
4108                 if (!ret) {
4109                         if (bit_off == offset) {
4110                                 ret = 1;
4111                                 goto out;
4112                         } else if (bit_off > offset &&
4113                                    offset + bytes > bit_off) {
4114                                 ret = 1;
4115                                 goto out;
4116                         }
4117                 }
4118
4119                 n = rb_prev(&info->offset_index);
4120                 while (n) {
4121                         tmp = rb_entry(n, struct btrfs_free_space,
4122                                        offset_index);
4123                         if (tmp->offset + tmp->bytes < offset)
4124                                 break;
4125                         if (offset + bytes < tmp->offset) {
4126                                 n = rb_prev(&tmp->offset_index);
4127                                 continue;
4128                         }
4129                         info = tmp;
4130                         goto have_info;
4131                 }
4132
4133                 n = rb_next(&info->offset_index);
4134                 while (n) {
4135                         tmp = rb_entry(n, struct btrfs_free_space,
4136                                        offset_index);
4137                         if (offset + bytes < tmp->offset)
4138                                 break;
4139                         if (tmp->offset + tmp->bytes < offset) {
4140                                 n = rb_next(&tmp->offset_index);
4141                                 continue;
4142                         }
4143                         info = tmp;
4144                         goto have_info;
4145                 }
4146
4147                 ret = 0;
4148                 goto out;
4149         }
4150
4151         if (info->offset == offset) {
4152                 ret = 1;
4153                 goto out;
4154         }
4155
4156         if (offset > info->offset && offset < info->offset + info->bytes)
4157                 ret = 1;
4158 out:
4159         spin_unlock(&ctl->tree_lock);
4160         return ret;
4161 }
4162 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */