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 "ctree.h"
  15 #include "fs.h"
  16 #include "messages.h"
  17 #include "misc.h"
  18 #include "free-space-cache.h"
  19 #include "transaction.h"
  20 #include "disk-io.h"
  21 #include "extent_io.h"
  22 #include "volumes.h"
  23 #include "space-info.h"
  24 #include "delalloc-space.h"
  25 #include "block-group.h"
  26 #include "discard.h"
  27 #include "subpage.h"
  28 #include "inode-item.h"
  29 #include "accessors.h"
  30 #include "file-item.h"
  31 #include "file.h"
  32 #include "super.h"
  33
  34 #define BITS_PER_BITMAP         (PAGE_SIZE * 8UL)
  35 #define MAX_CACHE_BYTES_PER_GIG SZ_64K
  36 #define FORCE_EXTENT_THRESHOLD  SZ_1M
  37
  38 static struct kmem_cache *btrfs_free_space_cachep;
  39 static struct kmem_cache *btrfs_free_space_bitmap_cachep;
  40
  41 struct btrfs_trim_range {
  42         u64 start;
  43         u64 bytes;
  44         struct list_head list;
  45 };
  46
  47 static int link_free_space(struct btrfs_free_space_ctl *ctl,
  48                            struct btrfs_free_space *info);
  49 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
  50                               struct btrfs_free_space *info, bool update_stat);
  51 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
  52                          struct btrfs_free_space *bitmap_info, u64 *offset,
  53                          u64 *bytes, bool for_alloc);
  54 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
  55                         struct btrfs_free_space *bitmap_info);
  56 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
  57                               struct btrfs_free_space *info, u64 offset,
  58                               u64 bytes, bool update_stats);
  59
  60 static void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
  61 {
  62         struct btrfs_free_space *info;
  63         struct rb_node *node;
  64
  65         while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
  66                 info = rb_entry(node, struct btrfs_free_space, offset_index);
  67                 if (!info->bitmap) {
  68                         unlink_free_space(ctl, info, true);
  69                         kmem_cache_free(btrfs_free_space_cachep, info);
  70                 } else {
  71                         free_bitmap(ctl, info);
  72                 }
  73
  74                 cond_resched_lock(&ctl->tree_lock);
  75         }
  76 }
  77
  78 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
  79                                                struct btrfs_path *path,
  80                                                u64 offset)
  81 {
  82         struct btrfs_fs_info *fs_info = root->fs_info;
  83         struct btrfs_key key;
  84         struct btrfs_key location;
  85         struct btrfs_disk_key disk_key;
  86         struct btrfs_free_space_header *header;
  87         struct extent_buffer *leaf;
  88         struct inode *inode = NULL;
  89         unsigned nofs_flag;
  90         int ret;
  91
  92         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
  93         key.offset = offset;
  94         key.type = 0;
  95
  96         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
  97         if (ret < 0)
  98                 return ERR_PTR(ret);
  99         if (ret > 0) {
 100                 btrfs_release_path(path);
 101                 return ERR_PTR(-ENOENT);
 102         }
 103
 104         leaf = path->nodes[0];
 105         header = btrfs_item_ptr(leaf, path->slots[0],
 106                                 struct btrfs_free_space_header);
 107         btrfs_free_space_key(leaf, header, &disk_key);
 108         btrfs_disk_key_to_cpu(&location, &disk_key);
 109         btrfs_release_path(path);
 110
 111         /*
 112          * We are often under a trans handle at this point, so we need to make
 113          * sure NOFS is set to keep us from deadlocking.
 114          */
 115         nofs_flag = memalloc_nofs_save();
 116         inode = btrfs_iget_path(fs_info->sb, location.objectid, root, path);
 117         btrfs_release_path(path);
 118         memalloc_nofs_restore(nofs_flag);
 119         if (IS_ERR(inode))
 120                 return inode;
 121
 122         mapping_set_gfp_mask(inode->i_mapping,
 123                         mapping_gfp_constraint(inode->i_mapping,
 124                         ~(__GFP_FS | __GFP_HIGHMEM)));
 125
 126         return inode;
 127 }
 128
 129 struct inode *lookup_free_space_inode(struct btrfs_block_group *block_group,
 130                 struct btrfs_path *path)
 131 {
 132         struct btrfs_fs_info *fs_info = block_group->fs_info;
 133         struct inode *inode = NULL;
 134         u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
 135
 136         spin_lock(&block_group->lock);
 137         if (block_group->inode)
 138                 inode = igrab(block_group->inode);
 139         spin_unlock(&block_group->lock);
 140         if (inode)
 141                 return inode;
 142
 143         inode = __lookup_free_space_inode(fs_info->tree_root, path,
 144                                           block_group->start);
 145         if (IS_ERR(inode))
 146                 return inode;
 147
 148         spin_lock(&block_group->lock);
 149         if (!((BTRFS_I(inode)->flags & flags) == flags)) {
 150                 btrfs_info(fs_info, "Old style space inode found, converting.");
 151                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
 152                         BTRFS_INODE_NODATACOW;
 153                 block_group->disk_cache_state = BTRFS_DC_CLEAR;
 154         }
 155
 156         if (!test_and_set_bit(BLOCK_GROUP_FLAG_IREF, &block_group->runtime_flags))
 157                 block_group->inode = igrab(inode);
 158         spin_unlock(&block_group->lock);
 159
 160         return inode;
 161 }
 162
 163 static int __create_free_space_inode(struct btrfs_root *root,
 164                                      struct btrfs_trans_handle *trans,
 165                                      struct btrfs_path *path,
 166                                      u64 ino, u64 offset)
 167 {
 168         struct btrfs_key key;
 169         struct btrfs_disk_key disk_key;
 170         struct btrfs_free_space_header *header;
 171         struct btrfs_inode_item *inode_item;
 172         struct extent_buffer *leaf;
 173         /* We inline CRCs for the free disk space cache */
 174         const u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC |
 175                           BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
 176         int ret;
 177
 178         ret = btrfs_insert_empty_inode(trans, root, path, ino);
 179         if (ret)
 180                 return ret;
 181
 182         leaf = path->nodes[0];
 183         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 184                                     struct btrfs_inode_item);
 185         btrfs_item_key(leaf, &disk_key, path->slots[0]);
 186         memzero_extent_buffer(leaf, (unsigned long)inode_item,
 187                              sizeof(*inode_item));
 188         btrfs_set_inode_generation(leaf, inode_item, trans->transid);
 189         btrfs_set_inode_size(leaf, inode_item, 0);
 190         btrfs_set_inode_nbytes(leaf, inode_item, 0);
 191         btrfs_set_inode_uid(leaf, inode_item, 0);
 192         btrfs_set_inode_gid(leaf, inode_item, 0);
 193         btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
 194         btrfs_set_inode_flags(leaf, inode_item, flags);
 195         btrfs_set_inode_nlink(leaf, inode_item, 1);
 196         btrfs_set_inode_transid(leaf, inode_item, trans->transid);
 197         btrfs_set_inode_block_group(leaf, inode_item, offset);
 198         btrfs_mark_buffer_dirty(leaf);
 199         btrfs_release_path(path);
 200
 201         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
 202         key.offset = offset;
 203         key.type = 0;
 204         ret = btrfs_insert_empty_item(trans, root, path, &key,
 205                                       sizeof(struct btrfs_free_space_header));
 206         if (ret < 0) {
 207                 btrfs_release_path(path);
 208                 return ret;
 209         }
 210
 211         leaf = path->nodes[0];
 212         header = btrfs_item_ptr(leaf, path->slots[0],
 213                                 struct btrfs_free_space_header);
 214         memzero_extent_buffer(leaf, (unsigned long)header, sizeof(*header));
 215         btrfs_set_free_space_key(leaf, header, &disk_key);
 216         btrfs_mark_buffer_dirty(leaf);
 217         btrfs_release_path(path);
 218
 219         return 0;
 220 }
 221
 222 int create_free_space_inode(struct btrfs_trans_handle *trans,
 223                             struct btrfs_block_group *block_group,
 224                             struct btrfs_path *path)
 225 {
 226         int ret;
 227         u64 ino;
 228
 229         ret = btrfs_get_free_objectid(trans->fs_info->tree_root, &ino);
 230         if (ret < 0)
 231                 return ret;
 232
 233         return __create_free_space_inode(trans->fs_info->tree_root, trans, path,
 234                                          ino, block_group->start);
 235 }
 236
 237 /*
 238  * inode is an optional sink: if it is NULL, btrfs_remove_free_space_inode
 239  * handles lookup, otherwise it takes ownership and iputs the inode.
 240  * Don't reuse an inode pointer after passing it into this function.
 241  */
 242 int btrfs_remove_free_space_inode(struct btrfs_trans_handle *trans,
 243                                   struct inode *inode,
 244                                   struct btrfs_block_group *block_group)
 245 {
 246         struct btrfs_path *path;
 247         struct btrfs_key key;
 248         int ret = 0;
 249
 250         path = btrfs_alloc_path();
 251         if (!path)
 252                 return -ENOMEM;
 253
 254         if (!inode)
 255                 inode = lookup_free_space_inode(block_group, path);
 256         if (IS_ERR(inode)) {
 257                 if (PTR_ERR(inode) != -ENOENT)
 258                         ret = PTR_ERR(inode);
 259                 goto out;
 260         }
 261         ret = btrfs_orphan_add(trans, BTRFS_I(inode));
 262         if (ret) {
 263                 btrfs_add_delayed_iput(BTRFS_I(inode));
 264                 goto out;
 265         }
 266         clear_nlink(inode);
 267         /* One for the block groups ref */
 268         spin_lock(&block_group->lock);
 269         if (test_and_clear_bit(BLOCK_GROUP_FLAG_IREF, &block_group->runtime_flags)) {
 270                 block_group->inode = NULL;
 271                 spin_unlock(&block_group->lock);
 272                 iput(inode);
 273         } else {
 274                 spin_unlock(&block_group->lock);
 275         }
 276         /* One for the lookup ref */
 277         btrfs_add_delayed_iput(BTRFS_I(inode));
 278
 279         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
 280         key.type = 0;
 281         key.offset = block_group->start;
 282         ret = btrfs_search_slot(trans, trans->fs_info->tree_root, &key, path,
 283                                 -1, 1);
 284         if (ret) {
 285                 if (ret > 0)
 286                         ret = 0;
 287                 goto out;
 288         }
 289         ret = btrfs_del_item(trans, trans->fs_info->tree_root, path);
 290 out:
 291         btrfs_free_path(path);
 292         return ret;
 293 }
 294
 295 int btrfs_check_trunc_cache_free_space(struct btrfs_fs_info *fs_info,
 296                                        struct btrfs_block_rsv *rsv)
 297 {
 298         u64 needed_bytes;
 299         int ret;
 300
 301         /* 1 for slack space, 1 for updating the inode */
 302         needed_bytes = btrfs_calc_insert_metadata_size(fs_info, 1) +
 303                 btrfs_calc_metadata_size(fs_info, 1);
 304
 305         spin_lock(&rsv->lock);
 306         if (rsv->reserved < needed_bytes)
 307                 ret = -ENOSPC;
 308         else
 309                 ret = 0;
 310         spin_unlock(&rsv->lock);
 311         return ret;
 312 }
 313
 314 int btrfs_truncate_free_space_cache(struct btrfs_trans_handle *trans,
 315                                     struct btrfs_block_group *block_group,
 316                                     struct inode *vfs_inode)
 317 {
 318         struct btrfs_truncate_control control = {
 319                 .inode = BTRFS_I(vfs_inode),
 320                 .new_size = 0,
 321                 .ino = btrfs_ino(BTRFS_I(vfs_inode)),
 322                 .min_type = BTRFS_EXTENT_DATA_KEY,
 323                 .clear_extent_range = true,
 324         };
 325         struct btrfs_inode *inode = BTRFS_I(vfs_inode);
 326         struct btrfs_root *root = inode->root;
 327         struct extent_state *cached_state = NULL;
 328         int ret = 0;
 329         bool locked = false;
 330
 331         if (block_group) {
 332                 struct btrfs_path *path = btrfs_alloc_path();
 333
 334                 if (!path) {
 335                         ret = -ENOMEM;
 336                         goto fail;
 337                 }
 338                 locked = true;
 339                 mutex_lock(&trans->transaction->cache_write_mutex);
 340                 if (!list_empty(&block_group->io_list)) {
 341                         list_del_init(&block_group->io_list);
 342
 343                         btrfs_wait_cache_io(trans, block_group, path);
 344                         btrfs_put_block_group(block_group);
 345                 }
 346
 347                 /*
 348                  * now that we've truncated the cache away, its no longer
 349                  * setup or written
 350                  */
 351                 spin_lock(&block_group->lock);
 352                 block_group->disk_cache_state = BTRFS_DC_CLEAR;
 353                 spin_unlock(&block_group->lock);
 354                 btrfs_free_path(path);
 355         }
 356
 357         btrfs_i_size_write(inode, 0);
 358         truncate_pagecache(vfs_inode, 0);
 359
 360         lock_extent(&inode->io_tree, 0, (u64)-1, &cached_state);
 361         btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
 362
 363         /*
 364          * We skip the throttling logic for free space cache inodes, so we don't
 365          * need to check for -EAGAIN.
 366          */
 367         ret = btrfs_truncate_inode_items(trans, root, &control);
 368
 369         inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
 370         btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
 371
 372         unlock_extent(&inode->io_tree, 0, (u64)-1, &cached_state);
 373         if (ret)
 374                 goto fail;
 375
 376         ret = btrfs_update_inode(trans, root, inode);
 377
 378 fail:
 379         if (locked)
 380                 mutex_unlock(&trans->transaction->cache_write_mutex);
 381         if (ret)
 382                 btrfs_abort_transaction(trans, ret);
 383
 384         return ret;
 385 }
 386
 387 static void readahead_cache(struct inode *inode)
 388 {
 389         struct file_ra_state ra;
 390         unsigned long last_index;
 391
 392         file_ra_state_init(&ra, inode->i_mapping);
 393         last_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
 394
 395         page_cache_sync_readahead(inode->i_mapping, &ra, NULL, 0, last_index);
 396 }
 397
 398 static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode,
 399                        int write)
 400 {
 401         int num_pages;
 402
 403         num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
 404
 405         /* Make sure we can fit our crcs and generation into the first page */
 406         if (write && (num_pages * sizeof(u32) + sizeof(u64)) > PAGE_SIZE)
 407                 return -ENOSPC;
 408
 409         memset(io_ctl, 0, sizeof(struct btrfs_io_ctl));
 410
 411         io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS);
 412         if (!io_ctl->pages)
 413                 return -ENOMEM;
 414
 415         io_ctl->num_pages = num_pages;
 416         io_ctl->fs_info = btrfs_sb(inode->i_sb);
 417         io_ctl->inode = inode;
 418
 419         return 0;
 420 }
 421 ALLOW_ERROR_INJECTION(io_ctl_init, ERRNO);
 422
 423 static void io_ctl_free(struct btrfs_io_ctl *io_ctl)
 424 {
 425         kfree(io_ctl->pages);
 426         io_ctl->pages = NULL;
 427 }
 428
 429 static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl)
 430 {
 431         if (io_ctl->cur) {
 432                 io_ctl->cur = NULL;
 433                 io_ctl->orig = NULL;
 434         }
 435 }
 436
 437 static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear)
 438 {
 439         ASSERT(io_ctl->index < io_ctl->num_pages);
 440         io_ctl->page = io_ctl->pages[io_ctl->index++];
 441         io_ctl->cur = page_address(io_ctl->page);
 442         io_ctl->orig = io_ctl->cur;
 443         io_ctl->size = PAGE_SIZE;
 444         if (clear)
 445                 clear_page(io_ctl->cur);
 446 }
 447
 448 static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl)
 449 {
 450         int i;
 451
 452         io_ctl_unmap_page(io_ctl);
 453
 454         for (i = 0; i < io_ctl->num_pages; i++) {
 455                 if (io_ctl->pages[i]) {
 456                         btrfs_page_clear_checked(io_ctl->fs_info,
 457                                         io_ctl->pages[i],
 458                                         page_offset(io_ctl->pages[i]),
 459                                         PAGE_SIZE);
 460                         unlock_page(io_ctl->pages[i]);
 461                         put_page(io_ctl->pages[i]);
 462                 }
 463         }
 464 }
 465
 466 static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, bool uptodate)
 467 {
 468         struct page *page;
 469         struct inode *inode = io_ctl->inode;
 470         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
 471         int i;
 472
 473         for (i = 0; i < io_ctl->num_pages; i++) {
 474                 int ret;
 475
 476                 page = find_or_create_page(inode->i_mapping, i, mask);
 477                 if (!page) {
 478                         io_ctl_drop_pages(io_ctl);
 479                         return -ENOMEM;
 480                 }
 481
 482                 ret = set_page_extent_mapped(page);
 483                 if (ret < 0) {
 484                         unlock_page(page);
 485                         put_page(page);
 486                         io_ctl_drop_pages(io_ctl);
 487                         return ret;
 488                 }
 489
 490                 io_ctl->pages[i] = page;
 491                 if (uptodate && !PageUptodate(page)) {
 492                         btrfs_read_folio(NULL, page_folio(page));
 493                         lock_page(page);
 494                         if (page->mapping != inode->i_mapping) {
 495                                 btrfs_err(BTRFS_I(inode)->root->fs_info,
 496                                           "free space cache page truncated");
 497                                 io_ctl_drop_pages(io_ctl);
 498                                 return -EIO;
 499                         }
 500                         if (!PageUptodate(page)) {
 501                                 btrfs_err(BTRFS_I(inode)->root->fs_info,
 502                                            "error reading free space cache");
 503                                 io_ctl_drop_pages(io_ctl);
 504                                 return -EIO;
 505                         }
 506                 }
 507         }
 508
 509         for (i = 0; i < io_ctl->num_pages; i++)
 510                 clear_page_dirty_for_io(io_ctl->pages[i]);
 511
 512         return 0;
 513 }
 514
 515 static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
 516 {
 517         io_ctl_map_page(io_ctl, 1);
 518
 519         /*
 520          * Skip the csum areas.  If we don't check crcs then we just have a
 521          * 64bit chunk at the front of the first page.
 522          */
 523         io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
 524         io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
 525
 526         put_unaligned_le64(generation, io_ctl->cur);
 527         io_ctl->cur += sizeof(u64);
 528 }
 529
 530 static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
 531 {
 532         u64 cache_gen;
 533
 534         /*
 535          * Skip the crc area.  If we don't check crcs then we just have a 64bit
 536          * chunk at the front of the first page.
 537          */
 538         io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
 539         io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
 540
 541         cache_gen = get_unaligned_le64(io_ctl->cur);
 542         if (cache_gen != generation) {
 543                 btrfs_err_rl(io_ctl->fs_info,
 544                         "space cache generation (%llu) does not match inode (%llu)",
 545                                 cache_gen, generation);
 546                 io_ctl_unmap_page(io_ctl);
 547                 return -EIO;
 548         }
 549         io_ctl->cur += sizeof(u64);
 550         return 0;
 551 }
 552
 553 static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index)
 554 {
 555         u32 *tmp;
 556         u32 crc = ~(u32)0;
 557         unsigned offset = 0;
 558
 559         if (index == 0)
 560                 offset = sizeof(u32) * io_ctl->num_pages;
 561
 562         crc = btrfs_crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
 563         btrfs_crc32c_final(crc, (u8 *)&crc);
 564         io_ctl_unmap_page(io_ctl);
 565         tmp = page_address(io_ctl->pages[0]);
 566         tmp += index;
 567         *tmp = crc;
 568 }
 569
 570 static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index)
 571 {
 572         u32 *tmp, val;
 573         u32 crc = ~(u32)0;
 574         unsigned offset = 0;
 575
 576         if (index == 0)
 577                 offset = sizeof(u32) * io_ctl->num_pages;
 578
 579         tmp = page_address(io_ctl->pages[0]);
 580         tmp += index;
 581         val = *tmp;
 582
 583         io_ctl_map_page(io_ctl, 0);
 584         crc = btrfs_crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
 585         btrfs_crc32c_final(crc, (u8 *)&crc);
 586         if (val != crc) {
 587                 btrfs_err_rl(io_ctl->fs_info,
 588                         "csum mismatch on free space cache");
 589                 io_ctl_unmap_page(io_ctl);
 590                 return -EIO;
 591         }
 592
 593         return 0;
 594 }
 595
 596 static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes,
 597                             void *bitmap)
 598 {
 599         struct btrfs_free_space_entry *entry;
 600
 601         if (!io_ctl->cur)
 602                 return -ENOSPC;
 603
 604         entry = io_ctl->cur;
 605         put_unaligned_le64(offset, &entry->offset);
 606         put_unaligned_le64(bytes, &entry->bytes);
 607         entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
 608                 BTRFS_FREE_SPACE_EXTENT;
 609         io_ctl->cur += sizeof(struct btrfs_free_space_entry);
 610         io_ctl->size -= sizeof(struct btrfs_free_space_entry);
 611
 612         if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
 613                 return 0;
 614
 615         io_ctl_set_crc(io_ctl, io_ctl->index - 1);
 616
 617         /* No more pages to map */
 618         if (io_ctl->index >= io_ctl->num_pages)
 619                 return 0;
 620
 621         /* map the next page */
 622         io_ctl_map_page(io_ctl, 1);
 623         return 0;
 624 }
 625
 626 static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap)
 627 {
 628         if (!io_ctl->cur)
 629                 return -ENOSPC;
 630
 631         /*
 632          * If we aren't at the start of the current page, unmap this one and
 633          * map the next one if there is any left.
 634          */
 635         if (io_ctl->cur != io_ctl->orig) {
 636                 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
 637                 if (io_ctl->index >= io_ctl->num_pages)
 638                         return -ENOSPC;
 639                 io_ctl_map_page(io_ctl, 0);
 640         }
 641
 642         copy_page(io_ctl->cur, bitmap);
 643         io_ctl_set_crc(io_ctl, io_ctl->index - 1);
 644         if (io_ctl->index < io_ctl->num_pages)
 645                 io_ctl_map_page(io_ctl, 0);
 646         return 0;
 647 }
 648
 649 static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl)
 650 {
 651         /*
 652          * If we're not on the boundary we know we've modified the page and we
 653          * need to crc the page.
 654          */
 655         if (io_ctl->cur != io_ctl->orig)
 656                 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
 657         else
 658                 io_ctl_unmap_page(io_ctl);
 659
 660         while (io_ctl->index < io_ctl->num_pages) {
 661                 io_ctl_map_page(io_ctl, 1);
 662                 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
 663         }
 664 }
 665
 666 static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl,
 667                             struct btrfs_free_space *entry, u8 *type)
 668 {
 669         struct btrfs_free_space_entry *e;
 670         int ret;
 671
 672         if (!io_ctl->cur) {
 673                 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
 674                 if (ret)
 675                         return ret;
 676         }
 677
 678         e = io_ctl->cur;
 679         entry->offset = get_unaligned_le64(&e->offset);
 680         entry->bytes = get_unaligned_le64(&e->bytes);
 681         *type = e->type;
 682         io_ctl->cur += sizeof(struct btrfs_free_space_entry);
 683         io_ctl->size -= sizeof(struct btrfs_free_space_entry);
 684
 685         if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
 686                 return 0;
 687
 688         io_ctl_unmap_page(io_ctl);
 689
 690         return 0;
 691 }
 692
 693 static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl,
 694                               struct btrfs_free_space *entry)
 695 {
 696         int ret;
 697
 698         ret = io_ctl_check_crc(io_ctl, io_ctl->index);
 699         if (ret)
 700                 return ret;
 701
 702         copy_page(entry->bitmap, io_ctl->cur);
 703         io_ctl_unmap_page(io_ctl);
 704
 705         return 0;
 706 }
 707
 708 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
 709 {
 710         struct btrfs_block_group *block_group = ctl->block_group;
 711         u64 max_bytes;
 712         u64 bitmap_bytes;
 713         u64 extent_bytes;
 714         u64 size = block_group->length;
 715         u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
 716         u64 max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
 717
 718         max_bitmaps = max_t(u64, max_bitmaps, 1);
 719
 720         if (ctl->total_bitmaps > max_bitmaps)
 721                 btrfs_err(block_group->fs_info,
 722 "invalid free space control: bg start=%llu len=%llu total_bitmaps=%u unit=%u max_bitmaps=%llu bytes_per_bg=%llu",
 723                           block_group->start, block_group->length,
 724                           ctl->total_bitmaps, ctl->unit, max_bitmaps,
 725                           bytes_per_bg);
 726         ASSERT(ctl->total_bitmaps <= max_bitmaps);
 727
 728         /*
 729          * We are trying to keep the total amount of memory used per 1GiB of
 730          * space to be MAX_CACHE_BYTES_PER_GIG.  However, with a reclamation
 731          * mechanism of pulling extents >= FORCE_EXTENT_THRESHOLD out of
 732          * bitmaps, we may end up using more memory than this.
 733          */
 734         if (size < SZ_1G)
 735                 max_bytes = MAX_CACHE_BYTES_PER_GIG;
 736         else
 737                 max_bytes = MAX_CACHE_BYTES_PER_GIG * div_u64(size, SZ_1G);
 738
 739         bitmap_bytes = ctl->total_bitmaps * ctl->unit;
 740
 741         /*
 742          * we want the extent entry threshold to always be at most 1/2 the max
 743          * bytes we can have, or whatever is less than that.
 744          */
 745         extent_bytes = max_bytes - bitmap_bytes;
 746         extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1);
 747
 748         ctl->extents_thresh =
 749                 div_u64(extent_bytes, sizeof(struct btrfs_free_space));
 750 }
 751
 752 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
 753                                    struct btrfs_free_space_ctl *ctl,
 754                                    struct btrfs_path *path, u64 offset)
 755 {
 756         struct btrfs_fs_info *fs_info = root->fs_info;
 757         struct btrfs_free_space_header *header;
 758         struct extent_buffer *leaf;
 759         struct btrfs_io_ctl io_ctl;
 760         struct btrfs_key key;
 761         struct btrfs_free_space *e, *n;
 762         LIST_HEAD(bitmaps);
 763         u64 num_entries;
 764         u64 num_bitmaps;
 765         u64 generation;
 766         u8 type;
 767         int ret = 0;
 768
 769         /* Nothing in the space cache, goodbye */
 770         if (!i_size_read(inode))
 771                 return 0;
 772
 773         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
 774         key.offset = offset;
 775         key.type = 0;
 776
 777         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 778         if (ret < 0)
 779                 return 0;
 780         else if (ret > 0) {
 781                 btrfs_release_path(path);
 782                 return 0;
 783         }
 784
 785         ret = -1;
 786
 787         leaf = path->nodes[0];
 788         header = btrfs_item_ptr(leaf, path->slots[0],
 789                                 struct btrfs_free_space_header);
 790         num_entries = btrfs_free_space_entries(leaf, header);
 791         num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
 792         generation = btrfs_free_space_generation(leaf, header);
 793         btrfs_release_path(path);
 794
 795         if (!BTRFS_I(inode)->generation) {
 796                 btrfs_info(fs_info,
 797                            "the free space cache file (%llu) is invalid, skip it",
 798                            offset);
 799                 return 0;
 800         }
 801
 802         if (BTRFS_I(inode)->generation != generation) {
 803                 btrfs_err(fs_info,
 804                           "free space inode generation (%llu) did not match free space cache generation (%llu)",
 805                           BTRFS_I(inode)->generation, generation);
 806                 return 0;
 807         }
 808
 809         if (!num_entries)
 810                 return 0;
 811
 812         ret = io_ctl_init(&io_ctl, inode, 0);
 813         if (ret)
 814                 return ret;
 815
 816         readahead_cache(inode);
 817
 818         ret = io_ctl_prepare_pages(&io_ctl, true);
 819         if (ret)
 820                 goto out;
 821
 822         ret = io_ctl_check_crc(&io_ctl, 0);
 823         if (ret)
 824                 goto free_cache;
 825
 826         ret = io_ctl_check_generation(&io_ctl, generation);
 827         if (ret)
 828                 goto free_cache;
 829
 830         while (num_entries) {
 831                 e = kmem_cache_zalloc(btrfs_free_space_cachep,
 832                                       GFP_NOFS);
 833                 if (!e) {
 834                         ret = -ENOMEM;
 835                         goto free_cache;
 836                 }
 837
 838                 ret = io_ctl_read_entry(&io_ctl, e, &type);
 839                 if (ret) {
 840                         kmem_cache_free(btrfs_free_space_cachep, e);
 841                         goto free_cache;
 842                 }
 843
 844                 if (!e->bytes) {
 845                         ret = -1;
 846                         kmem_cache_free(btrfs_free_space_cachep, e);
 847                         goto free_cache;
 848                 }
 849
 850                 if (type == BTRFS_FREE_SPACE_EXTENT) {
 851                         spin_lock(&ctl->tree_lock);
 852                         ret = link_free_space(ctl, e);
 853                         spin_unlock(&ctl->tree_lock);
 854                         if (ret) {
 855                                 btrfs_err(fs_info,
 856                                         "Duplicate entries in free space cache, dumping");
 857                                 kmem_cache_free(btrfs_free_space_cachep, e);
 858                                 goto free_cache;
 859                         }
 860                 } else {
 861                         ASSERT(num_bitmaps);
 862                         num_bitmaps--;
 863                         e->bitmap = kmem_cache_zalloc(
 864                                         btrfs_free_space_bitmap_cachep, GFP_NOFS);
 865                         if (!e->bitmap) {
 866                                 ret = -ENOMEM;
 867                                 kmem_cache_free(
 868                                         btrfs_free_space_cachep, e);
 869                                 goto free_cache;
 870                         }
 871                         spin_lock(&ctl->tree_lock);
 872                         ret = link_free_space(ctl, e);
 873                         ctl->total_bitmaps++;
 874                         recalculate_thresholds(ctl);
 875                         spin_unlock(&ctl->tree_lock);
 876                         if (ret) {
 877                                 btrfs_err(fs_info,
 878                                         "Duplicate entries in free space cache, dumping");
 879                                 kmem_cache_free(btrfs_free_space_cachep, e);
 880                                 goto free_cache;
 881                         }
 882                         list_add_tail(&e->list, &bitmaps);
 883                 }
 884
 885                 num_entries--;
 886         }
 887
 888         io_ctl_unmap_page(&io_ctl);
 889
 890         /*
 891          * We add the bitmaps at the end of the entries in order that
 892          * the bitmap entries are added to the cache.
 893          */
 894         list_for_each_entry_safe(e, n, &bitmaps, list) {
 895                 list_del_init(&e->list);
 896                 ret = io_ctl_read_bitmap(&io_ctl, e);
 897                 if (ret)
 898                         goto free_cache;
 899         }
 900
 901         io_ctl_drop_pages(&io_ctl);
 902         ret = 1;
 903 out:
 904         io_ctl_free(&io_ctl);
 905         return ret;
 906 free_cache:
 907         io_ctl_drop_pages(&io_ctl);
 908
 909         spin_lock(&ctl->tree_lock);
 910         __btrfs_remove_free_space_cache(ctl);
 911         spin_unlock(&ctl->tree_lock);
 912         goto out;
 913 }
 914
 915 static int copy_free_space_cache(struct btrfs_block_group *block_group,
 916                                  struct btrfs_free_space_ctl *ctl)
 917 {
 918         struct btrfs_free_space *info;
 919         struct rb_node *n;
 920         int ret = 0;
 921
 922         while (!ret && (n = rb_first(&ctl->free_space_offset)) != NULL) {
 923                 info = rb_entry(n, struct btrfs_free_space, offset_index);
 924                 if (!info->bitmap) {
 925                         unlink_free_space(ctl, info, true);
 926                         ret = btrfs_add_free_space(block_group, info->offset,
 927                                                    info->bytes);
 928                         kmem_cache_free(btrfs_free_space_cachep, info);
 929                 } else {
 930                         u64 offset = info->offset;
 931                         u64 bytes = ctl->unit;
 932
 933                         while (search_bitmap(ctl, info, &offset, &bytes,
 934                                              false) == 0) {
 935                                 ret = btrfs_add_free_space(block_group, offset,
 936                                                            bytes);
 937                                 if (ret)
 938                                         break;
 939                                 bitmap_clear_bits(ctl, info, offset, bytes, true);
 940                                 offset = info->offset;
 941                                 bytes = ctl->unit;
 942                         }
 943                         free_bitmap(ctl, info);
 944                 }
 945                 cond_resched();
 946         }
 947         return ret;
 948 }
 949
 950 static struct lock_class_key btrfs_free_space_inode_key;
 951
 952 int load_free_space_cache(struct btrfs_block_group *block_group)
 953 {
 954         struct btrfs_fs_info *fs_info = block_group->fs_info;
 955         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
 956         struct btrfs_free_space_ctl tmp_ctl = {};
 957         struct inode *inode;
 958         struct btrfs_path *path;
 959         int ret = 0;
 960         bool matched;
 961         u64 used = block_group->used;
 962
 963         /*
 964          * Because we could potentially discard our loaded free space, we want
 965          * to load everything into a temporary structure first, and then if it's
 966          * valid copy it all into the actual free space ctl.
 967          */
 968         btrfs_init_free_space_ctl(block_group, &tmp_ctl);
 969
 970         /*
 971          * If this block group has been marked to be cleared for one reason or
 972          * another then we can't trust the on disk cache, so just return.
 973          */
 974         spin_lock(&block_group->lock);
 975         if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
 976                 spin_unlock(&block_group->lock);
 977                 return 0;
 978         }
 979         spin_unlock(&block_group->lock);
 980
 981         path = btrfs_alloc_path();
 982         if (!path)
 983                 return 0;
 984         path->search_commit_root = 1;
 985         path->skip_locking = 1;
 986
 987         /*
 988          * We must pass a path with search_commit_root set to btrfs_iget in
 989          * order to avoid a deadlock when allocating extents for the tree root.
 990          *
 991          * When we are COWing an extent buffer from the tree root, when looking
 992          * for a free extent, at extent-tree.c:find_free_extent(), we can find
 993          * block group without its free space cache loaded. When we find one
 994          * we must load its space cache which requires reading its free space
 995          * cache's inode item from the root tree. If this inode item is located
 996          * in the same leaf that we started COWing before, then we end up in
 997          * deadlock on the extent buffer (trying to read lock it when we
 998          * previously write locked it).
 999          *
1000          * It's safe to read the inode item using the commit root because
1001          * block groups, once loaded, stay in memory forever (until they are
1002          * removed) as well as their space caches once loaded. New block groups
1003          * once created get their ->cached field set to BTRFS_CACHE_FINISHED so
1004          * we will never try to read their inode item while the fs is mounted.
1005          */
1006         inode = lookup_free_space_inode(block_group, path);
1007         if (IS_ERR(inode)) {
1008                 btrfs_free_path(path);
1009                 return 0;
1010         }
1011
1012         /* We may have converted the inode and made the cache invalid. */
1013         spin_lock(&block_group->lock);
1014         if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
1015                 spin_unlock(&block_group->lock);
1016                 btrfs_free_path(path);
1017                 goto out;
1018         }
1019         spin_unlock(&block_group->lock);
1020
1021         /*
1022          * Reinitialize the class of struct inode's mapping->invalidate_lock for
1023          * free space inodes to prevent false positives related to locks for normal
1024          * inodes.
1025          */
1026         lockdep_set_class(&(&inode->i_data)->invalidate_lock,
1027                           &btrfs_free_space_inode_key);
1028
1029         ret = __load_free_space_cache(fs_info->tree_root, inode, &tmp_ctl,
1030                                       path, block_group->start);
1031         btrfs_free_path(path);
1032         if (ret <= 0)
1033                 goto out;
1034
1035         matched = (tmp_ctl.free_space == (block_group->length - used -
1036                                           block_group->bytes_super));
1037
1038         if (matched) {
1039                 ret = copy_free_space_cache(block_group, &tmp_ctl);
1040                 /*
1041                  * ret == 1 means we successfully loaded the free space cache,
1042                  * so we need to re-set it here.
1043                  */
1044                 if (ret == 0)
1045                         ret = 1;
1046         } else {
1047                 /*
1048                  * We need to call the _locked variant so we don't try to update
1049                  * the discard counters.
1050                  */
1051                 spin_lock(&tmp_ctl.tree_lock);
1052                 __btrfs_remove_free_space_cache(&tmp_ctl);
1053                 spin_unlock(&tmp_ctl.tree_lock);
1054                 btrfs_warn(fs_info,
1055                            "block group %llu has wrong amount of free space",
1056                            block_group->start);
1057                 ret = -1;
1058         }
1059 out:
1060         if (ret < 0) {
1061                 /* This cache is bogus, make sure it gets cleared */
1062                 spin_lock(&block_group->lock);
1063                 block_group->disk_cache_state = BTRFS_DC_CLEAR;
1064                 spin_unlock(&block_group->lock);
1065                 ret = 0;
1066
1067                 btrfs_warn(fs_info,
1068                            "failed to load free space cache for block group %llu, rebuilding it now",
1069                            block_group->start);
1070         }
1071
1072         spin_lock(&ctl->tree_lock);
1073         btrfs_discard_update_discardable(block_group);
1074         spin_unlock(&ctl->tree_lock);
1075         iput(inode);
1076         return ret;
1077 }
1078
1079 static noinline_for_stack
1080 int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl,
1081                               struct btrfs_free_space_ctl *ctl,
1082                               struct btrfs_block_group *block_group,
1083                               int *entries, int *bitmaps,
1084                               struct list_head *bitmap_list)
1085 {
1086         int ret;
1087         struct btrfs_free_cluster *cluster = NULL;
1088         struct btrfs_free_cluster *cluster_locked = NULL;
1089         struct rb_node *node = rb_first(&ctl->free_space_offset);
1090         struct btrfs_trim_range *trim_entry;
1091
1092         /* Get the cluster for this block_group if it exists */
1093         if (block_group && !list_empty(&block_group->cluster_list)) {
1094                 cluster = list_entry(block_group->cluster_list.next,
1095                                      struct btrfs_free_cluster,
1096                                      block_group_list);
1097         }
1098
1099         if (!node && cluster) {
1100                 cluster_locked = cluster;
1101                 spin_lock(&cluster_locked->lock);
1102                 node = rb_first(&cluster->root);
1103                 cluster = NULL;
1104         }
1105
1106         /* Write out the extent entries */
1107         while (node) {
1108                 struct btrfs_free_space *e;
1109
1110                 e = rb_entry(node, struct btrfs_free_space, offset_index);
1111                 *entries += 1;
1112
1113                 ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
1114                                        e->bitmap);
1115                 if (ret)
1116                         goto fail;
1117
1118                 if (e->bitmap) {
1119                         list_add_tail(&e->list, bitmap_list);
1120                         *bitmaps += 1;
1121                 }
1122                 node = rb_next(node);
1123                 if (!node && cluster) {
1124                         node = rb_first(&cluster->root);
1125                         cluster_locked = cluster;
1126                         spin_lock(&cluster_locked->lock);
1127                         cluster = NULL;
1128                 }
1129         }
1130         if (cluster_locked) {
1131                 spin_unlock(&cluster_locked->lock);
1132                 cluster_locked = NULL;
1133         }
1134
1135         /*
1136          * Make sure we don't miss any range that was removed from our rbtree
1137          * because trimming is running. Otherwise after a umount+mount (or crash
1138          * after committing the transaction) we would leak free space and get
1139          * an inconsistent free space cache report from fsck.
1140          */
1141         list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) {
1142                 ret = io_ctl_add_entry(io_ctl, trim_entry->start,
1143                                        trim_entry->bytes, NULL);
1144                 if (ret)
1145                         goto fail;
1146                 *entries += 1;
1147         }
1148
1149         return 0;
1150 fail:
1151         if (cluster_locked)
1152                 spin_unlock(&cluster_locked->lock);
1153         return -ENOSPC;
1154 }
1155
1156 static noinline_for_stack int
1157 update_cache_item(struct btrfs_trans_handle *trans,
1158                   struct btrfs_root *root,
1159                   struct inode *inode,
1160                   struct btrfs_path *path, u64 offset,
1161                   int entries, int bitmaps)
1162 {
1163         struct btrfs_key key;
1164         struct btrfs_free_space_header *header;
1165         struct extent_buffer *leaf;
1166         int ret;
1167
1168         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1169         key.offset = offset;
1170         key.type = 0;
1171
1172         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1173         if (ret < 0) {
1174                 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1175                                  EXTENT_DELALLOC, NULL);
1176                 goto fail;
1177         }
1178         leaf = path->nodes[0];
1179         if (ret > 0) {
1180                 struct btrfs_key found_key;
1181                 ASSERT(path->slots[0]);
1182                 path->slots[0]--;
1183                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1184                 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1185                     found_key.offset != offset) {
1186                         clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1187                                          inode->i_size - 1, EXTENT_DELALLOC,
1188                                          NULL);
1189                         btrfs_release_path(path);
1190                         goto fail;
1191                 }
1192         }
1193
1194         BTRFS_I(inode)->generation = trans->transid;
1195         header = btrfs_item_ptr(leaf, path->slots[0],
1196                                 struct btrfs_free_space_header);
1197         btrfs_set_free_space_entries(leaf, header, entries);
1198         btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1199         btrfs_set_free_space_generation(leaf, header, trans->transid);
1200         btrfs_mark_buffer_dirty(leaf);
1201         btrfs_release_path(path);
1202
1203         return 0;
1204
1205 fail:
1206         return -1;
1207 }
1208
1209 static noinline_for_stack int write_pinned_extent_entries(
1210                             struct btrfs_trans_handle *trans,
1211                             struct btrfs_block_group *block_group,
1212                             struct btrfs_io_ctl *io_ctl,
1213                             int *entries)
1214 {
1215         u64 start, extent_start, extent_end, len;
1216         struct extent_io_tree *unpin = NULL;
1217         int ret;
1218
1219         if (!block_group)
1220                 return 0;
1221
1222         /*
1223          * We want to add any pinned extents to our free space cache
1224          * so we don't leak the space
1225          *
1226          * We shouldn't have switched the pinned extents yet so this is the
1227          * right one
1228          */
1229         unpin = &trans->transaction->pinned_extents;
1230
1231         start = block_group->start;
1232
1233         while (start < block_group->start + block_group->length) {
1234                 ret = find_first_extent_bit(unpin, start,
1235                                             &extent_start, &extent_end,
1236                                             EXTENT_DIRTY, NULL);
1237                 if (ret)
1238                         return 0;
1239
1240                 /* This pinned extent is out of our range */
1241                 if (extent_start >= block_group->start + block_group->length)
1242                         return 0;
1243
1244                 extent_start = max(extent_start, start);
1245                 extent_end = min(block_group->start + block_group->length,
1246                                  extent_end + 1);
1247                 len = extent_end - extent_start;
1248
1249                 *entries += 1;
1250                 ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
1251                 if (ret)
1252                         return -ENOSPC;
1253
1254                 start = extent_end;
1255         }
1256
1257         return 0;
1258 }
1259
1260 static noinline_for_stack int
1261 write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list)
1262 {
1263         struct btrfs_free_space *entry, *next;
1264         int ret;
1265
1266         /* Write out the bitmaps */
1267         list_for_each_entry_safe(entry, next, bitmap_list, list) {
1268                 ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
1269                 if (ret)
1270                         return -ENOSPC;
1271                 list_del_init(&entry->list);
1272         }
1273
1274         return 0;
1275 }
1276
1277 static int flush_dirty_cache(struct inode *inode)
1278 {
1279         int ret;
1280
1281         ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1282         if (ret)
1283                 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1284                                  EXTENT_DELALLOC, NULL);
1285
1286         return ret;
1287 }
1288
1289 static void noinline_for_stack
1290 cleanup_bitmap_list(struct list_head *bitmap_list)
1291 {
1292         struct btrfs_free_space *entry, *next;
1293
1294         list_for_each_entry_safe(entry, next, bitmap_list, list)
1295                 list_del_init(&entry->list);
1296 }
1297
1298 static void noinline_for_stack
1299 cleanup_write_cache_enospc(struct inode *inode,
1300                            struct btrfs_io_ctl *io_ctl,
1301                            struct extent_state **cached_state)
1302 {
1303         io_ctl_drop_pages(io_ctl);
1304         unlock_extent(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1305                       cached_state);
1306 }
1307
1308 static int __btrfs_wait_cache_io(struct btrfs_root *root,
1309                                  struct btrfs_trans_handle *trans,
1310                                  struct btrfs_block_group *block_group,
1311                                  struct btrfs_io_ctl *io_ctl,
1312                                  struct btrfs_path *path, u64 offset)
1313 {
1314         int ret;
1315         struct inode *inode = io_ctl->inode;
1316
1317         if (!inode)
1318                 return 0;
1319
1320         /* Flush the dirty pages in the cache file. */
1321         ret = flush_dirty_cache(inode);
1322         if (ret)
1323                 goto out;
1324
1325         /* Update the cache item to tell everyone this cache file is valid. */
1326         ret = update_cache_item(trans, root, inode, path, offset,
1327                                 io_ctl->entries, io_ctl->bitmaps);
1328 out:
1329         if (ret) {
1330                 invalidate_inode_pages2(inode->i_mapping);
1331                 BTRFS_I(inode)->generation = 0;
1332                 if (block_group)
1333                         btrfs_debug(root->fs_info,
1334           "failed to write free space cache for block group %llu error %d",
1335                                   block_group->start, ret);
1336         }
1337         btrfs_update_inode(trans, root, BTRFS_I(inode));
1338
1339         if (block_group) {
1340                 /* the dirty list is protected by the dirty_bgs_lock */
1341                 spin_lock(&trans->transaction->dirty_bgs_lock);
1342
1343                 /* the disk_cache_state is protected by the block group lock */
1344                 spin_lock(&block_group->lock);
1345
1346                 /*
1347                  * only mark this as written if we didn't get put back on
1348                  * the dirty list while waiting for IO.   Otherwise our
1349                  * cache state won't be right, and we won't get written again
1350                  */
1351                 if (!ret && list_empty(&block_group->dirty_list))
1352                         block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1353                 else if (ret)
1354                         block_group->disk_cache_state = BTRFS_DC_ERROR;
1355
1356                 spin_unlock(&block_group->lock);
1357                 spin_unlock(&trans->transaction->dirty_bgs_lock);
1358                 io_ctl->inode = NULL;
1359                 iput(inode);
1360         }
1361
1362         return ret;
1363
1364 }
1365
1366 int btrfs_wait_cache_io(struct btrfs_trans_handle *trans,
1367                         struct btrfs_block_group *block_group,
1368                         struct btrfs_path *path)
1369 {
1370         return __btrfs_wait_cache_io(block_group->fs_info->tree_root, trans,
1371                                      block_group, &block_group->io_ctl,
1372                                      path, block_group->start);
1373 }
1374
1375 /*
1376  * Write out cached info to an inode.
1377  *
1378  * @root:        root the inode belongs to
1379  * @inode:       freespace inode we are writing out
1380  * @ctl:         free space cache we are going to write out
1381  * @block_group: block_group for this cache if it belongs to a block_group
1382  * @io_ctl:      holds context for the io
1383  * @trans:       the trans handle
1384  *
1385  * This function writes out a free space cache struct to disk for quick recovery
1386  * on mount.  This will return 0 if it was successful in writing the cache out,
1387  * or an errno if it was not.
1388  */
1389 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
1390                                    struct btrfs_free_space_ctl *ctl,
1391                                    struct btrfs_block_group *block_group,
1392                                    struct btrfs_io_ctl *io_ctl,
1393                                    struct btrfs_trans_handle *trans)
1394 {
1395         struct extent_state *cached_state = NULL;
1396         LIST_HEAD(bitmap_list);
1397         int entries = 0;
1398         int bitmaps = 0;
1399         int ret;
1400         int must_iput = 0;
1401
1402         if (!i_size_read(inode))
1403                 return -EIO;
1404
1405         WARN_ON(io_ctl->pages);
1406         ret = io_ctl_init(io_ctl, inode, 1);
1407         if (ret)
1408                 return ret;
1409
1410         if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) {
1411                 down_write(&block_group->data_rwsem);
1412                 spin_lock(&block_group->lock);
1413                 if (block_group->delalloc_bytes) {
1414                         block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1415                         spin_unlock(&block_group->lock);
1416                         up_write(&block_group->data_rwsem);
1417                         BTRFS_I(inode)->generation = 0;
1418                         ret = 0;
1419                         must_iput = 1;
1420                         goto out;
1421                 }
1422                 spin_unlock(&block_group->lock);
1423         }
1424
1425         /* Lock all pages first so we can lock the extent safely. */
1426         ret = io_ctl_prepare_pages(io_ctl, false);
1427         if (ret)
1428                 goto out_unlock;
1429
1430         lock_extent(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1431                     &cached_state);
1432
1433         io_ctl_set_generation(io_ctl, trans->transid);
1434
1435         mutex_lock(&ctl->cache_writeout_mutex);
1436         /* Write out the extent entries in the free space cache */
1437         spin_lock(&ctl->tree_lock);
1438         ret = write_cache_extent_entries(io_ctl, ctl,
1439                                          block_group, &entries, &bitmaps,
1440                                          &bitmap_list);
1441         if (ret)
1442                 goto out_nospc_locked;
1443
1444         /*
1445          * Some spaces that are freed in the current transaction are pinned,
1446          * they will be added into free space cache after the transaction is
1447          * committed, we shouldn't lose them.
1448          *
1449          * If this changes while we are working we'll get added back to
1450          * the dirty list and redo it.  No locking needed
1451          */
1452         ret = write_pinned_extent_entries(trans, block_group, io_ctl, &entries);
1453         if (ret)
1454                 goto out_nospc_locked;
1455
1456         /*
1457          * At last, we write out all the bitmaps and keep cache_writeout_mutex
1458          * locked while doing it because a concurrent trim can be manipulating
1459          * or freeing the bitmap.
1460          */
1461         ret = write_bitmap_entries(io_ctl, &bitmap_list);
1462         spin_unlock(&ctl->tree_lock);
1463         mutex_unlock(&ctl->cache_writeout_mutex);
1464         if (ret)
1465                 goto out_nospc;
1466
1467         /* Zero out the rest of the pages just to make sure */
1468         io_ctl_zero_remaining_pages(io_ctl);
1469
1470         /* Everything is written out, now we dirty the pages in the file. */
1471         ret = btrfs_dirty_pages(BTRFS_I(inode), io_ctl->pages,
1472                                 io_ctl->num_pages, 0, i_size_read(inode),
1473                                 &cached_state, false);
1474         if (ret)
1475                 goto out_nospc;
1476
1477         if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1478                 up_write(&block_group->data_rwsem);
1479         /*
1480          * Release the pages and unlock the extent, we will flush
1481          * them out later
1482          */
1483         io_ctl_drop_pages(io_ctl);
1484         io_ctl_free(io_ctl);
1485
1486         unlock_extent(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1487                       &cached_state);
1488
1489         /*
1490          * at this point the pages are under IO and we're happy,
1491          * The caller is responsible for waiting on them and updating
1492          * the cache and the inode
1493          */
1494         io_ctl->entries = entries;
1495         io_ctl->bitmaps = bitmaps;
1496
1497         ret = btrfs_fdatawrite_range(inode, 0, (u64)-1);
1498         if (ret)
1499                 goto out;
1500
1501         return 0;
1502
1503 out_nospc_locked:
1504         cleanup_bitmap_list(&bitmap_list);
1505         spin_unlock(&ctl->tree_lock);
1506         mutex_unlock(&ctl->cache_writeout_mutex);
1507
1508 out_nospc:
1509         cleanup_write_cache_enospc(inode, io_ctl, &cached_state);
1510
1511 out_unlock:
1512         if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1513                 up_write(&block_group->data_rwsem);
1514
1515 out:
1516         io_ctl->inode = NULL;
1517         io_ctl_free(io_ctl);
1518         if (ret) {
1519                 invalidate_inode_pages2(inode->i_mapping);
1520                 BTRFS_I(inode)->generation = 0;
1521         }
1522         btrfs_update_inode(trans, root, BTRFS_I(inode));
1523         if (must_iput)
1524                 iput(inode);
1525         return ret;
1526 }
1527
1528 int btrfs_write_out_cache(struct btrfs_trans_handle *trans,
1529                           struct btrfs_block_group *block_group,
1530                           struct btrfs_path *path)
1531 {
1532         struct btrfs_fs_info *fs_info = trans->fs_info;
1533         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1534         struct inode *inode;
1535         int ret = 0;
1536
1537         spin_lock(&block_group->lock);
1538         if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1539                 spin_unlock(&block_group->lock);
1540                 return 0;
1541         }
1542         spin_unlock(&block_group->lock);
1543
1544         inode = lookup_free_space_inode(block_group, path);
1545         if (IS_ERR(inode))
1546                 return 0;
1547
1548         ret = __btrfs_write_out_cache(fs_info->tree_root, inode, ctl,
1549                                 block_group, &block_group->io_ctl, trans);
1550         if (ret) {
1551                 btrfs_debug(fs_info,
1552           "failed to write free space cache for block group %llu error %d",
1553                           block_group->start, ret);
1554                 spin_lock(&block_group->lock);
1555                 block_group->disk_cache_state = BTRFS_DC_ERROR;
1556                 spin_unlock(&block_group->lock);
1557
1558                 block_group->io_ctl.inode = NULL;
1559                 iput(inode);
1560         }
1561
1562         /*
1563          * if ret == 0 the caller is expected to call btrfs_wait_cache_io
1564          * to wait for IO and put the inode
1565          */
1566
1567         return ret;
1568 }
1569
1570 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1571                                           u64 offset)
1572 {
1573         ASSERT(offset >= bitmap_start);
1574         offset -= bitmap_start;
1575         return (unsigned long)(div_u64(offset, unit));
1576 }
1577
1578 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1579 {
1580         return (unsigned long)(div_u64(bytes, unit));
1581 }
1582
1583 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1584                                    u64 offset)
1585 {
1586         u64 bitmap_start;
1587         u64 bytes_per_bitmap;
1588
1589         bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1590         bitmap_start = offset - ctl->start;
1591         bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1592         bitmap_start *= bytes_per_bitmap;
1593         bitmap_start += ctl->start;
1594
1595         return bitmap_start;
1596 }
1597
1598 static int tree_insert_offset(struct rb_root *root, u64 offset,
1599                               struct rb_node *node, int bitmap)
1600 {
1601         struct rb_node **p = &root->rb_node;
1602         struct rb_node *parent = NULL;
1603         struct btrfs_free_space *info;
1604
1605         while (*p) {
1606                 parent = *p;
1607                 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1608
1609                 if (offset < info->offset) {
1610                         p = &(*p)->rb_left;
1611                 } else if (offset > info->offset) {
1612                         p = &(*p)->rb_right;
1613                 } else {
1614                         /*
1615                          * we could have a bitmap entry and an extent entry
1616                          * share the same offset.  If this is the case, we want
1617                          * the extent entry to always be found first if we do a
1618                          * linear search through the tree, since we want to have
1619                          * the quickest allocation time, and allocating from an
1620                          * extent is faster than allocating from a bitmap.  So
1621                          * if we're inserting a bitmap and we find an entry at
1622                          * this offset, we want to go right, or after this entry
1623                          * logically.  If we are inserting an extent and we've
1624                          * found a bitmap, we want to go left, or before
1625                          * logically.
1626                          */
1627                         if (bitmap) {
1628                                 if (info->bitmap) {
1629                                         WARN_ON_ONCE(1);
1630                                         return -EEXIST;
1631                                 }
1632                                 p = &(*p)->rb_right;
1633                         } else {
1634                                 if (!info->bitmap) {
1635                                         WARN_ON_ONCE(1);
1636                                         return -EEXIST;
1637                                 }
1638                                 p = &(*p)->rb_left;
1639                         }
1640                 }
1641         }
1642
1643         rb_link_node(node, parent, p);
1644         rb_insert_color(node, root);
1645
1646         return 0;
1647 }
1648
1649 /*
1650  * This is a little subtle.  We *only* have ->max_extent_size set if we actually
1651  * searched through the bitmap and figured out the largest ->max_extent_size,
1652  * otherwise it's 0.  In the case that it's 0 we don't want to tell the
1653  * allocator the wrong thing, we want to use the actual real max_extent_size
1654  * we've found already if it's larger, or we want to use ->bytes.
1655  *
1656  * This matters because find_free_space() will skip entries who's ->bytes is
1657  * less than the required bytes.  So if we didn't search down this bitmap, we
1658  * may pick some previous entry that has a smaller ->max_extent_size than we
1659  * have.  For example, assume we have two entries, one that has
1660  * ->max_extent_size set to 4K and ->bytes set to 1M.  A second entry hasn't set
1661  * ->max_extent_size yet, has ->bytes set to 8K and it's contiguous.  We will
1662  *  call into find_free_space(), and return with max_extent_size == 4K, because
1663  *  that first bitmap entry had ->max_extent_size set, but the second one did
1664  *  not.  If instead we returned 8K we'd come in searching for 8K, and find the
1665  *  8K contiguous range.
1666  *
1667  *  Consider the other case, we have 2 8K chunks in that second entry and still
1668  *  don't have ->max_extent_size set.  We'll return 16K, and the next time the
1669  *  allocator comes in it'll fully search our second bitmap, and this time it'll
1670  *  get an uptodate value of 8K as the maximum chunk size.  Then we'll get the
1671  *  right allocation the next loop through.
1672  */
1673 static inline u64 get_max_extent_size(const struct btrfs_free_space *entry)
1674 {
1675         if (entry->bitmap && entry->max_extent_size)
1676                 return entry->max_extent_size;
1677         return entry->bytes;
1678 }
1679
1680 /*
1681  * We want the largest entry to be leftmost, so this is inverted from what you'd
1682  * normally expect.
1683  */
1684 static bool entry_less(struct rb_node *node, const struct rb_node *parent)
1685 {
1686         const struct btrfs_free_space *entry, *exist;
1687
1688         entry = rb_entry(node, struct btrfs_free_space, bytes_index);
1689         exist = rb_entry(parent, struct btrfs_free_space, bytes_index);
1690         return get_max_extent_size(exist) < get_max_extent_size(entry);
1691 }
1692
1693 /*
1694  * searches the tree for the given offset.
1695  *
1696  * fuzzy - If this is set, then we are trying to make an allocation, and we just
1697  * want a section that has at least bytes size and comes at or after the given
1698  * offset.
1699  */
1700 static struct btrfs_free_space *
1701 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1702                    u64 offset, int bitmap_only, int fuzzy)
1703 {
1704         struct rb_node *n = ctl->free_space_offset.rb_node;
1705         struct btrfs_free_space *entry = NULL, *prev = NULL;
1706
1707         /* find entry that is closest to the 'offset' */
1708         while (n) {
1709                 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1710                 prev = entry;
1711
1712                 if (offset < entry->offset)
1713                         n = n->rb_left;
1714                 else if (offset > entry->offset)
1715                         n = n->rb_right;
1716                 else
1717                         break;
1718
1719                 entry = NULL;
1720         }
1721
1722         if (bitmap_only) {
1723                 if (!entry)
1724                         return NULL;
1725                 if (entry->bitmap)
1726                         return entry;
1727
1728                 /*
1729                  * bitmap entry and extent entry may share same offset,
1730                  * in that case, bitmap entry comes after extent entry.
1731                  */
1732                 n = rb_next(n);
1733                 if (!n)
1734                         return NULL;
1735                 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1736                 if (entry->offset != offset)
1737                         return NULL;
1738
1739                 WARN_ON(!entry->bitmap);
1740                 return entry;
1741         } else if (entry) {
1742                 if (entry->bitmap) {
1743                         /*
1744                          * if previous extent entry covers the offset,
1745                          * we should return it instead of the bitmap entry
1746                          */
1747                         n = rb_prev(&entry->offset_index);
1748                         if (n) {
1749                                 prev = rb_entry(n, struct btrfs_free_space,
1750                                                 offset_index);
1751                                 if (!prev->bitmap &&
1752                                     prev->offset + prev->bytes > offset)
1753                                         entry = prev;
1754                         }
1755                 }
1756                 return entry;
1757         }
1758
1759         if (!prev)
1760                 return NULL;
1761
1762         /* find last entry before the 'offset' */
1763         entry = prev;
1764         if (entry->offset > offset) {
1765                 n = rb_prev(&entry->offset_index);
1766                 if (n) {
1767                         entry = rb_entry(n, struct btrfs_free_space,
1768                                         offset_index);
1769                         ASSERT(entry->offset <= offset);
1770                 } else {
1771                         if (fuzzy)
1772                                 return entry;
1773                         else
1774                                 return NULL;
1775                 }
1776         }
1777
1778         if (entry->bitmap) {
1779                 n = rb_prev(&entry->offset_index);
1780                 if (n) {
1781                         prev = rb_entry(n, struct btrfs_free_space,
1782                                         offset_index);
1783                         if (!prev->bitmap &&
1784                             prev->offset + prev->bytes > offset)
1785                                 return prev;
1786                 }
1787                 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1788                         return entry;
1789         } else if (entry->offset + entry->bytes > offset)
1790                 return entry;
1791
1792         if (!fuzzy)
1793                 return NULL;
1794
1795         while (1) {
1796                 n = rb_next(&entry->offset_index);
1797                 if (!n)
1798                         return NULL;
1799                 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1800                 if (entry->bitmap) {
1801                         if (entry->offset + BITS_PER_BITMAP *
1802                             ctl->unit > offset)
1803                                 break;
1804                 } else {
1805                         if (entry->offset + entry->bytes > offset)
1806                                 break;
1807                 }
1808         }
1809         return entry;
1810 }
1811
1812 static inline void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1813                                      struct btrfs_free_space *info,
1814                                      bool update_stat)
1815 {
1816         rb_erase(&info->offset_index, &ctl->free_space_offset);
1817         rb_erase_cached(&info->bytes_index, &ctl->free_space_bytes);
1818         ctl->free_extents--;
1819
1820         if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1821                 ctl->discardable_extents[BTRFS_STAT_CURR]--;
1822                 ctl->discardable_bytes[BTRFS_STAT_CURR] -= info->bytes;
1823         }
1824
1825         if (update_stat)
1826                 ctl->free_space -= info->bytes;
1827 }
1828
1829 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1830                            struct btrfs_free_space *info)
1831 {
1832         int ret = 0;
1833
1834         ASSERT(info->bytes || info->bitmap);
1835         ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1836                                  &info->offset_index, (info->bitmap != NULL));
1837         if (ret)
1838                 return ret;
1839
1840         rb_add_cached(&info->bytes_index, &ctl->free_space_bytes, entry_less);
1841
1842         if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1843                 ctl->discardable_extents[BTRFS_STAT_CURR]++;
1844                 ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
1845         }
1846
1847         ctl->free_space += info->bytes;
1848         ctl->free_extents++;
1849         return ret;
1850 }
1851
1852 static void relink_bitmap_entry(struct btrfs_free_space_ctl *ctl,
1853                                 struct btrfs_free_space *info)
1854 {
1855         ASSERT(info->bitmap);
1856
1857         /*
1858          * If our entry is empty it's because we're on a cluster and we don't
1859          * want to re-link it into our ctl bytes index.
1860          */
1861         if (RB_EMPTY_NODE(&info->bytes_index))
1862                 return;
1863
1864         rb_erase_cached(&info->bytes_index, &ctl->free_space_bytes);
1865         rb_add_cached(&info->bytes_index, &ctl->free_space_bytes, entry_less);
1866 }
1867
1868 static inline void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1869                                      struct btrfs_free_space *info,
1870                                      u64 offset, u64 bytes, bool update_stat)
1871 {
1872         unsigned long start, count, end;
1873         int extent_delta = -1;
1874
1875         start = offset_to_bit(info->offset, ctl->unit, offset);
1876         count = bytes_to_bits(bytes, ctl->unit);
1877         end = start + count;
1878         ASSERT(end <= BITS_PER_BITMAP);
1879
1880         bitmap_clear(info->bitmap, start, count);
1881
1882         info->bytes -= bytes;
1883         if (info->max_extent_size > ctl->unit)
1884                 info->max_extent_size = 0;
1885
1886         relink_bitmap_entry(ctl, info);
1887
1888         if (start && test_bit(start - 1, info->bitmap))
1889                 extent_delta++;
1890
1891         if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
1892                 extent_delta++;
1893
1894         info->bitmap_extents += extent_delta;
1895         if (!btrfs_free_space_trimmed(info)) {
1896                 ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1897                 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
1898         }
1899
1900         if (update_stat)
1901                 ctl->free_space -= bytes;
1902 }
1903
1904 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1905                             struct btrfs_free_space *info, u64 offset,
1906                             u64 bytes)
1907 {
1908         unsigned long start, count, end;
1909         int extent_delta = 1;
1910
1911         start = offset_to_bit(info->offset, ctl->unit, offset);
1912         count = bytes_to_bits(bytes, ctl->unit);
1913         end = start + count;
1914         ASSERT(end <= BITS_PER_BITMAP);
1915
1916         bitmap_set(info->bitmap, start, count);
1917
1918         /*
1919          * We set some bytes, we have no idea what the max extent size is
1920          * anymore.
1921          */
1922         info->max_extent_size = 0;
1923         info->bytes += bytes;
1924         ctl->free_space += bytes;
1925
1926         relink_bitmap_entry(ctl, info);
1927
1928         if (start && test_bit(start - 1, info->bitmap))
1929                 extent_delta--;
1930
1931         if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
1932                 extent_delta--;
1933
1934         info->bitmap_extents += extent_delta;
1935         if (!btrfs_free_space_trimmed(info)) {
1936                 ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1937                 ctl->discardable_bytes[BTRFS_STAT_CURR] += bytes;
1938         }
1939 }
1940
1941 /*
1942  * If we can not find suitable extent, we will use bytes to record
1943  * the size of the max extent.
1944  */
1945 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1946                          struct btrfs_free_space *bitmap_info, u64 *offset,
1947                          u64 *bytes, bool for_alloc)
1948 {
1949         unsigned long found_bits = 0;
1950         unsigned long max_bits = 0;
1951         unsigned long bits, i;
1952         unsigned long next_zero;
1953         unsigned long extent_bits;
1954
1955         /*
1956          * Skip searching the bitmap if we don't have a contiguous section that
1957          * is large enough for this allocation.
1958          */
1959         if (for_alloc &&
1960             bitmap_info->max_extent_size &&
1961             bitmap_info->max_extent_size < *bytes) {
1962                 *bytes = bitmap_info->max_extent_size;
1963                 return -1;
1964         }
1965
1966         i = offset_to_bit(bitmap_info->offset, ctl->unit,
1967                           max_t(u64, *offset, bitmap_info->offset));
1968         bits = bytes_to_bits(*bytes, ctl->unit);
1969
1970         for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1971                 if (for_alloc && bits == 1) {
1972                         found_bits = 1;
1973                         break;
1974                 }
1975                 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1976                                                BITS_PER_BITMAP, i);
1977                 extent_bits = next_zero - i;
1978                 if (extent_bits >= bits) {
1979                         found_bits = extent_bits;
1980                         break;
1981                 } else if (extent_bits > max_bits) {
1982                         max_bits = extent_bits;
1983                 }
1984                 i = next_zero;
1985         }
1986
1987         if (found_bits) {
1988                 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1989                 *bytes = (u64)(found_bits) * ctl->unit;
1990                 return 0;
1991         }
1992
1993         *bytes = (u64)(max_bits) * ctl->unit;
1994         bitmap_info->max_extent_size = *bytes;
1995         relink_bitmap_entry(ctl, bitmap_info);
1996         return -1;
1997 }
1998
1999 /* Cache the size of the max extent in bytes */
2000 static struct btrfs_free_space *
2001 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
2002                 unsigned long align, u64 *max_extent_size, bool use_bytes_index)
2003 {
2004         struct btrfs_free_space *entry;
2005         struct rb_node *node;
2006         u64 tmp;
2007         u64 align_off;
2008         int ret;
2009
2010         if (!ctl->free_space_offset.rb_node)
2011                 goto out;
2012 again:
2013         if (use_bytes_index) {
2014                 node = rb_first_cached(&ctl->free_space_bytes);
2015         } else {
2016                 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset),
2017                                            0, 1);
2018                 if (!entry)
2019                         goto out;
2020                 node = &entry->offset_index;
2021         }
2022
2023         for (; node; node = rb_next(node)) {
2024                 if (use_bytes_index)
2025                         entry = rb_entry(node, struct btrfs_free_space,
2026                                          bytes_index);
2027                 else
2028                         entry = rb_entry(node, struct btrfs_free_space,
2029                                          offset_index);
2030
2031                 /*
2032                  * If we are using the bytes index then all subsequent entries
2033                  * in this tree are going to be < bytes, so simply set the max
2034                  * extent size and exit the loop.
2035                  *
2036                  * If we're using the offset index then we need to keep going
2037                  * through the rest of the tree.
2038                  */
2039                 if (entry->bytes < *bytes) {
2040                         *max_extent_size = max(get_max_extent_size(entry),
2041                                                *max_extent_size);
2042                         if (use_bytes_index)
2043                                 break;
2044                         continue;
2045                 }
2046
2047                 /* make sure the space returned is big enough
2048                  * to match our requested alignment
2049                  */
2050                 if (*bytes >= align) {
2051                         tmp = entry->offset - ctl->start + align - 1;
2052                         tmp = div64_u64(tmp, align);
2053                         tmp = tmp * align + ctl->start;
2054                         align_off = tmp - entry->offset;
2055                 } else {
2056                         align_off = 0;
2057                         tmp = entry->offset;
2058                 }
2059
2060                 /*
2061                  * We don't break here if we're using the bytes index because we
2062                  * may have another entry that has the correct alignment that is
2063                  * the right size, so we don't want to miss that possibility.
2064                  * At worst this adds another loop through the logic, but if we
2065                  * broke here we could prematurely ENOSPC.
2066                  */
2067                 if (entry->bytes < *bytes + align_off) {
2068                         *max_extent_size = max(get_max_extent_size(entry),
2069                                                *max_extent_size);
2070                         continue;
2071                 }
2072
2073                 if (entry->bitmap) {
2074                         struct rb_node *old_next = rb_next(node);
2075                         u64 size = *bytes;
2076
2077                         ret = search_bitmap(ctl, entry, &tmp, &size, true);
2078                         if (!ret) {
2079                                 *offset = tmp;
2080                                 *bytes = size;
2081                                 return entry;
2082                         } else {
2083                                 *max_extent_size =
2084                                         max(get_max_extent_size(entry),
2085                                             *max_extent_size);
2086                         }
2087
2088                         /*
2089                          * The bitmap may have gotten re-arranged in the space
2090                          * index here because the max_extent_size may have been
2091                          * updated.  Start from the beginning again if this
2092                          * happened.
2093                          */
2094                         if (use_bytes_index && old_next != rb_next(node))
2095                                 goto again;
2096                         continue;
2097                 }
2098
2099                 *offset = tmp;
2100                 *bytes = entry->bytes - align_off;
2101                 return entry;
2102         }
2103 out:
2104         return NULL;
2105 }
2106
2107 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
2108                            struct btrfs_free_space *info, u64 offset)
2109 {
2110         info->offset = offset_to_bitmap(ctl, offset);
2111         info->bytes = 0;
2112         info->bitmap_extents = 0;
2113         INIT_LIST_HEAD(&info->list);
2114         link_free_space(ctl, info);
2115         ctl->total_bitmaps++;
2116         recalculate_thresholds(ctl);
2117 }
2118
2119 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
2120                         struct btrfs_free_space *bitmap_info)
2121 {
2122         /*
2123          * Normally when this is called, the bitmap is completely empty. However,
2124          * if we are blowing up the free space cache for one reason or another
2125          * via __btrfs_remove_free_space_cache(), then it may not be freed and
2126          * we may leave stats on the table.
2127          */
2128         if (bitmap_info->bytes && !btrfs_free_space_trimmed(bitmap_info)) {
2129                 ctl->discardable_extents[BTRFS_STAT_CURR] -=
2130                         bitmap_info->bitmap_extents;
2131                 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bitmap_info->bytes;
2132
2133         }
2134         unlink_free_space(ctl, bitmap_info, true);
2135         kmem_cache_free(btrfs_free_space_bitmap_cachep, bitmap_info->bitmap);
2136         kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
2137         ctl->total_bitmaps--;
2138         recalculate_thresholds(ctl);
2139 }
2140
2141 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
2142                               struct btrfs_free_space *bitmap_info,
2143                               u64 *offset, u64 *bytes)
2144 {
2145         u64 end;
2146         u64 search_start, search_bytes;
2147         int ret;
2148
2149 again:
2150         end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
2151
2152         /*
2153          * We need to search for bits in this bitmap.  We could only cover some
2154          * of the extent in this bitmap thanks to how we add space, so we need
2155          * to search for as much as it as we can and clear that amount, and then
2156          * go searching for the next bit.
2157          */
2158         search_start = *offset;
2159         search_bytes = ctl->unit;
2160         search_bytes = min(search_bytes, end - search_start + 1);
2161         ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes,
2162                             false);
2163         if (ret < 0 || search_start != *offset)
2164                 return -EINVAL;
2165
2166         /* We may have found more bits than what we need */
2167         search_bytes = min(search_bytes, *bytes);
2168
2169         /* Cannot clear past the end of the bitmap */
2170         search_bytes = min(search_bytes, end - search_start + 1);
2171
2172         bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes, true);
2173         *offset += search_bytes;
2174         *bytes -= search_bytes;
2175
2176         if (*bytes) {
2177                 struct rb_node *next = rb_next(&bitmap_info->offset_index);
2178                 if (!bitmap_info->bytes)
2179                         free_bitmap(ctl, bitmap_info);
2180
2181                 /*
2182                  * no entry after this bitmap, but we still have bytes to
2183                  * remove, so something has gone wrong.
2184                  */
2185                 if (!next)
2186                         return -EINVAL;
2187
2188                 bitmap_info = rb_entry(next, struct btrfs_free_space,
2189                                        offset_index);
2190
2191                 /*
2192                  * if the next entry isn't a bitmap we need to return to let the
2193                  * extent stuff do its work.
2194                  */
2195                 if (!bitmap_info->bitmap)
2196                         return -EAGAIN;
2197
2198                 /*
2199                  * Ok the next item is a bitmap, but it may not actually hold
2200                  * the information for the rest of this free space stuff, so
2201                  * look for it, and if we don't find it return so we can try
2202                  * everything over again.
2203                  */
2204                 search_start = *offset;
2205                 search_bytes = ctl->unit;
2206                 ret = search_bitmap(ctl, bitmap_info, &search_start,
2207                                     &search_bytes, false);
2208                 if (ret < 0 || search_start != *offset)
2209                         return -EAGAIN;
2210
2211                 goto again;
2212         } else if (!bitmap_info->bytes)
2213                 free_bitmap(ctl, bitmap_info);
2214
2215         return 0;
2216 }
2217
2218 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
2219                                struct btrfs_free_space *info, u64 offset,
2220                                u64 bytes, enum btrfs_trim_state trim_state)
2221 {
2222         u64 bytes_to_set = 0;
2223         u64 end;
2224
2225         /*
2226          * This is a tradeoff to make bitmap trim state minimal.  We mark the
2227          * whole bitmap untrimmed if at any point we add untrimmed regions.
2228          */
2229         if (trim_state == BTRFS_TRIM_STATE_UNTRIMMED) {
2230                 if (btrfs_free_space_trimmed(info)) {
2231                         ctl->discardable_extents[BTRFS_STAT_CURR] +=
2232                                 info->bitmap_extents;
2233                         ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
2234                 }
2235                 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2236         }
2237
2238         end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
2239
2240         bytes_to_set = min(end - offset, bytes);
2241
2242         bitmap_set_bits(ctl, info, offset, bytes_to_set);
2243
2244         return bytes_to_set;
2245
2246 }
2247
2248 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
2249                       struct btrfs_free_space *info)
2250 {
2251         struct btrfs_block_group *block_group = ctl->block_group;
2252         struct btrfs_fs_info *fs_info = block_group->fs_info;
2253         bool forced = false;
2254
2255 #ifdef CONFIG_BTRFS_DEBUG
2256         if (btrfs_should_fragment_free_space(block_group))
2257                 forced = true;
2258 #endif
2259
2260         /* This is a way to reclaim large regions from the bitmaps. */
2261         if (!forced && info->bytes >= FORCE_EXTENT_THRESHOLD)
2262                 return false;
2263
2264         /*
2265          * If we are below the extents threshold then we can add this as an
2266          * extent, and don't have to deal with the bitmap
2267          */
2268         if (!forced && ctl->free_extents < ctl->extents_thresh) {
2269                 /*
2270                  * If this block group has some small extents we don't want to
2271                  * use up all of our free slots in the cache with them, we want
2272                  * to reserve them to larger extents, however if we have plenty
2273                  * of cache left then go ahead an dadd them, no sense in adding
2274                  * the overhead of a bitmap if we don't have to.
2275                  */
2276                 if (info->bytes <= fs_info->sectorsize * 8) {
2277                         if (ctl->free_extents * 3 <= ctl->extents_thresh)
2278                                 return false;
2279                 } else {
2280                         return false;
2281                 }
2282         }
2283
2284         /*
2285          * The original block groups from mkfs can be really small, like 8
2286          * megabytes, so don't bother with a bitmap for those entries.  However
2287          * some block groups can be smaller than what a bitmap would cover but
2288          * are still large enough that they could overflow the 32k memory limit,
2289          * so allow those block groups to still be allowed to have a bitmap
2290          * entry.
2291          */
2292         if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->length)
2293                 return false;
2294
2295         return true;
2296 }
2297
2298 static const struct btrfs_free_space_op free_space_op = {
2299         .use_bitmap             = use_bitmap,
2300 };
2301
2302 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
2303                               struct btrfs_free_space *info)
2304 {
2305         struct btrfs_free_space *bitmap_info;
2306         struct btrfs_block_group *block_group = NULL;
2307         int added = 0;
2308         u64 bytes, offset, bytes_added;
2309         enum btrfs_trim_state trim_state;
2310         int ret;
2311
2312         bytes = info->bytes;
2313         offset = info->offset;
2314         trim_state = info->trim_state;
2315
2316         if (!ctl->op->use_bitmap(ctl, info))
2317                 return 0;
2318
2319         if (ctl->op == &free_space_op)
2320                 block_group = ctl->block_group;
2321 again:
2322         /*
2323          * Since we link bitmaps right into the cluster we need to see if we
2324          * have a cluster here, and if so and it has our bitmap we need to add
2325          * the free space to that bitmap.
2326          */
2327         if (block_group && !list_empty(&block_group->cluster_list)) {
2328                 struct btrfs_free_cluster *cluster;
2329                 struct rb_node *node;
2330                 struct btrfs_free_space *entry;
2331
2332                 cluster = list_entry(block_group->cluster_list.next,
2333                                      struct btrfs_free_cluster,
2334                                      block_group_list);
2335                 spin_lock(&cluster->lock);
2336                 node = rb_first(&cluster->root);
2337                 if (!node) {
2338                         spin_unlock(&cluster->lock);
2339                         goto no_cluster_bitmap;
2340                 }
2341
2342                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2343                 if (!entry->bitmap) {
2344                         spin_unlock(&cluster->lock);
2345                         goto no_cluster_bitmap;
2346                 }
2347
2348                 if (entry->offset == offset_to_bitmap(ctl, offset)) {
2349                         bytes_added = add_bytes_to_bitmap(ctl, entry, offset,
2350                                                           bytes, trim_state);
2351                         bytes -= bytes_added;
2352                         offset += bytes_added;
2353                 }
2354                 spin_unlock(&cluster->lock);
2355                 if (!bytes) {
2356                         ret = 1;
2357                         goto out;
2358                 }
2359         }
2360
2361 no_cluster_bitmap:
2362         bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2363                                          1, 0);
2364         if (!bitmap_info) {
2365                 ASSERT(added == 0);
2366                 goto new_bitmap;
2367         }
2368
2369         bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes,
2370                                           trim_state);
2371         bytes -= bytes_added;
2372         offset += bytes_added;
2373         added = 0;
2374
2375         if (!bytes) {
2376                 ret = 1;
2377                 goto out;
2378         } else
2379                 goto again;
2380
2381 new_bitmap:
2382         if (info && info->bitmap) {
2383                 add_new_bitmap(ctl, info, offset);
2384                 added = 1;
2385                 info = NULL;
2386                 goto again;
2387         } else {
2388                 spin_unlock(&ctl->tree_lock);
2389
2390                 /* no pre-allocated info, allocate a new one */
2391                 if (!info) {
2392                         info = kmem_cache_zalloc(btrfs_free_space_cachep,
2393                                                  GFP_NOFS);
2394                         if (!info) {
2395                                 spin_lock(&ctl->tree_lock);
2396                                 ret = -ENOMEM;
2397                                 goto out;
2398                         }
2399                 }
2400
2401                 /* allocate the bitmap */
2402                 info->bitmap = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep,
2403                                                  GFP_NOFS);
2404                 info->trim_state = BTRFS_TRIM_STATE_TRIMMED;
2405                 spin_lock(&ctl->tree_lock);
2406                 if (!info->bitmap) {
2407                         ret = -ENOMEM;
2408                         goto out;
2409                 }
2410                 goto again;
2411         }
2412
2413 out:
2414         if (info) {
2415                 if (info->bitmap)
2416                         kmem_cache_free(btrfs_free_space_bitmap_cachep,
2417                                         info->bitmap);
2418                 kmem_cache_free(btrfs_free_space_cachep, info);
2419         }
2420
2421         return ret;
2422 }
2423
2424 /*
2425  * Free space merging rules:
2426  *  1) Merge trimmed areas together
2427  *  2) Let untrimmed areas coalesce with trimmed areas
2428  *  3) Always pull neighboring regions from bitmaps
2429  *
2430  * The above rules are for when we merge free space based on btrfs_trim_state.
2431  * Rules 2 and 3 are subtle because they are suboptimal, but are done for the
2432  * same reason: to promote larger extent regions which makes life easier for
2433  * find_free_extent().  Rule 2 enables coalescing based on the common path
2434  * being returning free space from btrfs_finish_extent_commit().  So when free
2435  * space is trimmed, it will prevent aggregating trimmed new region and
2436  * untrimmed regions in the rb_tree.  Rule 3 is purely to obtain larger extents
2437  * and provide find_free_extent() with the largest extents possible hoping for
2438  * the reuse path.
2439  */
2440 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
2441                           struct btrfs_free_space *info, bool update_stat)
2442 {
2443         struct btrfs_free_space *left_info = NULL;
2444         struct btrfs_free_space *right_info;
2445         bool merged = false;
2446         u64 offset = info->offset;
2447         u64 bytes = info->bytes;
2448         const bool is_trimmed = btrfs_free_space_trimmed(info);
2449
2450         /*
2451          * first we want to see if there is free space adjacent to the range we
2452          * are adding, if there is remove that struct and add a new one to
2453          * cover the entire range
2454          */
2455         right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
2456         if (right_info && rb_prev(&right_info->offset_index))
2457                 left_info = rb_entry(rb_prev(&right_info->offset_index),
2458                                      struct btrfs_free_space, offset_index);
2459         else if (!right_info)
2460                 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
2461
2462         /* See try_merge_free_space() comment. */
2463         if (right_info && !right_info->bitmap &&
2464             (!is_trimmed || btrfs_free_space_trimmed(right_info))) {
2465                 unlink_free_space(ctl, right_info, update_stat);
2466                 info->bytes += right_info->bytes;
2467                 kmem_cache_free(btrfs_free_space_cachep, right_info);
2468                 merged = true;
2469         }
2470
2471         /* See try_merge_free_space() comment. */
2472         if (left_info && !left_info->bitmap &&
2473             left_info->offset + left_info->bytes == offset &&
2474             (!is_trimmed || btrfs_free_space_trimmed(left_info))) {
2475                 unlink_free_space(ctl, left_info, update_stat);
2476                 info->offset = left_info->offset;
2477                 info->bytes += left_info->bytes;
2478                 kmem_cache_free(btrfs_free_space_cachep, left_info);
2479                 merged = true;
2480         }
2481
2482         return merged;
2483 }
2484
2485 static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl,
2486                                      struct btrfs_free_space *info,
2487                                      bool update_stat)
2488 {
2489         struct btrfs_free_space *bitmap;
2490         unsigned long i;
2491         unsigned long j;
2492         const u64 end = info->offset + info->bytes;
2493         const u64 bitmap_offset = offset_to_bitmap(ctl, end);
2494         u64 bytes;
2495
2496         bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2497         if (!bitmap)
2498                 return false;
2499
2500         i = offset_to_bit(bitmap->offset, ctl->unit, end);
2501         j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i);
2502         if (j == i)
2503                 return false;
2504         bytes = (j - i) * ctl->unit;
2505         info->bytes += bytes;
2506
2507         /* See try_merge_free_space() comment. */
2508         if (!btrfs_free_space_trimmed(bitmap))
2509                 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2510
2511         bitmap_clear_bits(ctl, bitmap, end, bytes, update_stat);
2512
2513         if (!bitmap->bytes)
2514                 free_bitmap(ctl, bitmap);
2515
2516         return true;
2517 }
2518
2519 static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl,
2520                                        struct btrfs_free_space *info,
2521                                        bool update_stat)
2522 {
2523         struct btrfs_free_space *bitmap;
2524         u64 bitmap_offset;
2525         unsigned long i;
2526         unsigned long j;
2527         unsigned long prev_j;
2528         u64 bytes;
2529
2530         bitmap_offset = offset_to_bitmap(ctl, info->offset);
2531         /* If we're on a boundary, try the previous logical bitmap. */
2532         if (bitmap_offset == info->offset) {
2533                 if (info->offset == 0)
2534                         return false;
2535                 bitmap_offset = offset_to_bitmap(ctl, info->offset - 1);
2536         }
2537
2538         bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2539         if (!bitmap)
2540                 return false;
2541
2542         i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1;
2543         j = 0;
2544         prev_j = (unsigned long)-1;
2545         for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) {
2546                 if (j > i)
2547                         break;
2548                 prev_j = j;
2549         }
2550         if (prev_j == i)
2551                 return false;
2552
2553         if (prev_j == (unsigned long)-1)
2554                 bytes = (i + 1) * ctl->unit;
2555         else
2556                 bytes = (i - prev_j) * ctl->unit;
2557
2558         info->offset -= bytes;
2559         info->bytes += bytes;
2560
2561         /* See try_merge_free_space() comment. */
2562         if (!btrfs_free_space_trimmed(bitmap))
2563                 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2564
2565         bitmap_clear_bits(ctl, bitmap, info->offset, bytes, update_stat);
2566
2567         if (!bitmap->bytes)
2568                 free_bitmap(ctl, bitmap);
2569
2570         return true;
2571 }
2572
2573 /*
2574  * We prefer always to allocate from extent entries, both for clustered and
2575  * non-clustered allocation requests. So when attempting to add a new extent
2576  * entry, try to see if there's adjacent free space in bitmap entries, and if
2577  * there is, migrate that space from the bitmaps to the extent.
2578  * Like this we get better chances of satisfying space allocation requests
2579  * because we attempt to satisfy them based on a single cache entry, and never
2580  * on 2 or more entries - even if the entries represent a contiguous free space
2581  * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
2582  * ends).
2583  */
2584 static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl,
2585                               struct btrfs_free_space *info,
2586                               bool update_stat)
2587 {
2588         /*
2589          * Only work with disconnected entries, as we can change their offset,
2590          * and must be extent entries.
2591          */
2592         ASSERT(!info->bitmap);
2593         ASSERT(RB_EMPTY_NODE(&info->offset_index));
2594
2595         if (ctl->total_bitmaps > 0) {
2596                 bool stole_end;
2597                 bool stole_front = false;
2598
2599                 stole_end = steal_from_bitmap_to_end(ctl, info, update_stat);
2600                 if (ctl->total_bitmaps > 0)
2601                         stole_front = steal_from_bitmap_to_front(ctl, info,
2602                                                                  update_stat);
2603
2604                 if (stole_end || stole_front)
2605                         try_merge_free_space(ctl, info, update_stat);
2606         }
2607 }
2608
2609 int __btrfs_add_free_space(struct btrfs_block_group *block_group,
2610                            u64 offset, u64 bytes,
2611                            enum btrfs_trim_state trim_state)
2612 {
2613         struct btrfs_fs_info *fs_info = block_group->fs_info;
2614         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2615         struct btrfs_free_space *info;
2616         int ret = 0;
2617         u64 filter_bytes = bytes;
2618
2619         ASSERT(!btrfs_is_zoned(fs_info));
2620
2621         info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2622         if (!info)
2623                 return -ENOMEM;
2624
2625         info->offset = offset;
2626         info->bytes = bytes;
2627         info->trim_state = trim_state;
2628         RB_CLEAR_NODE(&info->offset_index);
2629         RB_CLEAR_NODE(&info->bytes_index);
2630
2631         spin_lock(&ctl->tree_lock);
2632
2633         if (try_merge_free_space(ctl, info, true))
2634                 goto link;
2635
2636         /*
2637          * There was no extent directly to the left or right of this new
2638          * extent then we know we're going to have to allocate a new extent, so
2639          * before we do that see if we need to drop this into a bitmap
2640          */
2641         ret = insert_into_bitmap(ctl, info);
2642         if (ret < 0) {
2643                 goto out;
2644         } else if (ret) {
2645                 ret = 0;
2646                 goto out;
2647         }
2648 link:
2649         /*
2650          * Only steal free space from adjacent bitmaps if we're sure we're not
2651          * going to add the new free space to existing bitmap entries - because
2652          * that would mean unnecessary work that would be reverted. Therefore
2653          * attempt to steal space from bitmaps if we're adding an extent entry.
2654          */
2655         steal_from_bitmap(ctl, info, true);
2656
2657         filter_bytes = max(filter_bytes, info->bytes);
2658
2659         ret = link_free_space(ctl, info);
2660         if (ret)
2661                 kmem_cache_free(btrfs_free_space_cachep, info);
2662 out:
2663         btrfs_discard_update_discardable(block_group);
2664         spin_unlock(&ctl->tree_lock);
2665
2666         if (ret) {
2667                 btrfs_crit(fs_info, "unable to add free space :%d", ret);
2668                 ASSERT(ret != -EEXIST);
2669         }
2670
2671         if (trim_state != BTRFS_TRIM_STATE_TRIMMED) {
2672                 btrfs_discard_check_filter(block_group, filter_bytes);
2673                 btrfs_discard_queue_work(&fs_info->discard_ctl, block_group);
2674         }
2675
2676         return ret;
2677 }
2678
2679 static int __btrfs_add_free_space_zoned(struct btrfs_block_group *block_group,
2680                                         u64 bytenr, u64 size, bool used)
2681 {
2682         struct btrfs_space_info *sinfo = block_group->space_info;
2683         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2684         u64 offset = bytenr - block_group->start;
2685         u64 to_free, to_unusable;
2686         int bg_reclaim_threshold = 0;
2687         bool initial = (size == block_group->length);
2688         u64 reclaimable_unusable;
2689
2690         WARN_ON(!initial && offset + size > block_group->zone_capacity);
2691
2692         if (!initial)
2693                 bg_reclaim_threshold = READ_ONCE(sinfo->bg_reclaim_threshold);
2694
2695         spin_lock(&ctl->tree_lock);
2696         if (!used)
2697                 to_free = size;
2698         else if (initial)
2699                 to_free = block_group->zone_capacity;
2700         else if (offset >= block_group->alloc_offset)
2701                 to_free = size;
2702         else if (offset + size <= block_group->alloc_offset)
2703                 to_free = 0;
2704         else
2705                 to_free = offset + size - block_group->alloc_offset;
2706         to_unusable = size - to_free;
2707
2708         ctl->free_space += to_free;
2709         /*
2710          * If the block group is read-only, we should account freed space into
2711          * bytes_readonly.
2712          */
2713         if (!block_group->ro)
2714                 block_group->zone_unusable += to_unusable;
2715         spin_unlock(&ctl->tree_lock);
2716         if (!used) {
2717                 spin_lock(&block_group->lock);
2718                 block_group->alloc_offset -= size;
2719                 spin_unlock(&block_group->lock);
2720         }
2721
2722         reclaimable_unusable = block_group->zone_unusable -
2723                                (block_group->length - block_group->zone_capacity);
2724         /* All the region is now unusable. Mark it as unused and reclaim */
2725         if (block_group->zone_unusable == block_group->length) {
2726                 btrfs_mark_bg_unused(block_group);
2727         } else if (bg_reclaim_threshold &&
2728                    reclaimable_unusable >=
2729                    mult_perc(block_group->zone_capacity, bg_reclaim_threshold)) {
2730                 btrfs_mark_bg_to_reclaim(block_group);
2731         }
2732
2733         return 0;
2734 }
2735
2736 int btrfs_add_free_space(struct btrfs_block_group *block_group,
2737                          u64 bytenr, u64 size)
2738 {
2739         enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2740
2741         if (btrfs_is_zoned(block_group->fs_info))
2742                 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2743                                                     true);
2744
2745         if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC))
2746                 trim_state = BTRFS_TRIM_STATE_TRIMMED;
2747
2748         return __btrfs_add_free_space(block_group, bytenr, size, trim_state);
2749 }
2750
2751 int btrfs_add_free_space_unused(struct btrfs_block_group *block_group,
2752                                 u64 bytenr, u64 size)
2753 {
2754         if (btrfs_is_zoned(block_group->fs_info))
2755                 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2756                                                     false);
2757
2758         return btrfs_add_free_space(block_group, bytenr, size);
2759 }
2760
2761 /*
2762  * This is a subtle distinction because when adding free space back in general,
2763  * we want it to be added as untrimmed for async. But in the case where we add
2764  * it on loading of a block group, we want to consider it trimmed.
2765  */
2766 int btrfs_add_free_space_async_trimmed(struct btrfs_block_group *block_group,
2767                                        u64 bytenr, u64 size)
2768 {
2769         enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2770
2771         if (btrfs_is_zoned(block_group->fs_info))
2772                 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2773                                                     true);
2774
2775         if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC) ||
2776             btrfs_test_opt(block_group->fs_info, DISCARD_ASYNC))
2777                 trim_state = BTRFS_TRIM_STATE_TRIMMED;
2778
2779         return __btrfs_add_free_space(block_group, bytenr, size, trim_state);
2780 }
2781
2782 int btrfs_remove_free_space(struct btrfs_block_group *block_group,
2783                             u64 offset, u64 bytes)
2784 {
2785         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2786         struct btrfs_free_space *info;
2787         int ret;
2788         bool re_search = false;
2789
2790         if (btrfs_is_zoned(block_group->fs_info)) {
2791                 /*
2792                  * This can happen with conventional zones when replaying log.
2793                  * Since the allocation info of tree-log nodes are not recorded
2794                  * to the extent-tree, calculate_alloc_pointer() failed to
2795                  * advance the allocation pointer after last allocated tree log
2796                  * node blocks.
2797                  *
2798                  * This function is called from
2799                  * btrfs_pin_extent_for_log_replay() when replaying the log.
2800                  * Advance the pointer not to overwrite the tree-log nodes.
2801                  */
2802                 if (block_group->start + block_group->alloc_offset <
2803                     offset + bytes) {
2804                         block_group->alloc_offset =
2805                                 offset + bytes - block_group->start;
2806                 }
2807                 return 0;
2808         }
2809
2810         spin_lock(&ctl->tree_lock);
2811
2812 again:
2813         ret = 0;
2814         if (!bytes)
2815                 goto out_lock;
2816
2817         info = tree_search_offset(ctl, offset, 0, 0);
2818         if (!info) {
2819                 /*
2820                  * oops didn't find an extent that matched the space we wanted
2821                  * to remove, look for a bitmap instead
2822                  */
2823                 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2824                                           1, 0);
2825                 if (!info) {
2826                         /*
2827                          * If we found a partial bit of our free space in a
2828                          * bitmap but then couldn't find the other part this may
2829                          * be a problem, so WARN about it.
2830                          */
2831                         WARN_ON(re_search);
2832                         goto out_lock;
2833                 }
2834         }
2835
2836         re_search = false;
2837         if (!info->bitmap) {
2838                 unlink_free_space(ctl, info, true);
2839                 if (offset == info->offset) {
2840                         u64 to_free = min(bytes, info->bytes);
2841
2842                         info->bytes -= to_free;
2843                         info->offset += to_free;
2844                         if (info->bytes) {
2845                                 ret = link_free_space(ctl, info);
2846                                 WARN_ON(ret);
2847                         } else {
2848                                 kmem_cache_free(btrfs_free_space_cachep, info);
2849                         }
2850
2851                         offset += to_free;
2852                         bytes -= to_free;
2853                         goto again;
2854                 } else {
2855                         u64 old_end = info->bytes + info->offset;
2856
2857                         info->bytes = offset - info->offset;
2858                         ret = link_free_space(ctl, info);
2859                         WARN_ON(ret);
2860                         if (ret)
2861                                 goto out_lock;
2862
2863                         /* Not enough bytes in this entry to satisfy us */
2864                         if (old_end < offset + bytes) {
2865                                 bytes -= old_end - offset;
2866                                 offset = old_end;
2867                                 goto again;
2868                         } else if (old_end == offset + bytes) {
2869                                 /* all done */
2870                                 goto out_lock;
2871                         }
2872                         spin_unlock(&ctl->tree_lock);
2873
2874                         ret = __btrfs_add_free_space(block_group,
2875                                                      offset + bytes,
2876                                                      old_end - (offset + bytes),
2877                                                      info->trim_state);
2878                         WARN_ON(ret);
2879                         goto out;
2880                 }
2881         }
2882
2883         ret = remove_from_bitmap(ctl, info, &offset, &bytes);
2884         if (ret == -EAGAIN) {
2885                 re_search = true;
2886                 goto again;
2887         }
2888 out_lock:
2889         btrfs_discard_update_discardable(block_group);
2890         spin_unlock(&ctl->tree_lock);
2891 out:
2892         return ret;
2893 }
2894
2895 void btrfs_dump_free_space(struct btrfs_block_group *block_group,
2896                            u64 bytes)
2897 {
2898         struct btrfs_fs_info *fs_info = block_group->fs_info;
2899         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2900         struct btrfs_free_space *info;
2901         struct rb_node *n;
2902         int count = 0;
2903
2904         /*
2905          * Zoned btrfs does not use free space tree and cluster. Just print
2906          * out the free space after the allocation offset.
2907          */
2908         if (btrfs_is_zoned(fs_info)) {
2909                 btrfs_info(fs_info, "free space %llu active %d",
2910                            block_group->zone_capacity - block_group->alloc_offset,
2911                            test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE,
2912                                     &block_group->runtime_flags));
2913                 return;
2914         }
2915
2916         spin_lock(&ctl->tree_lock);
2917         for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2918                 info = rb_entry(n, struct btrfs_free_space, offset_index);
2919                 if (info->bytes >= bytes && !block_group->ro)
2920                         count++;
2921                 btrfs_crit(fs_info, "entry offset %llu, bytes %llu, bitmap %s",
2922                            info->offset, info->bytes,
2923                        (info->bitmap) ? "yes" : "no");
2924         }
2925         spin_unlock(&ctl->tree_lock);
2926         btrfs_info(fs_info, "block group has cluster?: %s",
2927                list_empty(&block_group->cluster_list) ? "no" : "yes");
2928         btrfs_info(fs_info,
2929                    "%d blocks of free space at or bigger than bytes is", count);
2930 }
2931
2932 void btrfs_init_free_space_ctl(struct btrfs_block_group *block_group,
2933                                struct btrfs_free_space_ctl *ctl)
2934 {
2935         struct btrfs_fs_info *fs_info = block_group->fs_info;
2936
2937         spin_lock_init(&ctl->tree_lock);
2938         ctl->unit = fs_info->sectorsize;
2939         ctl->start = block_group->start;
2940         ctl->block_group = block_group;
2941         ctl->op = &free_space_op;
2942         ctl->free_space_bytes = RB_ROOT_CACHED;
2943         INIT_LIST_HEAD(&ctl->trimming_ranges);
2944         mutex_init(&ctl->cache_writeout_mutex);
2945
2946         /*
2947          * we only want to have 32k of ram per block group for keeping
2948          * track of free space, and if we pass 1/2 of that we want to
2949          * start converting things over to using bitmaps
2950          */
2951         ctl->extents_thresh = (SZ_32K / 2) / sizeof(struct btrfs_free_space);
2952 }
2953
2954 /*
2955  * for a given cluster, put all of its extents back into the free
2956  * space cache.  If the block group passed doesn't match the block group
2957  * pointed to by the cluster, someone else raced in and freed the
2958  * cluster already.  In that case, we just return without changing anything
2959  */
2960 static void __btrfs_return_cluster_to_free_space(
2961                              struct btrfs_block_group *block_group,
2962                              struct btrfs_free_cluster *cluster)
2963 {
2964         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2965         struct btrfs_free_space *entry;
2966         struct rb_node *node;
2967
2968         spin_lock(&cluster->lock);
2969         if (cluster->block_group != block_group) {
2970                 spin_unlock(&cluster->lock);
2971                 return;
2972         }
2973
2974         cluster->block_group = NULL;
2975         cluster->window_start = 0;
2976         list_del_init(&cluster->block_group_list);
2977
2978         node = rb_first(&cluster->root);
2979         while (node) {
2980                 bool bitmap;
2981
2982                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2983                 node = rb_next(&entry->offset_index);
2984                 rb_erase(&entry->offset_index, &cluster->root);
2985                 RB_CLEAR_NODE(&entry->offset_index);
2986
2987                 bitmap = (entry->bitmap != NULL);
2988                 if (!bitmap) {
2989                         /* Merging treats extents as if they were new */
2990                         if (!btrfs_free_space_trimmed(entry)) {
2991                                 ctl->discardable_extents[BTRFS_STAT_CURR]--;
2992                                 ctl->discardable_bytes[BTRFS_STAT_CURR] -=
2993                                         entry->bytes;
2994                         }
2995
2996                         try_merge_free_space(ctl, entry, false);
2997                         steal_from_bitmap(ctl, entry, false);
2998
2999                         /* As we insert directly, update these statistics */
3000                         if (!btrfs_free_space_trimmed(entry)) {
3001                                 ctl->discardable_extents[BTRFS_STAT_CURR]++;
3002                                 ctl->discardable_bytes[BTRFS_STAT_CURR] +=
3003                                         entry->bytes;
3004                         }
3005                 }
3006                 tree_insert_offset(&ctl->free_space_offset,
3007                                    entry->offset, &entry->offset_index, bitmap);
3008                 rb_add_cached(&entry->bytes_index, &ctl->free_space_bytes,
3009                               entry_less);
3010         }
3011         cluster->root = RB_ROOT;
3012         spin_unlock(&cluster->lock);
3013         btrfs_put_block_group(block_group);
3014 }
3015
3016 void btrfs_remove_free_space_cache(struct btrfs_block_group *block_group)
3017 {
3018         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3019         struct btrfs_free_cluster *cluster;
3020         struct list_head *head;
3021
3022         spin_lock(&ctl->tree_lock);
3023         while ((head = block_group->cluster_list.next) !=
3024                &block_group->cluster_list) {
3025                 cluster = list_entry(head, struct btrfs_free_cluster,
3026                                      block_group_list);
3027
3028                 WARN_ON(cluster->block_group != block_group);
3029                 __btrfs_return_cluster_to_free_space(block_group, cluster);
3030
3031                 cond_resched_lock(&ctl->tree_lock);
3032         }
3033         __btrfs_remove_free_space_cache(ctl);
3034         btrfs_discard_update_discardable(block_group);
3035         spin_unlock(&ctl->tree_lock);
3036
3037 }
3038
3039 /*
3040  * Walk @block_group's free space rb_tree to determine if everything is trimmed.
3041  */
3042 bool btrfs_is_free_space_trimmed(struct btrfs_block_group *block_group)
3043 {
3044         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3045         struct btrfs_free_space *info;
3046         struct rb_node *node;
3047         bool ret = true;
3048
3049         spin_lock(&ctl->tree_lock);
3050         node = rb_first(&ctl->free_space_offset);
3051
3052         while (node) {
3053                 info = rb_entry(node, struct btrfs_free_space, offset_index);
3054
3055                 if (!btrfs_free_space_trimmed(info)) {
3056                         ret = false;
3057                         break;
3058                 }
3059
3060                 node = rb_next(node);
3061         }
3062
3063         spin_unlock(&ctl->tree_lock);
3064         return ret;
3065 }
3066
3067 u64 btrfs_find_space_for_alloc(struct btrfs_block_group *block_group,
3068                                u64 offset, u64 bytes, u64 empty_size,
3069                                u64 *max_extent_size)
3070 {
3071         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3072         struct btrfs_discard_ctl *discard_ctl =
3073                                         &block_group->fs_info->discard_ctl;
3074         struct btrfs_free_space *entry = NULL;
3075         u64 bytes_search = bytes + empty_size;
3076         u64 ret = 0;
3077         u64 align_gap = 0;
3078         u64 align_gap_len = 0;
3079         enum btrfs_trim_state align_gap_trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3080         bool use_bytes_index = (offset == block_group->start);
3081
3082         ASSERT(!btrfs_is_zoned(block_group->fs_info));
3083
3084         spin_lock(&ctl->tree_lock);
3085         entry = find_free_space(ctl, &offset, &bytes_search,
3086                                 block_group->full_stripe_len, max_extent_size,
3087                                 use_bytes_index);
3088         if (!entry)
3089                 goto out;
3090
3091         ret = offset;
3092         if (entry->bitmap) {
3093                 bitmap_clear_bits(ctl, entry, offset, bytes, true);
3094
3095                 if (!btrfs_free_space_trimmed(entry))
3096                         atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3097
3098                 if (!entry->bytes)
3099                         free_bitmap(ctl, entry);
3100         } else {
3101                 unlink_free_space(ctl, entry, true);
3102                 align_gap_len = offset - entry->offset;
3103                 align_gap = entry->offset;
3104                 align_gap_trim_state = entry->trim_state;
3105
3106                 if (!btrfs_free_space_trimmed(entry))
3107                         atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3108
3109                 entry->offset = offset + bytes;
3110                 WARN_ON(entry->bytes < bytes + align_gap_len);
3111
3112                 entry->bytes -= bytes + align_gap_len;
3113                 if (!entry->bytes)
3114                         kmem_cache_free(btrfs_free_space_cachep, entry);
3115                 else
3116                         link_free_space(ctl, entry);
3117         }
3118 out:
3119         btrfs_discard_update_discardable(block_group);
3120         spin_unlock(&ctl->tree_lock);
3121
3122         if (align_gap_len)
3123                 __btrfs_add_free_space(block_group, align_gap, align_gap_len,
3124                                        align_gap_trim_state);
3125         return ret;
3126 }
3127
3128 /*
3129  * given a cluster, put all of its extents back into the free space
3130  * cache.  If a block group is passed, this function will only free
3131  * a cluster that belongs to the passed block group.
3132  *
3133  * Otherwise, it'll get a reference on the block group pointed to by the
3134  * cluster and remove the cluster from it.
3135  */
3136 void btrfs_return_cluster_to_free_space(
3137                                struct btrfs_block_group *block_group,
3138                                struct btrfs_free_cluster *cluster)
3139 {
3140         struct btrfs_free_space_ctl *ctl;
3141
3142         /* first, get a safe pointer to the block group */
3143         spin_lock(&cluster->lock);
3144         if (!block_group) {
3145                 block_group = cluster->block_group;
3146                 if (!block_group) {
3147                         spin_unlock(&cluster->lock);
3148                         return;
3149                 }
3150         } else if (cluster->block_group != block_group) {
3151                 /* someone else has already freed it don't redo their work */
3152                 spin_unlock(&cluster->lock);
3153                 return;
3154         }
3155         btrfs_get_block_group(block_group);
3156         spin_unlock(&cluster->lock);
3157
3158         ctl = block_group->free_space_ctl;
3159
3160         /* now return any extents the cluster had on it */
3161         spin_lock(&ctl->tree_lock);
3162         __btrfs_return_cluster_to_free_space(block_group, cluster);
3163         spin_unlock(&ctl->tree_lock);
3164
3165         btrfs_discard_queue_work(&block_group->fs_info->discard_ctl, block_group);
3166
3167         /* finally drop our ref */
3168         btrfs_put_block_group(block_group);
3169 }
3170
3171 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group *block_group,
3172                                    struct btrfs_free_cluster *cluster,
3173                                    struct btrfs_free_space *entry,
3174                                    u64 bytes, u64 min_start,
3175                                    u64 *max_extent_size)
3176 {
3177         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3178         int err;
3179         u64 search_start = cluster->window_start;
3180         u64 search_bytes = bytes;
3181         u64 ret = 0;
3182
3183         search_start = min_start;
3184         search_bytes = bytes;
3185
3186         err = search_bitmap(ctl, entry, &search_start, &search_bytes, true);
3187         if (err) {
3188                 *max_extent_size = max(get_max_extent_size(entry),
3189                                        *max_extent_size);
3190                 return 0;
3191         }
3192
3193         ret = search_start;
3194         bitmap_clear_bits(ctl, entry, ret, bytes, false);
3195
3196         return ret;
3197 }
3198
3199 /*
3200  * given a cluster, try to allocate 'bytes' from it, returns 0
3201  * if it couldn't find anything suitably large, or a logical disk offset
3202  * if things worked out
3203  */
3204 u64 btrfs_alloc_from_cluster(struct btrfs_block_group *block_group,
3205                              struct btrfs_free_cluster *cluster, u64 bytes,
3206                              u64 min_start, u64 *max_extent_size)
3207 {
3208         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3209         struct btrfs_discard_ctl *discard_ctl =
3210                                         &block_group->fs_info->discard_ctl;
3211         struct btrfs_free_space *entry = NULL;
3212         struct rb_node *node;
3213         u64 ret = 0;
3214
3215         ASSERT(!btrfs_is_zoned(block_group->fs_info));
3216
3217         spin_lock(&cluster->lock);
3218         if (bytes > cluster->max_size)
3219                 goto out;
3220
3221         if (cluster->block_group != block_group)
3222                 goto out;
3223
3224         node = rb_first(&cluster->root);
3225         if (!node)
3226                 goto out;
3227
3228         entry = rb_entry(node, struct btrfs_free_space, offset_index);
3229         while (1) {
3230                 if (entry->bytes < bytes)
3231                         *max_extent_size = max(get_max_extent_size(entry),
3232                                                *max_extent_size);
3233
3234                 if (entry->bytes < bytes ||
3235                     (!entry->bitmap && entry->offset < min_start)) {
3236                         node = rb_next(&entry->offset_index);
3237                         if (!node)
3238                                 break;
3239                         entry = rb_entry(node, struct btrfs_free_space,
3240                                          offset_index);
3241                         continue;
3242                 }
3243
3244                 if (entry->bitmap) {
3245                         ret = btrfs_alloc_from_bitmap(block_group,
3246                                                       cluster, entry, bytes,
3247                                                       cluster->window_start,
3248                                                       max_extent_size);
3249                         if (ret == 0) {
3250                                 node = rb_next(&entry->offset_index);
3251                                 if (!node)
3252                                         break;
3253                                 entry = rb_entry(node, struct btrfs_free_space,
3254                                                  offset_index);
3255                                 continue;
3256                         }
3257                         cluster->window_start += bytes;
3258                 } else {
3259                         ret = entry->offset;
3260
3261                         entry->offset += bytes;
3262                         entry->bytes -= bytes;
3263                 }
3264
3265                 break;
3266         }
3267 out:
3268         spin_unlock(&cluster->lock);
3269
3270         if (!ret)
3271                 return 0;
3272
3273         spin_lock(&ctl->tree_lock);
3274
3275         if (!btrfs_free_space_trimmed(entry))
3276                 atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3277
3278         ctl->free_space -= bytes;
3279         if (!entry->bitmap && !btrfs_free_space_trimmed(entry))
3280                 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
3281
3282         spin_lock(&cluster->lock);
3283         if (entry->bytes == 0) {
3284                 rb_erase(&entry->offset_index, &cluster->root);
3285                 ctl->free_extents--;
3286                 if (entry->bitmap) {
3287                         kmem_cache_free(btrfs_free_space_bitmap_cachep,
3288                                         entry->bitmap);
3289                         ctl->total_bitmaps--;
3290                         recalculate_thresholds(ctl);
3291                 } else if (!btrfs_free_space_trimmed(entry)) {
3292                         ctl->discardable_extents[BTRFS_STAT_CURR]--;
3293                 }
3294                 kmem_cache_free(btrfs_free_space_cachep, entry);
3295         }
3296
3297         spin_unlock(&cluster->lock);
3298         spin_unlock(&ctl->tree_lock);
3299
3300         return ret;
3301 }
3302
3303 static int btrfs_bitmap_cluster(struct btrfs_block_group *block_group,
3304                                 struct btrfs_free_space *entry,
3305                                 struct btrfs_free_cluster *cluster,
3306                                 u64 offset, u64 bytes,
3307                                 u64 cont1_bytes, u64 min_bytes)
3308 {
3309         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3310         unsigned long next_zero;
3311         unsigned long i;
3312         unsigned long want_bits;
3313         unsigned long min_bits;
3314         unsigned long found_bits;
3315         unsigned long max_bits = 0;
3316         unsigned long start = 0;
3317         unsigned long total_found = 0;
3318         int ret;
3319
3320         i = offset_to_bit(entry->offset, ctl->unit,
3321                           max_t(u64, offset, entry->offset));
3322         want_bits = bytes_to_bits(bytes, ctl->unit);
3323         min_bits = bytes_to_bits(min_bytes, ctl->unit);
3324
3325         /*
3326          * Don't bother looking for a cluster in this bitmap if it's heavily
3327          * fragmented.
3328          */
3329         if (entry->max_extent_size &&
3330             entry->max_extent_size < cont1_bytes)
3331                 return -ENOSPC;
3332 again:
3333         found_bits = 0;
3334         for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
3335                 next_zero = find_next_zero_bit(entry->bitmap,
3336                                                BITS_PER_BITMAP, i);
3337                 if (next_zero - i >= min_bits) {
3338                         found_bits = next_zero - i;
3339                         if (found_bits > max_bits)
3340                                 max_bits = found_bits;
3341                         break;
3342                 }
3343                 if (next_zero - i > max_bits)
3344                         max_bits = next_zero - i;
3345                 i = next_zero;
3346         }
3347
3348         if (!found_bits) {
3349                 entry->max_extent_size = (u64)max_bits * ctl->unit;
3350                 return -ENOSPC;
3351         }
3352
3353         if (!total_found) {
3354                 start = i;
3355                 cluster->max_size = 0;
3356         }
3357
3358         total_found += found_bits;
3359
3360         if (cluster->max_size < found_bits * ctl->unit)
3361                 cluster->max_size = found_bits * ctl->unit;
3362
3363         if (total_found < want_bits || cluster->max_size < cont1_bytes) {
3364                 i = next_zero + 1;
3365                 goto again;
3366         }
3367
3368         cluster->window_start = start * ctl->unit + entry->offset;
3369         rb_erase(&entry->offset_index, &ctl->free_space_offset);
3370         rb_erase_cached(&entry->bytes_index, &ctl->free_space_bytes);
3371
3372         /*
3373          * We need to know if we're currently on the normal space index when we
3374          * manipulate the bitmap so that we know we need to remove and re-insert
3375          * it into the space_index tree.  Clear the bytes_index node here so the
3376          * bitmap manipulation helpers know not to mess with the space_index
3377          * until this bitmap entry is added back into the normal cache.
3378          */
3379         RB_CLEAR_NODE(&entry->bytes_index);
3380
3381         ret = tree_insert_offset(&cluster->root, entry->offset,
3382                                  &entry->offset_index, 1);
3383         ASSERT(!ret); /* -EEXIST; Logic error */
3384
3385         trace_btrfs_setup_cluster(block_group, cluster,
3386                                   total_found * ctl->unit, 1);
3387         return 0;
3388 }
3389
3390 /*
3391  * This searches the block group for just extents to fill the cluster with.
3392  * Try to find a cluster with at least bytes total bytes, at least one
3393  * extent of cont1_bytes, and other clusters of at least min_bytes.
3394  */
3395 static noinline int
3396 setup_cluster_no_bitmap(struct btrfs_block_group *block_group,
3397                         struct btrfs_free_cluster *cluster,
3398                         struct list_head *bitmaps, u64 offset, u64 bytes,
3399                         u64 cont1_bytes, u64 min_bytes)
3400 {
3401         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3402         struct btrfs_free_space *first = NULL;
3403         struct btrfs_free_space *entry = NULL;
3404         struct btrfs_free_space *last;
3405         struct rb_node *node;
3406         u64 window_free;
3407         u64 max_extent;
3408         u64 total_size = 0;
3409
3410         entry = tree_search_offset(ctl, offset, 0, 1);
3411         if (!entry)
3412                 return -ENOSPC;
3413
3414         /*
3415          * We don't want bitmaps, so just move along until we find a normal
3416          * extent entry.
3417          */
3418         while (entry->bitmap || entry->bytes < min_bytes) {
3419                 if (entry->bitmap && list_empty(&entry->list))
3420                         list_add_tail(&entry->list, bitmaps);
3421                 node = rb_next(&entry->offset_index);
3422                 if (!node)
3423                         return -ENOSPC;
3424                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3425         }
3426
3427         window_free = entry->bytes;
3428         max_extent = entry->bytes;
3429         first = entry;
3430         last = entry;
3431
3432         for (node = rb_next(&entry->offset_index); node;
3433              node = rb_next(&entry->offset_index)) {
3434                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3435
3436                 if (entry->bitmap) {
3437                         if (list_empty(&entry->list))
3438                                 list_add_tail(&entry->list, bitmaps);
3439                         continue;
3440                 }
3441
3442                 if (entry->bytes < min_bytes)
3443                         continue;
3444
3445                 last = entry;
3446                 window_free += entry->bytes;
3447                 if (entry->bytes > max_extent)
3448                         max_extent = entry->bytes;
3449         }
3450
3451         if (window_free < bytes || max_extent < cont1_bytes)
3452                 return -ENOSPC;
3453
3454         cluster->window_start = first->offset;
3455
3456         node = &first->offset_index;
3457
3458         /*
3459          * now we've found our entries, pull them out of the free space
3460          * cache and put them into the cluster rbtree
3461          */
3462         do {
3463                 int ret;
3464
3465                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3466                 node = rb_next(&entry->offset_index);
3467                 if (entry->bitmap || entry->bytes < min_bytes)
3468                         continue;
3469
3470                 rb_erase(&entry->offset_index, &ctl->free_space_offset);
3471                 rb_erase_cached(&entry->bytes_index, &ctl->free_space_bytes);
3472                 ret = tree_insert_offset(&cluster->root, entry->offset,
3473                                          &entry->offset_index, 0);
3474                 total_size += entry->bytes;
3475                 ASSERT(!ret); /* -EEXIST; Logic error */
3476         } while (node && entry != last);
3477
3478         cluster->max_size = max_extent;
3479         trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
3480         return 0;
3481 }
3482
3483 /*
3484  * This specifically looks for bitmaps that may work in the cluster, we assume
3485  * that we have already failed to find extents that will work.
3486  */
3487 static noinline int
3488 setup_cluster_bitmap(struct btrfs_block_group *block_group,
3489                      struct btrfs_free_cluster *cluster,
3490                      struct list_head *bitmaps, u64 offset, u64 bytes,
3491                      u64 cont1_bytes, u64 min_bytes)
3492 {
3493         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3494         struct btrfs_free_space *entry = NULL;
3495         int ret = -ENOSPC;
3496         u64 bitmap_offset = offset_to_bitmap(ctl, offset);
3497
3498         if (ctl->total_bitmaps == 0)
3499                 return -ENOSPC;
3500
3501         /*
3502          * The bitmap that covers offset won't be in the list unless offset
3503          * is just its start offset.
3504          */
3505         if (!list_empty(bitmaps))
3506                 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
3507
3508         if (!entry || entry->offset != bitmap_offset) {
3509                 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
3510                 if (entry && list_empty(&entry->list))
3511                         list_add(&entry->list, bitmaps);
3512         }
3513
3514         list_for_each_entry(entry, bitmaps, list) {
3515                 if (entry->bytes < bytes)
3516                         continue;
3517                 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
3518                                            bytes, cont1_bytes, min_bytes);
3519                 if (!ret)
3520                         return 0;
3521         }
3522
3523         /*
3524          * The bitmaps list has all the bitmaps that record free space
3525          * starting after offset, so no more search is required.
3526          */
3527         return -ENOSPC;
3528 }
3529
3530 /*
3531  * here we try to find a cluster of blocks in a block group.  The goal
3532  * is to find at least bytes+empty_size.
3533  * We might not find them all in one contiguous area.
3534  *
3535  * returns zero and sets up cluster if things worked out, otherwise
3536  * it returns -enospc
3537  */
3538 int btrfs_find_space_cluster(struct btrfs_block_group *block_group,
3539                              struct btrfs_free_cluster *cluster,
3540                              u64 offset, u64 bytes, u64 empty_size)
3541 {
3542         struct btrfs_fs_info *fs_info = block_group->fs_info;
3543         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3544         struct btrfs_free_space *entry, *tmp;
3545         LIST_HEAD(bitmaps);
3546         u64 min_bytes;
3547         u64 cont1_bytes;
3548         int ret;
3549
3550         /*
3551          * Choose the minimum extent size we'll require for this
3552          * cluster.  For SSD_SPREAD, don't allow any fragmentation.
3553          * For metadata, allow allocates with smaller extents.  For
3554          * data, keep it dense.
3555          */
3556         if (btrfs_test_opt(fs_info, SSD_SPREAD)) {
3557                 cont1_bytes = bytes + empty_size;
3558                 min_bytes = cont1_bytes;
3559         } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
3560                 cont1_bytes = bytes;
3561                 min_bytes = fs_info->sectorsize;
3562         } else {
3563                 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
3564                 min_bytes = fs_info->sectorsize;
3565         }
3566
3567         spin_lock(&ctl->tree_lock);
3568
3569         /*
3570          * If we know we don't have enough space to make a cluster don't even
3571          * bother doing all the work to try and find one.
3572          */
3573         if (ctl->free_space < bytes) {
3574                 spin_unlock(&ctl->tree_lock);
3575                 return -ENOSPC;
3576         }
3577
3578         spin_lock(&cluster->lock);
3579
3580         /* someone already found a cluster, hooray */
3581         if (cluster->block_group) {
3582                 ret = 0;
3583                 goto out;
3584         }
3585
3586         trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
3587                                  min_bytes);
3588
3589         ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
3590                                       bytes + empty_size,
3591                                       cont1_bytes, min_bytes);
3592         if (ret)
3593                 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
3594                                            offset, bytes + empty_size,
3595                                            cont1_bytes, min_bytes);
3596
3597         /* Clear our temporary list */
3598         list_for_each_entry_safe(entry, tmp, &bitmaps, list)
3599                 list_del_init(&entry->list);
3600
3601         if (!ret) {
3602                 btrfs_get_block_group(block_group);
3603                 list_add_tail(&cluster->block_group_list,
3604                               &block_group->cluster_list);
3605                 cluster->block_group = block_group;
3606         } else {
3607                 trace_btrfs_failed_cluster_setup(block_group);
3608         }
3609 out:
3610         spin_unlock(&cluster->lock);
3611         spin_unlock(&ctl->tree_lock);
3612
3613         return ret;
3614 }
3615
3616 /*
3617  * simple code to zero out a cluster
3618  */
3619 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
3620 {
3621         spin_lock_init(&cluster->lock);
3622         spin_lock_init(&cluster->refill_lock);
3623         cluster->root = RB_ROOT;
3624         cluster->max_size = 0;
3625         cluster->fragmented = false;
3626         INIT_LIST_HEAD(&cluster->block_group_list);
3627         cluster->block_group = NULL;
3628 }
3629
3630 static int do_trimming(struct btrfs_block_group *block_group,
3631                        u64 *total_trimmed, u64 start, u64 bytes,
3632                        u64 reserved_start, u64 reserved_bytes,
3633                        enum btrfs_trim_state reserved_trim_state,
3634                        struct btrfs_trim_range *trim_entry)
3635 {
3636         struct btrfs_space_info *space_info = block_group->space_info;
3637         struct btrfs_fs_info *fs_info = block_group->fs_info;
3638         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3639         int ret;
3640         int update = 0;
3641         const u64 end = start + bytes;
3642         const u64 reserved_end = reserved_start + reserved_bytes;
3643         enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3644         u64 trimmed = 0;
3645
3646         spin_lock(&space_info->lock);
3647         spin_lock(&block_group->lock);
3648         if (!block_group->ro) {
3649                 block_group->reserved += reserved_bytes;
3650                 space_info->bytes_reserved += reserved_bytes;
3651                 update = 1;
3652         }
3653         spin_unlock(&block_group->lock);
3654         spin_unlock(&space_info->lock);
3655
3656         ret = btrfs_discard_extent(fs_info, start, bytes, &trimmed);
3657         if (!ret) {
3658                 *total_trimmed += trimmed;
3659                 trim_state = BTRFS_TRIM_STATE_TRIMMED;
3660         }
3661
3662         mutex_lock(&ctl->cache_writeout_mutex);
3663         if (reserved_start < start)
3664                 __btrfs_add_free_space(block_group, reserved_start,
3665                                        start - reserved_start,
3666                                        reserved_trim_state);
3667         if (start + bytes < reserved_start + reserved_bytes)
3668                 __btrfs_add_free_space(block_group, end, reserved_end - end,
3669                                        reserved_trim_state);
3670         __btrfs_add_free_space(block_group, start, bytes, trim_state);
3671         list_del(&trim_entry->list);
3672         mutex_unlock(&ctl->cache_writeout_mutex);
3673
3674         if (update) {
3675                 spin_lock(&space_info->lock);
3676                 spin_lock(&block_group->lock);
3677                 if (block_group->ro)
3678                         space_info->bytes_readonly += reserved_bytes;
3679                 block_group->reserved -= reserved_bytes;
3680                 space_info->bytes_reserved -= reserved_bytes;
3681                 spin_unlock(&block_group->lock);
3682                 spin_unlock(&space_info->lock);
3683         }
3684
3685         return ret;
3686 }
3687
3688 /*
3689  * If @async is set, then we will trim 1 region and return.
3690  */
3691 static int trim_no_bitmap(struct btrfs_block_group *block_group,
3692                           u64 *total_trimmed, u64 start, u64 end, u64 minlen,
3693                           bool async)
3694 {
3695         struct btrfs_discard_ctl *discard_ctl =
3696                                         &block_group->fs_info->discard_ctl;
3697         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3698         struct btrfs_free_space *entry;
3699         struct rb_node *node;
3700         int ret = 0;
3701         u64 extent_start;
3702         u64 extent_bytes;
3703         enum btrfs_trim_state extent_trim_state;
3704         u64 bytes;
3705         const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3706
3707         while (start < end) {
3708                 struct btrfs_trim_range trim_entry;
3709
3710                 mutex_lock(&ctl->cache_writeout_mutex);
3711                 spin_lock(&ctl->tree_lock);
3712
3713                 if (ctl->free_space < minlen)
3714                         goto out_unlock;
3715
3716                 entry = tree_search_offset(ctl, start, 0, 1);
3717                 if (!entry)
3718                         goto out_unlock;
3719
3720                 /* Skip bitmaps and if async, already trimmed entries */
3721                 while (entry->bitmap ||
3722                        (async && btrfs_free_space_trimmed(entry))) {
3723                         node = rb_next(&entry->offset_index);
3724                         if (!node)
3725                                 goto out_unlock;
3726                         entry = rb_entry(node, struct btrfs_free_space,
3727                                          offset_index);
3728                 }
3729
3730                 if (entry->offset >= end)
3731                         goto out_unlock;
3732
3733                 extent_start = entry->offset;
3734                 extent_bytes = entry->bytes;
3735                 extent_trim_state = entry->trim_state;
3736                 if (async) {
3737                         start = entry->offset;
3738                         bytes = entry->bytes;
3739                         if (bytes < minlen) {
3740                                 spin_unlock(&ctl->tree_lock);
3741                                 mutex_unlock(&ctl->cache_writeout_mutex);
3742                                 goto next;
3743                         }
3744                         unlink_free_space(ctl, entry, true);
3745                         /*
3746                          * Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
3747                          * If X < BTRFS_ASYNC_DISCARD_MIN_FILTER, we won't trim
3748                          * X when we come back around.  So trim it now.
3749                          */
3750                         if (max_discard_size &&
3751                             bytes >= (max_discard_size +
3752                                       BTRFS_ASYNC_DISCARD_MIN_FILTER)) {
3753                                 bytes = max_discard_size;
3754                                 extent_bytes = max_discard_size;
3755                                 entry->offset += max_discard_size;
3756                                 entry->bytes -= max_discard_size;
3757                                 link_free_space(ctl, entry);
3758                         } else {
3759                                 kmem_cache_free(btrfs_free_space_cachep, entry);
3760                         }
3761                 } else {
3762                         start = max(start, extent_start);
3763                         bytes = min(extent_start + extent_bytes, end) - start;
3764                         if (bytes < minlen) {
3765                                 spin_unlock(&ctl->tree_lock);
3766                                 mutex_unlock(&ctl->cache_writeout_mutex);
3767                                 goto next;
3768                         }
3769
3770                         unlink_free_space(ctl, entry, true);
3771                         kmem_cache_free(btrfs_free_space_cachep, entry);
3772                 }
3773
3774                 spin_unlock(&ctl->tree_lock);
3775                 trim_entry.start = extent_start;
3776                 trim_entry.bytes = extent_bytes;
3777                 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3778                 mutex_unlock(&ctl->cache_writeout_mutex);
3779
3780                 ret = do_trimming(block_group, total_trimmed, start, bytes,
3781                                   extent_start, extent_bytes, extent_trim_state,
3782                                   &trim_entry);
3783                 if (ret) {
3784                         block_group->discard_cursor = start + bytes;
3785                         break;
3786                 }
3787 next:
3788                 start += bytes;
3789                 block_group->discard_cursor = start;
3790                 if (async && *total_trimmed)
3791                         break;
3792
3793                 if (fatal_signal_pending(current)) {
3794                         ret = -ERESTARTSYS;
3795                         break;
3796                 }
3797
3798                 cond_resched();
3799         }
3800
3801         return ret;
3802
3803 out_unlock:
3804         block_group->discard_cursor = btrfs_block_group_end(block_group);
3805         spin_unlock(&ctl->tree_lock);
3806         mutex_unlock(&ctl->cache_writeout_mutex);
3807
3808         return ret;
3809 }
3810
3811 /*
3812  * If we break out of trimming a bitmap prematurely, we should reset the
3813  * trimming bit.  In a rather contrieved case, it's possible to race here so
3814  * reset the state to BTRFS_TRIM_STATE_UNTRIMMED.
3815  *
3816  * start = start of bitmap
3817  * end = near end of bitmap
3818  *
3819  * Thread 1:                    Thread 2:
3820  * trim_bitmaps(start)
3821  *                              trim_bitmaps(end)
3822  *                              end_trimming_bitmap()
3823  * reset_trimming_bitmap()
3824  */
3825 static void reset_trimming_bitmap(struct btrfs_free_space_ctl *ctl, u64 offset)
3826 {
3827         struct btrfs_free_space *entry;
3828
3829         spin_lock(&ctl->tree_lock);
3830         entry = tree_search_offset(ctl, offset, 1, 0);
3831         if (entry) {
3832                 if (btrfs_free_space_trimmed(entry)) {
3833                         ctl->discardable_extents[BTRFS_STAT_CURR] +=
3834                                 entry->bitmap_extents;
3835                         ctl->discardable_bytes[BTRFS_STAT_CURR] += entry->bytes;
3836                 }
3837                 entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3838         }
3839
3840         spin_unlock(&ctl->tree_lock);
3841 }
3842
3843 static void end_trimming_bitmap(struct btrfs_free_space_ctl *ctl,
3844                                 struct btrfs_free_space *entry)
3845 {
3846         if (btrfs_free_space_trimming_bitmap(entry)) {
3847                 entry->trim_state = BTRFS_TRIM_STATE_TRIMMED;
3848                 ctl->discardable_extents[BTRFS_STAT_CURR] -=
3849                         entry->bitmap_extents;
3850                 ctl->discardable_bytes[BTRFS_STAT_CURR] -= entry->bytes;
3851         }
3852 }
3853
3854 /*
3855  * If @async is set, then we will trim 1 region and return.
3856  */
3857 static int trim_bitmaps(struct btrfs_block_group *block_group,
3858                         u64 *total_trimmed, u64 start, u64 end, u64 minlen,
3859                         u64 maxlen, bool async)
3860 {
3861         struct btrfs_discard_ctl *discard_ctl =
3862                                         &block_group->fs_info->discard_ctl;
3863         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3864         struct btrfs_free_space *entry;
3865         int ret = 0;
3866         int ret2;
3867         u64 bytes;
3868         u64 offset = offset_to_bitmap(ctl, start);
3869         const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3870
3871         while (offset < end) {
3872                 bool next_bitmap = false;
3873                 struct btrfs_trim_range trim_entry;
3874
3875                 mutex_lock(&ctl->cache_writeout_mutex);
3876                 spin_lock(&ctl->tree_lock);
3877
3878                 if (ctl->free_space < minlen) {
3879                         block_group->discard_cursor =
3880                                 btrfs_block_group_end(block_group);
3881                         spin_unlock(&ctl->tree_lock);
3882                         mutex_unlock(&ctl->cache_writeout_mutex);
3883                         break;
3884                 }
3885
3886                 entry = tree_search_offset(ctl, offset, 1, 0);
3887                 /*
3888                  * Bitmaps are marked trimmed lossily now to prevent constant
3889                  * discarding of the same bitmap (the reason why we are bound
3890                  * by the filters).  So, retrim the block group bitmaps when we
3891                  * are preparing to punt to the unused_bgs list.  This uses
3892                  * @minlen to determine if we are in BTRFS_DISCARD_INDEX_UNUSED
3893                  * which is the only discard index which sets minlen to 0.
3894                  */
3895                 if (!entry || (async && minlen && start == offset &&
3896                                btrfs_free_space_trimmed(entry))) {
3897                         spin_unlock(&ctl->tree_lock);
3898                         mutex_unlock(&ctl->cache_writeout_mutex);
3899                         next_bitmap = true;
3900                         goto next;
3901                 }
3902
3903                 /*
3904                  * Async discard bitmap trimming begins at by setting the start
3905                  * to be key.objectid and the offset_to_bitmap() aligns to the
3906                  * start of the bitmap.  This lets us know we are fully
3907                  * scanning the bitmap rather than only some portion of it.
3908                  */
3909                 if (start == offset)
3910                         entry->trim_state = BTRFS_TRIM_STATE_TRIMMING;
3911
3912                 bytes = minlen;
3913                 ret2 = search_bitmap(ctl, entry, &start, &bytes, false);
3914                 if (ret2 || start >= end) {
3915                         /*
3916                          * We lossily consider a bitmap trimmed if we only skip
3917                          * over regions <= BTRFS_ASYNC_DISCARD_MIN_FILTER.
3918                          */
3919                         if (ret2 && minlen <= BTRFS_ASYNC_DISCARD_MIN_FILTER)
3920                                 end_trimming_bitmap(ctl, entry);
3921                         else
3922                                 entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3923                         spin_unlock(&ctl->tree_lock);
3924                         mutex_unlock(&ctl->cache_writeout_mutex);
3925                         next_bitmap = true;
3926                         goto next;
3927                 }
3928
3929                 /*
3930                  * We already trimmed a region, but are using the locking above
3931                  * to reset the trim_state.
3932                  */
3933                 if (async && *total_trimmed) {
3934                         spin_unlock(&ctl->tree_lock);
3935                         mutex_unlock(&ctl->cache_writeout_mutex);
3936                         goto out;
3937                 }
3938
3939                 bytes = min(bytes, end - start);
3940                 if (bytes < minlen || (async && maxlen && bytes > maxlen)) {
3941                         spin_unlock(&ctl->tree_lock);
3942                         mutex_unlock(&ctl->cache_writeout_mutex);
3943                         goto next;
3944                 }
3945
3946                 /*
3947                  * Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
3948                  * If X < @minlen, we won't trim X when we come back around.
3949                  * So trim it now.  We differ here from trimming extents as we
3950                  * don't keep individual state per bit.
3951                  */
3952                 if (async &&
3953                     max_discard_size &&
3954                     bytes > (max_discard_size + minlen))
3955                         bytes = max_discard_size;
3956
3957                 bitmap_clear_bits(ctl, entry, start, bytes, true);
3958                 if (entry->bytes == 0)
3959                         free_bitmap(ctl, entry);
3960
3961                 spin_unlock(&ctl->tree_lock);
3962                 trim_entry.start = start;
3963                 trim_entry.bytes = bytes;
3964                 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3965                 mutex_unlock(&ctl->cache_writeout_mutex);
3966
3967                 ret = do_trimming(block_group, total_trimmed, start, bytes,
3968                                   start, bytes, 0, &trim_entry);
3969                 if (ret) {
3970                         reset_trimming_bitmap(ctl, offset);
3971                         block_group->discard_cursor =
3972                                 btrfs_block_group_end(block_group);
3973                         break;
3974                 }
3975 next:
3976                 if (next_bitmap) {
3977                         offset += BITS_PER_BITMAP * ctl->unit;
3978                         start = offset;
3979                 } else {
3980                         start += bytes;
3981                 }
3982                 block_group->discard_cursor = start;
3983
3984                 if (fatal_signal_pending(current)) {
3985                         if (start != offset)
3986                                 reset_trimming_bitmap(ctl, offset);
3987                         ret = -ERESTARTSYS;
3988                         break;
3989                 }
3990
3991                 cond_resched();
3992         }
3993
3994         if (offset >= end)
3995                 block_group->discard_cursor = end;
3996
3997 out:
3998         return ret;
3999 }
4000
4001 int btrfs_trim_block_group(struct btrfs_block_group *block_group,
4002                            u64 *trimmed, u64 start, u64 end, u64 minlen)
4003 {
4004         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
4005         int ret;
4006         u64 rem = 0;
4007
4008         ASSERT(!btrfs_is_zoned(block_group->fs_info));
4009
4010         *trimmed = 0;
4011
4012         spin_lock(&block_group->lock);
4013         if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4014                 spin_unlock(&block_group->lock);
4015                 return 0;
4016         }
4017         btrfs_freeze_block_group(block_group);
4018         spin_unlock(&block_group->lock);
4019
4020         ret = trim_no_bitmap(block_group, trimmed, start, end, minlen, false);
4021         if (ret)
4022                 goto out;
4023
4024         ret = trim_bitmaps(block_group, trimmed, start, end, minlen, 0, false);
4025         div64_u64_rem(end, BITS_PER_BITMAP * ctl->unit, &rem);
4026         /* If we ended in the middle of a bitmap, reset the trimming flag */
4027         if (rem)
4028                 reset_trimming_bitmap(ctl, offset_to_bitmap(ctl, end));
4029 out:
4030         btrfs_unfreeze_block_group(block_group);
4031         return ret;
4032 }
4033
4034 int btrfs_trim_block_group_extents(struct btrfs_block_group *block_group,
4035                                    u64 *trimmed, u64 start, u64 end, u64 minlen,
4036                                    bool async)
4037 {
4038         int ret;
4039
4040         *trimmed = 0;
4041
4042         spin_lock(&block_group->lock);
4043         if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4044                 spin_unlock(&block_group->lock);
4045                 return 0;
4046         }
4047         btrfs_freeze_block_group(block_group);
4048         spin_unlock(&block_group->lock);
4049
4050         ret = trim_no_bitmap(block_group, trimmed, start, end, minlen, async);
4051         btrfs_unfreeze_block_group(block_group);
4052
4053         return ret;
4054 }
4055
4056 int btrfs_trim_block_group_bitmaps(struct btrfs_block_group *block_group,
4057                                    u64 *trimmed, u64 start, u64 end, u64 minlen,
4058                                    u64 maxlen, bool async)
4059 {
4060         int ret;
4061
4062         *trimmed = 0;
4063
4064         spin_lock(&block_group->lock);
4065         if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4066                 spin_unlock(&block_group->lock);
4067                 return 0;
4068         }
4069         btrfs_freeze_block_group(block_group);
4070         spin_unlock(&block_group->lock);
4071
4072         ret = trim_bitmaps(block_group, trimmed, start, end, minlen, maxlen,
4073                            async);
4074
4075         btrfs_unfreeze_block_group(block_group);
4076
4077         return ret;
4078 }
4079
4080 bool btrfs_free_space_cache_v1_active(struct btrfs_fs_info *fs_info)
4081 {
4082         return btrfs_super_cache_generation(fs_info->super_copy);
4083 }
4084
4085 static int cleanup_free_space_cache_v1(struct btrfs_fs_info *fs_info,
4086                                        struct btrfs_trans_handle *trans)
4087 {
4088         struct btrfs_block_group *block_group;
4089         struct rb_node *node;
4090         int ret = 0;
4091
4092         btrfs_info(fs_info, "cleaning free space cache v1");
4093
4094         node = rb_first_cached(&fs_info->block_group_cache_tree);
4095         while (node) {
4096                 block_group = rb_entry(node, struct btrfs_block_group, cache_node);
4097                 ret = btrfs_remove_free_space_inode(trans, NULL, block_group);
4098                 if (ret)
4099                         goto out;
4100                 node = rb_next(node);
4101         }
4102 out:
4103         return ret;
4104 }
4105
4106 int btrfs_set_free_space_cache_v1_active(struct btrfs_fs_info *fs_info, bool active)
4107 {
4108         struct btrfs_trans_handle *trans;
4109         int ret;
4110
4111         /*
4112          * update_super_roots will appropriately set or unset
4113          * super_copy->cache_generation based on SPACE_CACHE and
4114          * BTRFS_FS_CLEANUP_SPACE_CACHE_V1. For this reason, we need a
4115          * transaction commit whether we are enabling space cache v1 and don't
4116          * have any other work to do, or are disabling it and removing free
4117          * space inodes.
4118          */
4119         trans = btrfs_start_transaction(fs_info->tree_root, 0);
4120         if (IS_ERR(trans))
4121                 return PTR_ERR(trans);
4122
4123         if (!active) {
4124                 set_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags);
4125                 ret = cleanup_free_space_cache_v1(fs_info, trans);
4126                 if (ret) {
4127                         btrfs_abort_transaction(trans, ret);
4128                         btrfs_end_transaction(trans);
4129                         goto out;
4130                 }
4131         }
4132
4133         ret = btrfs_commit_transaction(trans);
4134 out:
4135         clear_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags);
4136
4137         return ret;
4138 }
4139
4140 int __init btrfs_free_space_init(void)
4141 {
4142         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
4143                         sizeof(struct btrfs_free_space), 0,
4144                         SLAB_MEM_SPREAD, NULL);
4145         if (!btrfs_free_space_cachep)
4146                 return -ENOMEM;
4147
4148         btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
4149                                                         PAGE_SIZE, PAGE_SIZE,
4150                                                         SLAB_MEM_SPREAD, NULL);
4151         if (!btrfs_free_space_bitmap_cachep) {
4152                 kmem_cache_destroy(btrfs_free_space_cachep);
4153                 return -ENOMEM;
4154         }
4155
4156         return 0;
4157 }
4158
4159 void __cold btrfs_free_space_exit(void)
4160 {
4161         kmem_cache_destroy(btrfs_free_space_cachep);
4162         kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
4163 }
4164
4165 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4166 /*
4167  * Use this if you need to make a bitmap or extent entry specifically, it
4168  * doesn't do any of the merging that add_free_space does, this acts a lot like
4169  * how the free space cache loading stuff works, so you can get really weird
4170  * configurations.
4171  */
4172 int test_add_free_space_entry(struct btrfs_block_group *cache,
4173                               u64 offset, u64 bytes, bool bitmap)
4174 {
4175         struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
4176         struct btrfs_free_space *info = NULL, *bitmap_info;
4177         void *map = NULL;
4178         enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_TRIMMED;
4179         u64 bytes_added;
4180         int ret;
4181
4182 again:
4183         if (!info) {
4184                 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
4185                 if (!info)
4186                         return -ENOMEM;
4187         }
4188
4189         if (!bitmap) {
4190                 spin_lock(&ctl->tree_lock);
4191                 info->offset = offset;
4192                 info->bytes = bytes;
4193                 info->max_extent_size = 0;
4194                 ret = link_free_space(ctl, info);
4195                 spin_unlock(&ctl->tree_lock);
4196                 if (ret)
4197                         kmem_cache_free(btrfs_free_space_cachep, info);
4198                 return ret;
4199         }
4200
4201         if (!map) {
4202                 map = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep, GFP_NOFS);
4203                 if (!map) {
4204                         kmem_cache_free(btrfs_free_space_cachep, info);
4205                         return -ENOMEM;
4206                 }
4207         }
4208
4209         spin_lock(&ctl->tree_lock);
4210         bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
4211                                          1, 0);
4212         if (!bitmap_info) {
4213                 info->bitmap = map;
4214                 map = NULL;
4215                 add_new_bitmap(ctl, info, offset);
4216                 bitmap_info = info;
4217                 info = NULL;
4218         }
4219
4220         bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes,
4221                                           trim_state);
4222
4223         bytes -= bytes_added;
4224         offset += bytes_added;
4225         spin_unlock(&ctl->tree_lock);
4226
4227         if (bytes)
4228                 goto again;
4229
4230         if (info)
4231                 kmem_cache_free(btrfs_free_space_cachep, info);
4232         if (map)
4233                 kmem_cache_free(btrfs_free_space_bitmap_cachep, map);
4234         return 0;
4235 }
4236
4237 /*
4238  * Checks to see if the given range is in the free space cache.  This is really
4239  * just used to check the absence of space, so if there is free space in the
4240  * range at all we will return 1.
4241  */
4242 int test_check_exists(struct btrfs_block_group *cache,
4243                       u64 offset, u64 bytes)
4244 {
4245         struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
4246         struct btrfs_free_space *info;
4247         int ret = 0;
4248
4249         spin_lock(&ctl->tree_lock);
4250         info = tree_search_offset(ctl, offset, 0, 0);
4251         if (!info) {
4252                 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
4253                                           1, 0);
4254                 if (!info)
4255                         goto out;
4256         }
4257
4258 have_info:
4259         if (info->bitmap) {
4260                 u64 bit_off, bit_bytes;
4261                 struct rb_node *n;
4262                 struct btrfs_free_space *tmp;
4263
4264                 bit_off = offset;
4265                 bit_bytes = ctl->unit;
4266                 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes, false);
4267                 if (!ret) {
4268                         if (bit_off == offset) {
4269                                 ret = 1;
4270                                 goto out;
4271                         } else if (bit_off > offset &&
4272                                    offset + bytes > bit_off) {
4273                                 ret = 1;
4274                                 goto out;
4275                         }
4276                 }
4277
4278                 n = rb_prev(&info->offset_index);
4279                 while (n) {
4280                         tmp = rb_entry(n, struct btrfs_free_space,
4281                                        offset_index);
4282                         if (tmp->offset + tmp->bytes < offset)
4283                                 break;
4284                         if (offset + bytes < tmp->offset) {
4285                                 n = rb_prev(&tmp->offset_index);
4286                                 continue;
4287                         }
4288                         info = tmp;
4289                         goto have_info;
4290                 }
4291
4292                 n = rb_next(&info->offset_index);
4293                 while (n) {
4294                         tmp = rb_entry(n, struct btrfs_free_space,
4295                                        offset_index);
4296                         if (offset + bytes < tmp->offset)
4297                                 break;
4298                         if (tmp->offset + tmp->bytes < offset) {
4299                                 n = rb_next(&tmp->offset_index);
4300                                 continue;
4301                         }
4302                         info = tmp;
4303                         goto have_info;
4304                 }
4305
4306                 ret = 0;
4307                 goto out;
4308         }
4309
4310         if (info->offset == offset) {
4311                 ret = 1;
4312                 goto out;
4313         }
4314
4315         if (offset > info->offset && offset < info->offset + info->bytes)
4316                 ret = 1;
4317 out:
4318         spin_unlock(&ctl->tree_lock);
4319         return ret;
4320 }
4321 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */