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_crc32c_final(u32 crc, u8 *result)
  61 {
  62         put_unaligned_le32(~crc, result);
  63 }
  64
  65 static void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
  66 {
  67         struct btrfs_free_space *info;
  68         struct rb_node *node;
  69
  70         while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
  71                 info = rb_entry(node, struct btrfs_free_space, offset_index);
  72                 if (!info->bitmap) {
  73                         unlink_free_space(ctl, info, true);
  74                         kmem_cache_free(btrfs_free_space_cachep, info);
  75                 } else {
  76                         free_bitmap(ctl, info);
  77                 }
  78
  79                 cond_resched_lock(&ctl->tree_lock);
  80         }
  81 }
  82
  83 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
  84                                                struct btrfs_path *path,
  85                                                u64 offset)
  86 {
  87         struct btrfs_fs_info *fs_info = root->fs_info;
  88         struct btrfs_key key;
  89         struct btrfs_key location;
  90         struct btrfs_disk_key disk_key;
  91         struct btrfs_free_space_header *header;
  92         struct extent_buffer *leaf;
  93         struct inode *inode = NULL;
  94         unsigned nofs_flag;
  95         int ret;
  96
  97         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
  98         key.offset = offset;
  99         key.type = 0;
 100
 101         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 102         if (ret < 0)
 103                 return ERR_PTR(ret);
 104         if (ret > 0) {
 105                 btrfs_release_path(path);
 106                 return ERR_PTR(-ENOENT);
 107         }
 108
 109         leaf = path->nodes[0];
 110         header = btrfs_item_ptr(leaf, path->slots[0],
 111                                 struct btrfs_free_space_header);
 112         btrfs_free_space_key(leaf, header, &disk_key);
 113         btrfs_disk_key_to_cpu(&location, &disk_key);
 114         btrfs_release_path(path);
 115
 116         /*
 117          * We are often under a trans handle at this point, so we need to make
 118          * sure NOFS is set to keep us from deadlocking.
 119          */
 120         nofs_flag = memalloc_nofs_save();
 121         inode = btrfs_iget_path(fs_info->sb, location.objectid, root, path);
 122         btrfs_release_path(path);
 123         memalloc_nofs_restore(nofs_flag);
 124         if (IS_ERR(inode))
 125                 return inode;
 126
 127         mapping_set_gfp_mask(inode->i_mapping,
 128                         mapping_gfp_constraint(inode->i_mapping,
 129                         ~(__GFP_FS | __GFP_HIGHMEM)));
 130
 131         return inode;
 132 }
 133
 134 struct inode *lookup_free_space_inode(struct btrfs_block_group *block_group,
 135                 struct btrfs_path *path)
 136 {
 137         struct btrfs_fs_info *fs_info = block_group->fs_info;
 138         struct inode *inode = NULL;
 139         u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
 140
 141         spin_lock(&block_group->lock);
 142         if (block_group->inode)
 143                 inode = igrab(block_group->inode);
 144         spin_unlock(&block_group->lock);
 145         if (inode)
 146                 return inode;
 147
 148         inode = __lookup_free_space_inode(fs_info->tree_root, path,
 149                                           block_group->start);
 150         if (IS_ERR(inode))
 151                 return inode;
 152
 153         spin_lock(&block_group->lock);
 154         if (!((BTRFS_I(inode)->flags & flags) == flags)) {
 155                 btrfs_info(fs_info, "Old style space inode found, converting.");
 156                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
 157                         BTRFS_INODE_NODATACOW;
 158                 block_group->disk_cache_state = BTRFS_DC_CLEAR;
 159         }
 160
 161         if (!test_and_set_bit(BLOCK_GROUP_FLAG_IREF, &block_group->runtime_flags))
 162                 block_group->inode = igrab(inode);
 163         spin_unlock(&block_group->lock);
 164
 165         return inode;
 166 }
 167
 168 static int __create_free_space_inode(struct btrfs_root *root,
 169                                      struct btrfs_trans_handle *trans,
 170                                      struct btrfs_path *path,
 171                                      u64 ino, u64 offset)
 172 {
 173         struct btrfs_key key;
 174         struct btrfs_disk_key disk_key;
 175         struct btrfs_free_space_header *header;
 176         struct btrfs_inode_item *inode_item;
 177         struct extent_buffer *leaf;
 178         /* We inline CRCs for the free disk space cache */
 179         const u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC |
 180                           BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
 181         int ret;
 182
 183         ret = btrfs_insert_empty_inode(trans, root, path, ino);
 184         if (ret)
 185                 return ret;
 186
 187         leaf = path->nodes[0];
 188         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 189                                     struct btrfs_inode_item);
 190         btrfs_item_key(leaf, &disk_key, path->slots[0]);
 191         memzero_extent_buffer(leaf, (unsigned long)inode_item,
 192                              sizeof(*inode_item));
 193         btrfs_set_inode_generation(leaf, inode_item, trans->transid);
 194         btrfs_set_inode_size(leaf, inode_item, 0);
 195         btrfs_set_inode_nbytes(leaf, inode_item, 0);
 196         btrfs_set_inode_uid(leaf, inode_item, 0);
 197         btrfs_set_inode_gid(leaf, inode_item, 0);
 198         btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
 199         btrfs_set_inode_flags(leaf, inode_item, flags);
 200         btrfs_set_inode_nlink(leaf, inode_item, 1);
 201         btrfs_set_inode_transid(leaf, inode_item, trans->transid);
 202         btrfs_set_inode_block_group(leaf, inode_item, offset);
 203         btrfs_mark_buffer_dirty(trans, leaf);
 204         btrfs_release_path(path);
 205
 206         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
 207         key.offset = offset;
 208         key.type = 0;
 209         ret = btrfs_insert_empty_item(trans, root, path, &key,
 210                                       sizeof(struct btrfs_free_space_header));
 211         if (ret < 0) {
 212                 btrfs_release_path(path);
 213                 return ret;
 214         }
 215
 216         leaf = path->nodes[0];
 217         header = btrfs_item_ptr(leaf, path->slots[0],
 218                                 struct btrfs_free_space_header);
 219         memzero_extent_buffer(leaf, (unsigned long)header, sizeof(*header));
 220         btrfs_set_free_space_key(leaf, header, &disk_key);
 221         btrfs_mark_buffer_dirty(trans, leaf);
 222         btrfs_release_path(path);
 223
 224         return 0;
 225 }
 226
 227 int create_free_space_inode(struct btrfs_trans_handle *trans,
 228                             struct btrfs_block_group *block_group,
 229                             struct btrfs_path *path)
 230 {
 231         int ret;
 232         u64 ino;
 233
 234         ret = btrfs_get_free_objectid(trans->fs_info->tree_root, &ino);
 235         if (ret < 0)
 236                 return ret;
 237
 238         return __create_free_space_inode(trans->fs_info->tree_root, trans, path,
 239                                          ino, block_group->start);
 240 }
 241
 242 /*
 243  * inode is an optional sink: if it is NULL, btrfs_remove_free_space_inode
 244  * handles lookup, otherwise it takes ownership and iputs the inode.
 245  * Don't reuse an inode pointer after passing it into this function.
 246  */
 247 int btrfs_remove_free_space_inode(struct btrfs_trans_handle *trans,
 248                                   struct inode *inode,
 249                                   struct btrfs_block_group *block_group)
 250 {
 251         struct btrfs_path *path;
 252         struct btrfs_key key;
 253         int ret = 0;
 254
 255         path = btrfs_alloc_path();
 256         if (!path)
 257                 return -ENOMEM;
 258
 259         if (!inode)
 260                 inode = lookup_free_space_inode(block_group, path);
 261         if (IS_ERR(inode)) {
 262                 if (PTR_ERR(inode) != -ENOENT)
 263                         ret = PTR_ERR(inode);
 264                 goto out;
 265         }
 266         ret = btrfs_orphan_add(trans, BTRFS_I(inode));
 267         if (ret) {
 268                 btrfs_add_delayed_iput(BTRFS_I(inode));
 269                 goto out;
 270         }
 271         clear_nlink(inode);
 272         /* One for the block groups ref */
 273         spin_lock(&block_group->lock);
 274         if (test_and_clear_bit(BLOCK_GROUP_FLAG_IREF, &block_group->runtime_flags)) {
 275                 block_group->inode = NULL;
 276                 spin_unlock(&block_group->lock);
 277                 iput(inode);
 278         } else {
 279                 spin_unlock(&block_group->lock);
 280         }
 281         /* One for the lookup ref */
 282         btrfs_add_delayed_iput(BTRFS_I(inode));
 283
 284         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
 285         key.type = 0;
 286         key.offset = block_group->start;
 287         ret = btrfs_search_slot(trans, trans->fs_info->tree_root, &key, path,
 288                                 -1, 1);
 289         if (ret) {
 290                 if (ret > 0)
 291                         ret = 0;
 292                 goto out;
 293         }
 294         ret = btrfs_del_item(trans, trans->fs_info->tree_root, path);
 295 out:
 296         btrfs_free_path(path);
 297         return ret;
 298 }
 299
 300 int btrfs_truncate_free_space_cache(struct btrfs_trans_handle *trans,
 301                                     struct btrfs_block_group *block_group,
 302                                     struct inode *vfs_inode)
 303 {
 304         struct btrfs_truncate_control control = {
 305                 .inode = BTRFS_I(vfs_inode),
 306                 .new_size = 0,
 307                 .ino = btrfs_ino(BTRFS_I(vfs_inode)),
 308                 .min_type = BTRFS_EXTENT_DATA_KEY,
 309                 .clear_extent_range = true,
 310         };
 311         struct btrfs_inode *inode = BTRFS_I(vfs_inode);
 312         struct btrfs_root *root = inode->root;
 313         struct extent_state *cached_state = NULL;
 314         int ret = 0;
 315         bool locked = false;
 316
 317         if (block_group) {
 318                 struct btrfs_path *path = btrfs_alloc_path();
 319
 320                 if (!path) {
 321                         ret = -ENOMEM;
 322                         goto fail;
 323                 }
 324                 locked = true;
 325                 mutex_lock(&trans->transaction->cache_write_mutex);
 326                 if (!list_empty(&block_group->io_list)) {
 327                         list_del_init(&block_group->io_list);
 328
 329                         btrfs_wait_cache_io(trans, block_group, path);
 330                         btrfs_put_block_group(block_group);
 331                 }
 332
 333                 /*
 334                  * now that we've truncated the cache away, its no longer
 335                  * setup or written
 336                  */
 337                 spin_lock(&block_group->lock);
 338                 block_group->disk_cache_state = BTRFS_DC_CLEAR;
 339                 spin_unlock(&block_group->lock);
 340                 btrfs_free_path(path);
 341         }
 342
 343         btrfs_i_size_write(inode, 0);
 344         truncate_pagecache(vfs_inode, 0);
 345
 346         lock_extent(&inode->io_tree, 0, (u64)-1, &cached_state);
 347         btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
 348
 349         /*
 350          * We skip the throttling logic for free space cache inodes, so we don't
 351          * need to check for -EAGAIN.
 352          */
 353         ret = btrfs_truncate_inode_items(trans, root, &control);
 354
 355         inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
 356         btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
 357
 358         unlock_extent(&inode->io_tree, 0, (u64)-1, &cached_state);
 359         if (ret)
 360                 goto fail;
 361
 362         ret = btrfs_update_inode(trans, inode);
 363
 364 fail:
 365         if (locked)
 366                 mutex_unlock(&trans->transaction->cache_write_mutex);
 367         if (ret)
 368                 btrfs_abort_transaction(trans, ret);
 369
 370         return ret;
 371 }
 372
 373 static void readahead_cache(struct inode *inode)
 374 {
 375         struct file_ra_state ra;
 376         unsigned long last_index;
 377
 378         file_ra_state_init(&ra, inode->i_mapping);
 379         last_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
 380
 381         page_cache_sync_readahead(inode->i_mapping, &ra, NULL, 0, last_index);
 382 }
 383
 384 static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode,
 385                        int write)
 386 {
 387         int num_pages;
 388
 389         num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
 390
 391         /* Make sure we can fit our crcs and generation into the first page */
 392         if (write && (num_pages * sizeof(u32) + sizeof(u64)) > PAGE_SIZE)
 393                 return -ENOSPC;
 394
 395         memset(io_ctl, 0, sizeof(struct btrfs_io_ctl));
 396
 397         io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS);
 398         if (!io_ctl->pages)
 399                 return -ENOMEM;
 400
 401         io_ctl->num_pages = num_pages;
 402         io_ctl->fs_info = btrfs_sb(inode->i_sb);
 403         io_ctl->inode = inode;
 404
 405         return 0;
 406 }
 407 ALLOW_ERROR_INJECTION(io_ctl_init, ERRNO);
 408
 409 static void io_ctl_free(struct btrfs_io_ctl *io_ctl)
 410 {
 411         kfree(io_ctl->pages);
 412         io_ctl->pages = NULL;
 413 }
 414
 415 static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl)
 416 {
 417         if (io_ctl->cur) {
 418                 io_ctl->cur = NULL;
 419                 io_ctl->orig = NULL;
 420         }
 421 }
 422
 423 static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear)
 424 {
 425         ASSERT(io_ctl->index < io_ctl->num_pages);
 426         io_ctl->page = io_ctl->pages[io_ctl->index++];
 427         io_ctl->cur = page_address(io_ctl->page);
 428         io_ctl->orig = io_ctl->cur;
 429         io_ctl->size = PAGE_SIZE;
 430         if (clear)
 431                 clear_page(io_ctl->cur);
 432 }
 433
 434 static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl)
 435 {
 436         int i;
 437
 438         io_ctl_unmap_page(io_ctl);
 439
 440         for (i = 0; i < io_ctl->num_pages; i++) {
 441                 if (io_ctl->pages[i]) {
 442                         btrfs_folio_clear_checked(io_ctl->fs_info,
 443                                         page_folio(io_ctl->pages[i]),
 444                                         page_offset(io_ctl->pages[i]),
 445                                         PAGE_SIZE);
 446                         unlock_page(io_ctl->pages[i]);
 447                         put_page(io_ctl->pages[i]);
 448                 }
 449         }
 450 }
 451
 452 static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, bool uptodate)
 453 {
 454         struct page *page;
 455         struct inode *inode = io_ctl->inode;
 456         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
 457         int i;
 458
 459         for (i = 0; i < io_ctl->num_pages; i++) {
 460                 int ret;
 461
 462                 page = find_or_create_page(inode->i_mapping, i, mask);
 463                 if (!page) {
 464                         io_ctl_drop_pages(io_ctl);
 465                         return -ENOMEM;
 466                 }
 467
 468                 ret = set_page_extent_mapped(page);
 469                 if (ret < 0) {
 470                         unlock_page(page);
 471                         put_page(page);
 472                         io_ctl_drop_pages(io_ctl);
 473                         return ret;
 474                 }
 475
 476                 io_ctl->pages[i] = page;
 477                 if (uptodate && !PageUptodate(page)) {
 478                         btrfs_read_folio(NULL, page_folio(page));
 479                         lock_page(page);
 480                         if (page->mapping != inode->i_mapping) {
 481                                 btrfs_err(BTRFS_I(inode)->root->fs_info,
 482                                           "free space cache page truncated");
 483                                 io_ctl_drop_pages(io_ctl);
 484                                 return -EIO;
 485                         }
 486                         if (!PageUptodate(page)) {
 487                                 btrfs_err(BTRFS_I(inode)->root->fs_info,
 488                                            "error reading free space cache");
 489                                 io_ctl_drop_pages(io_ctl);
 490                                 return -EIO;
 491                         }
 492                 }
 493         }
 494
 495         for (i = 0; i < io_ctl->num_pages; i++)
 496                 clear_page_dirty_for_io(io_ctl->pages[i]);
 497
 498         return 0;
 499 }
 500
 501 static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
 502 {
 503         io_ctl_map_page(io_ctl, 1);
 504
 505         /*
 506          * Skip the csum areas.  If we don't check crcs then we just have a
 507          * 64bit chunk at the front of the first page.
 508          */
 509         io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
 510         io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
 511
 512         put_unaligned_le64(generation, io_ctl->cur);
 513         io_ctl->cur += sizeof(u64);
 514 }
 515
 516 static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
 517 {
 518         u64 cache_gen;
 519
 520         /*
 521          * Skip the crc area.  If we don't check crcs then we just have a 64bit
 522          * chunk at the front of the first page.
 523          */
 524         io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
 525         io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
 526
 527         cache_gen = get_unaligned_le64(io_ctl->cur);
 528         if (cache_gen != generation) {
 529                 btrfs_err_rl(io_ctl->fs_info,
 530                         "space cache generation (%llu) does not match inode (%llu)",
 531                                 cache_gen, generation);
 532                 io_ctl_unmap_page(io_ctl);
 533                 return -EIO;
 534         }
 535         io_ctl->cur += sizeof(u64);
 536         return 0;
 537 }
 538
 539 static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index)
 540 {
 541         u32 *tmp;
 542         u32 crc = ~(u32)0;
 543         unsigned offset = 0;
 544
 545         if (index == 0)
 546                 offset = sizeof(u32) * io_ctl->num_pages;
 547
 548         crc = crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
 549         btrfs_crc32c_final(crc, (u8 *)&crc);
 550         io_ctl_unmap_page(io_ctl);
 551         tmp = page_address(io_ctl->pages[0]);
 552         tmp += index;
 553         *tmp = crc;
 554 }
 555
 556 static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index)
 557 {
 558         u32 *tmp, val;
 559         u32 crc = ~(u32)0;
 560         unsigned offset = 0;
 561
 562         if (index == 0)
 563                 offset = sizeof(u32) * io_ctl->num_pages;
 564
 565         tmp = page_address(io_ctl->pages[0]);
 566         tmp += index;
 567         val = *tmp;
 568
 569         io_ctl_map_page(io_ctl, 0);
 570         crc = crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
 571         btrfs_crc32c_final(crc, (u8 *)&crc);
 572         if (val != crc) {
 573                 btrfs_err_rl(io_ctl->fs_info,
 574                         "csum mismatch on free space cache");
 575                 io_ctl_unmap_page(io_ctl);
 576                 return -EIO;
 577         }
 578
 579         return 0;
 580 }
 581
 582 static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes,
 583                             void *bitmap)
 584 {
 585         struct btrfs_free_space_entry *entry;
 586
 587         if (!io_ctl->cur)
 588                 return -ENOSPC;
 589
 590         entry = io_ctl->cur;
 591         put_unaligned_le64(offset, &entry->offset);
 592         put_unaligned_le64(bytes, &entry->bytes);
 593         entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
 594                 BTRFS_FREE_SPACE_EXTENT;
 595         io_ctl->cur += sizeof(struct btrfs_free_space_entry);
 596         io_ctl->size -= sizeof(struct btrfs_free_space_entry);
 597
 598         if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
 599                 return 0;
 600
 601         io_ctl_set_crc(io_ctl, io_ctl->index - 1);
 602
 603         /* No more pages to map */
 604         if (io_ctl->index >= io_ctl->num_pages)
 605                 return 0;
 606
 607         /* map the next page */
 608         io_ctl_map_page(io_ctl, 1);
 609         return 0;
 610 }
 611
 612 static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap)
 613 {
 614         if (!io_ctl->cur)
 615                 return -ENOSPC;
 616
 617         /*
 618          * If we aren't at the start of the current page, unmap this one and
 619          * map the next one if there is any left.
 620          */
 621         if (io_ctl->cur != io_ctl->orig) {
 622                 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
 623                 if (io_ctl->index >= io_ctl->num_pages)
 624                         return -ENOSPC;
 625                 io_ctl_map_page(io_ctl, 0);
 626         }
 627
 628         copy_page(io_ctl->cur, bitmap);
 629         io_ctl_set_crc(io_ctl, io_ctl->index - 1);
 630         if (io_ctl->index < io_ctl->num_pages)
 631                 io_ctl_map_page(io_ctl, 0);
 632         return 0;
 633 }
 634
 635 static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl)
 636 {
 637         /*
 638          * If we're not on the boundary we know we've modified the page and we
 639          * need to crc the page.
 640          */
 641         if (io_ctl->cur != io_ctl->orig)
 642                 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
 643         else
 644                 io_ctl_unmap_page(io_ctl);
 645
 646         while (io_ctl->index < io_ctl->num_pages) {
 647                 io_ctl_map_page(io_ctl, 1);
 648                 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
 649         }
 650 }
 651
 652 static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl,
 653                             struct btrfs_free_space *entry, u8 *type)
 654 {
 655         struct btrfs_free_space_entry *e;
 656         int ret;
 657
 658         if (!io_ctl->cur) {
 659                 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
 660                 if (ret)
 661                         return ret;
 662         }
 663
 664         e = io_ctl->cur;
 665         entry->offset = get_unaligned_le64(&e->offset);
 666         entry->bytes = get_unaligned_le64(&e->bytes);
 667         *type = e->type;
 668         io_ctl->cur += sizeof(struct btrfs_free_space_entry);
 669         io_ctl->size -= sizeof(struct btrfs_free_space_entry);
 670
 671         if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
 672                 return 0;
 673
 674         io_ctl_unmap_page(io_ctl);
 675
 676         return 0;
 677 }
 678
 679 static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl,
 680                               struct btrfs_free_space *entry)
 681 {
 682         int ret;
 683
 684         ret = io_ctl_check_crc(io_ctl, io_ctl->index);
 685         if (ret)
 686                 return ret;
 687
 688         copy_page(entry->bitmap, io_ctl->cur);
 689         io_ctl_unmap_page(io_ctl);
 690
 691         return 0;
 692 }
 693
 694 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
 695 {
 696         struct btrfs_block_group *block_group = ctl->block_group;
 697         u64 max_bytes;
 698         u64 bitmap_bytes;
 699         u64 extent_bytes;
 700         u64 size = block_group->length;
 701         u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
 702         u64 max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
 703
 704         max_bitmaps = max_t(u64, max_bitmaps, 1);
 705
 706         if (ctl->total_bitmaps > max_bitmaps)
 707                 btrfs_err(block_group->fs_info,
 708 "invalid free space control: bg start=%llu len=%llu total_bitmaps=%u unit=%u max_bitmaps=%llu bytes_per_bg=%llu",
 709                           block_group->start, block_group->length,
 710                           ctl->total_bitmaps, ctl->unit, max_bitmaps,
 711                           bytes_per_bg);
 712         ASSERT(ctl->total_bitmaps <= max_bitmaps);
 713
 714         /*
 715          * We are trying to keep the total amount of memory used per 1GiB of
 716          * space to be MAX_CACHE_BYTES_PER_GIG.  However, with a reclamation
 717          * mechanism of pulling extents >= FORCE_EXTENT_THRESHOLD out of
 718          * bitmaps, we may end up using more memory than this.
 719          */
 720         if (size < SZ_1G)
 721                 max_bytes = MAX_CACHE_BYTES_PER_GIG;
 722         else
 723                 max_bytes = MAX_CACHE_BYTES_PER_GIG * div_u64(size, SZ_1G);
 724
 725         bitmap_bytes = ctl->total_bitmaps * ctl->unit;
 726
 727         /*
 728          * we want the extent entry threshold to always be at most 1/2 the max
 729          * bytes we can have, or whatever is less than that.
 730          */
 731         extent_bytes = max_bytes - bitmap_bytes;
 732         extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1);
 733
 734         ctl->extents_thresh =
 735                 div_u64(extent_bytes, sizeof(struct btrfs_free_space));
 736 }
 737
 738 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
 739                                    struct btrfs_free_space_ctl *ctl,
 740                                    struct btrfs_path *path, u64 offset)
 741 {
 742         struct btrfs_fs_info *fs_info = root->fs_info;
 743         struct btrfs_free_space_header *header;
 744         struct extent_buffer *leaf;
 745         struct btrfs_io_ctl io_ctl;
 746         struct btrfs_key key;
 747         struct btrfs_free_space *e, *n;
 748         LIST_HEAD(bitmaps);
 749         u64 num_entries;
 750         u64 num_bitmaps;
 751         u64 generation;
 752         u8 type;
 753         int ret = 0;
 754
 755         /* Nothing in the space cache, goodbye */
 756         if (!i_size_read(inode))
 757                 return 0;
 758
 759         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
 760         key.offset = offset;
 761         key.type = 0;
 762
 763         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 764         if (ret < 0)
 765                 return 0;
 766         else if (ret > 0) {
 767                 btrfs_release_path(path);
 768                 return 0;
 769         }
 770
 771         ret = -1;
 772
 773         leaf = path->nodes[0];
 774         header = btrfs_item_ptr(leaf, path->slots[0],
 775                                 struct btrfs_free_space_header);
 776         num_entries = btrfs_free_space_entries(leaf, header);
 777         num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
 778         generation = btrfs_free_space_generation(leaf, header);
 779         btrfs_release_path(path);
 780
 781         if (!BTRFS_I(inode)->generation) {
 782                 btrfs_info(fs_info,
 783                            "the free space cache file (%llu) is invalid, skip it",
 784                            offset);
 785                 return 0;
 786         }
 787
 788         if (BTRFS_I(inode)->generation != generation) {
 789                 btrfs_err(fs_info,
 790                           "free space inode generation (%llu) did not match free space cache generation (%llu)",
 791                           BTRFS_I(inode)->generation, generation);
 792                 return 0;
 793         }
 794
 795         if (!num_entries)
 796                 return 0;
 797
 798         ret = io_ctl_init(&io_ctl, inode, 0);
 799         if (ret)
 800                 return ret;
 801
 802         readahead_cache(inode);
 803
 804         ret = io_ctl_prepare_pages(&io_ctl, true);
 805         if (ret)
 806                 goto out;
 807
 808         ret = io_ctl_check_crc(&io_ctl, 0);
 809         if (ret)
 810                 goto free_cache;
 811
 812         ret = io_ctl_check_generation(&io_ctl, generation);
 813         if (ret)
 814                 goto free_cache;
 815
 816         while (num_entries) {
 817                 e = kmem_cache_zalloc(btrfs_free_space_cachep,
 818                                       GFP_NOFS);
 819                 if (!e) {
 820                         ret = -ENOMEM;
 821                         goto free_cache;
 822                 }
 823
 824                 ret = io_ctl_read_entry(&io_ctl, e, &type);
 825                 if (ret) {
 826                         kmem_cache_free(btrfs_free_space_cachep, e);
 827                         goto free_cache;
 828                 }
 829
 830                 if (!e->bytes) {
 831                         ret = -1;
 832                         kmem_cache_free(btrfs_free_space_cachep, e);
 833                         goto free_cache;
 834                 }
 835
 836                 if (type == BTRFS_FREE_SPACE_EXTENT) {
 837                         spin_lock(&ctl->tree_lock);
 838                         ret = link_free_space(ctl, e);
 839                         spin_unlock(&ctl->tree_lock);
 840                         if (ret) {
 841                                 btrfs_err(fs_info,
 842                                         "Duplicate entries in free space cache, dumping");
 843                                 kmem_cache_free(btrfs_free_space_cachep, e);
 844                                 goto free_cache;
 845                         }
 846                 } else {
 847                         ASSERT(num_bitmaps);
 848                         num_bitmaps--;
 849                         e->bitmap = kmem_cache_zalloc(
 850                                         btrfs_free_space_bitmap_cachep, GFP_NOFS);
 851                         if (!e->bitmap) {
 852                                 ret = -ENOMEM;
 853                                 kmem_cache_free(
 854                                         btrfs_free_space_cachep, e);
 855                                 goto free_cache;
 856                         }
 857                         spin_lock(&ctl->tree_lock);
 858                         ret = link_free_space(ctl, e);
 859                         if (ret) {
 860                                 spin_unlock(&ctl->tree_lock);
 861                                 btrfs_err(fs_info,
 862                                         "Duplicate entries in free space cache, dumping");
 863                                 kmem_cache_free(btrfs_free_space_cachep, e);
 864                                 goto free_cache;
 865                         }
 866                         ctl->total_bitmaps++;
 867                         recalculate_thresholds(ctl);
 868                         spin_unlock(&ctl->tree_lock);
 869                         list_add_tail(&e->list, &bitmaps);
 870                 }
 871
 872                 num_entries--;
 873         }
 874
 875         io_ctl_unmap_page(&io_ctl);
 876
 877         /*
 878          * We add the bitmaps at the end of the entries in order that
 879          * the bitmap entries are added to the cache.
 880          */
 881         list_for_each_entry_safe(e, n, &bitmaps, list) {
 882                 list_del_init(&e->list);
 883                 ret = io_ctl_read_bitmap(&io_ctl, e);
 884                 if (ret)
 885                         goto free_cache;
 886         }
 887
 888         io_ctl_drop_pages(&io_ctl);
 889         ret = 1;
 890 out:
 891         io_ctl_free(&io_ctl);
 892         return ret;
 893 free_cache:
 894         io_ctl_drop_pages(&io_ctl);
 895
 896         spin_lock(&ctl->tree_lock);
 897         __btrfs_remove_free_space_cache(ctl);
 898         spin_unlock(&ctl->tree_lock);
 899         goto out;
 900 }
 901
 902 static int copy_free_space_cache(struct btrfs_block_group *block_group,
 903                                  struct btrfs_free_space_ctl *ctl)
 904 {
 905         struct btrfs_free_space *info;
 906         struct rb_node *n;
 907         int ret = 0;
 908
 909         while (!ret && (n = rb_first(&ctl->free_space_offset)) != NULL) {
 910                 info = rb_entry(n, struct btrfs_free_space, offset_index);
 911                 if (!info->bitmap) {
 912                         const u64 offset = info->offset;
 913                         const u64 bytes = info->bytes;
 914
 915                         unlink_free_space(ctl, info, true);
 916                         spin_unlock(&ctl->tree_lock);
 917                         kmem_cache_free(btrfs_free_space_cachep, info);
 918                         ret = btrfs_add_free_space(block_group, offset, bytes);
 919                         spin_lock(&ctl->tree_lock);
 920                 } else {
 921                         u64 offset = info->offset;
 922                         u64 bytes = ctl->unit;
 923
 924                         ret = search_bitmap(ctl, info, &offset, &bytes, false);
 925                         if (ret == 0) {
 926                                 bitmap_clear_bits(ctl, info, offset, bytes, true);
 927                                 spin_unlock(&ctl->tree_lock);
 928                                 ret = btrfs_add_free_space(block_group, offset,
 929                                                            bytes);
 930                                 spin_lock(&ctl->tree_lock);
 931                         } else {
 932                                 free_bitmap(ctl, info);
 933                                 ret = 0;
 934                         }
 935                 }
 936                 cond_resched_lock(&ctl->tree_lock);
 937         }
 938         return ret;
 939 }
 940
 941 static struct lock_class_key btrfs_free_space_inode_key;
 942
 943 int load_free_space_cache(struct btrfs_block_group *block_group)
 944 {
 945         struct btrfs_fs_info *fs_info = block_group->fs_info;
 946         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
 947         struct btrfs_free_space_ctl tmp_ctl = {};
 948         struct inode *inode;
 949         struct btrfs_path *path;
 950         int ret = 0;
 951         bool matched;
 952         u64 used = block_group->used;
 953
 954         /*
 955          * Because we could potentially discard our loaded free space, we want
 956          * to load everything into a temporary structure first, and then if it's
 957          * valid copy it all into the actual free space ctl.
 958          */
 959         btrfs_init_free_space_ctl(block_group, &tmp_ctl);
 960
 961         /*
 962          * If this block group has been marked to be cleared for one reason or
 963          * another then we can't trust the on disk cache, so just return.
 964          */
 965         spin_lock(&block_group->lock);
 966         if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
 967                 spin_unlock(&block_group->lock);
 968                 return 0;
 969         }
 970         spin_unlock(&block_group->lock);
 971
 972         path = btrfs_alloc_path();
 973         if (!path)
 974                 return 0;
 975         path->search_commit_root = 1;
 976         path->skip_locking = 1;
 977
 978         /*
 979          * We must pass a path with search_commit_root set to btrfs_iget in
 980          * order to avoid a deadlock when allocating extents for the tree root.
 981          *
 982          * When we are COWing an extent buffer from the tree root, when looking
 983          * for a free extent, at extent-tree.c:find_free_extent(), we can find
 984          * block group without its free space cache loaded. When we find one
 985          * we must load its space cache which requires reading its free space
 986          * cache's inode item from the root tree. If this inode item is located
 987          * in the same leaf that we started COWing before, then we end up in
 988          * deadlock on the extent buffer (trying to read lock it when we
 989          * previously write locked it).
 990          *
 991          * It's safe to read the inode item using the commit root because
 992          * block groups, once loaded, stay in memory forever (until they are
 993          * removed) as well as their space caches once loaded. New block groups
 994          * once created get their ->cached field set to BTRFS_CACHE_FINISHED so
 995          * we will never try to read their inode item while the fs is mounted.
 996          */
 997         inode = lookup_free_space_inode(block_group, path);
 998         if (IS_ERR(inode)) {
 999                 btrfs_free_path(path);
1000                 return 0;
1001         }
1002
1003         /* We may have converted the inode and made the cache invalid. */
1004         spin_lock(&block_group->lock);
1005         if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
1006                 spin_unlock(&block_group->lock);
1007                 btrfs_free_path(path);
1008                 goto out;
1009         }
1010         spin_unlock(&block_group->lock);
1011
1012         /*
1013          * Reinitialize the class of struct inode's mapping->invalidate_lock for
1014          * free space inodes to prevent false positives related to locks for normal
1015          * inodes.
1016          */
1017         lockdep_set_class(&(&inode->i_data)->invalidate_lock,
1018                           &btrfs_free_space_inode_key);
1019
1020         ret = __load_free_space_cache(fs_info->tree_root, inode, &tmp_ctl,
1021                                       path, block_group->start);
1022         btrfs_free_path(path);
1023         if (ret <= 0)
1024                 goto out;
1025
1026         matched = (tmp_ctl.free_space == (block_group->length - used -
1027                                           block_group->bytes_super));
1028
1029         if (matched) {
1030                 spin_lock(&tmp_ctl.tree_lock);
1031                 ret = copy_free_space_cache(block_group, &tmp_ctl);
1032                 spin_unlock(&tmp_ctl.tree_lock);
1033                 /*
1034                  * ret == 1 means we successfully loaded the free space cache,
1035                  * so we need to re-set it here.
1036                  */
1037                 if (ret == 0)
1038                         ret = 1;
1039         } else {
1040                 /*
1041                  * We need to call the _locked variant so we don't try to update
1042                  * the discard counters.
1043                  */
1044                 spin_lock(&tmp_ctl.tree_lock);
1045                 __btrfs_remove_free_space_cache(&tmp_ctl);
1046                 spin_unlock(&tmp_ctl.tree_lock);
1047                 btrfs_warn(fs_info,
1048                            "block group %llu has wrong amount of free space",
1049                            block_group->start);
1050                 ret = -1;
1051         }
1052 out:
1053         if (ret < 0) {
1054                 /* This cache is bogus, make sure it gets cleared */
1055                 spin_lock(&block_group->lock);
1056                 block_group->disk_cache_state = BTRFS_DC_CLEAR;
1057                 spin_unlock(&block_group->lock);
1058                 ret = 0;
1059
1060                 btrfs_warn(fs_info,
1061                            "failed to load free space cache for block group %llu, rebuilding it now",
1062                            block_group->start);
1063         }
1064
1065         spin_lock(&ctl->tree_lock);
1066         btrfs_discard_update_discardable(block_group);
1067         spin_unlock(&ctl->tree_lock);
1068         iput(inode);
1069         return ret;
1070 }
1071
1072 static noinline_for_stack
1073 int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl,
1074                               struct btrfs_free_space_ctl *ctl,
1075                               struct btrfs_block_group *block_group,
1076                               int *entries, int *bitmaps,
1077                               struct list_head *bitmap_list)
1078 {
1079         int ret;
1080         struct btrfs_free_cluster *cluster = NULL;
1081         struct btrfs_free_cluster *cluster_locked = NULL;
1082         struct rb_node *node = rb_first(&ctl->free_space_offset);
1083         struct btrfs_trim_range *trim_entry;
1084
1085         /* Get the cluster for this block_group if it exists */
1086         if (block_group && !list_empty(&block_group->cluster_list)) {
1087                 cluster = list_entry(block_group->cluster_list.next,
1088                                      struct btrfs_free_cluster,
1089                                      block_group_list);
1090         }
1091
1092         if (!node && cluster) {
1093                 cluster_locked = cluster;
1094                 spin_lock(&cluster_locked->lock);
1095                 node = rb_first(&cluster->root);
1096                 cluster = NULL;
1097         }
1098
1099         /* Write out the extent entries */
1100         while (node) {
1101                 struct btrfs_free_space *e;
1102
1103                 e = rb_entry(node, struct btrfs_free_space, offset_index);
1104                 *entries += 1;
1105
1106                 ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
1107                                        e->bitmap);
1108                 if (ret)
1109                         goto fail;
1110
1111                 if (e->bitmap) {
1112                         list_add_tail(&e->list, bitmap_list);
1113                         *bitmaps += 1;
1114                 }
1115                 node = rb_next(node);
1116                 if (!node && cluster) {
1117                         node = rb_first(&cluster->root);
1118                         cluster_locked = cluster;
1119                         spin_lock(&cluster_locked->lock);
1120                         cluster = NULL;
1121                 }
1122         }
1123         if (cluster_locked) {
1124                 spin_unlock(&cluster_locked->lock);
1125                 cluster_locked = NULL;
1126         }
1127
1128         /*
1129          * Make sure we don't miss any range that was removed from our rbtree
1130          * because trimming is running. Otherwise after a umount+mount (or crash
1131          * after committing the transaction) we would leak free space and get
1132          * an inconsistent free space cache report from fsck.
1133          */
1134         list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) {
1135                 ret = io_ctl_add_entry(io_ctl, trim_entry->start,
1136                                        trim_entry->bytes, NULL);
1137                 if (ret)
1138                         goto fail;
1139                 *entries += 1;
1140         }
1141
1142         return 0;
1143 fail:
1144         if (cluster_locked)
1145                 spin_unlock(&cluster_locked->lock);
1146         return -ENOSPC;
1147 }
1148
1149 static noinline_for_stack int
1150 update_cache_item(struct btrfs_trans_handle *trans,
1151                   struct btrfs_root *root,
1152                   struct inode *inode,
1153                   struct btrfs_path *path, u64 offset,
1154                   int entries, int bitmaps)
1155 {
1156         struct btrfs_key key;
1157         struct btrfs_free_space_header *header;
1158         struct extent_buffer *leaf;
1159         int ret;
1160
1161         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1162         key.offset = offset;
1163         key.type = 0;
1164
1165         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1166         if (ret < 0) {
1167                 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1168                                  EXTENT_DELALLOC, NULL);
1169                 goto fail;
1170         }
1171         leaf = path->nodes[0];
1172         if (ret > 0) {
1173                 struct btrfs_key found_key;
1174                 ASSERT(path->slots[0]);
1175                 path->slots[0]--;
1176                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1177                 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1178                     found_key.offset != offset) {
1179                         clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1180                                          inode->i_size - 1, EXTENT_DELALLOC,
1181                                          NULL);
1182                         btrfs_release_path(path);
1183                         goto fail;
1184                 }
1185         }
1186
1187         BTRFS_I(inode)->generation = trans->transid;
1188         header = btrfs_item_ptr(leaf, path->slots[0],
1189                                 struct btrfs_free_space_header);
1190         btrfs_set_free_space_entries(leaf, header, entries);
1191         btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1192         btrfs_set_free_space_generation(leaf, header, trans->transid);
1193         btrfs_mark_buffer_dirty(trans, leaf);
1194         btrfs_release_path(path);
1195
1196         return 0;
1197
1198 fail:
1199         return -1;
1200 }
1201
1202 static noinline_for_stack int write_pinned_extent_entries(
1203                             struct btrfs_trans_handle *trans,
1204                             struct btrfs_block_group *block_group,
1205                             struct btrfs_io_ctl *io_ctl,
1206                             int *entries)
1207 {
1208         u64 start, extent_start, extent_end, len;
1209         struct extent_io_tree *unpin = NULL;
1210         int ret;
1211
1212         if (!block_group)
1213                 return 0;
1214
1215         /*
1216          * We want to add any pinned extents to our free space cache
1217          * so we don't leak the space
1218          *
1219          * We shouldn't have switched the pinned extents yet so this is the
1220          * right one
1221          */
1222         unpin = &trans->transaction->pinned_extents;
1223
1224         start = block_group->start;
1225
1226         while (start < block_group->start + block_group->length) {
1227                 if (!find_first_extent_bit(unpin, start,
1228                                            &extent_start, &extent_end,
1229                                            EXTENT_DIRTY, NULL))
1230                         return 0;
1231
1232                 /* This pinned extent is out of our range */
1233                 if (extent_start >= block_group->start + block_group->length)
1234                         return 0;
1235
1236                 extent_start = max(extent_start, start);
1237                 extent_end = min(block_group->start + block_group->length,
1238                                  extent_end + 1);
1239                 len = extent_end - extent_start;
1240
1241                 *entries += 1;
1242                 ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
1243                 if (ret)
1244                         return -ENOSPC;
1245
1246                 start = extent_end;
1247         }
1248
1249         return 0;
1250 }
1251
1252 static noinline_for_stack int
1253 write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list)
1254 {
1255         struct btrfs_free_space *entry, *next;
1256         int ret;
1257
1258         /* Write out the bitmaps */
1259         list_for_each_entry_safe(entry, next, bitmap_list, list) {
1260                 ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
1261                 if (ret)
1262                         return -ENOSPC;
1263                 list_del_init(&entry->list);
1264         }
1265
1266         return 0;
1267 }
1268
1269 static int flush_dirty_cache(struct inode *inode)
1270 {
1271         int ret;
1272
1273         ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1274         if (ret)
1275                 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1276                                  EXTENT_DELALLOC, NULL);
1277
1278         return ret;
1279 }
1280
1281 static void noinline_for_stack
1282 cleanup_bitmap_list(struct list_head *bitmap_list)
1283 {
1284         struct btrfs_free_space *entry, *next;
1285
1286         list_for_each_entry_safe(entry, next, bitmap_list, list)
1287                 list_del_init(&entry->list);
1288 }
1289
1290 static void noinline_for_stack
1291 cleanup_write_cache_enospc(struct inode *inode,
1292                            struct btrfs_io_ctl *io_ctl,
1293                            struct extent_state **cached_state)
1294 {
1295         io_ctl_drop_pages(io_ctl);
1296         unlock_extent(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1297                       cached_state);
1298 }
1299
1300 static int __btrfs_wait_cache_io(struct btrfs_root *root,
1301                                  struct btrfs_trans_handle *trans,
1302                                  struct btrfs_block_group *block_group,
1303                                  struct btrfs_io_ctl *io_ctl,
1304                                  struct btrfs_path *path, u64 offset)
1305 {
1306         int ret;
1307         struct inode *inode = io_ctl->inode;
1308
1309         if (!inode)
1310                 return 0;
1311
1312         /* Flush the dirty pages in the cache file. */
1313         ret = flush_dirty_cache(inode);
1314         if (ret)
1315                 goto out;
1316
1317         /* Update the cache item to tell everyone this cache file is valid. */
1318         ret = update_cache_item(trans, root, inode, path, offset,
1319                                 io_ctl->entries, io_ctl->bitmaps);
1320 out:
1321         if (ret) {
1322                 invalidate_inode_pages2(inode->i_mapping);
1323                 BTRFS_I(inode)->generation = 0;
1324                 if (block_group)
1325                         btrfs_debug(root->fs_info,
1326           "failed to write free space cache for block group %llu error %d",
1327                                   block_group->start, ret);
1328         }
1329         btrfs_update_inode(trans, BTRFS_I(inode));
1330
1331         if (block_group) {
1332                 /* the dirty list is protected by the dirty_bgs_lock */
1333                 spin_lock(&trans->transaction->dirty_bgs_lock);
1334
1335                 /* the disk_cache_state is protected by the block group lock */
1336                 spin_lock(&block_group->lock);
1337
1338                 /*
1339                  * only mark this as written if we didn't get put back on
1340                  * the dirty list while waiting for IO.   Otherwise our
1341                  * cache state won't be right, and we won't get written again
1342                  */
1343                 if (!ret && list_empty(&block_group->dirty_list))
1344                         block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1345                 else if (ret)
1346                         block_group->disk_cache_state = BTRFS_DC_ERROR;
1347
1348                 spin_unlock(&block_group->lock);
1349                 spin_unlock(&trans->transaction->dirty_bgs_lock);
1350                 io_ctl->inode = NULL;
1351                 iput(inode);
1352         }
1353
1354         return ret;
1355
1356 }
1357
1358 int btrfs_wait_cache_io(struct btrfs_trans_handle *trans,
1359                         struct btrfs_block_group *block_group,
1360                         struct btrfs_path *path)
1361 {
1362         return __btrfs_wait_cache_io(block_group->fs_info->tree_root, trans,
1363                                      block_group, &block_group->io_ctl,
1364                                      path, block_group->start);
1365 }
1366
1367 /*
1368  * Write out cached info to an inode.
1369  *
1370  * @inode:       freespace inode we are writing out
1371  * @ctl:         free space cache we are going to write out
1372  * @block_group: block_group for this cache if it belongs to a block_group
1373  * @io_ctl:      holds context for the io
1374  * @trans:       the trans handle
1375  *
1376  * This function writes out a free space cache struct to disk for quick recovery
1377  * on mount.  This will return 0 if it was successful in writing the cache out,
1378  * or an errno if it was not.
1379  */
1380 static int __btrfs_write_out_cache(struct inode *inode,
1381                                    struct btrfs_free_space_ctl *ctl,
1382                                    struct btrfs_block_group *block_group,
1383                                    struct btrfs_io_ctl *io_ctl,
1384                                    struct btrfs_trans_handle *trans)
1385 {
1386         struct extent_state *cached_state = NULL;
1387         LIST_HEAD(bitmap_list);
1388         int entries = 0;
1389         int bitmaps = 0;
1390         int ret;
1391         int must_iput = 0;
1392
1393         if (!i_size_read(inode))
1394                 return -EIO;
1395
1396         WARN_ON(io_ctl->pages);
1397         ret = io_ctl_init(io_ctl, inode, 1);
1398         if (ret)
1399                 return ret;
1400
1401         if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) {
1402                 down_write(&block_group->data_rwsem);
1403                 spin_lock(&block_group->lock);
1404                 if (block_group->delalloc_bytes) {
1405                         block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1406                         spin_unlock(&block_group->lock);
1407                         up_write(&block_group->data_rwsem);
1408                         BTRFS_I(inode)->generation = 0;
1409                         ret = 0;
1410                         must_iput = 1;
1411                         goto out;
1412                 }
1413                 spin_unlock(&block_group->lock);
1414         }
1415
1416         /* Lock all pages first so we can lock the extent safely. */
1417         ret = io_ctl_prepare_pages(io_ctl, false);
1418         if (ret)
1419                 goto out_unlock;
1420
1421         lock_extent(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1422                     &cached_state);
1423
1424         io_ctl_set_generation(io_ctl, trans->transid);
1425
1426         mutex_lock(&ctl->cache_writeout_mutex);
1427         /* Write out the extent entries in the free space cache */
1428         spin_lock(&ctl->tree_lock);
1429         ret = write_cache_extent_entries(io_ctl, ctl,
1430                                          block_group, &entries, &bitmaps,
1431                                          &bitmap_list);
1432         if (ret)
1433                 goto out_nospc_locked;
1434
1435         /*
1436          * Some spaces that are freed in the current transaction are pinned,
1437          * they will be added into free space cache after the transaction is
1438          * committed, we shouldn't lose them.
1439          *
1440          * If this changes while we are working we'll get added back to
1441          * the dirty list and redo it.  No locking needed
1442          */
1443         ret = write_pinned_extent_entries(trans, block_group, io_ctl, &entries);
1444         if (ret)
1445                 goto out_nospc_locked;
1446
1447         /*
1448          * At last, we write out all the bitmaps and keep cache_writeout_mutex
1449          * locked while doing it because a concurrent trim can be manipulating
1450          * or freeing the bitmap.
1451          */
1452         ret = write_bitmap_entries(io_ctl, &bitmap_list);
1453         spin_unlock(&ctl->tree_lock);
1454         mutex_unlock(&ctl->cache_writeout_mutex);
1455         if (ret)
1456                 goto out_nospc;
1457
1458         /* Zero out the rest of the pages just to make sure */
1459         io_ctl_zero_remaining_pages(io_ctl);
1460
1461         /* Everything is written out, now we dirty the pages in the file. */
1462         ret = btrfs_dirty_pages(BTRFS_I(inode), io_ctl->pages,
1463                                 io_ctl->num_pages, 0, i_size_read(inode),
1464                                 &cached_state, false);
1465         if (ret)
1466                 goto out_nospc;
1467
1468         if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1469                 up_write(&block_group->data_rwsem);
1470         /*
1471          * Release the pages and unlock the extent, we will flush
1472          * them out later
1473          */
1474         io_ctl_drop_pages(io_ctl);
1475         io_ctl_free(io_ctl);
1476
1477         unlock_extent(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1478                       &cached_state);
1479
1480         /*
1481          * at this point the pages are under IO and we're happy,
1482          * The caller is responsible for waiting on them and updating
1483          * the cache and the inode
1484          */
1485         io_ctl->entries = entries;
1486         io_ctl->bitmaps = bitmaps;
1487
1488         ret = btrfs_fdatawrite_range(inode, 0, (u64)-1);
1489         if (ret)
1490                 goto out;
1491
1492         return 0;
1493
1494 out_nospc_locked:
1495         cleanup_bitmap_list(&bitmap_list);
1496         spin_unlock(&ctl->tree_lock);
1497         mutex_unlock(&ctl->cache_writeout_mutex);
1498
1499 out_nospc:
1500         cleanup_write_cache_enospc(inode, io_ctl, &cached_state);
1501
1502 out_unlock:
1503         if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1504                 up_write(&block_group->data_rwsem);
1505
1506 out:
1507         io_ctl->inode = NULL;
1508         io_ctl_free(io_ctl);
1509         if (ret) {
1510                 invalidate_inode_pages2(inode->i_mapping);
1511                 BTRFS_I(inode)->generation = 0;
1512         }
1513         btrfs_update_inode(trans, BTRFS_I(inode));
1514         if (must_iput)
1515                 iput(inode);
1516         return ret;
1517 }
1518
1519 int btrfs_write_out_cache(struct btrfs_trans_handle *trans,
1520                           struct btrfs_block_group *block_group,
1521                           struct btrfs_path *path)
1522 {
1523         struct btrfs_fs_info *fs_info = trans->fs_info;
1524         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1525         struct inode *inode;
1526         int ret = 0;
1527
1528         spin_lock(&block_group->lock);
1529         if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1530                 spin_unlock(&block_group->lock);
1531                 return 0;
1532         }
1533         spin_unlock(&block_group->lock);
1534
1535         inode = lookup_free_space_inode(block_group, path);
1536         if (IS_ERR(inode))
1537                 return 0;
1538
1539         ret = __btrfs_write_out_cache(inode, ctl, block_group,
1540                                       &block_group->io_ctl, trans);
1541         if (ret) {
1542                 btrfs_debug(fs_info,
1543           "failed to write free space cache for block group %llu error %d",
1544                           block_group->start, ret);
1545                 spin_lock(&block_group->lock);
1546                 block_group->disk_cache_state = BTRFS_DC_ERROR;
1547                 spin_unlock(&block_group->lock);
1548
1549                 block_group->io_ctl.inode = NULL;
1550                 iput(inode);
1551         }
1552
1553         /*
1554          * if ret == 0 the caller is expected to call btrfs_wait_cache_io
1555          * to wait for IO and put the inode
1556          */
1557
1558         return ret;
1559 }
1560
1561 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1562                                           u64 offset)
1563 {
1564         ASSERT(offset >= bitmap_start);
1565         offset -= bitmap_start;
1566         return (unsigned long)(div_u64(offset, unit));
1567 }
1568
1569 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1570 {
1571         return (unsigned long)(div_u64(bytes, unit));
1572 }
1573
1574 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1575                                    u64 offset)
1576 {
1577         u64 bitmap_start;
1578         u64 bytes_per_bitmap;
1579
1580         bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1581         bitmap_start = offset - ctl->start;
1582         bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1583         bitmap_start *= bytes_per_bitmap;
1584         bitmap_start += ctl->start;
1585
1586         return bitmap_start;
1587 }
1588
1589 static int tree_insert_offset(struct btrfs_free_space_ctl *ctl,
1590                               struct btrfs_free_cluster *cluster,
1591                               struct btrfs_free_space *new_entry)
1592 {
1593         struct rb_root *root;
1594         struct rb_node **p;
1595         struct rb_node *parent = NULL;
1596
1597         lockdep_assert_held(&ctl->tree_lock);
1598
1599         if (cluster) {
1600                 lockdep_assert_held(&cluster->lock);
1601                 root = &cluster->root;
1602         } else {
1603                 root = &ctl->free_space_offset;
1604         }
1605
1606         p = &root->rb_node;
1607
1608         while (*p) {
1609                 struct btrfs_free_space *info;
1610
1611                 parent = *p;
1612                 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1613
1614                 if (new_entry->offset < info->offset) {
1615                         p = &(*p)->rb_left;
1616                 } else if (new_entry->offset > info->offset) {
1617                         p = &(*p)->rb_right;
1618                 } else {
1619                         /*
1620                          * we could have a bitmap entry and an extent entry
1621                          * share the same offset.  If this is the case, we want
1622                          * the extent entry to always be found first if we do a
1623                          * linear search through the tree, since we want to have
1624                          * the quickest allocation time, and allocating from an
1625                          * extent is faster than allocating from a bitmap.  So
1626                          * if we're inserting a bitmap and we find an entry at
1627                          * this offset, we want to go right, or after this entry
1628                          * logically.  If we are inserting an extent and we've
1629                          * found a bitmap, we want to go left, or before
1630                          * logically.
1631                          */
1632                         if (new_entry->bitmap) {
1633                                 if (info->bitmap) {
1634                                         WARN_ON_ONCE(1);
1635                                         return -EEXIST;
1636                                 }
1637                                 p = &(*p)->rb_right;
1638                         } else {
1639                                 if (!info->bitmap) {
1640                                         WARN_ON_ONCE(1);
1641                                         return -EEXIST;
1642                                 }
1643                                 p = &(*p)->rb_left;
1644                         }
1645                 }
1646         }
1647
1648         rb_link_node(&new_entry->offset_index, parent, p);
1649         rb_insert_color(&new_entry->offset_index, root);
1650
1651         return 0;
1652 }
1653
1654 /*
1655  * This is a little subtle.  We *only* have ->max_extent_size set if we actually
1656  * searched through the bitmap and figured out the largest ->max_extent_size,
1657  * otherwise it's 0.  In the case that it's 0 we don't want to tell the
1658  * allocator the wrong thing, we want to use the actual real max_extent_size
1659  * we've found already if it's larger, or we want to use ->bytes.
1660  *
1661  * This matters because find_free_space() will skip entries who's ->bytes is
1662  * less than the required bytes.  So if we didn't search down this bitmap, we
1663  * may pick some previous entry that has a smaller ->max_extent_size than we
1664  * have.  For example, assume we have two entries, one that has
1665  * ->max_extent_size set to 4K and ->bytes set to 1M.  A second entry hasn't set
1666  * ->max_extent_size yet, has ->bytes set to 8K and it's contiguous.  We will
1667  *  call into find_free_space(), and return with max_extent_size == 4K, because
1668  *  that first bitmap entry had ->max_extent_size set, but the second one did
1669  *  not.  If instead we returned 8K we'd come in searching for 8K, and find the
1670  *  8K contiguous range.
1671  *
1672  *  Consider the other case, we have 2 8K chunks in that second entry and still
1673  *  don't have ->max_extent_size set.  We'll return 16K, and the next time the
1674  *  allocator comes in it'll fully search our second bitmap, and this time it'll
1675  *  get an uptodate value of 8K as the maximum chunk size.  Then we'll get the
1676  *  right allocation the next loop through.
1677  */
1678 static inline u64 get_max_extent_size(const struct btrfs_free_space *entry)
1679 {
1680         if (entry->bitmap && entry->max_extent_size)
1681                 return entry->max_extent_size;
1682         return entry->bytes;
1683 }
1684
1685 /*
1686  * We want the largest entry to be leftmost, so this is inverted from what you'd
1687  * normally expect.
1688  */
1689 static bool entry_less(struct rb_node *node, const struct rb_node *parent)
1690 {
1691         const struct btrfs_free_space *entry, *exist;
1692
1693         entry = rb_entry(node, struct btrfs_free_space, bytes_index);
1694         exist = rb_entry(parent, struct btrfs_free_space, bytes_index);
1695         return get_max_extent_size(exist) < get_max_extent_size(entry);
1696 }
1697
1698 /*
1699  * searches the tree for the given offset.
1700  *
1701  * fuzzy - If this is set, then we are trying to make an allocation, and we just
1702  * want a section that has at least bytes size and comes at or after the given
1703  * offset.
1704  */
1705 static struct btrfs_free_space *
1706 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1707                    u64 offset, int bitmap_only, int fuzzy)
1708 {
1709         struct rb_node *n = ctl->free_space_offset.rb_node;
1710         struct btrfs_free_space *entry = NULL, *prev = NULL;
1711
1712         lockdep_assert_held(&ctl->tree_lock);
1713
1714         /* find entry that is closest to the 'offset' */
1715         while (n) {
1716                 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1717                 prev = entry;
1718
1719                 if (offset < entry->offset)
1720                         n = n->rb_left;
1721                 else if (offset > entry->offset)
1722                         n = n->rb_right;
1723                 else
1724                         break;
1725
1726                 entry = NULL;
1727         }
1728
1729         if (bitmap_only) {
1730                 if (!entry)
1731                         return NULL;
1732                 if (entry->bitmap)
1733                         return entry;
1734
1735                 /*
1736                  * bitmap entry and extent entry may share same offset,
1737                  * in that case, bitmap entry comes after extent entry.
1738                  */
1739                 n = rb_next(n);
1740                 if (!n)
1741                         return NULL;
1742                 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1743                 if (entry->offset != offset)
1744                         return NULL;
1745
1746                 WARN_ON(!entry->bitmap);
1747                 return entry;
1748         } else if (entry) {
1749                 if (entry->bitmap) {
1750                         /*
1751                          * if previous extent entry covers the offset,
1752                          * we should return it instead of the bitmap entry
1753                          */
1754                         n = rb_prev(&entry->offset_index);
1755                         if (n) {
1756                                 prev = rb_entry(n, struct btrfs_free_space,
1757                                                 offset_index);
1758                                 if (!prev->bitmap &&
1759                                     prev->offset + prev->bytes > offset)
1760                                         entry = prev;
1761                         }
1762                 }
1763                 return entry;
1764         }
1765
1766         if (!prev)
1767                 return NULL;
1768
1769         /* find last entry before the 'offset' */
1770         entry = prev;
1771         if (entry->offset > offset) {
1772                 n = rb_prev(&entry->offset_index);
1773                 if (n) {
1774                         entry = rb_entry(n, struct btrfs_free_space,
1775                                         offset_index);
1776                         ASSERT(entry->offset <= offset);
1777                 } else {
1778                         if (fuzzy)
1779                                 return entry;
1780                         else
1781                                 return NULL;
1782                 }
1783         }
1784
1785         if (entry->bitmap) {
1786                 n = rb_prev(&entry->offset_index);
1787                 if (n) {
1788                         prev = rb_entry(n, struct btrfs_free_space,
1789                                         offset_index);
1790                         if (!prev->bitmap &&
1791                             prev->offset + prev->bytes > offset)
1792                                 return prev;
1793                 }
1794                 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1795                         return entry;
1796         } else if (entry->offset + entry->bytes > offset)
1797                 return entry;
1798
1799         if (!fuzzy)
1800                 return NULL;
1801
1802         while (1) {
1803                 n = rb_next(&entry->offset_index);
1804                 if (!n)
1805                         return NULL;
1806                 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1807                 if (entry->bitmap) {
1808                         if (entry->offset + BITS_PER_BITMAP *
1809                             ctl->unit > offset)
1810                                 break;
1811                 } else {
1812                         if (entry->offset + entry->bytes > offset)
1813                                 break;
1814                 }
1815         }
1816         return entry;
1817 }
1818
1819 static inline void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1820                                      struct btrfs_free_space *info,
1821                                      bool update_stat)
1822 {
1823         lockdep_assert_held(&ctl->tree_lock);
1824
1825         rb_erase(&info->offset_index, &ctl->free_space_offset);
1826         rb_erase_cached(&info->bytes_index, &ctl->free_space_bytes);
1827         ctl->free_extents--;
1828
1829         if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1830                 ctl->discardable_extents[BTRFS_STAT_CURR]--;
1831                 ctl->discardable_bytes[BTRFS_STAT_CURR] -= info->bytes;
1832         }
1833
1834         if (update_stat)
1835                 ctl->free_space -= info->bytes;
1836 }
1837
1838 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1839                            struct btrfs_free_space *info)
1840 {
1841         int ret = 0;
1842
1843         lockdep_assert_held(&ctl->tree_lock);
1844
1845         ASSERT(info->bytes || info->bitmap);
1846         ret = tree_insert_offset(ctl, NULL, info);
1847         if (ret)
1848                 return ret;
1849
1850         rb_add_cached(&info->bytes_index, &ctl->free_space_bytes, entry_less);
1851
1852         if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1853                 ctl->discardable_extents[BTRFS_STAT_CURR]++;
1854                 ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
1855         }
1856
1857         ctl->free_space += info->bytes;
1858         ctl->free_extents++;
1859         return ret;
1860 }
1861
1862 static void relink_bitmap_entry(struct btrfs_free_space_ctl *ctl,
1863                                 struct btrfs_free_space *info)
1864 {
1865         ASSERT(info->bitmap);
1866
1867         /*
1868          * If our entry is empty it's because we're on a cluster and we don't
1869          * want to re-link it into our ctl bytes index.
1870          */
1871         if (RB_EMPTY_NODE(&info->bytes_index))
1872                 return;
1873
1874         lockdep_assert_held(&ctl->tree_lock);
1875
1876         rb_erase_cached(&info->bytes_index, &ctl->free_space_bytes);
1877         rb_add_cached(&info->bytes_index, &ctl->free_space_bytes, entry_less);
1878 }
1879
1880 static inline void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1881                                      struct btrfs_free_space *info,
1882                                      u64 offset, u64 bytes, bool update_stat)
1883 {
1884         unsigned long start, count, end;
1885         int extent_delta = -1;
1886
1887         start = offset_to_bit(info->offset, ctl->unit, offset);
1888         count = bytes_to_bits(bytes, ctl->unit);
1889         end = start + count;
1890         ASSERT(end <= BITS_PER_BITMAP);
1891
1892         bitmap_clear(info->bitmap, start, count);
1893
1894         info->bytes -= bytes;
1895         if (info->max_extent_size > ctl->unit)
1896                 info->max_extent_size = 0;
1897
1898         relink_bitmap_entry(ctl, info);
1899
1900         if (start && test_bit(start - 1, info->bitmap))
1901                 extent_delta++;
1902
1903         if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
1904                 extent_delta++;
1905
1906         info->bitmap_extents += extent_delta;
1907         if (!btrfs_free_space_trimmed(info)) {
1908                 ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1909                 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
1910         }
1911
1912         if (update_stat)
1913                 ctl->free_space -= bytes;
1914 }
1915
1916 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1917                             struct btrfs_free_space *info, u64 offset,
1918                             u64 bytes)
1919 {
1920         unsigned long start, count, end;
1921         int extent_delta = 1;
1922
1923         start = offset_to_bit(info->offset, ctl->unit, offset);
1924         count = bytes_to_bits(bytes, ctl->unit);
1925         end = start + count;
1926         ASSERT(end <= BITS_PER_BITMAP);
1927
1928         bitmap_set(info->bitmap, start, count);
1929
1930         /*
1931          * We set some bytes, we have no idea what the max extent size is
1932          * anymore.
1933          */
1934         info->max_extent_size = 0;
1935         info->bytes += bytes;
1936         ctl->free_space += bytes;
1937
1938         relink_bitmap_entry(ctl, info);
1939
1940         if (start && test_bit(start - 1, info->bitmap))
1941                 extent_delta--;
1942
1943         if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
1944                 extent_delta--;
1945
1946         info->bitmap_extents += extent_delta;
1947         if (!btrfs_free_space_trimmed(info)) {
1948                 ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1949                 ctl->discardable_bytes[BTRFS_STAT_CURR] += bytes;
1950         }
1951 }
1952
1953 /*
1954  * If we can not find suitable extent, we will use bytes to record
1955  * the size of the max extent.
1956  */
1957 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1958                          struct btrfs_free_space *bitmap_info, u64 *offset,
1959                          u64 *bytes, bool for_alloc)
1960 {
1961         unsigned long found_bits = 0;
1962         unsigned long max_bits = 0;
1963         unsigned long bits, i;
1964         unsigned long next_zero;
1965         unsigned long extent_bits;
1966
1967         /*
1968          * Skip searching the bitmap if we don't have a contiguous section that
1969          * is large enough for this allocation.
1970          */
1971         if (for_alloc &&
1972             bitmap_info->max_extent_size &&
1973             bitmap_info->max_extent_size < *bytes) {
1974                 *bytes = bitmap_info->max_extent_size;
1975                 return -1;
1976         }
1977
1978         i = offset_to_bit(bitmap_info->offset, ctl->unit,
1979                           max_t(u64, *offset, bitmap_info->offset));
1980         bits = bytes_to_bits(*bytes, ctl->unit);
1981
1982         for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1983                 if (for_alloc && bits == 1) {
1984                         found_bits = 1;
1985                         break;
1986                 }
1987                 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1988                                                BITS_PER_BITMAP, i);
1989                 extent_bits = next_zero - i;
1990                 if (extent_bits >= bits) {
1991                         found_bits = extent_bits;
1992                         break;
1993                 } else if (extent_bits > max_bits) {
1994                         max_bits = extent_bits;
1995                 }
1996                 i = next_zero;
1997         }
1998
1999         if (found_bits) {
2000                 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
2001                 *bytes = (u64)(found_bits) * ctl->unit;
2002                 return 0;
2003         }
2004
2005         *bytes = (u64)(max_bits) * ctl->unit;
2006         bitmap_info->max_extent_size = *bytes;
2007         relink_bitmap_entry(ctl, bitmap_info);
2008         return -1;
2009 }
2010
2011 /* Cache the size of the max extent in bytes */
2012 static struct btrfs_free_space *
2013 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
2014                 unsigned long align, u64 *max_extent_size, bool use_bytes_index)
2015 {
2016         struct btrfs_free_space *entry;
2017         struct rb_node *node;
2018         u64 tmp;
2019         u64 align_off;
2020         int ret;
2021
2022         if (!ctl->free_space_offset.rb_node)
2023                 goto out;
2024 again:
2025         if (use_bytes_index) {
2026                 node = rb_first_cached(&ctl->free_space_bytes);
2027         } else {
2028                 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset),
2029                                            0, 1);
2030                 if (!entry)
2031                         goto out;
2032                 node = &entry->offset_index;
2033         }
2034
2035         for (; node; node = rb_next(node)) {
2036                 if (use_bytes_index)
2037                         entry = rb_entry(node, struct btrfs_free_space,
2038                                          bytes_index);
2039                 else
2040                         entry = rb_entry(node, struct btrfs_free_space,
2041                                          offset_index);
2042
2043                 /*
2044                  * If we are using the bytes index then all subsequent entries
2045                  * in this tree are going to be < bytes, so simply set the max
2046                  * extent size and exit the loop.
2047                  *
2048                  * If we're using the offset index then we need to keep going
2049                  * through the rest of the tree.
2050                  */
2051                 if (entry->bytes < *bytes) {
2052                         *max_extent_size = max(get_max_extent_size(entry),
2053                                                *max_extent_size);
2054                         if (use_bytes_index)
2055                                 break;
2056                         continue;
2057                 }
2058
2059                 /* make sure the space returned is big enough
2060                  * to match our requested alignment
2061                  */
2062                 if (*bytes >= align) {
2063                         tmp = entry->offset - ctl->start + align - 1;
2064                         tmp = div64_u64(tmp, align);
2065                         tmp = tmp * align + ctl->start;
2066                         align_off = tmp - entry->offset;
2067                 } else {
2068                         align_off = 0;
2069                         tmp = entry->offset;
2070                 }
2071
2072                 /*
2073                  * We don't break here if we're using the bytes index because we
2074                  * may have another entry that has the correct alignment that is
2075                  * the right size, so we don't want to miss that possibility.
2076                  * At worst this adds another loop through the logic, but if we
2077                  * broke here we could prematurely ENOSPC.
2078                  */
2079                 if (entry->bytes < *bytes + align_off) {
2080                         *max_extent_size = max(get_max_extent_size(entry),
2081                                                *max_extent_size);
2082                         continue;
2083                 }
2084
2085                 if (entry->bitmap) {
2086                         struct rb_node *old_next = rb_next(node);
2087                         u64 size = *bytes;
2088
2089                         ret = search_bitmap(ctl, entry, &tmp, &size, true);
2090                         if (!ret) {
2091                                 *offset = tmp;
2092                                 *bytes = size;
2093                                 return entry;
2094                         } else {
2095                                 *max_extent_size =
2096                                         max(get_max_extent_size(entry),
2097                                             *max_extent_size);
2098                         }
2099
2100                         /*
2101                          * The bitmap may have gotten re-arranged in the space
2102                          * index here because the max_extent_size may have been
2103                          * updated.  Start from the beginning again if this
2104                          * happened.
2105                          */
2106                         if (use_bytes_index && old_next != rb_next(node))
2107                                 goto again;
2108                         continue;
2109                 }
2110
2111                 *offset = tmp;
2112                 *bytes = entry->bytes - align_off;
2113                 return entry;
2114         }
2115 out:
2116         return NULL;
2117 }
2118
2119 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
2120                            struct btrfs_free_space *info, u64 offset)
2121 {
2122         info->offset = offset_to_bitmap(ctl, offset);
2123         info->bytes = 0;
2124         info->bitmap_extents = 0;
2125         INIT_LIST_HEAD(&info->list);
2126         link_free_space(ctl, info);
2127         ctl->total_bitmaps++;
2128         recalculate_thresholds(ctl);
2129 }
2130
2131 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
2132                         struct btrfs_free_space *bitmap_info)
2133 {
2134         /*
2135          * Normally when this is called, the bitmap is completely empty. However,
2136          * if we are blowing up the free space cache for one reason or another
2137          * via __btrfs_remove_free_space_cache(), then it may not be freed and
2138          * we may leave stats on the table.
2139          */
2140         if (bitmap_info->bytes && !btrfs_free_space_trimmed(bitmap_info)) {
2141                 ctl->discardable_extents[BTRFS_STAT_CURR] -=
2142                         bitmap_info->bitmap_extents;
2143                 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bitmap_info->bytes;
2144
2145         }
2146         unlink_free_space(ctl, bitmap_info, true);
2147         kmem_cache_free(btrfs_free_space_bitmap_cachep, bitmap_info->bitmap);
2148         kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
2149         ctl->total_bitmaps--;
2150         recalculate_thresholds(ctl);
2151 }
2152
2153 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
2154                               struct btrfs_free_space *bitmap_info,
2155                               u64 *offset, u64 *bytes)
2156 {
2157         u64 end;
2158         u64 search_start, search_bytes;
2159         int ret;
2160
2161 again:
2162         end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
2163
2164         /*
2165          * We need to search for bits in this bitmap.  We could only cover some
2166          * of the extent in this bitmap thanks to how we add space, so we need
2167          * to search for as much as it as we can and clear that amount, and then
2168          * go searching for the next bit.
2169          */
2170         search_start = *offset;
2171         search_bytes = ctl->unit;
2172         search_bytes = min(search_bytes, end - search_start + 1);
2173         ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes,
2174                             false);
2175         if (ret < 0 || search_start != *offset)
2176                 return -EINVAL;
2177
2178         /* We may have found more bits than what we need */
2179         search_bytes = min(search_bytes, *bytes);
2180
2181         /* Cannot clear past the end of the bitmap */
2182         search_bytes = min(search_bytes, end - search_start + 1);
2183
2184         bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes, true);
2185         *offset += search_bytes;
2186         *bytes -= search_bytes;
2187
2188         if (*bytes) {
2189                 struct rb_node *next = rb_next(&bitmap_info->offset_index);
2190                 if (!bitmap_info->bytes)
2191                         free_bitmap(ctl, bitmap_info);
2192
2193                 /*
2194                  * no entry after this bitmap, but we still have bytes to
2195                  * remove, so something has gone wrong.
2196                  */
2197                 if (!next)
2198                         return -EINVAL;
2199
2200                 bitmap_info = rb_entry(next, struct btrfs_free_space,
2201                                        offset_index);
2202
2203                 /*
2204                  * if the next entry isn't a bitmap we need to return to let the
2205                  * extent stuff do its work.
2206                  */
2207                 if (!bitmap_info->bitmap)
2208                         return -EAGAIN;
2209
2210                 /*
2211                  * Ok the next item is a bitmap, but it may not actually hold
2212                  * the information for the rest of this free space stuff, so
2213                  * look for it, and if we don't find it return so we can try
2214                  * everything over again.
2215                  */
2216                 search_start = *offset;
2217                 search_bytes = ctl->unit;
2218                 ret = search_bitmap(ctl, bitmap_info, &search_start,
2219                                     &search_bytes, false);
2220                 if (ret < 0 || search_start != *offset)
2221                         return -EAGAIN;
2222
2223                 goto again;
2224         } else if (!bitmap_info->bytes)
2225                 free_bitmap(ctl, bitmap_info);
2226
2227         return 0;
2228 }
2229
2230 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
2231                                struct btrfs_free_space *info, u64 offset,
2232                                u64 bytes, enum btrfs_trim_state trim_state)
2233 {
2234         u64 bytes_to_set = 0;
2235         u64 end;
2236
2237         /*
2238          * This is a tradeoff to make bitmap trim state minimal.  We mark the
2239          * whole bitmap untrimmed if at any point we add untrimmed regions.
2240          */
2241         if (trim_state == BTRFS_TRIM_STATE_UNTRIMMED) {
2242                 if (btrfs_free_space_trimmed(info)) {
2243                         ctl->discardable_extents[BTRFS_STAT_CURR] +=
2244                                 info->bitmap_extents;
2245                         ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
2246                 }
2247                 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2248         }
2249
2250         end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
2251
2252         bytes_to_set = min(end - offset, bytes);
2253
2254         bitmap_set_bits(ctl, info, offset, bytes_to_set);
2255
2256         return bytes_to_set;
2257
2258 }
2259
2260 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
2261                       struct btrfs_free_space *info)
2262 {
2263         struct btrfs_block_group *block_group = ctl->block_group;
2264         struct btrfs_fs_info *fs_info = block_group->fs_info;
2265         bool forced = false;
2266
2267 #ifdef CONFIG_BTRFS_DEBUG
2268         if (btrfs_should_fragment_free_space(block_group))
2269                 forced = true;
2270 #endif
2271
2272         /* This is a way to reclaim large regions from the bitmaps. */
2273         if (!forced && info->bytes >= FORCE_EXTENT_THRESHOLD)
2274                 return false;
2275
2276         /*
2277          * If we are below the extents threshold then we can add this as an
2278          * extent, and don't have to deal with the bitmap
2279          */
2280         if (!forced && ctl->free_extents < ctl->extents_thresh) {
2281                 /*
2282                  * If this block group has some small extents we don't want to
2283                  * use up all of our free slots in the cache with them, we want
2284                  * to reserve them to larger extents, however if we have plenty
2285                  * of cache left then go ahead an dadd them, no sense in adding
2286                  * the overhead of a bitmap if we don't have to.
2287                  */
2288                 if (info->bytes <= fs_info->sectorsize * 8) {
2289                         if (ctl->free_extents * 3 <= ctl->extents_thresh)
2290                                 return false;
2291                 } else {
2292                         return false;
2293                 }
2294         }
2295
2296         /*
2297          * The original block groups from mkfs can be really small, like 8
2298          * megabytes, so don't bother with a bitmap for those entries.  However
2299          * some block groups can be smaller than what a bitmap would cover but
2300          * are still large enough that they could overflow the 32k memory limit,
2301          * so allow those block groups to still be allowed to have a bitmap
2302          * entry.
2303          */
2304         if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->length)
2305                 return false;
2306
2307         return true;
2308 }
2309
2310 static const struct btrfs_free_space_op free_space_op = {
2311         .use_bitmap             = use_bitmap,
2312 };
2313
2314 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
2315                               struct btrfs_free_space *info)
2316 {
2317         struct btrfs_free_space *bitmap_info;
2318         struct btrfs_block_group *block_group = NULL;
2319         int added = 0;
2320         u64 bytes, offset, bytes_added;
2321         enum btrfs_trim_state trim_state;
2322         int ret;
2323
2324         bytes = info->bytes;
2325         offset = info->offset;
2326         trim_state = info->trim_state;
2327
2328         if (!ctl->op->use_bitmap(ctl, info))
2329                 return 0;
2330
2331         if (ctl->op == &free_space_op)
2332                 block_group = ctl->block_group;
2333 again:
2334         /*
2335          * Since we link bitmaps right into the cluster we need to see if we
2336          * have a cluster here, and if so and it has our bitmap we need to add
2337          * the free space to that bitmap.
2338          */
2339         if (block_group && !list_empty(&block_group->cluster_list)) {
2340                 struct btrfs_free_cluster *cluster;
2341                 struct rb_node *node;
2342                 struct btrfs_free_space *entry;
2343
2344                 cluster = list_entry(block_group->cluster_list.next,
2345                                      struct btrfs_free_cluster,
2346                                      block_group_list);
2347                 spin_lock(&cluster->lock);
2348                 node = rb_first(&cluster->root);
2349                 if (!node) {
2350                         spin_unlock(&cluster->lock);
2351                         goto no_cluster_bitmap;
2352                 }
2353
2354                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2355                 if (!entry->bitmap) {
2356                         spin_unlock(&cluster->lock);
2357                         goto no_cluster_bitmap;
2358                 }
2359
2360                 if (entry->offset == offset_to_bitmap(ctl, offset)) {
2361                         bytes_added = add_bytes_to_bitmap(ctl, entry, offset,
2362                                                           bytes, trim_state);
2363                         bytes -= bytes_added;
2364                         offset += bytes_added;
2365                 }
2366                 spin_unlock(&cluster->lock);
2367                 if (!bytes) {
2368                         ret = 1;
2369                         goto out;
2370                 }
2371         }
2372
2373 no_cluster_bitmap:
2374         bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2375                                          1, 0);
2376         if (!bitmap_info) {
2377                 ASSERT(added == 0);
2378                 goto new_bitmap;
2379         }
2380
2381         bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes,
2382                                           trim_state);
2383         bytes -= bytes_added;
2384         offset += bytes_added;
2385         added = 0;
2386
2387         if (!bytes) {
2388                 ret = 1;
2389                 goto out;
2390         } else
2391                 goto again;
2392
2393 new_bitmap:
2394         if (info && info->bitmap) {
2395                 add_new_bitmap(ctl, info, offset);
2396                 added = 1;
2397                 info = NULL;
2398                 goto again;
2399         } else {
2400                 spin_unlock(&ctl->tree_lock);
2401
2402                 /* no pre-allocated info, allocate a new one */
2403                 if (!info) {
2404                         info = kmem_cache_zalloc(btrfs_free_space_cachep,
2405                                                  GFP_NOFS);
2406                         if (!info) {
2407                                 spin_lock(&ctl->tree_lock);
2408                                 ret = -ENOMEM;
2409                                 goto out;
2410                         }
2411                 }
2412
2413                 /* allocate the bitmap */
2414                 info->bitmap = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep,
2415                                                  GFP_NOFS);
2416                 info->trim_state = BTRFS_TRIM_STATE_TRIMMED;
2417                 spin_lock(&ctl->tree_lock);
2418                 if (!info->bitmap) {
2419                         ret = -ENOMEM;
2420                         goto out;
2421                 }
2422                 goto again;
2423         }
2424
2425 out:
2426         if (info) {
2427                 if (info->bitmap)
2428                         kmem_cache_free(btrfs_free_space_bitmap_cachep,
2429                                         info->bitmap);
2430                 kmem_cache_free(btrfs_free_space_cachep, info);
2431         }
2432
2433         return ret;
2434 }
2435
2436 /*
2437  * Free space merging rules:
2438  *  1) Merge trimmed areas together
2439  *  2) Let untrimmed areas coalesce with trimmed areas
2440  *  3) Always pull neighboring regions from bitmaps
2441  *
2442  * The above rules are for when we merge free space based on btrfs_trim_state.
2443  * Rules 2 and 3 are subtle because they are suboptimal, but are done for the
2444  * same reason: to promote larger extent regions which makes life easier for
2445  * find_free_extent().  Rule 2 enables coalescing based on the common path
2446  * being returning free space from btrfs_finish_extent_commit().  So when free
2447  * space is trimmed, it will prevent aggregating trimmed new region and
2448  * untrimmed regions in the rb_tree.  Rule 3 is purely to obtain larger extents
2449  * and provide find_free_extent() with the largest extents possible hoping for
2450  * the reuse path.
2451  */
2452 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
2453                           struct btrfs_free_space *info, bool update_stat)
2454 {
2455         struct btrfs_free_space *left_info = NULL;
2456         struct btrfs_free_space *right_info;
2457         bool merged = false;
2458         u64 offset = info->offset;
2459         u64 bytes = info->bytes;
2460         const bool is_trimmed = btrfs_free_space_trimmed(info);
2461         struct rb_node *right_prev = NULL;
2462
2463         /*
2464          * first we want to see if there is free space adjacent to the range we
2465          * are adding, if there is remove that struct and add a new one to
2466          * cover the entire range
2467          */
2468         right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
2469         if (right_info)
2470                 right_prev = rb_prev(&right_info->offset_index);
2471
2472         if (right_prev)
2473                 left_info = rb_entry(right_prev, struct btrfs_free_space, offset_index);
2474         else if (!right_info)
2475                 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
2476
2477         /* See try_merge_free_space() comment. */
2478         if (right_info && !right_info->bitmap &&
2479             (!is_trimmed || btrfs_free_space_trimmed(right_info))) {
2480                 unlink_free_space(ctl, right_info, update_stat);
2481                 info->bytes += right_info->bytes;
2482                 kmem_cache_free(btrfs_free_space_cachep, right_info);
2483                 merged = true;
2484         }
2485
2486         /* See try_merge_free_space() comment. */
2487         if (left_info && !left_info->bitmap &&
2488             left_info->offset + left_info->bytes == offset &&
2489             (!is_trimmed || btrfs_free_space_trimmed(left_info))) {
2490                 unlink_free_space(ctl, left_info, update_stat);
2491                 info->offset = left_info->offset;
2492                 info->bytes += left_info->bytes;
2493                 kmem_cache_free(btrfs_free_space_cachep, left_info);
2494                 merged = true;
2495         }
2496
2497         return merged;
2498 }
2499
2500 static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl,
2501                                      struct btrfs_free_space *info,
2502                                      bool update_stat)
2503 {
2504         struct btrfs_free_space *bitmap;
2505         unsigned long i;
2506         unsigned long j;
2507         const u64 end = info->offset + info->bytes;
2508         const u64 bitmap_offset = offset_to_bitmap(ctl, end);
2509         u64 bytes;
2510
2511         bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2512         if (!bitmap)
2513                 return false;
2514
2515         i = offset_to_bit(bitmap->offset, ctl->unit, end);
2516         j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i);
2517         if (j == i)
2518                 return false;
2519         bytes = (j - i) * ctl->unit;
2520         info->bytes += bytes;
2521
2522         /* See try_merge_free_space() comment. */
2523         if (!btrfs_free_space_trimmed(bitmap))
2524                 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2525
2526         bitmap_clear_bits(ctl, bitmap, end, bytes, update_stat);
2527
2528         if (!bitmap->bytes)
2529                 free_bitmap(ctl, bitmap);
2530
2531         return true;
2532 }
2533
2534 static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl,
2535                                        struct btrfs_free_space *info,
2536                                        bool update_stat)
2537 {
2538         struct btrfs_free_space *bitmap;
2539         u64 bitmap_offset;
2540         unsigned long i;
2541         unsigned long j;
2542         unsigned long prev_j;
2543         u64 bytes;
2544
2545         bitmap_offset = offset_to_bitmap(ctl, info->offset);
2546         /* If we're on a boundary, try the previous logical bitmap. */
2547         if (bitmap_offset == info->offset) {
2548                 if (info->offset == 0)
2549                         return false;
2550                 bitmap_offset = offset_to_bitmap(ctl, info->offset - 1);
2551         }
2552
2553         bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2554         if (!bitmap)
2555                 return false;
2556
2557         i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1;
2558         j = 0;
2559         prev_j = (unsigned long)-1;
2560         for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) {
2561                 if (j > i)
2562                         break;
2563                 prev_j = j;
2564         }
2565         if (prev_j == i)
2566                 return false;
2567
2568         if (prev_j == (unsigned long)-1)
2569                 bytes = (i + 1) * ctl->unit;
2570         else
2571                 bytes = (i - prev_j) * ctl->unit;
2572
2573         info->offset -= bytes;
2574         info->bytes += bytes;
2575
2576         /* See try_merge_free_space() comment. */
2577         if (!btrfs_free_space_trimmed(bitmap))
2578                 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2579
2580         bitmap_clear_bits(ctl, bitmap, info->offset, bytes, update_stat);
2581
2582         if (!bitmap->bytes)
2583                 free_bitmap(ctl, bitmap);
2584
2585         return true;
2586 }
2587
2588 /*
2589  * We prefer always to allocate from extent entries, both for clustered and
2590  * non-clustered allocation requests. So when attempting to add a new extent
2591  * entry, try to see if there's adjacent free space in bitmap entries, and if
2592  * there is, migrate that space from the bitmaps to the extent.
2593  * Like this we get better chances of satisfying space allocation requests
2594  * because we attempt to satisfy them based on a single cache entry, and never
2595  * on 2 or more entries - even if the entries represent a contiguous free space
2596  * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
2597  * ends).
2598  */
2599 static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl,
2600                               struct btrfs_free_space *info,
2601                               bool update_stat)
2602 {
2603         /*
2604          * Only work with disconnected entries, as we can change their offset,
2605          * and must be extent entries.
2606          */
2607         ASSERT(!info->bitmap);
2608         ASSERT(RB_EMPTY_NODE(&info->offset_index));
2609
2610         if (ctl->total_bitmaps > 0) {
2611                 bool stole_end;
2612                 bool stole_front = false;
2613
2614                 stole_end = steal_from_bitmap_to_end(ctl, info, update_stat);
2615                 if (ctl->total_bitmaps > 0)
2616                         stole_front = steal_from_bitmap_to_front(ctl, info,
2617                                                                  update_stat);
2618
2619                 if (stole_end || stole_front)
2620                         try_merge_free_space(ctl, info, update_stat);
2621         }
2622 }
2623
2624 int __btrfs_add_free_space(struct btrfs_block_group *block_group,
2625                            u64 offset, u64 bytes,
2626                            enum btrfs_trim_state trim_state)
2627 {
2628         struct btrfs_fs_info *fs_info = block_group->fs_info;
2629         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2630         struct btrfs_free_space *info;
2631         int ret = 0;
2632         u64 filter_bytes = bytes;
2633
2634         ASSERT(!btrfs_is_zoned(fs_info));
2635
2636         info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2637         if (!info)
2638                 return -ENOMEM;
2639
2640         info->offset = offset;
2641         info->bytes = bytes;
2642         info->trim_state = trim_state;
2643         RB_CLEAR_NODE(&info->offset_index);
2644         RB_CLEAR_NODE(&info->bytes_index);
2645
2646         spin_lock(&ctl->tree_lock);
2647
2648         if (try_merge_free_space(ctl, info, true))
2649                 goto link;
2650
2651         /*
2652          * There was no extent directly to the left or right of this new
2653          * extent then we know we're going to have to allocate a new extent, so
2654          * before we do that see if we need to drop this into a bitmap
2655          */
2656         ret = insert_into_bitmap(ctl, info);
2657         if (ret < 0) {
2658                 goto out;
2659         } else if (ret) {
2660                 ret = 0;
2661                 goto out;
2662         }
2663 link:
2664         /*
2665          * Only steal free space from adjacent bitmaps if we're sure we're not
2666          * going to add the new free space to existing bitmap entries - because
2667          * that would mean unnecessary work that would be reverted. Therefore
2668          * attempt to steal space from bitmaps if we're adding an extent entry.
2669          */
2670         steal_from_bitmap(ctl, info, true);
2671
2672         filter_bytes = max(filter_bytes, info->bytes);
2673
2674         ret = link_free_space(ctl, info);
2675         if (ret)
2676                 kmem_cache_free(btrfs_free_space_cachep, info);
2677 out:
2678         btrfs_discard_update_discardable(block_group);
2679         spin_unlock(&ctl->tree_lock);
2680
2681         if (ret) {
2682                 btrfs_crit(fs_info, "unable to add free space :%d", ret);
2683                 ASSERT(ret != -EEXIST);
2684         }
2685
2686         if (trim_state != BTRFS_TRIM_STATE_TRIMMED) {
2687                 btrfs_discard_check_filter(block_group, filter_bytes);
2688                 btrfs_discard_queue_work(&fs_info->discard_ctl, block_group);
2689         }
2690
2691         return ret;
2692 }
2693
2694 static int __btrfs_add_free_space_zoned(struct btrfs_block_group *block_group,
2695                                         u64 bytenr, u64 size, bool used)
2696 {
2697         struct btrfs_space_info *sinfo = block_group->space_info;
2698         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2699         u64 offset = bytenr - block_group->start;
2700         u64 to_free, to_unusable;
2701         int bg_reclaim_threshold = 0;
2702         bool initial = (size == block_group->length);
2703         u64 reclaimable_unusable;
2704
2705         WARN_ON(!initial && offset + size > block_group->zone_capacity);
2706
2707         if (!initial)
2708                 bg_reclaim_threshold = READ_ONCE(sinfo->bg_reclaim_threshold);
2709
2710         spin_lock(&ctl->tree_lock);
2711         if (!used)
2712                 to_free = size;
2713         else if (initial)
2714                 to_free = block_group->zone_capacity;
2715         else if (offset >= block_group->alloc_offset)
2716                 to_free = size;
2717         else if (offset + size <= block_group->alloc_offset)
2718                 to_free = 0;
2719         else
2720                 to_free = offset + size - block_group->alloc_offset;
2721         to_unusable = size - to_free;
2722
2723         ctl->free_space += to_free;
2724         /*
2725          * If the block group is read-only, we should account freed space into
2726          * bytes_readonly.
2727          */
2728         if (!block_group->ro)
2729                 block_group->zone_unusable += to_unusable;
2730         spin_unlock(&ctl->tree_lock);
2731         if (!used) {
2732                 spin_lock(&block_group->lock);
2733                 block_group->alloc_offset -= size;
2734                 spin_unlock(&block_group->lock);
2735         }
2736
2737         reclaimable_unusable = block_group->zone_unusable -
2738                                (block_group->length - block_group->zone_capacity);
2739         /* All the region is now unusable. Mark it as unused and reclaim */
2740         if (block_group->zone_unusable == block_group->length) {
2741                 btrfs_mark_bg_unused(block_group);
2742         } else if (bg_reclaim_threshold &&
2743                    reclaimable_unusable >=
2744                    mult_perc(block_group->zone_capacity, bg_reclaim_threshold)) {
2745                 btrfs_mark_bg_to_reclaim(block_group);
2746         }
2747
2748         return 0;
2749 }
2750
2751 int btrfs_add_free_space(struct btrfs_block_group *block_group,
2752                          u64 bytenr, u64 size)
2753 {
2754         enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2755
2756         if (btrfs_is_zoned(block_group->fs_info))
2757                 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2758                                                     true);
2759
2760         if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC))
2761                 trim_state = BTRFS_TRIM_STATE_TRIMMED;
2762
2763         return __btrfs_add_free_space(block_group, bytenr, size, trim_state);
2764 }
2765
2766 int btrfs_add_free_space_unused(struct btrfs_block_group *block_group,
2767                                 u64 bytenr, u64 size)
2768 {
2769         if (btrfs_is_zoned(block_group->fs_info))
2770                 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2771                                                     false);
2772
2773         return btrfs_add_free_space(block_group, bytenr, size);
2774 }
2775
2776 /*
2777  * This is a subtle distinction because when adding free space back in general,
2778  * we want it to be added as untrimmed for async. But in the case where we add
2779  * it on loading of a block group, we want to consider it trimmed.
2780  */
2781 int btrfs_add_free_space_async_trimmed(struct btrfs_block_group *block_group,
2782                                        u64 bytenr, u64 size)
2783 {
2784         enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2785
2786         if (btrfs_is_zoned(block_group->fs_info))
2787                 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2788                                                     true);
2789
2790         if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC) ||
2791             btrfs_test_opt(block_group->fs_info, DISCARD_ASYNC))
2792                 trim_state = BTRFS_TRIM_STATE_TRIMMED;
2793
2794         return __btrfs_add_free_space(block_group, bytenr, size, trim_state);
2795 }
2796
2797 int btrfs_remove_free_space(struct btrfs_block_group *block_group,
2798                             u64 offset, u64 bytes)
2799 {
2800         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2801         struct btrfs_free_space *info;
2802         int ret;
2803         bool re_search = false;
2804
2805         if (btrfs_is_zoned(block_group->fs_info)) {
2806                 /*
2807                  * This can happen with conventional zones when replaying log.
2808                  * Since the allocation info of tree-log nodes are not recorded
2809                  * to the extent-tree, calculate_alloc_pointer() failed to
2810                  * advance the allocation pointer after last allocated tree log
2811                  * node blocks.
2812                  *
2813                  * This function is called from
2814                  * btrfs_pin_extent_for_log_replay() when replaying the log.
2815                  * Advance the pointer not to overwrite the tree-log nodes.
2816                  */
2817                 if (block_group->start + block_group->alloc_offset <
2818                     offset + bytes) {
2819                         block_group->alloc_offset =
2820                                 offset + bytes - block_group->start;
2821                 }
2822                 return 0;
2823         }
2824
2825         spin_lock(&ctl->tree_lock);
2826
2827 again:
2828         ret = 0;
2829         if (!bytes)
2830                 goto out_lock;
2831
2832         info = tree_search_offset(ctl, offset, 0, 0);
2833         if (!info) {
2834                 /*
2835                  * oops didn't find an extent that matched the space we wanted
2836                  * to remove, look for a bitmap instead
2837                  */
2838                 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2839                                           1, 0);
2840                 if (!info) {
2841                         /*
2842                          * If we found a partial bit of our free space in a
2843                          * bitmap but then couldn't find the other part this may
2844                          * be a problem, so WARN about it.
2845                          */
2846                         WARN_ON(re_search);
2847                         goto out_lock;
2848                 }
2849         }
2850
2851         re_search = false;
2852         if (!info->bitmap) {
2853                 unlink_free_space(ctl, info, true);
2854                 if (offset == info->offset) {
2855                         u64 to_free = min(bytes, info->bytes);
2856
2857                         info->bytes -= to_free;
2858                         info->offset += to_free;
2859                         if (info->bytes) {
2860                                 ret = link_free_space(ctl, info);
2861                                 WARN_ON(ret);
2862                         } else {
2863                                 kmem_cache_free(btrfs_free_space_cachep, info);
2864                         }
2865
2866                         offset += to_free;
2867                         bytes -= to_free;
2868                         goto again;
2869                 } else {
2870                         u64 old_end = info->bytes + info->offset;
2871
2872                         info->bytes = offset - info->offset;
2873                         ret = link_free_space(ctl, info);
2874                         WARN_ON(ret);
2875                         if (ret)
2876                                 goto out_lock;
2877
2878                         /* Not enough bytes in this entry to satisfy us */
2879                         if (old_end < offset + bytes) {
2880                                 bytes -= old_end - offset;
2881                                 offset = old_end;
2882                                 goto again;
2883                         } else if (old_end == offset + bytes) {
2884                                 /* all done */
2885                                 goto out_lock;
2886                         }
2887                         spin_unlock(&ctl->tree_lock);
2888
2889                         ret = __btrfs_add_free_space(block_group,
2890                                                      offset + bytes,
2891                                                      old_end - (offset + bytes),
2892                                                      info->trim_state);
2893                         WARN_ON(ret);
2894                         goto out;
2895                 }
2896         }
2897
2898         ret = remove_from_bitmap(ctl, info, &offset, &bytes);
2899         if (ret == -EAGAIN) {
2900                 re_search = true;
2901                 goto again;
2902         }
2903 out_lock:
2904         btrfs_discard_update_discardable(block_group);
2905         spin_unlock(&ctl->tree_lock);
2906 out:
2907         return ret;
2908 }
2909
2910 void btrfs_dump_free_space(struct btrfs_block_group *block_group,
2911                            u64 bytes)
2912 {
2913         struct btrfs_fs_info *fs_info = block_group->fs_info;
2914         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2915         struct btrfs_free_space *info;
2916         struct rb_node *n;
2917         int count = 0;
2918
2919         /*
2920          * Zoned btrfs does not use free space tree and cluster. Just print
2921          * out the free space after the allocation offset.
2922          */
2923         if (btrfs_is_zoned(fs_info)) {
2924                 btrfs_info(fs_info, "free space %llu active %d",
2925                            block_group->zone_capacity - block_group->alloc_offset,
2926                            test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE,
2927                                     &block_group->runtime_flags));
2928                 return;
2929         }
2930
2931         spin_lock(&ctl->tree_lock);
2932         for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2933                 info = rb_entry(n, struct btrfs_free_space, offset_index);
2934                 if (info->bytes >= bytes && !block_group->ro)
2935                         count++;
2936                 btrfs_crit(fs_info, "entry offset %llu, bytes %llu, bitmap %s",
2937                            info->offset, info->bytes,
2938                        (info->bitmap) ? "yes" : "no");
2939         }
2940         spin_unlock(&ctl->tree_lock);
2941         btrfs_info(fs_info, "block group has cluster?: %s",
2942                list_empty(&block_group->cluster_list) ? "no" : "yes");
2943         btrfs_info(fs_info,
2944                    "%d free space entries at or bigger than %llu bytes",
2945                    count, bytes);
2946 }
2947
2948 void btrfs_init_free_space_ctl(struct btrfs_block_group *block_group,
2949                                struct btrfs_free_space_ctl *ctl)
2950 {
2951         struct btrfs_fs_info *fs_info = block_group->fs_info;
2952
2953         spin_lock_init(&ctl->tree_lock);
2954         ctl->unit = fs_info->sectorsize;
2955         ctl->start = block_group->start;
2956         ctl->block_group = block_group;
2957         ctl->op = &free_space_op;
2958         ctl->free_space_bytes = RB_ROOT_CACHED;
2959         INIT_LIST_HEAD(&ctl->trimming_ranges);
2960         mutex_init(&ctl->cache_writeout_mutex);
2961
2962         /*
2963          * we only want to have 32k of ram per block group for keeping
2964          * track of free space, and if we pass 1/2 of that we want to
2965          * start converting things over to using bitmaps
2966          */
2967         ctl->extents_thresh = (SZ_32K / 2) / sizeof(struct btrfs_free_space);
2968 }
2969
2970 /*
2971  * for a given cluster, put all of its extents back into the free
2972  * space cache.  If the block group passed doesn't match the block group
2973  * pointed to by the cluster, someone else raced in and freed the
2974  * cluster already.  In that case, we just return without changing anything
2975  */
2976 static void __btrfs_return_cluster_to_free_space(
2977                              struct btrfs_block_group *block_group,
2978                              struct btrfs_free_cluster *cluster)
2979 {
2980         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2981         struct rb_node *node;
2982
2983         lockdep_assert_held(&ctl->tree_lock);
2984
2985         spin_lock(&cluster->lock);
2986         if (cluster->block_group != block_group) {
2987                 spin_unlock(&cluster->lock);
2988                 return;
2989         }
2990
2991         cluster->block_group = NULL;
2992         cluster->window_start = 0;
2993         list_del_init(&cluster->block_group_list);
2994
2995         node = rb_first(&cluster->root);
2996         while (node) {
2997                 struct btrfs_free_space *entry;
2998
2999                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3000                 node = rb_next(&entry->offset_index);
3001                 rb_erase(&entry->offset_index, &cluster->root);
3002                 RB_CLEAR_NODE(&entry->offset_index);
3003
3004                 if (!entry->bitmap) {
3005                         /* Merging treats extents as if they were new */
3006                         if (!btrfs_free_space_trimmed(entry)) {
3007                                 ctl->discardable_extents[BTRFS_STAT_CURR]--;
3008                                 ctl->discardable_bytes[BTRFS_STAT_CURR] -=
3009                                         entry->bytes;
3010                         }
3011
3012                         try_merge_free_space(ctl, entry, false);
3013                         steal_from_bitmap(ctl, entry, false);
3014
3015                         /* As we insert directly, update these statistics */
3016                         if (!btrfs_free_space_trimmed(entry)) {
3017                                 ctl->discardable_extents[BTRFS_STAT_CURR]++;
3018                                 ctl->discardable_bytes[BTRFS_STAT_CURR] +=
3019                                         entry->bytes;
3020                         }
3021                 }
3022                 tree_insert_offset(ctl, NULL, entry);
3023                 rb_add_cached(&entry->bytes_index, &ctl->free_space_bytes,
3024                               entry_less);
3025         }
3026         cluster->root = RB_ROOT;
3027         spin_unlock(&cluster->lock);
3028         btrfs_put_block_group(block_group);
3029 }
3030
3031 void btrfs_remove_free_space_cache(struct btrfs_block_group *block_group)
3032 {
3033         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3034         struct btrfs_free_cluster *cluster;
3035         struct list_head *head;
3036
3037         spin_lock(&ctl->tree_lock);
3038         while ((head = block_group->cluster_list.next) !=
3039                &block_group->cluster_list) {
3040                 cluster = list_entry(head, struct btrfs_free_cluster,
3041                                      block_group_list);
3042
3043                 WARN_ON(cluster->block_group != block_group);
3044                 __btrfs_return_cluster_to_free_space(block_group, cluster);
3045
3046                 cond_resched_lock(&ctl->tree_lock);
3047         }
3048         __btrfs_remove_free_space_cache(ctl);
3049         btrfs_discard_update_discardable(block_group);
3050         spin_unlock(&ctl->tree_lock);
3051
3052 }
3053
3054 /*
3055  * Walk @block_group's free space rb_tree to determine if everything is trimmed.
3056  */
3057 bool btrfs_is_free_space_trimmed(struct btrfs_block_group *block_group)
3058 {
3059         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3060         struct btrfs_free_space *info;
3061         struct rb_node *node;
3062         bool ret = true;
3063
3064         spin_lock(&ctl->tree_lock);
3065         node = rb_first(&ctl->free_space_offset);
3066
3067         while (node) {
3068                 info = rb_entry(node, struct btrfs_free_space, offset_index);
3069
3070                 if (!btrfs_free_space_trimmed(info)) {
3071                         ret = false;
3072                         break;
3073                 }
3074
3075                 node = rb_next(node);
3076         }
3077
3078         spin_unlock(&ctl->tree_lock);
3079         return ret;
3080 }
3081
3082 u64 btrfs_find_space_for_alloc(struct btrfs_block_group *block_group,
3083                                u64 offset, u64 bytes, u64 empty_size,
3084                                u64 *max_extent_size)
3085 {
3086         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3087         struct btrfs_discard_ctl *discard_ctl =
3088                                         &block_group->fs_info->discard_ctl;
3089         struct btrfs_free_space *entry = NULL;
3090         u64 bytes_search = bytes + empty_size;
3091         u64 ret = 0;
3092         u64 align_gap = 0;
3093         u64 align_gap_len = 0;
3094         enum btrfs_trim_state align_gap_trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3095         bool use_bytes_index = (offset == block_group->start);
3096
3097         ASSERT(!btrfs_is_zoned(block_group->fs_info));
3098
3099         spin_lock(&ctl->tree_lock);
3100         entry = find_free_space(ctl, &offset, &bytes_search,
3101                                 block_group->full_stripe_len, max_extent_size,
3102                                 use_bytes_index);
3103         if (!entry)
3104                 goto out;
3105
3106         ret = offset;
3107         if (entry->bitmap) {
3108                 bitmap_clear_bits(ctl, entry, offset, bytes, true);
3109
3110                 if (!btrfs_free_space_trimmed(entry))
3111                         atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3112
3113                 if (!entry->bytes)
3114                         free_bitmap(ctl, entry);
3115         } else {
3116                 unlink_free_space(ctl, entry, true);
3117                 align_gap_len = offset - entry->offset;
3118                 align_gap = entry->offset;
3119                 align_gap_trim_state = entry->trim_state;
3120
3121                 if (!btrfs_free_space_trimmed(entry))
3122                         atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3123
3124                 entry->offset = offset + bytes;
3125                 WARN_ON(entry->bytes < bytes + align_gap_len);
3126
3127                 entry->bytes -= bytes + align_gap_len;
3128                 if (!entry->bytes)
3129                         kmem_cache_free(btrfs_free_space_cachep, entry);
3130                 else
3131                         link_free_space(ctl, entry);
3132         }
3133 out:
3134         btrfs_discard_update_discardable(block_group);
3135         spin_unlock(&ctl->tree_lock);
3136
3137         if (align_gap_len)
3138                 __btrfs_add_free_space(block_group, align_gap, align_gap_len,
3139                                        align_gap_trim_state);
3140         return ret;
3141 }
3142
3143 /*
3144  * given a cluster, put all of its extents back into the free space
3145  * cache.  If a block group is passed, this function will only free
3146  * a cluster that belongs to the passed block group.
3147  *
3148  * Otherwise, it'll get a reference on the block group pointed to by the
3149  * cluster and remove the cluster from it.
3150  */
3151 void btrfs_return_cluster_to_free_space(
3152                                struct btrfs_block_group *block_group,
3153                                struct btrfs_free_cluster *cluster)
3154 {
3155         struct btrfs_free_space_ctl *ctl;
3156
3157         /* first, get a safe pointer to the block group */
3158         spin_lock(&cluster->lock);
3159         if (!block_group) {
3160                 block_group = cluster->block_group;
3161                 if (!block_group) {
3162                         spin_unlock(&cluster->lock);
3163                         return;
3164                 }
3165         } else if (cluster->block_group != block_group) {
3166                 /* someone else has already freed it don't redo their work */
3167                 spin_unlock(&cluster->lock);
3168                 return;
3169         }
3170         btrfs_get_block_group(block_group);
3171         spin_unlock(&cluster->lock);
3172
3173         ctl = block_group->free_space_ctl;
3174
3175         /* now return any extents the cluster had on it */
3176         spin_lock(&ctl->tree_lock);
3177         __btrfs_return_cluster_to_free_space(block_group, cluster);
3178         spin_unlock(&ctl->tree_lock);
3179
3180         btrfs_discard_queue_work(&block_group->fs_info->discard_ctl, block_group);
3181
3182         /* finally drop our ref */
3183         btrfs_put_block_group(block_group);
3184 }
3185
3186 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group *block_group,
3187                                    struct btrfs_free_cluster *cluster,
3188                                    struct btrfs_free_space *entry,
3189                                    u64 bytes, u64 min_start,
3190                                    u64 *max_extent_size)
3191 {
3192         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3193         int err;
3194         u64 search_start = cluster->window_start;
3195         u64 search_bytes = bytes;
3196         u64 ret = 0;
3197
3198         search_start = min_start;
3199         search_bytes = bytes;
3200
3201         err = search_bitmap(ctl, entry, &search_start, &search_bytes, true);
3202         if (err) {
3203                 *max_extent_size = max(get_max_extent_size(entry),
3204                                        *max_extent_size);
3205                 return 0;
3206         }
3207
3208         ret = search_start;
3209         bitmap_clear_bits(ctl, entry, ret, bytes, false);
3210
3211         return ret;
3212 }
3213
3214 /*
3215  * given a cluster, try to allocate 'bytes' from it, returns 0
3216  * if it couldn't find anything suitably large, or a logical disk offset
3217  * if things worked out
3218  */
3219 u64 btrfs_alloc_from_cluster(struct btrfs_block_group *block_group,
3220                              struct btrfs_free_cluster *cluster, u64 bytes,
3221                              u64 min_start, u64 *max_extent_size)
3222 {
3223         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3224         struct btrfs_discard_ctl *discard_ctl =
3225                                         &block_group->fs_info->discard_ctl;
3226         struct btrfs_free_space *entry = NULL;
3227         struct rb_node *node;
3228         u64 ret = 0;
3229
3230         ASSERT(!btrfs_is_zoned(block_group->fs_info));
3231
3232         spin_lock(&cluster->lock);
3233         if (bytes > cluster->max_size)
3234                 goto out;
3235
3236         if (cluster->block_group != block_group)
3237                 goto out;
3238
3239         node = rb_first(&cluster->root);
3240         if (!node)
3241                 goto out;
3242
3243         entry = rb_entry(node, struct btrfs_free_space, offset_index);
3244         while (1) {
3245                 if (entry->bytes < bytes)
3246                         *max_extent_size = max(get_max_extent_size(entry),
3247                                                *max_extent_size);
3248
3249                 if (entry->bytes < bytes ||
3250                     (!entry->bitmap && entry->offset < min_start)) {
3251                         node = rb_next(&entry->offset_index);
3252                         if (!node)
3253                                 break;
3254                         entry = rb_entry(node, struct btrfs_free_space,
3255                                          offset_index);
3256                         continue;
3257                 }
3258
3259                 if (entry->bitmap) {
3260                         ret = btrfs_alloc_from_bitmap(block_group,
3261                                                       cluster, entry, bytes,
3262                                                       cluster->window_start,
3263                                                       max_extent_size);
3264                         if (ret == 0) {
3265                                 node = rb_next(&entry->offset_index);
3266                                 if (!node)
3267                                         break;
3268                                 entry = rb_entry(node, struct btrfs_free_space,
3269                                                  offset_index);
3270                                 continue;
3271                         }
3272                         cluster->window_start += bytes;
3273                 } else {
3274                         ret = entry->offset;
3275
3276                         entry->offset += bytes;
3277                         entry->bytes -= bytes;
3278                 }
3279
3280                 break;
3281         }
3282 out:
3283         spin_unlock(&cluster->lock);
3284
3285         if (!ret)
3286                 return 0;
3287
3288         spin_lock(&ctl->tree_lock);
3289
3290         if (!btrfs_free_space_trimmed(entry))
3291                 atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3292
3293         ctl->free_space -= bytes;
3294         if (!entry->bitmap && !btrfs_free_space_trimmed(entry))
3295                 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
3296
3297         spin_lock(&cluster->lock);
3298         if (entry->bytes == 0) {
3299                 rb_erase(&entry->offset_index, &cluster->root);
3300                 ctl->free_extents--;
3301                 if (entry->bitmap) {
3302                         kmem_cache_free(btrfs_free_space_bitmap_cachep,
3303                                         entry->bitmap);
3304                         ctl->total_bitmaps--;
3305                         recalculate_thresholds(ctl);
3306                 } else if (!btrfs_free_space_trimmed(entry)) {
3307                         ctl->discardable_extents[BTRFS_STAT_CURR]--;
3308                 }
3309                 kmem_cache_free(btrfs_free_space_cachep, entry);
3310         }
3311
3312         spin_unlock(&cluster->lock);
3313         spin_unlock(&ctl->tree_lock);
3314
3315         return ret;
3316 }
3317
3318 static int btrfs_bitmap_cluster(struct btrfs_block_group *block_group,
3319                                 struct btrfs_free_space *entry,
3320                                 struct btrfs_free_cluster *cluster,
3321                                 u64 offset, u64 bytes,
3322                                 u64 cont1_bytes, u64 min_bytes)
3323 {
3324         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3325         unsigned long next_zero;
3326         unsigned long i;
3327         unsigned long want_bits;
3328         unsigned long min_bits;
3329         unsigned long found_bits;
3330         unsigned long max_bits = 0;
3331         unsigned long start = 0;
3332         unsigned long total_found = 0;
3333         int ret;
3334
3335         lockdep_assert_held(&ctl->tree_lock);
3336
3337         i = offset_to_bit(entry->offset, ctl->unit,
3338                           max_t(u64, offset, entry->offset));
3339         want_bits = bytes_to_bits(bytes, ctl->unit);
3340         min_bits = bytes_to_bits(min_bytes, ctl->unit);
3341
3342         /*
3343          * Don't bother looking for a cluster in this bitmap if it's heavily
3344          * fragmented.
3345          */
3346         if (entry->max_extent_size &&
3347             entry->max_extent_size < cont1_bytes)
3348                 return -ENOSPC;
3349 again:
3350         found_bits = 0;
3351         for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
3352                 next_zero = find_next_zero_bit(entry->bitmap,
3353                                                BITS_PER_BITMAP, i);
3354                 if (next_zero - i >= min_bits) {
3355                         found_bits = next_zero - i;
3356                         if (found_bits > max_bits)
3357                                 max_bits = found_bits;
3358                         break;
3359                 }
3360                 if (next_zero - i > max_bits)
3361                         max_bits = next_zero - i;
3362                 i = next_zero;
3363         }
3364
3365         if (!found_bits) {
3366                 entry->max_extent_size = (u64)max_bits * ctl->unit;
3367                 return -ENOSPC;
3368         }
3369
3370         if (!total_found) {
3371                 start = i;
3372                 cluster->max_size = 0;
3373         }
3374
3375         total_found += found_bits;
3376
3377         if (cluster->max_size < found_bits * ctl->unit)
3378                 cluster->max_size = found_bits * ctl->unit;
3379
3380         if (total_found < want_bits || cluster->max_size < cont1_bytes) {
3381                 i = next_zero + 1;
3382                 goto again;
3383         }
3384
3385         cluster->window_start = start * ctl->unit + entry->offset;
3386         rb_erase(&entry->offset_index, &ctl->free_space_offset);
3387         rb_erase_cached(&entry->bytes_index, &ctl->free_space_bytes);
3388
3389         /*
3390          * We need to know if we're currently on the normal space index when we
3391          * manipulate the bitmap so that we know we need to remove and re-insert
3392          * it into the space_index tree.  Clear the bytes_index node here so the
3393          * bitmap manipulation helpers know not to mess with the space_index
3394          * until this bitmap entry is added back into the normal cache.
3395          */
3396         RB_CLEAR_NODE(&entry->bytes_index);
3397
3398         ret = tree_insert_offset(ctl, cluster, entry);
3399         ASSERT(!ret); /* -EEXIST; Logic error */
3400
3401         trace_btrfs_setup_cluster(block_group, cluster,
3402                                   total_found * ctl->unit, 1);
3403         return 0;
3404 }
3405
3406 /*
3407  * This searches the block group for just extents to fill the cluster with.
3408  * Try to find a cluster with at least bytes total bytes, at least one
3409  * extent of cont1_bytes, and other clusters of at least min_bytes.
3410  */
3411 static noinline int
3412 setup_cluster_no_bitmap(struct btrfs_block_group *block_group,
3413                         struct btrfs_free_cluster *cluster,
3414                         struct list_head *bitmaps, u64 offset, u64 bytes,
3415                         u64 cont1_bytes, u64 min_bytes)
3416 {
3417         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3418         struct btrfs_free_space *first = NULL;
3419         struct btrfs_free_space *entry = NULL;
3420         struct btrfs_free_space *last;
3421         struct rb_node *node;
3422         u64 window_free;
3423         u64 max_extent;
3424         u64 total_size = 0;
3425
3426         lockdep_assert_held(&ctl->tree_lock);
3427
3428         entry = tree_search_offset(ctl, offset, 0, 1);
3429         if (!entry)
3430                 return -ENOSPC;
3431
3432         /*
3433          * We don't want bitmaps, so just move along until we find a normal
3434          * extent entry.
3435          */
3436         while (entry->bitmap || entry->bytes < min_bytes) {
3437                 if (entry->bitmap && list_empty(&entry->list))
3438                         list_add_tail(&entry->list, bitmaps);
3439                 node = rb_next(&entry->offset_index);
3440                 if (!node)
3441                         return -ENOSPC;
3442                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3443         }
3444
3445         window_free = entry->bytes;
3446         max_extent = entry->bytes;
3447         first = entry;
3448         last = entry;
3449
3450         for (node = rb_next(&entry->offset_index); node;
3451              node = rb_next(&entry->offset_index)) {
3452                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3453
3454                 if (entry->bitmap) {
3455                         if (list_empty(&entry->list))
3456                                 list_add_tail(&entry->list, bitmaps);
3457                         continue;
3458                 }
3459
3460                 if (entry->bytes < min_bytes)
3461                         continue;
3462
3463                 last = entry;
3464                 window_free += entry->bytes;
3465                 if (entry->bytes > max_extent)
3466                         max_extent = entry->bytes;
3467         }
3468
3469         if (window_free < bytes || max_extent < cont1_bytes)
3470                 return -ENOSPC;
3471
3472         cluster->window_start = first->offset;
3473
3474         node = &first->offset_index;
3475
3476         /*
3477          * now we've found our entries, pull them out of the free space
3478          * cache and put them into the cluster rbtree
3479          */
3480         do {
3481                 int ret;
3482
3483                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3484                 node = rb_next(&entry->offset_index);
3485                 if (entry->bitmap || entry->bytes < min_bytes)
3486                         continue;
3487
3488                 rb_erase(&entry->offset_index, &ctl->free_space_offset);
3489                 rb_erase_cached(&entry->bytes_index, &ctl->free_space_bytes);
3490                 ret = tree_insert_offset(ctl, cluster, entry);
3491                 total_size += entry->bytes;
3492                 ASSERT(!ret); /* -EEXIST; Logic error */
3493         } while (node && entry != last);
3494
3495         cluster->max_size = max_extent;
3496         trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
3497         return 0;
3498 }
3499
3500 /*
3501  * This specifically looks for bitmaps that may work in the cluster, we assume
3502  * that we have already failed to find extents that will work.
3503  */
3504 static noinline int
3505 setup_cluster_bitmap(struct btrfs_block_group *block_group,
3506                      struct btrfs_free_cluster *cluster,
3507                      struct list_head *bitmaps, u64 offset, u64 bytes,
3508                      u64 cont1_bytes, u64 min_bytes)
3509 {
3510         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3511         struct btrfs_free_space *entry = NULL;
3512         int ret = -ENOSPC;
3513         u64 bitmap_offset = offset_to_bitmap(ctl, offset);
3514
3515         if (ctl->total_bitmaps == 0)
3516                 return -ENOSPC;
3517
3518         /*
3519          * The bitmap that covers offset won't be in the list unless offset
3520          * is just its start offset.
3521          */
3522         if (!list_empty(bitmaps))
3523                 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
3524
3525         if (!entry || entry->offset != bitmap_offset) {
3526                 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
3527                 if (entry && list_empty(&entry->list))
3528                         list_add(&entry->list, bitmaps);
3529         }
3530
3531         list_for_each_entry(entry, bitmaps, list) {
3532                 if (entry->bytes < bytes)
3533                         continue;
3534                 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
3535                                            bytes, cont1_bytes, min_bytes);
3536                 if (!ret)
3537                         return 0;
3538         }
3539
3540         /*
3541          * The bitmaps list has all the bitmaps that record free space
3542          * starting after offset, so no more search is required.
3543          */
3544         return -ENOSPC;
3545 }
3546
3547 /*
3548  * here we try to find a cluster of blocks in a block group.  The goal
3549  * is to find at least bytes+empty_size.
3550  * We might not find them all in one contiguous area.
3551  *
3552  * returns zero and sets up cluster if things worked out, otherwise
3553  * it returns -enospc
3554  */
3555 int btrfs_find_space_cluster(struct btrfs_block_group *block_group,
3556                              struct btrfs_free_cluster *cluster,
3557                              u64 offset, u64 bytes, u64 empty_size)
3558 {
3559         struct btrfs_fs_info *fs_info = block_group->fs_info;
3560         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3561         struct btrfs_free_space *entry, *tmp;
3562         LIST_HEAD(bitmaps);
3563         u64 min_bytes;
3564         u64 cont1_bytes;
3565         int ret;
3566
3567         /*
3568          * Choose the minimum extent size we'll require for this
3569          * cluster.  For SSD_SPREAD, don't allow any fragmentation.
3570          * For metadata, allow allocates with smaller extents.  For
3571          * data, keep it dense.
3572          */
3573         if (btrfs_test_opt(fs_info, SSD_SPREAD)) {
3574                 cont1_bytes = bytes + empty_size;
3575                 min_bytes = cont1_bytes;
3576         } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
3577                 cont1_bytes = bytes;
3578                 min_bytes = fs_info->sectorsize;
3579         } else {
3580                 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
3581                 min_bytes = fs_info->sectorsize;
3582         }
3583
3584         spin_lock(&ctl->tree_lock);
3585
3586         /*
3587          * If we know we don't have enough space to make a cluster don't even
3588          * bother doing all the work to try and find one.
3589          */
3590         if (ctl->free_space < bytes) {
3591                 spin_unlock(&ctl->tree_lock);
3592                 return -ENOSPC;
3593         }
3594
3595         spin_lock(&cluster->lock);
3596
3597         /* someone already found a cluster, hooray */
3598         if (cluster->block_group) {
3599                 ret = 0;
3600                 goto out;
3601         }
3602
3603         trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
3604                                  min_bytes);
3605
3606         ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
3607                                       bytes + empty_size,
3608                                       cont1_bytes, min_bytes);
3609         if (ret)
3610                 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
3611                                            offset, bytes + empty_size,
3612                                            cont1_bytes, min_bytes);
3613
3614         /* Clear our temporary list */
3615         list_for_each_entry_safe(entry, tmp, &bitmaps, list)
3616                 list_del_init(&entry->list);
3617
3618         if (!ret) {
3619                 btrfs_get_block_group(block_group);
3620                 list_add_tail(&cluster->block_group_list,
3621                               &block_group->cluster_list);
3622                 cluster->block_group = block_group;
3623         } else {
3624                 trace_btrfs_failed_cluster_setup(block_group);
3625         }
3626 out:
3627         spin_unlock(&cluster->lock);
3628         spin_unlock(&ctl->tree_lock);
3629
3630         return ret;
3631 }
3632
3633 /*
3634  * simple code to zero out a cluster
3635  */
3636 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
3637 {
3638         spin_lock_init(&cluster->lock);
3639         spin_lock_init(&cluster->refill_lock);
3640         cluster->root = RB_ROOT;
3641         cluster->max_size = 0;
3642         cluster->fragmented = false;
3643         INIT_LIST_HEAD(&cluster->block_group_list);
3644         cluster->block_group = NULL;
3645 }
3646
3647 static int do_trimming(struct btrfs_block_group *block_group,
3648                        u64 *total_trimmed, u64 start, u64 bytes,
3649                        u64 reserved_start, u64 reserved_bytes,
3650                        enum btrfs_trim_state reserved_trim_state,
3651                        struct btrfs_trim_range *trim_entry)
3652 {
3653         struct btrfs_space_info *space_info = block_group->space_info;
3654         struct btrfs_fs_info *fs_info = block_group->fs_info;
3655         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3656         int ret;
3657         int update = 0;
3658         const u64 end = start + bytes;
3659         const u64 reserved_end = reserved_start + reserved_bytes;
3660         enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3661         u64 trimmed = 0;
3662
3663         spin_lock(&space_info->lock);
3664         spin_lock(&block_group->lock);
3665         if (!block_group->ro) {
3666                 block_group->reserved += reserved_bytes;
3667                 space_info->bytes_reserved += reserved_bytes;
3668                 update = 1;
3669         }
3670         spin_unlock(&block_group->lock);
3671         spin_unlock(&space_info->lock);
3672
3673         ret = btrfs_discard_extent(fs_info, start, bytes, &trimmed);
3674         if (!ret) {
3675                 *total_trimmed += trimmed;
3676                 trim_state = BTRFS_TRIM_STATE_TRIMMED;
3677         }
3678
3679         mutex_lock(&ctl->cache_writeout_mutex);
3680         if (reserved_start < start)
3681                 __btrfs_add_free_space(block_group, reserved_start,
3682                                        start - reserved_start,
3683                                        reserved_trim_state);
3684         if (end < reserved_end)
3685                 __btrfs_add_free_space(block_group, end, reserved_end - end,
3686                                        reserved_trim_state);
3687         __btrfs_add_free_space(block_group, start, bytes, trim_state);
3688         list_del(&trim_entry->list);
3689         mutex_unlock(&ctl->cache_writeout_mutex);
3690
3691         if (update) {
3692                 spin_lock(&space_info->lock);
3693                 spin_lock(&block_group->lock);
3694                 if (block_group->ro)
3695                         space_info->bytes_readonly += reserved_bytes;
3696                 block_group->reserved -= reserved_bytes;
3697                 space_info->bytes_reserved -= reserved_bytes;
3698                 spin_unlock(&block_group->lock);
3699                 spin_unlock(&space_info->lock);
3700         }
3701
3702         return ret;
3703 }
3704
3705 /*
3706  * If @async is set, then we will trim 1 region and return.
3707  */
3708 static int trim_no_bitmap(struct btrfs_block_group *block_group,
3709                           u64 *total_trimmed, u64 start, u64 end, u64 minlen,
3710                           bool async)
3711 {
3712         struct btrfs_discard_ctl *discard_ctl =
3713                                         &block_group->fs_info->discard_ctl;
3714         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3715         struct btrfs_free_space *entry;
3716         struct rb_node *node;
3717         int ret = 0;
3718         u64 extent_start;
3719         u64 extent_bytes;
3720         enum btrfs_trim_state extent_trim_state;
3721         u64 bytes;
3722         const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3723
3724         while (start < end) {
3725                 struct btrfs_trim_range trim_entry;
3726
3727                 mutex_lock(&ctl->cache_writeout_mutex);
3728                 spin_lock(&ctl->tree_lock);
3729
3730                 if (ctl->free_space < minlen)
3731                         goto out_unlock;
3732
3733                 entry = tree_search_offset(ctl, start, 0, 1);
3734                 if (!entry)
3735                         goto out_unlock;
3736
3737                 /* Skip bitmaps and if async, already trimmed entries */
3738                 while (entry->bitmap ||
3739                        (async && btrfs_free_space_trimmed(entry))) {
3740                         node = rb_next(&entry->offset_index);
3741                         if (!node)
3742                                 goto out_unlock;
3743                         entry = rb_entry(node, struct btrfs_free_space,
3744                                          offset_index);
3745                 }
3746
3747                 if (entry->offset >= end)
3748                         goto out_unlock;
3749
3750                 extent_start = entry->offset;
3751                 extent_bytes = entry->bytes;
3752                 extent_trim_state = entry->trim_state;
3753                 if (async) {
3754                         start = entry->offset;
3755                         bytes = entry->bytes;
3756                         if (bytes < minlen) {
3757                                 spin_unlock(&ctl->tree_lock);
3758                                 mutex_unlock(&ctl->cache_writeout_mutex);
3759                                 goto next;
3760                         }
3761                         unlink_free_space(ctl, entry, true);
3762                         /*
3763                          * Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
3764                          * If X < BTRFS_ASYNC_DISCARD_MIN_FILTER, we won't trim
3765                          * X when we come back around.  So trim it now.
3766                          */
3767                         if (max_discard_size &&
3768                             bytes >= (max_discard_size +
3769                                       BTRFS_ASYNC_DISCARD_MIN_FILTER)) {
3770                                 bytes = max_discard_size;
3771                                 extent_bytes = max_discard_size;
3772                                 entry->offset += max_discard_size;
3773                                 entry->bytes -= max_discard_size;
3774                                 link_free_space(ctl, entry);
3775                         } else {
3776                                 kmem_cache_free(btrfs_free_space_cachep, entry);
3777                         }
3778                 } else {
3779                         start = max(start, extent_start);
3780                         bytes = min(extent_start + extent_bytes, end) - start;
3781                         if (bytes < minlen) {
3782                                 spin_unlock(&ctl->tree_lock);
3783                                 mutex_unlock(&ctl->cache_writeout_mutex);
3784                                 goto next;
3785                         }
3786
3787                         unlink_free_space(ctl, entry, true);
3788                         kmem_cache_free(btrfs_free_space_cachep, entry);
3789                 }
3790
3791                 spin_unlock(&ctl->tree_lock);
3792                 trim_entry.start = extent_start;
3793                 trim_entry.bytes = extent_bytes;
3794                 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3795                 mutex_unlock(&ctl->cache_writeout_mutex);
3796
3797                 ret = do_trimming(block_group, total_trimmed, start, bytes,
3798                                   extent_start, extent_bytes, extent_trim_state,
3799                                   &trim_entry);
3800                 if (ret) {
3801                         block_group->discard_cursor = start + bytes;
3802                         break;
3803                 }
3804 next:
3805                 start += bytes;
3806                 block_group->discard_cursor = start;
3807                 if (async && *total_trimmed)
3808                         break;
3809
3810                 if (fatal_signal_pending(current)) {
3811                         ret = -ERESTARTSYS;
3812                         break;
3813                 }
3814
3815                 cond_resched();
3816         }
3817
3818         return ret;
3819
3820 out_unlock:
3821         block_group->discard_cursor = btrfs_block_group_end(block_group);
3822         spin_unlock(&ctl->tree_lock);
3823         mutex_unlock(&ctl->cache_writeout_mutex);
3824
3825         return ret;
3826 }
3827
3828 /*
3829  * If we break out of trimming a bitmap prematurely, we should reset the
3830  * trimming bit.  In a rather contrieved case, it's possible to race here so
3831  * reset the state to BTRFS_TRIM_STATE_UNTRIMMED.
3832  *
3833  * start = start of bitmap
3834  * end = near end of bitmap
3835  *
3836  * Thread 1:                    Thread 2:
3837  * trim_bitmaps(start)
3838  *                              trim_bitmaps(end)
3839  *                              end_trimming_bitmap()
3840  * reset_trimming_bitmap()
3841  */
3842 static void reset_trimming_bitmap(struct btrfs_free_space_ctl *ctl, u64 offset)
3843 {
3844         struct btrfs_free_space *entry;
3845
3846         spin_lock(&ctl->tree_lock);
3847         entry = tree_search_offset(ctl, offset, 1, 0);
3848         if (entry) {
3849                 if (btrfs_free_space_trimmed(entry)) {
3850                         ctl->discardable_extents[BTRFS_STAT_CURR] +=
3851                                 entry->bitmap_extents;
3852                         ctl->discardable_bytes[BTRFS_STAT_CURR] += entry->bytes;
3853                 }
3854                 entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3855         }
3856
3857         spin_unlock(&ctl->tree_lock);
3858 }
3859
3860 static void end_trimming_bitmap(struct btrfs_free_space_ctl *ctl,
3861                                 struct btrfs_free_space *entry)
3862 {
3863         if (btrfs_free_space_trimming_bitmap(entry)) {
3864                 entry->trim_state = BTRFS_TRIM_STATE_TRIMMED;
3865                 ctl->discardable_extents[BTRFS_STAT_CURR] -=
3866                         entry->bitmap_extents;
3867                 ctl->discardable_bytes[BTRFS_STAT_CURR] -= entry->bytes;
3868         }
3869 }
3870
3871 /*
3872  * If @async is set, then we will trim 1 region and return.
3873  */
3874 static int trim_bitmaps(struct btrfs_block_group *block_group,
3875                         u64 *total_trimmed, u64 start, u64 end, u64 minlen,
3876                         u64 maxlen, bool async)
3877 {
3878         struct btrfs_discard_ctl *discard_ctl =
3879                                         &block_group->fs_info->discard_ctl;
3880         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3881         struct btrfs_free_space *entry;
3882         int ret = 0;
3883         int ret2;
3884         u64 bytes;
3885         u64 offset = offset_to_bitmap(ctl, start);
3886         const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3887
3888         while (offset < end) {
3889                 bool next_bitmap = false;
3890                 struct btrfs_trim_range trim_entry;
3891
3892                 mutex_lock(&ctl->cache_writeout_mutex);
3893                 spin_lock(&ctl->tree_lock);
3894
3895                 if (ctl->free_space < minlen) {
3896                         block_group->discard_cursor =
3897                                 btrfs_block_group_end(block_group);
3898                         spin_unlock(&ctl->tree_lock);
3899                         mutex_unlock(&ctl->cache_writeout_mutex);
3900                         break;
3901                 }
3902
3903                 entry = tree_search_offset(ctl, offset, 1, 0);
3904                 /*
3905                  * Bitmaps are marked trimmed lossily now to prevent constant
3906                  * discarding of the same bitmap (the reason why we are bound
3907                  * by the filters).  So, retrim the block group bitmaps when we
3908                  * are preparing to punt to the unused_bgs list.  This uses
3909                  * @minlen to determine if we are in BTRFS_DISCARD_INDEX_UNUSED
3910                  * which is the only discard index which sets minlen to 0.
3911                  */
3912                 if (!entry || (async && minlen && start == offset &&
3913                                btrfs_free_space_trimmed(entry))) {
3914                         spin_unlock(&ctl->tree_lock);
3915                         mutex_unlock(&ctl->cache_writeout_mutex);
3916                         next_bitmap = true;
3917                         goto next;
3918                 }
3919
3920                 /*
3921                  * Async discard bitmap trimming begins at by setting the start
3922                  * to be key.objectid and the offset_to_bitmap() aligns to the
3923                  * start of the bitmap.  This lets us know we are fully
3924                  * scanning the bitmap rather than only some portion of it.
3925                  */
3926                 if (start == offset)
3927                         entry->trim_state = BTRFS_TRIM_STATE_TRIMMING;
3928
3929                 bytes = minlen;
3930                 ret2 = search_bitmap(ctl, entry, &start, &bytes, false);
3931                 if (ret2 || start >= end) {
3932                         /*
3933                          * We lossily consider a bitmap trimmed if we only skip
3934                          * over regions <= BTRFS_ASYNC_DISCARD_MIN_FILTER.
3935                          */
3936                         if (ret2 && minlen <= BTRFS_ASYNC_DISCARD_MIN_FILTER)
3937                                 end_trimming_bitmap(ctl, entry);
3938                         else
3939                                 entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3940                         spin_unlock(&ctl->tree_lock);
3941                         mutex_unlock(&ctl->cache_writeout_mutex);
3942                         next_bitmap = true;
3943                         goto next;
3944                 }
3945
3946                 /*
3947                  * We already trimmed a region, but are using the locking above
3948                  * to reset the trim_state.
3949                  */
3950                 if (async && *total_trimmed) {
3951                         spin_unlock(&ctl->tree_lock);
3952                         mutex_unlock(&ctl->cache_writeout_mutex);
3953                         goto out;
3954                 }
3955
3956                 bytes = min(bytes, end - start);
3957                 if (bytes < minlen || (async && maxlen && bytes > maxlen)) {
3958                         spin_unlock(&ctl->tree_lock);
3959                         mutex_unlock(&ctl->cache_writeout_mutex);
3960                         goto next;
3961                 }
3962
3963                 /*
3964                  * Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
3965                  * If X < @minlen, we won't trim X when we come back around.
3966                  * So trim it now.  We differ here from trimming extents as we
3967                  * don't keep individual state per bit.
3968                  */
3969                 if (async &&
3970                     max_discard_size &&
3971                     bytes > (max_discard_size + minlen))
3972                         bytes = max_discard_size;
3973
3974                 bitmap_clear_bits(ctl, entry, start, bytes, true);
3975                 if (entry->bytes == 0)
3976                         free_bitmap(ctl, entry);
3977
3978                 spin_unlock(&ctl->tree_lock);
3979                 trim_entry.start = start;
3980                 trim_entry.bytes = bytes;
3981                 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3982                 mutex_unlock(&ctl->cache_writeout_mutex);
3983
3984                 ret = do_trimming(block_group, total_trimmed, start, bytes,
3985                                   start, bytes, 0, &trim_entry);
3986                 if (ret) {
3987                         reset_trimming_bitmap(ctl, offset);
3988                         block_group->discard_cursor =
3989                                 btrfs_block_group_end(block_group);
3990                         break;
3991                 }
3992 next:
3993                 if (next_bitmap) {
3994                         offset += BITS_PER_BITMAP * ctl->unit;
3995                         start = offset;
3996                 } else {
3997                         start += bytes;
3998                 }
3999                 block_group->discard_cursor = start;
4000
4001                 if (fatal_signal_pending(current)) {
4002                         if (start != offset)
4003                                 reset_trimming_bitmap(ctl, offset);
4004                         ret = -ERESTARTSYS;
4005                         break;
4006                 }
4007
4008                 cond_resched();
4009         }
4010
4011         if (offset >= end)
4012                 block_group->discard_cursor = end;
4013
4014 out:
4015         return ret;
4016 }
4017
4018 int btrfs_trim_block_group(struct btrfs_block_group *block_group,
4019                            u64 *trimmed, u64 start, u64 end, u64 minlen)
4020 {
4021         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
4022         int ret;
4023         u64 rem = 0;
4024
4025         ASSERT(!btrfs_is_zoned(block_group->fs_info));
4026
4027         *trimmed = 0;
4028
4029         spin_lock(&block_group->lock);
4030         if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4031                 spin_unlock(&block_group->lock);
4032                 return 0;
4033         }
4034         btrfs_freeze_block_group(block_group);
4035         spin_unlock(&block_group->lock);
4036
4037         ret = trim_no_bitmap(block_group, trimmed, start, end, minlen, false);
4038         if (ret)
4039                 goto out;
4040
4041         ret = trim_bitmaps(block_group, trimmed, start, end, minlen, 0, false);
4042         div64_u64_rem(end, BITS_PER_BITMAP * ctl->unit, &rem);
4043         /* If we ended in the middle of a bitmap, reset the trimming flag */
4044         if (rem)
4045                 reset_trimming_bitmap(ctl, offset_to_bitmap(ctl, end));
4046 out:
4047         btrfs_unfreeze_block_group(block_group);
4048         return ret;
4049 }
4050
4051 int btrfs_trim_block_group_extents(struct btrfs_block_group *block_group,
4052                                    u64 *trimmed, u64 start, u64 end, u64 minlen,
4053                                    bool async)
4054 {
4055         int ret;
4056
4057         *trimmed = 0;
4058
4059         spin_lock(&block_group->lock);
4060         if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4061                 spin_unlock(&block_group->lock);
4062                 return 0;
4063         }
4064         btrfs_freeze_block_group(block_group);
4065         spin_unlock(&block_group->lock);
4066
4067         ret = trim_no_bitmap(block_group, trimmed, start, end, minlen, async);
4068         btrfs_unfreeze_block_group(block_group);
4069
4070         return ret;
4071 }
4072
4073 int btrfs_trim_block_group_bitmaps(struct btrfs_block_group *block_group,
4074                                    u64 *trimmed, u64 start, u64 end, u64 minlen,
4075                                    u64 maxlen, bool async)
4076 {
4077         int ret;
4078
4079         *trimmed = 0;
4080
4081         spin_lock(&block_group->lock);
4082         if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4083                 spin_unlock(&block_group->lock);
4084                 return 0;
4085         }
4086         btrfs_freeze_block_group(block_group);
4087         spin_unlock(&block_group->lock);
4088
4089         ret = trim_bitmaps(block_group, trimmed, start, end, minlen, maxlen,
4090                            async);
4091
4092         btrfs_unfreeze_block_group(block_group);
4093
4094         return ret;
4095 }
4096
4097 bool btrfs_free_space_cache_v1_active(struct btrfs_fs_info *fs_info)
4098 {
4099         return btrfs_super_cache_generation(fs_info->super_copy);
4100 }
4101
4102 static int cleanup_free_space_cache_v1(struct btrfs_fs_info *fs_info,
4103                                        struct btrfs_trans_handle *trans)
4104 {
4105         struct btrfs_block_group *block_group;
4106         struct rb_node *node;
4107         int ret = 0;
4108
4109         btrfs_info(fs_info, "cleaning free space cache v1");
4110
4111         node = rb_first_cached(&fs_info->block_group_cache_tree);
4112         while (node) {
4113                 block_group = rb_entry(node, struct btrfs_block_group, cache_node);
4114                 ret = btrfs_remove_free_space_inode(trans, NULL, block_group);
4115                 if (ret)
4116                         goto out;
4117                 node = rb_next(node);
4118         }
4119 out:
4120         return ret;
4121 }
4122
4123 int btrfs_set_free_space_cache_v1_active(struct btrfs_fs_info *fs_info, bool active)
4124 {
4125         struct btrfs_trans_handle *trans;
4126         int ret;
4127
4128         /*
4129          * update_super_roots will appropriately set or unset
4130          * super_copy->cache_generation based on SPACE_CACHE and
4131          * BTRFS_FS_CLEANUP_SPACE_CACHE_V1. For this reason, we need a
4132          * transaction commit whether we are enabling space cache v1 and don't
4133          * have any other work to do, or are disabling it and removing free
4134          * space inodes.
4135          */
4136         trans = btrfs_start_transaction(fs_info->tree_root, 0);
4137         if (IS_ERR(trans))
4138                 return PTR_ERR(trans);
4139
4140         if (!active) {
4141                 set_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags);
4142                 ret = cleanup_free_space_cache_v1(fs_info, trans);
4143                 if (ret) {
4144                         btrfs_abort_transaction(trans, ret);
4145                         btrfs_end_transaction(trans);
4146                         goto out;
4147                 }
4148         }
4149
4150         ret = btrfs_commit_transaction(trans);
4151 out:
4152         clear_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags);
4153
4154         return ret;
4155 }
4156
4157 int __init btrfs_free_space_init(void)
4158 {
4159         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
4160                         sizeof(struct btrfs_free_space), 0,
4161                         SLAB_MEM_SPREAD, NULL);
4162         if (!btrfs_free_space_cachep)
4163                 return -ENOMEM;
4164
4165         btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
4166                                                         PAGE_SIZE, PAGE_SIZE,
4167                                                         SLAB_MEM_SPREAD, NULL);
4168         if (!btrfs_free_space_bitmap_cachep) {
4169                 kmem_cache_destroy(btrfs_free_space_cachep);
4170                 return -ENOMEM;
4171         }
4172
4173         return 0;
4174 }
4175
4176 void __cold btrfs_free_space_exit(void)
4177 {
4178         kmem_cache_destroy(btrfs_free_space_cachep);
4179         kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
4180 }
4181
4182 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4183 /*
4184  * Use this if you need to make a bitmap or extent entry specifically, it
4185  * doesn't do any of the merging that add_free_space does, this acts a lot like
4186  * how the free space cache loading stuff works, so you can get really weird
4187  * configurations.
4188  */
4189 int test_add_free_space_entry(struct btrfs_block_group *cache,
4190                               u64 offset, u64 bytes, bool bitmap)
4191 {
4192         struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
4193         struct btrfs_free_space *info = NULL, *bitmap_info;
4194         void *map = NULL;
4195         enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_TRIMMED;
4196         u64 bytes_added;
4197         int ret;
4198
4199 again:
4200         if (!info) {
4201                 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
4202                 if (!info)
4203                         return -ENOMEM;
4204         }
4205
4206         if (!bitmap) {
4207                 spin_lock(&ctl->tree_lock);
4208                 info->offset = offset;
4209                 info->bytes = bytes;
4210                 info->max_extent_size = 0;
4211                 ret = link_free_space(ctl, info);
4212                 spin_unlock(&ctl->tree_lock);
4213                 if (ret)
4214                         kmem_cache_free(btrfs_free_space_cachep, info);
4215                 return ret;
4216         }
4217
4218         if (!map) {
4219                 map = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep, GFP_NOFS);
4220                 if (!map) {
4221                         kmem_cache_free(btrfs_free_space_cachep, info);
4222                         return -ENOMEM;
4223                 }
4224         }
4225
4226         spin_lock(&ctl->tree_lock);
4227         bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
4228                                          1, 0);
4229         if (!bitmap_info) {
4230                 info->bitmap = map;
4231                 map = NULL;
4232                 add_new_bitmap(ctl, info, offset);
4233                 bitmap_info = info;
4234                 info = NULL;
4235         }
4236
4237         bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes,
4238                                           trim_state);
4239
4240         bytes -= bytes_added;
4241         offset += bytes_added;
4242         spin_unlock(&ctl->tree_lock);
4243
4244         if (bytes)
4245                 goto again;
4246
4247         if (info)
4248                 kmem_cache_free(btrfs_free_space_cachep, info);
4249         if (map)
4250                 kmem_cache_free(btrfs_free_space_bitmap_cachep, map);
4251         return 0;
4252 }
4253
4254 /*
4255  * Checks to see if the given range is in the free space cache.  This is really
4256  * just used to check the absence of space, so if there is free space in the
4257  * range at all we will return 1.
4258  */
4259 int test_check_exists(struct btrfs_block_group *cache,
4260                       u64 offset, u64 bytes)
4261 {
4262         struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
4263         struct btrfs_free_space *info;
4264         int ret = 0;
4265
4266         spin_lock(&ctl->tree_lock);
4267         info = tree_search_offset(ctl, offset, 0, 0);
4268         if (!info) {
4269                 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
4270                                           1, 0);
4271                 if (!info)
4272                         goto out;
4273         }
4274
4275 have_info:
4276         if (info->bitmap) {
4277                 u64 bit_off, bit_bytes;
4278                 struct rb_node *n;
4279                 struct btrfs_free_space *tmp;
4280
4281                 bit_off = offset;
4282                 bit_bytes = ctl->unit;
4283                 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes, false);
4284                 if (!ret) {
4285                         if (bit_off == offset) {
4286                                 ret = 1;
4287                                 goto out;
4288                         } else if (bit_off > offset &&
4289                                    offset + bytes > bit_off) {
4290                                 ret = 1;
4291                                 goto out;
4292                         }
4293                 }
4294
4295                 n = rb_prev(&info->offset_index);
4296                 while (n) {
4297                         tmp = rb_entry(n, struct btrfs_free_space,
4298                                        offset_index);
4299                         if (tmp->offset + tmp->bytes < offset)
4300                                 break;
4301                         if (offset + bytes < tmp->offset) {
4302                                 n = rb_prev(&tmp->offset_index);
4303                                 continue;
4304                         }
4305                         info = tmp;
4306                         goto have_info;
4307                 }
4308
4309                 n = rb_next(&info->offset_index);
4310                 while (n) {
4311                         tmp = rb_entry(n, struct btrfs_free_space,
4312                                        offset_index);
4313                         if (offset + bytes < tmp->offset)
4314                                 break;
4315                         if (tmp->offset + tmp->bytes < offset) {
4316                                 n = rb_next(&tmp->offset_index);
4317                                 continue;
4318                         }
4319                         info = tmp;
4320                         goto have_info;
4321                 }
4322
4323                 ret = 0;
4324                 goto out;
4325         }
4326
4327         if (info->offset == offset) {
4328                 ret = 1;
4329                 goto out;
4330         }
4331
4332         if (offset > info->offset && offset < info->offset + info->bytes)
4333                 ret = 1;
4334 out:
4335         spin_unlock(&ctl->tree_lock);
4336         return ret;
4337 }
4338 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */