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