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