fs/f2fs/extent_cache.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * f2fs extent cache support
   4  *
   5  * Copyright (c) 2015 Motorola Mobility
   6  * Copyright (c) 2015 Samsung Electronics
   7  * Authors: Jaegeuk Kim <jaegeuk@kernel.org>
   8  *          Chao Yu <chao2.yu@samsung.com>
   9  */
  10
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13
  14 #include "f2fs.h"
  15 #include "node.h"
  16 #include <trace/events/f2fs.h>
  17
  18 static void __set_extent_info(struct extent_info *ei,
  19                                 unsigned int fofs, unsigned int len,
  20                                 block_t blk, bool keep_clen,
  21                                 enum extent_type type)
  22 {
  23         ei->fofs = fofs;
  24         ei->len = len;
  25
  26         if (type == EX_READ) {
  27                 ei->blk = blk;
  28                 if (keep_clen)
  29                         return;
  30 #ifdef CONFIG_F2FS_FS_COMPRESSION
  31                 ei->c_len = 0;
  32 #endif
  33         }
  34 }
  35
  36 static bool __may_read_extent_tree(struct inode *inode)
  37 {
  38         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  39
  40         if (!test_opt(sbi, READ_EXTENT_CACHE))
  41                 return false;
  42         if (is_inode_flag_set(inode, FI_NO_EXTENT))
  43                 return false;
  44         if (is_inode_flag_set(inode, FI_COMPRESSED_FILE) &&
  45                          !f2fs_sb_has_readonly(sbi))
  46                 return false;
  47         return S_ISREG(inode->i_mode);
  48 }
  49
  50 static bool __may_extent_tree(struct inode *inode, enum extent_type type)
  51 {
  52         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  53
  54         /*
  55          * for recovered files during mount do not create extents
  56          * if shrinker is not registered.
  57          */
  58         if (list_empty(&sbi->s_list))
  59                 return false;
  60
  61         if (type == EX_READ)
  62                 return __may_read_extent_tree(inode);
  63         return false;
  64 }
  65
  66 static void __try_update_largest_extent(struct extent_tree *et,
  67                                                 struct extent_node *en)
  68 {
  69         if (et->type != EX_READ)
  70                 return;
  71         if (en->ei.len <= et->largest.len)
  72                 return;
  73
  74         et->largest = en->ei;
  75         et->largest_updated = true;
  76 }
  77
  78 static bool __is_extent_mergeable(struct extent_info *back,
  79                 struct extent_info *front, enum extent_type type)
  80 {
  81         if (type == EX_READ) {
  82 #ifdef CONFIG_F2FS_FS_COMPRESSION
  83                 if (back->c_len && back->len != back->c_len)
  84                         return false;
  85                 if (front->c_len && front->len != front->c_len)
  86                         return false;
  87 #endif
  88                 return (back->fofs + back->len == front->fofs &&
  89                                 back->blk + back->len == front->blk);
  90         }
  91         return false;
  92 }
  93
  94 static bool __is_back_mergeable(struct extent_info *cur,
  95                 struct extent_info *back, enum extent_type type)
  96 {
  97         return __is_extent_mergeable(back, cur, type);
  98 }
  99
 100 static bool __is_front_mergeable(struct extent_info *cur,
 101                 struct extent_info *front, enum extent_type type)
 102 {
 103         return __is_extent_mergeable(cur, front, type);
 104 }
 105
 106 static struct rb_entry *__lookup_rb_tree_fast(struct rb_entry *cached_re,
 107                                                         unsigned int ofs)
 108 {
 109         if (cached_re) {
 110                 if (cached_re->ofs <= ofs &&
 111                                 cached_re->ofs + cached_re->len > ofs) {
 112                         return cached_re;
 113                 }
 114         }
 115         return NULL;
 116 }
 117
 118 static struct rb_entry *__lookup_rb_tree_slow(struct rb_root_cached *root,
 119                                                         unsigned int ofs)
 120 {
 121         struct rb_node *node = root->rb_root.rb_node;
 122         struct rb_entry *re;
 123
 124         while (node) {
 125                 re = rb_entry(node, struct rb_entry, rb_node);
 126
 127                 if (ofs < re->ofs)
 128                         node = node->rb_left;
 129                 else if (ofs >= re->ofs + re->len)
 130                         node = node->rb_right;
 131                 else
 132                         return re;
 133         }
 134         return NULL;
 135 }
 136
 137 struct rb_entry *f2fs_lookup_rb_tree(struct rb_root_cached *root,
 138                                 struct rb_entry *cached_re, unsigned int ofs)
 139 {
 140         struct rb_entry *re;
 141
 142         re = __lookup_rb_tree_fast(cached_re, ofs);
 143         if (!re)
 144                 return __lookup_rb_tree_slow(root, ofs);
 145
 146         return re;
 147 }
 148
 149 struct rb_node **f2fs_lookup_rb_tree_ext(struct f2fs_sb_info *sbi,
 150                                         struct rb_root_cached *root,
 151                                         struct rb_node **parent,
 152                                         unsigned long long key, bool *leftmost)
 153 {
 154         struct rb_node **p = &root->rb_root.rb_node;
 155         struct rb_entry *re;
 156
 157         while (*p) {
 158                 *parent = *p;
 159                 re = rb_entry(*parent, struct rb_entry, rb_node);
 160
 161                 if (key < re->key) {
 162                         p = &(*p)->rb_left;
 163                 } else {
 164                         p = &(*p)->rb_right;
 165                         *leftmost = false;
 166                 }
 167         }
 168
 169         return p;
 170 }
 171
 172 struct rb_node **f2fs_lookup_rb_tree_for_insert(struct f2fs_sb_info *sbi,
 173                                 struct rb_root_cached *root,
 174                                 struct rb_node **parent,
 175                                 unsigned int ofs, bool *leftmost)
 176 {
 177         struct rb_node **p = &root->rb_root.rb_node;
 178         struct rb_entry *re;
 179
 180         while (*p) {
 181                 *parent = *p;
 182                 re = rb_entry(*parent, struct rb_entry, rb_node);
 183
 184                 if (ofs < re->ofs) {
 185                         p = &(*p)->rb_left;
 186                 } else if (ofs >= re->ofs + re->len) {
 187                         p = &(*p)->rb_right;
 188                         *leftmost = false;
 189                 } else {
 190                         f2fs_bug_on(sbi, 1);
 191                 }
 192         }
 193
 194         return p;
 195 }
 196
 197 /*
 198  * lookup rb entry in position of @ofs in rb-tree,
 199  * if hit, return the entry, otherwise, return NULL
 200  * @prev_ex: extent before ofs
 201  * @next_ex: extent after ofs
 202  * @insert_p: insert point for new extent at ofs
 203  * in order to simpfy the insertion after.
 204  * tree must stay unchanged between lookup and insertion.
 205  */
 206 struct rb_entry *f2fs_lookup_rb_tree_ret(struct rb_root_cached *root,
 207                                 struct rb_entry *cached_re,
 208                                 unsigned int ofs,
 209                                 struct rb_entry **prev_entry,
 210                                 struct rb_entry **next_entry,
 211                                 struct rb_node ***insert_p,
 212                                 struct rb_node **insert_parent,
 213                                 bool force, bool *leftmost)
 214 {
 215         struct rb_node **pnode = &root->rb_root.rb_node;
 216         struct rb_node *parent = NULL, *tmp_node;
 217         struct rb_entry *re = cached_re;
 218
 219         *insert_p = NULL;
 220         *insert_parent = NULL;
 221         *prev_entry = NULL;
 222         *next_entry = NULL;
 223
 224         if (RB_EMPTY_ROOT(&root->rb_root))
 225                 return NULL;
 226
 227         if (re) {
 228                 if (re->ofs <= ofs && re->ofs + re->len > ofs)
 229                         goto lookup_neighbors;
 230         }
 231
 232         if (leftmost)
 233                 *leftmost = true;
 234
 235         while (*pnode) {
 236                 parent = *pnode;
 237                 re = rb_entry(*pnode, struct rb_entry, rb_node);
 238
 239                 if (ofs < re->ofs) {
 240                         pnode = &(*pnode)->rb_left;
 241                 } else if (ofs >= re->ofs + re->len) {
 242                         pnode = &(*pnode)->rb_right;
 243                         if (leftmost)
 244                                 *leftmost = false;
 245                 } else {
 246                         goto lookup_neighbors;
 247                 }
 248         }
 249
 250         *insert_p = pnode;
 251         *insert_parent = parent;
 252
 253         re = rb_entry(parent, struct rb_entry, rb_node);
 254         tmp_node = parent;
 255         if (parent && ofs > re->ofs)
 256                 tmp_node = rb_next(parent);
 257         *next_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
 258
 259         tmp_node = parent;
 260         if (parent && ofs < re->ofs)
 261                 tmp_node = rb_prev(parent);
 262         *prev_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
 263         return NULL;
 264
 265 lookup_neighbors:
 266         if (ofs == re->ofs || force) {
 267                 /* lookup prev node for merging backward later */
 268                 tmp_node = rb_prev(&re->rb_node);
 269                 *prev_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
 270         }
 271         if (ofs == re->ofs + re->len - 1 || force) {
 272                 /* lookup next node for merging frontward later */
 273                 tmp_node = rb_next(&re->rb_node);
 274                 *next_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
 275         }
 276         return re;
 277 }
 278
 279 bool f2fs_check_rb_tree_consistence(struct f2fs_sb_info *sbi,
 280                                 struct rb_root_cached *root, bool check_key)
 281 {
 282 #ifdef CONFIG_F2FS_CHECK_FS
 283         struct rb_node *cur = rb_first_cached(root), *next;
 284         struct rb_entry *cur_re, *next_re;
 285
 286         if (!cur)
 287                 return true;
 288
 289         while (cur) {
 290                 next = rb_next(cur);
 291                 if (!next)
 292                         return true;
 293
 294                 cur_re = rb_entry(cur, struct rb_entry, rb_node);
 295                 next_re = rb_entry(next, struct rb_entry, rb_node);
 296
 297                 if (check_key) {
 298                         if (cur_re->key > next_re->key) {
 299                                 f2fs_info(sbi, "inconsistent rbtree, "
 300                                         "cur(%llu) next(%llu)",
 301                                         cur_re->key, next_re->key);
 302                                 return false;
 303                         }
 304                         goto next;
 305                 }
 306
 307                 if (cur_re->ofs + cur_re->len > next_re->ofs) {
 308                         f2fs_info(sbi, "inconsistent rbtree, cur(%u, %u) next(%u, %u)",
 309                                   cur_re->ofs, cur_re->len,
 310                                   next_re->ofs, next_re->len);
 311                         return false;
 312                 }
 313 next:
 314                 cur = next;
 315         }
 316 #endif
 317         return true;
 318 }
 319
 320 static struct kmem_cache *extent_tree_slab;
 321 static struct kmem_cache *extent_node_slab;
 322
 323 static struct extent_node *__attach_extent_node(struct f2fs_sb_info *sbi,
 324                                 struct extent_tree *et, struct extent_info *ei,
 325                                 struct rb_node *parent, struct rb_node **p,
 326                                 bool leftmost)
 327 {
 328         struct extent_tree_info *eti = &sbi->extent_tree[et->type];
 329         struct extent_node *en;
 330
 331         en = f2fs_kmem_cache_alloc(extent_node_slab, GFP_ATOMIC, false, sbi);
 332         if (!en)
 333                 return NULL;
 334
 335         en->ei = *ei;
 336         INIT_LIST_HEAD(&en->list);
 337         en->et = et;
 338
 339         rb_link_node(&en->rb_node, parent, p);
 340         rb_insert_color_cached(&en->rb_node, &et->root, leftmost);
 341         atomic_inc(&et->node_cnt);
 342         atomic_inc(&eti->total_ext_node);
 343         return en;
 344 }
 345
 346 static void __detach_extent_node(struct f2fs_sb_info *sbi,
 347                                 struct extent_tree *et, struct extent_node *en)
 348 {
 349         struct extent_tree_info *eti = &sbi->extent_tree[et->type];
 350
 351         rb_erase_cached(&en->rb_node, &et->root);
 352         atomic_dec(&et->node_cnt);
 353         atomic_dec(&eti->total_ext_node);
 354
 355         if (et->cached_en == en)
 356                 et->cached_en = NULL;
 357         kmem_cache_free(extent_node_slab, en);
 358 }
 359
 360 /*
 361  * Flow to release an extent_node:
 362  * 1. list_del_init
 363  * 2. __detach_extent_node
 364  * 3. kmem_cache_free.
 365  */
 366 static void __release_extent_node(struct f2fs_sb_info *sbi,
 367                         struct extent_tree *et, struct extent_node *en)
 368 {
 369         struct extent_tree_info *eti = &sbi->extent_tree[et->type];
 370
 371         spin_lock(&eti->extent_lock);
 372         f2fs_bug_on(sbi, list_empty(&en->list));
 373         list_del_init(&en->list);
 374         spin_unlock(&eti->extent_lock);
 375
 376         __detach_extent_node(sbi, et, en);
 377 }
 378
 379 static struct extent_tree *__grab_extent_tree(struct inode *inode,
 380                                                 enum extent_type type)
 381 {
 382         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 383         struct extent_tree_info *eti = &sbi->extent_tree[type];
 384         struct extent_tree *et;
 385         nid_t ino = inode->i_ino;
 386
 387         mutex_lock(&eti->extent_tree_lock);
 388         et = radix_tree_lookup(&eti->extent_tree_root, ino);
 389         if (!et) {
 390                 et = f2fs_kmem_cache_alloc(extent_tree_slab,
 391                                         GFP_NOFS, true, NULL);
 392                 f2fs_radix_tree_insert(&eti->extent_tree_root, ino, et);
 393                 memset(et, 0, sizeof(struct extent_tree));
 394                 et->ino = ino;
 395                 et->type = type;
 396                 et->root = RB_ROOT_CACHED;
 397                 et->cached_en = NULL;
 398                 rwlock_init(&et->lock);
 399                 INIT_LIST_HEAD(&et->list);
 400                 atomic_set(&et->node_cnt, 0);
 401                 atomic_inc(&eti->total_ext_tree);
 402         } else {
 403                 atomic_dec(&eti->total_zombie_tree);
 404                 list_del_init(&et->list);
 405         }
 406         mutex_unlock(&eti->extent_tree_lock);
 407
 408         /* never died until evict_inode */
 409         F2FS_I(inode)->extent_tree[type] = et;
 410
 411         return et;
 412 }
 413
 414 static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi,
 415                                         struct extent_tree *et)
 416 {
 417         struct rb_node *node, *next;
 418         struct extent_node *en;
 419         unsigned int count = atomic_read(&et->node_cnt);
 420
 421         node = rb_first_cached(&et->root);
 422         while (node) {
 423                 next = rb_next(node);
 424                 en = rb_entry(node, struct extent_node, rb_node);
 425                 __release_extent_node(sbi, et, en);
 426                 node = next;
 427         }
 428
 429         return count - atomic_read(&et->node_cnt);
 430 }
 431
 432 static void __drop_largest_extent(struct extent_tree *et,
 433                                         pgoff_t fofs, unsigned int len)
 434 {
 435         if (fofs < et->largest.fofs + et->largest.len &&
 436                         fofs + len > et->largest.fofs) {
 437                 et->largest.len = 0;
 438                 et->largest_updated = true;
 439         }
 440 }
 441
 442 /* return true, if inode page is changed */
 443 static void __f2fs_init_extent_tree(struct inode *inode, struct page *ipage,
 444                                                         enum extent_type type)
 445 {
 446         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 447         struct extent_tree_info *eti = &sbi->extent_tree[type];
 448         struct f2fs_extent *i_ext = ipage ? &F2FS_INODE(ipage)->i_ext : NULL;
 449         struct extent_tree *et;
 450         struct extent_node *en;
 451         struct extent_info ei;
 452
 453         if (!__may_extent_tree(inode, type)) {
 454                 /* drop largest read extent */
 455                 if (type == EX_READ && i_ext && i_ext->len) {
 456                         f2fs_wait_on_page_writeback(ipage, NODE, true, true);
 457                         i_ext->len = 0;
 458                         set_page_dirty(ipage);
 459                 }
 460                 goto out;
 461         }
 462
 463         et = __grab_extent_tree(inode, type);
 464
 465         if (!i_ext || !i_ext->len)
 466                 goto out;
 467
 468         BUG_ON(type != EX_READ);
 469
 470         get_read_extent_info(&ei, i_ext);
 471
 472         write_lock(&et->lock);
 473         if (atomic_read(&et->node_cnt))
 474                 goto unlock_out;
 475
 476         en = __attach_extent_node(sbi, et, &ei, NULL,
 477                                 &et->root.rb_root.rb_node, true);
 478         if (en) {
 479                 et->largest = en->ei;
 480                 et->cached_en = en;
 481
 482                 spin_lock(&eti->extent_lock);
 483                 list_add_tail(&en->list, &eti->extent_list);
 484                 spin_unlock(&eti->extent_lock);
 485         }
 486 unlock_out:
 487         write_unlock(&et->lock);
 488 out:
 489         if (type == EX_READ && !F2FS_I(inode)->extent_tree[EX_READ])
 490                 set_inode_flag(inode, FI_NO_EXTENT);
 491 }
 492
 493 void f2fs_init_extent_tree(struct inode *inode, struct page *ipage)
 494 {
 495         /* initialize read cache */
 496         __f2fs_init_extent_tree(inode, ipage, EX_READ);
 497 }
 498
 499 static bool __lookup_extent_tree(struct inode *inode, pgoff_t pgofs,
 500                         struct extent_info *ei, enum extent_type type)
 501 {
 502         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 503         struct extent_tree_info *eti = &sbi->extent_tree[type];
 504         struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
 505         struct extent_node *en;
 506         bool ret = false;
 507
 508         if (!et)
 509                 return false;
 510
 511         trace_f2fs_lookup_extent_tree_start(inode, pgofs, type);
 512
 513         read_lock(&et->lock);
 514
 515         if (type == EX_READ &&
 516                         et->largest.fofs <= pgofs &&
 517                         et->largest.fofs + et->largest.len > pgofs) {
 518                 *ei = et->largest;
 519                 ret = true;
 520                 stat_inc_largest_node_hit(sbi);
 521                 goto out;
 522         }
 523
 524         en = (struct extent_node *)f2fs_lookup_rb_tree(&et->root,
 525                                 (struct rb_entry *)et->cached_en, pgofs);
 526         if (!en)
 527                 goto out;
 528
 529         if (en == et->cached_en)
 530                 stat_inc_cached_node_hit(sbi, type);
 531         else
 532                 stat_inc_rbtree_node_hit(sbi, type);
 533
 534         *ei = en->ei;
 535         spin_lock(&eti->extent_lock);
 536         if (!list_empty(&en->list)) {
 537                 list_move_tail(&en->list, &eti->extent_list);
 538                 et->cached_en = en;
 539         }
 540         spin_unlock(&eti->extent_lock);
 541         ret = true;
 542 out:
 543         stat_inc_total_hit(sbi, type);
 544         read_unlock(&et->lock);
 545
 546         if (type == EX_READ)
 547                 trace_f2fs_lookup_read_extent_tree_end(inode, pgofs, ei);
 548         return ret;
 549 }
 550
 551 static struct extent_node *__try_merge_extent_node(struct f2fs_sb_info *sbi,
 552                                 struct extent_tree *et, struct extent_info *ei,
 553                                 struct extent_node *prev_ex,
 554                                 struct extent_node *next_ex)
 555 {
 556         struct extent_tree_info *eti = &sbi->extent_tree[et->type];
 557         struct extent_node *en = NULL;
 558
 559         if (prev_ex && __is_back_mergeable(ei, &prev_ex->ei, et->type)) {
 560                 prev_ex->ei.len += ei->len;
 561                 ei = &prev_ex->ei;
 562                 en = prev_ex;
 563         }
 564
 565         if (next_ex && __is_front_mergeable(ei, &next_ex->ei, et->type)) {
 566                 next_ex->ei.fofs = ei->fofs;
 567                 next_ex->ei.len += ei->len;
 568                 if (et->type == EX_READ)
 569                         next_ex->ei.blk = ei->blk;
 570                 if (en)
 571                         __release_extent_node(sbi, et, prev_ex);
 572
 573                 en = next_ex;
 574         }
 575
 576         if (!en)
 577                 return NULL;
 578
 579         __try_update_largest_extent(et, en);
 580
 581         spin_lock(&eti->extent_lock);
 582         if (!list_empty(&en->list)) {
 583                 list_move_tail(&en->list, &eti->extent_list);
 584                 et->cached_en = en;
 585         }
 586         spin_unlock(&eti->extent_lock);
 587         return en;
 588 }
 589
 590 static struct extent_node *__insert_extent_tree(struct f2fs_sb_info *sbi,
 591                                 struct extent_tree *et, struct extent_info *ei,
 592                                 struct rb_node **insert_p,
 593                                 struct rb_node *insert_parent,
 594                                 bool leftmost)
 595 {
 596         struct extent_tree_info *eti = &sbi->extent_tree[et->type];
 597         struct rb_node **p;
 598         struct rb_node *parent = NULL;
 599         struct extent_node *en = NULL;
 600
 601         if (insert_p && insert_parent) {
 602                 parent = insert_parent;
 603                 p = insert_p;
 604                 goto do_insert;
 605         }
 606
 607         leftmost = true;
 608
 609         p = f2fs_lookup_rb_tree_for_insert(sbi, &et->root, &parent,
 610                                                 ei->fofs, &leftmost);
 611 do_insert:
 612         en = __attach_extent_node(sbi, et, ei, parent, p, leftmost);
 613         if (!en)
 614                 return NULL;
 615
 616         __try_update_largest_extent(et, en);
 617
 618         /* update in global extent list */
 619         spin_lock(&eti->extent_lock);
 620         list_add_tail(&en->list, &eti->extent_list);
 621         et->cached_en = en;
 622         spin_unlock(&eti->extent_lock);
 623         return en;
 624 }
 625
 626 static void __update_extent_tree_range(struct inode *inode,
 627                         struct extent_info *tei, enum extent_type type)
 628 {
 629         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 630         struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
 631         struct extent_node *en = NULL, *en1 = NULL;
 632         struct extent_node *prev_en = NULL, *next_en = NULL;
 633         struct extent_info ei, dei, prev;
 634         struct rb_node **insert_p = NULL, *insert_parent = NULL;
 635         unsigned int fofs = tei->fofs, len = tei->len;
 636         unsigned int end = fofs + len;
 637         bool updated = false;
 638         bool leftmost = false;
 639
 640         if (!et)
 641                 return;
 642
 643         if (type == EX_READ)
 644                 trace_f2fs_update_read_extent_tree_range(inode, fofs, len,
 645                                                 tei->blk, 0);
 646         write_lock(&et->lock);
 647
 648         if (type == EX_READ) {
 649                 if (is_inode_flag_set(inode, FI_NO_EXTENT)) {
 650                         write_unlock(&et->lock);
 651                         return;
 652                 }
 653
 654                 prev = et->largest;
 655                 dei.len = 0;
 656
 657                 /*
 658                  * drop largest extent before lookup, in case it's already
 659                  * been shrunk from extent tree
 660                  */
 661                 __drop_largest_extent(et, fofs, len);
 662         }
 663
 664         /* 1. lookup first extent node in range [fofs, fofs + len - 1] */
 665         en = (struct extent_node *)f2fs_lookup_rb_tree_ret(&et->root,
 666                                         (struct rb_entry *)et->cached_en, fofs,
 667                                         (struct rb_entry **)&prev_en,
 668                                         (struct rb_entry **)&next_en,
 669                                         &insert_p, &insert_parent, false,
 670                                         &leftmost);
 671         if (!en)
 672                 en = next_en;
 673
 674         /* 2. invlidate all extent nodes in range [fofs, fofs + len - 1] */
 675         while (en && en->ei.fofs < end) {
 676                 unsigned int org_end;
 677                 int parts = 0;  /* # of parts current extent split into */
 678
 679                 next_en = en1 = NULL;
 680
 681                 dei = en->ei;
 682                 org_end = dei.fofs + dei.len;
 683                 f2fs_bug_on(sbi, fofs >= org_end);
 684
 685                 if (fofs > dei.fofs && (type != EX_READ ||
 686                                 fofs - dei.fofs >= F2FS_MIN_EXTENT_LEN)) {
 687                         en->ei.len = fofs - en->ei.fofs;
 688                         prev_en = en;
 689                         parts = 1;
 690                 }
 691
 692                 if (end < org_end && (type != EX_READ ||
 693                                 org_end - end >= F2FS_MIN_EXTENT_LEN)) {
 694                         if (parts) {
 695                                 __set_extent_info(&ei,
 696                                         end, org_end - end,
 697                                         end - dei.fofs + dei.blk, false,
 698                                         type);
 699                                 en1 = __insert_extent_tree(sbi, et, &ei,
 700                                                         NULL, NULL, true);
 701                                 next_en = en1;
 702                         } else {
 703                                 __set_extent_info(&en->ei,
 704                                         end, en->ei.len - (end - dei.fofs),
 705                                         en->ei.blk + (end - dei.fofs), true,
 706                                         type);
 707                                 next_en = en;
 708                         }
 709                         parts++;
 710                 }
 711
 712                 if (!next_en) {
 713                         struct rb_node *node = rb_next(&en->rb_node);
 714
 715                         next_en = rb_entry_safe(node, struct extent_node,
 716                                                 rb_node);
 717                 }
 718
 719                 if (parts)
 720                         __try_update_largest_extent(et, en);
 721                 else
 722                         __release_extent_node(sbi, et, en);
 723
 724                 /*
 725                  * if original extent is split into zero or two parts, extent
 726                  * tree has been altered by deletion or insertion, therefore
 727                  * invalidate pointers regard to tree.
 728                  */
 729                 if (parts != 1) {
 730                         insert_p = NULL;
 731                         insert_parent = NULL;
 732                 }
 733                 en = next_en;
 734         }
 735
 736         /* 3. update extent in read extent cache */
 737         BUG_ON(type != EX_READ);
 738
 739         if (tei->blk) {
 740                 __set_extent_info(&ei, fofs, len, tei->blk, false, EX_READ);
 741                 if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
 742                         __insert_extent_tree(sbi, et, &ei,
 743                                         insert_p, insert_parent, leftmost);
 744
 745                 /* give up extent_cache, if split and small updates happen */
 746                 if (dei.len >= 1 &&
 747                                 prev.len < F2FS_MIN_EXTENT_LEN &&
 748                                 et->largest.len < F2FS_MIN_EXTENT_LEN) {
 749                         et->largest.len = 0;
 750                         et->largest_updated = true;
 751                         set_inode_flag(inode, FI_NO_EXTENT);
 752                 }
 753         }
 754
 755         if (is_inode_flag_set(inode, FI_NO_EXTENT))
 756                 __free_extent_tree(sbi, et);
 757
 758         if (et->largest_updated) {
 759                 et->largest_updated = false;
 760                 updated = true;
 761         }
 762
 763         write_unlock(&et->lock);
 764
 765         if (updated)
 766                 f2fs_mark_inode_dirty_sync(inode, true);
 767 }
 768
 769 #ifdef CONFIG_F2FS_FS_COMPRESSION
 770 void f2fs_update_read_extent_tree_range_compressed(struct inode *inode,
 771                                 pgoff_t fofs, block_t blkaddr, unsigned int llen,
 772                                 unsigned int c_len)
 773 {
 774         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 775         struct extent_tree *et = F2FS_I(inode)->extent_tree[EX_READ];
 776         struct extent_node *en = NULL;
 777         struct extent_node *prev_en = NULL, *next_en = NULL;
 778         struct extent_info ei;
 779         struct rb_node **insert_p = NULL, *insert_parent = NULL;
 780         bool leftmost = false;
 781
 782         trace_f2fs_update_read_extent_tree_range(inode, fofs, llen,
 783                                                 blkaddr, c_len);
 784
 785         /* it is safe here to check FI_NO_EXTENT w/o et->lock in ro image */
 786         if (is_inode_flag_set(inode, FI_NO_EXTENT))
 787                 return;
 788
 789         write_lock(&et->lock);
 790
 791         en = (struct extent_node *)f2fs_lookup_rb_tree_ret(&et->root,
 792                                 (struct rb_entry *)et->cached_en, fofs,
 793                                 (struct rb_entry **)&prev_en,
 794                                 (struct rb_entry **)&next_en,
 795                                 &insert_p, &insert_parent, false,
 796                                 &leftmost);
 797         if (en)
 798                 goto unlock_out;
 799
 800         __set_extent_info(&ei, fofs, llen, blkaddr, true, EX_READ);
 801         ei.c_len = c_len;
 802
 803         if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
 804                 __insert_extent_tree(sbi, et, &ei,
 805                                 insert_p, insert_parent, leftmost);
 806 unlock_out:
 807         write_unlock(&et->lock);
 808 }
 809 #endif
 810
 811 static void __update_extent_cache(struct dnode_of_data *dn, enum extent_type type)
 812 {
 813         struct extent_info ei;
 814
 815         if (!__may_extent_tree(dn->inode, type))
 816                 return;
 817
 818         ei.fofs = f2fs_start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 819                                                                 dn->ofs_in_node;
 820         ei.len = 1;
 821
 822         if (type == EX_READ) {
 823                 if (dn->data_blkaddr == NEW_ADDR)
 824                         ei.blk = NULL_ADDR;
 825                 else
 826                         ei.blk = dn->data_blkaddr;
 827         }
 828         __update_extent_tree_range(dn->inode, &ei, type);
 829 }
 830
 831 static unsigned int __shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink,
 832                                         enum extent_type type)
 833 {
 834         struct extent_tree_info *eti = &sbi->extent_tree[type];
 835         struct extent_tree *et, *next;
 836         struct extent_node *en;
 837         unsigned int node_cnt = 0, tree_cnt = 0;
 838         int remained;
 839
 840         if (!atomic_read(&eti->total_zombie_tree))
 841                 goto free_node;
 842
 843         if (!mutex_trylock(&eti->extent_tree_lock))
 844                 goto out;
 845
 846         /* 1. remove unreferenced extent tree */
 847         list_for_each_entry_safe(et, next, &eti->zombie_list, list) {
 848                 if (atomic_read(&et->node_cnt)) {
 849                         write_lock(&et->lock);
 850                         node_cnt += __free_extent_tree(sbi, et);
 851                         write_unlock(&et->lock);
 852                 }
 853                 f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
 854                 list_del_init(&et->list);
 855                 radix_tree_delete(&eti->extent_tree_root, et->ino);
 856                 kmem_cache_free(extent_tree_slab, et);
 857                 atomic_dec(&eti->total_ext_tree);
 858                 atomic_dec(&eti->total_zombie_tree);
 859                 tree_cnt++;
 860
 861                 if (node_cnt + tree_cnt >= nr_shrink)
 862                         goto unlock_out;
 863                 cond_resched();
 864         }
 865         mutex_unlock(&eti->extent_tree_lock);
 866
 867 free_node:
 868         /* 2. remove LRU extent entries */
 869         if (!mutex_trylock(&eti->extent_tree_lock))
 870                 goto out;
 871
 872         remained = nr_shrink - (node_cnt + tree_cnt);
 873
 874         spin_lock(&eti->extent_lock);
 875         for (; remained > 0; remained--) {
 876                 if (list_empty(&eti->extent_list))
 877                         break;
 878                 en = list_first_entry(&eti->extent_list,
 879                                         struct extent_node, list);
 880                 et = en->et;
 881                 if (!write_trylock(&et->lock)) {
 882                         /* refresh this extent node's position in extent list */
 883                         list_move_tail(&en->list, &eti->extent_list);
 884                         continue;
 885                 }
 886
 887                 list_del_init(&en->list);
 888                 spin_unlock(&eti->extent_lock);
 889
 890                 __detach_extent_node(sbi, et, en);
 891
 892                 write_unlock(&et->lock);
 893                 node_cnt++;
 894                 spin_lock(&eti->extent_lock);
 895         }
 896         spin_unlock(&eti->extent_lock);
 897
 898 unlock_out:
 899         mutex_unlock(&eti->extent_tree_lock);
 900 out:
 901         trace_f2fs_shrink_extent_tree(sbi, node_cnt, tree_cnt, type);
 902
 903         return node_cnt + tree_cnt;
 904 }
 905
 906 /* read extent cache operations */
 907 bool f2fs_lookup_read_extent_cache(struct inode *inode, pgoff_t pgofs,
 908                                 struct extent_info *ei)
 909 {
 910         if (!__may_extent_tree(inode, EX_READ))
 911                 return false;
 912
 913         return __lookup_extent_tree(inode, pgofs, ei, EX_READ);
 914 }
 915
 916 void f2fs_update_read_extent_cache(struct dnode_of_data *dn)
 917 {
 918         return __update_extent_cache(dn, EX_READ);
 919 }
 920
 921 void f2fs_update_read_extent_cache_range(struct dnode_of_data *dn,
 922                                 pgoff_t fofs, block_t blkaddr, unsigned int len)
 923 {
 924         struct extent_info ei = {
 925                 .fofs = fofs,
 926                 .len = len,
 927                 .blk = blkaddr,
 928         };
 929
 930         if (!__may_extent_tree(dn->inode, EX_READ))
 931                 return;
 932
 933         __update_extent_tree_range(dn->inode, &ei, EX_READ);
 934 }
 935
 936 unsigned int f2fs_shrink_read_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
 937 {
 938         if (!test_opt(sbi, READ_EXTENT_CACHE))
 939                 return 0;
 940
 941         return __shrink_extent_tree(sbi, nr_shrink, EX_READ);
 942 }
 943
 944 static unsigned int __destroy_extent_node(struct inode *inode,
 945                                         enum extent_type type)
 946 {
 947         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 948         struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
 949         unsigned int node_cnt = 0;
 950
 951         if (!et || !atomic_read(&et->node_cnt))
 952                 return 0;
 953
 954         write_lock(&et->lock);
 955         node_cnt = __free_extent_tree(sbi, et);
 956         write_unlock(&et->lock);
 957
 958         return node_cnt;
 959 }
 960
 961 void f2fs_destroy_extent_node(struct inode *inode)
 962 {
 963         __destroy_extent_node(inode, EX_READ);
 964 }
 965
 966 static void __drop_extent_tree(struct inode *inode, enum extent_type type)
 967 {
 968         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 969         struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
 970         bool updated = false;
 971
 972         if (!__may_extent_tree(inode, type))
 973                 return;
 974
 975         write_lock(&et->lock);
 976         __free_extent_tree(sbi, et);
 977         if (type == EX_READ) {
 978                 set_inode_flag(inode, FI_NO_EXTENT);
 979                 if (et->largest.len) {
 980                         et->largest.len = 0;
 981                         updated = true;
 982                 }
 983         }
 984         write_unlock(&et->lock);
 985         if (updated)
 986                 f2fs_mark_inode_dirty_sync(inode, true);
 987 }
 988
 989 void f2fs_drop_extent_tree(struct inode *inode)
 990 {
 991         __drop_extent_tree(inode, EX_READ);
 992 }
 993
 994 static void __destroy_extent_tree(struct inode *inode, enum extent_type type)
 995 {
 996         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 997         struct extent_tree_info *eti = &sbi->extent_tree[type];
 998         struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
 999         unsigned int node_cnt = 0;
1000
1001         if (!et)
1002                 return;
1003
1004         if (inode->i_nlink && !is_bad_inode(inode) &&
1005                                         atomic_read(&et->node_cnt)) {
1006                 mutex_lock(&eti->extent_tree_lock);
1007                 list_add_tail(&et->list, &eti->zombie_list);
1008                 atomic_inc(&eti->total_zombie_tree);
1009                 mutex_unlock(&eti->extent_tree_lock);
1010                 return;
1011         }
1012
1013         /* free all extent info belong to this extent tree */
1014         node_cnt = __destroy_extent_node(inode, type);
1015
1016         /* delete extent tree entry in radix tree */
1017         mutex_lock(&eti->extent_tree_lock);
1018         f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
1019         radix_tree_delete(&eti->extent_tree_root, inode->i_ino);
1020         kmem_cache_free(extent_tree_slab, et);
1021         atomic_dec(&eti->total_ext_tree);
1022         mutex_unlock(&eti->extent_tree_lock);
1023
1024         F2FS_I(inode)->extent_tree[type] = NULL;
1025
1026         trace_f2fs_destroy_extent_tree(inode, node_cnt, type);
1027 }
1028
1029 void f2fs_destroy_extent_tree(struct inode *inode)
1030 {
1031         __destroy_extent_tree(inode, EX_READ);
1032 }
1033
1034 static void __init_extent_tree_info(struct extent_tree_info *eti)
1035 {
1036         INIT_RADIX_TREE(&eti->extent_tree_root, GFP_NOIO);
1037         mutex_init(&eti->extent_tree_lock);
1038         INIT_LIST_HEAD(&eti->extent_list);
1039         spin_lock_init(&eti->extent_lock);
1040         atomic_set(&eti->total_ext_tree, 0);
1041         INIT_LIST_HEAD(&eti->zombie_list);
1042         atomic_set(&eti->total_zombie_tree, 0);
1043         atomic_set(&eti->total_ext_node, 0);
1044 }
1045
1046 void f2fs_init_extent_cache_info(struct f2fs_sb_info *sbi)
1047 {
1048         __init_extent_tree_info(&sbi->extent_tree[EX_READ]);
1049 }
1050
1051 int __init f2fs_create_extent_cache(void)
1052 {
1053         extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree",
1054                         sizeof(struct extent_tree));
1055         if (!extent_tree_slab)
1056                 return -ENOMEM;
1057         extent_node_slab = f2fs_kmem_cache_create("f2fs_extent_node",
1058                         sizeof(struct extent_node));
1059         if (!extent_node_slab) {
1060                 kmem_cache_destroy(extent_tree_slab);
1061                 return -ENOMEM;
1062         }
1063         return 0;
1064 }
1065
1066 void f2fs_destroy_extent_cache(void)
1067 {
1068         kmem_cache_destroy(extent_node_slab);
1069         kmem_cache_destroy(extent_tree_slab);
1070 }