1 /* 2 * fs/f2fs/gc.c 3 * 4 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 5 * http://www.samsung.com/ 6 * 7 * This program is free software; you can redistribute it and/or modify 8 * it under the terms of the GNU General Public License version 2 as 9 * published by the Free Software Foundation. 10 */ 11 #include <linux/fs.h> 12 #include <linux/module.h> 13 #include <linux/backing-dev.h> 14 #include <linux/init.h> 15 #include <linux/f2fs_fs.h> 16 #include <linux/kthread.h> 17 #include <linux/delay.h> 18 #include <linux/freezer.h> 19 20 #include "f2fs.h" 21 #include "node.h" 22 #include "segment.h" 23 #include "gc.h" 24 #include <trace/events/f2fs.h> 25 26 static int gc_thread_func(void *data) 27 { 28 struct f2fs_sb_info *sbi = data; 29 struct f2fs_gc_kthread *gc_th = sbi->gc_thread; 30 wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head; 31 unsigned int wait_ms; 32 33 wait_ms = gc_th->min_sleep_time; 34 35 set_freezable(); 36 do { 37 wait_event_interruptible_timeout(*wq, 38 kthread_should_stop() || freezing(current) || 39 gc_th->gc_wake, 40 msecs_to_jiffies(wait_ms)); 41 42 /* give it a try one time */ 43 if (gc_th->gc_wake) 44 gc_th->gc_wake = 0; 45 46 if (try_to_freeze()) 47 continue; 48 if (kthread_should_stop()) 49 break; 50 51 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) { 52 increase_sleep_time(gc_th, &wait_ms); 53 continue; 54 } 55 56 if (time_to_inject(sbi, FAULT_CHECKPOINT)) { 57 f2fs_show_injection_info(FAULT_CHECKPOINT); 58 f2fs_stop_checkpoint(sbi, false); 59 } 60 61 if (!sb_start_write_trylock(sbi->sb)) 62 continue; 63 64 /* 65 * [GC triggering condition] 66 * 0. GC is not conducted currently. 67 * 1. There are enough dirty segments. 68 * 2. IO subsystem is idle by checking the # of writeback pages. 69 * 3. IO subsystem is idle by checking the # of requests in 70 * bdev's request list. 71 * 72 * Note) We have to avoid triggering GCs frequently. 73 * Because it is possible that some segments can be 74 * invalidated soon after by user update or deletion. 75 * So, I'd like to wait some time to collect dirty segments. 76 */ 77 if (sbi->gc_mode == GC_URGENT) { 78 wait_ms = gc_th->urgent_sleep_time; 79 mutex_lock(&sbi->gc_mutex); 80 goto do_gc; 81 } 82 83 if (!mutex_trylock(&sbi->gc_mutex)) 84 goto next; 85 86 if (!is_idle(sbi)) { 87 increase_sleep_time(gc_th, &wait_ms); 88 mutex_unlock(&sbi->gc_mutex); 89 goto next; 90 } 91 92 if (has_enough_invalid_blocks(sbi)) 93 decrease_sleep_time(gc_th, &wait_ms); 94 else 95 increase_sleep_time(gc_th, &wait_ms); 96 do_gc: 97 stat_inc_bggc_count(sbi); 98 99 /* if return value is not zero, no victim was selected */ 100 if (f2fs_gc(sbi, test_opt(sbi, FORCE_FG_GC), true, NULL_SEGNO)) 101 wait_ms = gc_th->no_gc_sleep_time; 102 103 trace_f2fs_background_gc(sbi->sb, wait_ms, 104 prefree_segments(sbi), free_segments(sbi)); 105 106 /* balancing f2fs's metadata periodically */ 107 f2fs_balance_fs_bg(sbi); 108 next: 109 sb_end_write(sbi->sb); 110 111 } while (!kthread_should_stop()); 112 return 0; 113 } 114 115 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi) 116 { 117 struct f2fs_gc_kthread *gc_th; 118 dev_t dev = sbi->sb->s_bdev->bd_dev; 119 int err = 0; 120 121 gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL); 122 if (!gc_th) { 123 err = -ENOMEM; 124 goto out; 125 } 126 127 gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME; 128 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME; 129 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME; 130 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME; 131 132 gc_th->gc_wake= 0; 133 134 sbi->gc_thread = gc_th; 135 init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); 136 sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, 137 "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); 138 if (IS_ERR(gc_th->f2fs_gc_task)) { 139 err = PTR_ERR(gc_th->f2fs_gc_task); 140 kfree(gc_th); 141 sbi->gc_thread = NULL; 142 } 143 out: 144 return err; 145 } 146 147 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi) 148 { 149 struct f2fs_gc_kthread *gc_th = sbi->gc_thread; 150 if (!gc_th) 151 return; 152 kthread_stop(gc_th->f2fs_gc_task); 153 kfree(gc_th); 154 sbi->gc_thread = NULL; 155 } 156 157 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type) 158 { 159 int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY; 160 161 switch (sbi->gc_mode) { 162 case GC_IDLE_CB: 163 gc_mode = GC_CB; 164 break; 165 case GC_IDLE_GREEDY: 166 case GC_URGENT: 167 gc_mode = GC_GREEDY; 168 break; 169 } 170 return gc_mode; 171 } 172 173 static void select_policy(struct f2fs_sb_info *sbi, int gc_type, 174 int type, struct victim_sel_policy *p) 175 { 176 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 177 178 if (p->alloc_mode == SSR) { 179 p->gc_mode = GC_GREEDY; 180 p->dirty_segmap = dirty_i->dirty_segmap[type]; 181 p->max_search = dirty_i->nr_dirty[type]; 182 p->ofs_unit = 1; 183 } else { 184 p->gc_mode = select_gc_type(sbi, gc_type); 185 p->dirty_segmap = dirty_i->dirty_segmap[DIRTY]; 186 p->max_search = dirty_i->nr_dirty[DIRTY]; 187 p->ofs_unit = sbi->segs_per_sec; 188 } 189 190 /* we need to check every dirty segments in the FG_GC case */ 191 if (gc_type != FG_GC && 192 (sbi->gc_mode != GC_URGENT) && 193 p->max_search > sbi->max_victim_search) 194 p->max_search = sbi->max_victim_search; 195 196 /* let's select beginning hot/small space first in no_heap mode*/ 197 if (test_opt(sbi, NOHEAP) && 198 (type == CURSEG_HOT_DATA || IS_NODESEG(type))) 199 p->offset = 0; 200 else 201 p->offset = SIT_I(sbi)->last_victim[p->gc_mode]; 202 } 203 204 static unsigned int get_max_cost(struct f2fs_sb_info *sbi, 205 struct victim_sel_policy *p) 206 { 207 /* SSR allocates in a segment unit */ 208 if (p->alloc_mode == SSR) 209 return sbi->blocks_per_seg; 210 if (p->gc_mode == GC_GREEDY) 211 return 2 * sbi->blocks_per_seg * p->ofs_unit; 212 else if (p->gc_mode == GC_CB) 213 return UINT_MAX; 214 else /* No other gc_mode */ 215 return 0; 216 } 217 218 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) 219 { 220 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 221 unsigned int secno; 222 223 /* 224 * If the gc_type is FG_GC, we can select victim segments 225 * selected by background GC before. 226 * Those segments guarantee they have small valid blocks. 227 */ 228 for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) { 229 if (sec_usage_check(sbi, secno)) 230 continue; 231 clear_bit(secno, dirty_i->victim_secmap); 232 return GET_SEG_FROM_SEC(sbi, secno); 233 } 234 return NULL_SEGNO; 235 } 236 237 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno) 238 { 239 struct sit_info *sit_i = SIT_I(sbi); 240 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 241 unsigned int start = GET_SEG_FROM_SEC(sbi, secno); 242 unsigned long long mtime = 0; 243 unsigned int vblocks; 244 unsigned char age = 0; 245 unsigned char u; 246 unsigned int i; 247 248 for (i = 0; i < sbi->segs_per_sec; i++) 249 mtime += get_seg_entry(sbi, start + i)->mtime; 250 vblocks = get_valid_blocks(sbi, segno, true); 251 252 mtime = div_u64(mtime, sbi->segs_per_sec); 253 vblocks = div_u64(vblocks, sbi->segs_per_sec); 254 255 u = (vblocks * 100) >> sbi->log_blocks_per_seg; 256 257 /* Handle if the system time has changed by the user */ 258 if (mtime < sit_i->min_mtime) 259 sit_i->min_mtime = mtime; 260 if (mtime > sit_i->max_mtime) 261 sit_i->max_mtime = mtime; 262 if (sit_i->max_mtime != sit_i->min_mtime) 263 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime), 264 sit_i->max_mtime - sit_i->min_mtime); 265 266 return UINT_MAX - ((100 * (100 - u) * age) / (100 + u)); 267 } 268 269 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi, 270 unsigned int segno, struct victim_sel_policy *p) 271 { 272 if (p->alloc_mode == SSR) 273 return get_seg_entry(sbi, segno)->ckpt_valid_blocks; 274 275 /* alloc_mode == LFS */ 276 if (p->gc_mode == GC_GREEDY) 277 return get_valid_blocks(sbi, segno, true); 278 else 279 return get_cb_cost(sbi, segno); 280 } 281 282 static unsigned int count_bits(const unsigned long *addr, 283 unsigned int offset, unsigned int len) 284 { 285 unsigned int end = offset + len, sum = 0; 286 287 while (offset < end) { 288 if (test_bit(offset++, addr)) 289 ++sum; 290 } 291 return sum; 292 } 293 294 /* 295 * This function is called from two paths. 296 * One is garbage collection and the other is SSR segment selection. 297 * When it is called during GC, it just gets a victim segment 298 * and it does not remove it from dirty seglist. 299 * When it is called from SSR segment selection, it finds a segment 300 * which has minimum valid blocks and removes it from dirty seglist. 301 */ 302 static int get_victim_by_default(struct f2fs_sb_info *sbi, 303 unsigned int *result, int gc_type, int type, char alloc_mode) 304 { 305 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 306 struct sit_info *sm = SIT_I(sbi); 307 struct victim_sel_policy p; 308 unsigned int secno, last_victim; 309 unsigned int last_segment = MAIN_SEGS(sbi); 310 unsigned int nsearched = 0; 311 312 mutex_lock(&dirty_i->seglist_lock); 313 314 p.alloc_mode = alloc_mode; 315 select_policy(sbi, gc_type, type, &p); 316 317 p.min_segno = NULL_SEGNO; 318 p.min_cost = get_max_cost(sbi, &p); 319 320 if (*result != NULL_SEGNO) { 321 if (IS_DATASEG(get_seg_entry(sbi, *result)->type) && 322 get_valid_blocks(sbi, *result, false) && 323 !sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result))) 324 p.min_segno = *result; 325 goto out; 326 } 327 328 if (p.max_search == 0) 329 goto out; 330 331 last_victim = sm->last_victim[p.gc_mode]; 332 if (p.alloc_mode == LFS && gc_type == FG_GC) { 333 p.min_segno = check_bg_victims(sbi); 334 if (p.min_segno != NULL_SEGNO) 335 goto got_it; 336 } 337 338 while (1) { 339 unsigned long cost; 340 unsigned int segno; 341 342 segno = find_next_bit(p.dirty_segmap, last_segment, p.offset); 343 if (segno >= last_segment) { 344 if (sm->last_victim[p.gc_mode]) { 345 last_segment = 346 sm->last_victim[p.gc_mode]; 347 sm->last_victim[p.gc_mode] = 0; 348 p.offset = 0; 349 continue; 350 } 351 break; 352 } 353 354 p.offset = segno + p.ofs_unit; 355 if (p.ofs_unit > 1) { 356 p.offset -= segno % p.ofs_unit; 357 nsearched += count_bits(p.dirty_segmap, 358 p.offset - p.ofs_unit, 359 p.ofs_unit); 360 } else { 361 nsearched++; 362 } 363 364 secno = GET_SEC_FROM_SEG(sbi, segno); 365 366 if (sec_usage_check(sbi, secno)) 367 goto next; 368 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) 369 goto next; 370 371 cost = get_gc_cost(sbi, segno, &p); 372 373 if (p.min_cost > cost) { 374 p.min_segno = segno; 375 p.min_cost = cost; 376 } 377 next: 378 if (nsearched >= p.max_search) { 379 if (!sm->last_victim[p.gc_mode] && segno <= last_victim) 380 sm->last_victim[p.gc_mode] = last_victim + 1; 381 else 382 sm->last_victim[p.gc_mode] = segno + 1; 383 sm->last_victim[p.gc_mode] %= MAIN_SEGS(sbi); 384 break; 385 } 386 } 387 if (p.min_segno != NULL_SEGNO) { 388 got_it: 389 if (p.alloc_mode == LFS) { 390 secno = GET_SEC_FROM_SEG(sbi, p.min_segno); 391 if (gc_type == FG_GC) 392 sbi->cur_victim_sec = secno; 393 else 394 set_bit(secno, dirty_i->victim_secmap); 395 } 396 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; 397 398 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, 399 sbi->cur_victim_sec, 400 prefree_segments(sbi), free_segments(sbi)); 401 } 402 out: 403 mutex_unlock(&dirty_i->seglist_lock); 404 405 return (p.min_segno == NULL_SEGNO) ? 0 : 1; 406 } 407 408 static const struct victim_selection default_v_ops = { 409 .get_victim = get_victim_by_default, 410 }; 411 412 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino) 413 { 414 struct inode_entry *ie; 415 416 ie = radix_tree_lookup(&gc_list->iroot, ino); 417 if (ie) 418 return ie->inode; 419 return NULL; 420 } 421 422 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode) 423 { 424 struct inode_entry *new_ie; 425 426 if (inode == find_gc_inode(gc_list, inode->i_ino)) { 427 iput(inode); 428 return; 429 } 430 new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, GFP_NOFS); 431 new_ie->inode = inode; 432 433 f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie); 434 list_add_tail(&new_ie->list, &gc_list->ilist); 435 } 436 437 static void put_gc_inode(struct gc_inode_list *gc_list) 438 { 439 struct inode_entry *ie, *next_ie; 440 list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) { 441 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino); 442 iput(ie->inode); 443 list_del(&ie->list); 444 kmem_cache_free(f2fs_inode_entry_slab, ie); 445 } 446 } 447 448 static int check_valid_map(struct f2fs_sb_info *sbi, 449 unsigned int segno, int offset) 450 { 451 struct sit_info *sit_i = SIT_I(sbi); 452 struct seg_entry *sentry; 453 int ret; 454 455 down_read(&sit_i->sentry_lock); 456 sentry = get_seg_entry(sbi, segno); 457 ret = f2fs_test_bit(offset, sentry->cur_valid_map); 458 up_read(&sit_i->sentry_lock); 459 return ret; 460 } 461 462 /* 463 * This function compares node address got in summary with that in NAT. 464 * On validity, copy that node with cold status, otherwise (invalid node) 465 * ignore that. 466 */ 467 static void gc_node_segment(struct f2fs_sb_info *sbi, 468 struct f2fs_summary *sum, unsigned int segno, int gc_type) 469 { 470 struct f2fs_summary *entry; 471 block_t start_addr; 472 int off; 473 int phase = 0; 474 bool fggc = (gc_type == FG_GC); 475 476 start_addr = START_BLOCK(sbi, segno); 477 478 next_step: 479 entry = sum; 480 481 if (fggc && phase == 2) 482 atomic_inc(&sbi->wb_sync_req[NODE]); 483 484 for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { 485 nid_t nid = le32_to_cpu(entry->nid); 486 struct page *node_page; 487 struct node_info ni; 488 489 /* stop BG_GC if there is not enough free sections. */ 490 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) 491 return; 492 493 if (check_valid_map(sbi, segno, off) == 0) 494 continue; 495 496 if (phase == 0) { 497 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, 498 META_NAT, true); 499 continue; 500 } 501 502 if (phase == 1) { 503 f2fs_ra_node_page(sbi, nid); 504 continue; 505 } 506 507 /* phase == 2 */ 508 node_page = f2fs_get_node_page(sbi, nid); 509 if (IS_ERR(node_page)) 510 continue; 511 512 /* block may become invalid during f2fs_get_node_page */ 513 if (check_valid_map(sbi, segno, off) == 0) { 514 f2fs_put_page(node_page, 1); 515 continue; 516 } 517 518 if (f2fs_get_node_info(sbi, nid, &ni)) { 519 f2fs_put_page(node_page, 1); 520 continue; 521 } 522 523 if (ni.blk_addr != start_addr + off) { 524 f2fs_put_page(node_page, 1); 525 continue; 526 } 527 528 f2fs_move_node_page(node_page, gc_type); 529 stat_inc_node_blk_count(sbi, 1, gc_type); 530 } 531 532 if (++phase < 3) 533 goto next_step; 534 535 if (fggc) 536 atomic_dec(&sbi->wb_sync_req[NODE]); 537 } 538 539 /* 540 * Calculate start block index indicating the given node offset. 541 * Be careful, caller should give this node offset only indicating direct node 542 * blocks. If any node offsets, which point the other types of node blocks such 543 * as indirect or double indirect node blocks, are given, it must be a caller's 544 * bug. 545 */ 546 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode) 547 { 548 unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; 549 unsigned int bidx; 550 551 if (node_ofs == 0) 552 return 0; 553 554 if (node_ofs <= 2) { 555 bidx = node_ofs - 1; 556 } else if (node_ofs <= indirect_blks) { 557 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); 558 bidx = node_ofs - 2 - dec; 559 } else { 560 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); 561 bidx = node_ofs - 5 - dec; 562 } 563 return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(inode); 564 } 565 566 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 567 struct node_info *dni, block_t blkaddr, unsigned int *nofs) 568 { 569 struct page *node_page; 570 nid_t nid; 571 unsigned int ofs_in_node; 572 block_t source_blkaddr; 573 574 nid = le32_to_cpu(sum->nid); 575 ofs_in_node = le16_to_cpu(sum->ofs_in_node); 576 577 node_page = f2fs_get_node_page(sbi, nid); 578 if (IS_ERR(node_page)) 579 return false; 580 581 if (f2fs_get_node_info(sbi, nid, dni)) { 582 f2fs_put_page(node_page, 1); 583 return false; 584 } 585 586 if (sum->version != dni->version) { 587 f2fs_msg(sbi->sb, KERN_WARNING, 588 "%s: valid data with mismatched node version.", 589 __func__); 590 set_sbi_flag(sbi, SBI_NEED_FSCK); 591 } 592 593 *nofs = ofs_of_node(node_page); 594 source_blkaddr = datablock_addr(NULL, node_page, ofs_in_node); 595 f2fs_put_page(node_page, 1); 596 597 if (source_blkaddr != blkaddr) 598 return false; 599 return true; 600 } 601 602 static int ra_data_block(struct inode *inode, pgoff_t index) 603 { 604 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 605 struct address_space *mapping = inode->i_mapping; 606 struct dnode_of_data dn; 607 struct page *page; 608 struct extent_info ei = {0, 0, 0}; 609 struct f2fs_io_info fio = { 610 .sbi = sbi, 611 .ino = inode->i_ino, 612 .type = DATA, 613 .temp = COLD, 614 .op = REQ_OP_READ, 615 .op_flags = 0, 616 .encrypted_page = NULL, 617 .in_list = false, 618 .retry = false, 619 }; 620 int err; 621 622 page = f2fs_grab_cache_page(mapping, index, true); 623 if (!page) 624 return -ENOMEM; 625 626 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 627 dn.data_blkaddr = ei.blk + index - ei.fofs; 628 goto got_it; 629 } 630 631 set_new_dnode(&dn, inode, NULL, NULL, 0); 632 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 633 if (err) 634 goto put_page; 635 f2fs_put_dnode(&dn); 636 637 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr, 638 DATA_GENERIC))) { 639 err = -EFAULT; 640 goto put_page; 641 } 642 got_it: 643 /* read page */ 644 fio.page = page; 645 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; 646 647 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi), 648 dn.data_blkaddr, 649 FGP_LOCK | FGP_CREAT, GFP_NOFS); 650 if (!fio.encrypted_page) { 651 err = -ENOMEM; 652 goto put_page; 653 } 654 655 err = f2fs_submit_page_bio(&fio); 656 if (err) 657 goto put_encrypted_page; 658 f2fs_put_page(fio.encrypted_page, 0); 659 f2fs_put_page(page, 1); 660 return 0; 661 put_encrypted_page: 662 f2fs_put_page(fio.encrypted_page, 1); 663 put_page: 664 f2fs_put_page(page, 1); 665 return err; 666 } 667 668 /* 669 * Move data block via META_MAPPING while keeping locked data page. 670 * This can be used to move blocks, aka LBAs, directly on disk. 671 */ 672 static void move_data_block(struct inode *inode, block_t bidx, 673 int gc_type, unsigned int segno, int off) 674 { 675 struct f2fs_io_info fio = { 676 .sbi = F2FS_I_SB(inode), 677 .ino = inode->i_ino, 678 .type = DATA, 679 .temp = COLD, 680 .op = REQ_OP_READ, 681 .op_flags = 0, 682 .encrypted_page = NULL, 683 .in_list = false, 684 .retry = false, 685 }; 686 struct dnode_of_data dn; 687 struct f2fs_summary sum; 688 struct node_info ni; 689 struct page *page, *mpage; 690 block_t newaddr; 691 int err; 692 bool lfs_mode = test_opt(fio.sbi, LFS); 693 694 /* do not read out */ 695 page = f2fs_grab_cache_page(inode->i_mapping, bidx, false); 696 if (!page) 697 return; 698 699 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) 700 goto out; 701 702 if (f2fs_is_atomic_file(inode)) { 703 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++; 704 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++; 705 goto out; 706 } 707 708 if (f2fs_is_pinned_file(inode)) { 709 f2fs_pin_file_control(inode, true); 710 goto out; 711 } 712 713 set_new_dnode(&dn, inode, NULL, NULL, 0); 714 err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE); 715 if (err) 716 goto out; 717 718 if (unlikely(dn.data_blkaddr == NULL_ADDR)) { 719 ClearPageUptodate(page); 720 goto put_out; 721 } 722 723 /* 724 * don't cache encrypted data into meta inode until previous dirty 725 * data were writebacked to avoid racing between GC and flush. 726 */ 727 f2fs_wait_on_page_writeback(page, DATA, true); 728 729 err = f2fs_get_node_info(fio.sbi, dn.nid, &ni); 730 if (err) 731 goto put_out; 732 733 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); 734 735 /* read page */ 736 fio.page = page; 737 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; 738 739 if (lfs_mode) 740 down_write(&fio.sbi->io_order_lock); 741 742 f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr, 743 &sum, CURSEG_COLD_DATA, NULL, false); 744 745 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi), 746 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS); 747 if (!fio.encrypted_page) { 748 err = -ENOMEM; 749 goto recover_block; 750 } 751 752 mpage = f2fs_pagecache_get_page(META_MAPPING(fio.sbi), 753 fio.old_blkaddr, FGP_LOCK, GFP_NOFS); 754 if (mpage) { 755 bool updated = false; 756 757 if (PageUptodate(mpage)) { 758 memcpy(page_address(fio.encrypted_page), 759 page_address(mpage), PAGE_SIZE); 760 updated = true; 761 } 762 f2fs_put_page(mpage, 1); 763 invalidate_mapping_pages(META_MAPPING(fio.sbi), 764 fio.old_blkaddr, fio.old_blkaddr); 765 if (updated) 766 goto write_page; 767 } 768 769 err = f2fs_submit_page_bio(&fio); 770 if (err) 771 goto put_page_out; 772 773 /* write page */ 774 lock_page(fio.encrypted_page); 775 776 if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) { 777 err = -EIO; 778 goto put_page_out; 779 } 780 if (unlikely(!PageUptodate(fio.encrypted_page))) { 781 err = -EIO; 782 goto put_page_out; 783 } 784 785 write_page: 786 set_page_dirty(fio.encrypted_page); 787 f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true); 788 if (clear_page_dirty_for_io(fio.encrypted_page)) 789 dec_page_count(fio.sbi, F2FS_DIRTY_META); 790 791 set_page_writeback(fio.encrypted_page); 792 ClearPageError(page); 793 794 /* allocate block address */ 795 f2fs_wait_on_page_writeback(dn.node_page, NODE, true); 796 797 fio.op = REQ_OP_WRITE; 798 fio.op_flags = REQ_SYNC; 799 fio.new_blkaddr = newaddr; 800 f2fs_submit_page_write(&fio); 801 if (fio.retry) { 802 if (PageWriteback(fio.encrypted_page)) 803 end_page_writeback(fio.encrypted_page); 804 goto put_page_out; 805 } 806 807 f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE); 808 809 f2fs_update_data_blkaddr(&dn, newaddr); 810 set_inode_flag(inode, FI_APPEND_WRITE); 811 if (page->index == 0) 812 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN); 813 put_page_out: 814 f2fs_put_page(fio.encrypted_page, 1); 815 recover_block: 816 if (lfs_mode) 817 up_write(&fio.sbi->io_order_lock); 818 if (err) 819 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr, 820 true, true); 821 put_out: 822 f2fs_put_dnode(&dn); 823 out: 824 f2fs_put_page(page, 1); 825 } 826 827 static void move_data_page(struct inode *inode, block_t bidx, int gc_type, 828 unsigned int segno, int off) 829 { 830 struct page *page; 831 832 page = f2fs_get_lock_data_page(inode, bidx, true); 833 if (IS_ERR(page)) 834 return; 835 836 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) 837 goto out; 838 839 if (f2fs_is_atomic_file(inode)) { 840 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++; 841 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++; 842 goto out; 843 } 844 if (f2fs_is_pinned_file(inode)) { 845 if (gc_type == FG_GC) 846 f2fs_pin_file_control(inode, true); 847 goto out; 848 } 849 850 if (gc_type == BG_GC) { 851 if (PageWriteback(page)) 852 goto out; 853 set_page_dirty(page); 854 set_cold_data(page); 855 } else { 856 struct f2fs_io_info fio = { 857 .sbi = F2FS_I_SB(inode), 858 .ino = inode->i_ino, 859 .type = DATA, 860 .temp = COLD, 861 .op = REQ_OP_WRITE, 862 .op_flags = REQ_SYNC, 863 .old_blkaddr = NULL_ADDR, 864 .page = page, 865 .encrypted_page = NULL, 866 .need_lock = LOCK_REQ, 867 .io_type = FS_GC_DATA_IO, 868 }; 869 bool is_dirty = PageDirty(page); 870 int err; 871 872 retry: 873 set_page_dirty(page); 874 f2fs_wait_on_page_writeback(page, DATA, true); 875 if (clear_page_dirty_for_io(page)) { 876 inode_dec_dirty_pages(inode); 877 f2fs_remove_dirty_inode(inode); 878 } 879 880 set_cold_data(page); 881 882 err = f2fs_do_write_data_page(&fio); 883 if (err) { 884 clear_cold_data(page); 885 if (err == -ENOMEM) { 886 congestion_wait(BLK_RW_ASYNC, HZ/50); 887 goto retry; 888 } 889 if (is_dirty) 890 set_page_dirty(page); 891 } 892 } 893 out: 894 f2fs_put_page(page, 1); 895 } 896 897 /* 898 * This function tries to get parent node of victim data block, and identifies 899 * data block validity. If the block is valid, copy that with cold status and 900 * modify parent node. 901 * If the parent node is not valid or the data block address is different, 902 * the victim data block is ignored. 903 */ 904 static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 905 struct gc_inode_list *gc_list, unsigned int segno, int gc_type) 906 { 907 struct super_block *sb = sbi->sb; 908 struct f2fs_summary *entry; 909 block_t start_addr; 910 int off; 911 int phase = 0; 912 913 start_addr = START_BLOCK(sbi, segno); 914 915 next_step: 916 entry = sum; 917 918 for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { 919 struct page *data_page; 920 struct inode *inode; 921 struct node_info dni; /* dnode info for the data */ 922 unsigned int ofs_in_node, nofs; 923 block_t start_bidx; 924 nid_t nid = le32_to_cpu(entry->nid); 925 926 /* stop BG_GC if there is not enough free sections. */ 927 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) 928 return; 929 930 if (check_valid_map(sbi, segno, off) == 0) 931 continue; 932 933 if (phase == 0) { 934 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, 935 META_NAT, true); 936 continue; 937 } 938 939 if (phase == 1) { 940 f2fs_ra_node_page(sbi, nid); 941 continue; 942 } 943 944 /* Get an inode by ino with checking validity */ 945 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs)) 946 continue; 947 948 if (phase == 2) { 949 f2fs_ra_node_page(sbi, dni.ino); 950 continue; 951 } 952 953 ofs_in_node = le16_to_cpu(entry->ofs_in_node); 954 955 if (phase == 3) { 956 inode = f2fs_iget(sb, dni.ino); 957 if (IS_ERR(inode) || is_bad_inode(inode)) 958 continue; 959 960 if (!down_write_trylock( 961 &F2FS_I(inode)->i_gc_rwsem[WRITE])) { 962 iput(inode); 963 sbi->skipped_gc_rwsem++; 964 continue; 965 } 966 967 start_bidx = f2fs_start_bidx_of_node(nofs, inode) + 968 ofs_in_node; 969 970 if (f2fs_post_read_required(inode)) { 971 int err = ra_data_block(inode, start_bidx); 972 973 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 974 if (err) { 975 iput(inode); 976 continue; 977 } 978 add_gc_inode(gc_list, inode); 979 continue; 980 } 981 982 data_page = f2fs_get_read_data_page(inode, 983 start_bidx, REQ_RAHEAD, true); 984 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 985 if (IS_ERR(data_page)) { 986 iput(inode); 987 continue; 988 } 989 990 f2fs_put_page(data_page, 0); 991 add_gc_inode(gc_list, inode); 992 continue; 993 } 994 995 /* phase 4 */ 996 inode = find_gc_inode(gc_list, dni.ino); 997 if (inode) { 998 struct f2fs_inode_info *fi = F2FS_I(inode); 999 bool locked = false; 1000 1001 if (S_ISREG(inode->i_mode)) { 1002 if (!down_write_trylock(&fi->i_gc_rwsem[READ])) 1003 continue; 1004 if (!down_write_trylock( 1005 &fi->i_gc_rwsem[WRITE])) { 1006 sbi->skipped_gc_rwsem++; 1007 up_write(&fi->i_gc_rwsem[READ]); 1008 continue; 1009 } 1010 locked = true; 1011 1012 /* wait for all inflight aio data */ 1013 inode_dio_wait(inode); 1014 } 1015 1016 start_bidx = f2fs_start_bidx_of_node(nofs, inode) 1017 + ofs_in_node; 1018 if (f2fs_post_read_required(inode)) 1019 move_data_block(inode, start_bidx, gc_type, 1020 segno, off); 1021 else 1022 move_data_page(inode, start_bidx, gc_type, 1023 segno, off); 1024 1025 if (locked) { 1026 up_write(&fi->i_gc_rwsem[WRITE]); 1027 up_write(&fi->i_gc_rwsem[READ]); 1028 } 1029 1030 stat_inc_data_blk_count(sbi, 1, gc_type); 1031 } 1032 } 1033 1034 if (++phase < 5) 1035 goto next_step; 1036 } 1037 1038 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, 1039 int gc_type) 1040 { 1041 struct sit_info *sit_i = SIT_I(sbi); 1042 int ret; 1043 1044 down_write(&sit_i->sentry_lock); 1045 ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, 1046 NO_CHECK_TYPE, LFS); 1047 up_write(&sit_i->sentry_lock); 1048 return ret; 1049 } 1050 1051 static int do_garbage_collect(struct f2fs_sb_info *sbi, 1052 unsigned int start_segno, 1053 struct gc_inode_list *gc_list, int gc_type) 1054 { 1055 struct page *sum_page; 1056 struct f2fs_summary_block *sum; 1057 struct blk_plug plug; 1058 unsigned int segno = start_segno; 1059 unsigned int end_segno = start_segno + sbi->segs_per_sec; 1060 int seg_freed = 0; 1061 unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ? 1062 SUM_TYPE_DATA : SUM_TYPE_NODE; 1063 1064 /* readahead multi ssa blocks those have contiguous address */ 1065 if (sbi->segs_per_sec > 1) 1066 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), 1067 sbi->segs_per_sec, META_SSA, true); 1068 1069 /* reference all summary page */ 1070 while (segno < end_segno) { 1071 sum_page = f2fs_get_sum_page(sbi, segno++); 1072 unlock_page(sum_page); 1073 } 1074 1075 blk_start_plug(&plug); 1076 1077 for (segno = start_segno; segno < end_segno; segno++) { 1078 1079 /* find segment summary of victim */ 1080 sum_page = find_get_page(META_MAPPING(sbi), 1081 GET_SUM_BLOCK(sbi, segno)); 1082 f2fs_put_page(sum_page, 0); 1083 1084 if (get_valid_blocks(sbi, segno, false) == 0 || 1085 !PageUptodate(sum_page) || 1086 unlikely(f2fs_cp_error(sbi))) 1087 goto next; 1088 1089 sum = page_address(sum_page); 1090 if (type != GET_SUM_TYPE((&sum->footer))) { 1091 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent segment (%u) " 1092 "type [%d, %d] in SSA and SIT", 1093 segno, type, GET_SUM_TYPE((&sum->footer))); 1094 set_sbi_flag(sbi, SBI_NEED_FSCK); 1095 goto next; 1096 } 1097 1098 /* 1099 * this is to avoid deadlock: 1100 * - lock_page(sum_page) - f2fs_replace_block 1101 * - check_valid_map() - down_write(sentry_lock) 1102 * - down_read(sentry_lock) - change_curseg() 1103 * - lock_page(sum_page) 1104 */ 1105 if (type == SUM_TYPE_NODE) 1106 gc_node_segment(sbi, sum->entries, segno, gc_type); 1107 else 1108 gc_data_segment(sbi, sum->entries, gc_list, segno, 1109 gc_type); 1110 1111 stat_inc_seg_count(sbi, type, gc_type); 1112 1113 if (gc_type == FG_GC && 1114 get_valid_blocks(sbi, segno, false) == 0) 1115 seg_freed++; 1116 next: 1117 f2fs_put_page(sum_page, 0); 1118 } 1119 1120 if (gc_type == FG_GC) 1121 f2fs_submit_merged_write(sbi, 1122 (type == SUM_TYPE_NODE) ? NODE : DATA); 1123 1124 blk_finish_plug(&plug); 1125 1126 stat_inc_call_count(sbi->stat_info); 1127 1128 return seg_freed; 1129 } 1130 1131 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync, 1132 bool background, unsigned int segno) 1133 { 1134 int gc_type = sync ? FG_GC : BG_GC; 1135 int sec_freed = 0, seg_freed = 0, total_freed = 0; 1136 int ret = 0; 1137 struct cp_control cpc; 1138 unsigned int init_segno = segno; 1139 struct gc_inode_list gc_list = { 1140 .ilist = LIST_HEAD_INIT(gc_list.ilist), 1141 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS), 1142 }; 1143 unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC]; 1144 unsigned long long first_skipped; 1145 unsigned int skipped_round = 0, round = 0; 1146 1147 trace_f2fs_gc_begin(sbi->sb, sync, background, 1148 get_pages(sbi, F2FS_DIRTY_NODES), 1149 get_pages(sbi, F2FS_DIRTY_DENTS), 1150 get_pages(sbi, F2FS_DIRTY_IMETA), 1151 free_sections(sbi), 1152 free_segments(sbi), 1153 reserved_segments(sbi), 1154 prefree_segments(sbi)); 1155 1156 cpc.reason = __get_cp_reason(sbi); 1157 sbi->skipped_gc_rwsem = 0; 1158 first_skipped = last_skipped; 1159 gc_more: 1160 if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) { 1161 ret = -EINVAL; 1162 goto stop; 1163 } 1164 if (unlikely(f2fs_cp_error(sbi))) { 1165 ret = -EIO; 1166 goto stop; 1167 } 1168 1169 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) { 1170 /* 1171 * For example, if there are many prefree_segments below given 1172 * threshold, we can make them free by checkpoint. Then, we 1173 * secure free segments which doesn't need fggc any more. 1174 */ 1175 if (prefree_segments(sbi)) { 1176 ret = f2fs_write_checkpoint(sbi, &cpc); 1177 if (ret) 1178 goto stop; 1179 } 1180 if (has_not_enough_free_secs(sbi, 0, 0)) 1181 gc_type = FG_GC; 1182 } 1183 1184 /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */ 1185 if (gc_type == BG_GC && !background) { 1186 ret = -EINVAL; 1187 goto stop; 1188 } 1189 if (!__get_victim(sbi, &segno, gc_type)) { 1190 ret = -ENODATA; 1191 goto stop; 1192 } 1193 1194 seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type); 1195 if (gc_type == FG_GC && seg_freed == sbi->segs_per_sec) 1196 sec_freed++; 1197 total_freed += seg_freed; 1198 1199 if (gc_type == FG_GC) { 1200 if (sbi->skipped_atomic_files[FG_GC] > last_skipped || 1201 sbi->skipped_gc_rwsem) 1202 skipped_round++; 1203 last_skipped = sbi->skipped_atomic_files[FG_GC]; 1204 round++; 1205 } 1206 1207 if (gc_type == FG_GC) 1208 sbi->cur_victim_sec = NULL_SEGNO; 1209 1210 if (sync) 1211 goto stop; 1212 1213 if (has_not_enough_free_secs(sbi, sec_freed, 0)) { 1214 if (skipped_round <= MAX_SKIP_GC_COUNT || 1215 skipped_round * 2 < round) { 1216 segno = NULL_SEGNO; 1217 goto gc_more; 1218 } 1219 1220 if (first_skipped < last_skipped && 1221 (last_skipped - first_skipped) > 1222 sbi->skipped_gc_rwsem) { 1223 f2fs_drop_inmem_pages_all(sbi, true); 1224 segno = NULL_SEGNO; 1225 goto gc_more; 1226 } 1227 if (gc_type == FG_GC) 1228 ret = f2fs_write_checkpoint(sbi, &cpc); 1229 } 1230 stop: 1231 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0; 1232 SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno; 1233 1234 trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed, 1235 get_pages(sbi, F2FS_DIRTY_NODES), 1236 get_pages(sbi, F2FS_DIRTY_DENTS), 1237 get_pages(sbi, F2FS_DIRTY_IMETA), 1238 free_sections(sbi), 1239 free_segments(sbi), 1240 reserved_segments(sbi), 1241 prefree_segments(sbi)); 1242 1243 mutex_unlock(&sbi->gc_mutex); 1244 1245 put_gc_inode(&gc_list); 1246 1247 if (sync) 1248 ret = sec_freed ? 0 : -EAGAIN; 1249 return ret; 1250 } 1251 1252 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi) 1253 { 1254 DIRTY_I(sbi)->v_ops = &default_v_ops; 1255 1256 sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES; 1257 1258 /* give warm/cold data area from slower device */ 1259 if (sbi->s_ndevs && sbi->segs_per_sec == 1) 1260 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 1261 GET_SEGNO(sbi, FDEV(0).end_blk) + 1; 1262 } 1263