1 /* 2 * fs/f2fs/segment.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/f2fs_fs.h> 13 #include <linux/bio.h> 14 #include <linux/blkdev.h> 15 #include <linux/prefetch.h> 16 #include <linux/kthread.h> 17 #include <linux/swap.h> 18 #include <linux/timer.h> 19 20 #include "f2fs.h" 21 #include "segment.h" 22 #include "node.h" 23 #include "trace.h" 24 #include <trace/events/f2fs.h> 25 26 #define __reverse_ffz(x) __reverse_ffs(~(x)) 27 28 static struct kmem_cache *discard_entry_slab; 29 static struct kmem_cache *discard_cmd_slab; 30 static struct kmem_cache *sit_entry_set_slab; 31 static struct kmem_cache *inmem_entry_slab; 32 33 static unsigned long __reverse_ulong(unsigned char *str) 34 { 35 unsigned long tmp = 0; 36 int shift = 24, idx = 0; 37 38 #if BITS_PER_LONG == 64 39 shift = 56; 40 #endif 41 while (shift >= 0) { 42 tmp |= (unsigned long)str[idx++] << shift; 43 shift -= BITS_PER_BYTE; 44 } 45 return tmp; 46 } 47 48 /* 49 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since 50 * MSB and LSB are reversed in a byte by f2fs_set_bit. 51 */ 52 static inline unsigned long __reverse_ffs(unsigned long word) 53 { 54 int num = 0; 55 56 #if BITS_PER_LONG == 64 57 if ((word & 0xffffffff00000000UL) == 0) 58 num += 32; 59 else 60 word >>= 32; 61 #endif 62 if ((word & 0xffff0000) == 0) 63 num += 16; 64 else 65 word >>= 16; 66 67 if ((word & 0xff00) == 0) 68 num += 8; 69 else 70 word >>= 8; 71 72 if ((word & 0xf0) == 0) 73 num += 4; 74 else 75 word >>= 4; 76 77 if ((word & 0xc) == 0) 78 num += 2; 79 else 80 word >>= 2; 81 82 if ((word & 0x2) == 0) 83 num += 1; 84 return num; 85 } 86 87 /* 88 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because 89 * f2fs_set_bit makes MSB and LSB reversed in a byte. 90 * @size must be integral times of unsigned long. 91 * Example: 92 * MSB <--> LSB 93 * f2fs_set_bit(0, bitmap) => 1000 0000 94 * f2fs_set_bit(7, bitmap) => 0000 0001 95 */ 96 static unsigned long __find_rev_next_bit(const unsigned long *addr, 97 unsigned long size, unsigned long offset) 98 { 99 const unsigned long *p = addr + BIT_WORD(offset); 100 unsigned long result = size; 101 unsigned long tmp; 102 103 if (offset >= size) 104 return size; 105 106 size -= (offset & ~(BITS_PER_LONG - 1)); 107 offset %= BITS_PER_LONG; 108 109 while (1) { 110 if (*p == 0) 111 goto pass; 112 113 tmp = __reverse_ulong((unsigned char *)p); 114 115 tmp &= ~0UL >> offset; 116 if (size < BITS_PER_LONG) 117 tmp &= (~0UL << (BITS_PER_LONG - size)); 118 if (tmp) 119 goto found; 120 pass: 121 if (size <= BITS_PER_LONG) 122 break; 123 size -= BITS_PER_LONG; 124 offset = 0; 125 p++; 126 } 127 return result; 128 found: 129 return result - size + __reverse_ffs(tmp); 130 } 131 132 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr, 133 unsigned long size, unsigned long offset) 134 { 135 const unsigned long *p = addr + BIT_WORD(offset); 136 unsigned long result = size; 137 unsigned long tmp; 138 139 if (offset >= size) 140 return size; 141 142 size -= (offset & ~(BITS_PER_LONG - 1)); 143 offset %= BITS_PER_LONG; 144 145 while (1) { 146 if (*p == ~0UL) 147 goto pass; 148 149 tmp = __reverse_ulong((unsigned char *)p); 150 151 if (offset) 152 tmp |= ~0UL << (BITS_PER_LONG - offset); 153 if (size < BITS_PER_LONG) 154 tmp |= ~0UL >> size; 155 if (tmp != ~0UL) 156 goto found; 157 pass: 158 if (size <= BITS_PER_LONG) 159 break; 160 size -= BITS_PER_LONG; 161 offset = 0; 162 p++; 163 } 164 return result; 165 found: 166 return result - size + __reverse_ffz(tmp); 167 } 168 169 void register_inmem_page(struct inode *inode, struct page *page) 170 { 171 struct f2fs_inode_info *fi = F2FS_I(inode); 172 struct inmem_pages *new; 173 174 f2fs_trace_pid(page); 175 176 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE); 177 SetPagePrivate(page); 178 179 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS); 180 181 /* add atomic page indices to the list */ 182 new->page = page; 183 INIT_LIST_HEAD(&new->list); 184 185 /* increase reference count with clean state */ 186 mutex_lock(&fi->inmem_lock); 187 get_page(page); 188 list_add_tail(&new->list, &fi->inmem_pages); 189 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); 190 mutex_unlock(&fi->inmem_lock); 191 192 trace_f2fs_register_inmem_page(page, INMEM); 193 } 194 195 static int __revoke_inmem_pages(struct inode *inode, 196 struct list_head *head, bool drop, bool recover) 197 { 198 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 199 struct inmem_pages *cur, *tmp; 200 int err = 0; 201 202 list_for_each_entry_safe(cur, tmp, head, list) { 203 struct page *page = cur->page; 204 205 if (drop) 206 trace_f2fs_commit_inmem_page(page, INMEM_DROP); 207 208 lock_page(page); 209 210 if (recover) { 211 struct dnode_of_data dn; 212 struct node_info ni; 213 214 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE); 215 216 set_new_dnode(&dn, inode, NULL, NULL, 0); 217 if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) { 218 err = -EAGAIN; 219 goto next; 220 } 221 get_node_info(sbi, dn.nid, &ni); 222 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, 223 cur->old_addr, ni.version, true, true); 224 f2fs_put_dnode(&dn); 225 } 226 next: 227 /* we don't need to invalidate this in the sccessful status */ 228 if (drop || recover) 229 ClearPageUptodate(page); 230 set_page_private(page, 0); 231 ClearPagePrivate(page); 232 f2fs_put_page(page, 1); 233 234 list_del(&cur->list); 235 kmem_cache_free(inmem_entry_slab, cur); 236 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); 237 } 238 return err; 239 } 240 241 void drop_inmem_pages(struct inode *inode) 242 { 243 struct f2fs_inode_info *fi = F2FS_I(inode); 244 245 mutex_lock(&fi->inmem_lock); 246 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false); 247 mutex_unlock(&fi->inmem_lock); 248 249 clear_inode_flag(inode, FI_ATOMIC_FILE); 250 stat_dec_atomic_write(inode); 251 } 252 253 static int __commit_inmem_pages(struct inode *inode, 254 struct list_head *revoke_list) 255 { 256 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 257 struct f2fs_inode_info *fi = F2FS_I(inode); 258 struct inmem_pages *cur, *tmp; 259 struct f2fs_io_info fio = { 260 .sbi = sbi, 261 .type = DATA, 262 .op = REQ_OP_WRITE, 263 .op_flags = REQ_SYNC | REQ_PRIO, 264 .encrypted_page = NULL, 265 }; 266 pgoff_t last_idx = ULONG_MAX; 267 int err = 0; 268 269 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) { 270 struct page *page = cur->page; 271 272 lock_page(page); 273 if (page->mapping == inode->i_mapping) { 274 trace_f2fs_commit_inmem_page(page, INMEM); 275 276 set_page_dirty(page); 277 f2fs_wait_on_page_writeback(page, DATA, true); 278 if (clear_page_dirty_for_io(page)) { 279 inode_dec_dirty_pages(inode); 280 remove_dirty_inode(inode); 281 } 282 283 fio.page = page; 284 err = do_write_data_page(&fio); 285 if (err) { 286 unlock_page(page); 287 break; 288 } 289 290 /* record old blkaddr for revoking */ 291 cur->old_addr = fio.old_blkaddr; 292 last_idx = page->index; 293 } 294 unlock_page(page); 295 list_move_tail(&cur->list, revoke_list); 296 } 297 298 if (last_idx != ULONG_MAX) 299 f2fs_submit_merged_bio_cond(sbi, inode, 0, last_idx, 300 DATA, WRITE); 301 302 if (!err) 303 __revoke_inmem_pages(inode, revoke_list, false, false); 304 305 return err; 306 } 307 308 int commit_inmem_pages(struct inode *inode) 309 { 310 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 311 struct f2fs_inode_info *fi = F2FS_I(inode); 312 struct list_head revoke_list; 313 int err; 314 315 INIT_LIST_HEAD(&revoke_list); 316 f2fs_balance_fs(sbi, true); 317 f2fs_lock_op(sbi); 318 319 set_inode_flag(inode, FI_ATOMIC_COMMIT); 320 321 mutex_lock(&fi->inmem_lock); 322 err = __commit_inmem_pages(inode, &revoke_list); 323 if (err) { 324 int ret; 325 /* 326 * try to revoke all committed pages, but still we could fail 327 * due to no memory or other reason, if that happened, EAGAIN 328 * will be returned, which means in such case, transaction is 329 * already not integrity, caller should use journal to do the 330 * recovery or rewrite & commit last transaction. For other 331 * error number, revoking was done by filesystem itself. 332 */ 333 ret = __revoke_inmem_pages(inode, &revoke_list, false, true); 334 if (ret) 335 err = ret; 336 337 /* drop all uncommitted pages */ 338 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false); 339 } 340 mutex_unlock(&fi->inmem_lock); 341 342 clear_inode_flag(inode, FI_ATOMIC_COMMIT); 343 344 f2fs_unlock_op(sbi); 345 return err; 346 } 347 348 /* 349 * This function balances dirty node and dentry pages. 350 * In addition, it controls garbage collection. 351 */ 352 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need) 353 { 354 #ifdef CONFIG_F2FS_FAULT_INJECTION 355 if (time_to_inject(sbi, FAULT_CHECKPOINT)) 356 f2fs_stop_checkpoint(sbi, false); 357 #endif 358 359 if (!need) 360 return; 361 362 /* balance_fs_bg is able to be pending */ 363 if (excess_cached_nats(sbi)) 364 f2fs_balance_fs_bg(sbi); 365 366 /* 367 * We should do GC or end up with checkpoint, if there are so many dirty 368 * dir/node pages without enough free segments. 369 */ 370 if (has_not_enough_free_secs(sbi, 0, 0)) { 371 mutex_lock(&sbi->gc_mutex); 372 f2fs_gc(sbi, false, false); 373 } 374 } 375 376 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi) 377 { 378 /* try to shrink extent cache when there is no enough memory */ 379 if (!available_free_memory(sbi, EXTENT_CACHE)) 380 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER); 381 382 /* check the # of cached NAT entries */ 383 if (!available_free_memory(sbi, NAT_ENTRIES)) 384 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK); 385 386 if (!available_free_memory(sbi, FREE_NIDS)) 387 try_to_free_nids(sbi, MAX_FREE_NIDS); 388 else 389 build_free_nids(sbi, false, false); 390 391 if (!is_idle(sbi)) 392 return; 393 394 /* checkpoint is the only way to shrink partial cached entries */ 395 if (!available_free_memory(sbi, NAT_ENTRIES) || 396 !available_free_memory(sbi, INO_ENTRIES) || 397 excess_prefree_segs(sbi) || 398 excess_dirty_nats(sbi) || 399 f2fs_time_over(sbi, CP_TIME)) { 400 if (test_opt(sbi, DATA_FLUSH)) { 401 struct blk_plug plug; 402 403 blk_start_plug(&plug); 404 sync_dirty_inodes(sbi, FILE_INODE); 405 blk_finish_plug(&plug); 406 } 407 f2fs_sync_fs(sbi->sb, true); 408 stat_inc_bg_cp_count(sbi->stat_info); 409 } 410 } 411 412 static int __submit_flush_wait(struct block_device *bdev) 413 { 414 struct bio *bio = f2fs_bio_alloc(0); 415 int ret; 416 417 bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH; 418 bio->bi_bdev = bdev; 419 ret = submit_bio_wait(bio); 420 bio_put(bio); 421 return ret; 422 } 423 424 static int submit_flush_wait(struct f2fs_sb_info *sbi) 425 { 426 int ret = __submit_flush_wait(sbi->sb->s_bdev); 427 int i; 428 429 if (sbi->s_ndevs && !ret) { 430 for (i = 1; i < sbi->s_ndevs; i++) { 431 trace_f2fs_issue_flush(FDEV(i).bdev, 432 test_opt(sbi, NOBARRIER), 433 test_opt(sbi, FLUSH_MERGE)); 434 ret = __submit_flush_wait(FDEV(i).bdev); 435 if (ret) 436 break; 437 } 438 } 439 return ret; 440 } 441 442 static int issue_flush_thread(void *data) 443 { 444 struct f2fs_sb_info *sbi = data; 445 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 446 wait_queue_head_t *q = &fcc->flush_wait_queue; 447 repeat: 448 if (kthread_should_stop()) 449 return 0; 450 451 if (!llist_empty(&fcc->issue_list)) { 452 struct flush_cmd *cmd, *next; 453 int ret; 454 455 fcc->dispatch_list = llist_del_all(&fcc->issue_list); 456 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); 457 458 ret = submit_flush_wait(sbi); 459 llist_for_each_entry_safe(cmd, next, 460 fcc->dispatch_list, llnode) { 461 cmd->ret = ret; 462 complete(&cmd->wait); 463 } 464 fcc->dispatch_list = NULL; 465 } 466 467 wait_event_interruptible(*q, 468 kthread_should_stop() || !llist_empty(&fcc->issue_list)); 469 goto repeat; 470 } 471 472 int f2fs_issue_flush(struct f2fs_sb_info *sbi) 473 { 474 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 475 struct flush_cmd cmd; 476 477 if (test_opt(sbi, NOBARRIER)) 478 return 0; 479 480 if (!test_opt(sbi, FLUSH_MERGE)) 481 return submit_flush_wait(sbi); 482 483 if (!atomic_read(&fcc->submit_flush)) { 484 int ret; 485 486 atomic_inc(&fcc->submit_flush); 487 ret = submit_flush_wait(sbi); 488 atomic_dec(&fcc->submit_flush); 489 return ret; 490 } 491 492 init_completion(&cmd.wait); 493 494 atomic_inc(&fcc->submit_flush); 495 llist_add(&cmd.llnode, &fcc->issue_list); 496 497 if (!fcc->dispatch_list) 498 wake_up(&fcc->flush_wait_queue); 499 500 if (fcc->f2fs_issue_flush) { 501 wait_for_completion(&cmd.wait); 502 atomic_dec(&fcc->submit_flush); 503 } else { 504 llist_del_all(&fcc->issue_list); 505 atomic_set(&fcc->submit_flush, 0); 506 } 507 508 return cmd.ret; 509 } 510 511 int create_flush_cmd_control(struct f2fs_sb_info *sbi) 512 { 513 dev_t dev = sbi->sb->s_bdev->bd_dev; 514 struct flush_cmd_control *fcc; 515 int err = 0; 516 517 if (SM_I(sbi)->fcc_info) { 518 fcc = SM_I(sbi)->fcc_info; 519 goto init_thread; 520 } 521 522 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL); 523 if (!fcc) 524 return -ENOMEM; 525 atomic_set(&fcc->submit_flush, 0); 526 init_waitqueue_head(&fcc->flush_wait_queue); 527 init_llist_head(&fcc->issue_list); 528 SM_I(sbi)->fcc_info = fcc; 529 init_thread: 530 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, 531 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); 532 if (IS_ERR(fcc->f2fs_issue_flush)) { 533 err = PTR_ERR(fcc->f2fs_issue_flush); 534 kfree(fcc); 535 SM_I(sbi)->fcc_info = NULL; 536 return err; 537 } 538 539 return err; 540 } 541 542 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free) 543 { 544 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 545 546 if (fcc && fcc->f2fs_issue_flush) { 547 struct task_struct *flush_thread = fcc->f2fs_issue_flush; 548 549 fcc->f2fs_issue_flush = NULL; 550 kthread_stop(flush_thread); 551 } 552 if (free) { 553 kfree(fcc); 554 SM_I(sbi)->fcc_info = NULL; 555 } 556 } 557 558 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 559 enum dirty_type dirty_type) 560 { 561 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 562 563 /* need not be added */ 564 if (IS_CURSEG(sbi, segno)) 565 return; 566 567 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 568 dirty_i->nr_dirty[dirty_type]++; 569 570 if (dirty_type == DIRTY) { 571 struct seg_entry *sentry = get_seg_entry(sbi, segno); 572 enum dirty_type t = sentry->type; 573 574 if (unlikely(t >= DIRTY)) { 575 f2fs_bug_on(sbi, 1); 576 return; 577 } 578 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) 579 dirty_i->nr_dirty[t]++; 580 } 581 } 582 583 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 584 enum dirty_type dirty_type) 585 { 586 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 587 588 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) 589 dirty_i->nr_dirty[dirty_type]--; 590 591 if (dirty_type == DIRTY) { 592 struct seg_entry *sentry = get_seg_entry(sbi, segno); 593 enum dirty_type t = sentry->type; 594 595 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) 596 dirty_i->nr_dirty[t]--; 597 598 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0) 599 clear_bit(GET_SECNO(sbi, segno), 600 dirty_i->victim_secmap); 601 } 602 } 603 604 /* 605 * Should not occur error such as -ENOMEM. 606 * Adding dirty entry into seglist is not critical operation. 607 * If a given segment is one of current working segments, it won't be added. 608 */ 609 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) 610 { 611 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 612 unsigned short valid_blocks; 613 614 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) 615 return; 616 617 mutex_lock(&dirty_i->seglist_lock); 618 619 valid_blocks = get_valid_blocks(sbi, segno, 0); 620 621 if (valid_blocks == 0) { 622 __locate_dirty_segment(sbi, segno, PRE); 623 __remove_dirty_segment(sbi, segno, DIRTY); 624 } else if (valid_blocks < sbi->blocks_per_seg) { 625 __locate_dirty_segment(sbi, segno, DIRTY); 626 } else { 627 /* Recovery routine with SSR needs this */ 628 __remove_dirty_segment(sbi, segno, DIRTY); 629 } 630 631 mutex_unlock(&dirty_i->seglist_lock); 632 } 633 634 static void __add_discard_cmd(struct f2fs_sb_info *sbi, 635 struct bio *bio, block_t lstart, block_t len) 636 { 637 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 638 struct list_head *cmd_list = &(dcc->discard_cmd_list); 639 struct discard_cmd *dc; 640 641 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS); 642 INIT_LIST_HEAD(&dc->list); 643 dc->bio = bio; 644 bio->bi_private = dc; 645 dc->lstart = lstart; 646 dc->len = len; 647 dc->state = D_PREP; 648 init_completion(&dc->wait); 649 650 mutex_lock(&dcc->cmd_lock); 651 list_add_tail(&dc->list, cmd_list); 652 mutex_unlock(&dcc->cmd_lock); 653 } 654 655 static void __remove_discard_cmd(struct f2fs_sb_info *sbi, struct discard_cmd *dc) 656 { 657 int err = dc->bio->bi_error; 658 659 if (dc->state == D_DONE) 660 atomic_dec(&(SM_I(sbi)->dcc_info->submit_discard)); 661 662 if (err == -EOPNOTSUPP) 663 err = 0; 664 665 if (err) 666 f2fs_msg(sbi->sb, KERN_INFO, 667 "Issue discard failed, ret: %d", err); 668 bio_put(dc->bio); 669 list_del(&dc->list); 670 kmem_cache_free(discard_cmd_slab, dc); 671 } 672 673 /* This should be covered by global mutex, &sit_i->sentry_lock */ 674 void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr) 675 { 676 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 677 struct list_head *wait_list = &(dcc->discard_cmd_list); 678 struct discard_cmd *dc, *tmp; 679 struct blk_plug plug; 680 681 mutex_lock(&dcc->cmd_lock); 682 683 blk_start_plug(&plug); 684 685 list_for_each_entry_safe(dc, tmp, wait_list, list) { 686 687 if (blkaddr == NULL_ADDR) { 688 if (dc->state == D_PREP) { 689 dc->state = D_SUBMIT; 690 submit_bio(dc->bio); 691 atomic_inc(&dcc->submit_discard); 692 } 693 continue; 694 } 695 696 if (dc->lstart <= blkaddr && blkaddr < dc->lstart + dc->len) { 697 if (dc->state == D_SUBMIT) 698 wait_for_completion_io(&dc->wait); 699 else 700 __remove_discard_cmd(sbi, dc); 701 } 702 } 703 blk_finish_plug(&plug); 704 705 /* this comes from f2fs_put_super */ 706 if (blkaddr == NULL_ADDR) { 707 list_for_each_entry_safe(dc, tmp, wait_list, list) { 708 wait_for_completion_io(&dc->wait); 709 __remove_discard_cmd(sbi, dc); 710 } 711 } 712 mutex_unlock(&dcc->cmd_lock); 713 } 714 715 static void f2fs_submit_discard_endio(struct bio *bio) 716 { 717 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private; 718 719 complete(&dc->wait); 720 dc->state = D_DONE; 721 } 722 723 static int issue_discard_thread(void *data) 724 { 725 struct f2fs_sb_info *sbi = data; 726 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 727 wait_queue_head_t *q = &dcc->discard_wait_queue; 728 struct list_head *cmd_list = &dcc->discard_cmd_list; 729 struct discard_cmd *dc, *tmp; 730 struct blk_plug plug; 731 int iter = 0; 732 repeat: 733 if (kthread_should_stop()) 734 return 0; 735 736 blk_start_plug(&plug); 737 738 mutex_lock(&dcc->cmd_lock); 739 list_for_each_entry_safe(dc, tmp, cmd_list, list) { 740 if (dc->state == D_PREP) { 741 dc->state = D_SUBMIT; 742 submit_bio(dc->bio); 743 atomic_inc(&dcc->submit_discard); 744 if (iter++ > DISCARD_ISSUE_RATE) 745 break; 746 } else if (dc->state == D_DONE) { 747 __remove_discard_cmd(sbi, dc); 748 } 749 } 750 mutex_unlock(&dcc->cmd_lock); 751 752 blk_finish_plug(&plug); 753 754 iter = 0; 755 congestion_wait(BLK_RW_SYNC, HZ/50); 756 757 wait_event_interruptible(*q, 758 kthread_should_stop() || !list_empty(&dcc->discard_cmd_list)); 759 goto repeat; 760 } 761 762 763 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */ 764 static int __f2fs_issue_discard_async(struct f2fs_sb_info *sbi, 765 struct block_device *bdev, block_t blkstart, block_t blklen) 766 { 767 struct bio *bio = NULL; 768 block_t lblkstart = blkstart; 769 int err; 770 771 trace_f2fs_issue_discard(bdev, blkstart, blklen); 772 773 if (sbi->s_ndevs) { 774 int devi = f2fs_target_device_index(sbi, blkstart); 775 776 blkstart -= FDEV(devi).start_blk; 777 } 778 err = __blkdev_issue_discard(bdev, 779 SECTOR_FROM_BLOCK(blkstart), 780 SECTOR_FROM_BLOCK(blklen), 781 GFP_NOFS, 0, &bio); 782 if (!err && bio) { 783 bio->bi_end_io = f2fs_submit_discard_endio; 784 bio->bi_opf |= REQ_SYNC; 785 786 __add_discard_cmd(sbi, bio, lblkstart, blklen); 787 wake_up(&SM_I(sbi)->dcc_info->discard_wait_queue); 788 } 789 return err; 790 } 791 792 #ifdef CONFIG_BLK_DEV_ZONED 793 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi, 794 struct block_device *bdev, block_t blkstart, block_t blklen) 795 { 796 sector_t sector, nr_sects; 797 int devi = 0; 798 799 if (sbi->s_ndevs) { 800 devi = f2fs_target_device_index(sbi, blkstart); 801 blkstart -= FDEV(devi).start_blk; 802 } 803 804 /* 805 * We need to know the type of the zone: for conventional zones, 806 * use regular discard if the drive supports it. For sequential 807 * zones, reset the zone write pointer. 808 */ 809 switch (get_blkz_type(sbi, bdev, blkstart)) { 810 811 case BLK_ZONE_TYPE_CONVENTIONAL: 812 if (!blk_queue_discard(bdev_get_queue(bdev))) 813 return 0; 814 return __f2fs_issue_discard_async(sbi, bdev, blkstart, blklen); 815 case BLK_ZONE_TYPE_SEQWRITE_REQ: 816 case BLK_ZONE_TYPE_SEQWRITE_PREF: 817 sector = SECTOR_FROM_BLOCK(blkstart); 818 nr_sects = SECTOR_FROM_BLOCK(blklen); 819 820 if (sector & (bdev_zone_sectors(bdev) - 1) || 821 nr_sects != bdev_zone_sectors(bdev)) { 822 f2fs_msg(sbi->sb, KERN_INFO, 823 "(%d) %s: Unaligned discard attempted (block %x + %x)", 824 devi, sbi->s_ndevs ? FDEV(devi).path: "", 825 blkstart, blklen); 826 return -EIO; 827 } 828 trace_f2fs_issue_reset_zone(bdev, blkstart); 829 return blkdev_reset_zones(bdev, sector, 830 nr_sects, GFP_NOFS); 831 default: 832 /* Unknown zone type: broken device ? */ 833 return -EIO; 834 } 835 } 836 #endif 837 838 static int __issue_discard_async(struct f2fs_sb_info *sbi, 839 struct block_device *bdev, block_t blkstart, block_t blklen) 840 { 841 #ifdef CONFIG_BLK_DEV_ZONED 842 if (f2fs_sb_mounted_blkzoned(sbi->sb) && 843 bdev_zoned_model(bdev) != BLK_ZONED_NONE) 844 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen); 845 #endif 846 return __f2fs_issue_discard_async(sbi, bdev, blkstart, blklen); 847 } 848 849 static int f2fs_issue_discard(struct f2fs_sb_info *sbi, 850 block_t blkstart, block_t blklen) 851 { 852 sector_t start = blkstart, len = 0; 853 struct block_device *bdev; 854 struct seg_entry *se; 855 unsigned int offset; 856 block_t i; 857 int err = 0; 858 859 bdev = f2fs_target_device(sbi, blkstart, NULL); 860 861 for (i = blkstart; i < blkstart + blklen; i++, len++) { 862 if (i != start) { 863 struct block_device *bdev2 = 864 f2fs_target_device(sbi, i, NULL); 865 866 if (bdev2 != bdev) { 867 err = __issue_discard_async(sbi, bdev, 868 start, len); 869 if (err) 870 return err; 871 bdev = bdev2; 872 start = i; 873 len = 0; 874 } 875 } 876 877 se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); 878 offset = GET_BLKOFF_FROM_SEG0(sbi, i); 879 880 if (!f2fs_test_and_set_bit(offset, se->discard_map)) 881 sbi->discard_blks--; 882 } 883 884 if (len) 885 err = __issue_discard_async(sbi, bdev, start, len); 886 return err; 887 } 888 889 static void __add_discard_entry(struct f2fs_sb_info *sbi, 890 struct cp_control *cpc, struct seg_entry *se, 891 unsigned int start, unsigned int end) 892 { 893 struct list_head *head = &SM_I(sbi)->dcc_info->discard_entry_list; 894 struct discard_entry *new, *last; 895 896 if (!list_empty(head)) { 897 last = list_last_entry(head, struct discard_entry, list); 898 if (START_BLOCK(sbi, cpc->trim_start) + start == 899 last->blkaddr + last->len && 900 last->len < MAX_DISCARD_BLOCKS(sbi)) { 901 last->len += end - start; 902 goto done; 903 } 904 } 905 906 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS); 907 INIT_LIST_HEAD(&new->list); 908 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start; 909 new->len = end - start; 910 list_add_tail(&new->list, head); 911 done: 912 SM_I(sbi)->dcc_info->nr_discards += end - start; 913 } 914 915 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc, 916 bool check_only) 917 { 918 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 919 int max_blocks = sbi->blocks_per_seg; 920 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); 921 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 922 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 923 unsigned long *discard_map = (unsigned long *)se->discard_map; 924 unsigned long *dmap = SIT_I(sbi)->tmp_map; 925 unsigned int start = 0, end = -1; 926 bool force = (cpc->reason == CP_DISCARD); 927 int i; 928 929 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi)) 930 return false; 931 932 if (!force) { 933 if (!test_opt(sbi, DISCARD) || !se->valid_blocks || 934 SM_I(sbi)->dcc_info->nr_discards >= 935 SM_I(sbi)->dcc_info->max_discards) 936 return false; 937 } 938 939 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ 940 for (i = 0; i < entries; i++) 941 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : 942 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; 943 944 while (force || SM_I(sbi)->dcc_info->nr_discards <= 945 SM_I(sbi)->dcc_info->max_discards) { 946 start = __find_rev_next_bit(dmap, max_blocks, end + 1); 947 if (start >= max_blocks) 948 break; 949 950 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); 951 if (force && start && end != max_blocks 952 && (end - start) < cpc->trim_minlen) 953 continue; 954 955 if (check_only) 956 return true; 957 958 __add_discard_entry(sbi, cpc, se, start, end); 959 } 960 return false; 961 } 962 963 void release_discard_addrs(struct f2fs_sb_info *sbi) 964 { 965 struct list_head *head = &(SM_I(sbi)->dcc_info->discard_entry_list); 966 struct discard_entry *entry, *this; 967 968 /* drop caches */ 969 list_for_each_entry_safe(entry, this, head, list) { 970 list_del(&entry->list); 971 kmem_cache_free(discard_entry_slab, entry); 972 } 973 } 974 975 /* 976 * Should call clear_prefree_segments after checkpoint is done. 977 */ 978 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 979 { 980 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 981 unsigned int segno; 982 983 mutex_lock(&dirty_i->seglist_lock); 984 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) 985 __set_test_and_free(sbi, segno); 986 mutex_unlock(&dirty_i->seglist_lock); 987 } 988 989 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc) 990 { 991 struct list_head *head = &(SM_I(sbi)->dcc_info->discard_entry_list); 992 struct discard_entry *entry, *this; 993 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 994 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; 995 unsigned int start = 0, end = -1; 996 unsigned int secno, start_segno; 997 bool force = (cpc->reason == CP_DISCARD); 998 999 mutex_lock(&dirty_i->seglist_lock); 1000 1001 while (1) { 1002 int i; 1003 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); 1004 if (start >= MAIN_SEGS(sbi)) 1005 break; 1006 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), 1007 start + 1); 1008 1009 for (i = start; i < end; i++) 1010 clear_bit(i, prefree_map); 1011 1012 dirty_i->nr_dirty[PRE] -= end - start; 1013 1014 if (!test_opt(sbi, DISCARD)) 1015 continue; 1016 1017 if (force && start >= cpc->trim_start && 1018 (end - 1) <= cpc->trim_end) 1019 continue; 1020 1021 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) { 1022 f2fs_issue_discard(sbi, START_BLOCK(sbi, start), 1023 (end - start) << sbi->log_blocks_per_seg); 1024 continue; 1025 } 1026 next: 1027 secno = GET_SECNO(sbi, start); 1028 start_segno = secno * sbi->segs_per_sec; 1029 if (!IS_CURSEC(sbi, secno) && 1030 !get_valid_blocks(sbi, start, sbi->segs_per_sec)) 1031 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno), 1032 sbi->segs_per_sec << sbi->log_blocks_per_seg); 1033 1034 start = start_segno + sbi->segs_per_sec; 1035 if (start < end) 1036 goto next; 1037 } 1038 mutex_unlock(&dirty_i->seglist_lock); 1039 1040 /* send small discards */ 1041 list_for_each_entry_safe(entry, this, head, list) { 1042 if (force && entry->len < cpc->trim_minlen) 1043 goto skip; 1044 f2fs_issue_discard(sbi, entry->blkaddr, entry->len); 1045 cpc->trimmed += entry->len; 1046 skip: 1047 list_del(&entry->list); 1048 SM_I(sbi)->dcc_info->nr_discards -= entry->len; 1049 kmem_cache_free(discard_entry_slab, entry); 1050 } 1051 } 1052 1053 static int create_discard_cmd_control(struct f2fs_sb_info *sbi) 1054 { 1055 dev_t dev = sbi->sb->s_bdev->bd_dev; 1056 struct discard_cmd_control *dcc; 1057 int err = 0; 1058 1059 if (SM_I(sbi)->dcc_info) { 1060 dcc = SM_I(sbi)->dcc_info; 1061 goto init_thread; 1062 } 1063 1064 dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL); 1065 if (!dcc) 1066 return -ENOMEM; 1067 1068 INIT_LIST_HEAD(&dcc->discard_entry_list); 1069 INIT_LIST_HEAD(&dcc->discard_cmd_list); 1070 mutex_init(&dcc->cmd_lock); 1071 atomic_set(&dcc->submit_discard, 0); 1072 dcc->nr_discards = 0; 1073 dcc->max_discards = 0; 1074 1075 init_waitqueue_head(&dcc->discard_wait_queue); 1076 SM_I(sbi)->dcc_info = dcc; 1077 init_thread: 1078 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi, 1079 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev)); 1080 if (IS_ERR(dcc->f2fs_issue_discard)) { 1081 err = PTR_ERR(dcc->f2fs_issue_discard); 1082 kfree(dcc); 1083 SM_I(sbi)->dcc_info = NULL; 1084 return err; 1085 } 1086 1087 return err; 1088 } 1089 1090 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi, bool free) 1091 { 1092 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1093 1094 if (dcc && dcc->f2fs_issue_discard) { 1095 struct task_struct *discard_thread = dcc->f2fs_issue_discard; 1096 1097 dcc->f2fs_issue_discard = NULL; 1098 kthread_stop(discard_thread); 1099 } 1100 if (free) { 1101 kfree(dcc); 1102 SM_I(sbi)->dcc_info = NULL; 1103 } 1104 } 1105 1106 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 1107 { 1108 struct sit_info *sit_i = SIT_I(sbi); 1109 1110 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { 1111 sit_i->dirty_sentries++; 1112 return false; 1113 } 1114 1115 return true; 1116 } 1117 1118 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 1119 unsigned int segno, int modified) 1120 { 1121 struct seg_entry *se = get_seg_entry(sbi, segno); 1122 se->type = type; 1123 if (modified) 1124 __mark_sit_entry_dirty(sbi, segno); 1125 } 1126 1127 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 1128 { 1129 struct seg_entry *se; 1130 unsigned int segno, offset; 1131 long int new_vblocks; 1132 1133 segno = GET_SEGNO(sbi, blkaddr); 1134 1135 se = get_seg_entry(sbi, segno); 1136 new_vblocks = se->valid_blocks + del; 1137 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 1138 1139 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) || 1140 (new_vblocks > sbi->blocks_per_seg))); 1141 1142 se->valid_blocks = new_vblocks; 1143 se->mtime = get_mtime(sbi); 1144 SIT_I(sbi)->max_mtime = se->mtime; 1145 1146 /* Update valid block bitmap */ 1147 if (del > 0) { 1148 if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) { 1149 #ifdef CONFIG_F2FS_CHECK_FS 1150 if (f2fs_test_and_set_bit(offset, 1151 se->cur_valid_map_mir)) 1152 f2fs_bug_on(sbi, 1); 1153 else 1154 WARN_ON(1); 1155 #else 1156 f2fs_bug_on(sbi, 1); 1157 #endif 1158 } 1159 if (f2fs_discard_en(sbi) && 1160 !f2fs_test_and_set_bit(offset, se->discard_map)) 1161 sbi->discard_blks--; 1162 } else { 1163 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) { 1164 #ifdef CONFIG_F2FS_CHECK_FS 1165 if (!f2fs_test_and_clear_bit(offset, 1166 se->cur_valid_map_mir)) 1167 f2fs_bug_on(sbi, 1); 1168 else 1169 WARN_ON(1); 1170 #else 1171 f2fs_bug_on(sbi, 1); 1172 #endif 1173 } 1174 if (f2fs_discard_en(sbi) && 1175 f2fs_test_and_clear_bit(offset, se->discard_map)) 1176 sbi->discard_blks++; 1177 } 1178 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) 1179 se->ckpt_valid_blocks += del; 1180 1181 __mark_sit_entry_dirty(sbi, segno); 1182 1183 /* update total number of valid blocks to be written in ckpt area */ 1184 SIT_I(sbi)->written_valid_blocks += del; 1185 1186 if (sbi->segs_per_sec > 1) 1187 get_sec_entry(sbi, segno)->valid_blocks += del; 1188 } 1189 1190 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new) 1191 { 1192 update_sit_entry(sbi, new, 1); 1193 if (GET_SEGNO(sbi, old) != NULL_SEGNO) 1194 update_sit_entry(sbi, old, -1); 1195 1196 locate_dirty_segment(sbi, GET_SEGNO(sbi, old)); 1197 locate_dirty_segment(sbi, GET_SEGNO(sbi, new)); 1198 } 1199 1200 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) 1201 { 1202 unsigned int segno = GET_SEGNO(sbi, addr); 1203 struct sit_info *sit_i = SIT_I(sbi); 1204 1205 f2fs_bug_on(sbi, addr == NULL_ADDR); 1206 if (addr == NEW_ADDR) 1207 return; 1208 1209 /* add it into sit main buffer */ 1210 mutex_lock(&sit_i->sentry_lock); 1211 1212 update_sit_entry(sbi, addr, -1); 1213 1214 /* add it into dirty seglist */ 1215 locate_dirty_segment(sbi, segno); 1216 1217 mutex_unlock(&sit_i->sentry_lock); 1218 } 1219 1220 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr) 1221 { 1222 struct sit_info *sit_i = SIT_I(sbi); 1223 unsigned int segno, offset; 1224 struct seg_entry *se; 1225 bool is_cp = false; 1226 1227 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) 1228 return true; 1229 1230 mutex_lock(&sit_i->sentry_lock); 1231 1232 segno = GET_SEGNO(sbi, blkaddr); 1233 se = get_seg_entry(sbi, segno); 1234 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 1235 1236 if (f2fs_test_bit(offset, se->ckpt_valid_map)) 1237 is_cp = true; 1238 1239 mutex_unlock(&sit_i->sentry_lock); 1240 1241 return is_cp; 1242 } 1243 1244 /* 1245 * This function should be resided under the curseg_mutex lock 1246 */ 1247 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, 1248 struct f2fs_summary *sum) 1249 { 1250 struct curseg_info *curseg = CURSEG_I(sbi, type); 1251 void *addr = curseg->sum_blk; 1252 addr += curseg->next_blkoff * sizeof(struct f2fs_summary); 1253 memcpy(addr, sum, sizeof(struct f2fs_summary)); 1254 } 1255 1256 /* 1257 * Calculate the number of current summary pages for writing 1258 */ 1259 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) 1260 { 1261 int valid_sum_count = 0; 1262 int i, sum_in_page; 1263 1264 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 1265 if (sbi->ckpt->alloc_type[i] == SSR) 1266 valid_sum_count += sbi->blocks_per_seg; 1267 else { 1268 if (for_ra) 1269 valid_sum_count += le16_to_cpu( 1270 F2FS_CKPT(sbi)->cur_data_blkoff[i]); 1271 else 1272 valid_sum_count += curseg_blkoff(sbi, i); 1273 } 1274 } 1275 1276 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE - 1277 SUM_FOOTER_SIZE) / SUMMARY_SIZE; 1278 if (valid_sum_count <= sum_in_page) 1279 return 1; 1280 else if ((valid_sum_count - sum_in_page) <= 1281 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) 1282 return 2; 1283 return 3; 1284 } 1285 1286 /* 1287 * Caller should put this summary page 1288 */ 1289 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) 1290 { 1291 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno)); 1292 } 1293 1294 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr) 1295 { 1296 struct page *page = grab_meta_page(sbi, blk_addr); 1297 void *dst = page_address(page); 1298 1299 if (src) 1300 memcpy(dst, src, PAGE_SIZE); 1301 else 1302 memset(dst, 0, PAGE_SIZE); 1303 set_page_dirty(page); 1304 f2fs_put_page(page, 1); 1305 } 1306 1307 static void write_sum_page(struct f2fs_sb_info *sbi, 1308 struct f2fs_summary_block *sum_blk, block_t blk_addr) 1309 { 1310 update_meta_page(sbi, (void *)sum_blk, blk_addr); 1311 } 1312 1313 static void write_current_sum_page(struct f2fs_sb_info *sbi, 1314 int type, block_t blk_addr) 1315 { 1316 struct curseg_info *curseg = CURSEG_I(sbi, type); 1317 struct page *page = grab_meta_page(sbi, blk_addr); 1318 struct f2fs_summary_block *src = curseg->sum_blk; 1319 struct f2fs_summary_block *dst; 1320 1321 dst = (struct f2fs_summary_block *)page_address(page); 1322 1323 mutex_lock(&curseg->curseg_mutex); 1324 1325 down_read(&curseg->journal_rwsem); 1326 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE); 1327 up_read(&curseg->journal_rwsem); 1328 1329 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE); 1330 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE); 1331 1332 mutex_unlock(&curseg->curseg_mutex); 1333 1334 set_page_dirty(page); 1335 f2fs_put_page(page, 1); 1336 } 1337 1338 /* 1339 * Find a new segment from the free segments bitmap to right order 1340 * This function should be returned with success, otherwise BUG 1341 */ 1342 static void get_new_segment(struct f2fs_sb_info *sbi, 1343 unsigned int *newseg, bool new_sec, int dir) 1344 { 1345 struct free_segmap_info *free_i = FREE_I(sbi); 1346 unsigned int segno, secno, zoneno; 1347 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; 1348 unsigned int hint = *newseg / sbi->segs_per_sec; 1349 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg); 1350 unsigned int left_start = hint; 1351 bool init = true; 1352 int go_left = 0; 1353 int i; 1354 1355 spin_lock(&free_i->segmap_lock); 1356 1357 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { 1358 segno = find_next_zero_bit(free_i->free_segmap, 1359 (hint + 1) * sbi->segs_per_sec, *newseg + 1); 1360 if (segno < (hint + 1) * sbi->segs_per_sec) 1361 goto got_it; 1362 } 1363 find_other_zone: 1364 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 1365 if (secno >= MAIN_SECS(sbi)) { 1366 if (dir == ALLOC_RIGHT) { 1367 secno = find_next_zero_bit(free_i->free_secmap, 1368 MAIN_SECS(sbi), 0); 1369 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi)); 1370 } else { 1371 go_left = 1; 1372 left_start = hint - 1; 1373 } 1374 } 1375 if (go_left == 0) 1376 goto skip_left; 1377 1378 while (test_bit(left_start, free_i->free_secmap)) { 1379 if (left_start > 0) { 1380 left_start--; 1381 continue; 1382 } 1383 left_start = find_next_zero_bit(free_i->free_secmap, 1384 MAIN_SECS(sbi), 0); 1385 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi)); 1386 break; 1387 } 1388 secno = left_start; 1389 skip_left: 1390 hint = secno; 1391 segno = secno * sbi->segs_per_sec; 1392 zoneno = secno / sbi->secs_per_zone; 1393 1394 /* give up on finding another zone */ 1395 if (!init) 1396 goto got_it; 1397 if (sbi->secs_per_zone == 1) 1398 goto got_it; 1399 if (zoneno == old_zoneno) 1400 goto got_it; 1401 if (dir == ALLOC_LEFT) { 1402 if (!go_left && zoneno + 1 >= total_zones) 1403 goto got_it; 1404 if (go_left && zoneno == 0) 1405 goto got_it; 1406 } 1407 for (i = 0; i < NR_CURSEG_TYPE; i++) 1408 if (CURSEG_I(sbi, i)->zone == zoneno) 1409 break; 1410 1411 if (i < NR_CURSEG_TYPE) { 1412 /* zone is in user, try another */ 1413 if (go_left) 1414 hint = zoneno * sbi->secs_per_zone - 1; 1415 else if (zoneno + 1 >= total_zones) 1416 hint = 0; 1417 else 1418 hint = (zoneno + 1) * sbi->secs_per_zone; 1419 init = false; 1420 goto find_other_zone; 1421 } 1422 got_it: 1423 /* set it as dirty segment in free segmap */ 1424 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); 1425 __set_inuse(sbi, segno); 1426 *newseg = segno; 1427 spin_unlock(&free_i->segmap_lock); 1428 } 1429 1430 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 1431 { 1432 struct curseg_info *curseg = CURSEG_I(sbi, type); 1433 struct summary_footer *sum_footer; 1434 1435 curseg->segno = curseg->next_segno; 1436 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno); 1437 curseg->next_blkoff = 0; 1438 curseg->next_segno = NULL_SEGNO; 1439 1440 sum_footer = &(curseg->sum_blk->footer); 1441 memset(sum_footer, 0, sizeof(struct summary_footer)); 1442 if (IS_DATASEG(type)) 1443 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 1444 if (IS_NODESEG(type)) 1445 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 1446 __set_sit_entry_type(sbi, type, curseg->segno, modified); 1447 } 1448 1449 /* 1450 * Allocate a current working segment. 1451 * This function always allocates a free segment in LFS manner. 1452 */ 1453 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 1454 { 1455 struct curseg_info *curseg = CURSEG_I(sbi, type); 1456 unsigned int segno = curseg->segno; 1457 int dir = ALLOC_LEFT; 1458 1459 write_sum_page(sbi, curseg->sum_blk, 1460 GET_SUM_BLOCK(sbi, segno)); 1461 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA) 1462 dir = ALLOC_RIGHT; 1463 1464 if (test_opt(sbi, NOHEAP)) 1465 dir = ALLOC_RIGHT; 1466 1467 get_new_segment(sbi, &segno, new_sec, dir); 1468 curseg->next_segno = segno; 1469 reset_curseg(sbi, type, 1); 1470 curseg->alloc_type = LFS; 1471 } 1472 1473 static void __next_free_blkoff(struct f2fs_sb_info *sbi, 1474 struct curseg_info *seg, block_t start) 1475 { 1476 struct seg_entry *se = get_seg_entry(sbi, seg->segno); 1477 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 1478 unsigned long *target_map = SIT_I(sbi)->tmp_map; 1479 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 1480 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 1481 int i, pos; 1482 1483 for (i = 0; i < entries; i++) 1484 target_map[i] = ckpt_map[i] | cur_map[i]; 1485 1486 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); 1487 1488 seg->next_blkoff = pos; 1489 } 1490 1491 /* 1492 * If a segment is written by LFS manner, next block offset is just obtained 1493 * by increasing the current block offset. However, if a segment is written by 1494 * SSR manner, next block offset obtained by calling __next_free_blkoff 1495 */ 1496 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, 1497 struct curseg_info *seg) 1498 { 1499 if (seg->alloc_type == SSR) 1500 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1); 1501 else 1502 seg->next_blkoff++; 1503 } 1504 1505 /* 1506 * This function always allocates a used segment(from dirty seglist) by SSR 1507 * manner, so it should recover the existing segment information of valid blocks 1508 */ 1509 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse) 1510 { 1511 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1512 struct curseg_info *curseg = CURSEG_I(sbi, type); 1513 unsigned int new_segno = curseg->next_segno; 1514 struct f2fs_summary_block *sum_node; 1515 struct page *sum_page; 1516 1517 write_sum_page(sbi, curseg->sum_blk, 1518 GET_SUM_BLOCK(sbi, curseg->segno)); 1519 __set_test_and_inuse(sbi, new_segno); 1520 1521 mutex_lock(&dirty_i->seglist_lock); 1522 __remove_dirty_segment(sbi, new_segno, PRE); 1523 __remove_dirty_segment(sbi, new_segno, DIRTY); 1524 mutex_unlock(&dirty_i->seglist_lock); 1525 1526 reset_curseg(sbi, type, 1); 1527 curseg->alloc_type = SSR; 1528 __next_free_blkoff(sbi, curseg, 0); 1529 1530 if (reuse) { 1531 sum_page = get_sum_page(sbi, new_segno); 1532 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 1533 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); 1534 f2fs_put_page(sum_page, 1); 1535 } 1536 } 1537 1538 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type) 1539 { 1540 struct curseg_info *curseg = CURSEG_I(sbi, type); 1541 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; 1542 int i, n; 1543 1544 /* need_SSR() already forces to do this */ 1545 if (v_ops->get_victim(sbi, &(curseg)->next_segno, BG_GC, type, SSR)) 1546 return 1; 1547 1548 /* For node segments, let's do SSR more intensively */ 1549 if (IS_NODESEG(type)) { 1550 i = CURSEG_HOT_NODE; 1551 n = CURSEG_COLD_NODE; 1552 } else { 1553 i = CURSEG_HOT_DATA; 1554 n = CURSEG_COLD_DATA; 1555 } 1556 1557 for (; i <= n; i++) { 1558 if (i == type) 1559 continue; 1560 if (v_ops->get_victim(sbi, &(curseg)->next_segno, 1561 BG_GC, i, SSR)) 1562 return 1; 1563 } 1564 return 0; 1565 } 1566 1567 /* 1568 * flush out current segment and replace it with new segment 1569 * This function should be returned with success, otherwise BUG 1570 */ 1571 static void allocate_segment_by_default(struct f2fs_sb_info *sbi, 1572 int type, bool force) 1573 { 1574 if (force) 1575 new_curseg(sbi, type, true); 1576 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) && 1577 type == CURSEG_WARM_NODE) 1578 new_curseg(sbi, type, false); 1579 else if (need_SSR(sbi) && get_ssr_segment(sbi, type)) 1580 change_curseg(sbi, type, true); 1581 else 1582 new_curseg(sbi, type, false); 1583 1584 stat_inc_seg_type(sbi, CURSEG_I(sbi, type)); 1585 } 1586 1587 void allocate_new_segments(struct f2fs_sb_info *sbi) 1588 { 1589 struct curseg_info *curseg; 1590 unsigned int old_segno; 1591 int i; 1592 1593 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 1594 curseg = CURSEG_I(sbi, i); 1595 old_segno = curseg->segno; 1596 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true); 1597 locate_dirty_segment(sbi, old_segno); 1598 } 1599 } 1600 1601 static const struct segment_allocation default_salloc_ops = { 1602 .allocate_segment = allocate_segment_by_default, 1603 }; 1604 1605 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1606 { 1607 __u64 trim_start = cpc->trim_start; 1608 bool has_candidate = false; 1609 1610 mutex_lock(&SIT_I(sbi)->sentry_lock); 1611 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) { 1612 if (add_discard_addrs(sbi, cpc, true)) { 1613 has_candidate = true; 1614 break; 1615 } 1616 } 1617 mutex_unlock(&SIT_I(sbi)->sentry_lock); 1618 1619 cpc->trim_start = trim_start; 1620 return has_candidate; 1621 } 1622 1623 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) 1624 { 1625 __u64 start = F2FS_BYTES_TO_BLK(range->start); 1626 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; 1627 unsigned int start_segno, end_segno; 1628 struct cp_control cpc; 1629 int err = 0; 1630 1631 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize) 1632 return -EINVAL; 1633 1634 cpc.trimmed = 0; 1635 if (end <= MAIN_BLKADDR(sbi)) 1636 goto out; 1637 1638 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 1639 f2fs_msg(sbi->sb, KERN_WARNING, 1640 "Found FS corruption, run fsck to fix."); 1641 goto out; 1642 } 1643 1644 /* start/end segment number in main_area */ 1645 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); 1646 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : 1647 GET_SEGNO(sbi, end); 1648 cpc.reason = CP_DISCARD; 1649 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen)); 1650 1651 /* do checkpoint to issue discard commands safely */ 1652 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) { 1653 cpc.trim_start = start_segno; 1654 1655 if (sbi->discard_blks == 0) 1656 break; 1657 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi)) 1658 cpc.trim_end = end_segno; 1659 else 1660 cpc.trim_end = min_t(unsigned int, 1661 rounddown(start_segno + 1662 BATCHED_TRIM_SEGMENTS(sbi), 1663 sbi->segs_per_sec) - 1, end_segno); 1664 1665 mutex_lock(&sbi->gc_mutex); 1666 err = write_checkpoint(sbi, &cpc); 1667 mutex_unlock(&sbi->gc_mutex); 1668 if (err) 1669 break; 1670 1671 schedule(); 1672 } 1673 out: 1674 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed); 1675 return err; 1676 } 1677 1678 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type) 1679 { 1680 struct curseg_info *curseg = CURSEG_I(sbi, type); 1681 if (curseg->next_blkoff < sbi->blocks_per_seg) 1682 return true; 1683 return false; 1684 } 1685 1686 static int __get_segment_type_2(struct page *page, enum page_type p_type) 1687 { 1688 if (p_type == DATA) 1689 return CURSEG_HOT_DATA; 1690 else 1691 return CURSEG_HOT_NODE; 1692 } 1693 1694 static int __get_segment_type_4(struct page *page, enum page_type p_type) 1695 { 1696 if (p_type == DATA) { 1697 struct inode *inode = page->mapping->host; 1698 1699 if (S_ISDIR(inode->i_mode)) 1700 return CURSEG_HOT_DATA; 1701 else 1702 return CURSEG_COLD_DATA; 1703 } else { 1704 if (IS_DNODE(page) && is_cold_node(page)) 1705 return CURSEG_WARM_NODE; 1706 else 1707 return CURSEG_COLD_NODE; 1708 } 1709 } 1710 1711 static int __get_segment_type_6(struct page *page, enum page_type p_type) 1712 { 1713 if (p_type == DATA) { 1714 struct inode *inode = page->mapping->host; 1715 1716 if (S_ISDIR(inode->i_mode)) 1717 return CURSEG_HOT_DATA; 1718 else if (is_cold_data(page) || file_is_cold(inode)) 1719 return CURSEG_COLD_DATA; 1720 else 1721 return CURSEG_WARM_DATA; 1722 } else { 1723 if (IS_DNODE(page)) 1724 return is_cold_node(page) ? CURSEG_WARM_NODE : 1725 CURSEG_HOT_NODE; 1726 else 1727 return CURSEG_COLD_NODE; 1728 } 1729 } 1730 1731 static int __get_segment_type(struct page *page, enum page_type p_type) 1732 { 1733 switch (F2FS_P_SB(page)->active_logs) { 1734 case 2: 1735 return __get_segment_type_2(page, p_type); 1736 case 4: 1737 return __get_segment_type_4(page, p_type); 1738 } 1739 /* NR_CURSEG_TYPE(6) logs by default */ 1740 f2fs_bug_on(F2FS_P_SB(page), 1741 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE); 1742 return __get_segment_type_6(page, p_type); 1743 } 1744 1745 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, 1746 block_t old_blkaddr, block_t *new_blkaddr, 1747 struct f2fs_summary *sum, int type) 1748 { 1749 struct sit_info *sit_i = SIT_I(sbi); 1750 struct curseg_info *curseg = CURSEG_I(sbi, type); 1751 1752 mutex_lock(&curseg->curseg_mutex); 1753 mutex_lock(&sit_i->sentry_lock); 1754 1755 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 1756 1757 f2fs_wait_discard_bio(sbi, *new_blkaddr); 1758 1759 /* 1760 * __add_sum_entry should be resided under the curseg_mutex 1761 * because, this function updates a summary entry in the 1762 * current summary block. 1763 */ 1764 __add_sum_entry(sbi, type, sum); 1765 1766 __refresh_next_blkoff(sbi, curseg); 1767 1768 stat_inc_block_count(sbi, curseg); 1769 1770 /* 1771 * SIT information should be updated before segment allocation, 1772 * since SSR needs latest valid block information. 1773 */ 1774 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr); 1775 1776 if (!__has_curseg_space(sbi, type)) 1777 sit_i->s_ops->allocate_segment(sbi, type, false); 1778 1779 mutex_unlock(&sit_i->sentry_lock); 1780 1781 if (page && IS_NODESEG(type)) 1782 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); 1783 1784 mutex_unlock(&curseg->curseg_mutex); 1785 } 1786 1787 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) 1788 { 1789 int type = __get_segment_type(fio->page, fio->type); 1790 int err; 1791 1792 if (fio->type == NODE || fio->type == DATA) 1793 mutex_lock(&fio->sbi->wio_mutex[fio->type]); 1794 reallocate: 1795 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr, 1796 &fio->new_blkaddr, sum, type); 1797 1798 /* writeout dirty page into bdev */ 1799 err = f2fs_submit_page_mbio(fio); 1800 if (err == -EAGAIN) { 1801 fio->old_blkaddr = fio->new_blkaddr; 1802 goto reallocate; 1803 } 1804 1805 if (fio->type == NODE || fio->type == DATA) 1806 mutex_unlock(&fio->sbi->wio_mutex[fio->type]); 1807 } 1808 1809 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page) 1810 { 1811 struct f2fs_io_info fio = { 1812 .sbi = sbi, 1813 .type = META, 1814 .op = REQ_OP_WRITE, 1815 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO, 1816 .old_blkaddr = page->index, 1817 .new_blkaddr = page->index, 1818 .page = page, 1819 .encrypted_page = NULL, 1820 }; 1821 1822 if (unlikely(page->index >= MAIN_BLKADDR(sbi))) 1823 fio.op_flags &= ~REQ_META; 1824 1825 set_page_writeback(page); 1826 f2fs_submit_page_mbio(&fio); 1827 } 1828 1829 void write_node_page(unsigned int nid, struct f2fs_io_info *fio) 1830 { 1831 struct f2fs_summary sum; 1832 1833 set_summary(&sum, nid, 0, 0); 1834 do_write_page(&sum, fio); 1835 } 1836 1837 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio) 1838 { 1839 struct f2fs_sb_info *sbi = fio->sbi; 1840 struct f2fs_summary sum; 1841 struct node_info ni; 1842 1843 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); 1844 get_node_info(sbi, dn->nid, &ni); 1845 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); 1846 do_write_page(&sum, fio); 1847 f2fs_update_data_blkaddr(dn, fio->new_blkaddr); 1848 } 1849 1850 void rewrite_data_page(struct f2fs_io_info *fio) 1851 { 1852 fio->new_blkaddr = fio->old_blkaddr; 1853 stat_inc_inplace_blocks(fio->sbi); 1854 f2fs_submit_page_mbio(fio); 1855 } 1856 1857 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 1858 block_t old_blkaddr, block_t new_blkaddr, 1859 bool recover_curseg, bool recover_newaddr) 1860 { 1861 struct sit_info *sit_i = SIT_I(sbi); 1862 struct curseg_info *curseg; 1863 unsigned int segno, old_cursegno; 1864 struct seg_entry *se; 1865 int type; 1866 unsigned short old_blkoff; 1867 1868 segno = GET_SEGNO(sbi, new_blkaddr); 1869 se = get_seg_entry(sbi, segno); 1870 type = se->type; 1871 1872 if (!recover_curseg) { 1873 /* for recovery flow */ 1874 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { 1875 if (old_blkaddr == NULL_ADDR) 1876 type = CURSEG_COLD_DATA; 1877 else 1878 type = CURSEG_WARM_DATA; 1879 } 1880 } else { 1881 if (!IS_CURSEG(sbi, segno)) 1882 type = CURSEG_WARM_DATA; 1883 } 1884 1885 curseg = CURSEG_I(sbi, type); 1886 1887 mutex_lock(&curseg->curseg_mutex); 1888 mutex_lock(&sit_i->sentry_lock); 1889 1890 old_cursegno = curseg->segno; 1891 old_blkoff = curseg->next_blkoff; 1892 1893 /* change the current segment */ 1894 if (segno != curseg->segno) { 1895 curseg->next_segno = segno; 1896 change_curseg(sbi, type, true); 1897 } 1898 1899 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); 1900 __add_sum_entry(sbi, type, sum); 1901 1902 if (!recover_curseg || recover_newaddr) 1903 update_sit_entry(sbi, new_blkaddr, 1); 1904 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 1905 update_sit_entry(sbi, old_blkaddr, -1); 1906 1907 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 1908 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); 1909 1910 locate_dirty_segment(sbi, old_cursegno); 1911 1912 if (recover_curseg) { 1913 if (old_cursegno != curseg->segno) { 1914 curseg->next_segno = old_cursegno; 1915 change_curseg(sbi, type, true); 1916 } 1917 curseg->next_blkoff = old_blkoff; 1918 } 1919 1920 mutex_unlock(&sit_i->sentry_lock); 1921 mutex_unlock(&curseg->curseg_mutex); 1922 } 1923 1924 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, 1925 block_t old_addr, block_t new_addr, 1926 unsigned char version, bool recover_curseg, 1927 bool recover_newaddr) 1928 { 1929 struct f2fs_summary sum; 1930 1931 set_summary(&sum, dn->nid, dn->ofs_in_node, version); 1932 1933 __f2fs_replace_block(sbi, &sum, old_addr, new_addr, 1934 recover_curseg, recover_newaddr); 1935 1936 f2fs_update_data_blkaddr(dn, new_addr); 1937 } 1938 1939 void f2fs_wait_on_page_writeback(struct page *page, 1940 enum page_type type, bool ordered) 1941 { 1942 if (PageWriteback(page)) { 1943 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 1944 1945 f2fs_submit_merged_bio_cond(sbi, page->mapping->host, 1946 0, page->index, type, WRITE); 1947 if (ordered) 1948 wait_on_page_writeback(page); 1949 else 1950 wait_for_stable_page(page); 1951 } 1952 } 1953 1954 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi, 1955 block_t blkaddr) 1956 { 1957 struct page *cpage; 1958 1959 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) 1960 return; 1961 1962 cpage = find_lock_page(META_MAPPING(sbi), blkaddr); 1963 if (cpage) { 1964 f2fs_wait_on_page_writeback(cpage, DATA, true); 1965 f2fs_put_page(cpage, 1); 1966 } 1967 } 1968 1969 static int read_compacted_summaries(struct f2fs_sb_info *sbi) 1970 { 1971 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1972 struct curseg_info *seg_i; 1973 unsigned char *kaddr; 1974 struct page *page; 1975 block_t start; 1976 int i, j, offset; 1977 1978 start = start_sum_block(sbi); 1979 1980 page = get_meta_page(sbi, start++); 1981 kaddr = (unsigned char *)page_address(page); 1982 1983 /* Step 1: restore nat cache */ 1984 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 1985 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE); 1986 1987 /* Step 2: restore sit cache */ 1988 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 1989 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE); 1990 offset = 2 * SUM_JOURNAL_SIZE; 1991 1992 /* Step 3: restore summary entries */ 1993 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 1994 unsigned short blk_off; 1995 unsigned int segno; 1996 1997 seg_i = CURSEG_I(sbi, i); 1998 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 1999 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 2000 seg_i->next_segno = segno; 2001 reset_curseg(sbi, i, 0); 2002 seg_i->alloc_type = ckpt->alloc_type[i]; 2003 seg_i->next_blkoff = blk_off; 2004 2005 if (seg_i->alloc_type == SSR) 2006 blk_off = sbi->blocks_per_seg; 2007 2008 for (j = 0; j < blk_off; j++) { 2009 struct f2fs_summary *s; 2010 s = (struct f2fs_summary *)(kaddr + offset); 2011 seg_i->sum_blk->entries[j] = *s; 2012 offset += SUMMARY_SIZE; 2013 if (offset + SUMMARY_SIZE <= PAGE_SIZE - 2014 SUM_FOOTER_SIZE) 2015 continue; 2016 2017 f2fs_put_page(page, 1); 2018 page = NULL; 2019 2020 page = get_meta_page(sbi, start++); 2021 kaddr = (unsigned char *)page_address(page); 2022 offset = 0; 2023 } 2024 } 2025 f2fs_put_page(page, 1); 2026 return 0; 2027 } 2028 2029 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 2030 { 2031 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 2032 struct f2fs_summary_block *sum; 2033 struct curseg_info *curseg; 2034 struct page *new; 2035 unsigned short blk_off; 2036 unsigned int segno = 0; 2037 block_t blk_addr = 0; 2038 2039 /* get segment number and block addr */ 2040 if (IS_DATASEG(type)) { 2041 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 2042 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 2043 CURSEG_HOT_DATA]); 2044 if (__exist_node_summaries(sbi)) 2045 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type); 2046 else 2047 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 2048 } else { 2049 segno = le32_to_cpu(ckpt->cur_node_segno[type - 2050 CURSEG_HOT_NODE]); 2051 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 2052 CURSEG_HOT_NODE]); 2053 if (__exist_node_summaries(sbi)) 2054 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 2055 type - CURSEG_HOT_NODE); 2056 else 2057 blk_addr = GET_SUM_BLOCK(sbi, segno); 2058 } 2059 2060 new = get_meta_page(sbi, blk_addr); 2061 sum = (struct f2fs_summary_block *)page_address(new); 2062 2063 if (IS_NODESEG(type)) { 2064 if (__exist_node_summaries(sbi)) { 2065 struct f2fs_summary *ns = &sum->entries[0]; 2066 int i; 2067 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { 2068 ns->version = 0; 2069 ns->ofs_in_node = 0; 2070 } 2071 } else { 2072 int err; 2073 2074 err = restore_node_summary(sbi, segno, sum); 2075 if (err) { 2076 f2fs_put_page(new, 1); 2077 return err; 2078 } 2079 } 2080 } 2081 2082 /* set uncompleted segment to curseg */ 2083 curseg = CURSEG_I(sbi, type); 2084 mutex_lock(&curseg->curseg_mutex); 2085 2086 /* update journal info */ 2087 down_write(&curseg->journal_rwsem); 2088 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE); 2089 up_write(&curseg->journal_rwsem); 2090 2091 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE); 2092 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE); 2093 curseg->next_segno = segno; 2094 reset_curseg(sbi, type, 0); 2095 curseg->alloc_type = ckpt->alloc_type[type]; 2096 curseg->next_blkoff = blk_off; 2097 mutex_unlock(&curseg->curseg_mutex); 2098 f2fs_put_page(new, 1); 2099 return 0; 2100 } 2101 2102 static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 2103 { 2104 int type = CURSEG_HOT_DATA; 2105 int err; 2106 2107 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) { 2108 int npages = npages_for_summary_flush(sbi, true); 2109 2110 if (npages >= 2) 2111 ra_meta_pages(sbi, start_sum_block(sbi), npages, 2112 META_CP, true); 2113 2114 /* restore for compacted data summary */ 2115 if (read_compacted_summaries(sbi)) 2116 return -EINVAL; 2117 type = CURSEG_HOT_NODE; 2118 } 2119 2120 if (__exist_node_summaries(sbi)) 2121 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type), 2122 NR_CURSEG_TYPE - type, META_CP, true); 2123 2124 for (; type <= CURSEG_COLD_NODE; type++) { 2125 err = read_normal_summaries(sbi, type); 2126 if (err) 2127 return err; 2128 } 2129 2130 return 0; 2131 } 2132 2133 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 2134 { 2135 struct page *page; 2136 unsigned char *kaddr; 2137 struct f2fs_summary *summary; 2138 struct curseg_info *seg_i; 2139 int written_size = 0; 2140 int i, j; 2141 2142 page = grab_meta_page(sbi, blkaddr++); 2143 kaddr = (unsigned char *)page_address(page); 2144 2145 /* Step 1: write nat cache */ 2146 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 2147 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE); 2148 written_size += SUM_JOURNAL_SIZE; 2149 2150 /* Step 2: write sit cache */ 2151 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 2152 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE); 2153 written_size += SUM_JOURNAL_SIZE; 2154 2155 /* Step 3: write summary entries */ 2156 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2157 unsigned short blkoff; 2158 seg_i = CURSEG_I(sbi, i); 2159 if (sbi->ckpt->alloc_type[i] == SSR) 2160 blkoff = sbi->blocks_per_seg; 2161 else 2162 blkoff = curseg_blkoff(sbi, i); 2163 2164 for (j = 0; j < blkoff; j++) { 2165 if (!page) { 2166 page = grab_meta_page(sbi, blkaddr++); 2167 kaddr = (unsigned char *)page_address(page); 2168 written_size = 0; 2169 } 2170 summary = (struct f2fs_summary *)(kaddr + written_size); 2171 *summary = seg_i->sum_blk->entries[j]; 2172 written_size += SUMMARY_SIZE; 2173 2174 if (written_size + SUMMARY_SIZE <= PAGE_SIZE - 2175 SUM_FOOTER_SIZE) 2176 continue; 2177 2178 set_page_dirty(page); 2179 f2fs_put_page(page, 1); 2180 page = NULL; 2181 } 2182 } 2183 if (page) { 2184 set_page_dirty(page); 2185 f2fs_put_page(page, 1); 2186 } 2187 } 2188 2189 static void write_normal_summaries(struct f2fs_sb_info *sbi, 2190 block_t blkaddr, int type) 2191 { 2192 int i, end; 2193 if (IS_DATASEG(type)) 2194 end = type + NR_CURSEG_DATA_TYPE; 2195 else 2196 end = type + NR_CURSEG_NODE_TYPE; 2197 2198 for (i = type; i < end; i++) 2199 write_current_sum_page(sbi, i, blkaddr + (i - type)); 2200 } 2201 2202 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 2203 { 2204 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) 2205 write_compacted_summaries(sbi, start_blk); 2206 else 2207 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 2208 } 2209 2210 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 2211 { 2212 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 2213 } 2214 2215 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type, 2216 unsigned int val, int alloc) 2217 { 2218 int i; 2219 2220 if (type == NAT_JOURNAL) { 2221 for (i = 0; i < nats_in_cursum(journal); i++) { 2222 if (le32_to_cpu(nid_in_journal(journal, i)) == val) 2223 return i; 2224 } 2225 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL)) 2226 return update_nats_in_cursum(journal, 1); 2227 } else if (type == SIT_JOURNAL) { 2228 for (i = 0; i < sits_in_cursum(journal); i++) 2229 if (le32_to_cpu(segno_in_journal(journal, i)) == val) 2230 return i; 2231 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL)) 2232 return update_sits_in_cursum(journal, 1); 2233 } 2234 return -1; 2235 } 2236 2237 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, 2238 unsigned int segno) 2239 { 2240 return get_meta_page(sbi, current_sit_addr(sbi, segno)); 2241 } 2242 2243 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, 2244 unsigned int start) 2245 { 2246 struct sit_info *sit_i = SIT_I(sbi); 2247 struct page *src_page, *dst_page; 2248 pgoff_t src_off, dst_off; 2249 void *src_addr, *dst_addr; 2250 2251 src_off = current_sit_addr(sbi, start); 2252 dst_off = next_sit_addr(sbi, src_off); 2253 2254 /* get current sit block page without lock */ 2255 src_page = get_meta_page(sbi, src_off); 2256 dst_page = grab_meta_page(sbi, dst_off); 2257 f2fs_bug_on(sbi, PageDirty(src_page)); 2258 2259 src_addr = page_address(src_page); 2260 dst_addr = page_address(dst_page); 2261 memcpy(dst_addr, src_addr, PAGE_SIZE); 2262 2263 set_page_dirty(dst_page); 2264 f2fs_put_page(src_page, 1); 2265 2266 set_to_next_sit(sit_i, start); 2267 2268 return dst_page; 2269 } 2270 2271 static struct sit_entry_set *grab_sit_entry_set(void) 2272 { 2273 struct sit_entry_set *ses = 2274 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS); 2275 2276 ses->entry_cnt = 0; 2277 INIT_LIST_HEAD(&ses->set_list); 2278 return ses; 2279 } 2280 2281 static void release_sit_entry_set(struct sit_entry_set *ses) 2282 { 2283 list_del(&ses->set_list); 2284 kmem_cache_free(sit_entry_set_slab, ses); 2285 } 2286 2287 static void adjust_sit_entry_set(struct sit_entry_set *ses, 2288 struct list_head *head) 2289 { 2290 struct sit_entry_set *next = ses; 2291 2292 if (list_is_last(&ses->set_list, head)) 2293 return; 2294 2295 list_for_each_entry_continue(next, head, set_list) 2296 if (ses->entry_cnt <= next->entry_cnt) 2297 break; 2298 2299 list_move_tail(&ses->set_list, &next->set_list); 2300 } 2301 2302 static void add_sit_entry(unsigned int segno, struct list_head *head) 2303 { 2304 struct sit_entry_set *ses; 2305 unsigned int start_segno = START_SEGNO(segno); 2306 2307 list_for_each_entry(ses, head, set_list) { 2308 if (ses->start_segno == start_segno) { 2309 ses->entry_cnt++; 2310 adjust_sit_entry_set(ses, head); 2311 return; 2312 } 2313 } 2314 2315 ses = grab_sit_entry_set(); 2316 2317 ses->start_segno = start_segno; 2318 ses->entry_cnt++; 2319 list_add(&ses->set_list, head); 2320 } 2321 2322 static void add_sits_in_set(struct f2fs_sb_info *sbi) 2323 { 2324 struct f2fs_sm_info *sm_info = SM_I(sbi); 2325 struct list_head *set_list = &sm_info->sit_entry_set; 2326 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; 2327 unsigned int segno; 2328 2329 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) 2330 add_sit_entry(segno, set_list); 2331 } 2332 2333 static void remove_sits_in_journal(struct f2fs_sb_info *sbi) 2334 { 2335 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 2336 struct f2fs_journal *journal = curseg->journal; 2337 int i; 2338 2339 down_write(&curseg->journal_rwsem); 2340 for (i = 0; i < sits_in_cursum(journal); i++) { 2341 unsigned int segno; 2342 bool dirtied; 2343 2344 segno = le32_to_cpu(segno_in_journal(journal, i)); 2345 dirtied = __mark_sit_entry_dirty(sbi, segno); 2346 2347 if (!dirtied) 2348 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); 2349 } 2350 update_sits_in_cursum(journal, -i); 2351 up_write(&curseg->journal_rwsem); 2352 } 2353 2354 /* 2355 * CP calls this function, which flushes SIT entries including sit_journal, 2356 * and moves prefree segs to free segs. 2357 */ 2358 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 2359 { 2360 struct sit_info *sit_i = SIT_I(sbi); 2361 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 2362 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 2363 struct f2fs_journal *journal = curseg->journal; 2364 struct sit_entry_set *ses, *tmp; 2365 struct list_head *head = &SM_I(sbi)->sit_entry_set; 2366 bool to_journal = true; 2367 struct seg_entry *se; 2368 2369 mutex_lock(&sit_i->sentry_lock); 2370 2371 if (!sit_i->dirty_sentries) 2372 goto out; 2373 2374 /* 2375 * add and account sit entries of dirty bitmap in sit entry 2376 * set temporarily 2377 */ 2378 add_sits_in_set(sbi); 2379 2380 /* 2381 * if there are no enough space in journal to store dirty sit 2382 * entries, remove all entries from journal and add and account 2383 * them in sit entry set. 2384 */ 2385 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL)) 2386 remove_sits_in_journal(sbi); 2387 2388 /* 2389 * there are two steps to flush sit entries: 2390 * #1, flush sit entries to journal in current cold data summary block. 2391 * #2, flush sit entries to sit page. 2392 */ 2393 list_for_each_entry_safe(ses, tmp, head, set_list) { 2394 struct page *page = NULL; 2395 struct f2fs_sit_block *raw_sit = NULL; 2396 unsigned int start_segno = ses->start_segno; 2397 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, 2398 (unsigned long)MAIN_SEGS(sbi)); 2399 unsigned int segno = start_segno; 2400 2401 if (to_journal && 2402 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL)) 2403 to_journal = false; 2404 2405 if (to_journal) { 2406 down_write(&curseg->journal_rwsem); 2407 } else { 2408 page = get_next_sit_page(sbi, start_segno); 2409 raw_sit = page_address(page); 2410 } 2411 2412 /* flush dirty sit entries in region of current sit set */ 2413 for_each_set_bit_from(segno, bitmap, end) { 2414 int offset, sit_offset; 2415 2416 se = get_seg_entry(sbi, segno); 2417 2418 /* add discard candidates */ 2419 if (cpc->reason != CP_DISCARD) { 2420 cpc->trim_start = segno; 2421 add_discard_addrs(sbi, cpc, false); 2422 } 2423 2424 if (to_journal) { 2425 offset = lookup_journal_in_cursum(journal, 2426 SIT_JOURNAL, segno, 1); 2427 f2fs_bug_on(sbi, offset < 0); 2428 segno_in_journal(journal, offset) = 2429 cpu_to_le32(segno); 2430 seg_info_to_raw_sit(se, 2431 &sit_in_journal(journal, offset)); 2432 } else { 2433 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 2434 seg_info_to_raw_sit(se, 2435 &raw_sit->entries[sit_offset]); 2436 } 2437 2438 __clear_bit(segno, bitmap); 2439 sit_i->dirty_sentries--; 2440 ses->entry_cnt--; 2441 } 2442 2443 if (to_journal) 2444 up_write(&curseg->journal_rwsem); 2445 else 2446 f2fs_put_page(page, 1); 2447 2448 f2fs_bug_on(sbi, ses->entry_cnt); 2449 release_sit_entry_set(ses); 2450 } 2451 2452 f2fs_bug_on(sbi, !list_empty(head)); 2453 f2fs_bug_on(sbi, sit_i->dirty_sentries); 2454 out: 2455 if (cpc->reason == CP_DISCARD) { 2456 __u64 trim_start = cpc->trim_start; 2457 2458 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) 2459 add_discard_addrs(sbi, cpc, false); 2460 2461 cpc->trim_start = trim_start; 2462 } 2463 mutex_unlock(&sit_i->sentry_lock); 2464 2465 set_prefree_as_free_segments(sbi); 2466 } 2467 2468 static int build_sit_info(struct f2fs_sb_info *sbi) 2469 { 2470 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 2471 struct sit_info *sit_i; 2472 unsigned int sit_segs, start; 2473 char *src_bitmap; 2474 unsigned int bitmap_size; 2475 2476 /* allocate memory for SIT information */ 2477 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL); 2478 if (!sit_i) 2479 return -ENOMEM; 2480 2481 SM_I(sbi)->sit_info = sit_i; 2482 2483 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) * 2484 sizeof(struct seg_entry), GFP_KERNEL); 2485 if (!sit_i->sentries) 2486 return -ENOMEM; 2487 2488 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 2489 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL); 2490 if (!sit_i->dirty_sentries_bitmap) 2491 return -ENOMEM; 2492 2493 for (start = 0; start < MAIN_SEGS(sbi); start++) { 2494 sit_i->sentries[start].cur_valid_map 2495 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 2496 sit_i->sentries[start].ckpt_valid_map 2497 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 2498 if (!sit_i->sentries[start].cur_valid_map || 2499 !sit_i->sentries[start].ckpt_valid_map) 2500 return -ENOMEM; 2501 2502 #ifdef CONFIG_F2FS_CHECK_FS 2503 sit_i->sentries[start].cur_valid_map_mir 2504 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 2505 if (!sit_i->sentries[start].cur_valid_map_mir) 2506 return -ENOMEM; 2507 #endif 2508 2509 if (f2fs_discard_en(sbi)) { 2510 sit_i->sentries[start].discard_map 2511 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 2512 if (!sit_i->sentries[start].discard_map) 2513 return -ENOMEM; 2514 } 2515 } 2516 2517 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 2518 if (!sit_i->tmp_map) 2519 return -ENOMEM; 2520 2521 if (sbi->segs_per_sec > 1) { 2522 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) * 2523 sizeof(struct sec_entry), GFP_KERNEL); 2524 if (!sit_i->sec_entries) 2525 return -ENOMEM; 2526 } 2527 2528 /* get information related with SIT */ 2529 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 2530 2531 /* setup SIT bitmap from ckeckpoint pack */ 2532 bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 2533 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 2534 2535 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); 2536 if (!sit_i->sit_bitmap) 2537 return -ENOMEM; 2538 2539 #ifdef CONFIG_F2FS_CHECK_FS 2540 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); 2541 if (!sit_i->sit_bitmap_mir) 2542 return -ENOMEM; 2543 #endif 2544 2545 /* init SIT information */ 2546 sit_i->s_ops = &default_salloc_ops; 2547 2548 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 2549 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; 2550 sit_i->written_valid_blocks = 0; 2551 sit_i->bitmap_size = bitmap_size; 2552 sit_i->dirty_sentries = 0; 2553 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 2554 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 2555 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec; 2556 mutex_init(&sit_i->sentry_lock); 2557 return 0; 2558 } 2559 2560 static int build_free_segmap(struct f2fs_sb_info *sbi) 2561 { 2562 struct free_segmap_info *free_i; 2563 unsigned int bitmap_size, sec_bitmap_size; 2564 2565 /* allocate memory for free segmap information */ 2566 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL); 2567 if (!free_i) 2568 return -ENOMEM; 2569 2570 SM_I(sbi)->free_info = free_i; 2571 2572 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 2573 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL); 2574 if (!free_i->free_segmap) 2575 return -ENOMEM; 2576 2577 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 2578 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL); 2579 if (!free_i->free_secmap) 2580 return -ENOMEM; 2581 2582 /* set all segments as dirty temporarily */ 2583 memset(free_i->free_segmap, 0xff, bitmap_size); 2584 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 2585 2586 /* init free segmap information */ 2587 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); 2588 free_i->free_segments = 0; 2589 free_i->free_sections = 0; 2590 spin_lock_init(&free_i->segmap_lock); 2591 return 0; 2592 } 2593 2594 static int build_curseg(struct f2fs_sb_info *sbi) 2595 { 2596 struct curseg_info *array; 2597 int i; 2598 2599 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL); 2600 if (!array) 2601 return -ENOMEM; 2602 2603 SM_I(sbi)->curseg_array = array; 2604 2605 for (i = 0; i < NR_CURSEG_TYPE; i++) { 2606 mutex_init(&array[i].curseg_mutex); 2607 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL); 2608 if (!array[i].sum_blk) 2609 return -ENOMEM; 2610 init_rwsem(&array[i].journal_rwsem); 2611 array[i].journal = kzalloc(sizeof(struct f2fs_journal), 2612 GFP_KERNEL); 2613 if (!array[i].journal) 2614 return -ENOMEM; 2615 array[i].segno = NULL_SEGNO; 2616 array[i].next_blkoff = 0; 2617 } 2618 return restore_curseg_summaries(sbi); 2619 } 2620 2621 static void build_sit_entries(struct f2fs_sb_info *sbi) 2622 { 2623 struct sit_info *sit_i = SIT_I(sbi); 2624 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 2625 struct f2fs_journal *journal = curseg->journal; 2626 struct seg_entry *se; 2627 struct f2fs_sit_entry sit; 2628 int sit_blk_cnt = SIT_BLK_CNT(sbi); 2629 unsigned int i, start, end; 2630 unsigned int readed, start_blk = 0; 2631 2632 do { 2633 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES, 2634 META_SIT, true); 2635 2636 start = start_blk * sit_i->sents_per_block; 2637 end = (start_blk + readed) * sit_i->sents_per_block; 2638 2639 for (; start < end && start < MAIN_SEGS(sbi); start++) { 2640 struct f2fs_sit_block *sit_blk; 2641 struct page *page; 2642 2643 se = &sit_i->sentries[start]; 2644 page = get_current_sit_page(sbi, start); 2645 sit_blk = (struct f2fs_sit_block *)page_address(page); 2646 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 2647 f2fs_put_page(page, 1); 2648 2649 check_block_count(sbi, start, &sit); 2650 seg_info_from_raw_sit(se, &sit); 2651 2652 /* build discard map only one time */ 2653 if (f2fs_discard_en(sbi)) { 2654 memcpy(se->discard_map, se->cur_valid_map, 2655 SIT_VBLOCK_MAP_SIZE); 2656 sbi->discard_blks += sbi->blocks_per_seg - 2657 se->valid_blocks; 2658 } 2659 2660 if (sbi->segs_per_sec > 1) 2661 get_sec_entry(sbi, start)->valid_blocks += 2662 se->valid_blocks; 2663 } 2664 start_blk += readed; 2665 } while (start_blk < sit_blk_cnt); 2666 2667 down_read(&curseg->journal_rwsem); 2668 for (i = 0; i < sits_in_cursum(journal); i++) { 2669 unsigned int old_valid_blocks; 2670 2671 start = le32_to_cpu(segno_in_journal(journal, i)); 2672 se = &sit_i->sentries[start]; 2673 sit = sit_in_journal(journal, i); 2674 2675 old_valid_blocks = se->valid_blocks; 2676 2677 check_block_count(sbi, start, &sit); 2678 seg_info_from_raw_sit(se, &sit); 2679 2680 if (f2fs_discard_en(sbi)) { 2681 memcpy(se->discard_map, se->cur_valid_map, 2682 SIT_VBLOCK_MAP_SIZE); 2683 sbi->discard_blks += old_valid_blocks - 2684 se->valid_blocks; 2685 } 2686 2687 if (sbi->segs_per_sec > 1) 2688 get_sec_entry(sbi, start)->valid_blocks += 2689 se->valid_blocks - old_valid_blocks; 2690 } 2691 up_read(&curseg->journal_rwsem); 2692 } 2693 2694 static void init_free_segmap(struct f2fs_sb_info *sbi) 2695 { 2696 unsigned int start; 2697 int type; 2698 2699 for (start = 0; start < MAIN_SEGS(sbi); start++) { 2700 struct seg_entry *sentry = get_seg_entry(sbi, start); 2701 if (!sentry->valid_blocks) 2702 __set_free(sbi, start); 2703 else 2704 SIT_I(sbi)->written_valid_blocks += 2705 sentry->valid_blocks; 2706 } 2707 2708 /* set use the current segments */ 2709 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 2710 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 2711 __set_test_and_inuse(sbi, curseg_t->segno); 2712 } 2713 } 2714 2715 static void init_dirty_segmap(struct f2fs_sb_info *sbi) 2716 { 2717 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2718 struct free_segmap_info *free_i = FREE_I(sbi); 2719 unsigned int segno = 0, offset = 0; 2720 unsigned short valid_blocks; 2721 2722 while (1) { 2723 /* find dirty segment based on free segmap */ 2724 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); 2725 if (segno >= MAIN_SEGS(sbi)) 2726 break; 2727 offset = segno + 1; 2728 valid_blocks = get_valid_blocks(sbi, segno, 0); 2729 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks) 2730 continue; 2731 if (valid_blocks > sbi->blocks_per_seg) { 2732 f2fs_bug_on(sbi, 1); 2733 continue; 2734 } 2735 mutex_lock(&dirty_i->seglist_lock); 2736 __locate_dirty_segment(sbi, segno, DIRTY); 2737 mutex_unlock(&dirty_i->seglist_lock); 2738 } 2739 } 2740 2741 static int init_victim_secmap(struct f2fs_sb_info *sbi) 2742 { 2743 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2744 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 2745 2746 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL); 2747 if (!dirty_i->victim_secmap) 2748 return -ENOMEM; 2749 return 0; 2750 } 2751 2752 static int build_dirty_segmap(struct f2fs_sb_info *sbi) 2753 { 2754 struct dirty_seglist_info *dirty_i; 2755 unsigned int bitmap_size, i; 2756 2757 /* allocate memory for dirty segments list information */ 2758 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL); 2759 if (!dirty_i) 2760 return -ENOMEM; 2761 2762 SM_I(sbi)->dirty_info = dirty_i; 2763 mutex_init(&dirty_i->seglist_lock); 2764 2765 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 2766 2767 for (i = 0; i < NR_DIRTY_TYPE; i++) { 2768 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL); 2769 if (!dirty_i->dirty_segmap[i]) 2770 return -ENOMEM; 2771 } 2772 2773 init_dirty_segmap(sbi); 2774 return init_victim_secmap(sbi); 2775 } 2776 2777 /* 2778 * Update min, max modified time for cost-benefit GC algorithm 2779 */ 2780 static void init_min_max_mtime(struct f2fs_sb_info *sbi) 2781 { 2782 struct sit_info *sit_i = SIT_I(sbi); 2783 unsigned int segno; 2784 2785 mutex_lock(&sit_i->sentry_lock); 2786 2787 sit_i->min_mtime = LLONG_MAX; 2788 2789 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 2790 unsigned int i; 2791 unsigned long long mtime = 0; 2792 2793 for (i = 0; i < sbi->segs_per_sec; i++) 2794 mtime += get_seg_entry(sbi, segno + i)->mtime; 2795 2796 mtime = div_u64(mtime, sbi->segs_per_sec); 2797 2798 if (sit_i->min_mtime > mtime) 2799 sit_i->min_mtime = mtime; 2800 } 2801 sit_i->max_mtime = get_mtime(sbi); 2802 mutex_unlock(&sit_i->sentry_lock); 2803 } 2804 2805 int build_segment_manager(struct f2fs_sb_info *sbi) 2806 { 2807 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 2808 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 2809 struct f2fs_sm_info *sm_info; 2810 int err; 2811 2812 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL); 2813 if (!sm_info) 2814 return -ENOMEM; 2815 2816 /* init sm info */ 2817 sbi->sm_info = sm_info; 2818 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 2819 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 2820 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 2821 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 2822 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 2823 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 2824 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 2825 sm_info->rec_prefree_segments = sm_info->main_segments * 2826 DEF_RECLAIM_PREFREE_SEGMENTS / 100; 2827 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS) 2828 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS; 2829 2830 if (!test_opt(sbi, LFS)) 2831 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC; 2832 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; 2833 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; 2834 2835 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS; 2836 2837 INIT_LIST_HEAD(&sm_info->sit_entry_set); 2838 2839 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) { 2840 err = create_flush_cmd_control(sbi); 2841 if (err) 2842 return err; 2843 } 2844 2845 err = create_discard_cmd_control(sbi); 2846 if (err) 2847 return err; 2848 2849 err = build_sit_info(sbi); 2850 if (err) 2851 return err; 2852 err = build_free_segmap(sbi); 2853 if (err) 2854 return err; 2855 err = build_curseg(sbi); 2856 if (err) 2857 return err; 2858 2859 /* reinit free segmap based on SIT */ 2860 build_sit_entries(sbi); 2861 2862 init_free_segmap(sbi); 2863 err = build_dirty_segmap(sbi); 2864 if (err) 2865 return err; 2866 2867 init_min_max_mtime(sbi); 2868 return 0; 2869 } 2870 2871 static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 2872 enum dirty_type dirty_type) 2873 { 2874 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2875 2876 mutex_lock(&dirty_i->seglist_lock); 2877 kvfree(dirty_i->dirty_segmap[dirty_type]); 2878 dirty_i->nr_dirty[dirty_type] = 0; 2879 mutex_unlock(&dirty_i->seglist_lock); 2880 } 2881 2882 static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 2883 { 2884 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2885 kvfree(dirty_i->victim_secmap); 2886 } 2887 2888 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 2889 { 2890 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2891 int i; 2892 2893 if (!dirty_i) 2894 return; 2895 2896 /* discard pre-free/dirty segments list */ 2897 for (i = 0; i < NR_DIRTY_TYPE; i++) 2898 discard_dirty_segmap(sbi, i); 2899 2900 destroy_victim_secmap(sbi); 2901 SM_I(sbi)->dirty_info = NULL; 2902 kfree(dirty_i); 2903 } 2904 2905 static void destroy_curseg(struct f2fs_sb_info *sbi) 2906 { 2907 struct curseg_info *array = SM_I(sbi)->curseg_array; 2908 int i; 2909 2910 if (!array) 2911 return; 2912 SM_I(sbi)->curseg_array = NULL; 2913 for (i = 0; i < NR_CURSEG_TYPE; i++) { 2914 kfree(array[i].sum_blk); 2915 kfree(array[i].journal); 2916 } 2917 kfree(array); 2918 } 2919 2920 static void destroy_free_segmap(struct f2fs_sb_info *sbi) 2921 { 2922 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 2923 if (!free_i) 2924 return; 2925 SM_I(sbi)->free_info = NULL; 2926 kvfree(free_i->free_segmap); 2927 kvfree(free_i->free_secmap); 2928 kfree(free_i); 2929 } 2930 2931 static void destroy_sit_info(struct f2fs_sb_info *sbi) 2932 { 2933 struct sit_info *sit_i = SIT_I(sbi); 2934 unsigned int start; 2935 2936 if (!sit_i) 2937 return; 2938 2939 if (sit_i->sentries) { 2940 for (start = 0; start < MAIN_SEGS(sbi); start++) { 2941 kfree(sit_i->sentries[start].cur_valid_map); 2942 #ifdef CONFIG_F2FS_CHECK_FS 2943 kfree(sit_i->sentries[start].cur_valid_map_mir); 2944 #endif 2945 kfree(sit_i->sentries[start].ckpt_valid_map); 2946 kfree(sit_i->sentries[start].discard_map); 2947 } 2948 } 2949 kfree(sit_i->tmp_map); 2950 2951 kvfree(sit_i->sentries); 2952 kvfree(sit_i->sec_entries); 2953 kvfree(sit_i->dirty_sentries_bitmap); 2954 2955 SM_I(sbi)->sit_info = NULL; 2956 kfree(sit_i->sit_bitmap); 2957 #ifdef CONFIG_F2FS_CHECK_FS 2958 kfree(sit_i->sit_bitmap_mir); 2959 #endif 2960 kfree(sit_i); 2961 } 2962 2963 void destroy_segment_manager(struct f2fs_sb_info *sbi) 2964 { 2965 struct f2fs_sm_info *sm_info = SM_I(sbi); 2966 2967 if (!sm_info) 2968 return; 2969 destroy_flush_cmd_control(sbi, true); 2970 destroy_discard_cmd_control(sbi, true); 2971 destroy_dirty_segmap(sbi); 2972 destroy_curseg(sbi); 2973 destroy_free_segmap(sbi); 2974 destroy_sit_info(sbi); 2975 sbi->sm_info = NULL; 2976 kfree(sm_info); 2977 } 2978 2979 int __init create_segment_manager_caches(void) 2980 { 2981 discard_entry_slab = f2fs_kmem_cache_create("discard_entry", 2982 sizeof(struct discard_entry)); 2983 if (!discard_entry_slab) 2984 goto fail; 2985 2986 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd", 2987 sizeof(struct discard_cmd)); 2988 if (!discard_cmd_slab) 2989 goto destroy_discard_entry; 2990 2991 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set", 2992 sizeof(struct sit_entry_set)); 2993 if (!sit_entry_set_slab) 2994 goto destroy_discard_cmd; 2995 2996 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry", 2997 sizeof(struct inmem_pages)); 2998 if (!inmem_entry_slab) 2999 goto destroy_sit_entry_set; 3000 return 0; 3001 3002 destroy_sit_entry_set: 3003 kmem_cache_destroy(sit_entry_set_slab); 3004 destroy_discard_cmd: 3005 kmem_cache_destroy(discard_cmd_slab); 3006 destroy_discard_entry: 3007 kmem_cache_destroy(discard_entry_slab); 3008 fail: 3009 return -ENOMEM; 3010 } 3011 3012 void destroy_segment_manager_caches(void) 3013 { 3014 kmem_cache_destroy(sit_entry_set_slab); 3015 kmem_cache_destroy(discard_cmd_slab); 3016 kmem_cache_destroy(discard_entry_slab); 3017 kmem_cache_destroy(inmem_entry_slab); 3018 } 3019