1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * fs/f2fs/segment.c 4 * 5 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 6 * http://www.samsung.com/ 7 */ 8 #include <linux/fs.h> 9 #include <linux/f2fs_fs.h> 10 #include <linux/bio.h> 11 #include <linux/blkdev.h> 12 #include <linux/prefetch.h> 13 #include <linux/kthread.h> 14 #include <linux/swap.h> 15 #include <linux/timer.h> 16 #include <linux/freezer.h> 17 #include <linux/sched/signal.h> 18 19 #include "f2fs.h" 20 #include "segment.h" 21 #include "node.h" 22 #include "gc.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 bool f2fs_need_SSR(struct f2fs_sb_info *sbi) 170 { 171 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES); 172 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS); 173 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA); 174 175 if (f2fs_lfs_mode(sbi)) 176 return false; 177 if (sbi->gc_mode == GC_URGENT) 178 return true; 179 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 180 return true; 181 182 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs + 183 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi)); 184 } 185 186 void f2fs_register_inmem_page(struct inode *inode, struct page *page) 187 { 188 struct inmem_pages *new; 189 190 f2fs_trace_pid(page); 191 192 f2fs_set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE); 193 194 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS); 195 196 /* add atomic page indices to the list */ 197 new->page = page; 198 INIT_LIST_HEAD(&new->list); 199 200 /* increase reference count with clean state */ 201 get_page(page); 202 mutex_lock(&F2FS_I(inode)->inmem_lock); 203 list_add_tail(&new->list, &F2FS_I(inode)->inmem_pages); 204 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); 205 mutex_unlock(&F2FS_I(inode)->inmem_lock); 206 207 trace_f2fs_register_inmem_page(page, INMEM); 208 } 209 210 static int __revoke_inmem_pages(struct inode *inode, 211 struct list_head *head, bool drop, bool recover, 212 bool trylock) 213 { 214 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 215 struct inmem_pages *cur, *tmp; 216 int err = 0; 217 218 list_for_each_entry_safe(cur, tmp, head, list) { 219 struct page *page = cur->page; 220 221 if (drop) 222 trace_f2fs_commit_inmem_page(page, INMEM_DROP); 223 224 if (trylock) { 225 /* 226 * to avoid deadlock in between page lock and 227 * inmem_lock. 228 */ 229 if (!trylock_page(page)) 230 continue; 231 } else { 232 lock_page(page); 233 } 234 235 f2fs_wait_on_page_writeback(page, DATA, true, true); 236 237 if (recover) { 238 struct dnode_of_data dn; 239 struct node_info ni; 240 241 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE); 242 retry: 243 set_new_dnode(&dn, inode, NULL, NULL, 0); 244 err = f2fs_get_dnode_of_data(&dn, page->index, 245 LOOKUP_NODE); 246 if (err) { 247 if (err == -ENOMEM) { 248 congestion_wait(BLK_RW_ASYNC, 249 DEFAULT_IO_TIMEOUT); 250 cond_resched(); 251 goto retry; 252 } 253 err = -EAGAIN; 254 goto next; 255 } 256 257 err = f2fs_get_node_info(sbi, dn.nid, &ni); 258 if (err) { 259 f2fs_put_dnode(&dn); 260 return err; 261 } 262 263 if (cur->old_addr == NEW_ADDR) { 264 f2fs_invalidate_blocks(sbi, dn.data_blkaddr); 265 f2fs_update_data_blkaddr(&dn, NEW_ADDR); 266 } else 267 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, 268 cur->old_addr, ni.version, true, true); 269 f2fs_put_dnode(&dn); 270 } 271 next: 272 /* we don't need to invalidate this in the sccessful status */ 273 if (drop || recover) { 274 ClearPageUptodate(page); 275 clear_cold_data(page); 276 } 277 f2fs_clear_page_private(page); 278 f2fs_put_page(page, 1); 279 280 list_del(&cur->list); 281 kmem_cache_free(inmem_entry_slab, cur); 282 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); 283 } 284 return err; 285 } 286 287 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure) 288 { 289 struct list_head *head = &sbi->inode_list[ATOMIC_FILE]; 290 struct inode *inode; 291 struct f2fs_inode_info *fi; 292 unsigned int count = sbi->atomic_files; 293 unsigned int looped = 0; 294 next: 295 spin_lock(&sbi->inode_lock[ATOMIC_FILE]); 296 if (list_empty(head)) { 297 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 298 return; 299 } 300 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist); 301 inode = igrab(&fi->vfs_inode); 302 if (inode) 303 list_move_tail(&fi->inmem_ilist, head); 304 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 305 306 if (inode) { 307 if (gc_failure) { 308 if (!fi->i_gc_failures[GC_FAILURE_ATOMIC]) 309 goto skip; 310 } 311 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST); 312 f2fs_drop_inmem_pages(inode); 313 skip: 314 iput(inode); 315 } 316 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT); 317 cond_resched(); 318 if (gc_failure) { 319 if (++looped >= count) 320 return; 321 } 322 goto next; 323 } 324 325 void f2fs_drop_inmem_pages(struct inode *inode) 326 { 327 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 328 struct f2fs_inode_info *fi = F2FS_I(inode); 329 330 while (!list_empty(&fi->inmem_pages)) { 331 mutex_lock(&fi->inmem_lock); 332 __revoke_inmem_pages(inode, &fi->inmem_pages, 333 true, false, true); 334 mutex_unlock(&fi->inmem_lock); 335 } 336 337 fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0; 338 339 spin_lock(&sbi->inode_lock[ATOMIC_FILE]); 340 if (!list_empty(&fi->inmem_ilist)) 341 list_del_init(&fi->inmem_ilist); 342 if (f2fs_is_atomic_file(inode)) { 343 clear_inode_flag(inode, FI_ATOMIC_FILE); 344 sbi->atomic_files--; 345 } 346 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 347 } 348 349 void f2fs_drop_inmem_page(struct inode *inode, struct page *page) 350 { 351 struct f2fs_inode_info *fi = F2FS_I(inode); 352 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 353 struct list_head *head = &fi->inmem_pages; 354 struct inmem_pages *cur = NULL; 355 356 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page)); 357 358 mutex_lock(&fi->inmem_lock); 359 list_for_each_entry(cur, head, list) { 360 if (cur->page == page) 361 break; 362 } 363 364 f2fs_bug_on(sbi, list_empty(head) || cur->page != page); 365 list_del(&cur->list); 366 mutex_unlock(&fi->inmem_lock); 367 368 dec_page_count(sbi, F2FS_INMEM_PAGES); 369 kmem_cache_free(inmem_entry_slab, cur); 370 371 ClearPageUptodate(page); 372 f2fs_clear_page_private(page); 373 f2fs_put_page(page, 0); 374 375 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE); 376 } 377 378 static int __f2fs_commit_inmem_pages(struct inode *inode) 379 { 380 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 381 struct f2fs_inode_info *fi = F2FS_I(inode); 382 struct inmem_pages *cur, *tmp; 383 struct f2fs_io_info fio = { 384 .sbi = sbi, 385 .ino = inode->i_ino, 386 .type = DATA, 387 .op = REQ_OP_WRITE, 388 .op_flags = REQ_SYNC | REQ_PRIO, 389 .io_type = FS_DATA_IO, 390 }; 391 struct list_head revoke_list; 392 bool submit_bio = false; 393 int err = 0; 394 395 INIT_LIST_HEAD(&revoke_list); 396 397 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) { 398 struct page *page = cur->page; 399 400 lock_page(page); 401 if (page->mapping == inode->i_mapping) { 402 trace_f2fs_commit_inmem_page(page, INMEM); 403 404 f2fs_wait_on_page_writeback(page, DATA, true, true); 405 406 set_page_dirty(page); 407 if (clear_page_dirty_for_io(page)) { 408 inode_dec_dirty_pages(inode); 409 f2fs_remove_dirty_inode(inode); 410 } 411 retry: 412 fio.page = page; 413 fio.old_blkaddr = NULL_ADDR; 414 fio.encrypted_page = NULL; 415 fio.need_lock = LOCK_DONE; 416 err = f2fs_do_write_data_page(&fio); 417 if (err) { 418 if (err == -ENOMEM) { 419 congestion_wait(BLK_RW_ASYNC, 420 DEFAULT_IO_TIMEOUT); 421 cond_resched(); 422 goto retry; 423 } 424 unlock_page(page); 425 break; 426 } 427 /* record old blkaddr for revoking */ 428 cur->old_addr = fio.old_blkaddr; 429 submit_bio = true; 430 } 431 unlock_page(page); 432 list_move_tail(&cur->list, &revoke_list); 433 } 434 435 if (submit_bio) 436 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA); 437 438 if (err) { 439 /* 440 * try to revoke all committed pages, but still we could fail 441 * due to no memory or other reason, if that happened, EAGAIN 442 * will be returned, which means in such case, transaction is 443 * already not integrity, caller should use journal to do the 444 * recovery or rewrite & commit last transaction. For other 445 * error number, revoking was done by filesystem itself. 446 */ 447 err = __revoke_inmem_pages(inode, &revoke_list, 448 false, true, false); 449 450 /* drop all uncommitted pages */ 451 __revoke_inmem_pages(inode, &fi->inmem_pages, 452 true, false, false); 453 } else { 454 __revoke_inmem_pages(inode, &revoke_list, 455 false, false, false); 456 } 457 458 return err; 459 } 460 461 int f2fs_commit_inmem_pages(struct inode *inode) 462 { 463 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 464 struct f2fs_inode_info *fi = F2FS_I(inode); 465 int err; 466 467 f2fs_balance_fs(sbi, true); 468 469 down_write(&fi->i_gc_rwsem[WRITE]); 470 471 f2fs_lock_op(sbi); 472 set_inode_flag(inode, FI_ATOMIC_COMMIT); 473 474 mutex_lock(&fi->inmem_lock); 475 err = __f2fs_commit_inmem_pages(inode); 476 mutex_unlock(&fi->inmem_lock); 477 478 clear_inode_flag(inode, FI_ATOMIC_COMMIT); 479 480 f2fs_unlock_op(sbi); 481 up_write(&fi->i_gc_rwsem[WRITE]); 482 483 return err; 484 } 485 486 /* 487 * This function balances dirty node and dentry pages. 488 * In addition, it controls garbage collection. 489 */ 490 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need) 491 { 492 if (time_to_inject(sbi, FAULT_CHECKPOINT)) { 493 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT); 494 f2fs_stop_checkpoint(sbi, false); 495 } 496 497 /* balance_fs_bg is able to be pending */ 498 if (need && excess_cached_nats(sbi)) 499 f2fs_balance_fs_bg(sbi, false); 500 501 if (!f2fs_is_checkpoint_ready(sbi)) 502 return; 503 504 /* 505 * We should do GC or end up with checkpoint, if there are so many dirty 506 * dir/node pages without enough free segments. 507 */ 508 if (has_not_enough_free_secs(sbi, 0, 0)) { 509 down_write(&sbi->gc_lock); 510 f2fs_gc(sbi, false, false, NULL_SEGNO); 511 } 512 } 513 514 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg) 515 { 516 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 517 return; 518 519 /* try to shrink extent cache when there is no enough memory */ 520 if (!f2fs_available_free_memory(sbi, EXTENT_CACHE)) 521 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER); 522 523 /* check the # of cached NAT entries */ 524 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES)) 525 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK); 526 527 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) 528 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS); 529 else 530 f2fs_build_free_nids(sbi, false, false); 531 532 if (!is_idle(sbi, REQ_TIME) && 533 (!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi))) 534 return; 535 536 /* checkpoint is the only way to shrink partial cached entries */ 537 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) || 538 !f2fs_available_free_memory(sbi, INO_ENTRIES) || 539 excess_prefree_segs(sbi) || 540 excess_dirty_nats(sbi) || 541 excess_dirty_nodes(sbi) || 542 f2fs_time_over(sbi, CP_TIME)) { 543 if (test_opt(sbi, DATA_FLUSH) && from_bg) { 544 struct blk_plug plug; 545 546 mutex_lock(&sbi->flush_lock); 547 548 blk_start_plug(&plug); 549 f2fs_sync_dirty_inodes(sbi, FILE_INODE); 550 blk_finish_plug(&plug); 551 552 mutex_unlock(&sbi->flush_lock); 553 } 554 f2fs_sync_fs(sbi->sb, true); 555 stat_inc_bg_cp_count(sbi->stat_info); 556 } 557 } 558 559 static int __submit_flush_wait(struct f2fs_sb_info *sbi, 560 struct block_device *bdev) 561 { 562 struct bio *bio; 563 int ret; 564 565 bio = f2fs_bio_alloc(sbi, 0, false); 566 if (!bio) 567 return -ENOMEM; 568 569 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH; 570 bio_set_dev(bio, bdev); 571 ret = submit_bio_wait(bio); 572 bio_put(bio); 573 574 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER), 575 test_opt(sbi, FLUSH_MERGE), ret); 576 return ret; 577 } 578 579 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino) 580 { 581 int ret = 0; 582 int i; 583 584 if (!f2fs_is_multi_device(sbi)) 585 return __submit_flush_wait(sbi, sbi->sb->s_bdev); 586 587 for (i = 0; i < sbi->s_ndevs; i++) { 588 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO)) 589 continue; 590 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 591 if (ret) 592 break; 593 } 594 return ret; 595 } 596 597 static int issue_flush_thread(void *data) 598 { 599 struct f2fs_sb_info *sbi = data; 600 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 601 wait_queue_head_t *q = &fcc->flush_wait_queue; 602 repeat: 603 if (kthread_should_stop()) 604 return 0; 605 606 sb_start_intwrite(sbi->sb); 607 608 if (!llist_empty(&fcc->issue_list)) { 609 struct flush_cmd *cmd, *next; 610 int ret; 611 612 fcc->dispatch_list = llist_del_all(&fcc->issue_list); 613 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); 614 615 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode); 616 617 ret = submit_flush_wait(sbi, cmd->ino); 618 atomic_inc(&fcc->issued_flush); 619 620 llist_for_each_entry_safe(cmd, next, 621 fcc->dispatch_list, llnode) { 622 cmd->ret = ret; 623 complete(&cmd->wait); 624 } 625 fcc->dispatch_list = NULL; 626 } 627 628 sb_end_intwrite(sbi->sb); 629 630 wait_event_interruptible(*q, 631 kthread_should_stop() || !llist_empty(&fcc->issue_list)); 632 goto repeat; 633 } 634 635 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino) 636 { 637 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 638 struct flush_cmd cmd; 639 int ret; 640 641 if (test_opt(sbi, NOBARRIER)) 642 return 0; 643 644 if (!test_opt(sbi, FLUSH_MERGE)) { 645 atomic_inc(&fcc->queued_flush); 646 ret = submit_flush_wait(sbi, ino); 647 atomic_dec(&fcc->queued_flush); 648 atomic_inc(&fcc->issued_flush); 649 return ret; 650 } 651 652 if (atomic_inc_return(&fcc->queued_flush) == 1 || 653 f2fs_is_multi_device(sbi)) { 654 ret = submit_flush_wait(sbi, ino); 655 atomic_dec(&fcc->queued_flush); 656 657 atomic_inc(&fcc->issued_flush); 658 return ret; 659 } 660 661 cmd.ino = ino; 662 init_completion(&cmd.wait); 663 664 llist_add(&cmd.llnode, &fcc->issue_list); 665 666 /* update issue_list before we wake up issue_flush thread */ 667 smp_mb(); 668 669 if (waitqueue_active(&fcc->flush_wait_queue)) 670 wake_up(&fcc->flush_wait_queue); 671 672 if (fcc->f2fs_issue_flush) { 673 wait_for_completion(&cmd.wait); 674 atomic_dec(&fcc->queued_flush); 675 } else { 676 struct llist_node *list; 677 678 list = llist_del_all(&fcc->issue_list); 679 if (!list) { 680 wait_for_completion(&cmd.wait); 681 atomic_dec(&fcc->queued_flush); 682 } else { 683 struct flush_cmd *tmp, *next; 684 685 ret = submit_flush_wait(sbi, ino); 686 687 llist_for_each_entry_safe(tmp, next, list, llnode) { 688 if (tmp == &cmd) { 689 cmd.ret = ret; 690 atomic_dec(&fcc->queued_flush); 691 continue; 692 } 693 tmp->ret = ret; 694 complete(&tmp->wait); 695 } 696 } 697 } 698 699 return cmd.ret; 700 } 701 702 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi) 703 { 704 dev_t dev = sbi->sb->s_bdev->bd_dev; 705 struct flush_cmd_control *fcc; 706 int err = 0; 707 708 if (SM_I(sbi)->fcc_info) { 709 fcc = SM_I(sbi)->fcc_info; 710 if (fcc->f2fs_issue_flush) 711 return err; 712 goto init_thread; 713 } 714 715 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL); 716 if (!fcc) 717 return -ENOMEM; 718 atomic_set(&fcc->issued_flush, 0); 719 atomic_set(&fcc->queued_flush, 0); 720 init_waitqueue_head(&fcc->flush_wait_queue); 721 init_llist_head(&fcc->issue_list); 722 SM_I(sbi)->fcc_info = fcc; 723 if (!test_opt(sbi, FLUSH_MERGE)) 724 return err; 725 726 init_thread: 727 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, 728 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); 729 if (IS_ERR(fcc->f2fs_issue_flush)) { 730 err = PTR_ERR(fcc->f2fs_issue_flush); 731 kvfree(fcc); 732 SM_I(sbi)->fcc_info = NULL; 733 return err; 734 } 735 736 return err; 737 } 738 739 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free) 740 { 741 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 742 743 if (fcc && fcc->f2fs_issue_flush) { 744 struct task_struct *flush_thread = fcc->f2fs_issue_flush; 745 746 fcc->f2fs_issue_flush = NULL; 747 kthread_stop(flush_thread); 748 } 749 if (free) { 750 kvfree(fcc); 751 SM_I(sbi)->fcc_info = NULL; 752 } 753 } 754 755 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi) 756 { 757 int ret = 0, i; 758 759 if (!f2fs_is_multi_device(sbi)) 760 return 0; 761 762 for (i = 1; i < sbi->s_ndevs; i++) { 763 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device)) 764 continue; 765 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 766 if (ret) 767 break; 768 769 spin_lock(&sbi->dev_lock); 770 f2fs_clear_bit(i, (char *)&sbi->dirty_device); 771 spin_unlock(&sbi->dev_lock); 772 } 773 774 return ret; 775 } 776 777 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 778 enum dirty_type dirty_type) 779 { 780 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 781 782 /* need not be added */ 783 if (IS_CURSEG(sbi, segno)) 784 return; 785 786 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 787 dirty_i->nr_dirty[dirty_type]++; 788 789 if (dirty_type == DIRTY) { 790 struct seg_entry *sentry = get_seg_entry(sbi, segno); 791 enum dirty_type t = sentry->type; 792 793 if (unlikely(t >= DIRTY)) { 794 f2fs_bug_on(sbi, 1); 795 return; 796 } 797 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) 798 dirty_i->nr_dirty[t]++; 799 } 800 } 801 802 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 803 enum dirty_type dirty_type) 804 { 805 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 806 807 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) 808 dirty_i->nr_dirty[dirty_type]--; 809 810 if (dirty_type == DIRTY) { 811 struct seg_entry *sentry = get_seg_entry(sbi, segno); 812 enum dirty_type t = sentry->type; 813 814 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) 815 dirty_i->nr_dirty[t]--; 816 817 if (get_valid_blocks(sbi, segno, true) == 0) { 818 clear_bit(GET_SEC_FROM_SEG(sbi, segno), 819 dirty_i->victim_secmap); 820 #ifdef CONFIG_F2FS_CHECK_FS 821 clear_bit(segno, SIT_I(sbi)->invalid_segmap); 822 #endif 823 } 824 } 825 } 826 827 /* 828 * Should not occur error such as -ENOMEM. 829 * Adding dirty entry into seglist is not critical operation. 830 * If a given segment is one of current working segments, it won't be added. 831 */ 832 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) 833 { 834 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 835 unsigned short valid_blocks, ckpt_valid_blocks; 836 837 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) 838 return; 839 840 mutex_lock(&dirty_i->seglist_lock); 841 842 valid_blocks = get_valid_blocks(sbi, segno, false); 843 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno); 844 845 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) || 846 ckpt_valid_blocks == sbi->blocks_per_seg)) { 847 __locate_dirty_segment(sbi, segno, PRE); 848 __remove_dirty_segment(sbi, segno, DIRTY); 849 } else if (valid_blocks < sbi->blocks_per_seg) { 850 __locate_dirty_segment(sbi, segno, DIRTY); 851 } else { 852 /* Recovery routine with SSR needs this */ 853 __remove_dirty_segment(sbi, segno, DIRTY); 854 } 855 856 mutex_unlock(&dirty_i->seglist_lock); 857 } 858 859 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */ 860 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi) 861 { 862 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 863 unsigned int segno; 864 865 mutex_lock(&dirty_i->seglist_lock); 866 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 867 if (get_valid_blocks(sbi, segno, false)) 868 continue; 869 if (IS_CURSEG(sbi, segno)) 870 continue; 871 __locate_dirty_segment(sbi, segno, PRE); 872 __remove_dirty_segment(sbi, segno, DIRTY); 873 } 874 mutex_unlock(&dirty_i->seglist_lock); 875 } 876 877 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi) 878 { 879 int ovp_hole_segs = 880 (overprovision_segments(sbi) - reserved_segments(sbi)); 881 block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg; 882 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 883 block_t holes[2] = {0, 0}; /* DATA and NODE */ 884 block_t unusable; 885 struct seg_entry *se; 886 unsigned int segno; 887 888 mutex_lock(&dirty_i->seglist_lock); 889 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 890 se = get_seg_entry(sbi, segno); 891 if (IS_NODESEG(se->type)) 892 holes[NODE] += sbi->blocks_per_seg - se->valid_blocks; 893 else 894 holes[DATA] += sbi->blocks_per_seg - se->valid_blocks; 895 } 896 mutex_unlock(&dirty_i->seglist_lock); 897 898 unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE]; 899 if (unusable > ovp_holes) 900 return unusable - ovp_holes; 901 return 0; 902 } 903 904 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable) 905 { 906 int ovp_hole_segs = 907 (overprovision_segments(sbi) - reserved_segments(sbi)); 908 if (unusable > F2FS_OPTION(sbi).unusable_cap) 909 return -EAGAIN; 910 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) && 911 dirty_segments(sbi) > ovp_hole_segs) 912 return -EAGAIN; 913 return 0; 914 } 915 916 /* This is only used by SBI_CP_DISABLED */ 917 static unsigned int get_free_segment(struct f2fs_sb_info *sbi) 918 { 919 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 920 unsigned int segno = 0; 921 922 mutex_lock(&dirty_i->seglist_lock); 923 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 924 if (get_valid_blocks(sbi, segno, false)) 925 continue; 926 if (get_ckpt_valid_blocks(sbi, segno)) 927 continue; 928 mutex_unlock(&dirty_i->seglist_lock); 929 return segno; 930 } 931 mutex_unlock(&dirty_i->seglist_lock); 932 return NULL_SEGNO; 933 } 934 935 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi, 936 struct block_device *bdev, block_t lstart, 937 block_t start, block_t len) 938 { 939 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 940 struct list_head *pend_list; 941 struct discard_cmd *dc; 942 943 f2fs_bug_on(sbi, !len); 944 945 pend_list = &dcc->pend_list[plist_idx(len)]; 946 947 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS); 948 INIT_LIST_HEAD(&dc->list); 949 dc->bdev = bdev; 950 dc->lstart = lstart; 951 dc->start = start; 952 dc->len = len; 953 dc->ref = 0; 954 dc->state = D_PREP; 955 dc->queued = 0; 956 dc->error = 0; 957 init_completion(&dc->wait); 958 list_add_tail(&dc->list, pend_list); 959 spin_lock_init(&dc->lock); 960 dc->bio_ref = 0; 961 atomic_inc(&dcc->discard_cmd_cnt); 962 dcc->undiscard_blks += len; 963 964 return dc; 965 } 966 967 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi, 968 struct block_device *bdev, block_t lstart, 969 block_t start, block_t len, 970 struct rb_node *parent, struct rb_node **p, 971 bool leftmost) 972 { 973 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 974 struct discard_cmd *dc; 975 976 dc = __create_discard_cmd(sbi, bdev, lstart, start, len); 977 978 rb_link_node(&dc->rb_node, parent, p); 979 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost); 980 981 return dc; 982 } 983 984 static void __detach_discard_cmd(struct discard_cmd_control *dcc, 985 struct discard_cmd *dc) 986 { 987 if (dc->state == D_DONE) 988 atomic_sub(dc->queued, &dcc->queued_discard); 989 990 list_del(&dc->list); 991 rb_erase_cached(&dc->rb_node, &dcc->root); 992 dcc->undiscard_blks -= dc->len; 993 994 kmem_cache_free(discard_cmd_slab, dc); 995 996 atomic_dec(&dcc->discard_cmd_cnt); 997 } 998 999 static void __remove_discard_cmd(struct f2fs_sb_info *sbi, 1000 struct discard_cmd *dc) 1001 { 1002 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1003 unsigned long flags; 1004 1005 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len); 1006 1007 spin_lock_irqsave(&dc->lock, flags); 1008 if (dc->bio_ref) { 1009 spin_unlock_irqrestore(&dc->lock, flags); 1010 return; 1011 } 1012 spin_unlock_irqrestore(&dc->lock, flags); 1013 1014 f2fs_bug_on(sbi, dc->ref); 1015 1016 if (dc->error == -EOPNOTSUPP) 1017 dc->error = 0; 1018 1019 if (dc->error) 1020 printk_ratelimited( 1021 "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d", 1022 KERN_INFO, sbi->sb->s_id, 1023 dc->lstart, dc->start, dc->len, dc->error); 1024 __detach_discard_cmd(dcc, dc); 1025 } 1026 1027 static void f2fs_submit_discard_endio(struct bio *bio) 1028 { 1029 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private; 1030 unsigned long flags; 1031 1032 spin_lock_irqsave(&dc->lock, flags); 1033 if (!dc->error) 1034 dc->error = blk_status_to_errno(bio->bi_status); 1035 dc->bio_ref--; 1036 if (!dc->bio_ref && dc->state == D_SUBMIT) { 1037 dc->state = D_DONE; 1038 complete_all(&dc->wait); 1039 } 1040 spin_unlock_irqrestore(&dc->lock, flags); 1041 bio_put(bio); 1042 } 1043 1044 static void __check_sit_bitmap(struct f2fs_sb_info *sbi, 1045 block_t start, block_t end) 1046 { 1047 #ifdef CONFIG_F2FS_CHECK_FS 1048 struct seg_entry *sentry; 1049 unsigned int segno; 1050 block_t blk = start; 1051 unsigned long offset, size, max_blocks = sbi->blocks_per_seg; 1052 unsigned long *map; 1053 1054 while (blk < end) { 1055 segno = GET_SEGNO(sbi, blk); 1056 sentry = get_seg_entry(sbi, segno); 1057 offset = GET_BLKOFF_FROM_SEG0(sbi, blk); 1058 1059 if (end < START_BLOCK(sbi, segno + 1)) 1060 size = GET_BLKOFF_FROM_SEG0(sbi, end); 1061 else 1062 size = max_blocks; 1063 map = (unsigned long *)(sentry->cur_valid_map); 1064 offset = __find_rev_next_bit(map, size, offset); 1065 f2fs_bug_on(sbi, offset != size); 1066 blk = START_BLOCK(sbi, segno + 1); 1067 } 1068 #endif 1069 } 1070 1071 static void __init_discard_policy(struct f2fs_sb_info *sbi, 1072 struct discard_policy *dpolicy, 1073 int discard_type, unsigned int granularity) 1074 { 1075 /* common policy */ 1076 dpolicy->type = discard_type; 1077 dpolicy->sync = true; 1078 dpolicy->ordered = false; 1079 dpolicy->granularity = granularity; 1080 1081 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST; 1082 dpolicy->io_aware_gran = MAX_PLIST_NUM; 1083 dpolicy->timeout = false; 1084 1085 if (discard_type == DPOLICY_BG) { 1086 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME; 1087 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME; 1088 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME; 1089 dpolicy->io_aware = true; 1090 dpolicy->sync = false; 1091 dpolicy->ordered = true; 1092 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) { 1093 dpolicy->granularity = 1; 1094 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME; 1095 } 1096 } else if (discard_type == DPOLICY_FORCE) { 1097 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME; 1098 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME; 1099 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME; 1100 dpolicy->io_aware = false; 1101 } else if (discard_type == DPOLICY_FSTRIM) { 1102 dpolicy->io_aware = false; 1103 } else if (discard_type == DPOLICY_UMOUNT) { 1104 dpolicy->io_aware = false; 1105 /* we need to issue all to keep CP_TRIMMED_FLAG */ 1106 dpolicy->granularity = 1; 1107 dpolicy->timeout = true; 1108 } 1109 } 1110 1111 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1112 struct block_device *bdev, block_t lstart, 1113 block_t start, block_t len); 1114 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */ 1115 static int __submit_discard_cmd(struct f2fs_sb_info *sbi, 1116 struct discard_policy *dpolicy, 1117 struct discard_cmd *dc, 1118 unsigned int *issued) 1119 { 1120 struct block_device *bdev = dc->bdev; 1121 struct request_queue *q = bdev_get_queue(bdev); 1122 unsigned int max_discard_blocks = 1123 SECTOR_TO_BLOCK(q->limits.max_discard_sectors); 1124 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1125 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1126 &(dcc->fstrim_list) : &(dcc->wait_list); 1127 int flag = dpolicy->sync ? REQ_SYNC : 0; 1128 block_t lstart, start, len, total_len; 1129 int err = 0; 1130 1131 if (dc->state != D_PREP) 1132 return 0; 1133 1134 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 1135 return 0; 1136 1137 trace_f2fs_issue_discard(bdev, dc->start, dc->len); 1138 1139 lstart = dc->lstart; 1140 start = dc->start; 1141 len = dc->len; 1142 total_len = len; 1143 1144 dc->len = 0; 1145 1146 while (total_len && *issued < dpolicy->max_requests && !err) { 1147 struct bio *bio = NULL; 1148 unsigned long flags; 1149 bool last = true; 1150 1151 if (len > max_discard_blocks) { 1152 len = max_discard_blocks; 1153 last = false; 1154 } 1155 1156 (*issued)++; 1157 if (*issued == dpolicy->max_requests) 1158 last = true; 1159 1160 dc->len += len; 1161 1162 if (time_to_inject(sbi, FAULT_DISCARD)) { 1163 f2fs_show_injection_info(sbi, FAULT_DISCARD); 1164 err = -EIO; 1165 goto submit; 1166 } 1167 err = __blkdev_issue_discard(bdev, 1168 SECTOR_FROM_BLOCK(start), 1169 SECTOR_FROM_BLOCK(len), 1170 GFP_NOFS, 0, &bio); 1171 submit: 1172 if (err) { 1173 spin_lock_irqsave(&dc->lock, flags); 1174 if (dc->state == D_PARTIAL) 1175 dc->state = D_SUBMIT; 1176 spin_unlock_irqrestore(&dc->lock, flags); 1177 1178 break; 1179 } 1180 1181 f2fs_bug_on(sbi, !bio); 1182 1183 /* 1184 * should keep before submission to avoid D_DONE 1185 * right away 1186 */ 1187 spin_lock_irqsave(&dc->lock, flags); 1188 if (last) 1189 dc->state = D_SUBMIT; 1190 else 1191 dc->state = D_PARTIAL; 1192 dc->bio_ref++; 1193 spin_unlock_irqrestore(&dc->lock, flags); 1194 1195 atomic_inc(&dcc->queued_discard); 1196 dc->queued++; 1197 list_move_tail(&dc->list, wait_list); 1198 1199 /* sanity check on discard range */ 1200 __check_sit_bitmap(sbi, lstart, lstart + len); 1201 1202 bio->bi_private = dc; 1203 bio->bi_end_io = f2fs_submit_discard_endio; 1204 bio->bi_opf |= flag; 1205 submit_bio(bio); 1206 1207 atomic_inc(&dcc->issued_discard); 1208 1209 f2fs_update_iostat(sbi, FS_DISCARD, 1); 1210 1211 lstart += len; 1212 start += len; 1213 total_len -= len; 1214 len = total_len; 1215 } 1216 1217 if (!err && len) { 1218 dcc->undiscard_blks -= len; 1219 __update_discard_tree_range(sbi, bdev, lstart, start, len); 1220 } 1221 return err; 1222 } 1223 1224 static void __insert_discard_tree(struct f2fs_sb_info *sbi, 1225 struct block_device *bdev, block_t lstart, 1226 block_t start, block_t len, 1227 struct rb_node **insert_p, 1228 struct rb_node *insert_parent) 1229 { 1230 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1231 struct rb_node **p; 1232 struct rb_node *parent = NULL; 1233 bool leftmost = true; 1234 1235 if (insert_p && insert_parent) { 1236 parent = insert_parent; 1237 p = insert_p; 1238 goto do_insert; 1239 } 1240 1241 p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, 1242 lstart, &leftmost); 1243 do_insert: 1244 __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, 1245 p, leftmost); 1246 } 1247 1248 static void __relocate_discard_cmd(struct discard_cmd_control *dcc, 1249 struct discard_cmd *dc) 1250 { 1251 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]); 1252 } 1253 1254 static void __punch_discard_cmd(struct f2fs_sb_info *sbi, 1255 struct discard_cmd *dc, block_t blkaddr) 1256 { 1257 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1258 struct discard_info di = dc->di; 1259 bool modified = false; 1260 1261 if (dc->state == D_DONE || dc->len == 1) { 1262 __remove_discard_cmd(sbi, dc); 1263 return; 1264 } 1265 1266 dcc->undiscard_blks -= di.len; 1267 1268 if (blkaddr > di.lstart) { 1269 dc->len = blkaddr - dc->lstart; 1270 dcc->undiscard_blks += dc->len; 1271 __relocate_discard_cmd(dcc, dc); 1272 modified = true; 1273 } 1274 1275 if (blkaddr < di.lstart + di.len - 1) { 1276 if (modified) { 1277 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1, 1278 di.start + blkaddr + 1 - di.lstart, 1279 di.lstart + di.len - 1 - blkaddr, 1280 NULL, NULL); 1281 } else { 1282 dc->lstart++; 1283 dc->len--; 1284 dc->start++; 1285 dcc->undiscard_blks += dc->len; 1286 __relocate_discard_cmd(dcc, dc); 1287 } 1288 } 1289 } 1290 1291 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1292 struct block_device *bdev, block_t lstart, 1293 block_t start, block_t len) 1294 { 1295 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1296 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1297 struct discard_cmd *dc; 1298 struct discard_info di = {0}; 1299 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1300 struct request_queue *q = bdev_get_queue(bdev); 1301 unsigned int max_discard_blocks = 1302 SECTOR_TO_BLOCK(q->limits.max_discard_sectors); 1303 block_t end = lstart + len; 1304 1305 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root, 1306 NULL, lstart, 1307 (struct rb_entry **)&prev_dc, 1308 (struct rb_entry **)&next_dc, 1309 &insert_p, &insert_parent, true, NULL); 1310 if (dc) 1311 prev_dc = dc; 1312 1313 if (!prev_dc) { 1314 di.lstart = lstart; 1315 di.len = next_dc ? next_dc->lstart - lstart : len; 1316 di.len = min(di.len, len); 1317 di.start = start; 1318 } 1319 1320 while (1) { 1321 struct rb_node *node; 1322 bool merged = false; 1323 struct discard_cmd *tdc = NULL; 1324 1325 if (prev_dc) { 1326 di.lstart = prev_dc->lstart + prev_dc->len; 1327 if (di.lstart < lstart) 1328 di.lstart = lstart; 1329 if (di.lstart >= end) 1330 break; 1331 1332 if (!next_dc || next_dc->lstart > end) 1333 di.len = end - di.lstart; 1334 else 1335 di.len = next_dc->lstart - di.lstart; 1336 di.start = start + di.lstart - lstart; 1337 } 1338 1339 if (!di.len) 1340 goto next; 1341 1342 if (prev_dc && prev_dc->state == D_PREP && 1343 prev_dc->bdev == bdev && 1344 __is_discard_back_mergeable(&di, &prev_dc->di, 1345 max_discard_blocks)) { 1346 prev_dc->di.len += di.len; 1347 dcc->undiscard_blks += di.len; 1348 __relocate_discard_cmd(dcc, prev_dc); 1349 di = prev_dc->di; 1350 tdc = prev_dc; 1351 merged = true; 1352 } 1353 1354 if (next_dc && next_dc->state == D_PREP && 1355 next_dc->bdev == bdev && 1356 __is_discard_front_mergeable(&di, &next_dc->di, 1357 max_discard_blocks)) { 1358 next_dc->di.lstart = di.lstart; 1359 next_dc->di.len += di.len; 1360 next_dc->di.start = di.start; 1361 dcc->undiscard_blks += di.len; 1362 __relocate_discard_cmd(dcc, next_dc); 1363 if (tdc) 1364 __remove_discard_cmd(sbi, tdc); 1365 merged = true; 1366 } 1367 1368 if (!merged) { 1369 __insert_discard_tree(sbi, bdev, di.lstart, di.start, 1370 di.len, NULL, NULL); 1371 } 1372 next: 1373 prev_dc = next_dc; 1374 if (!prev_dc) 1375 break; 1376 1377 node = rb_next(&prev_dc->rb_node); 1378 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1379 } 1380 } 1381 1382 static int __queue_discard_cmd(struct f2fs_sb_info *sbi, 1383 struct block_device *bdev, block_t blkstart, block_t blklen) 1384 { 1385 block_t lblkstart = blkstart; 1386 1387 if (!f2fs_bdev_support_discard(bdev)) 1388 return 0; 1389 1390 trace_f2fs_queue_discard(bdev, blkstart, blklen); 1391 1392 if (f2fs_is_multi_device(sbi)) { 1393 int devi = f2fs_target_device_index(sbi, blkstart); 1394 1395 blkstart -= FDEV(devi).start_blk; 1396 } 1397 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock); 1398 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen); 1399 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock); 1400 return 0; 1401 } 1402 1403 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi, 1404 struct discard_policy *dpolicy) 1405 { 1406 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1407 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1408 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1409 struct discard_cmd *dc; 1410 struct blk_plug plug; 1411 unsigned int pos = dcc->next_pos; 1412 unsigned int issued = 0; 1413 bool io_interrupted = false; 1414 1415 mutex_lock(&dcc->cmd_lock); 1416 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root, 1417 NULL, pos, 1418 (struct rb_entry **)&prev_dc, 1419 (struct rb_entry **)&next_dc, 1420 &insert_p, &insert_parent, true, NULL); 1421 if (!dc) 1422 dc = next_dc; 1423 1424 blk_start_plug(&plug); 1425 1426 while (dc) { 1427 struct rb_node *node; 1428 int err = 0; 1429 1430 if (dc->state != D_PREP) 1431 goto next; 1432 1433 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) { 1434 io_interrupted = true; 1435 break; 1436 } 1437 1438 dcc->next_pos = dc->lstart + dc->len; 1439 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued); 1440 1441 if (issued >= dpolicy->max_requests) 1442 break; 1443 next: 1444 node = rb_next(&dc->rb_node); 1445 if (err) 1446 __remove_discard_cmd(sbi, dc); 1447 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1448 } 1449 1450 blk_finish_plug(&plug); 1451 1452 if (!dc) 1453 dcc->next_pos = 0; 1454 1455 mutex_unlock(&dcc->cmd_lock); 1456 1457 if (!issued && io_interrupted) 1458 issued = -1; 1459 1460 return issued; 1461 } 1462 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1463 struct discard_policy *dpolicy); 1464 1465 static int __issue_discard_cmd(struct f2fs_sb_info *sbi, 1466 struct discard_policy *dpolicy) 1467 { 1468 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1469 struct list_head *pend_list; 1470 struct discard_cmd *dc, *tmp; 1471 struct blk_plug plug; 1472 int i, issued; 1473 bool io_interrupted = false; 1474 1475 if (dpolicy->timeout) 1476 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT); 1477 1478 retry: 1479 issued = 0; 1480 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1481 if (dpolicy->timeout && 1482 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1483 break; 1484 1485 if (i + 1 < dpolicy->granularity) 1486 break; 1487 1488 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered) 1489 return __issue_discard_cmd_orderly(sbi, dpolicy); 1490 1491 pend_list = &dcc->pend_list[i]; 1492 1493 mutex_lock(&dcc->cmd_lock); 1494 if (list_empty(pend_list)) 1495 goto next; 1496 if (unlikely(dcc->rbtree_check)) 1497 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi, 1498 &dcc->root)); 1499 blk_start_plug(&plug); 1500 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1501 f2fs_bug_on(sbi, dc->state != D_PREP); 1502 1503 if (dpolicy->timeout && 1504 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1505 break; 1506 1507 if (dpolicy->io_aware && i < dpolicy->io_aware_gran && 1508 !is_idle(sbi, DISCARD_TIME)) { 1509 io_interrupted = true; 1510 break; 1511 } 1512 1513 __submit_discard_cmd(sbi, dpolicy, dc, &issued); 1514 1515 if (issued >= dpolicy->max_requests) 1516 break; 1517 } 1518 blk_finish_plug(&plug); 1519 next: 1520 mutex_unlock(&dcc->cmd_lock); 1521 1522 if (issued >= dpolicy->max_requests || io_interrupted) 1523 break; 1524 } 1525 1526 if (dpolicy->type == DPOLICY_UMOUNT && issued) { 1527 __wait_all_discard_cmd(sbi, dpolicy); 1528 goto retry; 1529 } 1530 1531 if (!issued && io_interrupted) 1532 issued = -1; 1533 1534 return issued; 1535 } 1536 1537 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi) 1538 { 1539 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1540 struct list_head *pend_list; 1541 struct discard_cmd *dc, *tmp; 1542 int i; 1543 bool dropped = false; 1544 1545 mutex_lock(&dcc->cmd_lock); 1546 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1547 pend_list = &dcc->pend_list[i]; 1548 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1549 f2fs_bug_on(sbi, dc->state != D_PREP); 1550 __remove_discard_cmd(sbi, dc); 1551 dropped = true; 1552 } 1553 } 1554 mutex_unlock(&dcc->cmd_lock); 1555 1556 return dropped; 1557 } 1558 1559 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi) 1560 { 1561 __drop_discard_cmd(sbi); 1562 } 1563 1564 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi, 1565 struct discard_cmd *dc) 1566 { 1567 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1568 unsigned int len = 0; 1569 1570 wait_for_completion_io(&dc->wait); 1571 mutex_lock(&dcc->cmd_lock); 1572 f2fs_bug_on(sbi, dc->state != D_DONE); 1573 dc->ref--; 1574 if (!dc->ref) { 1575 if (!dc->error) 1576 len = dc->len; 1577 __remove_discard_cmd(sbi, dc); 1578 } 1579 mutex_unlock(&dcc->cmd_lock); 1580 1581 return len; 1582 } 1583 1584 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi, 1585 struct discard_policy *dpolicy, 1586 block_t start, block_t end) 1587 { 1588 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1589 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1590 &(dcc->fstrim_list) : &(dcc->wait_list); 1591 struct discard_cmd *dc, *tmp; 1592 bool need_wait; 1593 unsigned int trimmed = 0; 1594 1595 next: 1596 need_wait = false; 1597 1598 mutex_lock(&dcc->cmd_lock); 1599 list_for_each_entry_safe(dc, tmp, wait_list, list) { 1600 if (dc->lstart + dc->len <= start || end <= dc->lstart) 1601 continue; 1602 if (dc->len < dpolicy->granularity) 1603 continue; 1604 if (dc->state == D_DONE && !dc->ref) { 1605 wait_for_completion_io(&dc->wait); 1606 if (!dc->error) 1607 trimmed += dc->len; 1608 __remove_discard_cmd(sbi, dc); 1609 } else { 1610 dc->ref++; 1611 need_wait = true; 1612 break; 1613 } 1614 } 1615 mutex_unlock(&dcc->cmd_lock); 1616 1617 if (need_wait) { 1618 trimmed += __wait_one_discard_bio(sbi, dc); 1619 goto next; 1620 } 1621 1622 return trimmed; 1623 } 1624 1625 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1626 struct discard_policy *dpolicy) 1627 { 1628 struct discard_policy dp; 1629 unsigned int discard_blks; 1630 1631 if (dpolicy) 1632 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX); 1633 1634 /* wait all */ 1635 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1); 1636 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1637 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1); 1638 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1639 1640 return discard_blks; 1641 } 1642 1643 /* This should be covered by global mutex, &sit_i->sentry_lock */ 1644 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr) 1645 { 1646 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1647 struct discard_cmd *dc; 1648 bool need_wait = false; 1649 1650 mutex_lock(&dcc->cmd_lock); 1651 dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root, 1652 NULL, blkaddr); 1653 if (dc) { 1654 if (dc->state == D_PREP) { 1655 __punch_discard_cmd(sbi, dc, blkaddr); 1656 } else { 1657 dc->ref++; 1658 need_wait = true; 1659 } 1660 } 1661 mutex_unlock(&dcc->cmd_lock); 1662 1663 if (need_wait) 1664 __wait_one_discard_bio(sbi, dc); 1665 } 1666 1667 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi) 1668 { 1669 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1670 1671 if (dcc && dcc->f2fs_issue_discard) { 1672 struct task_struct *discard_thread = dcc->f2fs_issue_discard; 1673 1674 dcc->f2fs_issue_discard = NULL; 1675 kthread_stop(discard_thread); 1676 } 1677 } 1678 1679 /* This comes from f2fs_put_super */ 1680 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi) 1681 { 1682 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1683 struct discard_policy dpolicy; 1684 bool dropped; 1685 1686 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT, 1687 dcc->discard_granularity); 1688 __issue_discard_cmd(sbi, &dpolicy); 1689 dropped = __drop_discard_cmd(sbi); 1690 1691 /* just to make sure there is no pending discard commands */ 1692 __wait_all_discard_cmd(sbi, NULL); 1693 1694 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt)); 1695 return dropped; 1696 } 1697 1698 static int issue_discard_thread(void *data) 1699 { 1700 struct f2fs_sb_info *sbi = data; 1701 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1702 wait_queue_head_t *q = &dcc->discard_wait_queue; 1703 struct discard_policy dpolicy; 1704 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME; 1705 int issued; 1706 1707 set_freezable(); 1708 1709 do { 1710 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG, 1711 dcc->discard_granularity); 1712 1713 wait_event_interruptible_timeout(*q, 1714 kthread_should_stop() || freezing(current) || 1715 dcc->discard_wake, 1716 msecs_to_jiffies(wait_ms)); 1717 1718 if (dcc->discard_wake) 1719 dcc->discard_wake = 0; 1720 1721 /* clean up pending candidates before going to sleep */ 1722 if (atomic_read(&dcc->queued_discard)) 1723 __wait_all_discard_cmd(sbi, NULL); 1724 1725 if (try_to_freeze()) 1726 continue; 1727 if (f2fs_readonly(sbi->sb)) 1728 continue; 1729 if (kthread_should_stop()) 1730 return 0; 1731 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 1732 wait_ms = dpolicy.max_interval; 1733 continue; 1734 } 1735 1736 if (sbi->gc_mode == GC_URGENT) 1737 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1); 1738 1739 sb_start_intwrite(sbi->sb); 1740 1741 issued = __issue_discard_cmd(sbi, &dpolicy); 1742 if (issued > 0) { 1743 __wait_all_discard_cmd(sbi, &dpolicy); 1744 wait_ms = dpolicy.min_interval; 1745 } else if (issued == -1){ 1746 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME); 1747 if (!wait_ms) 1748 wait_ms = dpolicy.mid_interval; 1749 } else { 1750 wait_ms = dpolicy.max_interval; 1751 } 1752 1753 sb_end_intwrite(sbi->sb); 1754 1755 } while (!kthread_should_stop()); 1756 return 0; 1757 } 1758 1759 #ifdef CONFIG_BLK_DEV_ZONED 1760 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi, 1761 struct block_device *bdev, block_t blkstart, block_t blklen) 1762 { 1763 sector_t sector, nr_sects; 1764 block_t lblkstart = blkstart; 1765 int devi = 0; 1766 1767 if (f2fs_is_multi_device(sbi)) { 1768 devi = f2fs_target_device_index(sbi, blkstart); 1769 if (blkstart < FDEV(devi).start_blk || 1770 blkstart > FDEV(devi).end_blk) { 1771 f2fs_err(sbi, "Invalid block %x", blkstart); 1772 return -EIO; 1773 } 1774 blkstart -= FDEV(devi).start_blk; 1775 } 1776 1777 /* For sequential zones, reset the zone write pointer */ 1778 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) { 1779 sector = SECTOR_FROM_BLOCK(blkstart); 1780 nr_sects = SECTOR_FROM_BLOCK(blklen); 1781 1782 if (sector & (bdev_zone_sectors(bdev) - 1) || 1783 nr_sects != bdev_zone_sectors(bdev)) { 1784 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)", 1785 devi, sbi->s_ndevs ? FDEV(devi).path : "", 1786 blkstart, blklen); 1787 return -EIO; 1788 } 1789 trace_f2fs_issue_reset_zone(bdev, blkstart); 1790 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, 1791 sector, nr_sects, GFP_NOFS); 1792 } 1793 1794 /* For conventional zones, use regular discard if supported */ 1795 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen); 1796 } 1797 #endif 1798 1799 static int __issue_discard_async(struct f2fs_sb_info *sbi, 1800 struct block_device *bdev, block_t blkstart, block_t blklen) 1801 { 1802 #ifdef CONFIG_BLK_DEV_ZONED 1803 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) 1804 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen); 1805 #endif 1806 return __queue_discard_cmd(sbi, bdev, blkstart, blklen); 1807 } 1808 1809 static int f2fs_issue_discard(struct f2fs_sb_info *sbi, 1810 block_t blkstart, block_t blklen) 1811 { 1812 sector_t start = blkstart, len = 0; 1813 struct block_device *bdev; 1814 struct seg_entry *se; 1815 unsigned int offset; 1816 block_t i; 1817 int err = 0; 1818 1819 bdev = f2fs_target_device(sbi, blkstart, NULL); 1820 1821 for (i = blkstart; i < blkstart + blklen; i++, len++) { 1822 if (i != start) { 1823 struct block_device *bdev2 = 1824 f2fs_target_device(sbi, i, NULL); 1825 1826 if (bdev2 != bdev) { 1827 err = __issue_discard_async(sbi, bdev, 1828 start, len); 1829 if (err) 1830 return err; 1831 bdev = bdev2; 1832 start = i; 1833 len = 0; 1834 } 1835 } 1836 1837 se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); 1838 offset = GET_BLKOFF_FROM_SEG0(sbi, i); 1839 1840 if (!f2fs_test_and_set_bit(offset, se->discard_map)) 1841 sbi->discard_blks--; 1842 } 1843 1844 if (len) 1845 err = __issue_discard_async(sbi, bdev, start, len); 1846 return err; 1847 } 1848 1849 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc, 1850 bool check_only) 1851 { 1852 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 1853 int max_blocks = sbi->blocks_per_seg; 1854 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); 1855 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 1856 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 1857 unsigned long *discard_map = (unsigned long *)se->discard_map; 1858 unsigned long *dmap = SIT_I(sbi)->tmp_map; 1859 unsigned int start = 0, end = -1; 1860 bool force = (cpc->reason & CP_DISCARD); 1861 struct discard_entry *de = NULL; 1862 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list; 1863 int i; 1864 1865 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi)) 1866 return false; 1867 1868 if (!force) { 1869 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks || 1870 SM_I(sbi)->dcc_info->nr_discards >= 1871 SM_I(sbi)->dcc_info->max_discards) 1872 return false; 1873 } 1874 1875 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ 1876 for (i = 0; i < entries; i++) 1877 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : 1878 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; 1879 1880 while (force || SM_I(sbi)->dcc_info->nr_discards <= 1881 SM_I(sbi)->dcc_info->max_discards) { 1882 start = __find_rev_next_bit(dmap, max_blocks, end + 1); 1883 if (start >= max_blocks) 1884 break; 1885 1886 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); 1887 if (force && start && end != max_blocks 1888 && (end - start) < cpc->trim_minlen) 1889 continue; 1890 1891 if (check_only) 1892 return true; 1893 1894 if (!de) { 1895 de = f2fs_kmem_cache_alloc(discard_entry_slab, 1896 GFP_F2FS_ZERO); 1897 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start); 1898 list_add_tail(&de->list, head); 1899 } 1900 1901 for (i = start; i < end; i++) 1902 __set_bit_le(i, (void *)de->discard_map); 1903 1904 SM_I(sbi)->dcc_info->nr_discards += end - start; 1905 } 1906 return false; 1907 } 1908 1909 static void release_discard_addr(struct discard_entry *entry) 1910 { 1911 list_del(&entry->list); 1912 kmem_cache_free(discard_entry_slab, entry); 1913 } 1914 1915 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi) 1916 { 1917 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); 1918 struct discard_entry *entry, *this; 1919 1920 /* drop caches */ 1921 list_for_each_entry_safe(entry, this, head, list) 1922 release_discard_addr(entry); 1923 } 1924 1925 /* 1926 * Should call f2fs_clear_prefree_segments after checkpoint is done. 1927 */ 1928 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 1929 { 1930 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1931 unsigned int segno; 1932 1933 mutex_lock(&dirty_i->seglist_lock); 1934 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) 1935 __set_test_and_free(sbi, segno); 1936 mutex_unlock(&dirty_i->seglist_lock); 1937 } 1938 1939 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi, 1940 struct cp_control *cpc) 1941 { 1942 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1943 struct list_head *head = &dcc->entry_list; 1944 struct discard_entry *entry, *this; 1945 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1946 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; 1947 unsigned int start = 0, end = -1; 1948 unsigned int secno, start_segno; 1949 bool force = (cpc->reason & CP_DISCARD); 1950 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi); 1951 1952 mutex_lock(&dirty_i->seglist_lock); 1953 1954 while (1) { 1955 int i; 1956 1957 if (need_align && end != -1) 1958 end--; 1959 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); 1960 if (start >= MAIN_SEGS(sbi)) 1961 break; 1962 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), 1963 start + 1); 1964 1965 if (need_align) { 1966 start = rounddown(start, sbi->segs_per_sec); 1967 end = roundup(end, sbi->segs_per_sec); 1968 } 1969 1970 for (i = start; i < end; i++) { 1971 if (test_and_clear_bit(i, prefree_map)) 1972 dirty_i->nr_dirty[PRE]--; 1973 } 1974 1975 if (!f2fs_realtime_discard_enable(sbi)) 1976 continue; 1977 1978 if (force && start >= cpc->trim_start && 1979 (end - 1) <= cpc->trim_end) 1980 continue; 1981 1982 if (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi)) { 1983 f2fs_issue_discard(sbi, START_BLOCK(sbi, start), 1984 (end - start) << sbi->log_blocks_per_seg); 1985 continue; 1986 } 1987 next: 1988 secno = GET_SEC_FROM_SEG(sbi, start); 1989 start_segno = GET_SEG_FROM_SEC(sbi, secno); 1990 if (!IS_CURSEC(sbi, secno) && 1991 !get_valid_blocks(sbi, start, true)) 1992 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno), 1993 sbi->segs_per_sec << sbi->log_blocks_per_seg); 1994 1995 start = start_segno + sbi->segs_per_sec; 1996 if (start < end) 1997 goto next; 1998 else 1999 end = start - 1; 2000 } 2001 mutex_unlock(&dirty_i->seglist_lock); 2002 2003 /* send small discards */ 2004 list_for_each_entry_safe(entry, this, head, list) { 2005 unsigned int cur_pos = 0, next_pos, len, total_len = 0; 2006 bool is_valid = test_bit_le(0, entry->discard_map); 2007 2008 find_next: 2009 if (is_valid) { 2010 next_pos = find_next_zero_bit_le(entry->discard_map, 2011 sbi->blocks_per_seg, cur_pos); 2012 len = next_pos - cur_pos; 2013 2014 if (f2fs_sb_has_blkzoned(sbi) || 2015 (force && len < cpc->trim_minlen)) 2016 goto skip; 2017 2018 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos, 2019 len); 2020 total_len += len; 2021 } else { 2022 next_pos = find_next_bit_le(entry->discard_map, 2023 sbi->blocks_per_seg, cur_pos); 2024 } 2025 skip: 2026 cur_pos = next_pos; 2027 is_valid = !is_valid; 2028 2029 if (cur_pos < sbi->blocks_per_seg) 2030 goto find_next; 2031 2032 release_discard_addr(entry); 2033 dcc->nr_discards -= total_len; 2034 } 2035 2036 wake_up_discard_thread(sbi, false); 2037 } 2038 2039 static int create_discard_cmd_control(struct f2fs_sb_info *sbi) 2040 { 2041 dev_t dev = sbi->sb->s_bdev->bd_dev; 2042 struct discard_cmd_control *dcc; 2043 int err = 0, i; 2044 2045 if (SM_I(sbi)->dcc_info) { 2046 dcc = SM_I(sbi)->dcc_info; 2047 goto init_thread; 2048 } 2049 2050 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL); 2051 if (!dcc) 2052 return -ENOMEM; 2053 2054 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY; 2055 INIT_LIST_HEAD(&dcc->entry_list); 2056 for (i = 0; i < MAX_PLIST_NUM; i++) 2057 INIT_LIST_HEAD(&dcc->pend_list[i]); 2058 INIT_LIST_HEAD(&dcc->wait_list); 2059 INIT_LIST_HEAD(&dcc->fstrim_list); 2060 mutex_init(&dcc->cmd_lock); 2061 atomic_set(&dcc->issued_discard, 0); 2062 atomic_set(&dcc->queued_discard, 0); 2063 atomic_set(&dcc->discard_cmd_cnt, 0); 2064 dcc->nr_discards = 0; 2065 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg; 2066 dcc->undiscard_blks = 0; 2067 dcc->next_pos = 0; 2068 dcc->root = RB_ROOT_CACHED; 2069 dcc->rbtree_check = false; 2070 2071 init_waitqueue_head(&dcc->discard_wait_queue); 2072 SM_I(sbi)->dcc_info = dcc; 2073 init_thread: 2074 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi, 2075 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev)); 2076 if (IS_ERR(dcc->f2fs_issue_discard)) { 2077 err = PTR_ERR(dcc->f2fs_issue_discard); 2078 kvfree(dcc); 2079 SM_I(sbi)->dcc_info = NULL; 2080 return err; 2081 } 2082 2083 return err; 2084 } 2085 2086 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi) 2087 { 2088 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2089 2090 if (!dcc) 2091 return; 2092 2093 f2fs_stop_discard_thread(sbi); 2094 2095 /* 2096 * Recovery can cache discard commands, so in error path of 2097 * fill_super(), it needs to give a chance to handle them. 2098 */ 2099 if (unlikely(atomic_read(&dcc->discard_cmd_cnt))) 2100 f2fs_issue_discard_timeout(sbi); 2101 2102 kvfree(dcc); 2103 SM_I(sbi)->dcc_info = NULL; 2104 } 2105 2106 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 2107 { 2108 struct sit_info *sit_i = SIT_I(sbi); 2109 2110 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { 2111 sit_i->dirty_sentries++; 2112 return false; 2113 } 2114 2115 return true; 2116 } 2117 2118 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 2119 unsigned int segno, int modified) 2120 { 2121 struct seg_entry *se = get_seg_entry(sbi, segno); 2122 se->type = type; 2123 if (modified) 2124 __mark_sit_entry_dirty(sbi, segno); 2125 } 2126 2127 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 2128 { 2129 struct seg_entry *se; 2130 unsigned int segno, offset; 2131 long int new_vblocks; 2132 bool exist; 2133 #ifdef CONFIG_F2FS_CHECK_FS 2134 bool mir_exist; 2135 #endif 2136 2137 segno = GET_SEGNO(sbi, blkaddr); 2138 2139 se = get_seg_entry(sbi, segno); 2140 new_vblocks = se->valid_blocks + del; 2141 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2142 2143 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) || 2144 (new_vblocks > sbi->blocks_per_seg))); 2145 2146 se->valid_blocks = new_vblocks; 2147 se->mtime = get_mtime(sbi, false); 2148 if (se->mtime > SIT_I(sbi)->max_mtime) 2149 SIT_I(sbi)->max_mtime = se->mtime; 2150 2151 /* Update valid block bitmap */ 2152 if (del > 0) { 2153 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map); 2154 #ifdef CONFIG_F2FS_CHECK_FS 2155 mir_exist = f2fs_test_and_set_bit(offset, 2156 se->cur_valid_map_mir); 2157 if (unlikely(exist != mir_exist)) { 2158 f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d", 2159 blkaddr, exist); 2160 f2fs_bug_on(sbi, 1); 2161 } 2162 #endif 2163 if (unlikely(exist)) { 2164 f2fs_err(sbi, "Bitmap was wrongly set, blk:%u", 2165 blkaddr); 2166 f2fs_bug_on(sbi, 1); 2167 se->valid_blocks--; 2168 del = 0; 2169 } 2170 2171 if (!f2fs_test_and_set_bit(offset, se->discard_map)) 2172 sbi->discard_blks--; 2173 2174 /* 2175 * SSR should never reuse block which is checkpointed 2176 * or newly invalidated. 2177 */ 2178 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) { 2179 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map)) 2180 se->ckpt_valid_blocks++; 2181 } 2182 } else { 2183 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map); 2184 #ifdef CONFIG_F2FS_CHECK_FS 2185 mir_exist = f2fs_test_and_clear_bit(offset, 2186 se->cur_valid_map_mir); 2187 if (unlikely(exist != mir_exist)) { 2188 f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d", 2189 blkaddr, exist); 2190 f2fs_bug_on(sbi, 1); 2191 } 2192 #endif 2193 if (unlikely(!exist)) { 2194 f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u", 2195 blkaddr); 2196 f2fs_bug_on(sbi, 1); 2197 se->valid_blocks++; 2198 del = 0; 2199 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 2200 /* 2201 * If checkpoints are off, we must not reuse data that 2202 * was used in the previous checkpoint. If it was used 2203 * before, we must track that to know how much space we 2204 * really have. 2205 */ 2206 if (f2fs_test_bit(offset, se->ckpt_valid_map)) { 2207 spin_lock(&sbi->stat_lock); 2208 sbi->unusable_block_count++; 2209 spin_unlock(&sbi->stat_lock); 2210 } 2211 } 2212 2213 if (f2fs_test_and_clear_bit(offset, se->discard_map)) 2214 sbi->discard_blks++; 2215 } 2216 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) 2217 se->ckpt_valid_blocks += del; 2218 2219 __mark_sit_entry_dirty(sbi, segno); 2220 2221 /* update total number of valid blocks to be written in ckpt area */ 2222 SIT_I(sbi)->written_valid_blocks += del; 2223 2224 if (__is_large_section(sbi)) 2225 get_sec_entry(sbi, segno)->valid_blocks += del; 2226 } 2227 2228 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) 2229 { 2230 unsigned int segno = GET_SEGNO(sbi, addr); 2231 struct sit_info *sit_i = SIT_I(sbi); 2232 2233 f2fs_bug_on(sbi, addr == NULL_ADDR); 2234 if (addr == NEW_ADDR || addr == COMPRESS_ADDR) 2235 return; 2236 2237 invalidate_mapping_pages(META_MAPPING(sbi), addr, addr); 2238 2239 /* add it into sit main buffer */ 2240 down_write(&sit_i->sentry_lock); 2241 2242 update_sit_entry(sbi, addr, -1); 2243 2244 /* add it into dirty seglist */ 2245 locate_dirty_segment(sbi, segno); 2246 2247 up_write(&sit_i->sentry_lock); 2248 } 2249 2250 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr) 2251 { 2252 struct sit_info *sit_i = SIT_I(sbi); 2253 unsigned int segno, offset; 2254 struct seg_entry *se; 2255 bool is_cp = false; 2256 2257 if (!__is_valid_data_blkaddr(blkaddr)) 2258 return true; 2259 2260 down_read(&sit_i->sentry_lock); 2261 2262 segno = GET_SEGNO(sbi, blkaddr); 2263 se = get_seg_entry(sbi, segno); 2264 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2265 2266 if (f2fs_test_bit(offset, se->ckpt_valid_map)) 2267 is_cp = true; 2268 2269 up_read(&sit_i->sentry_lock); 2270 2271 return is_cp; 2272 } 2273 2274 /* 2275 * This function should be resided under the curseg_mutex lock 2276 */ 2277 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, 2278 struct f2fs_summary *sum) 2279 { 2280 struct curseg_info *curseg = CURSEG_I(sbi, type); 2281 void *addr = curseg->sum_blk; 2282 addr += curseg->next_blkoff * sizeof(struct f2fs_summary); 2283 memcpy(addr, sum, sizeof(struct f2fs_summary)); 2284 } 2285 2286 /* 2287 * Calculate the number of current summary pages for writing 2288 */ 2289 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) 2290 { 2291 int valid_sum_count = 0; 2292 int i, sum_in_page; 2293 2294 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2295 if (sbi->ckpt->alloc_type[i] == SSR) 2296 valid_sum_count += sbi->blocks_per_seg; 2297 else { 2298 if (for_ra) 2299 valid_sum_count += le16_to_cpu( 2300 F2FS_CKPT(sbi)->cur_data_blkoff[i]); 2301 else 2302 valid_sum_count += curseg_blkoff(sbi, i); 2303 } 2304 } 2305 2306 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE - 2307 SUM_FOOTER_SIZE) / SUMMARY_SIZE; 2308 if (valid_sum_count <= sum_in_page) 2309 return 1; 2310 else if ((valid_sum_count - sum_in_page) <= 2311 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) 2312 return 2; 2313 return 3; 2314 } 2315 2316 /* 2317 * Caller should put this summary page 2318 */ 2319 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) 2320 { 2321 return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno)); 2322 } 2323 2324 void f2fs_update_meta_page(struct f2fs_sb_info *sbi, 2325 void *src, block_t blk_addr) 2326 { 2327 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2328 2329 memcpy(page_address(page), src, PAGE_SIZE); 2330 set_page_dirty(page); 2331 f2fs_put_page(page, 1); 2332 } 2333 2334 static void write_sum_page(struct f2fs_sb_info *sbi, 2335 struct f2fs_summary_block *sum_blk, block_t blk_addr) 2336 { 2337 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr); 2338 } 2339 2340 static void write_current_sum_page(struct f2fs_sb_info *sbi, 2341 int type, block_t blk_addr) 2342 { 2343 struct curseg_info *curseg = CURSEG_I(sbi, type); 2344 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2345 struct f2fs_summary_block *src = curseg->sum_blk; 2346 struct f2fs_summary_block *dst; 2347 2348 dst = (struct f2fs_summary_block *)page_address(page); 2349 memset(dst, 0, PAGE_SIZE); 2350 2351 mutex_lock(&curseg->curseg_mutex); 2352 2353 down_read(&curseg->journal_rwsem); 2354 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE); 2355 up_read(&curseg->journal_rwsem); 2356 2357 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE); 2358 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE); 2359 2360 mutex_unlock(&curseg->curseg_mutex); 2361 2362 set_page_dirty(page); 2363 f2fs_put_page(page, 1); 2364 } 2365 2366 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type) 2367 { 2368 struct curseg_info *curseg = CURSEG_I(sbi, type); 2369 unsigned int segno = curseg->segno + 1; 2370 struct free_segmap_info *free_i = FREE_I(sbi); 2371 2372 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec) 2373 return !test_bit(segno, free_i->free_segmap); 2374 return 0; 2375 } 2376 2377 /* 2378 * Find a new segment from the free segments bitmap to right order 2379 * This function should be returned with success, otherwise BUG 2380 */ 2381 static void get_new_segment(struct f2fs_sb_info *sbi, 2382 unsigned int *newseg, bool new_sec, int dir) 2383 { 2384 struct free_segmap_info *free_i = FREE_I(sbi); 2385 unsigned int segno, secno, zoneno; 2386 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; 2387 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg); 2388 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg); 2389 unsigned int left_start = hint; 2390 bool init = true; 2391 int go_left = 0; 2392 int i; 2393 2394 spin_lock(&free_i->segmap_lock); 2395 2396 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { 2397 segno = find_next_zero_bit(free_i->free_segmap, 2398 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1); 2399 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1)) 2400 goto got_it; 2401 } 2402 find_other_zone: 2403 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 2404 if (secno >= MAIN_SECS(sbi)) { 2405 if (dir == ALLOC_RIGHT) { 2406 secno = find_next_zero_bit(free_i->free_secmap, 2407 MAIN_SECS(sbi), 0); 2408 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi)); 2409 } else { 2410 go_left = 1; 2411 left_start = hint - 1; 2412 } 2413 } 2414 if (go_left == 0) 2415 goto skip_left; 2416 2417 while (test_bit(left_start, free_i->free_secmap)) { 2418 if (left_start > 0) { 2419 left_start--; 2420 continue; 2421 } 2422 left_start = find_next_zero_bit(free_i->free_secmap, 2423 MAIN_SECS(sbi), 0); 2424 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi)); 2425 break; 2426 } 2427 secno = left_start; 2428 skip_left: 2429 segno = GET_SEG_FROM_SEC(sbi, secno); 2430 zoneno = GET_ZONE_FROM_SEC(sbi, secno); 2431 2432 /* give up on finding another zone */ 2433 if (!init) 2434 goto got_it; 2435 if (sbi->secs_per_zone == 1) 2436 goto got_it; 2437 if (zoneno == old_zoneno) 2438 goto got_it; 2439 if (dir == ALLOC_LEFT) { 2440 if (!go_left && zoneno + 1 >= total_zones) 2441 goto got_it; 2442 if (go_left && zoneno == 0) 2443 goto got_it; 2444 } 2445 for (i = 0; i < NR_CURSEG_TYPE; i++) 2446 if (CURSEG_I(sbi, i)->zone == zoneno) 2447 break; 2448 2449 if (i < NR_CURSEG_TYPE) { 2450 /* zone is in user, try another */ 2451 if (go_left) 2452 hint = zoneno * sbi->secs_per_zone - 1; 2453 else if (zoneno + 1 >= total_zones) 2454 hint = 0; 2455 else 2456 hint = (zoneno + 1) * sbi->secs_per_zone; 2457 init = false; 2458 goto find_other_zone; 2459 } 2460 got_it: 2461 /* set it as dirty segment in free segmap */ 2462 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); 2463 __set_inuse(sbi, segno); 2464 *newseg = segno; 2465 spin_unlock(&free_i->segmap_lock); 2466 } 2467 2468 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 2469 { 2470 struct curseg_info *curseg = CURSEG_I(sbi, type); 2471 struct summary_footer *sum_footer; 2472 2473 curseg->segno = curseg->next_segno; 2474 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno); 2475 curseg->next_blkoff = 0; 2476 curseg->next_segno = NULL_SEGNO; 2477 2478 sum_footer = &(curseg->sum_blk->footer); 2479 memset(sum_footer, 0, sizeof(struct summary_footer)); 2480 if (IS_DATASEG(type)) 2481 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 2482 if (IS_NODESEG(type)) 2483 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 2484 __set_sit_entry_type(sbi, type, curseg->segno, modified); 2485 } 2486 2487 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type) 2488 { 2489 /* if segs_per_sec is large than 1, we need to keep original policy. */ 2490 if (__is_large_section(sbi)) 2491 return CURSEG_I(sbi, type)->segno; 2492 2493 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 2494 return 0; 2495 2496 if (test_opt(sbi, NOHEAP) && 2497 (type == CURSEG_HOT_DATA || IS_NODESEG(type))) 2498 return 0; 2499 2500 if (SIT_I(sbi)->last_victim[ALLOC_NEXT]) 2501 return SIT_I(sbi)->last_victim[ALLOC_NEXT]; 2502 2503 /* find segments from 0 to reuse freed segments */ 2504 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE) 2505 return 0; 2506 2507 return CURSEG_I(sbi, type)->segno; 2508 } 2509 2510 /* 2511 * Allocate a current working segment. 2512 * This function always allocates a free segment in LFS manner. 2513 */ 2514 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 2515 { 2516 struct curseg_info *curseg = CURSEG_I(sbi, type); 2517 unsigned int segno = curseg->segno; 2518 int dir = ALLOC_LEFT; 2519 2520 write_sum_page(sbi, curseg->sum_blk, 2521 GET_SUM_BLOCK(sbi, segno)); 2522 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA) 2523 dir = ALLOC_RIGHT; 2524 2525 if (test_opt(sbi, NOHEAP)) 2526 dir = ALLOC_RIGHT; 2527 2528 segno = __get_next_segno(sbi, type); 2529 get_new_segment(sbi, &segno, new_sec, dir); 2530 curseg->next_segno = segno; 2531 reset_curseg(sbi, type, 1); 2532 curseg->alloc_type = LFS; 2533 } 2534 2535 static void __next_free_blkoff(struct f2fs_sb_info *sbi, 2536 struct curseg_info *seg, block_t start) 2537 { 2538 struct seg_entry *se = get_seg_entry(sbi, seg->segno); 2539 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 2540 unsigned long *target_map = SIT_I(sbi)->tmp_map; 2541 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 2542 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 2543 int i, pos; 2544 2545 for (i = 0; i < entries; i++) 2546 target_map[i] = ckpt_map[i] | cur_map[i]; 2547 2548 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); 2549 2550 seg->next_blkoff = pos; 2551 } 2552 2553 /* 2554 * If a segment is written by LFS manner, next block offset is just obtained 2555 * by increasing the current block offset. However, if a segment is written by 2556 * SSR manner, next block offset obtained by calling __next_free_blkoff 2557 */ 2558 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, 2559 struct curseg_info *seg) 2560 { 2561 if (seg->alloc_type == SSR) 2562 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1); 2563 else 2564 seg->next_blkoff++; 2565 } 2566 2567 /* 2568 * This function always allocates a used segment(from dirty seglist) by SSR 2569 * manner, so it should recover the existing segment information of valid blocks 2570 */ 2571 static void change_curseg(struct f2fs_sb_info *sbi, int type) 2572 { 2573 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2574 struct curseg_info *curseg = CURSEG_I(sbi, type); 2575 unsigned int new_segno = curseg->next_segno; 2576 struct f2fs_summary_block *sum_node; 2577 struct page *sum_page; 2578 2579 write_sum_page(sbi, curseg->sum_blk, 2580 GET_SUM_BLOCK(sbi, curseg->segno)); 2581 __set_test_and_inuse(sbi, new_segno); 2582 2583 mutex_lock(&dirty_i->seglist_lock); 2584 __remove_dirty_segment(sbi, new_segno, PRE); 2585 __remove_dirty_segment(sbi, new_segno, DIRTY); 2586 mutex_unlock(&dirty_i->seglist_lock); 2587 2588 reset_curseg(sbi, type, 1); 2589 curseg->alloc_type = SSR; 2590 __next_free_blkoff(sbi, curseg, 0); 2591 2592 sum_page = f2fs_get_sum_page(sbi, new_segno); 2593 f2fs_bug_on(sbi, IS_ERR(sum_page)); 2594 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 2595 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); 2596 f2fs_put_page(sum_page, 1); 2597 } 2598 2599 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type) 2600 { 2601 struct curseg_info *curseg = CURSEG_I(sbi, type); 2602 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; 2603 unsigned segno = NULL_SEGNO; 2604 int i, cnt; 2605 bool reversed = false; 2606 2607 /* f2fs_need_SSR() already forces to do this */ 2608 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) { 2609 curseg->next_segno = segno; 2610 return 1; 2611 } 2612 2613 /* For node segments, let's do SSR more intensively */ 2614 if (IS_NODESEG(type)) { 2615 if (type >= CURSEG_WARM_NODE) { 2616 reversed = true; 2617 i = CURSEG_COLD_NODE; 2618 } else { 2619 i = CURSEG_HOT_NODE; 2620 } 2621 cnt = NR_CURSEG_NODE_TYPE; 2622 } else { 2623 if (type >= CURSEG_WARM_DATA) { 2624 reversed = true; 2625 i = CURSEG_COLD_DATA; 2626 } else { 2627 i = CURSEG_HOT_DATA; 2628 } 2629 cnt = NR_CURSEG_DATA_TYPE; 2630 } 2631 2632 for (; cnt-- > 0; reversed ? i-- : i++) { 2633 if (i == type) 2634 continue; 2635 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) { 2636 curseg->next_segno = segno; 2637 return 1; 2638 } 2639 } 2640 2641 /* find valid_blocks=0 in dirty list */ 2642 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 2643 segno = get_free_segment(sbi); 2644 if (segno != NULL_SEGNO) { 2645 curseg->next_segno = segno; 2646 return 1; 2647 } 2648 } 2649 return 0; 2650 } 2651 2652 /* 2653 * flush out current segment and replace it with new segment 2654 * This function should be returned with success, otherwise BUG 2655 */ 2656 static void allocate_segment_by_default(struct f2fs_sb_info *sbi, 2657 int type, bool force) 2658 { 2659 struct curseg_info *curseg = CURSEG_I(sbi, type); 2660 2661 if (force) 2662 new_curseg(sbi, type, true); 2663 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) && 2664 type == CURSEG_WARM_NODE) 2665 new_curseg(sbi, type, false); 2666 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) && 2667 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 2668 new_curseg(sbi, type, false); 2669 else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type)) 2670 change_curseg(sbi, type); 2671 else 2672 new_curseg(sbi, type, false); 2673 2674 stat_inc_seg_type(sbi, curseg); 2675 } 2676 2677 void allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type, 2678 unsigned int start, unsigned int end) 2679 { 2680 struct curseg_info *curseg = CURSEG_I(sbi, type); 2681 unsigned int segno; 2682 2683 down_read(&SM_I(sbi)->curseg_lock); 2684 mutex_lock(&curseg->curseg_mutex); 2685 down_write(&SIT_I(sbi)->sentry_lock); 2686 2687 segno = CURSEG_I(sbi, type)->segno; 2688 if (segno < start || segno > end) 2689 goto unlock; 2690 2691 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type)) 2692 change_curseg(sbi, type); 2693 else 2694 new_curseg(sbi, type, true); 2695 2696 stat_inc_seg_type(sbi, curseg); 2697 2698 locate_dirty_segment(sbi, segno); 2699 unlock: 2700 up_write(&SIT_I(sbi)->sentry_lock); 2701 2702 if (segno != curseg->segno) 2703 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u", 2704 type, segno, curseg->segno); 2705 2706 mutex_unlock(&curseg->curseg_mutex); 2707 up_read(&SM_I(sbi)->curseg_lock); 2708 } 2709 2710 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi, int type) 2711 { 2712 struct curseg_info *curseg; 2713 unsigned int old_segno; 2714 int i; 2715 2716 down_write(&SIT_I(sbi)->sentry_lock); 2717 2718 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2719 if (type != NO_CHECK_TYPE && i != type) 2720 continue; 2721 2722 curseg = CURSEG_I(sbi, i); 2723 if (type == NO_CHECK_TYPE || curseg->next_blkoff || 2724 get_valid_blocks(sbi, curseg->segno, false) || 2725 get_ckpt_valid_blocks(sbi, curseg->segno)) { 2726 old_segno = curseg->segno; 2727 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true); 2728 locate_dirty_segment(sbi, old_segno); 2729 } 2730 } 2731 2732 up_write(&SIT_I(sbi)->sentry_lock); 2733 } 2734 2735 static const struct segment_allocation default_salloc_ops = { 2736 .allocate_segment = allocate_segment_by_default, 2737 }; 2738 2739 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi, 2740 struct cp_control *cpc) 2741 { 2742 __u64 trim_start = cpc->trim_start; 2743 bool has_candidate = false; 2744 2745 down_write(&SIT_I(sbi)->sentry_lock); 2746 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) { 2747 if (add_discard_addrs(sbi, cpc, true)) { 2748 has_candidate = true; 2749 break; 2750 } 2751 } 2752 up_write(&SIT_I(sbi)->sentry_lock); 2753 2754 cpc->trim_start = trim_start; 2755 return has_candidate; 2756 } 2757 2758 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi, 2759 struct discard_policy *dpolicy, 2760 unsigned int start, unsigned int end) 2761 { 2762 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2763 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 2764 struct rb_node **insert_p = NULL, *insert_parent = NULL; 2765 struct discard_cmd *dc; 2766 struct blk_plug plug; 2767 int issued; 2768 unsigned int trimmed = 0; 2769 2770 next: 2771 issued = 0; 2772 2773 mutex_lock(&dcc->cmd_lock); 2774 if (unlikely(dcc->rbtree_check)) 2775 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi, 2776 &dcc->root)); 2777 2778 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root, 2779 NULL, start, 2780 (struct rb_entry **)&prev_dc, 2781 (struct rb_entry **)&next_dc, 2782 &insert_p, &insert_parent, true, NULL); 2783 if (!dc) 2784 dc = next_dc; 2785 2786 blk_start_plug(&plug); 2787 2788 while (dc && dc->lstart <= end) { 2789 struct rb_node *node; 2790 int err = 0; 2791 2792 if (dc->len < dpolicy->granularity) 2793 goto skip; 2794 2795 if (dc->state != D_PREP) { 2796 list_move_tail(&dc->list, &dcc->fstrim_list); 2797 goto skip; 2798 } 2799 2800 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued); 2801 2802 if (issued >= dpolicy->max_requests) { 2803 start = dc->lstart + dc->len; 2804 2805 if (err) 2806 __remove_discard_cmd(sbi, dc); 2807 2808 blk_finish_plug(&plug); 2809 mutex_unlock(&dcc->cmd_lock); 2810 trimmed += __wait_all_discard_cmd(sbi, NULL); 2811 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT); 2812 goto next; 2813 } 2814 skip: 2815 node = rb_next(&dc->rb_node); 2816 if (err) 2817 __remove_discard_cmd(sbi, dc); 2818 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 2819 2820 if (fatal_signal_pending(current)) 2821 break; 2822 } 2823 2824 blk_finish_plug(&plug); 2825 mutex_unlock(&dcc->cmd_lock); 2826 2827 return trimmed; 2828 } 2829 2830 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) 2831 { 2832 __u64 start = F2FS_BYTES_TO_BLK(range->start); 2833 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; 2834 unsigned int start_segno, end_segno; 2835 block_t start_block, end_block; 2836 struct cp_control cpc; 2837 struct discard_policy dpolicy; 2838 unsigned long long trimmed = 0; 2839 int err = 0; 2840 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi); 2841 2842 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize) 2843 return -EINVAL; 2844 2845 if (end < MAIN_BLKADDR(sbi)) 2846 goto out; 2847 2848 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 2849 f2fs_warn(sbi, "Found FS corruption, run fsck to fix."); 2850 return -EFSCORRUPTED; 2851 } 2852 2853 /* start/end segment number in main_area */ 2854 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); 2855 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : 2856 GET_SEGNO(sbi, end); 2857 if (need_align) { 2858 start_segno = rounddown(start_segno, sbi->segs_per_sec); 2859 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1; 2860 } 2861 2862 cpc.reason = CP_DISCARD; 2863 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen)); 2864 cpc.trim_start = start_segno; 2865 cpc.trim_end = end_segno; 2866 2867 if (sbi->discard_blks == 0) 2868 goto out; 2869 2870 down_write(&sbi->gc_lock); 2871 err = f2fs_write_checkpoint(sbi, &cpc); 2872 up_write(&sbi->gc_lock); 2873 if (err) 2874 goto out; 2875 2876 /* 2877 * We filed discard candidates, but actually we don't need to wait for 2878 * all of them, since they'll be issued in idle time along with runtime 2879 * discard option. User configuration looks like using runtime discard 2880 * or periodic fstrim instead of it. 2881 */ 2882 if (f2fs_realtime_discard_enable(sbi)) 2883 goto out; 2884 2885 start_block = START_BLOCK(sbi, start_segno); 2886 end_block = START_BLOCK(sbi, end_segno + 1); 2887 2888 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen); 2889 trimmed = __issue_discard_cmd_range(sbi, &dpolicy, 2890 start_block, end_block); 2891 2892 trimmed += __wait_discard_cmd_range(sbi, &dpolicy, 2893 start_block, end_block); 2894 out: 2895 if (!err) 2896 range->len = F2FS_BLK_TO_BYTES(trimmed); 2897 return err; 2898 } 2899 2900 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type) 2901 { 2902 struct curseg_info *curseg = CURSEG_I(sbi, type); 2903 if (curseg->next_blkoff < sbi->blocks_per_seg) 2904 return true; 2905 return false; 2906 } 2907 2908 int f2fs_rw_hint_to_seg_type(enum rw_hint hint) 2909 { 2910 switch (hint) { 2911 case WRITE_LIFE_SHORT: 2912 return CURSEG_HOT_DATA; 2913 case WRITE_LIFE_EXTREME: 2914 return CURSEG_COLD_DATA; 2915 default: 2916 return CURSEG_WARM_DATA; 2917 } 2918 } 2919 2920 /* This returns write hints for each segment type. This hints will be 2921 * passed down to block layer. There are mapping tables which depend on 2922 * the mount option 'whint_mode'. 2923 * 2924 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET. 2925 * 2926 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users. 2927 * 2928 * User F2FS Block 2929 * ---- ---- ----- 2930 * META WRITE_LIFE_NOT_SET 2931 * HOT_NODE " 2932 * WARM_NODE " 2933 * COLD_NODE " 2934 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME 2935 * extension list " " 2936 * 2937 * -- buffered io 2938 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 2939 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 2940 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 2941 * WRITE_LIFE_NONE " " 2942 * WRITE_LIFE_MEDIUM " " 2943 * WRITE_LIFE_LONG " " 2944 * 2945 * -- direct io 2946 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 2947 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 2948 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 2949 * WRITE_LIFE_NONE " WRITE_LIFE_NONE 2950 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM 2951 * WRITE_LIFE_LONG " WRITE_LIFE_LONG 2952 * 2953 * 3) whint_mode=fs-based. F2FS passes down hints with its policy. 2954 * 2955 * User F2FS Block 2956 * ---- ---- ----- 2957 * META WRITE_LIFE_MEDIUM; 2958 * HOT_NODE WRITE_LIFE_NOT_SET 2959 * WARM_NODE " 2960 * COLD_NODE WRITE_LIFE_NONE 2961 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME 2962 * extension list " " 2963 * 2964 * -- buffered io 2965 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 2966 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 2967 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG 2968 * WRITE_LIFE_NONE " " 2969 * WRITE_LIFE_MEDIUM " " 2970 * WRITE_LIFE_LONG " " 2971 * 2972 * -- direct io 2973 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 2974 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 2975 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 2976 * WRITE_LIFE_NONE " WRITE_LIFE_NONE 2977 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM 2978 * WRITE_LIFE_LONG " WRITE_LIFE_LONG 2979 */ 2980 2981 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi, 2982 enum page_type type, enum temp_type temp) 2983 { 2984 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) { 2985 if (type == DATA) { 2986 if (temp == WARM) 2987 return WRITE_LIFE_NOT_SET; 2988 else if (temp == HOT) 2989 return WRITE_LIFE_SHORT; 2990 else if (temp == COLD) 2991 return WRITE_LIFE_EXTREME; 2992 } else { 2993 return WRITE_LIFE_NOT_SET; 2994 } 2995 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) { 2996 if (type == DATA) { 2997 if (temp == WARM) 2998 return WRITE_LIFE_LONG; 2999 else if (temp == HOT) 3000 return WRITE_LIFE_SHORT; 3001 else if (temp == COLD) 3002 return WRITE_LIFE_EXTREME; 3003 } else if (type == NODE) { 3004 if (temp == WARM || temp == HOT) 3005 return WRITE_LIFE_NOT_SET; 3006 else if (temp == COLD) 3007 return WRITE_LIFE_NONE; 3008 } else if (type == META) { 3009 return WRITE_LIFE_MEDIUM; 3010 } 3011 } 3012 return WRITE_LIFE_NOT_SET; 3013 } 3014 3015 static int __get_segment_type_2(struct f2fs_io_info *fio) 3016 { 3017 if (fio->type == DATA) 3018 return CURSEG_HOT_DATA; 3019 else 3020 return CURSEG_HOT_NODE; 3021 } 3022 3023 static int __get_segment_type_4(struct f2fs_io_info *fio) 3024 { 3025 if (fio->type == DATA) { 3026 struct inode *inode = fio->page->mapping->host; 3027 3028 if (S_ISDIR(inode->i_mode)) 3029 return CURSEG_HOT_DATA; 3030 else 3031 return CURSEG_COLD_DATA; 3032 } else { 3033 if (IS_DNODE(fio->page) && is_cold_node(fio->page)) 3034 return CURSEG_WARM_NODE; 3035 else 3036 return CURSEG_COLD_NODE; 3037 } 3038 } 3039 3040 static int __get_segment_type_6(struct f2fs_io_info *fio) 3041 { 3042 if (fio->type == DATA) { 3043 struct inode *inode = fio->page->mapping->host; 3044 3045 if (is_cold_data(fio->page) || file_is_cold(inode) || 3046 f2fs_compressed_file(inode)) 3047 return CURSEG_COLD_DATA; 3048 if (file_is_hot(inode) || 3049 is_inode_flag_set(inode, FI_HOT_DATA) || 3050 f2fs_is_atomic_file(inode) || 3051 f2fs_is_volatile_file(inode)) 3052 return CURSEG_HOT_DATA; 3053 return f2fs_rw_hint_to_seg_type(inode->i_write_hint); 3054 } else { 3055 if (IS_DNODE(fio->page)) 3056 return is_cold_node(fio->page) ? CURSEG_WARM_NODE : 3057 CURSEG_HOT_NODE; 3058 return CURSEG_COLD_NODE; 3059 } 3060 } 3061 3062 static int __get_segment_type(struct f2fs_io_info *fio) 3063 { 3064 int type = 0; 3065 3066 switch (F2FS_OPTION(fio->sbi).active_logs) { 3067 case 2: 3068 type = __get_segment_type_2(fio); 3069 break; 3070 case 4: 3071 type = __get_segment_type_4(fio); 3072 break; 3073 case 6: 3074 type = __get_segment_type_6(fio); 3075 break; 3076 default: 3077 f2fs_bug_on(fio->sbi, true); 3078 } 3079 3080 if (IS_HOT(type)) 3081 fio->temp = HOT; 3082 else if (IS_WARM(type)) 3083 fio->temp = WARM; 3084 else 3085 fio->temp = COLD; 3086 return type; 3087 } 3088 3089 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, 3090 block_t old_blkaddr, block_t *new_blkaddr, 3091 struct f2fs_summary *sum, int type, 3092 struct f2fs_io_info *fio, bool add_list) 3093 { 3094 struct sit_info *sit_i = SIT_I(sbi); 3095 struct curseg_info *curseg = CURSEG_I(sbi, type); 3096 bool put_pin_sem = false; 3097 3098 if (type == CURSEG_COLD_DATA) { 3099 /* GC during CURSEG_COLD_DATA_PINNED allocation */ 3100 if (down_read_trylock(&sbi->pin_sem)) { 3101 put_pin_sem = true; 3102 } else { 3103 type = CURSEG_WARM_DATA; 3104 curseg = CURSEG_I(sbi, type); 3105 } 3106 } else if (type == CURSEG_COLD_DATA_PINNED) { 3107 type = CURSEG_COLD_DATA; 3108 } 3109 3110 /* 3111 * We need to wait for node_write to avoid block allocation during 3112 * checkpoint. This can only happen to quota writes which can cause 3113 * the below discard race condition. 3114 */ 3115 if (IS_DATASEG(type)) 3116 down_write(&sbi->node_write); 3117 3118 down_read(&SM_I(sbi)->curseg_lock); 3119 3120 mutex_lock(&curseg->curseg_mutex); 3121 down_write(&sit_i->sentry_lock); 3122 3123 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 3124 3125 f2fs_wait_discard_bio(sbi, *new_blkaddr); 3126 3127 /* 3128 * __add_sum_entry should be resided under the curseg_mutex 3129 * because, this function updates a summary entry in the 3130 * current summary block. 3131 */ 3132 __add_sum_entry(sbi, type, sum); 3133 3134 __refresh_next_blkoff(sbi, curseg); 3135 3136 stat_inc_block_count(sbi, curseg); 3137 3138 /* 3139 * SIT information should be updated before segment allocation, 3140 * since SSR needs latest valid block information. 3141 */ 3142 update_sit_entry(sbi, *new_blkaddr, 1); 3143 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 3144 update_sit_entry(sbi, old_blkaddr, -1); 3145 3146 if (!__has_curseg_space(sbi, type)) 3147 sit_i->s_ops->allocate_segment(sbi, type, false); 3148 3149 /* 3150 * segment dirty status should be updated after segment allocation, 3151 * so we just need to update status only one time after previous 3152 * segment being closed. 3153 */ 3154 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3155 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr)); 3156 3157 up_write(&sit_i->sentry_lock); 3158 3159 if (page && IS_NODESEG(type)) { 3160 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); 3161 3162 f2fs_inode_chksum_set(sbi, page); 3163 } 3164 3165 if (F2FS_IO_ALIGNED(sbi)) 3166 fio->retry = false; 3167 3168 if (add_list) { 3169 struct f2fs_bio_info *io; 3170 3171 INIT_LIST_HEAD(&fio->list); 3172 fio->in_list = true; 3173 io = sbi->write_io[fio->type] + fio->temp; 3174 spin_lock(&io->io_lock); 3175 list_add_tail(&fio->list, &io->io_list); 3176 spin_unlock(&io->io_lock); 3177 } 3178 3179 mutex_unlock(&curseg->curseg_mutex); 3180 3181 up_read(&SM_I(sbi)->curseg_lock); 3182 3183 if (IS_DATASEG(type)) 3184 up_write(&sbi->node_write); 3185 3186 if (put_pin_sem) 3187 up_read(&sbi->pin_sem); 3188 } 3189 3190 static void update_device_state(struct f2fs_io_info *fio) 3191 { 3192 struct f2fs_sb_info *sbi = fio->sbi; 3193 unsigned int devidx; 3194 3195 if (!f2fs_is_multi_device(sbi)) 3196 return; 3197 3198 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr); 3199 3200 /* update device state for fsync */ 3201 f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO); 3202 3203 /* update device state for checkpoint */ 3204 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) { 3205 spin_lock(&sbi->dev_lock); 3206 f2fs_set_bit(devidx, (char *)&sbi->dirty_device); 3207 spin_unlock(&sbi->dev_lock); 3208 } 3209 } 3210 3211 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) 3212 { 3213 int type = __get_segment_type(fio); 3214 bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA); 3215 3216 if (keep_order) 3217 down_read(&fio->sbi->io_order_lock); 3218 reallocate: 3219 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr, 3220 &fio->new_blkaddr, sum, type, fio, true); 3221 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO) 3222 invalidate_mapping_pages(META_MAPPING(fio->sbi), 3223 fio->old_blkaddr, fio->old_blkaddr); 3224 3225 /* writeout dirty page into bdev */ 3226 f2fs_submit_page_write(fio); 3227 if (fio->retry) { 3228 fio->old_blkaddr = fio->new_blkaddr; 3229 goto reallocate; 3230 } 3231 3232 update_device_state(fio); 3233 3234 if (keep_order) 3235 up_read(&fio->sbi->io_order_lock); 3236 } 3237 3238 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page, 3239 enum iostat_type io_type) 3240 { 3241 struct f2fs_io_info fio = { 3242 .sbi = sbi, 3243 .type = META, 3244 .temp = HOT, 3245 .op = REQ_OP_WRITE, 3246 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO, 3247 .old_blkaddr = page->index, 3248 .new_blkaddr = page->index, 3249 .page = page, 3250 .encrypted_page = NULL, 3251 .in_list = false, 3252 }; 3253 3254 if (unlikely(page->index >= MAIN_BLKADDR(sbi))) 3255 fio.op_flags &= ~REQ_META; 3256 3257 set_page_writeback(page); 3258 ClearPageError(page); 3259 f2fs_submit_page_write(&fio); 3260 3261 stat_inc_meta_count(sbi, page->index); 3262 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE); 3263 } 3264 3265 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio) 3266 { 3267 struct f2fs_summary sum; 3268 3269 set_summary(&sum, nid, 0, 0); 3270 do_write_page(&sum, fio); 3271 3272 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE); 3273 } 3274 3275 void f2fs_outplace_write_data(struct dnode_of_data *dn, 3276 struct f2fs_io_info *fio) 3277 { 3278 struct f2fs_sb_info *sbi = fio->sbi; 3279 struct f2fs_summary sum; 3280 3281 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); 3282 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version); 3283 do_write_page(&sum, fio); 3284 f2fs_update_data_blkaddr(dn, fio->new_blkaddr); 3285 3286 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE); 3287 } 3288 3289 int f2fs_inplace_write_data(struct f2fs_io_info *fio) 3290 { 3291 int err; 3292 struct f2fs_sb_info *sbi = fio->sbi; 3293 unsigned int segno; 3294 3295 fio->new_blkaddr = fio->old_blkaddr; 3296 /* i/o temperature is needed for passing down write hints */ 3297 __get_segment_type(fio); 3298 3299 segno = GET_SEGNO(sbi, fio->new_blkaddr); 3300 3301 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) { 3302 set_sbi_flag(sbi, SBI_NEED_FSCK); 3303 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.", 3304 __func__, segno); 3305 return -EFSCORRUPTED; 3306 } 3307 3308 stat_inc_inplace_blocks(fio->sbi); 3309 3310 if (fio->bio && !(SM_I(sbi)->ipu_policy & (1 << F2FS_IPU_NOCACHE))) 3311 err = f2fs_merge_page_bio(fio); 3312 else 3313 err = f2fs_submit_page_bio(fio); 3314 if (!err) { 3315 update_device_state(fio); 3316 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE); 3317 } 3318 3319 return err; 3320 } 3321 3322 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi, 3323 unsigned int segno) 3324 { 3325 int i; 3326 3327 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 3328 if (CURSEG_I(sbi, i)->segno == segno) 3329 break; 3330 } 3331 return i; 3332 } 3333 3334 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 3335 block_t old_blkaddr, block_t new_blkaddr, 3336 bool recover_curseg, bool recover_newaddr) 3337 { 3338 struct sit_info *sit_i = SIT_I(sbi); 3339 struct curseg_info *curseg; 3340 unsigned int segno, old_cursegno; 3341 struct seg_entry *se; 3342 int type; 3343 unsigned short old_blkoff; 3344 3345 segno = GET_SEGNO(sbi, new_blkaddr); 3346 se = get_seg_entry(sbi, segno); 3347 type = se->type; 3348 3349 down_write(&SM_I(sbi)->curseg_lock); 3350 3351 if (!recover_curseg) { 3352 /* for recovery flow */ 3353 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { 3354 if (old_blkaddr == NULL_ADDR) 3355 type = CURSEG_COLD_DATA; 3356 else 3357 type = CURSEG_WARM_DATA; 3358 } 3359 } else { 3360 if (IS_CURSEG(sbi, segno)) { 3361 /* se->type is volatile as SSR allocation */ 3362 type = __f2fs_get_curseg(sbi, segno); 3363 f2fs_bug_on(sbi, type == NO_CHECK_TYPE); 3364 } else { 3365 type = CURSEG_WARM_DATA; 3366 } 3367 } 3368 3369 f2fs_bug_on(sbi, !IS_DATASEG(type)); 3370 curseg = CURSEG_I(sbi, type); 3371 3372 mutex_lock(&curseg->curseg_mutex); 3373 down_write(&sit_i->sentry_lock); 3374 3375 old_cursegno = curseg->segno; 3376 old_blkoff = curseg->next_blkoff; 3377 3378 /* change the current segment */ 3379 if (segno != curseg->segno) { 3380 curseg->next_segno = segno; 3381 change_curseg(sbi, type); 3382 } 3383 3384 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); 3385 __add_sum_entry(sbi, type, sum); 3386 3387 if (!recover_curseg || recover_newaddr) 3388 update_sit_entry(sbi, new_blkaddr, 1); 3389 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) { 3390 invalidate_mapping_pages(META_MAPPING(sbi), 3391 old_blkaddr, old_blkaddr); 3392 update_sit_entry(sbi, old_blkaddr, -1); 3393 } 3394 3395 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3396 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); 3397 3398 locate_dirty_segment(sbi, old_cursegno); 3399 3400 if (recover_curseg) { 3401 if (old_cursegno != curseg->segno) { 3402 curseg->next_segno = old_cursegno; 3403 change_curseg(sbi, type); 3404 } 3405 curseg->next_blkoff = old_blkoff; 3406 } 3407 3408 up_write(&sit_i->sentry_lock); 3409 mutex_unlock(&curseg->curseg_mutex); 3410 up_write(&SM_I(sbi)->curseg_lock); 3411 } 3412 3413 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, 3414 block_t old_addr, block_t new_addr, 3415 unsigned char version, bool recover_curseg, 3416 bool recover_newaddr) 3417 { 3418 struct f2fs_summary sum; 3419 3420 set_summary(&sum, dn->nid, dn->ofs_in_node, version); 3421 3422 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr, 3423 recover_curseg, recover_newaddr); 3424 3425 f2fs_update_data_blkaddr(dn, new_addr); 3426 } 3427 3428 void f2fs_wait_on_page_writeback(struct page *page, 3429 enum page_type type, bool ordered, bool locked) 3430 { 3431 if (PageWriteback(page)) { 3432 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 3433 3434 /* submit cached LFS IO */ 3435 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type); 3436 /* sbumit cached IPU IO */ 3437 f2fs_submit_merged_ipu_write(sbi, NULL, page); 3438 if (ordered) { 3439 wait_on_page_writeback(page); 3440 f2fs_bug_on(sbi, locked && PageWriteback(page)); 3441 } else { 3442 wait_for_stable_page(page); 3443 } 3444 } 3445 } 3446 3447 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr) 3448 { 3449 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3450 struct page *cpage; 3451 3452 if (!f2fs_post_read_required(inode)) 3453 return; 3454 3455 if (!__is_valid_data_blkaddr(blkaddr)) 3456 return; 3457 3458 cpage = find_lock_page(META_MAPPING(sbi), blkaddr); 3459 if (cpage) { 3460 f2fs_wait_on_page_writeback(cpage, DATA, true, true); 3461 f2fs_put_page(cpage, 1); 3462 } 3463 } 3464 3465 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr, 3466 block_t len) 3467 { 3468 block_t i; 3469 3470 for (i = 0; i < len; i++) 3471 f2fs_wait_on_block_writeback(inode, blkaddr + i); 3472 } 3473 3474 static int read_compacted_summaries(struct f2fs_sb_info *sbi) 3475 { 3476 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3477 struct curseg_info *seg_i; 3478 unsigned char *kaddr; 3479 struct page *page; 3480 block_t start; 3481 int i, j, offset; 3482 3483 start = start_sum_block(sbi); 3484 3485 page = f2fs_get_meta_page(sbi, start++); 3486 if (IS_ERR(page)) 3487 return PTR_ERR(page); 3488 kaddr = (unsigned char *)page_address(page); 3489 3490 /* Step 1: restore nat cache */ 3491 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 3492 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE); 3493 3494 /* Step 2: restore sit cache */ 3495 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 3496 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE); 3497 offset = 2 * SUM_JOURNAL_SIZE; 3498 3499 /* Step 3: restore summary entries */ 3500 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 3501 unsigned short blk_off; 3502 unsigned int segno; 3503 3504 seg_i = CURSEG_I(sbi, i); 3505 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 3506 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 3507 seg_i->next_segno = segno; 3508 reset_curseg(sbi, i, 0); 3509 seg_i->alloc_type = ckpt->alloc_type[i]; 3510 seg_i->next_blkoff = blk_off; 3511 3512 if (seg_i->alloc_type == SSR) 3513 blk_off = sbi->blocks_per_seg; 3514 3515 for (j = 0; j < blk_off; j++) { 3516 struct f2fs_summary *s; 3517 s = (struct f2fs_summary *)(kaddr + offset); 3518 seg_i->sum_blk->entries[j] = *s; 3519 offset += SUMMARY_SIZE; 3520 if (offset + SUMMARY_SIZE <= PAGE_SIZE - 3521 SUM_FOOTER_SIZE) 3522 continue; 3523 3524 f2fs_put_page(page, 1); 3525 page = NULL; 3526 3527 page = f2fs_get_meta_page(sbi, start++); 3528 if (IS_ERR(page)) 3529 return PTR_ERR(page); 3530 kaddr = (unsigned char *)page_address(page); 3531 offset = 0; 3532 } 3533 } 3534 f2fs_put_page(page, 1); 3535 return 0; 3536 } 3537 3538 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 3539 { 3540 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3541 struct f2fs_summary_block *sum; 3542 struct curseg_info *curseg; 3543 struct page *new; 3544 unsigned short blk_off; 3545 unsigned int segno = 0; 3546 block_t blk_addr = 0; 3547 int err = 0; 3548 3549 /* get segment number and block addr */ 3550 if (IS_DATASEG(type)) { 3551 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 3552 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 3553 CURSEG_HOT_DATA]); 3554 if (__exist_node_summaries(sbi)) 3555 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type); 3556 else 3557 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 3558 } else { 3559 segno = le32_to_cpu(ckpt->cur_node_segno[type - 3560 CURSEG_HOT_NODE]); 3561 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 3562 CURSEG_HOT_NODE]); 3563 if (__exist_node_summaries(sbi)) 3564 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 3565 type - CURSEG_HOT_NODE); 3566 else 3567 blk_addr = GET_SUM_BLOCK(sbi, segno); 3568 } 3569 3570 new = f2fs_get_meta_page(sbi, blk_addr); 3571 if (IS_ERR(new)) 3572 return PTR_ERR(new); 3573 sum = (struct f2fs_summary_block *)page_address(new); 3574 3575 if (IS_NODESEG(type)) { 3576 if (__exist_node_summaries(sbi)) { 3577 struct f2fs_summary *ns = &sum->entries[0]; 3578 int i; 3579 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { 3580 ns->version = 0; 3581 ns->ofs_in_node = 0; 3582 } 3583 } else { 3584 err = f2fs_restore_node_summary(sbi, segno, sum); 3585 if (err) 3586 goto out; 3587 } 3588 } 3589 3590 /* set uncompleted segment to curseg */ 3591 curseg = CURSEG_I(sbi, type); 3592 mutex_lock(&curseg->curseg_mutex); 3593 3594 /* update journal info */ 3595 down_write(&curseg->journal_rwsem); 3596 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE); 3597 up_write(&curseg->journal_rwsem); 3598 3599 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE); 3600 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE); 3601 curseg->next_segno = segno; 3602 reset_curseg(sbi, type, 0); 3603 curseg->alloc_type = ckpt->alloc_type[type]; 3604 curseg->next_blkoff = blk_off; 3605 mutex_unlock(&curseg->curseg_mutex); 3606 out: 3607 f2fs_put_page(new, 1); 3608 return err; 3609 } 3610 3611 static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 3612 { 3613 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal; 3614 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal; 3615 int type = CURSEG_HOT_DATA; 3616 int err; 3617 3618 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) { 3619 int npages = f2fs_npages_for_summary_flush(sbi, true); 3620 3621 if (npages >= 2) 3622 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages, 3623 META_CP, true); 3624 3625 /* restore for compacted data summary */ 3626 err = read_compacted_summaries(sbi); 3627 if (err) 3628 return err; 3629 type = CURSEG_HOT_NODE; 3630 } 3631 3632 if (__exist_node_summaries(sbi)) 3633 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type), 3634 NR_CURSEG_TYPE - type, META_CP, true); 3635 3636 for (; type <= CURSEG_COLD_NODE; type++) { 3637 err = read_normal_summaries(sbi, type); 3638 if (err) 3639 return err; 3640 } 3641 3642 /* sanity check for summary blocks */ 3643 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES || 3644 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) { 3645 f2fs_err(sbi, "invalid journal entries nats %u sits %u\n", 3646 nats_in_cursum(nat_j), sits_in_cursum(sit_j)); 3647 return -EINVAL; 3648 } 3649 3650 return 0; 3651 } 3652 3653 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 3654 { 3655 struct page *page; 3656 unsigned char *kaddr; 3657 struct f2fs_summary *summary; 3658 struct curseg_info *seg_i; 3659 int written_size = 0; 3660 int i, j; 3661 3662 page = f2fs_grab_meta_page(sbi, blkaddr++); 3663 kaddr = (unsigned char *)page_address(page); 3664 memset(kaddr, 0, PAGE_SIZE); 3665 3666 /* Step 1: write nat cache */ 3667 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 3668 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE); 3669 written_size += SUM_JOURNAL_SIZE; 3670 3671 /* Step 2: write sit cache */ 3672 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 3673 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE); 3674 written_size += SUM_JOURNAL_SIZE; 3675 3676 /* Step 3: write summary entries */ 3677 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 3678 unsigned short blkoff; 3679 seg_i = CURSEG_I(sbi, i); 3680 if (sbi->ckpt->alloc_type[i] == SSR) 3681 blkoff = sbi->blocks_per_seg; 3682 else 3683 blkoff = curseg_blkoff(sbi, i); 3684 3685 for (j = 0; j < blkoff; j++) { 3686 if (!page) { 3687 page = f2fs_grab_meta_page(sbi, blkaddr++); 3688 kaddr = (unsigned char *)page_address(page); 3689 memset(kaddr, 0, PAGE_SIZE); 3690 written_size = 0; 3691 } 3692 summary = (struct f2fs_summary *)(kaddr + written_size); 3693 *summary = seg_i->sum_blk->entries[j]; 3694 written_size += SUMMARY_SIZE; 3695 3696 if (written_size + SUMMARY_SIZE <= PAGE_SIZE - 3697 SUM_FOOTER_SIZE) 3698 continue; 3699 3700 set_page_dirty(page); 3701 f2fs_put_page(page, 1); 3702 page = NULL; 3703 } 3704 } 3705 if (page) { 3706 set_page_dirty(page); 3707 f2fs_put_page(page, 1); 3708 } 3709 } 3710 3711 static void write_normal_summaries(struct f2fs_sb_info *sbi, 3712 block_t blkaddr, int type) 3713 { 3714 int i, end; 3715 if (IS_DATASEG(type)) 3716 end = type + NR_CURSEG_DATA_TYPE; 3717 else 3718 end = type + NR_CURSEG_NODE_TYPE; 3719 3720 for (i = type; i < end; i++) 3721 write_current_sum_page(sbi, i, blkaddr + (i - type)); 3722 } 3723 3724 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 3725 { 3726 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) 3727 write_compacted_summaries(sbi, start_blk); 3728 else 3729 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 3730 } 3731 3732 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 3733 { 3734 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 3735 } 3736 3737 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type, 3738 unsigned int val, int alloc) 3739 { 3740 int i; 3741 3742 if (type == NAT_JOURNAL) { 3743 for (i = 0; i < nats_in_cursum(journal); i++) { 3744 if (le32_to_cpu(nid_in_journal(journal, i)) == val) 3745 return i; 3746 } 3747 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL)) 3748 return update_nats_in_cursum(journal, 1); 3749 } else if (type == SIT_JOURNAL) { 3750 for (i = 0; i < sits_in_cursum(journal); i++) 3751 if (le32_to_cpu(segno_in_journal(journal, i)) == val) 3752 return i; 3753 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL)) 3754 return update_sits_in_cursum(journal, 1); 3755 } 3756 return -1; 3757 } 3758 3759 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, 3760 unsigned int segno) 3761 { 3762 return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno)); 3763 } 3764 3765 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, 3766 unsigned int start) 3767 { 3768 struct sit_info *sit_i = SIT_I(sbi); 3769 struct page *page; 3770 pgoff_t src_off, dst_off; 3771 3772 src_off = current_sit_addr(sbi, start); 3773 dst_off = next_sit_addr(sbi, src_off); 3774 3775 page = f2fs_grab_meta_page(sbi, dst_off); 3776 seg_info_to_sit_page(sbi, page, start); 3777 3778 set_page_dirty(page); 3779 set_to_next_sit(sit_i, start); 3780 3781 return page; 3782 } 3783 3784 static struct sit_entry_set *grab_sit_entry_set(void) 3785 { 3786 struct sit_entry_set *ses = 3787 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS); 3788 3789 ses->entry_cnt = 0; 3790 INIT_LIST_HEAD(&ses->set_list); 3791 return ses; 3792 } 3793 3794 static void release_sit_entry_set(struct sit_entry_set *ses) 3795 { 3796 list_del(&ses->set_list); 3797 kmem_cache_free(sit_entry_set_slab, ses); 3798 } 3799 3800 static void adjust_sit_entry_set(struct sit_entry_set *ses, 3801 struct list_head *head) 3802 { 3803 struct sit_entry_set *next = ses; 3804 3805 if (list_is_last(&ses->set_list, head)) 3806 return; 3807 3808 list_for_each_entry_continue(next, head, set_list) 3809 if (ses->entry_cnt <= next->entry_cnt) 3810 break; 3811 3812 list_move_tail(&ses->set_list, &next->set_list); 3813 } 3814 3815 static void add_sit_entry(unsigned int segno, struct list_head *head) 3816 { 3817 struct sit_entry_set *ses; 3818 unsigned int start_segno = START_SEGNO(segno); 3819 3820 list_for_each_entry(ses, head, set_list) { 3821 if (ses->start_segno == start_segno) { 3822 ses->entry_cnt++; 3823 adjust_sit_entry_set(ses, head); 3824 return; 3825 } 3826 } 3827 3828 ses = grab_sit_entry_set(); 3829 3830 ses->start_segno = start_segno; 3831 ses->entry_cnt++; 3832 list_add(&ses->set_list, head); 3833 } 3834 3835 static void add_sits_in_set(struct f2fs_sb_info *sbi) 3836 { 3837 struct f2fs_sm_info *sm_info = SM_I(sbi); 3838 struct list_head *set_list = &sm_info->sit_entry_set; 3839 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; 3840 unsigned int segno; 3841 3842 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) 3843 add_sit_entry(segno, set_list); 3844 } 3845 3846 static void remove_sits_in_journal(struct f2fs_sb_info *sbi) 3847 { 3848 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 3849 struct f2fs_journal *journal = curseg->journal; 3850 int i; 3851 3852 down_write(&curseg->journal_rwsem); 3853 for (i = 0; i < sits_in_cursum(journal); i++) { 3854 unsigned int segno; 3855 bool dirtied; 3856 3857 segno = le32_to_cpu(segno_in_journal(journal, i)); 3858 dirtied = __mark_sit_entry_dirty(sbi, segno); 3859 3860 if (!dirtied) 3861 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); 3862 } 3863 update_sits_in_cursum(journal, -i); 3864 up_write(&curseg->journal_rwsem); 3865 } 3866 3867 /* 3868 * CP calls this function, which flushes SIT entries including sit_journal, 3869 * and moves prefree segs to free segs. 3870 */ 3871 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 3872 { 3873 struct sit_info *sit_i = SIT_I(sbi); 3874 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 3875 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 3876 struct f2fs_journal *journal = curseg->journal; 3877 struct sit_entry_set *ses, *tmp; 3878 struct list_head *head = &SM_I(sbi)->sit_entry_set; 3879 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS); 3880 struct seg_entry *se; 3881 3882 down_write(&sit_i->sentry_lock); 3883 3884 if (!sit_i->dirty_sentries) 3885 goto out; 3886 3887 /* 3888 * add and account sit entries of dirty bitmap in sit entry 3889 * set temporarily 3890 */ 3891 add_sits_in_set(sbi); 3892 3893 /* 3894 * if there are no enough space in journal to store dirty sit 3895 * entries, remove all entries from journal and add and account 3896 * them in sit entry set. 3897 */ 3898 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) || 3899 !to_journal) 3900 remove_sits_in_journal(sbi); 3901 3902 /* 3903 * there are two steps to flush sit entries: 3904 * #1, flush sit entries to journal in current cold data summary block. 3905 * #2, flush sit entries to sit page. 3906 */ 3907 list_for_each_entry_safe(ses, tmp, head, set_list) { 3908 struct page *page = NULL; 3909 struct f2fs_sit_block *raw_sit = NULL; 3910 unsigned int start_segno = ses->start_segno; 3911 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, 3912 (unsigned long)MAIN_SEGS(sbi)); 3913 unsigned int segno = start_segno; 3914 3915 if (to_journal && 3916 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL)) 3917 to_journal = false; 3918 3919 if (to_journal) { 3920 down_write(&curseg->journal_rwsem); 3921 } else { 3922 page = get_next_sit_page(sbi, start_segno); 3923 raw_sit = page_address(page); 3924 } 3925 3926 /* flush dirty sit entries in region of current sit set */ 3927 for_each_set_bit_from(segno, bitmap, end) { 3928 int offset, sit_offset; 3929 3930 se = get_seg_entry(sbi, segno); 3931 #ifdef CONFIG_F2FS_CHECK_FS 3932 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir, 3933 SIT_VBLOCK_MAP_SIZE)) 3934 f2fs_bug_on(sbi, 1); 3935 #endif 3936 3937 /* add discard candidates */ 3938 if (!(cpc->reason & CP_DISCARD)) { 3939 cpc->trim_start = segno; 3940 add_discard_addrs(sbi, cpc, false); 3941 } 3942 3943 if (to_journal) { 3944 offset = f2fs_lookup_journal_in_cursum(journal, 3945 SIT_JOURNAL, segno, 1); 3946 f2fs_bug_on(sbi, offset < 0); 3947 segno_in_journal(journal, offset) = 3948 cpu_to_le32(segno); 3949 seg_info_to_raw_sit(se, 3950 &sit_in_journal(journal, offset)); 3951 check_block_count(sbi, segno, 3952 &sit_in_journal(journal, offset)); 3953 } else { 3954 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 3955 seg_info_to_raw_sit(se, 3956 &raw_sit->entries[sit_offset]); 3957 check_block_count(sbi, segno, 3958 &raw_sit->entries[sit_offset]); 3959 } 3960 3961 __clear_bit(segno, bitmap); 3962 sit_i->dirty_sentries--; 3963 ses->entry_cnt--; 3964 } 3965 3966 if (to_journal) 3967 up_write(&curseg->journal_rwsem); 3968 else 3969 f2fs_put_page(page, 1); 3970 3971 f2fs_bug_on(sbi, ses->entry_cnt); 3972 release_sit_entry_set(ses); 3973 } 3974 3975 f2fs_bug_on(sbi, !list_empty(head)); 3976 f2fs_bug_on(sbi, sit_i->dirty_sentries); 3977 out: 3978 if (cpc->reason & CP_DISCARD) { 3979 __u64 trim_start = cpc->trim_start; 3980 3981 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) 3982 add_discard_addrs(sbi, cpc, false); 3983 3984 cpc->trim_start = trim_start; 3985 } 3986 up_write(&sit_i->sentry_lock); 3987 3988 set_prefree_as_free_segments(sbi); 3989 } 3990 3991 static int build_sit_info(struct f2fs_sb_info *sbi) 3992 { 3993 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 3994 struct sit_info *sit_i; 3995 unsigned int sit_segs, start; 3996 char *src_bitmap, *bitmap; 3997 unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size; 3998 3999 /* allocate memory for SIT information */ 4000 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL); 4001 if (!sit_i) 4002 return -ENOMEM; 4003 4004 SM_I(sbi)->sit_info = sit_i; 4005 4006 sit_i->sentries = 4007 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry), 4008 MAIN_SEGS(sbi)), 4009 GFP_KERNEL); 4010 if (!sit_i->sentries) 4011 return -ENOMEM; 4012 4013 main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4014 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size, 4015 GFP_KERNEL); 4016 if (!sit_i->dirty_sentries_bitmap) 4017 return -ENOMEM; 4018 4019 #ifdef CONFIG_F2FS_CHECK_FS 4020 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 4; 4021 #else 4022 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 3; 4023 #endif 4024 sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4025 if (!sit_i->bitmap) 4026 return -ENOMEM; 4027 4028 bitmap = sit_i->bitmap; 4029 4030 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4031 sit_i->sentries[start].cur_valid_map = bitmap; 4032 bitmap += SIT_VBLOCK_MAP_SIZE; 4033 4034 sit_i->sentries[start].ckpt_valid_map = bitmap; 4035 bitmap += SIT_VBLOCK_MAP_SIZE; 4036 4037 #ifdef CONFIG_F2FS_CHECK_FS 4038 sit_i->sentries[start].cur_valid_map_mir = bitmap; 4039 bitmap += SIT_VBLOCK_MAP_SIZE; 4040 #endif 4041 4042 sit_i->sentries[start].discard_map = bitmap; 4043 bitmap += SIT_VBLOCK_MAP_SIZE; 4044 } 4045 4046 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 4047 if (!sit_i->tmp_map) 4048 return -ENOMEM; 4049 4050 if (__is_large_section(sbi)) { 4051 sit_i->sec_entries = 4052 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry), 4053 MAIN_SECS(sbi)), 4054 GFP_KERNEL); 4055 if (!sit_i->sec_entries) 4056 return -ENOMEM; 4057 } 4058 4059 /* get information related with SIT */ 4060 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 4061 4062 /* setup SIT bitmap from ckeckpoint pack */ 4063 sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 4064 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 4065 4066 sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL); 4067 if (!sit_i->sit_bitmap) 4068 return -ENOMEM; 4069 4070 #ifdef CONFIG_F2FS_CHECK_FS 4071 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, 4072 sit_bitmap_size, GFP_KERNEL); 4073 if (!sit_i->sit_bitmap_mir) 4074 return -ENOMEM; 4075 4076 sit_i->invalid_segmap = f2fs_kvzalloc(sbi, 4077 main_bitmap_size, GFP_KERNEL); 4078 if (!sit_i->invalid_segmap) 4079 return -ENOMEM; 4080 #endif 4081 4082 /* init SIT information */ 4083 sit_i->s_ops = &default_salloc_ops; 4084 4085 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 4086 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; 4087 sit_i->written_valid_blocks = 0; 4088 sit_i->bitmap_size = sit_bitmap_size; 4089 sit_i->dirty_sentries = 0; 4090 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 4091 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 4092 sit_i->mounted_time = ktime_get_boottime_seconds(); 4093 init_rwsem(&sit_i->sentry_lock); 4094 return 0; 4095 } 4096 4097 static int build_free_segmap(struct f2fs_sb_info *sbi) 4098 { 4099 struct free_segmap_info *free_i; 4100 unsigned int bitmap_size, sec_bitmap_size; 4101 4102 /* allocate memory for free segmap information */ 4103 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL); 4104 if (!free_i) 4105 return -ENOMEM; 4106 4107 SM_I(sbi)->free_info = free_i; 4108 4109 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4110 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL); 4111 if (!free_i->free_segmap) 4112 return -ENOMEM; 4113 4114 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4115 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL); 4116 if (!free_i->free_secmap) 4117 return -ENOMEM; 4118 4119 /* set all segments as dirty temporarily */ 4120 memset(free_i->free_segmap, 0xff, bitmap_size); 4121 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 4122 4123 /* init free segmap information */ 4124 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); 4125 free_i->free_segments = 0; 4126 free_i->free_sections = 0; 4127 spin_lock_init(&free_i->segmap_lock); 4128 return 0; 4129 } 4130 4131 static int build_curseg(struct f2fs_sb_info *sbi) 4132 { 4133 struct curseg_info *array; 4134 int i; 4135 4136 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)), 4137 GFP_KERNEL); 4138 if (!array) 4139 return -ENOMEM; 4140 4141 SM_I(sbi)->curseg_array = array; 4142 4143 for (i = 0; i < NR_CURSEG_TYPE; i++) { 4144 mutex_init(&array[i].curseg_mutex); 4145 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL); 4146 if (!array[i].sum_blk) 4147 return -ENOMEM; 4148 init_rwsem(&array[i].journal_rwsem); 4149 array[i].journal = f2fs_kzalloc(sbi, 4150 sizeof(struct f2fs_journal), GFP_KERNEL); 4151 if (!array[i].journal) 4152 return -ENOMEM; 4153 array[i].segno = NULL_SEGNO; 4154 array[i].next_blkoff = 0; 4155 } 4156 return restore_curseg_summaries(sbi); 4157 } 4158 4159 static int build_sit_entries(struct f2fs_sb_info *sbi) 4160 { 4161 struct sit_info *sit_i = SIT_I(sbi); 4162 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4163 struct f2fs_journal *journal = curseg->journal; 4164 struct seg_entry *se; 4165 struct f2fs_sit_entry sit; 4166 int sit_blk_cnt = SIT_BLK_CNT(sbi); 4167 unsigned int i, start, end; 4168 unsigned int readed, start_blk = 0; 4169 int err = 0; 4170 block_t total_node_blocks = 0; 4171 4172 do { 4173 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES, 4174 META_SIT, true); 4175 4176 start = start_blk * sit_i->sents_per_block; 4177 end = (start_blk + readed) * sit_i->sents_per_block; 4178 4179 for (; start < end && start < MAIN_SEGS(sbi); start++) { 4180 struct f2fs_sit_block *sit_blk; 4181 struct page *page; 4182 4183 se = &sit_i->sentries[start]; 4184 page = get_current_sit_page(sbi, start); 4185 if (IS_ERR(page)) 4186 return PTR_ERR(page); 4187 sit_blk = (struct f2fs_sit_block *)page_address(page); 4188 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 4189 f2fs_put_page(page, 1); 4190 4191 err = check_block_count(sbi, start, &sit); 4192 if (err) 4193 return err; 4194 seg_info_from_raw_sit(se, &sit); 4195 if (IS_NODESEG(se->type)) 4196 total_node_blocks += se->valid_blocks; 4197 4198 /* build discard map only one time */ 4199 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4200 memset(se->discard_map, 0xff, 4201 SIT_VBLOCK_MAP_SIZE); 4202 } else { 4203 memcpy(se->discard_map, 4204 se->cur_valid_map, 4205 SIT_VBLOCK_MAP_SIZE); 4206 sbi->discard_blks += 4207 sbi->blocks_per_seg - 4208 se->valid_blocks; 4209 } 4210 4211 if (__is_large_section(sbi)) 4212 get_sec_entry(sbi, start)->valid_blocks += 4213 se->valid_blocks; 4214 } 4215 start_blk += readed; 4216 } while (start_blk < sit_blk_cnt); 4217 4218 down_read(&curseg->journal_rwsem); 4219 for (i = 0; i < sits_in_cursum(journal); i++) { 4220 unsigned int old_valid_blocks; 4221 4222 start = le32_to_cpu(segno_in_journal(journal, i)); 4223 if (start >= MAIN_SEGS(sbi)) { 4224 f2fs_err(sbi, "Wrong journal entry on segno %u", 4225 start); 4226 err = -EFSCORRUPTED; 4227 break; 4228 } 4229 4230 se = &sit_i->sentries[start]; 4231 sit = sit_in_journal(journal, i); 4232 4233 old_valid_blocks = se->valid_blocks; 4234 if (IS_NODESEG(se->type)) 4235 total_node_blocks -= old_valid_blocks; 4236 4237 err = check_block_count(sbi, start, &sit); 4238 if (err) 4239 break; 4240 seg_info_from_raw_sit(se, &sit); 4241 if (IS_NODESEG(se->type)) 4242 total_node_blocks += se->valid_blocks; 4243 4244 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4245 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE); 4246 } else { 4247 memcpy(se->discard_map, se->cur_valid_map, 4248 SIT_VBLOCK_MAP_SIZE); 4249 sbi->discard_blks += old_valid_blocks; 4250 sbi->discard_blks -= se->valid_blocks; 4251 } 4252 4253 if (__is_large_section(sbi)) { 4254 get_sec_entry(sbi, start)->valid_blocks += 4255 se->valid_blocks; 4256 get_sec_entry(sbi, start)->valid_blocks -= 4257 old_valid_blocks; 4258 } 4259 } 4260 up_read(&curseg->journal_rwsem); 4261 4262 if (!err && total_node_blocks != valid_node_count(sbi)) { 4263 f2fs_err(sbi, "SIT is corrupted node# %u vs %u", 4264 total_node_blocks, valid_node_count(sbi)); 4265 err = -EFSCORRUPTED; 4266 } 4267 4268 return err; 4269 } 4270 4271 static void init_free_segmap(struct f2fs_sb_info *sbi) 4272 { 4273 unsigned int start; 4274 int type; 4275 4276 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4277 struct seg_entry *sentry = get_seg_entry(sbi, start); 4278 if (!sentry->valid_blocks) 4279 __set_free(sbi, start); 4280 else 4281 SIT_I(sbi)->written_valid_blocks += 4282 sentry->valid_blocks; 4283 } 4284 4285 /* set use the current segments */ 4286 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 4287 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 4288 __set_test_and_inuse(sbi, curseg_t->segno); 4289 } 4290 } 4291 4292 static void init_dirty_segmap(struct f2fs_sb_info *sbi) 4293 { 4294 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4295 struct free_segmap_info *free_i = FREE_I(sbi); 4296 unsigned int segno = 0, offset = 0; 4297 unsigned short valid_blocks; 4298 4299 while (1) { 4300 /* find dirty segment based on free segmap */ 4301 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); 4302 if (segno >= MAIN_SEGS(sbi)) 4303 break; 4304 offset = segno + 1; 4305 valid_blocks = get_valid_blocks(sbi, segno, false); 4306 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks) 4307 continue; 4308 if (valid_blocks > sbi->blocks_per_seg) { 4309 f2fs_bug_on(sbi, 1); 4310 continue; 4311 } 4312 mutex_lock(&dirty_i->seglist_lock); 4313 __locate_dirty_segment(sbi, segno, DIRTY); 4314 mutex_unlock(&dirty_i->seglist_lock); 4315 } 4316 } 4317 4318 static int init_victim_secmap(struct f2fs_sb_info *sbi) 4319 { 4320 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4321 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4322 4323 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4324 if (!dirty_i->victim_secmap) 4325 return -ENOMEM; 4326 return 0; 4327 } 4328 4329 static int build_dirty_segmap(struct f2fs_sb_info *sbi) 4330 { 4331 struct dirty_seglist_info *dirty_i; 4332 unsigned int bitmap_size, i; 4333 4334 /* allocate memory for dirty segments list information */ 4335 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info), 4336 GFP_KERNEL); 4337 if (!dirty_i) 4338 return -ENOMEM; 4339 4340 SM_I(sbi)->dirty_info = dirty_i; 4341 mutex_init(&dirty_i->seglist_lock); 4342 4343 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4344 4345 for (i = 0; i < NR_DIRTY_TYPE; i++) { 4346 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size, 4347 GFP_KERNEL); 4348 if (!dirty_i->dirty_segmap[i]) 4349 return -ENOMEM; 4350 } 4351 4352 init_dirty_segmap(sbi); 4353 return init_victim_secmap(sbi); 4354 } 4355 4356 static int sanity_check_curseg(struct f2fs_sb_info *sbi) 4357 { 4358 int i; 4359 4360 /* 4361 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr; 4362 * In LFS curseg, all blkaddr after .next_blkoff should be unused. 4363 */ 4364 for (i = 0; i < NO_CHECK_TYPE; i++) { 4365 struct curseg_info *curseg = CURSEG_I(sbi, i); 4366 struct seg_entry *se = get_seg_entry(sbi, curseg->segno); 4367 unsigned int blkofs = curseg->next_blkoff; 4368 4369 if (f2fs_test_bit(blkofs, se->cur_valid_map)) 4370 goto out; 4371 4372 if (curseg->alloc_type == SSR) 4373 continue; 4374 4375 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) { 4376 if (!f2fs_test_bit(blkofs, se->cur_valid_map)) 4377 continue; 4378 out: 4379 f2fs_err(sbi, 4380 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u", 4381 i, curseg->segno, curseg->alloc_type, 4382 curseg->next_blkoff, blkofs); 4383 return -EFSCORRUPTED; 4384 } 4385 } 4386 return 0; 4387 } 4388 4389 #ifdef CONFIG_BLK_DEV_ZONED 4390 4391 static int check_zone_write_pointer(struct f2fs_sb_info *sbi, 4392 struct f2fs_dev_info *fdev, 4393 struct blk_zone *zone) 4394 { 4395 unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno; 4396 block_t zone_block, wp_block, last_valid_block; 4397 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 4398 int i, s, b, ret; 4399 struct seg_entry *se; 4400 4401 if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4402 return 0; 4403 4404 wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block); 4405 wp_segno = GET_SEGNO(sbi, wp_block); 4406 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno); 4407 zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block); 4408 zone_segno = GET_SEGNO(sbi, zone_block); 4409 zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno); 4410 4411 if (zone_segno >= MAIN_SEGS(sbi)) 4412 return 0; 4413 4414 /* 4415 * Skip check of zones cursegs point to, since 4416 * fix_curseg_write_pointer() checks them. 4417 */ 4418 for (i = 0; i < NO_CHECK_TYPE; i++) 4419 if (zone_secno == GET_SEC_FROM_SEG(sbi, 4420 CURSEG_I(sbi, i)->segno)) 4421 return 0; 4422 4423 /* 4424 * Get last valid block of the zone. 4425 */ 4426 last_valid_block = zone_block - 1; 4427 for (s = sbi->segs_per_sec - 1; s >= 0; s--) { 4428 segno = zone_segno + s; 4429 se = get_seg_entry(sbi, segno); 4430 for (b = sbi->blocks_per_seg - 1; b >= 0; b--) 4431 if (f2fs_test_bit(b, se->cur_valid_map)) { 4432 last_valid_block = START_BLOCK(sbi, segno) + b; 4433 break; 4434 } 4435 if (last_valid_block >= zone_block) 4436 break; 4437 } 4438 4439 /* 4440 * If last valid block is beyond the write pointer, report the 4441 * inconsistency. This inconsistency does not cause write error 4442 * because the zone will not be selected for write operation until 4443 * it get discarded. Just report it. 4444 */ 4445 if (last_valid_block >= wp_block) { 4446 f2fs_notice(sbi, "Valid block beyond write pointer: " 4447 "valid block[0x%x,0x%x] wp[0x%x,0x%x]", 4448 GET_SEGNO(sbi, last_valid_block), 4449 GET_BLKOFF_FROM_SEG0(sbi, last_valid_block), 4450 wp_segno, wp_blkoff); 4451 return 0; 4452 } 4453 4454 /* 4455 * If there is no valid block in the zone and if write pointer is 4456 * not at zone start, reset the write pointer. 4457 */ 4458 if (last_valid_block + 1 == zone_block && zone->wp != zone->start) { 4459 f2fs_notice(sbi, 4460 "Zone without valid block has non-zero write " 4461 "pointer. Reset the write pointer: wp[0x%x,0x%x]", 4462 wp_segno, wp_blkoff); 4463 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block, 4464 zone->len >> log_sectors_per_block); 4465 if (ret) { 4466 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 4467 fdev->path, ret); 4468 return ret; 4469 } 4470 } 4471 4472 return 0; 4473 } 4474 4475 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi, 4476 block_t zone_blkaddr) 4477 { 4478 int i; 4479 4480 for (i = 0; i < sbi->s_ndevs; i++) { 4481 if (!bdev_is_zoned(FDEV(i).bdev)) 4482 continue; 4483 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr && 4484 zone_blkaddr <= FDEV(i).end_blk)) 4485 return &FDEV(i); 4486 } 4487 4488 return NULL; 4489 } 4490 4491 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx, 4492 void *data) { 4493 memcpy(data, zone, sizeof(struct blk_zone)); 4494 return 0; 4495 } 4496 4497 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type) 4498 { 4499 struct curseg_info *cs = CURSEG_I(sbi, type); 4500 struct f2fs_dev_info *zbd; 4501 struct blk_zone zone; 4502 unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off; 4503 block_t cs_zone_block, wp_block; 4504 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 4505 sector_t zone_sector; 4506 int err; 4507 4508 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 4509 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 4510 4511 zbd = get_target_zoned_dev(sbi, cs_zone_block); 4512 if (!zbd) 4513 return 0; 4514 4515 /* report zone for the sector the curseg points to */ 4516 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 4517 << log_sectors_per_block; 4518 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 4519 report_one_zone_cb, &zone); 4520 if (err != 1) { 4521 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 4522 zbd->path, err); 4523 return err; 4524 } 4525 4526 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4527 return 0; 4528 4529 wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block); 4530 wp_segno = GET_SEGNO(sbi, wp_block); 4531 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno); 4532 wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0); 4533 4534 if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff && 4535 wp_sector_off == 0) 4536 return 0; 4537 4538 f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: " 4539 "curseg[0x%x,0x%x] wp[0x%x,0x%x]", 4540 type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff); 4541 4542 f2fs_notice(sbi, "Assign new section to curseg[%d]: " 4543 "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff); 4544 allocate_segment_by_default(sbi, type, true); 4545 4546 /* check consistency of the zone curseg pointed to */ 4547 if (check_zone_write_pointer(sbi, zbd, &zone)) 4548 return -EIO; 4549 4550 /* check newly assigned zone */ 4551 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 4552 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 4553 4554 zbd = get_target_zoned_dev(sbi, cs_zone_block); 4555 if (!zbd) 4556 return 0; 4557 4558 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 4559 << log_sectors_per_block; 4560 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 4561 report_one_zone_cb, &zone); 4562 if (err != 1) { 4563 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 4564 zbd->path, err); 4565 return err; 4566 } 4567 4568 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4569 return 0; 4570 4571 if (zone.wp != zone.start) { 4572 f2fs_notice(sbi, 4573 "New zone for curseg[%d] is not yet discarded. " 4574 "Reset the zone: curseg[0x%x,0x%x]", 4575 type, cs->segno, cs->next_blkoff); 4576 err = __f2fs_issue_discard_zone(sbi, zbd->bdev, 4577 zone_sector >> log_sectors_per_block, 4578 zone.len >> log_sectors_per_block); 4579 if (err) { 4580 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 4581 zbd->path, err); 4582 return err; 4583 } 4584 } 4585 4586 return 0; 4587 } 4588 4589 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 4590 { 4591 int i, ret; 4592 4593 for (i = 0; i < NO_CHECK_TYPE; i++) { 4594 ret = fix_curseg_write_pointer(sbi, i); 4595 if (ret) 4596 return ret; 4597 } 4598 4599 return 0; 4600 } 4601 4602 struct check_zone_write_pointer_args { 4603 struct f2fs_sb_info *sbi; 4604 struct f2fs_dev_info *fdev; 4605 }; 4606 4607 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx, 4608 void *data) { 4609 struct check_zone_write_pointer_args *args; 4610 args = (struct check_zone_write_pointer_args *)data; 4611 4612 return check_zone_write_pointer(args->sbi, args->fdev, zone); 4613 } 4614 4615 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi) 4616 { 4617 int i, ret; 4618 struct check_zone_write_pointer_args args; 4619 4620 for (i = 0; i < sbi->s_ndevs; i++) { 4621 if (!bdev_is_zoned(FDEV(i).bdev)) 4622 continue; 4623 4624 args.sbi = sbi; 4625 args.fdev = &FDEV(i); 4626 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES, 4627 check_zone_write_pointer_cb, &args); 4628 if (ret < 0) 4629 return ret; 4630 } 4631 4632 return 0; 4633 } 4634 #else 4635 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 4636 { 4637 return 0; 4638 } 4639 4640 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi) 4641 { 4642 return 0; 4643 } 4644 #endif 4645 4646 /* 4647 * Update min, max modified time for cost-benefit GC algorithm 4648 */ 4649 static void init_min_max_mtime(struct f2fs_sb_info *sbi) 4650 { 4651 struct sit_info *sit_i = SIT_I(sbi); 4652 unsigned int segno; 4653 4654 down_write(&sit_i->sentry_lock); 4655 4656 sit_i->min_mtime = ULLONG_MAX; 4657 4658 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 4659 unsigned int i; 4660 unsigned long long mtime = 0; 4661 4662 for (i = 0; i < sbi->segs_per_sec; i++) 4663 mtime += get_seg_entry(sbi, segno + i)->mtime; 4664 4665 mtime = div_u64(mtime, sbi->segs_per_sec); 4666 4667 if (sit_i->min_mtime > mtime) 4668 sit_i->min_mtime = mtime; 4669 } 4670 sit_i->max_mtime = get_mtime(sbi, false); 4671 up_write(&sit_i->sentry_lock); 4672 } 4673 4674 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi) 4675 { 4676 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 4677 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 4678 struct f2fs_sm_info *sm_info; 4679 int err; 4680 4681 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL); 4682 if (!sm_info) 4683 return -ENOMEM; 4684 4685 /* init sm info */ 4686 sbi->sm_info = sm_info; 4687 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 4688 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 4689 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 4690 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 4691 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 4692 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 4693 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 4694 sm_info->rec_prefree_segments = sm_info->main_segments * 4695 DEF_RECLAIM_PREFREE_SEGMENTS / 100; 4696 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS) 4697 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS; 4698 4699 if (!f2fs_lfs_mode(sbi)) 4700 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC; 4701 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; 4702 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; 4703 sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec; 4704 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS; 4705 sm_info->min_ssr_sections = reserved_sections(sbi); 4706 4707 INIT_LIST_HEAD(&sm_info->sit_entry_set); 4708 4709 init_rwsem(&sm_info->curseg_lock); 4710 4711 if (!f2fs_readonly(sbi->sb)) { 4712 err = f2fs_create_flush_cmd_control(sbi); 4713 if (err) 4714 return err; 4715 } 4716 4717 err = create_discard_cmd_control(sbi); 4718 if (err) 4719 return err; 4720 4721 err = build_sit_info(sbi); 4722 if (err) 4723 return err; 4724 err = build_free_segmap(sbi); 4725 if (err) 4726 return err; 4727 err = build_curseg(sbi); 4728 if (err) 4729 return err; 4730 4731 /* reinit free segmap based on SIT */ 4732 err = build_sit_entries(sbi); 4733 if (err) 4734 return err; 4735 4736 init_free_segmap(sbi); 4737 err = build_dirty_segmap(sbi); 4738 if (err) 4739 return err; 4740 4741 err = sanity_check_curseg(sbi); 4742 if (err) 4743 return err; 4744 4745 init_min_max_mtime(sbi); 4746 return 0; 4747 } 4748 4749 static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 4750 enum dirty_type dirty_type) 4751 { 4752 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4753 4754 mutex_lock(&dirty_i->seglist_lock); 4755 kvfree(dirty_i->dirty_segmap[dirty_type]); 4756 dirty_i->nr_dirty[dirty_type] = 0; 4757 mutex_unlock(&dirty_i->seglist_lock); 4758 } 4759 4760 static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 4761 { 4762 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4763 kvfree(dirty_i->victim_secmap); 4764 } 4765 4766 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 4767 { 4768 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4769 int i; 4770 4771 if (!dirty_i) 4772 return; 4773 4774 /* discard pre-free/dirty segments list */ 4775 for (i = 0; i < NR_DIRTY_TYPE; i++) 4776 discard_dirty_segmap(sbi, i); 4777 4778 destroy_victim_secmap(sbi); 4779 SM_I(sbi)->dirty_info = NULL; 4780 kvfree(dirty_i); 4781 } 4782 4783 static void destroy_curseg(struct f2fs_sb_info *sbi) 4784 { 4785 struct curseg_info *array = SM_I(sbi)->curseg_array; 4786 int i; 4787 4788 if (!array) 4789 return; 4790 SM_I(sbi)->curseg_array = NULL; 4791 for (i = 0; i < NR_CURSEG_TYPE; i++) { 4792 kvfree(array[i].sum_blk); 4793 kvfree(array[i].journal); 4794 } 4795 kvfree(array); 4796 } 4797 4798 static void destroy_free_segmap(struct f2fs_sb_info *sbi) 4799 { 4800 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 4801 if (!free_i) 4802 return; 4803 SM_I(sbi)->free_info = NULL; 4804 kvfree(free_i->free_segmap); 4805 kvfree(free_i->free_secmap); 4806 kvfree(free_i); 4807 } 4808 4809 static void destroy_sit_info(struct f2fs_sb_info *sbi) 4810 { 4811 struct sit_info *sit_i = SIT_I(sbi); 4812 4813 if (!sit_i) 4814 return; 4815 4816 if (sit_i->sentries) 4817 kvfree(sit_i->bitmap); 4818 kvfree(sit_i->tmp_map); 4819 4820 kvfree(sit_i->sentries); 4821 kvfree(sit_i->sec_entries); 4822 kvfree(sit_i->dirty_sentries_bitmap); 4823 4824 SM_I(sbi)->sit_info = NULL; 4825 kvfree(sit_i->sit_bitmap); 4826 #ifdef CONFIG_F2FS_CHECK_FS 4827 kvfree(sit_i->sit_bitmap_mir); 4828 kvfree(sit_i->invalid_segmap); 4829 #endif 4830 kvfree(sit_i); 4831 } 4832 4833 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi) 4834 { 4835 struct f2fs_sm_info *sm_info = SM_I(sbi); 4836 4837 if (!sm_info) 4838 return; 4839 f2fs_destroy_flush_cmd_control(sbi, true); 4840 destroy_discard_cmd_control(sbi); 4841 destroy_dirty_segmap(sbi); 4842 destroy_curseg(sbi); 4843 destroy_free_segmap(sbi); 4844 destroy_sit_info(sbi); 4845 sbi->sm_info = NULL; 4846 kvfree(sm_info); 4847 } 4848 4849 int __init f2fs_create_segment_manager_caches(void) 4850 { 4851 discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry", 4852 sizeof(struct discard_entry)); 4853 if (!discard_entry_slab) 4854 goto fail; 4855 4856 discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd", 4857 sizeof(struct discard_cmd)); 4858 if (!discard_cmd_slab) 4859 goto destroy_discard_entry; 4860 4861 sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set", 4862 sizeof(struct sit_entry_set)); 4863 if (!sit_entry_set_slab) 4864 goto destroy_discard_cmd; 4865 4866 inmem_entry_slab = f2fs_kmem_cache_create("f2fs_inmem_page_entry", 4867 sizeof(struct inmem_pages)); 4868 if (!inmem_entry_slab) 4869 goto destroy_sit_entry_set; 4870 return 0; 4871 4872 destroy_sit_entry_set: 4873 kmem_cache_destroy(sit_entry_set_slab); 4874 destroy_discard_cmd: 4875 kmem_cache_destroy(discard_cmd_slab); 4876 destroy_discard_entry: 4877 kmem_cache_destroy(discard_entry_slab); 4878 fail: 4879 return -ENOMEM; 4880 } 4881 4882 void f2fs_destroy_segment_manager_caches(void) 4883 { 4884 kmem_cache_destroy(sit_entry_set_slab); 4885 kmem_cache_destroy(discard_cmd_slab); 4886 kmem_cache_destroy(discard_entry_slab); 4887 kmem_cache_destroy(inmem_entry_slab); 4888 } 4889