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 dc->error = blk_status_to_errno(bio->bi_status); 1033 1034 spin_lock_irqsave(&dc->lock, flags); 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->max_requests = UINT_MAX; 1105 dpolicy->io_aware = false; 1106 /* we need to issue all to keep CP_TRIMMED_FLAG */ 1107 dpolicy->granularity = 1; 1108 dpolicy->timeout = true; 1109 } 1110 } 1111 1112 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1113 struct block_device *bdev, block_t lstart, 1114 block_t start, block_t len); 1115 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */ 1116 static int __submit_discard_cmd(struct f2fs_sb_info *sbi, 1117 struct discard_policy *dpolicy, 1118 struct discard_cmd *dc, 1119 unsigned int *issued) 1120 { 1121 struct block_device *bdev = dc->bdev; 1122 struct request_queue *q = bdev_get_queue(bdev); 1123 unsigned int max_discard_blocks = 1124 SECTOR_TO_BLOCK(q->limits.max_discard_sectors); 1125 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1126 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1127 &(dcc->fstrim_list) : &(dcc->wait_list); 1128 int flag = dpolicy->sync ? REQ_SYNC : 0; 1129 block_t lstart, start, len, total_len; 1130 int err = 0; 1131 1132 if (dc->state != D_PREP) 1133 return 0; 1134 1135 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 1136 return 0; 1137 1138 trace_f2fs_issue_discard(bdev, dc->start, dc->len); 1139 1140 lstart = dc->lstart; 1141 start = dc->start; 1142 len = dc->len; 1143 total_len = len; 1144 1145 dc->len = 0; 1146 1147 while (total_len && *issued < dpolicy->max_requests && !err) { 1148 struct bio *bio = NULL; 1149 unsigned long flags; 1150 bool last = true; 1151 1152 if (len > max_discard_blocks) { 1153 len = max_discard_blocks; 1154 last = false; 1155 } 1156 1157 (*issued)++; 1158 if (*issued == dpolicy->max_requests) 1159 last = true; 1160 1161 dc->len += len; 1162 1163 if (time_to_inject(sbi, FAULT_DISCARD)) { 1164 f2fs_show_injection_info(sbi, FAULT_DISCARD); 1165 err = -EIO; 1166 goto submit; 1167 } 1168 err = __blkdev_issue_discard(bdev, 1169 SECTOR_FROM_BLOCK(start), 1170 SECTOR_FROM_BLOCK(len), 1171 GFP_NOFS, 0, &bio); 1172 submit: 1173 if (err) { 1174 spin_lock_irqsave(&dc->lock, flags); 1175 if (dc->state == D_PARTIAL) 1176 dc->state = D_SUBMIT; 1177 spin_unlock_irqrestore(&dc->lock, flags); 1178 1179 break; 1180 } 1181 1182 f2fs_bug_on(sbi, !bio); 1183 1184 /* 1185 * should keep before submission to avoid D_DONE 1186 * right away 1187 */ 1188 spin_lock_irqsave(&dc->lock, flags); 1189 if (last) 1190 dc->state = D_SUBMIT; 1191 else 1192 dc->state = D_PARTIAL; 1193 dc->bio_ref++; 1194 spin_unlock_irqrestore(&dc->lock, flags); 1195 1196 atomic_inc(&dcc->queued_discard); 1197 dc->queued++; 1198 list_move_tail(&dc->list, wait_list); 1199 1200 /* sanity check on discard range */ 1201 __check_sit_bitmap(sbi, lstart, lstart + len); 1202 1203 bio->bi_private = dc; 1204 bio->bi_end_io = f2fs_submit_discard_endio; 1205 bio->bi_opf |= flag; 1206 submit_bio(bio); 1207 1208 atomic_inc(&dcc->issued_discard); 1209 1210 f2fs_update_iostat(sbi, FS_DISCARD, 1); 1211 1212 lstart += len; 1213 start += len; 1214 total_len -= len; 1215 len = total_len; 1216 } 1217 1218 if (!err && len) 1219 __update_discard_tree_range(sbi, bdev, lstart, start, len); 1220 return err; 1221 } 1222 1223 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi, 1224 struct block_device *bdev, block_t lstart, 1225 block_t start, block_t len, 1226 struct rb_node **insert_p, 1227 struct rb_node *insert_parent) 1228 { 1229 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1230 struct rb_node **p; 1231 struct rb_node *parent = NULL; 1232 struct discard_cmd *dc = 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 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, 1245 p, leftmost); 1246 if (!dc) 1247 return NULL; 1248 1249 return dc; 1250 } 1251 1252 static void __relocate_discard_cmd(struct discard_cmd_control *dcc, 1253 struct discard_cmd *dc) 1254 { 1255 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]); 1256 } 1257 1258 static void __punch_discard_cmd(struct f2fs_sb_info *sbi, 1259 struct discard_cmd *dc, block_t blkaddr) 1260 { 1261 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1262 struct discard_info di = dc->di; 1263 bool modified = false; 1264 1265 if (dc->state == D_DONE || dc->len == 1) { 1266 __remove_discard_cmd(sbi, dc); 1267 return; 1268 } 1269 1270 dcc->undiscard_blks -= di.len; 1271 1272 if (blkaddr > di.lstart) { 1273 dc->len = blkaddr - dc->lstart; 1274 dcc->undiscard_blks += dc->len; 1275 __relocate_discard_cmd(dcc, dc); 1276 modified = true; 1277 } 1278 1279 if (blkaddr < di.lstart + di.len - 1) { 1280 if (modified) { 1281 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1, 1282 di.start + blkaddr + 1 - di.lstart, 1283 di.lstart + di.len - 1 - blkaddr, 1284 NULL, NULL); 1285 } else { 1286 dc->lstart++; 1287 dc->len--; 1288 dc->start++; 1289 dcc->undiscard_blks += dc->len; 1290 __relocate_discard_cmd(dcc, dc); 1291 } 1292 } 1293 } 1294 1295 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1296 struct block_device *bdev, block_t lstart, 1297 block_t start, block_t len) 1298 { 1299 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1300 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1301 struct discard_cmd *dc; 1302 struct discard_info di = {0}; 1303 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1304 struct request_queue *q = bdev_get_queue(bdev); 1305 unsigned int max_discard_blocks = 1306 SECTOR_TO_BLOCK(q->limits.max_discard_sectors); 1307 block_t end = lstart + len; 1308 1309 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root, 1310 NULL, lstart, 1311 (struct rb_entry **)&prev_dc, 1312 (struct rb_entry **)&next_dc, 1313 &insert_p, &insert_parent, true, NULL); 1314 if (dc) 1315 prev_dc = dc; 1316 1317 if (!prev_dc) { 1318 di.lstart = lstart; 1319 di.len = next_dc ? next_dc->lstart - lstart : len; 1320 di.len = min(di.len, len); 1321 di.start = start; 1322 } 1323 1324 while (1) { 1325 struct rb_node *node; 1326 bool merged = false; 1327 struct discard_cmd *tdc = NULL; 1328 1329 if (prev_dc) { 1330 di.lstart = prev_dc->lstart + prev_dc->len; 1331 if (di.lstart < lstart) 1332 di.lstart = lstart; 1333 if (di.lstart >= end) 1334 break; 1335 1336 if (!next_dc || next_dc->lstart > end) 1337 di.len = end - di.lstart; 1338 else 1339 di.len = next_dc->lstart - di.lstart; 1340 di.start = start + di.lstart - lstart; 1341 } 1342 1343 if (!di.len) 1344 goto next; 1345 1346 if (prev_dc && prev_dc->state == D_PREP && 1347 prev_dc->bdev == bdev && 1348 __is_discard_back_mergeable(&di, &prev_dc->di, 1349 max_discard_blocks)) { 1350 prev_dc->di.len += di.len; 1351 dcc->undiscard_blks += di.len; 1352 __relocate_discard_cmd(dcc, prev_dc); 1353 di = prev_dc->di; 1354 tdc = prev_dc; 1355 merged = true; 1356 } 1357 1358 if (next_dc && next_dc->state == D_PREP && 1359 next_dc->bdev == bdev && 1360 __is_discard_front_mergeable(&di, &next_dc->di, 1361 max_discard_blocks)) { 1362 next_dc->di.lstart = di.lstart; 1363 next_dc->di.len += di.len; 1364 next_dc->di.start = di.start; 1365 dcc->undiscard_blks += di.len; 1366 __relocate_discard_cmd(dcc, next_dc); 1367 if (tdc) 1368 __remove_discard_cmd(sbi, tdc); 1369 merged = true; 1370 } 1371 1372 if (!merged) { 1373 __insert_discard_tree(sbi, bdev, di.lstart, di.start, 1374 di.len, NULL, NULL); 1375 } 1376 next: 1377 prev_dc = next_dc; 1378 if (!prev_dc) 1379 break; 1380 1381 node = rb_next(&prev_dc->rb_node); 1382 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1383 } 1384 } 1385 1386 static int __queue_discard_cmd(struct f2fs_sb_info *sbi, 1387 struct block_device *bdev, block_t blkstart, block_t blklen) 1388 { 1389 block_t lblkstart = blkstart; 1390 1391 if (!f2fs_bdev_support_discard(bdev)) 1392 return 0; 1393 1394 trace_f2fs_queue_discard(bdev, blkstart, blklen); 1395 1396 if (f2fs_is_multi_device(sbi)) { 1397 int devi = f2fs_target_device_index(sbi, blkstart); 1398 1399 blkstart -= FDEV(devi).start_blk; 1400 } 1401 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock); 1402 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen); 1403 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock); 1404 return 0; 1405 } 1406 1407 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi, 1408 struct discard_policy *dpolicy) 1409 { 1410 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1411 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1412 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1413 struct discard_cmd *dc; 1414 struct blk_plug plug; 1415 unsigned int pos = dcc->next_pos; 1416 unsigned int issued = 0; 1417 bool io_interrupted = false; 1418 1419 mutex_lock(&dcc->cmd_lock); 1420 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root, 1421 NULL, pos, 1422 (struct rb_entry **)&prev_dc, 1423 (struct rb_entry **)&next_dc, 1424 &insert_p, &insert_parent, true, NULL); 1425 if (!dc) 1426 dc = next_dc; 1427 1428 blk_start_plug(&plug); 1429 1430 while (dc) { 1431 struct rb_node *node; 1432 int err = 0; 1433 1434 if (dc->state != D_PREP) 1435 goto next; 1436 1437 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) { 1438 io_interrupted = true; 1439 break; 1440 } 1441 1442 dcc->next_pos = dc->lstart + dc->len; 1443 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued); 1444 1445 if (issued >= dpolicy->max_requests) 1446 break; 1447 next: 1448 node = rb_next(&dc->rb_node); 1449 if (err) 1450 __remove_discard_cmd(sbi, dc); 1451 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1452 } 1453 1454 blk_finish_plug(&plug); 1455 1456 if (!dc) 1457 dcc->next_pos = 0; 1458 1459 mutex_unlock(&dcc->cmd_lock); 1460 1461 if (!issued && io_interrupted) 1462 issued = -1; 1463 1464 return issued; 1465 } 1466 1467 static int __issue_discard_cmd(struct f2fs_sb_info *sbi, 1468 struct discard_policy *dpolicy) 1469 { 1470 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1471 struct list_head *pend_list; 1472 struct discard_cmd *dc, *tmp; 1473 struct blk_plug plug; 1474 int i, issued = 0; 1475 bool io_interrupted = false; 1476 1477 if (dpolicy->timeout) 1478 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT); 1479 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 (!issued && io_interrupted) 1527 issued = -1; 1528 1529 return issued; 1530 } 1531 1532 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi) 1533 { 1534 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1535 struct list_head *pend_list; 1536 struct discard_cmd *dc, *tmp; 1537 int i; 1538 bool dropped = false; 1539 1540 mutex_lock(&dcc->cmd_lock); 1541 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1542 pend_list = &dcc->pend_list[i]; 1543 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1544 f2fs_bug_on(sbi, dc->state != D_PREP); 1545 __remove_discard_cmd(sbi, dc); 1546 dropped = true; 1547 } 1548 } 1549 mutex_unlock(&dcc->cmd_lock); 1550 1551 return dropped; 1552 } 1553 1554 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi) 1555 { 1556 __drop_discard_cmd(sbi); 1557 } 1558 1559 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi, 1560 struct discard_cmd *dc) 1561 { 1562 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1563 unsigned int len = 0; 1564 1565 wait_for_completion_io(&dc->wait); 1566 mutex_lock(&dcc->cmd_lock); 1567 f2fs_bug_on(sbi, dc->state != D_DONE); 1568 dc->ref--; 1569 if (!dc->ref) { 1570 if (!dc->error) 1571 len = dc->len; 1572 __remove_discard_cmd(sbi, dc); 1573 } 1574 mutex_unlock(&dcc->cmd_lock); 1575 1576 return len; 1577 } 1578 1579 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi, 1580 struct discard_policy *dpolicy, 1581 block_t start, block_t end) 1582 { 1583 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1584 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1585 &(dcc->fstrim_list) : &(dcc->wait_list); 1586 struct discard_cmd *dc, *tmp; 1587 bool need_wait; 1588 unsigned int trimmed = 0; 1589 1590 next: 1591 need_wait = false; 1592 1593 mutex_lock(&dcc->cmd_lock); 1594 list_for_each_entry_safe(dc, tmp, wait_list, list) { 1595 if (dc->lstart + dc->len <= start || end <= dc->lstart) 1596 continue; 1597 if (dc->len < dpolicy->granularity) 1598 continue; 1599 if (dc->state == D_DONE && !dc->ref) { 1600 wait_for_completion_io(&dc->wait); 1601 if (!dc->error) 1602 trimmed += dc->len; 1603 __remove_discard_cmd(sbi, dc); 1604 } else { 1605 dc->ref++; 1606 need_wait = true; 1607 break; 1608 } 1609 } 1610 mutex_unlock(&dcc->cmd_lock); 1611 1612 if (need_wait) { 1613 trimmed += __wait_one_discard_bio(sbi, dc); 1614 goto next; 1615 } 1616 1617 return trimmed; 1618 } 1619 1620 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1621 struct discard_policy *dpolicy) 1622 { 1623 struct discard_policy dp; 1624 unsigned int discard_blks; 1625 1626 if (dpolicy) 1627 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX); 1628 1629 /* wait all */ 1630 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1); 1631 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1632 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1); 1633 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1634 1635 return discard_blks; 1636 } 1637 1638 /* This should be covered by global mutex, &sit_i->sentry_lock */ 1639 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr) 1640 { 1641 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1642 struct discard_cmd *dc; 1643 bool need_wait = false; 1644 1645 mutex_lock(&dcc->cmd_lock); 1646 dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root, 1647 NULL, blkaddr); 1648 if (dc) { 1649 if (dc->state == D_PREP) { 1650 __punch_discard_cmd(sbi, dc, blkaddr); 1651 } else { 1652 dc->ref++; 1653 need_wait = true; 1654 } 1655 } 1656 mutex_unlock(&dcc->cmd_lock); 1657 1658 if (need_wait) 1659 __wait_one_discard_bio(sbi, dc); 1660 } 1661 1662 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi) 1663 { 1664 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1665 1666 if (dcc && dcc->f2fs_issue_discard) { 1667 struct task_struct *discard_thread = dcc->f2fs_issue_discard; 1668 1669 dcc->f2fs_issue_discard = NULL; 1670 kthread_stop(discard_thread); 1671 } 1672 } 1673 1674 /* This comes from f2fs_put_super */ 1675 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi) 1676 { 1677 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1678 struct discard_policy dpolicy; 1679 bool dropped; 1680 1681 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT, 1682 dcc->discard_granularity); 1683 __issue_discard_cmd(sbi, &dpolicy); 1684 dropped = __drop_discard_cmd(sbi); 1685 1686 /* just to make sure there is no pending discard commands */ 1687 __wait_all_discard_cmd(sbi, NULL); 1688 1689 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt)); 1690 return dropped; 1691 } 1692 1693 static int issue_discard_thread(void *data) 1694 { 1695 struct f2fs_sb_info *sbi = data; 1696 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1697 wait_queue_head_t *q = &dcc->discard_wait_queue; 1698 struct discard_policy dpolicy; 1699 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME; 1700 int issued; 1701 1702 set_freezable(); 1703 1704 do { 1705 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG, 1706 dcc->discard_granularity); 1707 1708 wait_event_interruptible_timeout(*q, 1709 kthread_should_stop() || freezing(current) || 1710 dcc->discard_wake, 1711 msecs_to_jiffies(wait_ms)); 1712 1713 if (dcc->discard_wake) 1714 dcc->discard_wake = 0; 1715 1716 /* clean up pending candidates before going to sleep */ 1717 if (atomic_read(&dcc->queued_discard)) 1718 __wait_all_discard_cmd(sbi, NULL); 1719 1720 if (try_to_freeze()) 1721 continue; 1722 if (f2fs_readonly(sbi->sb)) 1723 continue; 1724 if (kthread_should_stop()) 1725 return 0; 1726 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 1727 wait_ms = dpolicy.max_interval; 1728 continue; 1729 } 1730 1731 if (sbi->gc_mode == GC_URGENT) 1732 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1); 1733 1734 sb_start_intwrite(sbi->sb); 1735 1736 issued = __issue_discard_cmd(sbi, &dpolicy); 1737 if (issued > 0) { 1738 __wait_all_discard_cmd(sbi, &dpolicy); 1739 wait_ms = dpolicy.min_interval; 1740 } else if (issued == -1){ 1741 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME); 1742 if (!wait_ms) 1743 wait_ms = dpolicy.mid_interval; 1744 } else { 1745 wait_ms = dpolicy.max_interval; 1746 } 1747 1748 sb_end_intwrite(sbi->sb); 1749 1750 } while (!kthread_should_stop()); 1751 return 0; 1752 } 1753 1754 #ifdef CONFIG_BLK_DEV_ZONED 1755 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi, 1756 struct block_device *bdev, block_t blkstart, block_t blklen) 1757 { 1758 sector_t sector, nr_sects; 1759 block_t lblkstart = blkstart; 1760 int devi = 0; 1761 1762 if (f2fs_is_multi_device(sbi)) { 1763 devi = f2fs_target_device_index(sbi, blkstart); 1764 if (blkstart < FDEV(devi).start_blk || 1765 blkstart > FDEV(devi).end_blk) { 1766 f2fs_err(sbi, "Invalid block %x", blkstart); 1767 return -EIO; 1768 } 1769 blkstart -= FDEV(devi).start_blk; 1770 } 1771 1772 /* For sequential zones, reset the zone write pointer */ 1773 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) { 1774 sector = SECTOR_FROM_BLOCK(blkstart); 1775 nr_sects = SECTOR_FROM_BLOCK(blklen); 1776 1777 if (sector & (bdev_zone_sectors(bdev) - 1) || 1778 nr_sects != bdev_zone_sectors(bdev)) { 1779 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)", 1780 devi, sbi->s_ndevs ? FDEV(devi).path : "", 1781 blkstart, blklen); 1782 return -EIO; 1783 } 1784 trace_f2fs_issue_reset_zone(bdev, blkstart); 1785 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, 1786 sector, nr_sects, GFP_NOFS); 1787 } 1788 1789 /* For conventional zones, use regular discard if supported */ 1790 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen); 1791 } 1792 #endif 1793 1794 static int __issue_discard_async(struct f2fs_sb_info *sbi, 1795 struct block_device *bdev, block_t blkstart, block_t blklen) 1796 { 1797 #ifdef CONFIG_BLK_DEV_ZONED 1798 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) 1799 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen); 1800 #endif 1801 return __queue_discard_cmd(sbi, bdev, blkstart, blklen); 1802 } 1803 1804 static int f2fs_issue_discard(struct f2fs_sb_info *sbi, 1805 block_t blkstart, block_t blklen) 1806 { 1807 sector_t start = blkstart, len = 0; 1808 struct block_device *bdev; 1809 struct seg_entry *se; 1810 unsigned int offset; 1811 block_t i; 1812 int err = 0; 1813 1814 bdev = f2fs_target_device(sbi, blkstart, NULL); 1815 1816 for (i = blkstart; i < blkstart + blklen; i++, len++) { 1817 if (i != start) { 1818 struct block_device *bdev2 = 1819 f2fs_target_device(sbi, i, NULL); 1820 1821 if (bdev2 != bdev) { 1822 err = __issue_discard_async(sbi, bdev, 1823 start, len); 1824 if (err) 1825 return err; 1826 bdev = bdev2; 1827 start = i; 1828 len = 0; 1829 } 1830 } 1831 1832 se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); 1833 offset = GET_BLKOFF_FROM_SEG0(sbi, i); 1834 1835 if (!f2fs_test_and_set_bit(offset, se->discard_map)) 1836 sbi->discard_blks--; 1837 } 1838 1839 if (len) 1840 err = __issue_discard_async(sbi, bdev, start, len); 1841 return err; 1842 } 1843 1844 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc, 1845 bool check_only) 1846 { 1847 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 1848 int max_blocks = sbi->blocks_per_seg; 1849 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); 1850 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 1851 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 1852 unsigned long *discard_map = (unsigned long *)se->discard_map; 1853 unsigned long *dmap = SIT_I(sbi)->tmp_map; 1854 unsigned int start = 0, end = -1; 1855 bool force = (cpc->reason & CP_DISCARD); 1856 struct discard_entry *de = NULL; 1857 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list; 1858 int i; 1859 1860 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi)) 1861 return false; 1862 1863 if (!force) { 1864 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks || 1865 SM_I(sbi)->dcc_info->nr_discards >= 1866 SM_I(sbi)->dcc_info->max_discards) 1867 return false; 1868 } 1869 1870 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ 1871 for (i = 0; i < entries; i++) 1872 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : 1873 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; 1874 1875 while (force || SM_I(sbi)->dcc_info->nr_discards <= 1876 SM_I(sbi)->dcc_info->max_discards) { 1877 start = __find_rev_next_bit(dmap, max_blocks, end + 1); 1878 if (start >= max_blocks) 1879 break; 1880 1881 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); 1882 if (force && start && end != max_blocks 1883 && (end - start) < cpc->trim_minlen) 1884 continue; 1885 1886 if (check_only) 1887 return true; 1888 1889 if (!de) { 1890 de = f2fs_kmem_cache_alloc(discard_entry_slab, 1891 GFP_F2FS_ZERO); 1892 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start); 1893 list_add_tail(&de->list, head); 1894 } 1895 1896 for (i = start; i < end; i++) 1897 __set_bit_le(i, (void *)de->discard_map); 1898 1899 SM_I(sbi)->dcc_info->nr_discards += end - start; 1900 } 1901 return false; 1902 } 1903 1904 static void release_discard_addr(struct discard_entry *entry) 1905 { 1906 list_del(&entry->list); 1907 kmem_cache_free(discard_entry_slab, entry); 1908 } 1909 1910 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi) 1911 { 1912 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); 1913 struct discard_entry *entry, *this; 1914 1915 /* drop caches */ 1916 list_for_each_entry_safe(entry, this, head, list) 1917 release_discard_addr(entry); 1918 } 1919 1920 /* 1921 * Should call f2fs_clear_prefree_segments after checkpoint is done. 1922 */ 1923 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 1924 { 1925 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1926 unsigned int segno; 1927 1928 mutex_lock(&dirty_i->seglist_lock); 1929 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) 1930 __set_test_and_free(sbi, segno); 1931 mutex_unlock(&dirty_i->seglist_lock); 1932 } 1933 1934 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi, 1935 struct cp_control *cpc) 1936 { 1937 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1938 struct list_head *head = &dcc->entry_list; 1939 struct discard_entry *entry, *this; 1940 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1941 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; 1942 unsigned int start = 0, end = -1; 1943 unsigned int secno, start_segno; 1944 bool force = (cpc->reason & CP_DISCARD); 1945 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi); 1946 1947 mutex_lock(&dirty_i->seglist_lock); 1948 1949 while (1) { 1950 int i; 1951 1952 if (need_align && end != -1) 1953 end--; 1954 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); 1955 if (start >= MAIN_SEGS(sbi)) 1956 break; 1957 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), 1958 start + 1); 1959 1960 if (need_align) { 1961 start = rounddown(start, sbi->segs_per_sec); 1962 end = roundup(end, sbi->segs_per_sec); 1963 } 1964 1965 for (i = start; i < end; i++) { 1966 if (test_and_clear_bit(i, prefree_map)) 1967 dirty_i->nr_dirty[PRE]--; 1968 } 1969 1970 if (!f2fs_realtime_discard_enable(sbi)) 1971 continue; 1972 1973 if (force && start >= cpc->trim_start && 1974 (end - 1) <= cpc->trim_end) 1975 continue; 1976 1977 if (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi)) { 1978 f2fs_issue_discard(sbi, START_BLOCK(sbi, start), 1979 (end - start) << sbi->log_blocks_per_seg); 1980 continue; 1981 } 1982 next: 1983 secno = GET_SEC_FROM_SEG(sbi, start); 1984 start_segno = GET_SEG_FROM_SEC(sbi, secno); 1985 if (!IS_CURSEC(sbi, secno) && 1986 !get_valid_blocks(sbi, start, true)) 1987 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno), 1988 sbi->segs_per_sec << sbi->log_blocks_per_seg); 1989 1990 start = start_segno + sbi->segs_per_sec; 1991 if (start < end) 1992 goto next; 1993 else 1994 end = start - 1; 1995 } 1996 mutex_unlock(&dirty_i->seglist_lock); 1997 1998 /* send small discards */ 1999 list_for_each_entry_safe(entry, this, head, list) { 2000 unsigned int cur_pos = 0, next_pos, len, total_len = 0; 2001 bool is_valid = test_bit_le(0, entry->discard_map); 2002 2003 find_next: 2004 if (is_valid) { 2005 next_pos = find_next_zero_bit_le(entry->discard_map, 2006 sbi->blocks_per_seg, cur_pos); 2007 len = next_pos - cur_pos; 2008 2009 if (f2fs_sb_has_blkzoned(sbi) || 2010 (force && len < cpc->trim_minlen)) 2011 goto skip; 2012 2013 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos, 2014 len); 2015 total_len += len; 2016 } else { 2017 next_pos = find_next_bit_le(entry->discard_map, 2018 sbi->blocks_per_seg, cur_pos); 2019 } 2020 skip: 2021 cur_pos = next_pos; 2022 is_valid = !is_valid; 2023 2024 if (cur_pos < sbi->blocks_per_seg) 2025 goto find_next; 2026 2027 release_discard_addr(entry); 2028 dcc->nr_discards -= total_len; 2029 } 2030 2031 wake_up_discard_thread(sbi, false); 2032 } 2033 2034 static int create_discard_cmd_control(struct f2fs_sb_info *sbi) 2035 { 2036 dev_t dev = sbi->sb->s_bdev->bd_dev; 2037 struct discard_cmd_control *dcc; 2038 int err = 0, i; 2039 2040 if (SM_I(sbi)->dcc_info) { 2041 dcc = SM_I(sbi)->dcc_info; 2042 goto init_thread; 2043 } 2044 2045 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL); 2046 if (!dcc) 2047 return -ENOMEM; 2048 2049 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY; 2050 INIT_LIST_HEAD(&dcc->entry_list); 2051 for (i = 0; i < MAX_PLIST_NUM; i++) 2052 INIT_LIST_HEAD(&dcc->pend_list[i]); 2053 INIT_LIST_HEAD(&dcc->wait_list); 2054 INIT_LIST_HEAD(&dcc->fstrim_list); 2055 mutex_init(&dcc->cmd_lock); 2056 atomic_set(&dcc->issued_discard, 0); 2057 atomic_set(&dcc->queued_discard, 0); 2058 atomic_set(&dcc->discard_cmd_cnt, 0); 2059 dcc->nr_discards = 0; 2060 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg; 2061 dcc->undiscard_blks = 0; 2062 dcc->next_pos = 0; 2063 dcc->root = RB_ROOT_CACHED; 2064 dcc->rbtree_check = false; 2065 2066 init_waitqueue_head(&dcc->discard_wait_queue); 2067 SM_I(sbi)->dcc_info = dcc; 2068 init_thread: 2069 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi, 2070 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev)); 2071 if (IS_ERR(dcc->f2fs_issue_discard)) { 2072 err = PTR_ERR(dcc->f2fs_issue_discard); 2073 kvfree(dcc); 2074 SM_I(sbi)->dcc_info = NULL; 2075 return err; 2076 } 2077 2078 return err; 2079 } 2080 2081 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi) 2082 { 2083 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2084 2085 if (!dcc) 2086 return; 2087 2088 f2fs_stop_discard_thread(sbi); 2089 2090 /* 2091 * Recovery can cache discard commands, so in error path of 2092 * fill_super(), it needs to give a chance to handle them. 2093 */ 2094 if (unlikely(atomic_read(&dcc->discard_cmd_cnt))) 2095 f2fs_issue_discard_timeout(sbi); 2096 2097 kvfree(dcc); 2098 SM_I(sbi)->dcc_info = NULL; 2099 } 2100 2101 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 2102 { 2103 struct sit_info *sit_i = SIT_I(sbi); 2104 2105 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { 2106 sit_i->dirty_sentries++; 2107 return false; 2108 } 2109 2110 return true; 2111 } 2112 2113 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 2114 unsigned int segno, int modified) 2115 { 2116 struct seg_entry *se = get_seg_entry(sbi, segno); 2117 se->type = type; 2118 if (modified) 2119 __mark_sit_entry_dirty(sbi, segno); 2120 } 2121 2122 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 2123 { 2124 struct seg_entry *se; 2125 unsigned int segno, offset; 2126 long int new_vblocks; 2127 bool exist; 2128 #ifdef CONFIG_F2FS_CHECK_FS 2129 bool mir_exist; 2130 #endif 2131 2132 segno = GET_SEGNO(sbi, blkaddr); 2133 2134 se = get_seg_entry(sbi, segno); 2135 new_vblocks = se->valid_blocks + del; 2136 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2137 2138 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) || 2139 (new_vblocks > sbi->blocks_per_seg))); 2140 2141 se->valid_blocks = new_vblocks; 2142 se->mtime = get_mtime(sbi, false); 2143 if (se->mtime > SIT_I(sbi)->max_mtime) 2144 SIT_I(sbi)->max_mtime = se->mtime; 2145 2146 /* Update valid block bitmap */ 2147 if (del > 0) { 2148 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map); 2149 #ifdef CONFIG_F2FS_CHECK_FS 2150 mir_exist = f2fs_test_and_set_bit(offset, 2151 se->cur_valid_map_mir); 2152 if (unlikely(exist != mir_exist)) { 2153 f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d", 2154 blkaddr, exist); 2155 f2fs_bug_on(sbi, 1); 2156 } 2157 #endif 2158 if (unlikely(exist)) { 2159 f2fs_err(sbi, "Bitmap was wrongly set, blk:%u", 2160 blkaddr); 2161 f2fs_bug_on(sbi, 1); 2162 se->valid_blocks--; 2163 del = 0; 2164 } 2165 2166 if (!f2fs_test_and_set_bit(offset, se->discard_map)) 2167 sbi->discard_blks--; 2168 2169 /* 2170 * SSR should never reuse block which is checkpointed 2171 * or newly invalidated. 2172 */ 2173 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) { 2174 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map)) 2175 se->ckpt_valid_blocks++; 2176 } 2177 } else { 2178 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map); 2179 #ifdef CONFIG_F2FS_CHECK_FS 2180 mir_exist = f2fs_test_and_clear_bit(offset, 2181 se->cur_valid_map_mir); 2182 if (unlikely(exist != mir_exist)) { 2183 f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d", 2184 blkaddr, exist); 2185 f2fs_bug_on(sbi, 1); 2186 } 2187 #endif 2188 if (unlikely(!exist)) { 2189 f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u", 2190 blkaddr); 2191 f2fs_bug_on(sbi, 1); 2192 se->valid_blocks++; 2193 del = 0; 2194 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 2195 /* 2196 * If checkpoints are off, we must not reuse data that 2197 * was used in the previous checkpoint. If it was used 2198 * before, we must track that to know how much space we 2199 * really have. 2200 */ 2201 if (f2fs_test_bit(offset, se->ckpt_valid_map)) { 2202 spin_lock(&sbi->stat_lock); 2203 sbi->unusable_block_count++; 2204 spin_unlock(&sbi->stat_lock); 2205 } 2206 } 2207 2208 if (f2fs_test_and_clear_bit(offset, se->discard_map)) 2209 sbi->discard_blks++; 2210 } 2211 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) 2212 se->ckpt_valid_blocks += del; 2213 2214 __mark_sit_entry_dirty(sbi, segno); 2215 2216 /* update total number of valid blocks to be written in ckpt area */ 2217 SIT_I(sbi)->written_valid_blocks += del; 2218 2219 if (__is_large_section(sbi)) 2220 get_sec_entry(sbi, segno)->valid_blocks += del; 2221 } 2222 2223 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) 2224 { 2225 unsigned int segno = GET_SEGNO(sbi, addr); 2226 struct sit_info *sit_i = SIT_I(sbi); 2227 2228 f2fs_bug_on(sbi, addr == NULL_ADDR); 2229 if (addr == NEW_ADDR || addr == COMPRESS_ADDR) 2230 return; 2231 2232 invalidate_mapping_pages(META_MAPPING(sbi), addr, addr); 2233 2234 /* add it into sit main buffer */ 2235 down_write(&sit_i->sentry_lock); 2236 2237 update_sit_entry(sbi, addr, -1); 2238 2239 /* add it into dirty seglist */ 2240 locate_dirty_segment(sbi, segno); 2241 2242 up_write(&sit_i->sentry_lock); 2243 } 2244 2245 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr) 2246 { 2247 struct sit_info *sit_i = SIT_I(sbi); 2248 unsigned int segno, offset; 2249 struct seg_entry *se; 2250 bool is_cp = false; 2251 2252 if (!__is_valid_data_blkaddr(blkaddr)) 2253 return true; 2254 2255 down_read(&sit_i->sentry_lock); 2256 2257 segno = GET_SEGNO(sbi, blkaddr); 2258 se = get_seg_entry(sbi, segno); 2259 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2260 2261 if (f2fs_test_bit(offset, se->ckpt_valid_map)) 2262 is_cp = true; 2263 2264 up_read(&sit_i->sentry_lock); 2265 2266 return is_cp; 2267 } 2268 2269 /* 2270 * This function should be resided under the curseg_mutex lock 2271 */ 2272 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, 2273 struct f2fs_summary *sum) 2274 { 2275 struct curseg_info *curseg = CURSEG_I(sbi, type); 2276 void *addr = curseg->sum_blk; 2277 addr += curseg->next_blkoff * sizeof(struct f2fs_summary); 2278 memcpy(addr, sum, sizeof(struct f2fs_summary)); 2279 } 2280 2281 /* 2282 * Calculate the number of current summary pages for writing 2283 */ 2284 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) 2285 { 2286 int valid_sum_count = 0; 2287 int i, sum_in_page; 2288 2289 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2290 if (sbi->ckpt->alloc_type[i] == SSR) 2291 valid_sum_count += sbi->blocks_per_seg; 2292 else { 2293 if (for_ra) 2294 valid_sum_count += le16_to_cpu( 2295 F2FS_CKPT(sbi)->cur_data_blkoff[i]); 2296 else 2297 valid_sum_count += curseg_blkoff(sbi, i); 2298 } 2299 } 2300 2301 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE - 2302 SUM_FOOTER_SIZE) / SUMMARY_SIZE; 2303 if (valid_sum_count <= sum_in_page) 2304 return 1; 2305 else if ((valid_sum_count - sum_in_page) <= 2306 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) 2307 return 2; 2308 return 3; 2309 } 2310 2311 /* 2312 * Caller should put this summary page 2313 */ 2314 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) 2315 { 2316 return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno)); 2317 } 2318 2319 void f2fs_update_meta_page(struct f2fs_sb_info *sbi, 2320 void *src, block_t blk_addr) 2321 { 2322 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2323 2324 memcpy(page_address(page), src, PAGE_SIZE); 2325 set_page_dirty(page); 2326 f2fs_put_page(page, 1); 2327 } 2328 2329 static void write_sum_page(struct f2fs_sb_info *sbi, 2330 struct f2fs_summary_block *sum_blk, block_t blk_addr) 2331 { 2332 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr); 2333 } 2334 2335 static void write_current_sum_page(struct f2fs_sb_info *sbi, 2336 int type, block_t blk_addr) 2337 { 2338 struct curseg_info *curseg = CURSEG_I(sbi, type); 2339 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2340 struct f2fs_summary_block *src = curseg->sum_blk; 2341 struct f2fs_summary_block *dst; 2342 2343 dst = (struct f2fs_summary_block *)page_address(page); 2344 memset(dst, 0, PAGE_SIZE); 2345 2346 mutex_lock(&curseg->curseg_mutex); 2347 2348 down_read(&curseg->journal_rwsem); 2349 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE); 2350 up_read(&curseg->journal_rwsem); 2351 2352 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE); 2353 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE); 2354 2355 mutex_unlock(&curseg->curseg_mutex); 2356 2357 set_page_dirty(page); 2358 f2fs_put_page(page, 1); 2359 } 2360 2361 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type) 2362 { 2363 struct curseg_info *curseg = CURSEG_I(sbi, type); 2364 unsigned int segno = curseg->segno + 1; 2365 struct free_segmap_info *free_i = FREE_I(sbi); 2366 2367 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec) 2368 return !test_bit(segno, free_i->free_segmap); 2369 return 0; 2370 } 2371 2372 /* 2373 * Find a new segment from the free segments bitmap to right order 2374 * This function should be returned with success, otherwise BUG 2375 */ 2376 static void get_new_segment(struct f2fs_sb_info *sbi, 2377 unsigned int *newseg, bool new_sec, int dir) 2378 { 2379 struct free_segmap_info *free_i = FREE_I(sbi); 2380 unsigned int segno, secno, zoneno; 2381 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; 2382 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg); 2383 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg); 2384 unsigned int left_start = hint; 2385 bool init = true; 2386 int go_left = 0; 2387 int i; 2388 2389 spin_lock(&free_i->segmap_lock); 2390 2391 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { 2392 segno = find_next_zero_bit(free_i->free_segmap, 2393 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1); 2394 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1)) 2395 goto got_it; 2396 } 2397 find_other_zone: 2398 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 2399 if (secno >= MAIN_SECS(sbi)) { 2400 if (dir == ALLOC_RIGHT) { 2401 secno = find_next_zero_bit(free_i->free_secmap, 2402 MAIN_SECS(sbi), 0); 2403 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi)); 2404 } else { 2405 go_left = 1; 2406 left_start = hint - 1; 2407 } 2408 } 2409 if (go_left == 0) 2410 goto skip_left; 2411 2412 while (test_bit(left_start, free_i->free_secmap)) { 2413 if (left_start > 0) { 2414 left_start--; 2415 continue; 2416 } 2417 left_start = find_next_zero_bit(free_i->free_secmap, 2418 MAIN_SECS(sbi), 0); 2419 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi)); 2420 break; 2421 } 2422 secno = left_start; 2423 skip_left: 2424 segno = GET_SEG_FROM_SEC(sbi, secno); 2425 zoneno = GET_ZONE_FROM_SEC(sbi, secno); 2426 2427 /* give up on finding another zone */ 2428 if (!init) 2429 goto got_it; 2430 if (sbi->secs_per_zone == 1) 2431 goto got_it; 2432 if (zoneno == old_zoneno) 2433 goto got_it; 2434 if (dir == ALLOC_LEFT) { 2435 if (!go_left && zoneno + 1 >= total_zones) 2436 goto got_it; 2437 if (go_left && zoneno == 0) 2438 goto got_it; 2439 } 2440 for (i = 0; i < NR_CURSEG_TYPE; i++) 2441 if (CURSEG_I(sbi, i)->zone == zoneno) 2442 break; 2443 2444 if (i < NR_CURSEG_TYPE) { 2445 /* zone is in user, try another */ 2446 if (go_left) 2447 hint = zoneno * sbi->secs_per_zone - 1; 2448 else if (zoneno + 1 >= total_zones) 2449 hint = 0; 2450 else 2451 hint = (zoneno + 1) * sbi->secs_per_zone; 2452 init = false; 2453 goto find_other_zone; 2454 } 2455 got_it: 2456 /* set it as dirty segment in free segmap */ 2457 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); 2458 __set_inuse(sbi, segno); 2459 *newseg = segno; 2460 spin_unlock(&free_i->segmap_lock); 2461 } 2462 2463 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 2464 { 2465 struct curseg_info *curseg = CURSEG_I(sbi, type); 2466 struct summary_footer *sum_footer; 2467 2468 curseg->segno = curseg->next_segno; 2469 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno); 2470 curseg->next_blkoff = 0; 2471 curseg->next_segno = NULL_SEGNO; 2472 2473 sum_footer = &(curseg->sum_blk->footer); 2474 memset(sum_footer, 0, sizeof(struct summary_footer)); 2475 if (IS_DATASEG(type)) 2476 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 2477 if (IS_NODESEG(type)) 2478 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 2479 __set_sit_entry_type(sbi, type, curseg->segno, modified); 2480 } 2481 2482 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type) 2483 { 2484 /* if segs_per_sec is large than 1, we need to keep original policy. */ 2485 if (__is_large_section(sbi)) 2486 return CURSEG_I(sbi, type)->segno; 2487 2488 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 2489 return 0; 2490 2491 if (test_opt(sbi, NOHEAP) && 2492 (type == CURSEG_HOT_DATA || IS_NODESEG(type))) 2493 return 0; 2494 2495 if (SIT_I(sbi)->last_victim[ALLOC_NEXT]) 2496 return SIT_I(sbi)->last_victim[ALLOC_NEXT]; 2497 2498 /* find segments from 0 to reuse freed segments */ 2499 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE) 2500 return 0; 2501 2502 return CURSEG_I(sbi, type)->segno; 2503 } 2504 2505 /* 2506 * Allocate a current working segment. 2507 * This function always allocates a free segment in LFS manner. 2508 */ 2509 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 2510 { 2511 struct curseg_info *curseg = CURSEG_I(sbi, type); 2512 unsigned int segno = curseg->segno; 2513 int dir = ALLOC_LEFT; 2514 2515 write_sum_page(sbi, curseg->sum_blk, 2516 GET_SUM_BLOCK(sbi, segno)); 2517 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA) 2518 dir = ALLOC_RIGHT; 2519 2520 if (test_opt(sbi, NOHEAP)) 2521 dir = ALLOC_RIGHT; 2522 2523 segno = __get_next_segno(sbi, type); 2524 get_new_segment(sbi, &segno, new_sec, dir); 2525 curseg->next_segno = segno; 2526 reset_curseg(sbi, type, 1); 2527 curseg->alloc_type = LFS; 2528 } 2529 2530 static void __next_free_blkoff(struct f2fs_sb_info *sbi, 2531 struct curseg_info *seg, block_t start) 2532 { 2533 struct seg_entry *se = get_seg_entry(sbi, seg->segno); 2534 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 2535 unsigned long *target_map = SIT_I(sbi)->tmp_map; 2536 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 2537 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 2538 int i, pos; 2539 2540 for (i = 0; i < entries; i++) 2541 target_map[i] = ckpt_map[i] | cur_map[i]; 2542 2543 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); 2544 2545 seg->next_blkoff = pos; 2546 } 2547 2548 /* 2549 * If a segment is written by LFS manner, next block offset is just obtained 2550 * by increasing the current block offset. However, if a segment is written by 2551 * SSR manner, next block offset obtained by calling __next_free_blkoff 2552 */ 2553 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, 2554 struct curseg_info *seg) 2555 { 2556 if (seg->alloc_type == SSR) 2557 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1); 2558 else 2559 seg->next_blkoff++; 2560 } 2561 2562 /* 2563 * This function always allocates a used segment(from dirty seglist) by SSR 2564 * manner, so it should recover the existing segment information of valid blocks 2565 */ 2566 static void change_curseg(struct f2fs_sb_info *sbi, int type) 2567 { 2568 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2569 struct curseg_info *curseg = CURSEG_I(sbi, type); 2570 unsigned int new_segno = curseg->next_segno; 2571 struct f2fs_summary_block *sum_node; 2572 struct page *sum_page; 2573 2574 write_sum_page(sbi, curseg->sum_blk, 2575 GET_SUM_BLOCK(sbi, curseg->segno)); 2576 __set_test_and_inuse(sbi, new_segno); 2577 2578 mutex_lock(&dirty_i->seglist_lock); 2579 __remove_dirty_segment(sbi, new_segno, PRE); 2580 __remove_dirty_segment(sbi, new_segno, DIRTY); 2581 mutex_unlock(&dirty_i->seglist_lock); 2582 2583 reset_curseg(sbi, type, 1); 2584 curseg->alloc_type = SSR; 2585 __next_free_blkoff(sbi, curseg, 0); 2586 2587 sum_page = f2fs_get_sum_page(sbi, new_segno); 2588 f2fs_bug_on(sbi, IS_ERR(sum_page)); 2589 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 2590 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); 2591 f2fs_put_page(sum_page, 1); 2592 } 2593 2594 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type) 2595 { 2596 struct curseg_info *curseg = CURSEG_I(sbi, type); 2597 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; 2598 unsigned segno = NULL_SEGNO; 2599 int i, cnt; 2600 bool reversed = false; 2601 2602 /* f2fs_need_SSR() already forces to do this */ 2603 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) { 2604 curseg->next_segno = segno; 2605 return 1; 2606 } 2607 2608 /* For node segments, let's do SSR more intensively */ 2609 if (IS_NODESEG(type)) { 2610 if (type >= CURSEG_WARM_NODE) { 2611 reversed = true; 2612 i = CURSEG_COLD_NODE; 2613 } else { 2614 i = CURSEG_HOT_NODE; 2615 } 2616 cnt = NR_CURSEG_NODE_TYPE; 2617 } else { 2618 if (type >= CURSEG_WARM_DATA) { 2619 reversed = true; 2620 i = CURSEG_COLD_DATA; 2621 } else { 2622 i = CURSEG_HOT_DATA; 2623 } 2624 cnt = NR_CURSEG_DATA_TYPE; 2625 } 2626 2627 for (; cnt-- > 0; reversed ? i-- : i++) { 2628 if (i == type) 2629 continue; 2630 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) { 2631 curseg->next_segno = segno; 2632 return 1; 2633 } 2634 } 2635 2636 /* find valid_blocks=0 in dirty list */ 2637 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 2638 segno = get_free_segment(sbi); 2639 if (segno != NULL_SEGNO) { 2640 curseg->next_segno = segno; 2641 return 1; 2642 } 2643 } 2644 return 0; 2645 } 2646 2647 /* 2648 * flush out current segment and replace it with new segment 2649 * This function should be returned with success, otherwise BUG 2650 */ 2651 static void allocate_segment_by_default(struct f2fs_sb_info *sbi, 2652 int type, bool force) 2653 { 2654 struct curseg_info *curseg = CURSEG_I(sbi, type); 2655 2656 if (force) 2657 new_curseg(sbi, type, true); 2658 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) && 2659 type == CURSEG_WARM_NODE) 2660 new_curseg(sbi, type, false); 2661 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) && 2662 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 2663 new_curseg(sbi, type, false); 2664 else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type)) 2665 change_curseg(sbi, type); 2666 else 2667 new_curseg(sbi, type, false); 2668 2669 stat_inc_seg_type(sbi, curseg); 2670 } 2671 2672 void allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type, 2673 unsigned int start, unsigned int end) 2674 { 2675 struct curseg_info *curseg = CURSEG_I(sbi, type); 2676 unsigned int segno; 2677 2678 down_read(&SM_I(sbi)->curseg_lock); 2679 mutex_lock(&curseg->curseg_mutex); 2680 down_write(&SIT_I(sbi)->sentry_lock); 2681 2682 segno = CURSEG_I(sbi, type)->segno; 2683 if (segno < start || segno > end) 2684 goto unlock; 2685 2686 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type)) 2687 change_curseg(sbi, type); 2688 else 2689 new_curseg(sbi, type, true); 2690 2691 stat_inc_seg_type(sbi, curseg); 2692 2693 locate_dirty_segment(sbi, segno); 2694 unlock: 2695 up_write(&SIT_I(sbi)->sentry_lock); 2696 2697 if (segno != curseg->segno) 2698 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u", 2699 type, segno, curseg->segno); 2700 2701 mutex_unlock(&curseg->curseg_mutex); 2702 up_read(&SM_I(sbi)->curseg_lock); 2703 } 2704 2705 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi, int type) 2706 { 2707 struct curseg_info *curseg; 2708 unsigned int old_segno; 2709 int i; 2710 2711 down_write(&SIT_I(sbi)->sentry_lock); 2712 2713 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2714 if (type != NO_CHECK_TYPE && i != type) 2715 continue; 2716 2717 curseg = CURSEG_I(sbi, i); 2718 if (type == NO_CHECK_TYPE || curseg->next_blkoff || 2719 get_valid_blocks(sbi, curseg->segno, false) || 2720 get_ckpt_valid_blocks(sbi, curseg->segno)) { 2721 old_segno = curseg->segno; 2722 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true); 2723 locate_dirty_segment(sbi, old_segno); 2724 } 2725 } 2726 2727 up_write(&SIT_I(sbi)->sentry_lock); 2728 } 2729 2730 static const struct segment_allocation default_salloc_ops = { 2731 .allocate_segment = allocate_segment_by_default, 2732 }; 2733 2734 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi, 2735 struct cp_control *cpc) 2736 { 2737 __u64 trim_start = cpc->trim_start; 2738 bool has_candidate = false; 2739 2740 down_write(&SIT_I(sbi)->sentry_lock); 2741 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) { 2742 if (add_discard_addrs(sbi, cpc, true)) { 2743 has_candidate = true; 2744 break; 2745 } 2746 } 2747 up_write(&SIT_I(sbi)->sentry_lock); 2748 2749 cpc->trim_start = trim_start; 2750 return has_candidate; 2751 } 2752 2753 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi, 2754 struct discard_policy *dpolicy, 2755 unsigned int start, unsigned int end) 2756 { 2757 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2758 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 2759 struct rb_node **insert_p = NULL, *insert_parent = NULL; 2760 struct discard_cmd *dc; 2761 struct blk_plug plug; 2762 int issued; 2763 unsigned int trimmed = 0; 2764 2765 next: 2766 issued = 0; 2767 2768 mutex_lock(&dcc->cmd_lock); 2769 if (unlikely(dcc->rbtree_check)) 2770 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi, 2771 &dcc->root)); 2772 2773 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root, 2774 NULL, start, 2775 (struct rb_entry **)&prev_dc, 2776 (struct rb_entry **)&next_dc, 2777 &insert_p, &insert_parent, true, NULL); 2778 if (!dc) 2779 dc = next_dc; 2780 2781 blk_start_plug(&plug); 2782 2783 while (dc && dc->lstart <= end) { 2784 struct rb_node *node; 2785 int err = 0; 2786 2787 if (dc->len < dpolicy->granularity) 2788 goto skip; 2789 2790 if (dc->state != D_PREP) { 2791 list_move_tail(&dc->list, &dcc->fstrim_list); 2792 goto skip; 2793 } 2794 2795 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued); 2796 2797 if (issued >= dpolicy->max_requests) { 2798 start = dc->lstart + dc->len; 2799 2800 if (err) 2801 __remove_discard_cmd(sbi, dc); 2802 2803 blk_finish_plug(&plug); 2804 mutex_unlock(&dcc->cmd_lock); 2805 trimmed += __wait_all_discard_cmd(sbi, NULL); 2806 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT); 2807 goto next; 2808 } 2809 skip: 2810 node = rb_next(&dc->rb_node); 2811 if (err) 2812 __remove_discard_cmd(sbi, dc); 2813 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 2814 2815 if (fatal_signal_pending(current)) 2816 break; 2817 } 2818 2819 blk_finish_plug(&plug); 2820 mutex_unlock(&dcc->cmd_lock); 2821 2822 return trimmed; 2823 } 2824 2825 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) 2826 { 2827 __u64 start = F2FS_BYTES_TO_BLK(range->start); 2828 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; 2829 unsigned int start_segno, end_segno; 2830 block_t start_block, end_block; 2831 struct cp_control cpc; 2832 struct discard_policy dpolicy; 2833 unsigned long long trimmed = 0; 2834 int err = 0; 2835 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi); 2836 2837 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize) 2838 return -EINVAL; 2839 2840 if (end < MAIN_BLKADDR(sbi)) 2841 goto out; 2842 2843 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 2844 f2fs_warn(sbi, "Found FS corruption, run fsck to fix."); 2845 return -EFSCORRUPTED; 2846 } 2847 2848 /* start/end segment number in main_area */ 2849 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); 2850 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : 2851 GET_SEGNO(sbi, end); 2852 if (need_align) { 2853 start_segno = rounddown(start_segno, sbi->segs_per_sec); 2854 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1; 2855 } 2856 2857 cpc.reason = CP_DISCARD; 2858 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen)); 2859 cpc.trim_start = start_segno; 2860 cpc.trim_end = end_segno; 2861 2862 if (sbi->discard_blks == 0) 2863 goto out; 2864 2865 down_write(&sbi->gc_lock); 2866 err = f2fs_write_checkpoint(sbi, &cpc); 2867 up_write(&sbi->gc_lock); 2868 if (err) 2869 goto out; 2870 2871 /* 2872 * We filed discard candidates, but actually we don't need to wait for 2873 * all of them, since they'll be issued in idle time along with runtime 2874 * discard option. User configuration looks like using runtime discard 2875 * or periodic fstrim instead of it. 2876 */ 2877 if (f2fs_realtime_discard_enable(sbi)) 2878 goto out; 2879 2880 start_block = START_BLOCK(sbi, start_segno); 2881 end_block = START_BLOCK(sbi, end_segno + 1); 2882 2883 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen); 2884 trimmed = __issue_discard_cmd_range(sbi, &dpolicy, 2885 start_block, end_block); 2886 2887 trimmed += __wait_discard_cmd_range(sbi, &dpolicy, 2888 start_block, end_block); 2889 out: 2890 if (!err) 2891 range->len = F2FS_BLK_TO_BYTES(trimmed); 2892 return err; 2893 } 2894 2895 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type) 2896 { 2897 struct curseg_info *curseg = CURSEG_I(sbi, type); 2898 if (curseg->next_blkoff < sbi->blocks_per_seg) 2899 return true; 2900 return false; 2901 } 2902 2903 int f2fs_rw_hint_to_seg_type(enum rw_hint hint) 2904 { 2905 switch (hint) { 2906 case WRITE_LIFE_SHORT: 2907 return CURSEG_HOT_DATA; 2908 case WRITE_LIFE_EXTREME: 2909 return CURSEG_COLD_DATA; 2910 default: 2911 return CURSEG_WARM_DATA; 2912 } 2913 } 2914 2915 /* This returns write hints for each segment type. This hints will be 2916 * passed down to block layer. There are mapping tables which depend on 2917 * the mount option 'whint_mode'. 2918 * 2919 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET. 2920 * 2921 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users. 2922 * 2923 * User F2FS Block 2924 * ---- ---- ----- 2925 * META WRITE_LIFE_NOT_SET 2926 * HOT_NODE " 2927 * WARM_NODE " 2928 * COLD_NODE " 2929 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME 2930 * extension list " " 2931 * 2932 * -- buffered io 2933 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 2934 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 2935 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 2936 * WRITE_LIFE_NONE " " 2937 * WRITE_LIFE_MEDIUM " " 2938 * WRITE_LIFE_LONG " " 2939 * 2940 * -- direct io 2941 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 2942 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 2943 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 2944 * WRITE_LIFE_NONE " WRITE_LIFE_NONE 2945 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM 2946 * WRITE_LIFE_LONG " WRITE_LIFE_LONG 2947 * 2948 * 3) whint_mode=fs-based. F2FS passes down hints with its policy. 2949 * 2950 * User F2FS Block 2951 * ---- ---- ----- 2952 * META WRITE_LIFE_MEDIUM; 2953 * HOT_NODE WRITE_LIFE_NOT_SET 2954 * WARM_NODE " 2955 * COLD_NODE WRITE_LIFE_NONE 2956 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME 2957 * extension list " " 2958 * 2959 * -- buffered io 2960 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 2961 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 2962 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG 2963 * WRITE_LIFE_NONE " " 2964 * WRITE_LIFE_MEDIUM " " 2965 * WRITE_LIFE_LONG " " 2966 * 2967 * -- direct io 2968 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 2969 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 2970 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 2971 * WRITE_LIFE_NONE " WRITE_LIFE_NONE 2972 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM 2973 * WRITE_LIFE_LONG " WRITE_LIFE_LONG 2974 */ 2975 2976 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi, 2977 enum page_type type, enum temp_type temp) 2978 { 2979 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) { 2980 if (type == DATA) { 2981 if (temp == WARM) 2982 return WRITE_LIFE_NOT_SET; 2983 else if (temp == HOT) 2984 return WRITE_LIFE_SHORT; 2985 else if (temp == COLD) 2986 return WRITE_LIFE_EXTREME; 2987 } else { 2988 return WRITE_LIFE_NOT_SET; 2989 } 2990 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) { 2991 if (type == DATA) { 2992 if (temp == WARM) 2993 return WRITE_LIFE_LONG; 2994 else if (temp == HOT) 2995 return WRITE_LIFE_SHORT; 2996 else if (temp == COLD) 2997 return WRITE_LIFE_EXTREME; 2998 } else if (type == NODE) { 2999 if (temp == WARM || temp == HOT) 3000 return WRITE_LIFE_NOT_SET; 3001 else if (temp == COLD) 3002 return WRITE_LIFE_NONE; 3003 } else if (type == META) { 3004 return WRITE_LIFE_MEDIUM; 3005 } 3006 } 3007 return WRITE_LIFE_NOT_SET; 3008 } 3009 3010 static int __get_segment_type_2(struct f2fs_io_info *fio) 3011 { 3012 if (fio->type == DATA) 3013 return CURSEG_HOT_DATA; 3014 else 3015 return CURSEG_HOT_NODE; 3016 } 3017 3018 static int __get_segment_type_4(struct f2fs_io_info *fio) 3019 { 3020 if (fio->type == DATA) { 3021 struct inode *inode = fio->page->mapping->host; 3022 3023 if (S_ISDIR(inode->i_mode)) 3024 return CURSEG_HOT_DATA; 3025 else 3026 return CURSEG_COLD_DATA; 3027 } else { 3028 if (IS_DNODE(fio->page) && is_cold_node(fio->page)) 3029 return CURSEG_WARM_NODE; 3030 else 3031 return CURSEG_COLD_NODE; 3032 } 3033 } 3034 3035 static int __get_segment_type_6(struct f2fs_io_info *fio) 3036 { 3037 if (fio->type == DATA) { 3038 struct inode *inode = fio->page->mapping->host; 3039 3040 if (is_cold_data(fio->page) || file_is_cold(inode) || 3041 f2fs_compressed_file(inode)) 3042 return CURSEG_COLD_DATA; 3043 if (file_is_hot(inode) || 3044 is_inode_flag_set(inode, FI_HOT_DATA) || 3045 f2fs_is_atomic_file(inode) || 3046 f2fs_is_volatile_file(inode)) 3047 return CURSEG_HOT_DATA; 3048 return f2fs_rw_hint_to_seg_type(inode->i_write_hint); 3049 } else { 3050 if (IS_DNODE(fio->page)) 3051 return is_cold_node(fio->page) ? CURSEG_WARM_NODE : 3052 CURSEG_HOT_NODE; 3053 return CURSEG_COLD_NODE; 3054 } 3055 } 3056 3057 static int __get_segment_type(struct f2fs_io_info *fio) 3058 { 3059 int type = 0; 3060 3061 switch (F2FS_OPTION(fio->sbi).active_logs) { 3062 case 2: 3063 type = __get_segment_type_2(fio); 3064 break; 3065 case 4: 3066 type = __get_segment_type_4(fio); 3067 break; 3068 case 6: 3069 type = __get_segment_type_6(fio); 3070 break; 3071 default: 3072 f2fs_bug_on(fio->sbi, true); 3073 } 3074 3075 if (IS_HOT(type)) 3076 fio->temp = HOT; 3077 else if (IS_WARM(type)) 3078 fio->temp = WARM; 3079 else 3080 fio->temp = COLD; 3081 return type; 3082 } 3083 3084 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, 3085 block_t old_blkaddr, block_t *new_blkaddr, 3086 struct f2fs_summary *sum, int type, 3087 struct f2fs_io_info *fio, bool add_list) 3088 { 3089 struct sit_info *sit_i = SIT_I(sbi); 3090 struct curseg_info *curseg = CURSEG_I(sbi, type); 3091 bool put_pin_sem = false; 3092 3093 if (type == CURSEG_COLD_DATA) { 3094 /* GC during CURSEG_COLD_DATA_PINNED allocation */ 3095 if (down_read_trylock(&sbi->pin_sem)) { 3096 put_pin_sem = true; 3097 } else { 3098 type = CURSEG_WARM_DATA; 3099 curseg = CURSEG_I(sbi, type); 3100 } 3101 } else if (type == CURSEG_COLD_DATA_PINNED) { 3102 type = CURSEG_COLD_DATA; 3103 } 3104 3105 down_read(&SM_I(sbi)->curseg_lock); 3106 3107 mutex_lock(&curseg->curseg_mutex); 3108 down_write(&sit_i->sentry_lock); 3109 3110 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 3111 3112 f2fs_wait_discard_bio(sbi, *new_blkaddr); 3113 3114 /* 3115 * __add_sum_entry should be resided under the curseg_mutex 3116 * because, this function updates a summary entry in the 3117 * current summary block. 3118 */ 3119 __add_sum_entry(sbi, type, sum); 3120 3121 __refresh_next_blkoff(sbi, curseg); 3122 3123 stat_inc_block_count(sbi, curseg); 3124 3125 /* 3126 * SIT information should be updated before segment allocation, 3127 * since SSR needs latest valid block information. 3128 */ 3129 update_sit_entry(sbi, *new_blkaddr, 1); 3130 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 3131 update_sit_entry(sbi, old_blkaddr, -1); 3132 3133 if (!__has_curseg_space(sbi, type)) 3134 sit_i->s_ops->allocate_segment(sbi, type, false); 3135 3136 /* 3137 * segment dirty status should be updated after segment allocation, 3138 * so we just need to update status only one time after previous 3139 * segment being closed. 3140 */ 3141 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3142 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr)); 3143 3144 up_write(&sit_i->sentry_lock); 3145 3146 if (page && IS_NODESEG(type)) { 3147 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); 3148 3149 f2fs_inode_chksum_set(sbi, page); 3150 } 3151 3152 if (F2FS_IO_ALIGNED(sbi)) 3153 fio->retry = false; 3154 3155 if (add_list) { 3156 struct f2fs_bio_info *io; 3157 3158 INIT_LIST_HEAD(&fio->list); 3159 fio->in_list = true; 3160 io = sbi->write_io[fio->type] + fio->temp; 3161 spin_lock(&io->io_lock); 3162 list_add_tail(&fio->list, &io->io_list); 3163 spin_unlock(&io->io_lock); 3164 } 3165 3166 mutex_unlock(&curseg->curseg_mutex); 3167 3168 up_read(&SM_I(sbi)->curseg_lock); 3169 3170 if (put_pin_sem) 3171 up_read(&sbi->pin_sem); 3172 } 3173 3174 static void update_device_state(struct f2fs_io_info *fio) 3175 { 3176 struct f2fs_sb_info *sbi = fio->sbi; 3177 unsigned int devidx; 3178 3179 if (!f2fs_is_multi_device(sbi)) 3180 return; 3181 3182 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr); 3183 3184 /* update device state for fsync */ 3185 f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO); 3186 3187 /* update device state for checkpoint */ 3188 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) { 3189 spin_lock(&sbi->dev_lock); 3190 f2fs_set_bit(devidx, (char *)&sbi->dirty_device); 3191 spin_unlock(&sbi->dev_lock); 3192 } 3193 } 3194 3195 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) 3196 { 3197 int type = __get_segment_type(fio); 3198 bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA); 3199 3200 if (keep_order) 3201 down_read(&fio->sbi->io_order_lock); 3202 reallocate: 3203 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr, 3204 &fio->new_blkaddr, sum, type, fio, true); 3205 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO) 3206 invalidate_mapping_pages(META_MAPPING(fio->sbi), 3207 fio->old_blkaddr, fio->old_blkaddr); 3208 3209 /* writeout dirty page into bdev */ 3210 f2fs_submit_page_write(fio); 3211 if (fio->retry) { 3212 fio->old_blkaddr = fio->new_blkaddr; 3213 goto reallocate; 3214 } 3215 3216 update_device_state(fio); 3217 3218 if (keep_order) 3219 up_read(&fio->sbi->io_order_lock); 3220 } 3221 3222 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page, 3223 enum iostat_type io_type) 3224 { 3225 struct f2fs_io_info fio = { 3226 .sbi = sbi, 3227 .type = META, 3228 .temp = HOT, 3229 .op = REQ_OP_WRITE, 3230 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO, 3231 .old_blkaddr = page->index, 3232 .new_blkaddr = page->index, 3233 .page = page, 3234 .encrypted_page = NULL, 3235 .in_list = false, 3236 }; 3237 3238 if (unlikely(page->index >= MAIN_BLKADDR(sbi))) 3239 fio.op_flags &= ~REQ_META; 3240 3241 set_page_writeback(page); 3242 ClearPageError(page); 3243 f2fs_submit_page_write(&fio); 3244 3245 stat_inc_meta_count(sbi, page->index); 3246 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE); 3247 } 3248 3249 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio) 3250 { 3251 struct f2fs_summary sum; 3252 3253 set_summary(&sum, nid, 0, 0); 3254 do_write_page(&sum, fio); 3255 3256 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE); 3257 } 3258 3259 void f2fs_outplace_write_data(struct dnode_of_data *dn, 3260 struct f2fs_io_info *fio) 3261 { 3262 struct f2fs_sb_info *sbi = fio->sbi; 3263 struct f2fs_summary sum; 3264 3265 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); 3266 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version); 3267 do_write_page(&sum, fio); 3268 f2fs_update_data_blkaddr(dn, fio->new_blkaddr); 3269 3270 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE); 3271 } 3272 3273 int f2fs_inplace_write_data(struct f2fs_io_info *fio) 3274 { 3275 int err; 3276 struct f2fs_sb_info *sbi = fio->sbi; 3277 unsigned int segno; 3278 3279 fio->new_blkaddr = fio->old_blkaddr; 3280 /* i/o temperature is needed for passing down write hints */ 3281 __get_segment_type(fio); 3282 3283 segno = GET_SEGNO(sbi, fio->new_blkaddr); 3284 3285 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) { 3286 set_sbi_flag(sbi, SBI_NEED_FSCK); 3287 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.", 3288 __func__, segno); 3289 return -EFSCORRUPTED; 3290 } 3291 3292 stat_inc_inplace_blocks(fio->sbi); 3293 3294 if (fio->bio && !(SM_I(sbi)->ipu_policy & (1 << F2FS_IPU_NOCACHE))) 3295 err = f2fs_merge_page_bio(fio); 3296 else 3297 err = f2fs_submit_page_bio(fio); 3298 if (!err) { 3299 update_device_state(fio); 3300 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE); 3301 } 3302 3303 return err; 3304 } 3305 3306 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi, 3307 unsigned int segno) 3308 { 3309 int i; 3310 3311 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 3312 if (CURSEG_I(sbi, i)->segno == segno) 3313 break; 3314 } 3315 return i; 3316 } 3317 3318 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 3319 block_t old_blkaddr, block_t new_blkaddr, 3320 bool recover_curseg, bool recover_newaddr) 3321 { 3322 struct sit_info *sit_i = SIT_I(sbi); 3323 struct curseg_info *curseg; 3324 unsigned int segno, old_cursegno; 3325 struct seg_entry *se; 3326 int type; 3327 unsigned short old_blkoff; 3328 3329 segno = GET_SEGNO(sbi, new_blkaddr); 3330 se = get_seg_entry(sbi, segno); 3331 type = se->type; 3332 3333 down_write(&SM_I(sbi)->curseg_lock); 3334 3335 if (!recover_curseg) { 3336 /* for recovery flow */ 3337 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { 3338 if (old_blkaddr == NULL_ADDR) 3339 type = CURSEG_COLD_DATA; 3340 else 3341 type = CURSEG_WARM_DATA; 3342 } 3343 } else { 3344 if (IS_CURSEG(sbi, segno)) { 3345 /* se->type is volatile as SSR allocation */ 3346 type = __f2fs_get_curseg(sbi, segno); 3347 f2fs_bug_on(sbi, type == NO_CHECK_TYPE); 3348 } else { 3349 type = CURSEG_WARM_DATA; 3350 } 3351 } 3352 3353 f2fs_bug_on(sbi, !IS_DATASEG(type)); 3354 curseg = CURSEG_I(sbi, type); 3355 3356 mutex_lock(&curseg->curseg_mutex); 3357 down_write(&sit_i->sentry_lock); 3358 3359 old_cursegno = curseg->segno; 3360 old_blkoff = curseg->next_blkoff; 3361 3362 /* change the current segment */ 3363 if (segno != curseg->segno) { 3364 curseg->next_segno = segno; 3365 change_curseg(sbi, type); 3366 } 3367 3368 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); 3369 __add_sum_entry(sbi, type, sum); 3370 3371 if (!recover_curseg || recover_newaddr) 3372 update_sit_entry(sbi, new_blkaddr, 1); 3373 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) { 3374 invalidate_mapping_pages(META_MAPPING(sbi), 3375 old_blkaddr, old_blkaddr); 3376 update_sit_entry(sbi, old_blkaddr, -1); 3377 } 3378 3379 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3380 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); 3381 3382 locate_dirty_segment(sbi, old_cursegno); 3383 3384 if (recover_curseg) { 3385 if (old_cursegno != curseg->segno) { 3386 curseg->next_segno = old_cursegno; 3387 change_curseg(sbi, type); 3388 } 3389 curseg->next_blkoff = old_blkoff; 3390 } 3391 3392 up_write(&sit_i->sentry_lock); 3393 mutex_unlock(&curseg->curseg_mutex); 3394 up_write(&SM_I(sbi)->curseg_lock); 3395 } 3396 3397 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, 3398 block_t old_addr, block_t new_addr, 3399 unsigned char version, bool recover_curseg, 3400 bool recover_newaddr) 3401 { 3402 struct f2fs_summary sum; 3403 3404 set_summary(&sum, dn->nid, dn->ofs_in_node, version); 3405 3406 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr, 3407 recover_curseg, recover_newaddr); 3408 3409 f2fs_update_data_blkaddr(dn, new_addr); 3410 } 3411 3412 void f2fs_wait_on_page_writeback(struct page *page, 3413 enum page_type type, bool ordered, bool locked) 3414 { 3415 if (PageWriteback(page)) { 3416 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 3417 3418 /* submit cached LFS IO */ 3419 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type); 3420 /* sbumit cached IPU IO */ 3421 f2fs_submit_merged_ipu_write(sbi, NULL, page); 3422 if (ordered) { 3423 wait_on_page_writeback(page); 3424 f2fs_bug_on(sbi, locked && PageWriteback(page)); 3425 } else { 3426 wait_for_stable_page(page); 3427 } 3428 } 3429 } 3430 3431 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr) 3432 { 3433 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3434 struct page *cpage; 3435 3436 if (!f2fs_post_read_required(inode)) 3437 return; 3438 3439 if (!__is_valid_data_blkaddr(blkaddr)) 3440 return; 3441 3442 cpage = find_lock_page(META_MAPPING(sbi), blkaddr); 3443 if (cpage) { 3444 f2fs_wait_on_page_writeback(cpage, DATA, true, true); 3445 f2fs_put_page(cpage, 1); 3446 } 3447 } 3448 3449 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr, 3450 block_t len) 3451 { 3452 block_t i; 3453 3454 for (i = 0; i < len; i++) 3455 f2fs_wait_on_block_writeback(inode, blkaddr + i); 3456 } 3457 3458 static int read_compacted_summaries(struct f2fs_sb_info *sbi) 3459 { 3460 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3461 struct curseg_info *seg_i; 3462 unsigned char *kaddr; 3463 struct page *page; 3464 block_t start; 3465 int i, j, offset; 3466 3467 start = start_sum_block(sbi); 3468 3469 page = f2fs_get_meta_page(sbi, start++); 3470 if (IS_ERR(page)) 3471 return PTR_ERR(page); 3472 kaddr = (unsigned char *)page_address(page); 3473 3474 /* Step 1: restore nat cache */ 3475 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 3476 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE); 3477 3478 /* Step 2: restore sit cache */ 3479 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 3480 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE); 3481 offset = 2 * SUM_JOURNAL_SIZE; 3482 3483 /* Step 3: restore summary entries */ 3484 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 3485 unsigned short blk_off; 3486 unsigned int segno; 3487 3488 seg_i = CURSEG_I(sbi, i); 3489 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 3490 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 3491 seg_i->next_segno = segno; 3492 reset_curseg(sbi, i, 0); 3493 seg_i->alloc_type = ckpt->alloc_type[i]; 3494 seg_i->next_blkoff = blk_off; 3495 3496 if (seg_i->alloc_type == SSR) 3497 blk_off = sbi->blocks_per_seg; 3498 3499 for (j = 0; j < blk_off; j++) { 3500 struct f2fs_summary *s; 3501 s = (struct f2fs_summary *)(kaddr + offset); 3502 seg_i->sum_blk->entries[j] = *s; 3503 offset += SUMMARY_SIZE; 3504 if (offset + SUMMARY_SIZE <= PAGE_SIZE - 3505 SUM_FOOTER_SIZE) 3506 continue; 3507 3508 f2fs_put_page(page, 1); 3509 page = NULL; 3510 3511 page = f2fs_get_meta_page(sbi, start++); 3512 if (IS_ERR(page)) 3513 return PTR_ERR(page); 3514 kaddr = (unsigned char *)page_address(page); 3515 offset = 0; 3516 } 3517 } 3518 f2fs_put_page(page, 1); 3519 return 0; 3520 } 3521 3522 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 3523 { 3524 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3525 struct f2fs_summary_block *sum; 3526 struct curseg_info *curseg; 3527 struct page *new; 3528 unsigned short blk_off; 3529 unsigned int segno = 0; 3530 block_t blk_addr = 0; 3531 int err = 0; 3532 3533 /* get segment number and block addr */ 3534 if (IS_DATASEG(type)) { 3535 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 3536 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 3537 CURSEG_HOT_DATA]); 3538 if (__exist_node_summaries(sbi)) 3539 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type); 3540 else 3541 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 3542 } else { 3543 segno = le32_to_cpu(ckpt->cur_node_segno[type - 3544 CURSEG_HOT_NODE]); 3545 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 3546 CURSEG_HOT_NODE]); 3547 if (__exist_node_summaries(sbi)) 3548 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 3549 type - CURSEG_HOT_NODE); 3550 else 3551 blk_addr = GET_SUM_BLOCK(sbi, segno); 3552 } 3553 3554 new = f2fs_get_meta_page(sbi, blk_addr); 3555 if (IS_ERR(new)) 3556 return PTR_ERR(new); 3557 sum = (struct f2fs_summary_block *)page_address(new); 3558 3559 if (IS_NODESEG(type)) { 3560 if (__exist_node_summaries(sbi)) { 3561 struct f2fs_summary *ns = &sum->entries[0]; 3562 int i; 3563 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { 3564 ns->version = 0; 3565 ns->ofs_in_node = 0; 3566 } 3567 } else { 3568 err = f2fs_restore_node_summary(sbi, segno, sum); 3569 if (err) 3570 goto out; 3571 } 3572 } 3573 3574 /* set uncompleted segment to curseg */ 3575 curseg = CURSEG_I(sbi, type); 3576 mutex_lock(&curseg->curseg_mutex); 3577 3578 /* update journal info */ 3579 down_write(&curseg->journal_rwsem); 3580 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE); 3581 up_write(&curseg->journal_rwsem); 3582 3583 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE); 3584 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE); 3585 curseg->next_segno = segno; 3586 reset_curseg(sbi, type, 0); 3587 curseg->alloc_type = ckpt->alloc_type[type]; 3588 curseg->next_blkoff = blk_off; 3589 mutex_unlock(&curseg->curseg_mutex); 3590 out: 3591 f2fs_put_page(new, 1); 3592 return err; 3593 } 3594 3595 static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 3596 { 3597 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal; 3598 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal; 3599 int type = CURSEG_HOT_DATA; 3600 int err; 3601 3602 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) { 3603 int npages = f2fs_npages_for_summary_flush(sbi, true); 3604 3605 if (npages >= 2) 3606 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages, 3607 META_CP, true); 3608 3609 /* restore for compacted data summary */ 3610 err = read_compacted_summaries(sbi); 3611 if (err) 3612 return err; 3613 type = CURSEG_HOT_NODE; 3614 } 3615 3616 if (__exist_node_summaries(sbi)) 3617 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type), 3618 NR_CURSEG_TYPE - type, META_CP, true); 3619 3620 for (; type <= CURSEG_COLD_NODE; type++) { 3621 err = read_normal_summaries(sbi, type); 3622 if (err) 3623 return err; 3624 } 3625 3626 /* sanity check for summary blocks */ 3627 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES || 3628 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) { 3629 f2fs_err(sbi, "invalid journal entries nats %u sits %u\n", 3630 nats_in_cursum(nat_j), sits_in_cursum(sit_j)); 3631 return -EINVAL; 3632 } 3633 3634 return 0; 3635 } 3636 3637 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 3638 { 3639 struct page *page; 3640 unsigned char *kaddr; 3641 struct f2fs_summary *summary; 3642 struct curseg_info *seg_i; 3643 int written_size = 0; 3644 int i, j; 3645 3646 page = f2fs_grab_meta_page(sbi, blkaddr++); 3647 kaddr = (unsigned char *)page_address(page); 3648 memset(kaddr, 0, PAGE_SIZE); 3649 3650 /* Step 1: write nat cache */ 3651 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 3652 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE); 3653 written_size += SUM_JOURNAL_SIZE; 3654 3655 /* Step 2: write sit cache */ 3656 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 3657 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE); 3658 written_size += SUM_JOURNAL_SIZE; 3659 3660 /* Step 3: write summary entries */ 3661 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 3662 unsigned short blkoff; 3663 seg_i = CURSEG_I(sbi, i); 3664 if (sbi->ckpt->alloc_type[i] == SSR) 3665 blkoff = sbi->blocks_per_seg; 3666 else 3667 blkoff = curseg_blkoff(sbi, i); 3668 3669 for (j = 0; j < blkoff; j++) { 3670 if (!page) { 3671 page = f2fs_grab_meta_page(sbi, blkaddr++); 3672 kaddr = (unsigned char *)page_address(page); 3673 memset(kaddr, 0, PAGE_SIZE); 3674 written_size = 0; 3675 } 3676 summary = (struct f2fs_summary *)(kaddr + written_size); 3677 *summary = seg_i->sum_blk->entries[j]; 3678 written_size += SUMMARY_SIZE; 3679 3680 if (written_size + SUMMARY_SIZE <= PAGE_SIZE - 3681 SUM_FOOTER_SIZE) 3682 continue; 3683 3684 set_page_dirty(page); 3685 f2fs_put_page(page, 1); 3686 page = NULL; 3687 } 3688 } 3689 if (page) { 3690 set_page_dirty(page); 3691 f2fs_put_page(page, 1); 3692 } 3693 } 3694 3695 static void write_normal_summaries(struct f2fs_sb_info *sbi, 3696 block_t blkaddr, int type) 3697 { 3698 int i, end; 3699 if (IS_DATASEG(type)) 3700 end = type + NR_CURSEG_DATA_TYPE; 3701 else 3702 end = type + NR_CURSEG_NODE_TYPE; 3703 3704 for (i = type; i < end; i++) 3705 write_current_sum_page(sbi, i, blkaddr + (i - type)); 3706 } 3707 3708 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 3709 { 3710 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) 3711 write_compacted_summaries(sbi, start_blk); 3712 else 3713 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 3714 } 3715 3716 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 3717 { 3718 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 3719 } 3720 3721 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type, 3722 unsigned int val, int alloc) 3723 { 3724 int i; 3725 3726 if (type == NAT_JOURNAL) { 3727 for (i = 0; i < nats_in_cursum(journal); i++) { 3728 if (le32_to_cpu(nid_in_journal(journal, i)) == val) 3729 return i; 3730 } 3731 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL)) 3732 return update_nats_in_cursum(journal, 1); 3733 } else if (type == SIT_JOURNAL) { 3734 for (i = 0; i < sits_in_cursum(journal); i++) 3735 if (le32_to_cpu(segno_in_journal(journal, i)) == val) 3736 return i; 3737 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL)) 3738 return update_sits_in_cursum(journal, 1); 3739 } 3740 return -1; 3741 } 3742 3743 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, 3744 unsigned int segno) 3745 { 3746 return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno)); 3747 } 3748 3749 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, 3750 unsigned int start) 3751 { 3752 struct sit_info *sit_i = SIT_I(sbi); 3753 struct page *page; 3754 pgoff_t src_off, dst_off; 3755 3756 src_off = current_sit_addr(sbi, start); 3757 dst_off = next_sit_addr(sbi, src_off); 3758 3759 page = f2fs_grab_meta_page(sbi, dst_off); 3760 seg_info_to_sit_page(sbi, page, start); 3761 3762 set_page_dirty(page); 3763 set_to_next_sit(sit_i, start); 3764 3765 return page; 3766 } 3767 3768 static struct sit_entry_set *grab_sit_entry_set(void) 3769 { 3770 struct sit_entry_set *ses = 3771 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS); 3772 3773 ses->entry_cnt = 0; 3774 INIT_LIST_HEAD(&ses->set_list); 3775 return ses; 3776 } 3777 3778 static void release_sit_entry_set(struct sit_entry_set *ses) 3779 { 3780 list_del(&ses->set_list); 3781 kmem_cache_free(sit_entry_set_slab, ses); 3782 } 3783 3784 static void adjust_sit_entry_set(struct sit_entry_set *ses, 3785 struct list_head *head) 3786 { 3787 struct sit_entry_set *next = ses; 3788 3789 if (list_is_last(&ses->set_list, head)) 3790 return; 3791 3792 list_for_each_entry_continue(next, head, set_list) 3793 if (ses->entry_cnt <= next->entry_cnt) 3794 break; 3795 3796 list_move_tail(&ses->set_list, &next->set_list); 3797 } 3798 3799 static void add_sit_entry(unsigned int segno, struct list_head *head) 3800 { 3801 struct sit_entry_set *ses; 3802 unsigned int start_segno = START_SEGNO(segno); 3803 3804 list_for_each_entry(ses, head, set_list) { 3805 if (ses->start_segno == start_segno) { 3806 ses->entry_cnt++; 3807 adjust_sit_entry_set(ses, head); 3808 return; 3809 } 3810 } 3811 3812 ses = grab_sit_entry_set(); 3813 3814 ses->start_segno = start_segno; 3815 ses->entry_cnt++; 3816 list_add(&ses->set_list, head); 3817 } 3818 3819 static void add_sits_in_set(struct f2fs_sb_info *sbi) 3820 { 3821 struct f2fs_sm_info *sm_info = SM_I(sbi); 3822 struct list_head *set_list = &sm_info->sit_entry_set; 3823 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; 3824 unsigned int segno; 3825 3826 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) 3827 add_sit_entry(segno, set_list); 3828 } 3829 3830 static void remove_sits_in_journal(struct f2fs_sb_info *sbi) 3831 { 3832 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 3833 struct f2fs_journal *journal = curseg->journal; 3834 int i; 3835 3836 down_write(&curseg->journal_rwsem); 3837 for (i = 0; i < sits_in_cursum(journal); i++) { 3838 unsigned int segno; 3839 bool dirtied; 3840 3841 segno = le32_to_cpu(segno_in_journal(journal, i)); 3842 dirtied = __mark_sit_entry_dirty(sbi, segno); 3843 3844 if (!dirtied) 3845 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); 3846 } 3847 update_sits_in_cursum(journal, -i); 3848 up_write(&curseg->journal_rwsem); 3849 } 3850 3851 /* 3852 * CP calls this function, which flushes SIT entries including sit_journal, 3853 * and moves prefree segs to free segs. 3854 */ 3855 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 3856 { 3857 struct sit_info *sit_i = SIT_I(sbi); 3858 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 3859 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 3860 struct f2fs_journal *journal = curseg->journal; 3861 struct sit_entry_set *ses, *tmp; 3862 struct list_head *head = &SM_I(sbi)->sit_entry_set; 3863 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS); 3864 struct seg_entry *se; 3865 3866 down_write(&sit_i->sentry_lock); 3867 3868 if (!sit_i->dirty_sentries) 3869 goto out; 3870 3871 /* 3872 * add and account sit entries of dirty bitmap in sit entry 3873 * set temporarily 3874 */ 3875 add_sits_in_set(sbi); 3876 3877 /* 3878 * if there are no enough space in journal to store dirty sit 3879 * entries, remove all entries from journal and add and account 3880 * them in sit entry set. 3881 */ 3882 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) || 3883 !to_journal) 3884 remove_sits_in_journal(sbi); 3885 3886 /* 3887 * there are two steps to flush sit entries: 3888 * #1, flush sit entries to journal in current cold data summary block. 3889 * #2, flush sit entries to sit page. 3890 */ 3891 list_for_each_entry_safe(ses, tmp, head, set_list) { 3892 struct page *page = NULL; 3893 struct f2fs_sit_block *raw_sit = NULL; 3894 unsigned int start_segno = ses->start_segno; 3895 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, 3896 (unsigned long)MAIN_SEGS(sbi)); 3897 unsigned int segno = start_segno; 3898 3899 if (to_journal && 3900 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL)) 3901 to_journal = false; 3902 3903 if (to_journal) { 3904 down_write(&curseg->journal_rwsem); 3905 } else { 3906 page = get_next_sit_page(sbi, start_segno); 3907 raw_sit = page_address(page); 3908 } 3909 3910 /* flush dirty sit entries in region of current sit set */ 3911 for_each_set_bit_from(segno, bitmap, end) { 3912 int offset, sit_offset; 3913 3914 se = get_seg_entry(sbi, segno); 3915 #ifdef CONFIG_F2FS_CHECK_FS 3916 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir, 3917 SIT_VBLOCK_MAP_SIZE)) 3918 f2fs_bug_on(sbi, 1); 3919 #endif 3920 3921 /* add discard candidates */ 3922 if (!(cpc->reason & CP_DISCARD)) { 3923 cpc->trim_start = segno; 3924 add_discard_addrs(sbi, cpc, false); 3925 } 3926 3927 if (to_journal) { 3928 offset = f2fs_lookup_journal_in_cursum(journal, 3929 SIT_JOURNAL, segno, 1); 3930 f2fs_bug_on(sbi, offset < 0); 3931 segno_in_journal(journal, offset) = 3932 cpu_to_le32(segno); 3933 seg_info_to_raw_sit(se, 3934 &sit_in_journal(journal, offset)); 3935 check_block_count(sbi, segno, 3936 &sit_in_journal(journal, offset)); 3937 } else { 3938 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 3939 seg_info_to_raw_sit(se, 3940 &raw_sit->entries[sit_offset]); 3941 check_block_count(sbi, segno, 3942 &raw_sit->entries[sit_offset]); 3943 } 3944 3945 __clear_bit(segno, bitmap); 3946 sit_i->dirty_sentries--; 3947 ses->entry_cnt--; 3948 } 3949 3950 if (to_journal) 3951 up_write(&curseg->journal_rwsem); 3952 else 3953 f2fs_put_page(page, 1); 3954 3955 f2fs_bug_on(sbi, ses->entry_cnt); 3956 release_sit_entry_set(ses); 3957 } 3958 3959 f2fs_bug_on(sbi, !list_empty(head)); 3960 f2fs_bug_on(sbi, sit_i->dirty_sentries); 3961 out: 3962 if (cpc->reason & CP_DISCARD) { 3963 __u64 trim_start = cpc->trim_start; 3964 3965 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) 3966 add_discard_addrs(sbi, cpc, false); 3967 3968 cpc->trim_start = trim_start; 3969 } 3970 up_write(&sit_i->sentry_lock); 3971 3972 set_prefree_as_free_segments(sbi); 3973 } 3974 3975 static int build_sit_info(struct f2fs_sb_info *sbi) 3976 { 3977 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 3978 struct sit_info *sit_i; 3979 unsigned int sit_segs, start; 3980 char *src_bitmap, *bitmap; 3981 unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size; 3982 3983 /* allocate memory for SIT information */ 3984 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL); 3985 if (!sit_i) 3986 return -ENOMEM; 3987 3988 SM_I(sbi)->sit_info = sit_i; 3989 3990 sit_i->sentries = 3991 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry), 3992 MAIN_SEGS(sbi)), 3993 GFP_KERNEL); 3994 if (!sit_i->sentries) 3995 return -ENOMEM; 3996 3997 main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 3998 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size, 3999 GFP_KERNEL); 4000 if (!sit_i->dirty_sentries_bitmap) 4001 return -ENOMEM; 4002 4003 #ifdef CONFIG_F2FS_CHECK_FS 4004 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 4; 4005 #else 4006 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 3; 4007 #endif 4008 sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4009 if (!sit_i->bitmap) 4010 return -ENOMEM; 4011 4012 bitmap = sit_i->bitmap; 4013 4014 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4015 sit_i->sentries[start].cur_valid_map = bitmap; 4016 bitmap += SIT_VBLOCK_MAP_SIZE; 4017 4018 sit_i->sentries[start].ckpt_valid_map = bitmap; 4019 bitmap += SIT_VBLOCK_MAP_SIZE; 4020 4021 #ifdef CONFIG_F2FS_CHECK_FS 4022 sit_i->sentries[start].cur_valid_map_mir = bitmap; 4023 bitmap += SIT_VBLOCK_MAP_SIZE; 4024 #endif 4025 4026 sit_i->sentries[start].discard_map = bitmap; 4027 bitmap += SIT_VBLOCK_MAP_SIZE; 4028 } 4029 4030 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 4031 if (!sit_i->tmp_map) 4032 return -ENOMEM; 4033 4034 if (__is_large_section(sbi)) { 4035 sit_i->sec_entries = 4036 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry), 4037 MAIN_SECS(sbi)), 4038 GFP_KERNEL); 4039 if (!sit_i->sec_entries) 4040 return -ENOMEM; 4041 } 4042 4043 /* get information related with SIT */ 4044 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 4045 4046 /* setup SIT bitmap from ckeckpoint pack */ 4047 sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 4048 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 4049 4050 sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL); 4051 if (!sit_i->sit_bitmap) 4052 return -ENOMEM; 4053 4054 #ifdef CONFIG_F2FS_CHECK_FS 4055 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, 4056 sit_bitmap_size, GFP_KERNEL); 4057 if (!sit_i->sit_bitmap_mir) 4058 return -ENOMEM; 4059 4060 sit_i->invalid_segmap = f2fs_kvzalloc(sbi, 4061 main_bitmap_size, GFP_KERNEL); 4062 if (!sit_i->invalid_segmap) 4063 return -ENOMEM; 4064 #endif 4065 4066 /* init SIT information */ 4067 sit_i->s_ops = &default_salloc_ops; 4068 4069 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 4070 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; 4071 sit_i->written_valid_blocks = 0; 4072 sit_i->bitmap_size = sit_bitmap_size; 4073 sit_i->dirty_sentries = 0; 4074 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 4075 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 4076 sit_i->mounted_time = ktime_get_boottime_seconds(); 4077 init_rwsem(&sit_i->sentry_lock); 4078 return 0; 4079 } 4080 4081 static int build_free_segmap(struct f2fs_sb_info *sbi) 4082 { 4083 struct free_segmap_info *free_i; 4084 unsigned int bitmap_size, sec_bitmap_size; 4085 4086 /* allocate memory for free segmap information */ 4087 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL); 4088 if (!free_i) 4089 return -ENOMEM; 4090 4091 SM_I(sbi)->free_info = free_i; 4092 4093 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4094 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL); 4095 if (!free_i->free_segmap) 4096 return -ENOMEM; 4097 4098 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4099 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL); 4100 if (!free_i->free_secmap) 4101 return -ENOMEM; 4102 4103 /* set all segments as dirty temporarily */ 4104 memset(free_i->free_segmap, 0xff, bitmap_size); 4105 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 4106 4107 /* init free segmap information */ 4108 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); 4109 free_i->free_segments = 0; 4110 free_i->free_sections = 0; 4111 spin_lock_init(&free_i->segmap_lock); 4112 return 0; 4113 } 4114 4115 static int build_curseg(struct f2fs_sb_info *sbi) 4116 { 4117 struct curseg_info *array; 4118 int i; 4119 4120 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)), 4121 GFP_KERNEL); 4122 if (!array) 4123 return -ENOMEM; 4124 4125 SM_I(sbi)->curseg_array = array; 4126 4127 for (i = 0; i < NR_CURSEG_TYPE; i++) { 4128 mutex_init(&array[i].curseg_mutex); 4129 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL); 4130 if (!array[i].sum_blk) 4131 return -ENOMEM; 4132 init_rwsem(&array[i].journal_rwsem); 4133 array[i].journal = f2fs_kzalloc(sbi, 4134 sizeof(struct f2fs_journal), GFP_KERNEL); 4135 if (!array[i].journal) 4136 return -ENOMEM; 4137 array[i].segno = NULL_SEGNO; 4138 array[i].next_blkoff = 0; 4139 } 4140 return restore_curseg_summaries(sbi); 4141 } 4142 4143 static int build_sit_entries(struct f2fs_sb_info *sbi) 4144 { 4145 struct sit_info *sit_i = SIT_I(sbi); 4146 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4147 struct f2fs_journal *journal = curseg->journal; 4148 struct seg_entry *se; 4149 struct f2fs_sit_entry sit; 4150 int sit_blk_cnt = SIT_BLK_CNT(sbi); 4151 unsigned int i, start, end; 4152 unsigned int readed, start_blk = 0; 4153 int err = 0; 4154 block_t total_node_blocks = 0; 4155 4156 do { 4157 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES, 4158 META_SIT, true); 4159 4160 start = start_blk * sit_i->sents_per_block; 4161 end = (start_blk + readed) * sit_i->sents_per_block; 4162 4163 for (; start < end && start < MAIN_SEGS(sbi); start++) { 4164 struct f2fs_sit_block *sit_blk; 4165 struct page *page; 4166 4167 se = &sit_i->sentries[start]; 4168 page = get_current_sit_page(sbi, start); 4169 if (IS_ERR(page)) 4170 return PTR_ERR(page); 4171 sit_blk = (struct f2fs_sit_block *)page_address(page); 4172 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 4173 f2fs_put_page(page, 1); 4174 4175 err = check_block_count(sbi, start, &sit); 4176 if (err) 4177 return err; 4178 seg_info_from_raw_sit(se, &sit); 4179 if (IS_NODESEG(se->type)) 4180 total_node_blocks += se->valid_blocks; 4181 4182 /* build discard map only one time */ 4183 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4184 memset(se->discard_map, 0xff, 4185 SIT_VBLOCK_MAP_SIZE); 4186 } else { 4187 memcpy(se->discard_map, 4188 se->cur_valid_map, 4189 SIT_VBLOCK_MAP_SIZE); 4190 sbi->discard_blks += 4191 sbi->blocks_per_seg - 4192 se->valid_blocks; 4193 } 4194 4195 if (__is_large_section(sbi)) 4196 get_sec_entry(sbi, start)->valid_blocks += 4197 se->valid_blocks; 4198 } 4199 start_blk += readed; 4200 } while (start_blk < sit_blk_cnt); 4201 4202 down_read(&curseg->journal_rwsem); 4203 for (i = 0; i < sits_in_cursum(journal); i++) { 4204 unsigned int old_valid_blocks; 4205 4206 start = le32_to_cpu(segno_in_journal(journal, i)); 4207 if (start >= MAIN_SEGS(sbi)) { 4208 f2fs_err(sbi, "Wrong journal entry on segno %u", 4209 start); 4210 err = -EFSCORRUPTED; 4211 break; 4212 } 4213 4214 se = &sit_i->sentries[start]; 4215 sit = sit_in_journal(journal, i); 4216 4217 old_valid_blocks = se->valid_blocks; 4218 if (IS_NODESEG(se->type)) 4219 total_node_blocks -= old_valid_blocks; 4220 4221 err = check_block_count(sbi, start, &sit); 4222 if (err) 4223 break; 4224 seg_info_from_raw_sit(se, &sit); 4225 if (IS_NODESEG(se->type)) 4226 total_node_blocks += se->valid_blocks; 4227 4228 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4229 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE); 4230 } else { 4231 memcpy(se->discard_map, se->cur_valid_map, 4232 SIT_VBLOCK_MAP_SIZE); 4233 sbi->discard_blks += old_valid_blocks; 4234 sbi->discard_blks -= se->valid_blocks; 4235 } 4236 4237 if (__is_large_section(sbi)) { 4238 get_sec_entry(sbi, start)->valid_blocks += 4239 se->valid_blocks; 4240 get_sec_entry(sbi, start)->valid_blocks -= 4241 old_valid_blocks; 4242 } 4243 } 4244 up_read(&curseg->journal_rwsem); 4245 4246 if (!err && total_node_blocks != valid_node_count(sbi)) { 4247 f2fs_err(sbi, "SIT is corrupted node# %u vs %u", 4248 total_node_blocks, valid_node_count(sbi)); 4249 err = -EFSCORRUPTED; 4250 } 4251 4252 return err; 4253 } 4254 4255 static void init_free_segmap(struct f2fs_sb_info *sbi) 4256 { 4257 unsigned int start; 4258 int type; 4259 4260 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4261 struct seg_entry *sentry = get_seg_entry(sbi, start); 4262 if (!sentry->valid_blocks) 4263 __set_free(sbi, start); 4264 else 4265 SIT_I(sbi)->written_valid_blocks += 4266 sentry->valid_blocks; 4267 } 4268 4269 /* set use the current segments */ 4270 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 4271 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 4272 __set_test_and_inuse(sbi, curseg_t->segno); 4273 } 4274 } 4275 4276 static void init_dirty_segmap(struct f2fs_sb_info *sbi) 4277 { 4278 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4279 struct free_segmap_info *free_i = FREE_I(sbi); 4280 unsigned int segno = 0, offset = 0; 4281 unsigned short valid_blocks; 4282 4283 while (1) { 4284 /* find dirty segment based on free segmap */ 4285 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); 4286 if (segno >= MAIN_SEGS(sbi)) 4287 break; 4288 offset = segno + 1; 4289 valid_blocks = get_valid_blocks(sbi, segno, false); 4290 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks) 4291 continue; 4292 if (valid_blocks > sbi->blocks_per_seg) { 4293 f2fs_bug_on(sbi, 1); 4294 continue; 4295 } 4296 mutex_lock(&dirty_i->seglist_lock); 4297 __locate_dirty_segment(sbi, segno, DIRTY); 4298 mutex_unlock(&dirty_i->seglist_lock); 4299 } 4300 } 4301 4302 static int init_victim_secmap(struct f2fs_sb_info *sbi) 4303 { 4304 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4305 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4306 4307 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4308 if (!dirty_i->victim_secmap) 4309 return -ENOMEM; 4310 return 0; 4311 } 4312 4313 static int build_dirty_segmap(struct f2fs_sb_info *sbi) 4314 { 4315 struct dirty_seglist_info *dirty_i; 4316 unsigned int bitmap_size, i; 4317 4318 /* allocate memory for dirty segments list information */ 4319 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info), 4320 GFP_KERNEL); 4321 if (!dirty_i) 4322 return -ENOMEM; 4323 4324 SM_I(sbi)->dirty_info = dirty_i; 4325 mutex_init(&dirty_i->seglist_lock); 4326 4327 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4328 4329 for (i = 0; i < NR_DIRTY_TYPE; i++) { 4330 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size, 4331 GFP_KERNEL); 4332 if (!dirty_i->dirty_segmap[i]) 4333 return -ENOMEM; 4334 } 4335 4336 init_dirty_segmap(sbi); 4337 return init_victim_secmap(sbi); 4338 } 4339 4340 static int sanity_check_curseg(struct f2fs_sb_info *sbi) 4341 { 4342 int i; 4343 4344 /* 4345 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr; 4346 * In LFS curseg, all blkaddr after .next_blkoff should be unused. 4347 */ 4348 for (i = 0; i < NO_CHECK_TYPE; i++) { 4349 struct curseg_info *curseg = CURSEG_I(sbi, i); 4350 struct seg_entry *se = get_seg_entry(sbi, curseg->segno); 4351 unsigned int blkofs = curseg->next_blkoff; 4352 4353 if (f2fs_test_bit(blkofs, se->cur_valid_map)) 4354 goto out; 4355 4356 if (curseg->alloc_type == SSR) 4357 continue; 4358 4359 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) { 4360 if (!f2fs_test_bit(blkofs, se->cur_valid_map)) 4361 continue; 4362 out: 4363 f2fs_err(sbi, 4364 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u", 4365 i, curseg->segno, curseg->alloc_type, 4366 curseg->next_blkoff, blkofs); 4367 return -EFSCORRUPTED; 4368 } 4369 } 4370 return 0; 4371 } 4372 4373 #ifdef CONFIG_BLK_DEV_ZONED 4374 4375 static int check_zone_write_pointer(struct f2fs_sb_info *sbi, 4376 struct f2fs_dev_info *fdev, 4377 struct blk_zone *zone) 4378 { 4379 unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno; 4380 block_t zone_block, wp_block, last_valid_block; 4381 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 4382 int i, s, b, ret; 4383 struct seg_entry *se; 4384 4385 if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4386 return 0; 4387 4388 wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block); 4389 wp_segno = GET_SEGNO(sbi, wp_block); 4390 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno); 4391 zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block); 4392 zone_segno = GET_SEGNO(sbi, zone_block); 4393 zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno); 4394 4395 if (zone_segno >= MAIN_SEGS(sbi)) 4396 return 0; 4397 4398 /* 4399 * Skip check of zones cursegs point to, since 4400 * fix_curseg_write_pointer() checks them. 4401 */ 4402 for (i = 0; i < NO_CHECK_TYPE; i++) 4403 if (zone_secno == GET_SEC_FROM_SEG(sbi, 4404 CURSEG_I(sbi, i)->segno)) 4405 return 0; 4406 4407 /* 4408 * Get last valid block of the zone. 4409 */ 4410 last_valid_block = zone_block - 1; 4411 for (s = sbi->segs_per_sec - 1; s >= 0; s--) { 4412 segno = zone_segno + s; 4413 se = get_seg_entry(sbi, segno); 4414 for (b = sbi->blocks_per_seg - 1; b >= 0; b--) 4415 if (f2fs_test_bit(b, se->cur_valid_map)) { 4416 last_valid_block = START_BLOCK(sbi, segno) + b; 4417 break; 4418 } 4419 if (last_valid_block >= zone_block) 4420 break; 4421 } 4422 4423 /* 4424 * If last valid block is beyond the write pointer, report the 4425 * inconsistency. This inconsistency does not cause write error 4426 * because the zone will not be selected for write operation until 4427 * it get discarded. Just report it. 4428 */ 4429 if (last_valid_block >= wp_block) { 4430 f2fs_notice(sbi, "Valid block beyond write pointer: " 4431 "valid block[0x%x,0x%x] wp[0x%x,0x%x]", 4432 GET_SEGNO(sbi, last_valid_block), 4433 GET_BLKOFF_FROM_SEG0(sbi, last_valid_block), 4434 wp_segno, wp_blkoff); 4435 return 0; 4436 } 4437 4438 /* 4439 * If there is no valid block in the zone and if write pointer is 4440 * not at zone start, reset the write pointer. 4441 */ 4442 if (last_valid_block + 1 == zone_block && zone->wp != zone->start) { 4443 f2fs_notice(sbi, 4444 "Zone without valid block has non-zero write " 4445 "pointer. Reset the write pointer: wp[0x%x,0x%x]", 4446 wp_segno, wp_blkoff); 4447 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block, 4448 zone->len >> log_sectors_per_block); 4449 if (ret) { 4450 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 4451 fdev->path, ret); 4452 return ret; 4453 } 4454 } 4455 4456 return 0; 4457 } 4458 4459 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi, 4460 block_t zone_blkaddr) 4461 { 4462 int i; 4463 4464 for (i = 0; i < sbi->s_ndevs; i++) { 4465 if (!bdev_is_zoned(FDEV(i).bdev)) 4466 continue; 4467 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr && 4468 zone_blkaddr <= FDEV(i).end_blk)) 4469 return &FDEV(i); 4470 } 4471 4472 return NULL; 4473 } 4474 4475 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx, 4476 void *data) { 4477 memcpy(data, zone, sizeof(struct blk_zone)); 4478 return 0; 4479 } 4480 4481 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type) 4482 { 4483 struct curseg_info *cs = CURSEG_I(sbi, type); 4484 struct f2fs_dev_info *zbd; 4485 struct blk_zone zone; 4486 unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off; 4487 block_t cs_zone_block, wp_block; 4488 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 4489 sector_t zone_sector; 4490 int err; 4491 4492 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 4493 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 4494 4495 zbd = get_target_zoned_dev(sbi, cs_zone_block); 4496 if (!zbd) 4497 return 0; 4498 4499 /* report zone for the sector the curseg points to */ 4500 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 4501 << log_sectors_per_block; 4502 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 4503 report_one_zone_cb, &zone); 4504 if (err != 1) { 4505 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 4506 zbd->path, err); 4507 return err; 4508 } 4509 4510 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4511 return 0; 4512 4513 wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block); 4514 wp_segno = GET_SEGNO(sbi, wp_block); 4515 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno); 4516 wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0); 4517 4518 if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff && 4519 wp_sector_off == 0) 4520 return 0; 4521 4522 f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: " 4523 "curseg[0x%x,0x%x] wp[0x%x,0x%x]", 4524 type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff); 4525 4526 f2fs_notice(sbi, "Assign new section to curseg[%d]: " 4527 "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff); 4528 allocate_segment_by_default(sbi, type, true); 4529 4530 /* check consistency of the zone curseg pointed to */ 4531 if (check_zone_write_pointer(sbi, zbd, &zone)) 4532 return -EIO; 4533 4534 /* check newly assigned zone */ 4535 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 4536 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 4537 4538 zbd = get_target_zoned_dev(sbi, cs_zone_block); 4539 if (!zbd) 4540 return 0; 4541 4542 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 4543 << log_sectors_per_block; 4544 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 4545 report_one_zone_cb, &zone); 4546 if (err != 1) { 4547 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 4548 zbd->path, err); 4549 return err; 4550 } 4551 4552 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4553 return 0; 4554 4555 if (zone.wp != zone.start) { 4556 f2fs_notice(sbi, 4557 "New zone for curseg[%d] is not yet discarded. " 4558 "Reset the zone: curseg[0x%x,0x%x]", 4559 type, cs->segno, cs->next_blkoff); 4560 err = __f2fs_issue_discard_zone(sbi, zbd->bdev, 4561 zone_sector >> log_sectors_per_block, 4562 zone.len >> log_sectors_per_block); 4563 if (err) { 4564 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 4565 zbd->path, err); 4566 return err; 4567 } 4568 } 4569 4570 return 0; 4571 } 4572 4573 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 4574 { 4575 int i, ret; 4576 4577 for (i = 0; i < NO_CHECK_TYPE; i++) { 4578 ret = fix_curseg_write_pointer(sbi, i); 4579 if (ret) 4580 return ret; 4581 } 4582 4583 return 0; 4584 } 4585 4586 struct check_zone_write_pointer_args { 4587 struct f2fs_sb_info *sbi; 4588 struct f2fs_dev_info *fdev; 4589 }; 4590 4591 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx, 4592 void *data) { 4593 struct check_zone_write_pointer_args *args; 4594 args = (struct check_zone_write_pointer_args *)data; 4595 4596 return check_zone_write_pointer(args->sbi, args->fdev, zone); 4597 } 4598 4599 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi) 4600 { 4601 int i, ret; 4602 struct check_zone_write_pointer_args args; 4603 4604 for (i = 0; i < sbi->s_ndevs; i++) { 4605 if (!bdev_is_zoned(FDEV(i).bdev)) 4606 continue; 4607 4608 args.sbi = sbi; 4609 args.fdev = &FDEV(i); 4610 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES, 4611 check_zone_write_pointer_cb, &args); 4612 if (ret < 0) 4613 return ret; 4614 } 4615 4616 return 0; 4617 } 4618 #else 4619 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 4620 { 4621 return 0; 4622 } 4623 4624 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi) 4625 { 4626 return 0; 4627 } 4628 #endif 4629 4630 /* 4631 * Update min, max modified time for cost-benefit GC algorithm 4632 */ 4633 static void init_min_max_mtime(struct f2fs_sb_info *sbi) 4634 { 4635 struct sit_info *sit_i = SIT_I(sbi); 4636 unsigned int segno; 4637 4638 down_write(&sit_i->sentry_lock); 4639 4640 sit_i->min_mtime = ULLONG_MAX; 4641 4642 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 4643 unsigned int i; 4644 unsigned long long mtime = 0; 4645 4646 for (i = 0; i < sbi->segs_per_sec; i++) 4647 mtime += get_seg_entry(sbi, segno + i)->mtime; 4648 4649 mtime = div_u64(mtime, sbi->segs_per_sec); 4650 4651 if (sit_i->min_mtime > mtime) 4652 sit_i->min_mtime = mtime; 4653 } 4654 sit_i->max_mtime = get_mtime(sbi, false); 4655 up_write(&sit_i->sentry_lock); 4656 } 4657 4658 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi) 4659 { 4660 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 4661 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 4662 struct f2fs_sm_info *sm_info; 4663 int err; 4664 4665 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL); 4666 if (!sm_info) 4667 return -ENOMEM; 4668 4669 /* init sm info */ 4670 sbi->sm_info = sm_info; 4671 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 4672 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 4673 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 4674 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 4675 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 4676 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 4677 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 4678 sm_info->rec_prefree_segments = sm_info->main_segments * 4679 DEF_RECLAIM_PREFREE_SEGMENTS / 100; 4680 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS) 4681 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS; 4682 4683 if (!f2fs_lfs_mode(sbi)) 4684 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC; 4685 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; 4686 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; 4687 sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec; 4688 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS; 4689 sm_info->min_ssr_sections = reserved_sections(sbi); 4690 4691 INIT_LIST_HEAD(&sm_info->sit_entry_set); 4692 4693 init_rwsem(&sm_info->curseg_lock); 4694 4695 if (!f2fs_readonly(sbi->sb)) { 4696 err = f2fs_create_flush_cmd_control(sbi); 4697 if (err) 4698 return err; 4699 } 4700 4701 err = create_discard_cmd_control(sbi); 4702 if (err) 4703 return err; 4704 4705 err = build_sit_info(sbi); 4706 if (err) 4707 return err; 4708 err = build_free_segmap(sbi); 4709 if (err) 4710 return err; 4711 err = build_curseg(sbi); 4712 if (err) 4713 return err; 4714 4715 /* reinit free segmap based on SIT */ 4716 err = build_sit_entries(sbi); 4717 if (err) 4718 return err; 4719 4720 init_free_segmap(sbi); 4721 err = build_dirty_segmap(sbi); 4722 if (err) 4723 return err; 4724 4725 err = sanity_check_curseg(sbi); 4726 if (err) 4727 return err; 4728 4729 init_min_max_mtime(sbi); 4730 return 0; 4731 } 4732 4733 static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 4734 enum dirty_type dirty_type) 4735 { 4736 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4737 4738 mutex_lock(&dirty_i->seglist_lock); 4739 kvfree(dirty_i->dirty_segmap[dirty_type]); 4740 dirty_i->nr_dirty[dirty_type] = 0; 4741 mutex_unlock(&dirty_i->seglist_lock); 4742 } 4743 4744 static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 4745 { 4746 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4747 kvfree(dirty_i->victim_secmap); 4748 } 4749 4750 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 4751 { 4752 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4753 int i; 4754 4755 if (!dirty_i) 4756 return; 4757 4758 /* discard pre-free/dirty segments list */ 4759 for (i = 0; i < NR_DIRTY_TYPE; i++) 4760 discard_dirty_segmap(sbi, i); 4761 4762 destroy_victim_secmap(sbi); 4763 SM_I(sbi)->dirty_info = NULL; 4764 kvfree(dirty_i); 4765 } 4766 4767 static void destroy_curseg(struct f2fs_sb_info *sbi) 4768 { 4769 struct curseg_info *array = SM_I(sbi)->curseg_array; 4770 int i; 4771 4772 if (!array) 4773 return; 4774 SM_I(sbi)->curseg_array = NULL; 4775 for (i = 0; i < NR_CURSEG_TYPE; i++) { 4776 kvfree(array[i].sum_blk); 4777 kvfree(array[i].journal); 4778 } 4779 kvfree(array); 4780 } 4781 4782 static void destroy_free_segmap(struct f2fs_sb_info *sbi) 4783 { 4784 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 4785 if (!free_i) 4786 return; 4787 SM_I(sbi)->free_info = NULL; 4788 kvfree(free_i->free_segmap); 4789 kvfree(free_i->free_secmap); 4790 kvfree(free_i); 4791 } 4792 4793 static void destroy_sit_info(struct f2fs_sb_info *sbi) 4794 { 4795 struct sit_info *sit_i = SIT_I(sbi); 4796 4797 if (!sit_i) 4798 return; 4799 4800 if (sit_i->sentries) 4801 kvfree(sit_i->bitmap); 4802 kvfree(sit_i->tmp_map); 4803 4804 kvfree(sit_i->sentries); 4805 kvfree(sit_i->sec_entries); 4806 kvfree(sit_i->dirty_sentries_bitmap); 4807 4808 SM_I(sbi)->sit_info = NULL; 4809 kvfree(sit_i->sit_bitmap); 4810 #ifdef CONFIG_F2FS_CHECK_FS 4811 kvfree(sit_i->sit_bitmap_mir); 4812 kvfree(sit_i->invalid_segmap); 4813 #endif 4814 kvfree(sit_i); 4815 } 4816 4817 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi) 4818 { 4819 struct f2fs_sm_info *sm_info = SM_I(sbi); 4820 4821 if (!sm_info) 4822 return; 4823 f2fs_destroy_flush_cmd_control(sbi, true); 4824 destroy_discard_cmd_control(sbi); 4825 destroy_dirty_segmap(sbi); 4826 destroy_curseg(sbi); 4827 destroy_free_segmap(sbi); 4828 destroy_sit_info(sbi); 4829 sbi->sm_info = NULL; 4830 kvfree(sm_info); 4831 } 4832 4833 int __init f2fs_create_segment_manager_caches(void) 4834 { 4835 discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry", 4836 sizeof(struct discard_entry)); 4837 if (!discard_entry_slab) 4838 goto fail; 4839 4840 discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd", 4841 sizeof(struct discard_cmd)); 4842 if (!discard_cmd_slab) 4843 goto destroy_discard_entry; 4844 4845 sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set", 4846 sizeof(struct sit_entry_set)); 4847 if (!sit_entry_set_slab) 4848 goto destroy_discard_cmd; 4849 4850 inmem_entry_slab = f2fs_kmem_cache_create("f2fs_inmem_page_entry", 4851 sizeof(struct inmem_pages)); 4852 if (!inmem_entry_slab) 4853 goto destroy_sit_entry_set; 4854 return 0; 4855 4856 destroy_sit_entry_set: 4857 kmem_cache_destroy(sit_entry_set_slab); 4858 destroy_discard_cmd: 4859 kmem_cache_destroy(discard_cmd_slab); 4860 destroy_discard_entry: 4861 kmem_cache_destroy(discard_entry_slab); 4862 fail: 4863 return -ENOMEM; 4864 } 4865 4866 void f2fs_destroy_segment_manager_caches(void) 4867 { 4868 kmem_cache_destroy(sit_entry_set_slab); 4869 kmem_cache_destroy(discard_cmd_slab); 4870 kmem_cache_destroy(discard_entry_slab); 4871 kmem_cache_destroy(inmem_entry_slab); 4872 } 4873