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 (test_opt(sbi, LFS)) 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 f2fs_sb_info *sbi = F2FS_I_SB(inode); 189 struct f2fs_inode_info *fi = F2FS_I(inode); 190 struct inmem_pages *new; 191 192 f2fs_trace_pid(page); 193 194 f2fs_set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE); 195 196 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS); 197 198 /* add atomic page indices to the list */ 199 new->page = page; 200 INIT_LIST_HEAD(&new->list); 201 202 /* increase reference count with clean state */ 203 mutex_lock(&fi->inmem_lock); 204 get_page(page); 205 list_add_tail(&new->list, &fi->inmem_pages); 206 spin_lock(&sbi->inode_lock[ATOMIC_FILE]); 207 if (list_empty(&fi->inmem_ilist)) 208 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]); 209 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 210 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); 211 mutex_unlock(&fi->inmem_lock); 212 213 trace_f2fs_register_inmem_page(page, INMEM); 214 } 215 216 static int __revoke_inmem_pages(struct inode *inode, 217 struct list_head *head, bool drop, bool recover, 218 bool trylock) 219 { 220 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 221 struct inmem_pages *cur, *tmp; 222 int err = 0; 223 224 list_for_each_entry_safe(cur, tmp, head, list) { 225 struct page *page = cur->page; 226 227 if (drop) 228 trace_f2fs_commit_inmem_page(page, INMEM_DROP); 229 230 if (trylock) { 231 /* 232 * to avoid deadlock in between page lock and 233 * inmem_lock. 234 */ 235 if (!trylock_page(page)) 236 continue; 237 } else { 238 lock_page(page); 239 } 240 241 f2fs_wait_on_page_writeback(page, DATA, true, true); 242 243 if (recover) { 244 struct dnode_of_data dn; 245 struct node_info ni; 246 247 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE); 248 retry: 249 set_new_dnode(&dn, inode, NULL, NULL, 0); 250 err = f2fs_get_dnode_of_data(&dn, page->index, 251 LOOKUP_NODE); 252 if (err) { 253 if (err == -ENOMEM) { 254 congestion_wait(BLK_RW_ASYNC, HZ/50); 255 cond_resched(); 256 goto retry; 257 } 258 err = -EAGAIN; 259 goto next; 260 } 261 262 err = f2fs_get_node_info(sbi, dn.nid, &ni); 263 if (err) { 264 f2fs_put_dnode(&dn); 265 return err; 266 } 267 268 if (cur->old_addr == NEW_ADDR) { 269 f2fs_invalidate_blocks(sbi, dn.data_blkaddr); 270 f2fs_update_data_blkaddr(&dn, NEW_ADDR); 271 } else 272 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, 273 cur->old_addr, ni.version, true, true); 274 f2fs_put_dnode(&dn); 275 } 276 next: 277 /* we don't need to invalidate this in the sccessful status */ 278 if (drop || recover) { 279 ClearPageUptodate(page); 280 clear_cold_data(page); 281 } 282 f2fs_clear_page_private(page); 283 f2fs_put_page(page, 1); 284 285 list_del(&cur->list); 286 kmem_cache_free(inmem_entry_slab, cur); 287 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); 288 } 289 return err; 290 } 291 292 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure) 293 { 294 struct list_head *head = &sbi->inode_list[ATOMIC_FILE]; 295 struct inode *inode; 296 struct f2fs_inode_info *fi; 297 next: 298 spin_lock(&sbi->inode_lock[ATOMIC_FILE]); 299 if (list_empty(head)) { 300 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 301 return; 302 } 303 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist); 304 inode = igrab(&fi->vfs_inode); 305 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 306 307 if (inode) { 308 if (gc_failure) { 309 if (fi->i_gc_failures[GC_FAILURE_ATOMIC]) 310 goto drop; 311 goto skip; 312 } 313 drop: 314 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST); 315 f2fs_drop_inmem_pages(inode); 316 iput(inode); 317 } 318 skip: 319 congestion_wait(BLK_RW_ASYNC, HZ/50); 320 cond_resched(); 321 goto next; 322 } 323 324 void f2fs_drop_inmem_pages(struct inode *inode) 325 { 326 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 327 struct f2fs_inode_info *fi = F2FS_I(inode); 328 329 while (!list_empty(&fi->inmem_pages)) { 330 mutex_lock(&fi->inmem_lock); 331 __revoke_inmem_pages(inode, &fi->inmem_pages, 332 true, false, true); 333 334 if (list_empty(&fi->inmem_pages)) { 335 spin_lock(&sbi->inode_lock[ATOMIC_FILE]); 336 if (!list_empty(&fi->inmem_ilist)) 337 list_del_init(&fi->inmem_ilist); 338 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 339 } 340 mutex_unlock(&fi->inmem_lock); 341 } 342 343 clear_inode_flag(inode, FI_ATOMIC_FILE); 344 fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0; 345 stat_dec_atomic_write(inode); 346 } 347 348 void f2fs_drop_inmem_page(struct inode *inode, struct page *page) 349 { 350 struct f2fs_inode_info *fi = F2FS_I(inode); 351 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 352 struct list_head *head = &fi->inmem_pages; 353 struct inmem_pages *cur = NULL; 354 355 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page)); 356 357 mutex_lock(&fi->inmem_lock); 358 list_for_each_entry(cur, head, list) { 359 if (cur->page == page) 360 break; 361 } 362 363 f2fs_bug_on(sbi, list_empty(head) || cur->page != page); 364 list_del(&cur->list); 365 mutex_unlock(&fi->inmem_lock); 366 367 dec_page_count(sbi, F2FS_INMEM_PAGES); 368 kmem_cache_free(inmem_entry_slab, cur); 369 370 ClearPageUptodate(page); 371 f2fs_clear_page_private(page); 372 f2fs_put_page(page, 0); 373 374 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE); 375 } 376 377 static int __f2fs_commit_inmem_pages(struct inode *inode) 378 { 379 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 380 struct f2fs_inode_info *fi = F2FS_I(inode); 381 struct inmem_pages *cur, *tmp; 382 struct f2fs_io_info fio = { 383 .sbi = sbi, 384 .ino = inode->i_ino, 385 .type = DATA, 386 .op = REQ_OP_WRITE, 387 .op_flags = REQ_SYNC | REQ_PRIO, 388 .io_type = FS_DATA_IO, 389 }; 390 struct list_head revoke_list; 391 bool submit_bio = false; 392 int err = 0; 393 394 INIT_LIST_HEAD(&revoke_list); 395 396 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) { 397 struct page *page = cur->page; 398 399 lock_page(page); 400 if (page->mapping == inode->i_mapping) { 401 trace_f2fs_commit_inmem_page(page, INMEM); 402 403 f2fs_wait_on_page_writeback(page, DATA, true, true); 404 405 set_page_dirty(page); 406 if (clear_page_dirty_for_io(page)) { 407 inode_dec_dirty_pages(inode); 408 f2fs_remove_dirty_inode(inode); 409 } 410 retry: 411 fio.page = page; 412 fio.old_blkaddr = NULL_ADDR; 413 fio.encrypted_page = NULL; 414 fio.need_lock = LOCK_DONE; 415 err = f2fs_do_write_data_page(&fio); 416 if (err) { 417 if (err == -ENOMEM) { 418 congestion_wait(BLK_RW_ASYNC, HZ/50); 419 cond_resched(); 420 goto retry; 421 } 422 unlock_page(page); 423 break; 424 } 425 /* record old blkaddr for revoking */ 426 cur->old_addr = fio.old_blkaddr; 427 submit_bio = true; 428 } 429 unlock_page(page); 430 list_move_tail(&cur->list, &revoke_list); 431 } 432 433 if (submit_bio) 434 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA); 435 436 if (err) { 437 /* 438 * try to revoke all committed pages, but still we could fail 439 * due to no memory or other reason, if that happened, EAGAIN 440 * will be returned, which means in such case, transaction is 441 * already not integrity, caller should use journal to do the 442 * recovery or rewrite & commit last transaction. For other 443 * error number, revoking was done by filesystem itself. 444 */ 445 err = __revoke_inmem_pages(inode, &revoke_list, 446 false, true, false); 447 448 /* drop all uncommitted pages */ 449 __revoke_inmem_pages(inode, &fi->inmem_pages, 450 true, false, false); 451 } else { 452 __revoke_inmem_pages(inode, &revoke_list, 453 false, false, false); 454 } 455 456 return err; 457 } 458 459 int f2fs_commit_inmem_pages(struct inode *inode) 460 { 461 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 462 struct f2fs_inode_info *fi = F2FS_I(inode); 463 int err; 464 465 f2fs_balance_fs(sbi, true); 466 467 down_write(&fi->i_gc_rwsem[WRITE]); 468 469 f2fs_lock_op(sbi); 470 set_inode_flag(inode, FI_ATOMIC_COMMIT); 471 472 mutex_lock(&fi->inmem_lock); 473 err = __f2fs_commit_inmem_pages(inode); 474 475 spin_lock(&sbi->inode_lock[ATOMIC_FILE]); 476 if (!list_empty(&fi->inmem_ilist)) 477 list_del_init(&fi->inmem_ilist); 478 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 479 mutex_unlock(&fi->inmem_lock); 480 481 clear_inode_flag(inode, FI_ATOMIC_COMMIT); 482 483 f2fs_unlock_op(sbi); 484 up_write(&fi->i_gc_rwsem[WRITE]); 485 486 return err; 487 } 488 489 /* 490 * This function balances dirty node and dentry pages. 491 * In addition, it controls garbage collection. 492 */ 493 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need) 494 { 495 if (time_to_inject(sbi, FAULT_CHECKPOINT)) { 496 f2fs_show_injection_info(FAULT_CHECKPOINT); 497 f2fs_stop_checkpoint(sbi, false); 498 } 499 500 /* balance_fs_bg is able to be pending */ 501 if (need && excess_cached_nats(sbi)) 502 f2fs_balance_fs_bg(sbi); 503 504 if (f2fs_is_checkpoint_ready(sbi)) 505 return; 506 507 /* 508 * We should do GC or end up with checkpoint, if there are so many dirty 509 * dir/node pages without enough free segments. 510 */ 511 if (has_not_enough_free_secs(sbi, 0, 0)) { 512 mutex_lock(&sbi->gc_mutex); 513 f2fs_gc(sbi, false, false, NULL_SEGNO); 514 } 515 } 516 517 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi) 518 { 519 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 520 return; 521 522 /* try to shrink extent cache when there is no enough memory */ 523 if (!f2fs_available_free_memory(sbi, EXTENT_CACHE)) 524 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER); 525 526 /* check the # of cached NAT entries */ 527 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES)) 528 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK); 529 530 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) 531 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS); 532 else 533 f2fs_build_free_nids(sbi, false, false); 534 535 if (!is_idle(sbi, REQ_TIME) && 536 (!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi))) 537 return; 538 539 /* checkpoint is the only way to shrink partial cached entries */ 540 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) || 541 !f2fs_available_free_memory(sbi, INO_ENTRIES) || 542 excess_prefree_segs(sbi) || 543 excess_dirty_nats(sbi) || 544 excess_dirty_nodes(sbi) || 545 f2fs_time_over(sbi, CP_TIME)) { 546 if (test_opt(sbi, DATA_FLUSH)) { 547 struct blk_plug plug; 548 549 mutex_lock(&sbi->flush_lock); 550 551 blk_start_plug(&plug); 552 f2fs_sync_dirty_inodes(sbi, FILE_INODE); 553 blk_finish_plug(&plug); 554 555 mutex_unlock(&sbi->flush_lock); 556 } 557 f2fs_sync_fs(sbi->sb, true); 558 stat_inc_bg_cp_count(sbi->stat_info); 559 } 560 } 561 562 static int __submit_flush_wait(struct f2fs_sb_info *sbi, 563 struct block_device *bdev) 564 { 565 struct bio *bio; 566 int ret; 567 568 bio = f2fs_bio_alloc(sbi, 0, false); 569 if (!bio) 570 return -ENOMEM; 571 572 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH; 573 bio_set_dev(bio, bdev); 574 ret = submit_bio_wait(bio); 575 bio_put(bio); 576 577 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER), 578 test_opt(sbi, FLUSH_MERGE), ret); 579 return ret; 580 } 581 582 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino) 583 { 584 int ret = 0; 585 int i; 586 587 if (!f2fs_is_multi_device(sbi)) 588 return __submit_flush_wait(sbi, sbi->sb->s_bdev); 589 590 for (i = 0; i < sbi->s_ndevs; i++) { 591 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO)) 592 continue; 593 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 594 if (ret) 595 break; 596 } 597 return ret; 598 } 599 600 static int issue_flush_thread(void *data) 601 { 602 struct f2fs_sb_info *sbi = data; 603 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 604 wait_queue_head_t *q = &fcc->flush_wait_queue; 605 repeat: 606 if (kthread_should_stop()) 607 return 0; 608 609 sb_start_intwrite(sbi->sb); 610 611 if (!llist_empty(&fcc->issue_list)) { 612 struct flush_cmd *cmd, *next; 613 int ret; 614 615 fcc->dispatch_list = llist_del_all(&fcc->issue_list); 616 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); 617 618 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode); 619 620 ret = submit_flush_wait(sbi, cmd->ino); 621 atomic_inc(&fcc->issued_flush); 622 623 llist_for_each_entry_safe(cmd, next, 624 fcc->dispatch_list, llnode) { 625 cmd->ret = ret; 626 complete(&cmd->wait); 627 } 628 fcc->dispatch_list = NULL; 629 } 630 631 sb_end_intwrite(sbi->sb); 632 633 wait_event_interruptible(*q, 634 kthread_should_stop() || !llist_empty(&fcc->issue_list)); 635 goto repeat; 636 } 637 638 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino) 639 { 640 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 641 struct flush_cmd cmd; 642 int ret; 643 644 if (test_opt(sbi, NOBARRIER)) 645 return 0; 646 647 if (!test_opt(sbi, FLUSH_MERGE)) { 648 atomic_inc(&fcc->queued_flush); 649 ret = submit_flush_wait(sbi, ino); 650 atomic_dec(&fcc->queued_flush); 651 atomic_inc(&fcc->issued_flush); 652 return ret; 653 } 654 655 if (atomic_inc_return(&fcc->queued_flush) == 1 || 656 f2fs_is_multi_device(sbi)) { 657 ret = submit_flush_wait(sbi, ino); 658 atomic_dec(&fcc->queued_flush); 659 660 atomic_inc(&fcc->issued_flush); 661 return ret; 662 } 663 664 cmd.ino = ino; 665 init_completion(&cmd.wait); 666 667 llist_add(&cmd.llnode, &fcc->issue_list); 668 669 /* update issue_list before we wake up issue_flush thread */ 670 smp_mb(); 671 672 if (waitqueue_active(&fcc->flush_wait_queue)) 673 wake_up(&fcc->flush_wait_queue); 674 675 if (fcc->f2fs_issue_flush) { 676 wait_for_completion(&cmd.wait); 677 atomic_dec(&fcc->queued_flush); 678 } else { 679 struct llist_node *list; 680 681 list = llist_del_all(&fcc->issue_list); 682 if (!list) { 683 wait_for_completion(&cmd.wait); 684 atomic_dec(&fcc->queued_flush); 685 } else { 686 struct flush_cmd *tmp, *next; 687 688 ret = submit_flush_wait(sbi, ino); 689 690 llist_for_each_entry_safe(tmp, next, list, llnode) { 691 if (tmp == &cmd) { 692 cmd.ret = ret; 693 atomic_dec(&fcc->queued_flush); 694 continue; 695 } 696 tmp->ret = ret; 697 complete(&tmp->wait); 698 } 699 } 700 } 701 702 return cmd.ret; 703 } 704 705 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi) 706 { 707 dev_t dev = sbi->sb->s_bdev->bd_dev; 708 struct flush_cmd_control *fcc; 709 int err = 0; 710 711 if (SM_I(sbi)->fcc_info) { 712 fcc = SM_I(sbi)->fcc_info; 713 if (fcc->f2fs_issue_flush) 714 return err; 715 goto init_thread; 716 } 717 718 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL); 719 if (!fcc) 720 return -ENOMEM; 721 atomic_set(&fcc->issued_flush, 0); 722 atomic_set(&fcc->queued_flush, 0); 723 init_waitqueue_head(&fcc->flush_wait_queue); 724 init_llist_head(&fcc->issue_list); 725 SM_I(sbi)->fcc_info = fcc; 726 if (!test_opt(sbi, FLUSH_MERGE)) 727 return err; 728 729 init_thread: 730 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, 731 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); 732 if (IS_ERR(fcc->f2fs_issue_flush)) { 733 err = PTR_ERR(fcc->f2fs_issue_flush); 734 kvfree(fcc); 735 SM_I(sbi)->fcc_info = NULL; 736 return err; 737 } 738 739 return err; 740 } 741 742 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free) 743 { 744 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 745 746 if (fcc && fcc->f2fs_issue_flush) { 747 struct task_struct *flush_thread = fcc->f2fs_issue_flush; 748 749 fcc->f2fs_issue_flush = NULL; 750 kthread_stop(flush_thread); 751 } 752 if (free) { 753 kvfree(fcc); 754 SM_I(sbi)->fcc_info = NULL; 755 } 756 } 757 758 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi) 759 { 760 int ret = 0, i; 761 762 if (!f2fs_is_multi_device(sbi)) 763 return 0; 764 765 for (i = 1; i < sbi->s_ndevs; i++) { 766 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device)) 767 continue; 768 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 769 if (ret) 770 break; 771 772 spin_lock(&sbi->dev_lock); 773 f2fs_clear_bit(i, (char *)&sbi->dirty_device); 774 spin_unlock(&sbi->dev_lock); 775 } 776 777 return ret; 778 } 779 780 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 781 enum dirty_type dirty_type) 782 { 783 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 784 785 /* need not be added */ 786 if (IS_CURSEG(sbi, segno)) 787 return; 788 789 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 790 dirty_i->nr_dirty[dirty_type]++; 791 792 if (dirty_type == DIRTY) { 793 struct seg_entry *sentry = get_seg_entry(sbi, segno); 794 enum dirty_type t = sentry->type; 795 796 if (unlikely(t >= DIRTY)) { 797 f2fs_bug_on(sbi, 1); 798 return; 799 } 800 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) 801 dirty_i->nr_dirty[t]++; 802 } 803 } 804 805 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 806 enum dirty_type dirty_type) 807 { 808 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 809 810 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) 811 dirty_i->nr_dirty[dirty_type]--; 812 813 if (dirty_type == DIRTY) { 814 struct seg_entry *sentry = get_seg_entry(sbi, segno); 815 enum dirty_type t = sentry->type; 816 817 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) 818 dirty_i->nr_dirty[t]--; 819 820 if (get_valid_blocks(sbi, segno, true) == 0) 821 clear_bit(GET_SEC_FROM_SEG(sbi, segno), 822 dirty_i->victim_secmap); 823 } 824 } 825 826 /* 827 * Should not occur error such as -ENOMEM. 828 * Adding dirty entry into seglist is not critical operation. 829 * If a given segment is one of current working segments, it won't be added. 830 */ 831 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) 832 { 833 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 834 unsigned short valid_blocks, ckpt_valid_blocks; 835 836 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) 837 return; 838 839 mutex_lock(&dirty_i->seglist_lock); 840 841 valid_blocks = get_valid_blocks(sbi, segno, false); 842 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno); 843 844 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) || 845 ckpt_valid_blocks == sbi->blocks_per_seg)) { 846 __locate_dirty_segment(sbi, segno, PRE); 847 __remove_dirty_segment(sbi, segno, DIRTY); 848 } else if (valid_blocks < sbi->blocks_per_seg) { 849 __locate_dirty_segment(sbi, segno, DIRTY); 850 } else { 851 /* Recovery routine with SSR needs this */ 852 __remove_dirty_segment(sbi, segno, DIRTY); 853 } 854 855 mutex_unlock(&dirty_i->seglist_lock); 856 } 857 858 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */ 859 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi) 860 { 861 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 862 unsigned int segno; 863 864 mutex_lock(&dirty_i->seglist_lock); 865 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 866 if (get_valid_blocks(sbi, segno, false)) 867 continue; 868 if (IS_CURSEG(sbi, segno)) 869 continue; 870 __locate_dirty_segment(sbi, segno, PRE); 871 __remove_dirty_segment(sbi, segno, DIRTY); 872 } 873 mutex_unlock(&dirty_i->seglist_lock); 874 } 875 876 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi) 877 { 878 int ovp_hole_segs = 879 (overprovision_segments(sbi) - reserved_segments(sbi)); 880 block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg; 881 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 882 block_t holes[2] = {0, 0}; /* DATA and NODE */ 883 block_t unusable; 884 struct seg_entry *se; 885 unsigned int segno; 886 887 mutex_lock(&dirty_i->seglist_lock); 888 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 889 se = get_seg_entry(sbi, segno); 890 if (IS_NODESEG(se->type)) 891 holes[NODE] += sbi->blocks_per_seg - se->valid_blocks; 892 else 893 holes[DATA] += sbi->blocks_per_seg - se->valid_blocks; 894 } 895 mutex_unlock(&dirty_i->seglist_lock); 896 897 unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE]; 898 if (unusable > ovp_holes) 899 return unusable - ovp_holes; 900 return 0; 901 } 902 903 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable) 904 { 905 int ovp_hole_segs = 906 (overprovision_segments(sbi) - reserved_segments(sbi)); 907 if (unusable > F2FS_OPTION(sbi).unusable_cap) 908 return -EAGAIN; 909 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) && 910 dirty_segments(sbi) > ovp_hole_segs) 911 return -EAGAIN; 912 return 0; 913 } 914 915 /* This is only used by SBI_CP_DISABLED */ 916 static unsigned int get_free_segment(struct f2fs_sb_info *sbi) 917 { 918 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 919 unsigned int segno = 0; 920 921 mutex_lock(&dirty_i->seglist_lock); 922 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 923 if (get_valid_blocks(sbi, segno, false)) 924 continue; 925 if (get_ckpt_valid_blocks(sbi, segno)) 926 continue; 927 mutex_unlock(&dirty_i->seglist_lock); 928 return segno; 929 } 930 mutex_unlock(&dirty_i->seglist_lock); 931 return NULL_SEGNO; 932 } 933 934 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi, 935 struct block_device *bdev, block_t lstart, 936 block_t start, block_t len) 937 { 938 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 939 struct list_head *pend_list; 940 struct discard_cmd *dc; 941 942 f2fs_bug_on(sbi, !len); 943 944 pend_list = &dcc->pend_list[plist_idx(len)]; 945 946 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS); 947 INIT_LIST_HEAD(&dc->list); 948 dc->bdev = bdev; 949 dc->lstart = lstart; 950 dc->start = start; 951 dc->len = len; 952 dc->ref = 0; 953 dc->state = D_PREP; 954 dc->queued = 0; 955 dc->error = 0; 956 init_completion(&dc->wait); 957 list_add_tail(&dc->list, pend_list); 958 spin_lock_init(&dc->lock); 959 dc->bio_ref = 0; 960 atomic_inc(&dcc->discard_cmd_cnt); 961 dcc->undiscard_blks += len; 962 963 return dc; 964 } 965 966 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi, 967 struct block_device *bdev, block_t lstart, 968 block_t start, block_t len, 969 struct rb_node *parent, struct rb_node **p, 970 bool leftmost) 971 { 972 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 973 struct discard_cmd *dc; 974 975 dc = __create_discard_cmd(sbi, bdev, lstart, start, len); 976 977 rb_link_node(&dc->rb_node, parent, p); 978 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost); 979 980 return dc; 981 } 982 983 static void __detach_discard_cmd(struct discard_cmd_control *dcc, 984 struct discard_cmd *dc) 985 { 986 if (dc->state == D_DONE) 987 atomic_sub(dc->queued, &dcc->queued_discard); 988 989 list_del(&dc->list); 990 rb_erase_cached(&dc->rb_node, &dcc->root); 991 dcc->undiscard_blks -= dc->len; 992 993 kmem_cache_free(discard_cmd_slab, dc); 994 995 atomic_dec(&dcc->discard_cmd_cnt); 996 } 997 998 static void __remove_discard_cmd(struct f2fs_sb_info *sbi, 999 struct discard_cmd *dc) 1000 { 1001 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1002 unsigned long flags; 1003 1004 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len); 1005 1006 spin_lock_irqsave(&dc->lock, flags); 1007 if (dc->bio_ref) { 1008 spin_unlock_irqrestore(&dc->lock, flags); 1009 return; 1010 } 1011 spin_unlock_irqrestore(&dc->lock, flags); 1012 1013 f2fs_bug_on(sbi, dc->ref); 1014 1015 if (dc->error == -EOPNOTSUPP) 1016 dc->error = 0; 1017 1018 if (dc->error) 1019 printk_ratelimited( 1020 "%sF2FS-fs: Issue discard(%u, %u, %u) failed, ret: %d", 1021 KERN_INFO, dc->lstart, dc->start, dc->len, dc->error); 1022 __detach_discard_cmd(dcc, dc); 1023 } 1024 1025 static void f2fs_submit_discard_endio(struct bio *bio) 1026 { 1027 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private; 1028 unsigned long flags; 1029 1030 dc->error = blk_status_to_errno(bio->bi_status); 1031 1032 spin_lock_irqsave(&dc->lock, flags); 1033 dc->bio_ref--; 1034 if (!dc->bio_ref && dc->state == D_SUBMIT) { 1035 dc->state = D_DONE; 1036 complete_all(&dc->wait); 1037 } 1038 spin_unlock_irqrestore(&dc->lock, flags); 1039 bio_put(bio); 1040 } 1041 1042 static void __check_sit_bitmap(struct f2fs_sb_info *sbi, 1043 block_t start, block_t end) 1044 { 1045 #ifdef CONFIG_F2FS_CHECK_FS 1046 struct seg_entry *sentry; 1047 unsigned int segno; 1048 block_t blk = start; 1049 unsigned long offset, size, max_blocks = sbi->blocks_per_seg; 1050 unsigned long *map; 1051 1052 while (blk < end) { 1053 segno = GET_SEGNO(sbi, blk); 1054 sentry = get_seg_entry(sbi, segno); 1055 offset = GET_BLKOFF_FROM_SEG0(sbi, blk); 1056 1057 if (end < START_BLOCK(sbi, segno + 1)) 1058 size = GET_BLKOFF_FROM_SEG0(sbi, end); 1059 else 1060 size = max_blocks; 1061 map = (unsigned long *)(sentry->cur_valid_map); 1062 offset = __find_rev_next_bit(map, size, offset); 1063 f2fs_bug_on(sbi, offset != size); 1064 blk = START_BLOCK(sbi, segno + 1); 1065 } 1066 #endif 1067 } 1068 1069 static void __init_discard_policy(struct f2fs_sb_info *sbi, 1070 struct discard_policy *dpolicy, 1071 int discard_type, unsigned int granularity) 1072 { 1073 /* common policy */ 1074 dpolicy->type = discard_type; 1075 dpolicy->sync = true; 1076 dpolicy->ordered = false; 1077 dpolicy->granularity = granularity; 1078 1079 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST; 1080 dpolicy->io_aware_gran = MAX_PLIST_NUM; 1081 dpolicy->timeout = 0; 1082 1083 if (discard_type == DPOLICY_BG) { 1084 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME; 1085 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME; 1086 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME; 1087 dpolicy->io_aware = true; 1088 dpolicy->sync = false; 1089 dpolicy->ordered = true; 1090 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) { 1091 dpolicy->granularity = 1; 1092 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME; 1093 } 1094 } else if (discard_type == DPOLICY_FORCE) { 1095 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME; 1096 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME; 1097 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME; 1098 dpolicy->io_aware = false; 1099 } else if (discard_type == DPOLICY_FSTRIM) { 1100 dpolicy->io_aware = false; 1101 } else if (discard_type == DPOLICY_UMOUNT) { 1102 dpolicy->max_requests = UINT_MAX; 1103 dpolicy->io_aware = false; 1104 /* we need to issue all to keep CP_TRIMMED_FLAG */ 1105 dpolicy->granularity = 1; 1106 } 1107 } 1108 1109 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1110 struct block_device *bdev, block_t lstart, 1111 block_t start, block_t len); 1112 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */ 1113 static int __submit_discard_cmd(struct f2fs_sb_info *sbi, 1114 struct discard_policy *dpolicy, 1115 struct discard_cmd *dc, 1116 unsigned int *issued) 1117 { 1118 struct block_device *bdev = dc->bdev; 1119 struct request_queue *q = bdev_get_queue(bdev); 1120 unsigned int max_discard_blocks = 1121 SECTOR_TO_BLOCK(q->limits.max_discard_sectors); 1122 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1123 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1124 &(dcc->fstrim_list) : &(dcc->wait_list); 1125 int flag = dpolicy->sync ? REQ_SYNC : 0; 1126 block_t lstart, start, len, total_len; 1127 int err = 0; 1128 1129 if (dc->state != D_PREP) 1130 return 0; 1131 1132 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 1133 return 0; 1134 1135 trace_f2fs_issue_discard(bdev, dc->start, dc->len); 1136 1137 lstart = dc->lstart; 1138 start = dc->start; 1139 len = dc->len; 1140 total_len = len; 1141 1142 dc->len = 0; 1143 1144 while (total_len && *issued < dpolicy->max_requests && !err) { 1145 struct bio *bio = NULL; 1146 unsigned long flags; 1147 bool last = true; 1148 1149 if (len > max_discard_blocks) { 1150 len = max_discard_blocks; 1151 last = false; 1152 } 1153 1154 (*issued)++; 1155 if (*issued == dpolicy->max_requests) 1156 last = true; 1157 1158 dc->len += len; 1159 1160 if (time_to_inject(sbi, FAULT_DISCARD)) { 1161 f2fs_show_injection_info(FAULT_DISCARD); 1162 err = -EIO; 1163 goto submit; 1164 } 1165 err = __blkdev_issue_discard(bdev, 1166 SECTOR_FROM_BLOCK(start), 1167 SECTOR_FROM_BLOCK(len), 1168 GFP_NOFS, 0, &bio); 1169 submit: 1170 if (err) { 1171 spin_lock_irqsave(&dc->lock, flags); 1172 if (dc->state == D_PARTIAL) 1173 dc->state = D_SUBMIT; 1174 spin_unlock_irqrestore(&dc->lock, flags); 1175 1176 break; 1177 } 1178 1179 f2fs_bug_on(sbi, !bio); 1180 1181 /* 1182 * should keep before submission to avoid D_DONE 1183 * right away 1184 */ 1185 spin_lock_irqsave(&dc->lock, flags); 1186 if (last) 1187 dc->state = D_SUBMIT; 1188 else 1189 dc->state = D_PARTIAL; 1190 dc->bio_ref++; 1191 spin_unlock_irqrestore(&dc->lock, flags); 1192 1193 atomic_inc(&dcc->queued_discard); 1194 dc->queued++; 1195 list_move_tail(&dc->list, wait_list); 1196 1197 /* sanity check on discard range */ 1198 __check_sit_bitmap(sbi, lstart, lstart + len); 1199 1200 bio->bi_private = dc; 1201 bio->bi_end_io = f2fs_submit_discard_endio; 1202 bio->bi_opf |= flag; 1203 submit_bio(bio); 1204 1205 atomic_inc(&dcc->issued_discard); 1206 1207 f2fs_update_iostat(sbi, FS_DISCARD, 1); 1208 1209 lstart += len; 1210 start += len; 1211 total_len -= len; 1212 len = total_len; 1213 } 1214 1215 if (!err && len) 1216 __update_discard_tree_range(sbi, bdev, lstart, start, len); 1217 return err; 1218 } 1219 1220 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi, 1221 struct block_device *bdev, block_t lstart, 1222 block_t start, block_t len, 1223 struct rb_node **insert_p, 1224 struct rb_node *insert_parent) 1225 { 1226 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1227 struct rb_node **p; 1228 struct rb_node *parent = NULL; 1229 struct discard_cmd *dc = NULL; 1230 bool leftmost = true; 1231 1232 if (insert_p && insert_parent) { 1233 parent = insert_parent; 1234 p = insert_p; 1235 goto do_insert; 1236 } 1237 1238 p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, 1239 lstart, &leftmost); 1240 do_insert: 1241 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, 1242 p, leftmost); 1243 if (!dc) 1244 return NULL; 1245 1246 return dc; 1247 } 1248 1249 static void __relocate_discard_cmd(struct discard_cmd_control *dcc, 1250 struct discard_cmd *dc) 1251 { 1252 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]); 1253 } 1254 1255 static void __punch_discard_cmd(struct f2fs_sb_info *sbi, 1256 struct discard_cmd *dc, block_t blkaddr) 1257 { 1258 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1259 struct discard_info di = dc->di; 1260 bool modified = false; 1261 1262 if (dc->state == D_DONE || dc->len == 1) { 1263 __remove_discard_cmd(sbi, dc); 1264 return; 1265 } 1266 1267 dcc->undiscard_blks -= di.len; 1268 1269 if (blkaddr > di.lstart) { 1270 dc->len = blkaddr - dc->lstart; 1271 dcc->undiscard_blks += dc->len; 1272 __relocate_discard_cmd(dcc, dc); 1273 modified = true; 1274 } 1275 1276 if (blkaddr < di.lstart + di.len - 1) { 1277 if (modified) { 1278 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1, 1279 di.start + blkaddr + 1 - di.lstart, 1280 di.lstart + di.len - 1 - blkaddr, 1281 NULL, NULL); 1282 } else { 1283 dc->lstart++; 1284 dc->len--; 1285 dc->start++; 1286 dcc->undiscard_blks += dc->len; 1287 __relocate_discard_cmd(dcc, dc); 1288 } 1289 } 1290 } 1291 1292 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1293 struct block_device *bdev, block_t lstart, 1294 block_t start, block_t len) 1295 { 1296 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1297 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1298 struct discard_cmd *dc; 1299 struct discard_info di = {0}; 1300 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1301 struct request_queue *q = bdev_get_queue(bdev); 1302 unsigned int max_discard_blocks = 1303 SECTOR_TO_BLOCK(q->limits.max_discard_sectors); 1304 block_t end = lstart + len; 1305 1306 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root, 1307 NULL, lstart, 1308 (struct rb_entry **)&prev_dc, 1309 (struct rb_entry **)&next_dc, 1310 &insert_p, &insert_parent, true, NULL); 1311 if (dc) 1312 prev_dc = dc; 1313 1314 if (!prev_dc) { 1315 di.lstart = lstart; 1316 di.len = next_dc ? next_dc->lstart - lstart : len; 1317 di.len = min(di.len, len); 1318 di.start = start; 1319 } 1320 1321 while (1) { 1322 struct rb_node *node; 1323 bool merged = false; 1324 struct discard_cmd *tdc = NULL; 1325 1326 if (prev_dc) { 1327 di.lstart = prev_dc->lstart + prev_dc->len; 1328 if (di.lstart < lstart) 1329 di.lstart = lstart; 1330 if (di.lstart >= end) 1331 break; 1332 1333 if (!next_dc || next_dc->lstart > end) 1334 di.len = end - di.lstart; 1335 else 1336 di.len = next_dc->lstart - di.lstart; 1337 di.start = start + di.lstart - lstart; 1338 } 1339 1340 if (!di.len) 1341 goto next; 1342 1343 if (prev_dc && prev_dc->state == D_PREP && 1344 prev_dc->bdev == bdev && 1345 __is_discard_back_mergeable(&di, &prev_dc->di, 1346 max_discard_blocks)) { 1347 prev_dc->di.len += di.len; 1348 dcc->undiscard_blks += di.len; 1349 __relocate_discard_cmd(dcc, prev_dc); 1350 di = prev_dc->di; 1351 tdc = prev_dc; 1352 merged = true; 1353 } 1354 1355 if (next_dc && next_dc->state == D_PREP && 1356 next_dc->bdev == bdev && 1357 __is_discard_front_mergeable(&di, &next_dc->di, 1358 max_discard_blocks)) { 1359 next_dc->di.lstart = di.lstart; 1360 next_dc->di.len += di.len; 1361 next_dc->di.start = di.start; 1362 dcc->undiscard_blks += di.len; 1363 __relocate_discard_cmd(dcc, next_dc); 1364 if (tdc) 1365 __remove_discard_cmd(sbi, tdc); 1366 merged = true; 1367 } 1368 1369 if (!merged) { 1370 __insert_discard_tree(sbi, bdev, di.lstart, di.start, 1371 di.len, NULL, NULL); 1372 } 1373 next: 1374 prev_dc = next_dc; 1375 if (!prev_dc) 1376 break; 1377 1378 node = rb_next(&prev_dc->rb_node); 1379 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1380 } 1381 } 1382 1383 static int __queue_discard_cmd(struct f2fs_sb_info *sbi, 1384 struct block_device *bdev, block_t blkstart, block_t blklen) 1385 { 1386 block_t lblkstart = blkstart; 1387 1388 if (!f2fs_bdev_support_discard(bdev)) 1389 return 0; 1390 1391 trace_f2fs_queue_discard(bdev, blkstart, blklen); 1392 1393 if (f2fs_is_multi_device(sbi)) { 1394 int devi = f2fs_target_device_index(sbi, blkstart); 1395 1396 blkstart -= FDEV(devi).start_blk; 1397 } 1398 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock); 1399 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen); 1400 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock); 1401 return 0; 1402 } 1403 1404 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi, 1405 struct discard_policy *dpolicy) 1406 { 1407 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1408 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1409 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1410 struct discard_cmd *dc; 1411 struct blk_plug plug; 1412 unsigned int pos = dcc->next_pos; 1413 unsigned int issued = 0; 1414 bool io_interrupted = false; 1415 1416 mutex_lock(&dcc->cmd_lock); 1417 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root, 1418 NULL, pos, 1419 (struct rb_entry **)&prev_dc, 1420 (struct rb_entry **)&next_dc, 1421 &insert_p, &insert_parent, true, NULL); 1422 if (!dc) 1423 dc = next_dc; 1424 1425 blk_start_plug(&plug); 1426 1427 while (dc) { 1428 struct rb_node *node; 1429 int err = 0; 1430 1431 if (dc->state != D_PREP) 1432 goto next; 1433 1434 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) { 1435 io_interrupted = true; 1436 break; 1437 } 1438 1439 dcc->next_pos = dc->lstart + dc->len; 1440 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued); 1441 1442 if (issued >= dpolicy->max_requests) 1443 break; 1444 next: 1445 node = rb_next(&dc->rb_node); 1446 if (err) 1447 __remove_discard_cmd(sbi, dc); 1448 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1449 } 1450 1451 blk_finish_plug(&plug); 1452 1453 if (!dc) 1454 dcc->next_pos = 0; 1455 1456 mutex_unlock(&dcc->cmd_lock); 1457 1458 if (!issued && io_interrupted) 1459 issued = -1; 1460 1461 return issued; 1462 } 1463 1464 static int __issue_discard_cmd(struct f2fs_sb_info *sbi, 1465 struct discard_policy *dpolicy) 1466 { 1467 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1468 struct list_head *pend_list; 1469 struct discard_cmd *dc, *tmp; 1470 struct blk_plug plug; 1471 int i, issued = 0; 1472 bool io_interrupted = false; 1473 1474 if (dpolicy->timeout != 0) 1475 f2fs_update_time(sbi, dpolicy->timeout); 1476 1477 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1478 if (dpolicy->timeout != 0 && 1479 f2fs_time_over(sbi, dpolicy->timeout)) 1480 break; 1481 1482 if (i + 1 < dpolicy->granularity) 1483 break; 1484 1485 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered) 1486 return __issue_discard_cmd_orderly(sbi, dpolicy); 1487 1488 pend_list = &dcc->pend_list[i]; 1489 1490 mutex_lock(&dcc->cmd_lock); 1491 if (list_empty(pend_list)) 1492 goto next; 1493 if (unlikely(dcc->rbtree_check)) 1494 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi, 1495 &dcc->root)); 1496 blk_start_plug(&plug); 1497 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1498 f2fs_bug_on(sbi, dc->state != D_PREP); 1499 1500 if (dpolicy->io_aware && i < dpolicy->io_aware_gran && 1501 !is_idle(sbi, DISCARD_TIME)) { 1502 io_interrupted = true; 1503 break; 1504 } 1505 1506 __submit_discard_cmd(sbi, dpolicy, dc, &issued); 1507 1508 if (issued >= dpolicy->max_requests) 1509 break; 1510 } 1511 blk_finish_plug(&plug); 1512 next: 1513 mutex_unlock(&dcc->cmd_lock); 1514 1515 if (issued >= dpolicy->max_requests || io_interrupted) 1516 break; 1517 } 1518 1519 if (!issued && io_interrupted) 1520 issued = -1; 1521 1522 return issued; 1523 } 1524 1525 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi) 1526 { 1527 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1528 struct list_head *pend_list; 1529 struct discard_cmd *dc, *tmp; 1530 int i; 1531 bool dropped = false; 1532 1533 mutex_lock(&dcc->cmd_lock); 1534 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1535 pend_list = &dcc->pend_list[i]; 1536 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1537 f2fs_bug_on(sbi, dc->state != D_PREP); 1538 __remove_discard_cmd(sbi, dc); 1539 dropped = true; 1540 } 1541 } 1542 mutex_unlock(&dcc->cmd_lock); 1543 1544 return dropped; 1545 } 1546 1547 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi) 1548 { 1549 __drop_discard_cmd(sbi); 1550 } 1551 1552 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi, 1553 struct discard_cmd *dc) 1554 { 1555 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1556 unsigned int len = 0; 1557 1558 wait_for_completion_io(&dc->wait); 1559 mutex_lock(&dcc->cmd_lock); 1560 f2fs_bug_on(sbi, dc->state != D_DONE); 1561 dc->ref--; 1562 if (!dc->ref) { 1563 if (!dc->error) 1564 len = dc->len; 1565 __remove_discard_cmd(sbi, dc); 1566 } 1567 mutex_unlock(&dcc->cmd_lock); 1568 1569 return len; 1570 } 1571 1572 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi, 1573 struct discard_policy *dpolicy, 1574 block_t start, block_t end) 1575 { 1576 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1577 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1578 &(dcc->fstrim_list) : &(dcc->wait_list); 1579 struct discard_cmd *dc, *tmp; 1580 bool need_wait; 1581 unsigned int trimmed = 0; 1582 1583 next: 1584 need_wait = false; 1585 1586 mutex_lock(&dcc->cmd_lock); 1587 list_for_each_entry_safe(dc, tmp, wait_list, list) { 1588 if (dc->lstart + dc->len <= start || end <= dc->lstart) 1589 continue; 1590 if (dc->len < dpolicy->granularity) 1591 continue; 1592 if (dc->state == D_DONE && !dc->ref) { 1593 wait_for_completion_io(&dc->wait); 1594 if (!dc->error) 1595 trimmed += dc->len; 1596 __remove_discard_cmd(sbi, dc); 1597 } else { 1598 dc->ref++; 1599 need_wait = true; 1600 break; 1601 } 1602 } 1603 mutex_unlock(&dcc->cmd_lock); 1604 1605 if (need_wait) { 1606 trimmed += __wait_one_discard_bio(sbi, dc); 1607 goto next; 1608 } 1609 1610 return trimmed; 1611 } 1612 1613 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1614 struct discard_policy *dpolicy) 1615 { 1616 struct discard_policy dp; 1617 unsigned int discard_blks; 1618 1619 if (dpolicy) 1620 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX); 1621 1622 /* wait all */ 1623 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1); 1624 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1625 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1); 1626 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1627 1628 return discard_blks; 1629 } 1630 1631 /* This should be covered by global mutex, &sit_i->sentry_lock */ 1632 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr) 1633 { 1634 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1635 struct discard_cmd *dc; 1636 bool need_wait = false; 1637 1638 mutex_lock(&dcc->cmd_lock); 1639 dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root, 1640 NULL, blkaddr); 1641 if (dc) { 1642 if (dc->state == D_PREP) { 1643 __punch_discard_cmd(sbi, dc, blkaddr); 1644 } else { 1645 dc->ref++; 1646 need_wait = true; 1647 } 1648 } 1649 mutex_unlock(&dcc->cmd_lock); 1650 1651 if (need_wait) 1652 __wait_one_discard_bio(sbi, dc); 1653 } 1654 1655 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi) 1656 { 1657 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1658 1659 if (dcc && dcc->f2fs_issue_discard) { 1660 struct task_struct *discard_thread = dcc->f2fs_issue_discard; 1661 1662 dcc->f2fs_issue_discard = NULL; 1663 kthread_stop(discard_thread); 1664 } 1665 } 1666 1667 /* This comes from f2fs_put_super */ 1668 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi) 1669 { 1670 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1671 struct discard_policy dpolicy; 1672 bool dropped; 1673 1674 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT, 1675 dcc->discard_granularity); 1676 dpolicy.timeout = UMOUNT_DISCARD_TIMEOUT; 1677 __issue_discard_cmd(sbi, &dpolicy); 1678 dropped = __drop_discard_cmd(sbi); 1679 1680 /* just to make sure there is no pending discard commands */ 1681 __wait_all_discard_cmd(sbi, NULL); 1682 1683 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt)); 1684 return dropped; 1685 } 1686 1687 static int issue_discard_thread(void *data) 1688 { 1689 struct f2fs_sb_info *sbi = data; 1690 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1691 wait_queue_head_t *q = &dcc->discard_wait_queue; 1692 struct discard_policy dpolicy; 1693 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME; 1694 int issued; 1695 1696 set_freezable(); 1697 1698 do { 1699 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG, 1700 dcc->discard_granularity); 1701 1702 wait_event_interruptible_timeout(*q, 1703 kthread_should_stop() || freezing(current) || 1704 dcc->discard_wake, 1705 msecs_to_jiffies(wait_ms)); 1706 1707 if (dcc->discard_wake) 1708 dcc->discard_wake = 0; 1709 1710 /* clean up pending candidates before going to sleep */ 1711 if (atomic_read(&dcc->queued_discard)) 1712 __wait_all_discard_cmd(sbi, NULL); 1713 1714 if (try_to_freeze()) 1715 continue; 1716 if (f2fs_readonly(sbi->sb)) 1717 continue; 1718 if (kthread_should_stop()) 1719 return 0; 1720 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 1721 wait_ms = dpolicy.max_interval; 1722 continue; 1723 } 1724 1725 if (sbi->gc_mode == GC_URGENT) 1726 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1); 1727 1728 sb_start_intwrite(sbi->sb); 1729 1730 issued = __issue_discard_cmd(sbi, &dpolicy); 1731 if (issued > 0) { 1732 __wait_all_discard_cmd(sbi, &dpolicy); 1733 wait_ms = dpolicy.min_interval; 1734 } else if (issued == -1){ 1735 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME); 1736 if (!wait_ms) 1737 wait_ms = dpolicy.mid_interval; 1738 } else { 1739 wait_ms = dpolicy.max_interval; 1740 } 1741 1742 sb_end_intwrite(sbi->sb); 1743 1744 } while (!kthread_should_stop()); 1745 return 0; 1746 } 1747 1748 #ifdef CONFIG_BLK_DEV_ZONED 1749 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi, 1750 struct block_device *bdev, block_t blkstart, block_t blklen) 1751 { 1752 sector_t sector, nr_sects; 1753 block_t lblkstart = blkstart; 1754 int devi = 0; 1755 1756 if (f2fs_is_multi_device(sbi)) { 1757 devi = f2fs_target_device_index(sbi, blkstart); 1758 if (blkstart < FDEV(devi).start_blk || 1759 blkstart > FDEV(devi).end_blk) { 1760 f2fs_msg(sbi->sb, KERN_ERR, "Invalid block %x", 1761 blkstart); 1762 return -EIO; 1763 } 1764 blkstart -= FDEV(devi).start_blk; 1765 } 1766 1767 /* For sequential zones, reset the zone write pointer */ 1768 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) { 1769 sector = SECTOR_FROM_BLOCK(blkstart); 1770 nr_sects = SECTOR_FROM_BLOCK(blklen); 1771 1772 if (sector & (bdev_zone_sectors(bdev) - 1) || 1773 nr_sects != bdev_zone_sectors(bdev)) { 1774 f2fs_msg(sbi->sb, KERN_ERR, 1775 "(%d) %s: Unaligned zone reset attempted (block %x + %x)", 1776 devi, sbi->s_ndevs ? FDEV(devi).path: "", 1777 blkstart, blklen); 1778 return -EIO; 1779 } 1780 trace_f2fs_issue_reset_zone(bdev, blkstart); 1781 return blkdev_reset_zones(bdev, sector, nr_sects, GFP_NOFS); 1782 } 1783 1784 /* For conventional zones, use regular discard if supported */ 1785 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen); 1786 } 1787 #endif 1788 1789 static int __issue_discard_async(struct f2fs_sb_info *sbi, 1790 struct block_device *bdev, block_t blkstart, block_t blklen) 1791 { 1792 #ifdef CONFIG_BLK_DEV_ZONED 1793 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) 1794 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen); 1795 #endif 1796 return __queue_discard_cmd(sbi, bdev, blkstart, blklen); 1797 } 1798 1799 static int f2fs_issue_discard(struct f2fs_sb_info *sbi, 1800 block_t blkstart, block_t blklen) 1801 { 1802 sector_t start = blkstart, len = 0; 1803 struct block_device *bdev; 1804 struct seg_entry *se; 1805 unsigned int offset; 1806 block_t i; 1807 int err = 0; 1808 1809 bdev = f2fs_target_device(sbi, blkstart, NULL); 1810 1811 for (i = blkstart; i < blkstart + blklen; i++, len++) { 1812 if (i != start) { 1813 struct block_device *bdev2 = 1814 f2fs_target_device(sbi, i, NULL); 1815 1816 if (bdev2 != bdev) { 1817 err = __issue_discard_async(sbi, bdev, 1818 start, len); 1819 if (err) 1820 return err; 1821 bdev = bdev2; 1822 start = i; 1823 len = 0; 1824 } 1825 } 1826 1827 se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); 1828 offset = GET_BLKOFF_FROM_SEG0(sbi, i); 1829 1830 if (!f2fs_test_and_set_bit(offset, se->discard_map)) 1831 sbi->discard_blks--; 1832 } 1833 1834 if (len) 1835 err = __issue_discard_async(sbi, bdev, start, len); 1836 return err; 1837 } 1838 1839 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc, 1840 bool check_only) 1841 { 1842 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 1843 int max_blocks = sbi->blocks_per_seg; 1844 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); 1845 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 1846 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 1847 unsigned long *discard_map = (unsigned long *)se->discard_map; 1848 unsigned long *dmap = SIT_I(sbi)->tmp_map; 1849 unsigned int start = 0, end = -1; 1850 bool force = (cpc->reason & CP_DISCARD); 1851 struct discard_entry *de = NULL; 1852 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list; 1853 int i; 1854 1855 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi)) 1856 return false; 1857 1858 if (!force) { 1859 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks || 1860 SM_I(sbi)->dcc_info->nr_discards >= 1861 SM_I(sbi)->dcc_info->max_discards) 1862 return false; 1863 } 1864 1865 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ 1866 for (i = 0; i < entries; i++) 1867 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : 1868 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; 1869 1870 while (force || SM_I(sbi)->dcc_info->nr_discards <= 1871 SM_I(sbi)->dcc_info->max_discards) { 1872 start = __find_rev_next_bit(dmap, max_blocks, end + 1); 1873 if (start >= max_blocks) 1874 break; 1875 1876 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); 1877 if (force && start && end != max_blocks 1878 && (end - start) < cpc->trim_minlen) 1879 continue; 1880 1881 if (check_only) 1882 return true; 1883 1884 if (!de) { 1885 de = f2fs_kmem_cache_alloc(discard_entry_slab, 1886 GFP_F2FS_ZERO); 1887 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start); 1888 list_add_tail(&de->list, head); 1889 } 1890 1891 for (i = start; i < end; i++) 1892 __set_bit_le(i, (void *)de->discard_map); 1893 1894 SM_I(sbi)->dcc_info->nr_discards += end - start; 1895 } 1896 return false; 1897 } 1898 1899 static void release_discard_addr(struct discard_entry *entry) 1900 { 1901 list_del(&entry->list); 1902 kmem_cache_free(discard_entry_slab, entry); 1903 } 1904 1905 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi) 1906 { 1907 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); 1908 struct discard_entry *entry, *this; 1909 1910 /* drop caches */ 1911 list_for_each_entry_safe(entry, this, head, list) 1912 release_discard_addr(entry); 1913 } 1914 1915 /* 1916 * Should call f2fs_clear_prefree_segments after checkpoint is done. 1917 */ 1918 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 1919 { 1920 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1921 unsigned int segno; 1922 1923 mutex_lock(&dirty_i->seglist_lock); 1924 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) 1925 __set_test_and_free(sbi, segno); 1926 mutex_unlock(&dirty_i->seglist_lock); 1927 } 1928 1929 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi, 1930 struct cp_control *cpc) 1931 { 1932 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1933 struct list_head *head = &dcc->entry_list; 1934 struct discard_entry *entry, *this; 1935 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1936 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; 1937 unsigned int start = 0, end = -1; 1938 unsigned int secno, start_segno; 1939 bool force = (cpc->reason & CP_DISCARD); 1940 bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi); 1941 1942 mutex_lock(&dirty_i->seglist_lock); 1943 1944 while (1) { 1945 int i; 1946 1947 if (need_align && end != -1) 1948 end--; 1949 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); 1950 if (start >= MAIN_SEGS(sbi)) 1951 break; 1952 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), 1953 start + 1); 1954 1955 if (need_align) { 1956 start = rounddown(start, sbi->segs_per_sec); 1957 end = roundup(end, sbi->segs_per_sec); 1958 } 1959 1960 for (i = start; i < end; i++) { 1961 if (test_and_clear_bit(i, prefree_map)) 1962 dirty_i->nr_dirty[PRE]--; 1963 } 1964 1965 if (!f2fs_realtime_discard_enable(sbi)) 1966 continue; 1967 1968 if (force && start >= cpc->trim_start && 1969 (end - 1) <= cpc->trim_end) 1970 continue; 1971 1972 if (!test_opt(sbi, LFS) || !__is_large_section(sbi)) { 1973 f2fs_issue_discard(sbi, START_BLOCK(sbi, start), 1974 (end - start) << sbi->log_blocks_per_seg); 1975 continue; 1976 } 1977 next: 1978 secno = GET_SEC_FROM_SEG(sbi, start); 1979 start_segno = GET_SEG_FROM_SEC(sbi, secno); 1980 if (!IS_CURSEC(sbi, secno) && 1981 !get_valid_blocks(sbi, start, true)) 1982 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno), 1983 sbi->segs_per_sec << sbi->log_blocks_per_seg); 1984 1985 start = start_segno + sbi->segs_per_sec; 1986 if (start < end) 1987 goto next; 1988 else 1989 end = start - 1; 1990 } 1991 mutex_unlock(&dirty_i->seglist_lock); 1992 1993 /* send small discards */ 1994 list_for_each_entry_safe(entry, this, head, list) { 1995 unsigned int cur_pos = 0, next_pos, len, total_len = 0; 1996 bool is_valid = test_bit_le(0, entry->discard_map); 1997 1998 find_next: 1999 if (is_valid) { 2000 next_pos = find_next_zero_bit_le(entry->discard_map, 2001 sbi->blocks_per_seg, cur_pos); 2002 len = next_pos - cur_pos; 2003 2004 if (f2fs_sb_has_blkzoned(sbi) || 2005 (force && len < cpc->trim_minlen)) 2006 goto skip; 2007 2008 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos, 2009 len); 2010 total_len += len; 2011 } else { 2012 next_pos = find_next_bit_le(entry->discard_map, 2013 sbi->blocks_per_seg, cur_pos); 2014 } 2015 skip: 2016 cur_pos = next_pos; 2017 is_valid = !is_valid; 2018 2019 if (cur_pos < sbi->blocks_per_seg) 2020 goto find_next; 2021 2022 release_discard_addr(entry); 2023 dcc->nr_discards -= total_len; 2024 } 2025 2026 wake_up_discard_thread(sbi, false); 2027 } 2028 2029 static int create_discard_cmd_control(struct f2fs_sb_info *sbi) 2030 { 2031 dev_t dev = sbi->sb->s_bdev->bd_dev; 2032 struct discard_cmd_control *dcc; 2033 int err = 0, i; 2034 2035 if (SM_I(sbi)->dcc_info) { 2036 dcc = SM_I(sbi)->dcc_info; 2037 goto init_thread; 2038 } 2039 2040 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL); 2041 if (!dcc) 2042 return -ENOMEM; 2043 2044 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY; 2045 INIT_LIST_HEAD(&dcc->entry_list); 2046 for (i = 0; i < MAX_PLIST_NUM; i++) 2047 INIT_LIST_HEAD(&dcc->pend_list[i]); 2048 INIT_LIST_HEAD(&dcc->wait_list); 2049 INIT_LIST_HEAD(&dcc->fstrim_list); 2050 mutex_init(&dcc->cmd_lock); 2051 atomic_set(&dcc->issued_discard, 0); 2052 atomic_set(&dcc->queued_discard, 0); 2053 atomic_set(&dcc->discard_cmd_cnt, 0); 2054 dcc->nr_discards = 0; 2055 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg; 2056 dcc->undiscard_blks = 0; 2057 dcc->next_pos = 0; 2058 dcc->root = RB_ROOT_CACHED; 2059 dcc->rbtree_check = false; 2060 2061 init_waitqueue_head(&dcc->discard_wait_queue); 2062 SM_I(sbi)->dcc_info = dcc; 2063 init_thread: 2064 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi, 2065 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev)); 2066 if (IS_ERR(dcc->f2fs_issue_discard)) { 2067 err = PTR_ERR(dcc->f2fs_issue_discard); 2068 kvfree(dcc); 2069 SM_I(sbi)->dcc_info = NULL; 2070 return err; 2071 } 2072 2073 return err; 2074 } 2075 2076 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi) 2077 { 2078 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2079 2080 if (!dcc) 2081 return; 2082 2083 f2fs_stop_discard_thread(sbi); 2084 2085 kvfree(dcc); 2086 SM_I(sbi)->dcc_info = NULL; 2087 } 2088 2089 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 2090 { 2091 struct sit_info *sit_i = SIT_I(sbi); 2092 2093 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { 2094 sit_i->dirty_sentries++; 2095 return false; 2096 } 2097 2098 return true; 2099 } 2100 2101 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 2102 unsigned int segno, int modified) 2103 { 2104 struct seg_entry *se = get_seg_entry(sbi, segno); 2105 se->type = type; 2106 if (modified) 2107 __mark_sit_entry_dirty(sbi, segno); 2108 } 2109 2110 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 2111 { 2112 struct seg_entry *se; 2113 unsigned int segno, offset; 2114 long int new_vblocks; 2115 bool exist; 2116 #ifdef CONFIG_F2FS_CHECK_FS 2117 bool mir_exist; 2118 #endif 2119 2120 segno = GET_SEGNO(sbi, blkaddr); 2121 2122 se = get_seg_entry(sbi, segno); 2123 new_vblocks = se->valid_blocks + del; 2124 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2125 2126 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) || 2127 (new_vblocks > sbi->blocks_per_seg))); 2128 2129 se->valid_blocks = new_vblocks; 2130 se->mtime = get_mtime(sbi, false); 2131 if (se->mtime > SIT_I(sbi)->max_mtime) 2132 SIT_I(sbi)->max_mtime = se->mtime; 2133 2134 /* Update valid block bitmap */ 2135 if (del > 0) { 2136 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map); 2137 #ifdef CONFIG_F2FS_CHECK_FS 2138 mir_exist = f2fs_test_and_set_bit(offset, 2139 se->cur_valid_map_mir); 2140 if (unlikely(exist != mir_exist)) { 2141 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error " 2142 "when setting bitmap, blk:%u, old bit:%d", 2143 blkaddr, exist); 2144 f2fs_bug_on(sbi, 1); 2145 } 2146 #endif 2147 if (unlikely(exist)) { 2148 f2fs_msg(sbi->sb, KERN_ERR, 2149 "Bitmap was wrongly set, blk:%u", blkaddr); 2150 f2fs_bug_on(sbi, 1); 2151 se->valid_blocks--; 2152 del = 0; 2153 } 2154 2155 if (!f2fs_test_and_set_bit(offset, se->discard_map)) 2156 sbi->discard_blks--; 2157 2158 /* don't overwrite by SSR to keep node chain */ 2159 if (IS_NODESEG(se->type) && 2160 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) { 2161 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map)) 2162 se->ckpt_valid_blocks++; 2163 } 2164 } else { 2165 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map); 2166 #ifdef CONFIG_F2FS_CHECK_FS 2167 mir_exist = f2fs_test_and_clear_bit(offset, 2168 se->cur_valid_map_mir); 2169 if (unlikely(exist != mir_exist)) { 2170 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error " 2171 "when clearing bitmap, blk:%u, old bit:%d", 2172 blkaddr, exist); 2173 f2fs_bug_on(sbi, 1); 2174 } 2175 #endif 2176 if (unlikely(!exist)) { 2177 f2fs_msg(sbi->sb, KERN_ERR, 2178 "Bitmap was wrongly cleared, blk:%u", blkaddr); 2179 f2fs_bug_on(sbi, 1); 2180 se->valid_blocks++; 2181 del = 0; 2182 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 2183 /* 2184 * If checkpoints are off, we must not reuse data that 2185 * was used in the previous checkpoint. If it was used 2186 * before, we must track that to know how much space we 2187 * really have. 2188 */ 2189 if (f2fs_test_bit(offset, se->ckpt_valid_map)) { 2190 spin_lock(&sbi->stat_lock); 2191 sbi->unusable_block_count++; 2192 spin_unlock(&sbi->stat_lock); 2193 } 2194 } 2195 2196 if (f2fs_test_and_clear_bit(offset, se->discard_map)) 2197 sbi->discard_blks++; 2198 } 2199 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) 2200 se->ckpt_valid_blocks += del; 2201 2202 __mark_sit_entry_dirty(sbi, segno); 2203 2204 /* update total number of valid blocks to be written in ckpt area */ 2205 SIT_I(sbi)->written_valid_blocks += del; 2206 2207 if (__is_large_section(sbi)) 2208 get_sec_entry(sbi, segno)->valid_blocks += del; 2209 } 2210 2211 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) 2212 { 2213 unsigned int segno = GET_SEGNO(sbi, addr); 2214 struct sit_info *sit_i = SIT_I(sbi); 2215 2216 f2fs_bug_on(sbi, addr == NULL_ADDR); 2217 if (addr == NEW_ADDR) 2218 return; 2219 2220 invalidate_mapping_pages(META_MAPPING(sbi), addr, addr); 2221 2222 /* add it into sit main buffer */ 2223 down_write(&sit_i->sentry_lock); 2224 2225 update_sit_entry(sbi, addr, -1); 2226 2227 /* add it into dirty seglist */ 2228 locate_dirty_segment(sbi, segno); 2229 2230 up_write(&sit_i->sentry_lock); 2231 } 2232 2233 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr) 2234 { 2235 struct sit_info *sit_i = SIT_I(sbi); 2236 unsigned int segno, offset; 2237 struct seg_entry *se; 2238 bool is_cp = false; 2239 2240 if (!__is_valid_data_blkaddr(blkaddr)) 2241 return true; 2242 2243 down_read(&sit_i->sentry_lock); 2244 2245 segno = GET_SEGNO(sbi, blkaddr); 2246 se = get_seg_entry(sbi, segno); 2247 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2248 2249 if (f2fs_test_bit(offset, se->ckpt_valid_map)) 2250 is_cp = true; 2251 2252 up_read(&sit_i->sentry_lock); 2253 2254 return is_cp; 2255 } 2256 2257 /* 2258 * This function should be resided under the curseg_mutex lock 2259 */ 2260 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, 2261 struct f2fs_summary *sum) 2262 { 2263 struct curseg_info *curseg = CURSEG_I(sbi, type); 2264 void *addr = curseg->sum_blk; 2265 addr += curseg->next_blkoff * sizeof(struct f2fs_summary); 2266 memcpy(addr, sum, sizeof(struct f2fs_summary)); 2267 } 2268 2269 /* 2270 * Calculate the number of current summary pages for writing 2271 */ 2272 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) 2273 { 2274 int valid_sum_count = 0; 2275 int i, sum_in_page; 2276 2277 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2278 if (sbi->ckpt->alloc_type[i] == SSR) 2279 valid_sum_count += sbi->blocks_per_seg; 2280 else { 2281 if (for_ra) 2282 valid_sum_count += le16_to_cpu( 2283 F2FS_CKPT(sbi)->cur_data_blkoff[i]); 2284 else 2285 valid_sum_count += curseg_blkoff(sbi, i); 2286 } 2287 } 2288 2289 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE - 2290 SUM_FOOTER_SIZE) / SUMMARY_SIZE; 2291 if (valid_sum_count <= sum_in_page) 2292 return 1; 2293 else if ((valid_sum_count - sum_in_page) <= 2294 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) 2295 return 2; 2296 return 3; 2297 } 2298 2299 /* 2300 * Caller should put this summary page 2301 */ 2302 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) 2303 { 2304 return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno)); 2305 } 2306 2307 void f2fs_update_meta_page(struct f2fs_sb_info *sbi, 2308 void *src, block_t blk_addr) 2309 { 2310 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2311 2312 memcpy(page_address(page), src, PAGE_SIZE); 2313 set_page_dirty(page); 2314 f2fs_put_page(page, 1); 2315 } 2316 2317 static void write_sum_page(struct f2fs_sb_info *sbi, 2318 struct f2fs_summary_block *sum_blk, block_t blk_addr) 2319 { 2320 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr); 2321 } 2322 2323 static void write_current_sum_page(struct f2fs_sb_info *sbi, 2324 int type, block_t blk_addr) 2325 { 2326 struct curseg_info *curseg = CURSEG_I(sbi, type); 2327 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2328 struct f2fs_summary_block *src = curseg->sum_blk; 2329 struct f2fs_summary_block *dst; 2330 2331 dst = (struct f2fs_summary_block *)page_address(page); 2332 memset(dst, 0, PAGE_SIZE); 2333 2334 mutex_lock(&curseg->curseg_mutex); 2335 2336 down_read(&curseg->journal_rwsem); 2337 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE); 2338 up_read(&curseg->journal_rwsem); 2339 2340 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE); 2341 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE); 2342 2343 mutex_unlock(&curseg->curseg_mutex); 2344 2345 set_page_dirty(page); 2346 f2fs_put_page(page, 1); 2347 } 2348 2349 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type) 2350 { 2351 struct curseg_info *curseg = CURSEG_I(sbi, type); 2352 unsigned int segno = curseg->segno + 1; 2353 struct free_segmap_info *free_i = FREE_I(sbi); 2354 2355 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec) 2356 return !test_bit(segno, free_i->free_segmap); 2357 return 0; 2358 } 2359 2360 /* 2361 * Find a new segment from the free segments bitmap to right order 2362 * This function should be returned with success, otherwise BUG 2363 */ 2364 static void get_new_segment(struct f2fs_sb_info *sbi, 2365 unsigned int *newseg, bool new_sec, int dir) 2366 { 2367 struct free_segmap_info *free_i = FREE_I(sbi); 2368 unsigned int segno, secno, zoneno; 2369 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; 2370 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg); 2371 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg); 2372 unsigned int left_start = hint; 2373 bool init = true; 2374 int go_left = 0; 2375 int i; 2376 2377 spin_lock(&free_i->segmap_lock); 2378 2379 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { 2380 segno = find_next_zero_bit(free_i->free_segmap, 2381 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1); 2382 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1)) 2383 goto got_it; 2384 } 2385 find_other_zone: 2386 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 2387 if (secno >= MAIN_SECS(sbi)) { 2388 if (dir == ALLOC_RIGHT) { 2389 secno = find_next_zero_bit(free_i->free_secmap, 2390 MAIN_SECS(sbi), 0); 2391 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi)); 2392 } else { 2393 go_left = 1; 2394 left_start = hint - 1; 2395 } 2396 } 2397 if (go_left == 0) 2398 goto skip_left; 2399 2400 while (test_bit(left_start, free_i->free_secmap)) { 2401 if (left_start > 0) { 2402 left_start--; 2403 continue; 2404 } 2405 left_start = find_next_zero_bit(free_i->free_secmap, 2406 MAIN_SECS(sbi), 0); 2407 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi)); 2408 break; 2409 } 2410 secno = left_start; 2411 skip_left: 2412 segno = GET_SEG_FROM_SEC(sbi, secno); 2413 zoneno = GET_ZONE_FROM_SEC(sbi, secno); 2414 2415 /* give up on finding another zone */ 2416 if (!init) 2417 goto got_it; 2418 if (sbi->secs_per_zone == 1) 2419 goto got_it; 2420 if (zoneno == old_zoneno) 2421 goto got_it; 2422 if (dir == ALLOC_LEFT) { 2423 if (!go_left && zoneno + 1 >= total_zones) 2424 goto got_it; 2425 if (go_left && zoneno == 0) 2426 goto got_it; 2427 } 2428 for (i = 0; i < NR_CURSEG_TYPE; i++) 2429 if (CURSEG_I(sbi, i)->zone == zoneno) 2430 break; 2431 2432 if (i < NR_CURSEG_TYPE) { 2433 /* zone is in user, try another */ 2434 if (go_left) 2435 hint = zoneno * sbi->secs_per_zone - 1; 2436 else if (zoneno + 1 >= total_zones) 2437 hint = 0; 2438 else 2439 hint = (zoneno + 1) * sbi->secs_per_zone; 2440 init = false; 2441 goto find_other_zone; 2442 } 2443 got_it: 2444 /* set it as dirty segment in free segmap */ 2445 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); 2446 __set_inuse(sbi, segno); 2447 *newseg = segno; 2448 spin_unlock(&free_i->segmap_lock); 2449 } 2450 2451 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 2452 { 2453 struct curseg_info *curseg = CURSEG_I(sbi, type); 2454 struct summary_footer *sum_footer; 2455 2456 curseg->segno = curseg->next_segno; 2457 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno); 2458 curseg->next_blkoff = 0; 2459 curseg->next_segno = NULL_SEGNO; 2460 2461 sum_footer = &(curseg->sum_blk->footer); 2462 memset(sum_footer, 0, sizeof(struct summary_footer)); 2463 if (IS_DATASEG(type)) 2464 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 2465 if (IS_NODESEG(type)) 2466 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 2467 __set_sit_entry_type(sbi, type, curseg->segno, modified); 2468 } 2469 2470 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type) 2471 { 2472 /* if segs_per_sec is large than 1, we need to keep original policy. */ 2473 if (__is_large_section(sbi)) 2474 return CURSEG_I(sbi, type)->segno; 2475 2476 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 2477 return 0; 2478 2479 if (test_opt(sbi, NOHEAP) && 2480 (type == CURSEG_HOT_DATA || IS_NODESEG(type))) 2481 return 0; 2482 2483 if (SIT_I(sbi)->last_victim[ALLOC_NEXT]) 2484 return SIT_I(sbi)->last_victim[ALLOC_NEXT]; 2485 2486 /* find segments from 0 to reuse freed segments */ 2487 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE) 2488 return 0; 2489 2490 return CURSEG_I(sbi, type)->segno; 2491 } 2492 2493 /* 2494 * Allocate a current working segment. 2495 * This function always allocates a free segment in LFS manner. 2496 */ 2497 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 2498 { 2499 struct curseg_info *curseg = CURSEG_I(sbi, type); 2500 unsigned int segno = curseg->segno; 2501 int dir = ALLOC_LEFT; 2502 2503 write_sum_page(sbi, curseg->sum_blk, 2504 GET_SUM_BLOCK(sbi, segno)); 2505 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA) 2506 dir = ALLOC_RIGHT; 2507 2508 if (test_opt(sbi, NOHEAP)) 2509 dir = ALLOC_RIGHT; 2510 2511 segno = __get_next_segno(sbi, type); 2512 get_new_segment(sbi, &segno, new_sec, dir); 2513 curseg->next_segno = segno; 2514 reset_curseg(sbi, type, 1); 2515 curseg->alloc_type = LFS; 2516 } 2517 2518 static void __next_free_blkoff(struct f2fs_sb_info *sbi, 2519 struct curseg_info *seg, block_t start) 2520 { 2521 struct seg_entry *se = get_seg_entry(sbi, seg->segno); 2522 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 2523 unsigned long *target_map = SIT_I(sbi)->tmp_map; 2524 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 2525 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 2526 int i, pos; 2527 2528 for (i = 0; i < entries; i++) 2529 target_map[i] = ckpt_map[i] | cur_map[i]; 2530 2531 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); 2532 2533 seg->next_blkoff = pos; 2534 } 2535 2536 /* 2537 * If a segment is written by LFS manner, next block offset is just obtained 2538 * by increasing the current block offset. However, if a segment is written by 2539 * SSR manner, next block offset obtained by calling __next_free_blkoff 2540 */ 2541 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, 2542 struct curseg_info *seg) 2543 { 2544 if (seg->alloc_type == SSR) 2545 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1); 2546 else 2547 seg->next_blkoff++; 2548 } 2549 2550 /* 2551 * This function always allocates a used segment(from dirty seglist) by SSR 2552 * manner, so it should recover the existing segment information of valid blocks 2553 */ 2554 static void change_curseg(struct f2fs_sb_info *sbi, int type) 2555 { 2556 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2557 struct curseg_info *curseg = CURSEG_I(sbi, type); 2558 unsigned int new_segno = curseg->next_segno; 2559 struct f2fs_summary_block *sum_node; 2560 struct page *sum_page; 2561 2562 write_sum_page(sbi, curseg->sum_blk, 2563 GET_SUM_BLOCK(sbi, curseg->segno)); 2564 __set_test_and_inuse(sbi, new_segno); 2565 2566 mutex_lock(&dirty_i->seglist_lock); 2567 __remove_dirty_segment(sbi, new_segno, PRE); 2568 __remove_dirty_segment(sbi, new_segno, DIRTY); 2569 mutex_unlock(&dirty_i->seglist_lock); 2570 2571 reset_curseg(sbi, type, 1); 2572 curseg->alloc_type = SSR; 2573 __next_free_blkoff(sbi, curseg, 0); 2574 2575 sum_page = f2fs_get_sum_page(sbi, new_segno); 2576 f2fs_bug_on(sbi, IS_ERR(sum_page)); 2577 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 2578 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); 2579 f2fs_put_page(sum_page, 1); 2580 } 2581 2582 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type) 2583 { 2584 struct curseg_info *curseg = CURSEG_I(sbi, type); 2585 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; 2586 unsigned segno = NULL_SEGNO; 2587 int i, cnt; 2588 bool reversed = false; 2589 2590 /* f2fs_need_SSR() already forces to do this */ 2591 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) { 2592 curseg->next_segno = segno; 2593 return 1; 2594 } 2595 2596 /* For node segments, let's do SSR more intensively */ 2597 if (IS_NODESEG(type)) { 2598 if (type >= CURSEG_WARM_NODE) { 2599 reversed = true; 2600 i = CURSEG_COLD_NODE; 2601 } else { 2602 i = CURSEG_HOT_NODE; 2603 } 2604 cnt = NR_CURSEG_NODE_TYPE; 2605 } else { 2606 if (type >= CURSEG_WARM_DATA) { 2607 reversed = true; 2608 i = CURSEG_COLD_DATA; 2609 } else { 2610 i = CURSEG_HOT_DATA; 2611 } 2612 cnt = NR_CURSEG_DATA_TYPE; 2613 } 2614 2615 for (; cnt-- > 0; reversed ? i-- : i++) { 2616 if (i == type) 2617 continue; 2618 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) { 2619 curseg->next_segno = segno; 2620 return 1; 2621 } 2622 } 2623 2624 /* find valid_blocks=0 in dirty list */ 2625 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 2626 segno = get_free_segment(sbi); 2627 if (segno != NULL_SEGNO) { 2628 curseg->next_segno = segno; 2629 return 1; 2630 } 2631 } 2632 return 0; 2633 } 2634 2635 /* 2636 * flush out current segment and replace it with new segment 2637 * This function should be returned with success, otherwise BUG 2638 */ 2639 static void allocate_segment_by_default(struct f2fs_sb_info *sbi, 2640 int type, bool force) 2641 { 2642 struct curseg_info *curseg = CURSEG_I(sbi, type); 2643 2644 if (force) 2645 new_curseg(sbi, type, true); 2646 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) && 2647 type == CURSEG_WARM_NODE) 2648 new_curseg(sbi, type, false); 2649 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) && 2650 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 2651 new_curseg(sbi, type, false); 2652 else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type)) 2653 change_curseg(sbi, type); 2654 else 2655 new_curseg(sbi, type, false); 2656 2657 stat_inc_seg_type(sbi, curseg); 2658 } 2659 2660 void allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type, 2661 unsigned int start, unsigned int end) 2662 { 2663 struct curseg_info *curseg = CURSEG_I(sbi, type); 2664 unsigned int segno; 2665 2666 down_read(&SM_I(sbi)->curseg_lock); 2667 mutex_lock(&curseg->curseg_mutex); 2668 down_write(&SIT_I(sbi)->sentry_lock); 2669 2670 segno = CURSEG_I(sbi, type)->segno; 2671 if (segno < start || segno > end) 2672 goto unlock; 2673 2674 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type)) 2675 change_curseg(sbi, type); 2676 else 2677 new_curseg(sbi, type, true); 2678 2679 stat_inc_seg_type(sbi, curseg); 2680 2681 locate_dirty_segment(sbi, segno); 2682 unlock: 2683 up_write(&SIT_I(sbi)->sentry_lock); 2684 2685 if (segno != curseg->segno) 2686 f2fs_msg(sbi->sb, KERN_NOTICE, 2687 "For resize: curseg of type %d: %u ==> %u", 2688 type, segno, curseg->segno); 2689 2690 mutex_unlock(&curseg->curseg_mutex); 2691 up_read(&SM_I(sbi)->curseg_lock); 2692 } 2693 2694 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi) 2695 { 2696 struct curseg_info *curseg; 2697 unsigned int old_segno; 2698 int i; 2699 2700 down_write(&SIT_I(sbi)->sentry_lock); 2701 2702 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2703 curseg = CURSEG_I(sbi, i); 2704 old_segno = curseg->segno; 2705 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true); 2706 locate_dirty_segment(sbi, old_segno); 2707 } 2708 2709 up_write(&SIT_I(sbi)->sentry_lock); 2710 } 2711 2712 static const struct segment_allocation default_salloc_ops = { 2713 .allocate_segment = allocate_segment_by_default, 2714 }; 2715 2716 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi, 2717 struct cp_control *cpc) 2718 { 2719 __u64 trim_start = cpc->trim_start; 2720 bool has_candidate = false; 2721 2722 down_write(&SIT_I(sbi)->sentry_lock); 2723 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) { 2724 if (add_discard_addrs(sbi, cpc, true)) { 2725 has_candidate = true; 2726 break; 2727 } 2728 } 2729 up_write(&SIT_I(sbi)->sentry_lock); 2730 2731 cpc->trim_start = trim_start; 2732 return has_candidate; 2733 } 2734 2735 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi, 2736 struct discard_policy *dpolicy, 2737 unsigned int start, unsigned int end) 2738 { 2739 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2740 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 2741 struct rb_node **insert_p = NULL, *insert_parent = NULL; 2742 struct discard_cmd *dc; 2743 struct blk_plug plug; 2744 int issued; 2745 unsigned int trimmed = 0; 2746 2747 next: 2748 issued = 0; 2749 2750 mutex_lock(&dcc->cmd_lock); 2751 if (unlikely(dcc->rbtree_check)) 2752 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi, 2753 &dcc->root)); 2754 2755 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root, 2756 NULL, start, 2757 (struct rb_entry **)&prev_dc, 2758 (struct rb_entry **)&next_dc, 2759 &insert_p, &insert_parent, true, NULL); 2760 if (!dc) 2761 dc = next_dc; 2762 2763 blk_start_plug(&plug); 2764 2765 while (dc && dc->lstart <= end) { 2766 struct rb_node *node; 2767 int err = 0; 2768 2769 if (dc->len < dpolicy->granularity) 2770 goto skip; 2771 2772 if (dc->state != D_PREP) { 2773 list_move_tail(&dc->list, &dcc->fstrim_list); 2774 goto skip; 2775 } 2776 2777 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued); 2778 2779 if (issued >= dpolicy->max_requests) { 2780 start = dc->lstart + dc->len; 2781 2782 if (err) 2783 __remove_discard_cmd(sbi, dc); 2784 2785 blk_finish_plug(&plug); 2786 mutex_unlock(&dcc->cmd_lock); 2787 trimmed += __wait_all_discard_cmd(sbi, NULL); 2788 congestion_wait(BLK_RW_ASYNC, HZ/50); 2789 goto next; 2790 } 2791 skip: 2792 node = rb_next(&dc->rb_node); 2793 if (err) 2794 __remove_discard_cmd(sbi, dc); 2795 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 2796 2797 if (fatal_signal_pending(current)) 2798 break; 2799 } 2800 2801 blk_finish_plug(&plug); 2802 mutex_unlock(&dcc->cmd_lock); 2803 2804 return trimmed; 2805 } 2806 2807 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) 2808 { 2809 __u64 start = F2FS_BYTES_TO_BLK(range->start); 2810 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; 2811 unsigned int start_segno, end_segno; 2812 block_t start_block, end_block; 2813 struct cp_control cpc; 2814 struct discard_policy dpolicy; 2815 unsigned long long trimmed = 0; 2816 int err = 0; 2817 bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi); 2818 2819 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize) 2820 return -EINVAL; 2821 2822 if (end < MAIN_BLKADDR(sbi)) 2823 goto out; 2824 2825 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 2826 f2fs_msg(sbi->sb, KERN_WARNING, 2827 "Found FS corruption, run fsck to fix."); 2828 return -EIO; 2829 } 2830 2831 /* start/end segment number in main_area */ 2832 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); 2833 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : 2834 GET_SEGNO(sbi, end); 2835 if (need_align) { 2836 start_segno = rounddown(start_segno, sbi->segs_per_sec); 2837 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1; 2838 } 2839 2840 cpc.reason = CP_DISCARD; 2841 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen)); 2842 cpc.trim_start = start_segno; 2843 cpc.trim_end = end_segno; 2844 2845 if (sbi->discard_blks == 0) 2846 goto out; 2847 2848 mutex_lock(&sbi->gc_mutex); 2849 err = f2fs_write_checkpoint(sbi, &cpc); 2850 mutex_unlock(&sbi->gc_mutex); 2851 if (err) 2852 goto out; 2853 2854 /* 2855 * We filed discard candidates, but actually we don't need to wait for 2856 * all of them, since they'll be issued in idle time along with runtime 2857 * discard option. User configuration looks like using runtime discard 2858 * or periodic fstrim instead of it. 2859 */ 2860 if (f2fs_realtime_discard_enable(sbi)) 2861 goto out; 2862 2863 start_block = START_BLOCK(sbi, start_segno); 2864 end_block = START_BLOCK(sbi, end_segno + 1); 2865 2866 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen); 2867 trimmed = __issue_discard_cmd_range(sbi, &dpolicy, 2868 start_block, end_block); 2869 2870 trimmed += __wait_discard_cmd_range(sbi, &dpolicy, 2871 start_block, end_block); 2872 out: 2873 if (!err) 2874 range->len = F2FS_BLK_TO_BYTES(trimmed); 2875 return err; 2876 } 2877 2878 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type) 2879 { 2880 struct curseg_info *curseg = CURSEG_I(sbi, type); 2881 if (curseg->next_blkoff < sbi->blocks_per_seg) 2882 return true; 2883 return false; 2884 } 2885 2886 int f2fs_rw_hint_to_seg_type(enum rw_hint hint) 2887 { 2888 switch (hint) { 2889 case WRITE_LIFE_SHORT: 2890 return CURSEG_HOT_DATA; 2891 case WRITE_LIFE_EXTREME: 2892 return CURSEG_COLD_DATA; 2893 default: 2894 return CURSEG_WARM_DATA; 2895 } 2896 } 2897 2898 /* This returns write hints for each segment type. This hints will be 2899 * passed down to block layer. There are mapping tables which depend on 2900 * the mount option 'whint_mode'. 2901 * 2902 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET. 2903 * 2904 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users. 2905 * 2906 * User F2FS Block 2907 * ---- ---- ----- 2908 * META WRITE_LIFE_NOT_SET 2909 * HOT_NODE " 2910 * WARM_NODE " 2911 * COLD_NODE " 2912 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME 2913 * extension list " " 2914 * 2915 * -- buffered io 2916 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 2917 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 2918 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 2919 * WRITE_LIFE_NONE " " 2920 * WRITE_LIFE_MEDIUM " " 2921 * WRITE_LIFE_LONG " " 2922 * 2923 * -- direct io 2924 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 2925 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 2926 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 2927 * WRITE_LIFE_NONE " WRITE_LIFE_NONE 2928 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM 2929 * WRITE_LIFE_LONG " WRITE_LIFE_LONG 2930 * 2931 * 3) whint_mode=fs-based. F2FS passes down hints with its policy. 2932 * 2933 * User F2FS Block 2934 * ---- ---- ----- 2935 * META WRITE_LIFE_MEDIUM; 2936 * HOT_NODE WRITE_LIFE_NOT_SET 2937 * WARM_NODE " 2938 * COLD_NODE WRITE_LIFE_NONE 2939 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME 2940 * extension list " " 2941 * 2942 * -- buffered io 2943 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 2944 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 2945 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG 2946 * WRITE_LIFE_NONE " " 2947 * WRITE_LIFE_MEDIUM " " 2948 * WRITE_LIFE_LONG " " 2949 * 2950 * -- direct io 2951 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 2952 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 2953 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 2954 * WRITE_LIFE_NONE " WRITE_LIFE_NONE 2955 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM 2956 * WRITE_LIFE_LONG " WRITE_LIFE_LONG 2957 */ 2958 2959 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi, 2960 enum page_type type, enum temp_type temp) 2961 { 2962 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) { 2963 if (type == DATA) { 2964 if (temp == WARM) 2965 return WRITE_LIFE_NOT_SET; 2966 else if (temp == HOT) 2967 return WRITE_LIFE_SHORT; 2968 else if (temp == COLD) 2969 return WRITE_LIFE_EXTREME; 2970 } else { 2971 return WRITE_LIFE_NOT_SET; 2972 } 2973 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) { 2974 if (type == DATA) { 2975 if (temp == WARM) 2976 return WRITE_LIFE_LONG; 2977 else if (temp == HOT) 2978 return WRITE_LIFE_SHORT; 2979 else if (temp == COLD) 2980 return WRITE_LIFE_EXTREME; 2981 } else if (type == NODE) { 2982 if (temp == WARM || temp == HOT) 2983 return WRITE_LIFE_NOT_SET; 2984 else if (temp == COLD) 2985 return WRITE_LIFE_NONE; 2986 } else if (type == META) { 2987 return WRITE_LIFE_MEDIUM; 2988 } 2989 } 2990 return WRITE_LIFE_NOT_SET; 2991 } 2992 2993 static int __get_segment_type_2(struct f2fs_io_info *fio) 2994 { 2995 if (fio->type == DATA) 2996 return CURSEG_HOT_DATA; 2997 else 2998 return CURSEG_HOT_NODE; 2999 } 3000 3001 static int __get_segment_type_4(struct f2fs_io_info *fio) 3002 { 3003 if (fio->type == DATA) { 3004 struct inode *inode = fio->page->mapping->host; 3005 3006 if (S_ISDIR(inode->i_mode)) 3007 return CURSEG_HOT_DATA; 3008 else 3009 return CURSEG_COLD_DATA; 3010 } else { 3011 if (IS_DNODE(fio->page) && is_cold_node(fio->page)) 3012 return CURSEG_WARM_NODE; 3013 else 3014 return CURSEG_COLD_NODE; 3015 } 3016 } 3017 3018 static int __get_segment_type_6(struct f2fs_io_info *fio) 3019 { 3020 if (fio->type == DATA) { 3021 struct inode *inode = fio->page->mapping->host; 3022 3023 if (is_cold_data(fio->page) || file_is_cold(inode)) 3024 return CURSEG_COLD_DATA; 3025 if (file_is_hot(inode) || 3026 is_inode_flag_set(inode, FI_HOT_DATA) || 3027 f2fs_is_atomic_file(inode) || 3028 f2fs_is_volatile_file(inode)) 3029 return CURSEG_HOT_DATA; 3030 return f2fs_rw_hint_to_seg_type(inode->i_write_hint); 3031 } else { 3032 if (IS_DNODE(fio->page)) 3033 return is_cold_node(fio->page) ? CURSEG_WARM_NODE : 3034 CURSEG_HOT_NODE; 3035 return CURSEG_COLD_NODE; 3036 } 3037 } 3038 3039 static int __get_segment_type(struct f2fs_io_info *fio) 3040 { 3041 int type = 0; 3042 3043 switch (F2FS_OPTION(fio->sbi).active_logs) { 3044 case 2: 3045 type = __get_segment_type_2(fio); 3046 break; 3047 case 4: 3048 type = __get_segment_type_4(fio); 3049 break; 3050 case 6: 3051 type = __get_segment_type_6(fio); 3052 break; 3053 default: 3054 f2fs_bug_on(fio->sbi, true); 3055 } 3056 3057 if (IS_HOT(type)) 3058 fio->temp = HOT; 3059 else if (IS_WARM(type)) 3060 fio->temp = WARM; 3061 else 3062 fio->temp = COLD; 3063 return type; 3064 } 3065 3066 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, 3067 block_t old_blkaddr, block_t *new_blkaddr, 3068 struct f2fs_summary *sum, int type, 3069 struct f2fs_io_info *fio, bool add_list) 3070 { 3071 struct sit_info *sit_i = SIT_I(sbi); 3072 struct curseg_info *curseg = CURSEG_I(sbi, type); 3073 3074 down_read(&SM_I(sbi)->curseg_lock); 3075 3076 mutex_lock(&curseg->curseg_mutex); 3077 down_write(&sit_i->sentry_lock); 3078 3079 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 3080 3081 f2fs_wait_discard_bio(sbi, *new_blkaddr); 3082 3083 /* 3084 * __add_sum_entry should be resided under the curseg_mutex 3085 * because, this function updates a summary entry in the 3086 * current summary block. 3087 */ 3088 __add_sum_entry(sbi, type, sum); 3089 3090 __refresh_next_blkoff(sbi, curseg); 3091 3092 stat_inc_block_count(sbi, curseg); 3093 3094 /* 3095 * SIT information should be updated before segment allocation, 3096 * since SSR needs latest valid block information. 3097 */ 3098 update_sit_entry(sbi, *new_blkaddr, 1); 3099 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 3100 update_sit_entry(sbi, old_blkaddr, -1); 3101 3102 if (!__has_curseg_space(sbi, type)) 3103 sit_i->s_ops->allocate_segment(sbi, type, false); 3104 3105 /* 3106 * segment dirty status should be updated after segment allocation, 3107 * so we just need to update status only one time after previous 3108 * segment being closed. 3109 */ 3110 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3111 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr)); 3112 3113 up_write(&sit_i->sentry_lock); 3114 3115 if (page && IS_NODESEG(type)) { 3116 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); 3117 3118 f2fs_inode_chksum_set(sbi, page); 3119 } 3120 3121 if (add_list) { 3122 struct f2fs_bio_info *io; 3123 3124 INIT_LIST_HEAD(&fio->list); 3125 fio->in_list = true; 3126 fio->retry = false; 3127 io = sbi->write_io[fio->type] + fio->temp; 3128 spin_lock(&io->io_lock); 3129 list_add_tail(&fio->list, &io->io_list); 3130 spin_unlock(&io->io_lock); 3131 } 3132 3133 mutex_unlock(&curseg->curseg_mutex); 3134 3135 up_read(&SM_I(sbi)->curseg_lock); 3136 } 3137 3138 static void update_device_state(struct f2fs_io_info *fio) 3139 { 3140 struct f2fs_sb_info *sbi = fio->sbi; 3141 unsigned int devidx; 3142 3143 if (!f2fs_is_multi_device(sbi)) 3144 return; 3145 3146 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr); 3147 3148 /* update device state for fsync */ 3149 f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO); 3150 3151 /* update device state for checkpoint */ 3152 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) { 3153 spin_lock(&sbi->dev_lock); 3154 f2fs_set_bit(devidx, (char *)&sbi->dirty_device); 3155 spin_unlock(&sbi->dev_lock); 3156 } 3157 } 3158 3159 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) 3160 { 3161 int type = __get_segment_type(fio); 3162 bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA); 3163 3164 if (keep_order) 3165 down_read(&fio->sbi->io_order_lock); 3166 reallocate: 3167 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr, 3168 &fio->new_blkaddr, sum, type, fio, true); 3169 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO) 3170 invalidate_mapping_pages(META_MAPPING(fio->sbi), 3171 fio->old_blkaddr, fio->old_blkaddr); 3172 3173 /* writeout dirty page into bdev */ 3174 f2fs_submit_page_write(fio); 3175 if (fio->retry) { 3176 fio->old_blkaddr = fio->new_blkaddr; 3177 goto reallocate; 3178 } 3179 3180 update_device_state(fio); 3181 3182 if (keep_order) 3183 up_read(&fio->sbi->io_order_lock); 3184 } 3185 3186 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page, 3187 enum iostat_type io_type) 3188 { 3189 struct f2fs_io_info fio = { 3190 .sbi = sbi, 3191 .type = META, 3192 .temp = HOT, 3193 .op = REQ_OP_WRITE, 3194 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO, 3195 .old_blkaddr = page->index, 3196 .new_blkaddr = page->index, 3197 .page = page, 3198 .encrypted_page = NULL, 3199 .in_list = false, 3200 }; 3201 3202 if (unlikely(page->index >= MAIN_BLKADDR(sbi))) 3203 fio.op_flags &= ~REQ_META; 3204 3205 set_page_writeback(page); 3206 ClearPageError(page); 3207 f2fs_submit_page_write(&fio); 3208 3209 stat_inc_meta_count(sbi, page->index); 3210 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE); 3211 } 3212 3213 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio) 3214 { 3215 struct f2fs_summary sum; 3216 3217 set_summary(&sum, nid, 0, 0); 3218 do_write_page(&sum, fio); 3219 3220 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE); 3221 } 3222 3223 void f2fs_outplace_write_data(struct dnode_of_data *dn, 3224 struct f2fs_io_info *fio) 3225 { 3226 struct f2fs_sb_info *sbi = fio->sbi; 3227 struct f2fs_summary sum; 3228 3229 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); 3230 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version); 3231 do_write_page(&sum, fio); 3232 f2fs_update_data_blkaddr(dn, fio->new_blkaddr); 3233 3234 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE); 3235 } 3236 3237 int f2fs_inplace_write_data(struct f2fs_io_info *fio) 3238 { 3239 int err; 3240 struct f2fs_sb_info *sbi = fio->sbi; 3241 unsigned int segno; 3242 3243 fio->new_blkaddr = fio->old_blkaddr; 3244 /* i/o temperature is needed for passing down write hints */ 3245 __get_segment_type(fio); 3246 3247 segno = GET_SEGNO(sbi, fio->new_blkaddr); 3248 3249 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) { 3250 set_sbi_flag(sbi, SBI_NEED_FSCK); 3251 return -EFAULT; 3252 } 3253 3254 stat_inc_inplace_blocks(fio->sbi); 3255 3256 if (fio->bio) 3257 err = f2fs_merge_page_bio(fio); 3258 else 3259 err = f2fs_submit_page_bio(fio); 3260 if (!err) { 3261 update_device_state(fio); 3262 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE); 3263 } 3264 3265 return err; 3266 } 3267 3268 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi, 3269 unsigned int segno) 3270 { 3271 int i; 3272 3273 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 3274 if (CURSEG_I(sbi, i)->segno == segno) 3275 break; 3276 } 3277 return i; 3278 } 3279 3280 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 3281 block_t old_blkaddr, block_t new_blkaddr, 3282 bool recover_curseg, bool recover_newaddr) 3283 { 3284 struct sit_info *sit_i = SIT_I(sbi); 3285 struct curseg_info *curseg; 3286 unsigned int segno, old_cursegno; 3287 struct seg_entry *se; 3288 int type; 3289 unsigned short old_blkoff; 3290 3291 segno = GET_SEGNO(sbi, new_blkaddr); 3292 se = get_seg_entry(sbi, segno); 3293 type = se->type; 3294 3295 down_write(&SM_I(sbi)->curseg_lock); 3296 3297 if (!recover_curseg) { 3298 /* for recovery flow */ 3299 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { 3300 if (old_blkaddr == NULL_ADDR) 3301 type = CURSEG_COLD_DATA; 3302 else 3303 type = CURSEG_WARM_DATA; 3304 } 3305 } else { 3306 if (IS_CURSEG(sbi, segno)) { 3307 /* se->type is volatile as SSR allocation */ 3308 type = __f2fs_get_curseg(sbi, segno); 3309 f2fs_bug_on(sbi, type == NO_CHECK_TYPE); 3310 } else { 3311 type = CURSEG_WARM_DATA; 3312 } 3313 } 3314 3315 f2fs_bug_on(sbi, !IS_DATASEG(type)); 3316 curseg = CURSEG_I(sbi, type); 3317 3318 mutex_lock(&curseg->curseg_mutex); 3319 down_write(&sit_i->sentry_lock); 3320 3321 old_cursegno = curseg->segno; 3322 old_blkoff = curseg->next_blkoff; 3323 3324 /* change the current segment */ 3325 if (segno != curseg->segno) { 3326 curseg->next_segno = segno; 3327 change_curseg(sbi, type); 3328 } 3329 3330 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); 3331 __add_sum_entry(sbi, type, sum); 3332 3333 if (!recover_curseg || recover_newaddr) 3334 update_sit_entry(sbi, new_blkaddr, 1); 3335 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) { 3336 invalidate_mapping_pages(META_MAPPING(sbi), 3337 old_blkaddr, old_blkaddr); 3338 update_sit_entry(sbi, old_blkaddr, -1); 3339 } 3340 3341 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3342 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); 3343 3344 locate_dirty_segment(sbi, old_cursegno); 3345 3346 if (recover_curseg) { 3347 if (old_cursegno != curseg->segno) { 3348 curseg->next_segno = old_cursegno; 3349 change_curseg(sbi, type); 3350 } 3351 curseg->next_blkoff = old_blkoff; 3352 } 3353 3354 up_write(&sit_i->sentry_lock); 3355 mutex_unlock(&curseg->curseg_mutex); 3356 up_write(&SM_I(sbi)->curseg_lock); 3357 } 3358 3359 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, 3360 block_t old_addr, block_t new_addr, 3361 unsigned char version, bool recover_curseg, 3362 bool recover_newaddr) 3363 { 3364 struct f2fs_summary sum; 3365 3366 set_summary(&sum, dn->nid, dn->ofs_in_node, version); 3367 3368 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr, 3369 recover_curseg, recover_newaddr); 3370 3371 f2fs_update_data_blkaddr(dn, new_addr); 3372 } 3373 3374 void f2fs_wait_on_page_writeback(struct page *page, 3375 enum page_type type, bool ordered, bool locked) 3376 { 3377 if (PageWriteback(page)) { 3378 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 3379 3380 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type); 3381 if (ordered) { 3382 wait_on_page_writeback(page); 3383 f2fs_bug_on(sbi, locked && PageWriteback(page)); 3384 } else { 3385 wait_for_stable_page(page); 3386 } 3387 } 3388 } 3389 3390 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr) 3391 { 3392 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3393 struct page *cpage; 3394 3395 if (!f2fs_post_read_required(inode)) 3396 return; 3397 3398 if (!__is_valid_data_blkaddr(blkaddr)) 3399 return; 3400 3401 cpage = find_lock_page(META_MAPPING(sbi), blkaddr); 3402 if (cpage) { 3403 f2fs_wait_on_page_writeback(cpage, DATA, true, true); 3404 f2fs_put_page(cpage, 1); 3405 } 3406 } 3407 3408 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr, 3409 block_t len) 3410 { 3411 block_t i; 3412 3413 for (i = 0; i < len; i++) 3414 f2fs_wait_on_block_writeback(inode, blkaddr + i); 3415 } 3416 3417 static int read_compacted_summaries(struct f2fs_sb_info *sbi) 3418 { 3419 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3420 struct curseg_info *seg_i; 3421 unsigned char *kaddr; 3422 struct page *page; 3423 block_t start; 3424 int i, j, offset; 3425 3426 start = start_sum_block(sbi); 3427 3428 page = f2fs_get_meta_page(sbi, start++); 3429 if (IS_ERR(page)) 3430 return PTR_ERR(page); 3431 kaddr = (unsigned char *)page_address(page); 3432 3433 /* Step 1: restore nat cache */ 3434 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 3435 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE); 3436 3437 /* Step 2: restore sit cache */ 3438 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 3439 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE); 3440 offset = 2 * SUM_JOURNAL_SIZE; 3441 3442 /* Step 3: restore summary entries */ 3443 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 3444 unsigned short blk_off; 3445 unsigned int segno; 3446 3447 seg_i = CURSEG_I(sbi, i); 3448 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 3449 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 3450 seg_i->next_segno = segno; 3451 reset_curseg(sbi, i, 0); 3452 seg_i->alloc_type = ckpt->alloc_type[i]; 3453 seg_i->next_blkoff = blk_off; 3454 3455 if (seg_i->alloc_type == SSR) 3456 blk_off = sbi->blocks_per_seg; 3457 3458 for (j = 0; j < blk_off; j++) { 3459 struct f2fs_summary *s; 3460 s = (struct f2fs_summary *)(kaddr + offset); 3461 seg_i->sum_blk->entries[j] = *s; 3462 offset += SUMMARY_SIZE; 3463 if (offset + SUMMARY_SIZE <= PAGE_SIZE - 3464 SUM_FOOTER_SIZE) 3465 continue; 3466 3467 f2fs_put_page(page, 1); 3468 page = NULL; 3469 3470 page = f2fs_get_meta_page(sbi, start++); 3471 if (IS_ERR(page)) 3472 return PTR_ERR(page); 3473 kaddr = (unsigned char *)page_address(page); 3474 offset = 0; 3475 } 3476 } 3477 f2fs_put_page(page, 1); 3478 return 0; 3479 } 3480 3481 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 3482 { 3483 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3484 struct f2fs_summary_block *sum; 3485 struct curseg_info *curseg; 3486 struct page *new; 3487 unsigned short blk_off; 3488 unsigned int segno = 0; 3489 block_t blk_addr = 0; 3490 int err = 0; 3491 3492 /* get segment number and block addr */ 3493 if (IS_DATASEG(type)) { 3494 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 3495 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 3496 CURSEG_HOT_DATA]); 3497 if (__exist_node_summaries(sbi)) 3498 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type); 3499 else 3500 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 3501 } else { 3502 segno = le32_to_cpu(ckpt->cur_node_segno[type - 3503 CURSEG_HOT_NODE]); 3504 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 3505 CURSEG_HOT_NODE]); 3506 if (__exist_node_summaries(sbi)) 3507 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 3508 type - CURSEG_HOT_NODE); 3509 else 3510 blk_addr = GET_SUM_BLOCK(sbi, segno); 3511 } 3512 3513 new = f2fs_get_meta_page(sbi, blk_addr); 3514 if (IS_ERR(new)) 3515 return PTR_ERR(new); 3516 sum = (struct f2fs_summary_block *)page_address(new); 3517 3518 if (IS_NODESEG(type)) { 3519 if (__exist_node_summaries(sbi)) { 3520 struct f2fs_summary *ns = &sum->entries[0]; 3521 int i; 3522 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { 3523 ns->version = 0; 3524 ns->ofs_in_node = 0; 3525 } 3526 } else { 3527 err = f2fs_restore_node_summary(sbi, segno, sum); 3528 if (err) 3529 goto out; 3530 } 3531 } 3532 3533 /* set uncompleted segment to curseg */ 3534 curseg = CURSEG_I(sbi, type); 3535 mutex_lock(&curseg->curseg_mutex); 3536 3537 /* update journal info */ 3538 down_write(&curseg->journal_rwsem); 3539 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE); 3540 up_write(&curseg->journal_rwsem); 3541 3542 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE); 3543 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE); 3544 curseg->next_segno = segno; 3545 reset_curseg(sbi, type, 0); 3546 curseg->alloc_type = ckpt->alloc_type[type]; 3547 curseg->next_blkoff = blk_off; 3548 mutex_unlock(&curseg->curseg_mutex); 3549 out: 3550 f2fs_put_page(new, 1); 3551 return err; 3552 } 3553 3554 static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 3555 { 3556 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal; 3557 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal; 3558 int type = CURSEG_HOT_DATA; 3559 int err; 3560 3561 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) { 3562 int npages = f2fs_npages_for_summary_flush(sbi, true); 3563 3564 if (npages >= 2) 3565 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages, 3566 META_CP, true); 3567 3568 /* restore for compacted data summary */ 3569 err = read_compacted_summaries(sbi); 3570 if (err) 3571 return err; 3572 type = CURSEG_HOT_NODE; 3573 } 3574 3575 if (__exist_node_summaries(sbi)) 3576 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type), 3577 NR_CURSEG_TYPE - type, META_CP, true); 3578 3579 for (; type <= CURSEG_COLD_NODE; type++) { 3580 err = read_normal_summaries(sbi, type); 3581 if (err) 3582 return err; 3583 } 3584 3585 /* sanity check for summary blocks */ 3586 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES || 3587 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) { 3588 f2fs_msg(sbi->sb, KERN_ERR, 3589 "invalid journal entries nats %u sits %u\n", 3590 nats_in_cursum(nat_j), sits_in_cursum(sit_j)); 3591 return -EINVAL; 3592 } 3593 3594 return 0; 3595 } 3596 3597 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 3598 { 3599 struct page *page; 3600 unsigned char *kaddr; 3601 struct f2fs_summary *summary; 3602 struct curseg_info *seg_i; 3603 int written_size = 0; 3604 int i, j; 3605 3606 page = f2fs_grab_meta_page(sbi, blkaddr++); 3607 kaddr = (unsigned char *)page_address(page); 3608 memset(kaddr, 0, PAGE_SIZE); 3609 3610 /* Step 1: write nat cache */ 3611 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 3612 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE); 3613 written_size += SUM_JOURNAL_SIZE; 3614 3615 /* Step 2: write sit cache */ 3616 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 3617 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE); 3618 written_size += SUM_JOURNAL_SIZE; 3619 3620 /* Step 3: write summary entries */ 3621 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 3622 unsigned short blkoff; 3623 seg_i = CURSEG_I(sbi, i); 3624 if (sbi->ckpt->alloc_type[i] == SSR) 3625 blkoff = sbi->blocks_per_seg; 3626 else 3627 blkoff = curseg_blkoff(sbi, i); 3628 3629 for (j = 0; j < blkoff; j++) { 3630 if (!page) { 3631 page = f2fs_grab_meta_page(sbi, blkaddr++); 3632 kaddr = (unsigned char *)page_address(page); 3633 memset(kaddr, 0, PAGE_SIZE); 3634 written_size = 0; 3635 } 3636 summary = (struct f2fs_summary *)(kaddr + written_size); 3637 *summary = seg_i->sum_blk->entries[j]; 3638 written_size += SUMMARY_SIZE; 3639 3640 if (written_size + SUMMARY_SIZE <= PAGE_SIZE - 3641 SUM_FOOTER_SIZE) 3642 continue; 3643 3644 set_page_dirty(page); 3645 f2fs_put_page(page, 1); 3646 page = NULL; 3647 } 3648 } 3649 if (page) { 3650 set_page_dirty(page); 3651 f2fs_put_page(page, 1); 3652 } 3653 } 3654 3655 static void write_normal_summaries(struct f2fs_sb_info *sbi, 3656 block_t blkaddr, int type) 3657 { 3658 int i, end; 3659 if (IS_DATASEG(type)) 3660 end = type + NR_CURSEG_DATA_TYPE; 3661 else 3662 end = type + NR_CURSEG_NODE_TYPE; 3663 3664 for (i = type; i < end; i++) 3665 write_current_sum_page(sbi, i, blkaddr + (i - type)); 3666 } 3667 3668 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 3669 { 3670 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) 3671 write_compacted_summaries(sbi, start_blk); 3672 else 3673 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 3674 } 3675 3676 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 3677 { 3678 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 3679 } 3680 3681 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type, 3682 unsigned int val, int alloc) 3683 { 3684 int i; 3685 3686 if (type == NAT_JOURNAL) { 3687 for (i = 0; i < nats_in_cursum(journal); i++) { 3688 if (le32_to_cpu(nid_in_journal(journal, i)) == val) 3689 return i; 3690 } 3691 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL)) 3692 return update_nats_in_cursum(journal, 1); 3693 } else if (type == SIT_JOURNAL) { 3694 for (i = 0; i < sits_in_cursum(journal); i++) 3695 if (le32_to_cpu(segno_in_journal(journal, i)) == val) 3696 return i; 3697 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL)) 3698 return update_sits_in_cursum(journal, 1); 3699 } 3700 return -1; 3701 } 3702 3703 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, 3704 unsigned int segno) 3705 { 3706 return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno)); 3707 } 3708 3709 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, 3710 unsigned int start) 3711 { 3712 struct sit_info *sit_i = SIT_I(sbi); 3713 struct page *page; 3714 pgoff_t src_off, dst_off; 3715 3716 src_off = current_sit_addr(sbi, start); 3717 dst_off = next_sit_addr(sbi, src_off); 3718 3719 page = f2fs_grab_meta_page(sbi, dst_off); 3720 seg_info_to_sit_page(sbi, page, start); 3721 3722 set_page_dirty(page); 3723 set_to_next_sit(sit_i, start); 3724 3725 return page; 3726 } 3727 3728 static struct sit_entry_set *grab_sit_entry_set(void) 3729 { 3730 struct sit_entry_set *ses = 3731 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS); 3732 3733 ses->entry_cnt = 0; 3734 INIT_LIST_HEAD(&ses->set_list); 3735 return ses; 3736 } 3737 3738 static void release_sit_entry_set(struct sit_entry_set *ses) 3739 { 3740 list_del(&ses->set_list); 3741 kmem_cache_free(sit_entry_set_slab, ses); 3742 } 3743 3744 static void adjust_sit_entry_set(struct sit_entry_set *ses, 3745 struct list_head *head) 3746 { 3747 struct sit_entry_set *next = ses; 3748 3749 if (list_is_last(&ses->set_list, head)) 3750 return; 3751 3752 list_for_each_entry_continue(next, head, set_list) 3753 if (ses->entry_cnt <= next->entry_cnt) 3754 break; 3755 3756 list_move_tail(&ses->set_list, &next->set_list); 3757 } 3758 3759 static void add_sit_entry(unsigned int segno, struct list_head *head) 3760 { 3761 struct sit_entry_set *ses; 3762 unsigned int start_segno = START_SEGNO(segno); 3763 3764 list_for_each_entry(ses, head, set_list) { 3765 if (ses->start_segno == start_segno) { 3766 ses->entry_cnt++; 3767 adjust_sit_entry_set(ses, head); 3768 return; 3769 } 3770 } 3771 3772 ses = grab_sit_entry_set(); 3773 3774 ses->start_segno = start_segno; 3775 ses->entry_cnt++; 3776 list_add(&ses->set_list, head); 3777 } 3778 3779 static void add_sits_in_set(struct f2fs_sb_info *sbi) 3780 { 3781 struct f2fs_sm_info *sm_info = SM_I(sbi); 3782 struct list_head *set_list = &sm_info->sit_entry_set; 3783 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; 3784 unsigned int segno; 3785 3786 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) 3787 add_sit_entry(segno, set_list); 3788 } 3789 3790 static void remove_sits_in_journal(struct f2fs_sb_info *sbi) 3791 { 3792 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 3793 struct f2fs_journal *journal = curseg->journal; 3794 int i; 3795 3796 down_write(&curseg->journal_rwsem); 3797 for (i = 0; i < sits_in_cursum(journal); i++) { 3798 unsigned int segno; 3799 bool dirtied; 3800 3801 segno = le32_to_cpu(segno_in_journal(journal, i)); 3802 dirtied = __mark_sit_entry_dirty(sbi, segno); 3803 3804 if (!dirtied) 3805 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); 3806 } 3807 update_sits_in_cursum(journal, -i); 3808 up_write(&curseg->journal_rwsem); 3809 } 3810 3811 /* 3812 * CP calls this function, which flushes SIT entries including sit_journal, 3813 * and moves prefree segs to free segs. 3814 */ 3815 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 3816 { 3817 struct sit_info *sit_i = SIT_I(sbi); 3818 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 3819 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 3820 struct f2fs_journal *journal = curseg->journal; 3821 struct sit_entry_set *ses, *tmp; 3822 struct list_head *head = &SM_I(sbi)->sit_entry_set; 3823 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS); 3824 struct seg_entry *se; 3825 3826 down_write(&sit_i->sentry_lock); 3827 3828 if (!sit_i->dirty_sentries) 3829 goto out; 3830 3831 /* 3832 * add and account sit entries of dirty bitmap in sit entry 3833 * set temporarily 3834 */ 3835 add_sits_in_set(sbi); 3836 3837 /* 3838 * if there are no enough space in journal to store dirty sit 3839 * entries, remove all entries from journal and add and account 3840 * them in sit entry set. 3841 */ 3842 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) || 3843 !to_journal) 3844 remove_sits_in_journal(sbi); 3845 3846 /* 3847 * there are two steps to flush sit entries: 3848 * #1, flush sit entries to journal in current cold data summary block. 3849 * #2, flush sit entries to sit page. 3850 */ 3851 list_for_each_entry_safe(ses, tmp, head, set_list) { 3852 struct page *page = NULL; 3853 struct f2fs_sit_block *raw_sit = NULL; 3854 unsigned int start_segno = ses->start_segno; 3855 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, 3856 (unsigned long)MAIN_SEGS(sbi)); 3857 unsigned int segno = start_segno; 3858 3859 if (to_journal && 3860 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL)) 3861 to_journal = false; 3862 3863 if (to_journal) { 3864 down_write(&curseg->journal_rwsem); 3865 } else { 3866 page = get_next_sit_page(sbi, start_segno); 3867 raw_sit = page_address(page); 3868 } 3869 3870 /* flush dirty sit entries in region of current sit set */ 3871 for_each_set_bit_from(segno, bitmap, end) { 3872 int offset, sit_offset; 3873 3874 se = get_seg_entry(sbi, segno); 3875 #ifdef CONFIG_F2FS_CHECK_FS 3876 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir, 3877 SIT_VBLOCK_MAP_SIZE)) 3878 f2fs_bug_on(sbi, 1); 3879 #endif 3880 3881 /* add discard candidates */ 3882 if (!(cpc->reason & CP_DISCARD)) { 3883 cpc->trim_start = segno; 3884 add_discard_addrs(sbi, cpc, false); 3885 } 3886 3887 if (to_journal) { 3888 offset = f2fs_lookup_journal_in_cursum(journal, 3889 SIT_JOURNAL, segno, 1); 3890 f2fs_bug_on(sbi, offset < 0); 3891 segno_in_journal(journal, offset) = 3892 cpu_to_le32(segno); 3893 seg_info_to_raw_sit(se, 3894 &sit_in_journal(journal, offset)); 3895 check_block_count(sbi, segno, 3896 &sit_in_journal(journal, offset)); 3897 } else { 3898 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 3899 seg_info_to_raw_sit(se, 3900 &raw_sit->entries[sit_offset]); 3901 check_block_count(sbi, segno, 3902 &raw_sit->entries[sit_offset]); 3903 } 3904 3905 __clear_bit(segno, bitmap); 3906 sit_i->dirty_sentries--; 3907 ses->entry_cnt--; 3908 } 3909 3910 if (to_journal) 3911 up_write(&curseg->journal_rwsem); 3912 else 3913 f2fs_put_page(page, 1); 3914 3915 f2fs_bug_on(sbi, ses->entry_cnt); 3916 release_sit_entry_set(ses); 3917 } 3918 3919 f2fs_bug_on(sbi, !list_empty(head)); 3920 f2fs_bug_on(sbi, sit_i->dirty_sentries); 3921 out: 3922 if (cpc->reason & CP_DISCARD) { 3923 __u64 trim_start = cpc->trim_start; 3924 3925 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) 3926 add_discard_addrs(sbi, cpc, false); 3927 3928 cpc->trim_start = trim_start; 3929 } 3930 up_write(&sit_i->sentry_lock); 3931 3932 set_prefree_as_free_segments(sbi); 3933 } 3934 3935 static int build_sit_info(struct f2fs_sb_info *sbi) 3936 { 3937 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 3938 struct sit_info *sit_i; 3939 unsigned int sit_segs, start; 3940 char *src_bitmap; 3941 unsigned int bitmap_size; 3942 3943 /* allocate memory for SIT information */ 3944 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL); 3945 if (!sit_i) 3946 return -ENOMEM; 3947 3948 SM_I(sbi)->sit_info = sit_i; 3949 3950 sit_i->sentries = 3951 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry), 3952 MAIN_SEGS(sbi)), 3953 GFP_KERNEL); 3954 if (!sit_i->sentries) 3955 return -ENOMEM; 3956 3957 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 3958 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size, 3959 GFP_KERNEL); 3960 if (!sit_i->dirty_sentries_bitmap) 3961 return -ENOMEM; 3962 3963 for (start = 0; start < MAIN_SEGS(sbi); start++) { 3964 sit_i->sentries[start].cur_valid_map 3965 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 3966 sit_i->sentries[start].ckpt_valid_map 3967 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 3968 if (!sit_i->sentries[start].cur_valid_map || 3969 !sit_i->sentries[start].ckpt_valid_map) 3970 return -ENOMEM; 3971 3972 #ifdef CONFIG_F2FS_CHECK_FS 3973 sit_i->sentries[start].cur_valid_map_mir 3974 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 3975 if (!sit_i->sentries[start].cur_valid_map_mir) 3976 return -ENOMEM; 3977 #endif 3978 3979 sit_i->sentries[start].discard_map 3980 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, 3981 GFP_KERNEL); 3982 if (!sit_i->sentries[start].discard_map) 3983 return -ENOMEM; 3984 } 3985 3986 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 3987 if (!sit_i->tmp_map) 3988 return -ENOMEM; 3989 3990 if (__is_large_section(sbi)) { 3991 sit_i->sec_entries = 3992 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry), 3993 MAIN_SECS(sbi)), 3994 GFP_KERNEL); 3995 if (!sit_i->sec_entries) 3996 return -ENOMEM; 3997 } 3998 3999 /* get information related with SIT */ 4000 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 4001 4002 /* setup SIT bitmap from ckeckpoint pack */ 4003 bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 4004 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 4005 4006 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); 4007 if (!sit_i->sit_bitmap) 4008 return -ENOMEM; 4009 4010 #ifdef CONFIG_F2FS_CHECK_FS 4011 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); 4012 if (!sit_i->sit_bitmap_mir) 4013 return -ENOMEM; 4014 #endif 4015 4016 /* init SIT information */ 4017 sit_i->s_ops = &default_salloc_ops; 4018 4019 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 4020 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; 4021 sit_i->written_valid_blocks = 0; 4022 sit_i->bitmap_size = bitmap_size; 4023 sit_i->dirty_sentries = 0; 4024 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 4025 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 4026 sit_i->mounted_time = ktime_get_real_seconds(); 4027 init_rwsem(&sit_i->sentry_lock); 4028 return 0; 4029 } 4030 4031 static int build_free_segmap(struct f2fs_sb_info *sbi) 4032 { 4033 struct free_segmap_info *free_i; 4034 unsigned int bitmap_size, sec_bitmap_size; 4035 4036 /* allocate memory for free segmap information */ 4037 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL); 4038 if (!free_i) 4039 return -ENOMEM; 4040 4041 SM_I(sbi)->free_info = free_i; 4042 4043 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4044 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL); 4045 if (!free_i->free_segmap) 4046 return -ENOMEM; 4047 4048 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4049 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL); 4050 if (!free_i->free_secmap) 4051 return -ENOMEM; 4052 4053 /* set all segments as dirty temporarily */ 4054 memset(free_i->free_segmap, 0xff, bitmap_size); 4055 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 4056 4057 /* init free segmap information */ 4058 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); 4059 free_i->free_segments = 0; 4060 free_i->free_sections = 0; 4061 spin_lock_init(&free_i->segmap_lock); 4062 return 0; 4063 } 4064 4065 static int build_curseg(struct f2fs_sb_info *sbi) 4066 { 4067 struct curseg_info *array; 4068 int i; 4069 4070 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)), 4071 GFP_KERNEL); 4072 if (!array) 4073 return -ENOMEM; 4074 4075 SM_I(sbi)->curseg_array = array; 4076 4077 for (i = 0; i < NR_CURSEG_TYPE; i++) { 4078 mutex_init(&array[i].curseg_mutex); 4079 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL); 4080 if (!array[i].sum_blk) 4081 return -ENOMEM; 4082 init_rwsem(&array[i].journal_rwsem); 4083 array[i].journal = f2fs_kzalloc(sbi, 4084 sizeof(struct f2fs_journal), GFP_KERNEL); 4085 if (!array[i].journal) 4086 return -ENOMEM; 4087 array[i].segno = NULL_SEGNO; 4088 array[i].next_blkoff = 0; 4089 } 4090 return restore_curseg_summaries(sbi); 4091 } 4092 4093 static int build_sit_entries(struct f2fs_sb_info *sbi) 4094 { 4095 struct sit_info *sit_i = SIT_I(sbi); 4096 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4097 struct f2fs_journal *journal = curseg->journal; 4098 struct seg_entry *se; 4099 struct f2fs_sit_entry sit; 4100 int sit_blk_cnt = SIT_BLK_CNT(sbi); 4101 unsigned int i, start, end; 4102 unsigned int readed, start_blk = 0; 4103 int err = 0; 4104 block_t total_node_blocks = 0; 4105 4106 do { 4107 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES, 4108 META_SIT, true); 4109 4110 start = start_blk * sit_i->sents_per_block; 4111 end = (start_blk + readed) * sit_i->sents_per_block; 4112 4113 for (; start < end && start < MAIN_SEGS(sbi); start++) { 4114 struct f2fs_sit_block *sit_blk; 4115 struct page *page; 4116 4117 se = &sit_i->sentries[start]; 4118 page = get_current_sit_page(sbi, start); 4119 if (IS_ERR(page)) 4120 return PTR_ERR(page); 4121 sit_blk = (struct f2fs_sit_block *)page_address(page); 4122 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 4123 f2fs_put_page(page, 1); 4124 4125 err = check_block_count(sbi, start, &sit); 4126 if (err) 4127 return err; 4128 seg_info_from_raw_sit(se, &sit); 4129 if (IS_NODESEG(se->type)) 4130 total_node_blocks += se->valid_blocks; 4131 4132 /* build discard map only one time */ 4133 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4134 memset(se->discard_map, 0xff, 4135 SIT_VBLOCK_MAP_SIZE); 4136 } else { 4137 memcpy(se->discard_map, 4138 se->cur_valid_map, 4139 SIT_VBLOCK_MAP_SIZE); 4140 sbi->discard_blks += 4141 sbi->blocks_per_seg - 4142 se->valid_blocks; 4143 } 4144 4145 if (__is_large_section(sbi)) 4146 get_sec_entry(sbi, start)->valid_blocks += 4147 se->valid_blocks; 4148 } 4149 start_blk += readed; 4150 } while (start_blk < sit_blk_cnt); 4151 4152 down_read(&curseg->journal_rwsem); 4153 for (i = 0; i < sits_in_cursum(journal); i++) { 4154 unsigned int old_valid_blocks; 4155 4156 start = le32_to_cpu(segno_in_journal(journal, i)); 4157 if (start >= MAIN_SEGS(sbi)) { 4158 f2fs_msg(sbi->sb, KERN_ERR, 4159 "Wrong journal entry on segno %u", 4160 start); 4161 set_sbi_flag(sbi, SBI_NEED_FSCK); 4162 err = -EINVAL; 4163 break; 4164 } 4165 4166 se = &sit_i->sentries[start]; 4167 sit = sit_in_journal(journal, i); 4168 4169 old_valid_blocks = se->valid_blocks; 4170 if (IS_NODESEG(se->type)) 4171 total_node_blocks -= old_valid_blocks; 4172 4173 err = check_block_count(sbi, start, &sit); 4174 if (err) 4175 break; 4176 seg_info_from_raw_sit(se, &sit); 4177 if (IS_NODESEG(se->type)) 4178 total_node_blocks += se->valid_blocks; 4179 4180 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4181 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE); 4182 } else { 4183 memcpy(se->discard_map, se->cur_valid_map, 4184 SIT_VBLOCK_MAP_SIZE); 4185 sbi->discard_blks += old_valid_blocks; 4186 sbi->discard_blks -= se->valid_blocks; 4187 } 4188 4189 if (__is_large_section(sbi)) { 4190 get_sec_entry(sbi, start)->valid_blocks += 4191 se->valid_blocks; 4192 get_sec_entry(sbi, start)->valid_blocks -= 4193 old_valid_blocks; 4194 } 4195 } 4196 up_read(&curseg->journal_rwsem); 4197 4198 if (!err && total_node_blocks != valid_node_count(sbi)) { 4199 f2fs_msg(sbi->sb, KERN_ERR, 4200 "SIT is corrupted node# %u vs %u", 4201 total_node_blocks, valid_node_count(sbi)); 4202 set_sbi_flag(sbi, SBI_NEED_FSCK); 4203 err = -EINVAL; 4204 } 4205 4206 return err; 4207 } 4208 4209 static void init_free_segmap(struct f2fs_sb_info *sbi) 4210 { 4211 unsigned int start; 4212 int type; 4213 4214 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4215 struct seg_entry *sentry = get_seg_entry(sbi, start); 4216 if (!sentry->valid_blocks) 4217 __set_free(sbi, start); 4218 else 4219 SIT_I(sbi)->written_valid_blocks += 4220 sentry->valid_blocks; 4221 } 4222 4223 /* set use the current segments */ 4224 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 4225 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 4226 __set_test_and_inuse(sbi, curseg_t->segno); 4227 } 4228 } 4229 4230 static void init_dirty_segmap(struct f2fs_sb_info *sbi) 4231 { 4232 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4233 struct free_segmap_info *free_i = FREE_I(sbi); 4234 unsigned int segno = 0, offset = 0; 4235 unsigned short valid_blocks; 4236 4237 while (1) { 4238 /* find dirty segment based on free segmap */ 4239 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); 4240 if (segno >= MAIN_SEGS(sbi)) 4241 break; 4242 offset = segno + 1; 4243 valid_blocks = get_valid_blocks(sbi, segno, false); 4244 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks) 4245 continue; 4246 if (valid_blocks > sbi->blocks_per_seg) { 4247 f2fs_bug_on(sbi, 1); 4248 continue; 4249 } 4250 mutex_lock(&dirty_i->seglist_lock); 4251 __locate_dirty_segment(sbi, segno, DIRTY); 4252 mutex_unlock(&dirty_i->seglist_lock); 4253 } 4254 } 4255 4256 static int init_victim_secmap(struct f2fs_sb_info *sbi) 4257 { 4258 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4259 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4260 4261 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4262 if (!dirty_i->victim_secmap) 4263 return -ENOMEM; 4264 return 0; 4265 } 4266 4267 static int build_dirty_segmap(struct f2fs_sb_info *sbi) 4268 { 4269 struct dirty_seglist_info *dirty_i; 4270 unsigned int bitmap_size, i; 4271 4272 /* allocate memory for dirty segments list information */ 4273 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info), 4274 GFP_KERNEL); 4275 if (!dirty_i) 4276 return -ENOMEM; 4277 4278 SM_I(sbi)->dirty_info = dirty_i; 4279 mutex_init(&dirty_i->seglist_lock); 4280 4281 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4282 4283 for (i = 0; i < NR_DIRTY_TYPE; i++) { 4284 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size, 4285 GFP_KERNEL); 4286 if (!dirty_i->dirty_segmap[i]) 4287 return -ENOMEM; 4288 } 4289 4290 init_dirty_segmap(sbi); 4291 return init_victim_secmap(sbi); 4292 } 4293 4294 static int sanity_check_curseg(struct f2fs_sb_info *sbi) 4295 { 4296 int i; 4297 4298 /* 4299 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr; 4300 * In LFS curseg, all blkaddr after .next_blkoff should be unused. 4301 */ 4302 for (i = 0; i < NO_CHECK_TYPE; i++) { 4303 struct curseg_info *curseg = CURSEG_I(sbi, i); 4304 struct seg_entry *se = get_seg_entry(sbi, curseg->segno); 4305 unsigned int blkofs = curseg->next_blkoff; 4306 4307 if (f2fs_test_bit(blkofs, se->cur_valid_map)) 4308 goto out; 4309 4310 if (curseg->alloc_type == SSR) 4311 continue; 4312 4313 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) { 4314 if (!f2fs_test_bit(blkofs, se->cur_valid_map)) 4315 continue; 4316 out: 4317 f2fs_msg(sbi->sb, KERN_ERR, 4318 "Current segment's next free block offset is " 4319 "inconsistent with bitmap, logtype:%u, " 4320 "segno:%u, type:%u, next_blkoff:%u, blkofs:%u", 4321 i, curseg->segno, curseg->alloc_type, 4322 curseg->next_blkoff, blkofs); 4323 return -EINVAL; 4324 } 4325 } 4326 return 0; 4327 } 4328 4329 /* 4330 * Update min, max modified time for cost-benefit GC algorithm 4331 */ 4332 static void init_min_max_mtime(struct f2fs_sb_info *sbi) 4333 { 4334 struct sit_info *sit_i = SIT_I(sbi); 4335 unsigned int segno; 4336 4337 down_write(&sit_i->sentry_lock); 4338 4339 sit_i->min_mtime = ULLONG_MAX; 4340 4341 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 4342 unsigned int i; 4343 unsigned long long mtime = 0; 4344 4345 for (i = 0; i < sbi->segs_per_sec; i++) 4346 mtime += get_seg_entry(sbi, segno + i)->mtime; 4347 4348 mtime = div_u64(mtime, sbi->segs_per_sec); 4349 4350 if (sit_i->min_mtime > mtime) 4351 sit_i->min_mtime = mtime; 4352 } 4353 sit_i->max_mtime = get_mtime(sbi, false); 4354 up_write(&sit_i->sentry_lock); 4355 } 4356 4357 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi) 4358 { 4359 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 4360 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 4361 struct f2fs_sm_info *sm_info; 4362 int err; 4363 4364 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL); 4365 if (!sm_info) 4366 return -ENOMEM; 4367 4368 /* init sm info */ 4369 sbi->sm_info = sm_info; 4370 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 4371 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 4372 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 4373 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 4374 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 4375 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 4376 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 4377 sm_info->rec_prefree_segments = sm_info->main_segments * 4378 DEF_RECLAIM_PREFREE_SEGMENTS / 100; 4379 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS) 4380 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS; 4381 4382 if (!test_opt(sbi, LFS)) 4383 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC; 4384 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; 4385 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; 4386 sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec; 4387 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS; 4388 sm_info->min_ssr_sections = reserved_sections(sbi); 4389 4390 INIT_LIST_HEAD(&sm_info->sit_entry_set); 4391 4392 init_rwsem(&sm_info->curseg_lock); 4393 4394 if (!f2fs_readonly(sbi->sb)) { 4395 err = f2fs_create_flush_cmd_control(sbi); 4396 if (err) 4397 return err; 4398 } 4399 4400 err = create_discard_cmd_control(sbi); 4401 if (err) 4402 return err; 4403 4404 err = build_sit_info(sbi); 4405 if (err) 4406 return err; 4407 err = build_free_segmap(sbi); 4408 if (err) 4409 return err; 4410 err = build_curseg(sbi); 4411 if (err) 4412 return err; 4413 4414 /* reinit free segmap based on SIT */ 4415 err = build_sit_entries(sbi); 4416 if (err) 4417 return err; 4418 4419 init_free_segmap(sbi); 4420 err = build_dirty_segmap(sbi); 4421 if (err) 4422 return err; 4423 4424 err = sanity_check_curseg(sbi); 4425 if (err) 4426 return err; 4427 4428 init_min_max_mtime(sbi); 4429 return 0; 4430 } 4431 4432 static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 4433 enum dirty_type dirty_type) 4434 { 4435 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4436 4437 mutex_lock(&dirty_i->seglist_lock); 4438 kvfree(dirty_i->dirty_segmap[dirty_type]); 4439 dirty_i->nr_dirty[dirty_type] = 0; 4440 mutex_unlock(&dirty_i->seglist_lock); 4441 } 4442 4443 static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 4444 { 4445 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4446 kvfree(dirty_i->victim_secmap); 4447 } 4448 4449 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 4450 { 4451 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4452 int i; 4453 4454 if (!dirty_i) 4455 return; 4456 4457 /* discard pre-free/dirty segments list */ 4458 for (i = 0; i < NR_DIRTY_TYPE; i++) 4459 discard_dirty_segmap(sbi, i); 4460 4461 destroy_victim_secmap(sbi); 4462 SM_I(sbi)->dirty_info = NULL; 4463 kvfree(dirty_i); 4464 } 4465 4466 static void destroy_curseg(struct f2fs_sb_info *sbi) 4467 { 4468 struct curseg_info *array = SM_I(sbi)->curseg_array; 4469 int i; 4470 4471 if (!array) 4472 return; 4473 SM_I(sbi)->curseg_array = NULL; 4474 for (i = 0; i < NR_CURSEG_TYPE; i++) { 4475 kvfree(array[i].sum_blk); 4476 kvfree(array[i].journal); 4477 } 4478 kvfree(array); 4479 } 4480 4481 static void destroy_free_segmap(struct f2fs_sb_info *sbi) 4482 { 4483 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 4484 if (!free_i) 4485 return; 4486 SM_I(sbi)->free_info = NULL; 4487 kvfree(free_i->free_segmap); 4488 kvfree(free_i->free_secmap); 4489 kvfree(free_i); 4490 } 4491 4492 static void destroy_sit_info(struct f2fs_sb_info *sbi) 4493 { 4494 struct sit_info *sit_i = SIT_I(sbi); 4495 unsigned int start; 4496 4497 if (!sit_i) 4498 return; 4499 4500 if (sit_i->sentries) { 4501 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4502 kvfree(sit_i->sentries[start].cur_valid_map); 4503 #ifdef CONFIG_F2FS_CHECK_FS 4504 kvfree(sit_i->sentries[start].cur_valid_map_mir); 4505 #endif 4506 kvfree(sit_i->sentries[start].ckpt_valid_map); 4507 kvfree(sit_i->sentries[start].discard_map); 4508 } 4509 } 4510 kvfree(sit_i->tmp_map); 4511 4512 kvfree(sit_i->sentries); 4513 kvfree(sit_i->sec_entries); 4514 kvfree(sit_i->dirty_sentries_bitmap); 4515 4516 SM_I(sbi)->sit_info = NULL; 4517 kvfree(sit_i->sit_bitmap); 4518 #ifdef CONFIG_F2FS_CHECK_FS 4519 kvfree(sit_i->sit_bitmap_mir); 4520 #endif 4521 kvfree(sit_i); 4522 } 4523 4524 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi) 4525 { 4526 struct f2fs_sm_info *sm_info = SM_I(sbi); 4527 4528 if (!sm_info) 4529 return; 4530 f2fs_destroy_flush_cmd_control(sbi, true); 4531 destroy_discard_cmd_control(sbi); 4532 destroy_dirty_segmap(sbi); 4533 destroy_curseg(sbi); 4534 destroy_free_segmap(sbi); 4535 destroy_sit_info(sbi); 4536 sbi->sm_info = NULL; 4537 kvfree(sm_info); 4538 } 4539 4540 int __init f2fs_create_segment_manager_caches(void) 4541 { 4542 discard_entry_slab = f2fs_kmem_cache_create("discard_entry", 4543 sizeof(struct discard_entry)); 4544 if (!discard_entry_slab) 4545 goto fail; 4546 4547 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd", 4548 sizeof(struct discard_cmd)); 4549 if (!discard_cmd_slab) 4550 goto destroy_discard_entry; 4551 4552 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set", 4553 sizeof(struct sit_entry_set)); 4554 if (!sit_entry_set_slab) 4555 goto destroy_discard_cmd; 4556 4557 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry", 4558 sizeof(struct inmem_pages)); 4559 if (!inmem_entry_slab) 4560 goto destroy_sit_entry_set; 4561 return 0; 4562 4563 destroy_sit_entry_set: 4564 kmem_cache_destroy(sit_entry_set_slab); 4565 destroy_discard_cmd: 4566 kmem_cache_destroy(discard_cmd_slab); 4567 destroy_discard_entry: 4568 kmem_cache_destroy(discard_entry_slab); 4569 fail: 4570 return -ENOMEM; 4571 } 4572 4573 void f2fs_destroy_segment_manager_caches(void) 4574 { 4575 kmem_cache_destroy(sit_entry_set_slab); 4576 kmem_cache_destroy(discard_cmd_slab); 4577 kmem_cache_destroy(discard_entry_slab); 4578 kmem_cache_destroy(inmem_entry_slab); 4579 } 4580