1 /* 2 * fs/f2fs/segment.c 3 * 4 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 5 * http://www.samsung.com/ 6 * 7 * This program is free software; you can redistribute it and/or modify 8 * it under the terms of the GNU General Public License version 2 as 9 * published by the Free Software Foundation. 10 */ 11 #include <linux/fs.h> 12 #include <linux/f2fs_fs.h> 13 #include <linux/bio.h> 14 #include <linux/blkdev.h> 15 #include <linux/prefetch.h> 16 #include <linux/kthread.h> 17 #include <linux/swap.h> 18 #include <linux/timer.h> 19 #include <linux/freezer.h> 20 #include <linux/sched/signal.h> 21 22 #include "f2fs.h" 23 #include "segment.h" 24 #include "node.h" 25 #include "gc.h" 26 #include "trace.h" 27 #include <trace/events/f2fs.h> 28 29 #define __reverse_ffz(x) __reverse_ffs(~(x)) 30 31 static struct kmem_cache *discard_entry_slab; 32 static struct kmem_cache *discard_cmd_slab; 33 static struct kmem_cache *sit_entry_set_slab; 34 static struct kmem_cache *inmem_entry_slab; 35 36 static unsigned long __reverse_ulong(unsigned char *str) 37 { 38 unsigned long tmp = 0; 39 int shift = 24, idx = 0; 40 41 #if BITS_PER_LONG == 64 42 shift = 56; 43 #endif 44 while (shift >= 0) { 45 tmp |= (unsigned long)str[idx++] << shift; 46 shift -= BITS_PER_BYTE; 47 } 48 return tmp; 49 } 50 51 /* 52 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since 53 * MSB and LSB are reversed in a byte by f2fs_set_bit. 54 */ 55 static inline unsigned long __reverse_ffs(unsigned long word) 56 { 57 int num = 0; 58 59 #if BITS_PER_LONG == 64 60 if ((word & 0xffffffff00000000UL) == 0) 61 num += 32; 62 else 63 word >>= 32; 64 #endif 65 if ((word & 0xffff0000) == 0) 66 num += 16; 67 else 68 word >>= 16; 69 70 if ((word & 0xff00) == 0) 71 num += 8; 72 else 73 word >>= 8; 74 75 if ((word & 0xf0) == 0) 76 num += 4; 77 else 78 word >>= 4; 79 80 if ((word & 0xc) == 0) 81 num += 2; 82 else 83 word >>= 2; 84 85 if ((word & 0x2) == 0) 86 num += 1; 87 return num; 88 } 89 90 /* 91 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because 92 * f2fs_set_bit makes MSB and LSB reversed in a byte. 93 * @size must be integral times of unsigned long. 94 * Example: 95 * MSB <--> LSB 96 * f2fs_set_bit(0, bitmap) => 1000 0000 97 * f2fs_set_bit(7, bitmap) => 0000 0001 98 */ 99 static unsigned long __find_rev_next_bit(const unsigned long *addr, 100 unsigned long size, unsigned long offset) 101 { 102 const unsigned long *p = addr + BIT_WORD(offset); 103 unsigned long result = size; 104 unsigned long tmp; 105 106 if (offset >= size) 107 return size; 108 109 size -= (offset & ~(BITS_PER_LONG - 1)); 110 offset %= BITS_PER_LONG; 111 112 while (1) { 113 if (*p == 0) 114 goto pass; 115 116 tmp = __reverse_ulong((unsigned char *)p); 117 118 tmp &= ~0UL >> offset; 119 if (size < BITS_PER_LONG) 120 tmp &= (~0UL << (BITS_PER_LONG - size)); 121 if (tmp) 122 goto found; 123 pass: 124 if (size <= BITS_PER_LONG) 125 break; 126 size -= BITS_PER_LONG; 127 offset = 0; 128 p++; 129 } 130 return result; 131 found: 132 return result - size + __reverse_ffs(tmp); 133 } 134 135 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr, 136 unsigned long size, unsigned long offset) 137 { 138 const unsigned long *p = addr + BIT_WORD(offset); 139 unsigned long result = size; 140 unsigned long tmp; 141 142 if (offset >= size) 143 return size; 144 145 size -= (offset & ~(BITS_PER_LONG - 1)); 146 offset %= BITS_PER_LONG; 147 148 while (1) { 149 if (*p == ~0UL) 150 goto pass; 151 152 tmp = __reverse_ulong((unsigned char *)p); 153 154 if (offset) 155 tmp |= ~0UL << (BITS_PER_LONG - offset); 156 if (size < BITS_PER_LONG) 157 tmp |= ~0UL >> size; 158 if (tmp != ~0UL) 159 goto found; 160 pass: 161 if (size <= BITS_PER_LONG) 162 break; 163 size -= BITS_PER_LONG; 164 offset = 0; 165 p++; 166 } 167 return result; 168 found: 169 return result - size + __reverse_ffz(tmp); 170 } 171 172 bool need_SSR(struct f2fs_sb_info *sbi) 173 { 174 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES); 175 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS); 176 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA); 177 178 if (test_opt(sbi, LFS)) 179 return false; 180 if (sbi->gc_mode == GC_URGENT) 181 return true; 182 183 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs + 184 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi)); 185 } 186 187 void register_inmem_page(struct inode *inode, struct page *page) 188 { 189 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 190 struct f2fs_inode_info *fi = F2FS_I(inode); 191 struct inmem_pages *new; 192 193 f2fs_trace_pid(page); 194 195 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE); 196 SetPagePrivate(page); 197 198 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS); 199 200 /* add atomic page indices to the list */ 201 new->page = page; 202 INIT_LIST_HEAD(&new->list); 203 204 /* increase reference count with clean state */ 205 mutex_lock(&fi->inmem_lock); 206 get_page(page); 207 list_add_tail(&new->list, &fi->inmem_pages); 208 spin_lock(&sbi->inode_lock[ATOMIC_FILE]); 209 if (list_empty(&fi->inmem_ilist)) 210 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]); 211 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 212 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); 213 mutex_unlock(&fi->inmem_lock); 214 215 trace_f2fs_register_inmem_page(page, INMEM); 216 } 217 218 static int __revoke_inmem_pages(struct inode *inode, 219 struct list_head *head, bool drop, bool recover) 220 { 221 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 222 struct inmem_pages *cur, *tmp; 223 int err = 0; 224 225 list_for_each_entry_safe(cur, tmp, head, list) { 226 struct page *page = cur->page; 227 228 if (drop) 229 trace_f2fs_commit_inmem_page(page, INMEM_DROP); 230 231 lock_page(page); 232 233 f2fs_wait_on_page_writeback(page, DATA, true); 234 235 if (recover) { 236 struct dnode_of_data dn; 237 struct node_info ni; 238 239 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE); 240 retry: 241 set_new_dnode(&dn, inode, NULL, NULL, 0); 242 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE); 243 if (err) { 244 if (err == -ENOMEM) { 245 congestion_wait(BLK_RW_ASYNC, HZ/50); 246 cond_resched(); 247 goto retry; 248 } 249 err = -EAGAIN; 250 goto next; 251 } 252 get_node_info(sbi, dn.nid, &ni); 253 if (cur->old_addr == NEW_ADDR) { 254 invalidate_blocks(sbi, dn.data_blkaddr); 255 f2fs_update_data_blkaddr(&dn, NEW_ADDR); 256 } else 257 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, 258 cur->old_addr, ni.version, true, true); 259 f2fs_put_dnode(&dn); 260 } 261 next: 262 /* we don't need to invalidate this in the sccessful status */ 263 if (drop || recover) 264 ClearPageUptodate(page); 265 set_page_private(page, 0); 266 ClearPagePrivate(page); 267 f2fs_put_page(page, 1); 268 269 list_del(&cur->list); 270 kmem_cache_free(inmem_entry_slab, cur); 271 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); 272 } 273 return err; 274 } 275 276 void drop_inmem_pages_all(struct f2fs_sb_info *sbi) 277 { 278 struct list_head *head = &sbi->inode_list[ATOMIC_FILE]; 279 struct inode *inode; 280 struct f2fs_inode_info *fi; 281 next: 282 spin_lock(&sbi->inode_lock[ATOMIC_FILE]); 283 if (list_empty(head)) { 284 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 285 return; 286 } 287 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist); 288 inode = igrab(&fi->vfs_inode); 289 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 290 291 if (inode) { 292 drop_inmem_pages(inode); 293 iput(inode); 294 } 295 congestion_wait(BLK_RW_ASYNC, HZ/50); 296 cond_resched(); 297 goto next; 298 } 299 300 void drop_inmem_pages(struct inode *inode) 301 { 302 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 303 struct f2fs_inode_info *fi = F2FS_I(inode); 304 305 mutex_lock(&fi->inmem_lock); 306 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false); 307 spin_lock(&sbi->inode_lock[ATOMIC_FILE]); 308 if (!list_empty(&fi->inmem_ilist)) 309 list_del_init(&fi->inmem_ilist); 310 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 311 mutex_unlock(&fi->inmem_lock); 312 313 clear_inode_flag(inode, FI_ATOMIC_FILE); 314 stat_dec_atomic_write(inode); 315 } 316 317 void drop_inmem_page(struct inode *inode, struct page *page) 318 { 319 struct f2fs_inode_info *fi = F2FS_I(inode); 320 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 321 struct list_head *head = &fi->inmem_pages; 322 struct inmem_pages *cur = NULL; 323 324 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page)); 325 326 mutex_lock(&fi->inmem_lock); 327 list_for_each_entry(cur, head, list) { 328 if (cur->page == page) 329 break; 330 } 331 332 f2fs_bug_on(sbi, list_empty(head) || cur->page != page); 333 list_del(&cur->list); 334 mutex_unlock(&fi->inmem_lock); 335 336 dec_page_count(sbi, F2FS_INMEM_PAGES); 337 kmem_cache_free(inmem_entry_slab, cur); 338 339 ClearPageUptodate(page); 340 set_page_private(page, 0); 341 ClearPagePrivate(page); 342 f2fs_put_page(page, 0); 343 344 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE); 345 } 346 347 static int __commit_inmem_pages(struct inode *inode) 348 { 349 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 350 struct f2fs_inode_info *fi = F2FS_I(inode); 351 struct inmem_pages *cur, *tmp; 352 struct f2fs_io_info fio = { 353 .sbi = sbi, 354 .ino = inode->i_ino, 355 .type = DATA, 356 .op = REQ_OP_WRITE, 357 .op_flags = REQ_SYNC | REQ_PRIO, 358 .io_type = FS_DATA_IO, 359 }; 360 struct list_head revoke_list; 361 pgoff_t last_idx = ULONG_MAX; 362 int err = 0; 363 364 INIT_LIST_HEAD(&revoke_list); 365 366 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) { 367 struct page *page = cur->page; 368 369 lock_page(page); 370 if (page->mapping == inode->i_mapping) { 371 trace_f2fs_commit_inmem_page(page, INMEM); 372 373 set_page_dirty(page); 374 f2fs_wait_on_page_writeback(page, DATA, true); 375 if (clear_page_dirty_for_io(page)) { 376 inode_dec_dirty_pages(inode); 377 remove_dirty_inode(inode); 378 } 379 retry: 380 fio.page = page; 381 fio.old_blkaddr = NULL_ADDR; 382 fio.encrypted_page = NULL; 383 fio.need_lock = LOCK_DONE; 384 err = do_write_data_page(&fio); 385 if (err) { 386 if (err == -ENOMEM) { 387 congestion_wait(BLK_RW_ASYNC, HZ/50); 388 cond_resched(); 389 goto retry; 390 } 391 unlock_page(page); 392 break; 393 } 394 /* record old blkaddr for revoking */ 395 cur->old_addr = fio.old_blkaddr; 396 last_idx = page->index; 397 } 398 unlock_page(page); 399 list_move_tail(&cur->list, &revoke_list); 400 } 401 402 if (last_idx != ULONG_MAX) 403 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA); 404 405 if (err) { 406 /* 407 * try to revoke all committed pages, but still we could fail 408 * due to no memory or other reason, if that happened, EAGAIN 409 * will be returned, which means in such case, transaction is 410 * already not integrity, caller should use journal to do the 411 * recovery or rewrite & commit last transaction. For other 412 * error number, revoking was done by filesystem itself. 413 */ 414 err = __revoke_inmem_pages(inode, &revoke_list, false, true); 415 416 /* drop all uncommitted pages */ 417 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false); 418 } else { 419 __revoke_inmem_pages(inode, &revoke_list, false, false); 420 } 421 422 return err; 423 } 424 425 int commit_inmem_pages(struct inode *inode) 426 { 427 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 428 struct f2fs_inode_info *fi = F2FS_I(inode); 429 int err; 430 431 f2fs_balance_fs(sbi, true); 432 f2fs_lock_op(sbi); 433 434 set_inode_flag(inode, FI_ATOMIC_COMMIT); 435 436 mutex_lock(&fi->inmem_lock); 437 err = __commit_inmem_pages(inode); 438 439 spin_lock(&sbi->inode_lock[ATOMIC_FILE]); 440 if (!list_empty(&fi->inmem_ilist)) 441 list_del_init(&fi->inmem_ilist); 442 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 443 mutex_unlock(&fi->inmem_lock); 444 445 clear_inode_flag(inode, FI_ATOMIC_COMMIT); 446 447 f2fs_unlock_op(sbi); 448 return err; 449 } 450 451 /* 452 * This function balances dirty node and dentry pages. 453 * In addition, it controls garbage collection. 454 */ 455 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need) 456 { 457 #ifdef CONFIG_F2FS_FAULT_INJECTION 458 if (time_to_inject(sbi, FAULT_CHECKPOINT)) { 459 f2fs_show_injection_info(FAULT_CHECKPOINT); 460 f2fs_stop_checkpoint(sbi, false); 461 } 462 #endif 463 464 /* balance_fs_bg is able to be pending */ 465 if (need && excess_cached_nats(sbi)) 466 f2fs_balance_fs_bg(sbi); 467 468 /* 469 * We should do GC or end up with checkpoint, if there are so many dirty 470 * dir/node pages without enough free segments. 471 */ 472 if (has_not_enough_free_secs(sbi, 0, 0)) { 473 mutex_lock(&sbi->gc_mutex); 474 f2fs_gc(sbi, false, false, NULL_SEGNO); 475 } 476 } 477 478 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi) 479 { 480 /* try to shrink extent cache when there is no enough memory */ 481 if (!available_free_memory(sbi, EXTENT_CACHE)) 482 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER); 483 484 /* check the # of cached NAT entries */ 485 if (!available_free_memory(sbi, NAT_ENTRIES)) 486 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK); 487 488 if (!available_free_memory(sbi, FREE_NIDS)) 489 try_to_free_nids(sbi, MAX_FREE_NIDS); 490 else 491 build_free_nids(sbi, false, false); 492 493 if (!is_idle(sbi) && !excess_dirty_nats(sbi)) 494 return; 495 496 /* checkpoint is the only way to shrink partial cached entries */ 497 if (!available_free_memory(sbi, NAT_ENTRIES) || 498 !available_free_memory(sbi, INO_ENTRIES) || 499 excess_prefree_segs(sbi) || 500 excess_dirty_nats(sbi) || 501 f2fs_time_over(sbi, CP_TIME)) { 502 if (test_opt(sbi, DATA_FLUSH)) { 503 struct blk_plug plug; 504 505 blk_start_plug(&plug); 506 sync_dirty_inodes(sbi, FILE_INODE); 507 blk_finish_plug(&plug); 508 } 509 f2fs_sync_fs(sbi->sb, true); 510 stat_inc_bg_cp_count(sbi->stat_info); 511 } 512 } 513 514 static int __submit_flush_wait(struct f2fs_sb_info *sbi, 515 struct block_device *bdev) 516 { 517 struct bio *bio = f2fs_bio_alloc(sbi, 0, true); 518 int ret; 519 520 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH; 521 bio_set_dev(bio, bdev); 522 ret = submit_bio_wait(bio); 523 bio_put(bio); 524 525 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER), 526 test_opt(sbi, FLUSH_MERGE), ret); 527 return ret; 528 } 529 530 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino) 531 { 532 int ret = 0; 533 int i; 534 535 if (!sbi->s_ndevs) 536 return __submit_flush_wait(sbi, sbi->sb->s_bdev); 537 538 for (i = 0; i < sbi->s_ndevs; i++) { 539 if (!is_dirty_device(sbi, ino, i, FLUSH_INO)) 540 continue; 541 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 542 if (ret) 543 break; 544 } 545 return ret; 546 } 547 548 static int issue_flush_thread(void *data) 549 { 550 struct f2fs_sb_info *sbi = data; 551 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 552 wait_queue_head_t *q = &fcc->flush_wait_queue; 553 repeat: 554 if (kthread_should_stop()) 555 return 0; 556 557 sb_start_intwrite(sbi->sb); 558 559 if (!llist_empty(&fcc->issue_list)) { 560 struct flush_cmd *cmd, *next; 561 int ret; 562 563 fcc->dispatch_list = llist_del_all(&fcc->issue_list); 564 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); 565 566 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode); 567 568 ret = submit_flush_wait(sbi, cmd->ino); 569 atomic_inc(&fcc->issued_flush); 570 571 llist_for_each_entry_safe(cmd, next, 572 fcc->dispatch_list, llnode) { 573 cmd->ret = ret; 574 complete(&cmd->wait); 575 } 576 fcc->dispatch_list = NULL; 577 } 578 579 sb_end_intwrite(sbi->sb); 580 581 wait_event_interruptible(*q, 582 kthread_should_stop() || !llist_empty(&fcc->issue_list)); 583 goto repeat; 584 } 585 586 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino) 587 { 588 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 589 struct flush_cmd cmd; 590 int ret; 591 592 if (test_opt(sbi, NOBARRIER)) 593 return 0; 594 595 if (!test_opt(sbi, FLUSH_MERGE)) { 596 ret = submit_flush_wait(sbi, ino); 597 atomic_inc(&fcc->issued_flush); 598 return ret; 599 } 600 601 if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) { 602 ret = submit_flush_wait(sbi, ino); 603 atomic_dec(&fcc->issing_flush); 604 605 atomic_inc(&fcc->issued_flush); 606 return ret; 607 } 608 609 cmd.ino = ino; 610 init_completion(&cmd.wait); 611 612 llist_add(&cmd.llnode, &fcc->issue_list); 613 614 /* update issue_list before we wake up issue_flush thread */ 615 smp_mb(); 616 617 if (waitqueue_active(&fcc->flush_wait_queue)) 618 wake_up(&fcc->flush_wait_queue); 619 620 if (fcc->f2fs_issue_flush) { 621 wait_for_completion(&cmd.wait); 622 atomic_dec(&fcc->issing_flush); 623 } else { 624 struct llist_node *list; 625 626 list = llist_del_all(&fcc->issue_list); 627 if (!list) { 628 wait_for_completion(&cmd.wait); 629 atomic_dec(&fcc->issing_flush); 630 } else { 631 struct flush_cmd *tmp, *next; 632 633 ret = submit_flush_wait(sbi, ino); 634 635 llist_for_each_entry_safe(tmp, next, list, llnode) { 636 if (tmp == &cmd) { 637 cmd.ret = ret; 638 atomic_dec(&fcc->issing_flush); 639 continue; 640 } 641 tmp->ret = ret; 642 complete(&tmp->wait); 643 } 644 } 645 } 646 647 return cmd.ret; 648 } 649 650 int create_flush_cmd_control(struct f2fs_sb_info *sbi) 651 { 652 dev_t dev = sbi->sb->s_bdev->bd_dev; 653 struct flush_cmd_control *fcc; 654 int err = 0; 655 656 if (SM_I(sbi)->fcc_info) { 657 fcc = SM_I(sbi)->fcc_info; 658 if (fcc->f2fs_issue_flush) 659 return err; 660 goto init_thread; 661 } 662 663 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL); 664 if (!fcc) 665 return -ENOMEM; 666 atomic_set(&fcc->issued_flush, 0); 667 atomic_set(&fcc->issing_flush, 0); 668 init_waitqueue_head(&fcc->flush_wait_queue); 669 init_llist_head(&fcc->issue_list); 670 SM_I(sbi)->fcc_info = fcc; 671 if (!test_opt(sbi, FLUSH_MERGE)) 672 return err; 673 674 init_thread: 675 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, 676 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); 677 if (IS_ERR(fcc->f2fs_issue_flush)) { 678 err = PTR_ERR(fcc->f2fs_issue_flush); 679 kfree(fcc); 680 SM_I(sbi)->fcc_info = NULL; 681 return err; 682 } 683 684 return err; 685 } 686 687 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free) 688 { 689 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 690 691 if (fcc && fcc->f2fs_issue_flush) { 692 struct task_struct *flush_thread = fcc->f2fs_issue_flush; 693 694 fcc->f2fs_issue_flush = NULL; 695 kthread_stop(flush_thread); 696 } 697 if (free) { 698 kfree(fcc); 699 SM_I(sbi)->fcc_info = NULL; 700 } 701 } 702 703 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi) 704 { 705 int ret = 0, i; 706 707 if (!sbi->s_ndevs) 708 return 0; 709 710 for (i = 1; i < sbi->s_ndevs; i++) { 711 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device)) 712 continue; 713 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 714 if (ret) 715 break; 716 717 spin_lock(&sbi->dev_lock); 718 f2fs_clear_bit(i, (char *)&sbi->dirty_device); 719 spin_unlock(&sbi->dev_lock); 720 } 721 722 return ret; 723 } 724 725 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 726 enum dirty_type dirty_type) 727 { 728 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 729 730 /* need not be added */ 731 if (IS_CURSEG(sbi, segno)) 732 return; 733 734 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 735 dirty_i->nr_dirty[dirty_type]++; 736 737 if (dirty_type == DIRTY) { 738 struct seg_entry *sentry = get_seg_entry(sbi, segno); 739 enum dirty_type t = sentry->type; 740 741 if (unlikely(t >= DIRTY)) { 742 f2fs_bug_on(sbi, 1); 743 return; 744 } 745 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) 746 dirty_i->nr_dirty[t]++; 747 } 748 } 749 750 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 751 enum dirty_type dirty_type) 752 { 753 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 754 755 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) 756 dirty_i->nr_dirty[dirty_type]--; 757 758 if (dirty_type == DIRTY) { 759 struct seg_entry *sentry = get_seg_entry(sbi, segno); 760 enum dirty_type t = sentry->type; 761 762 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) 763 dirty_i->nr_dirty[t]--; 764 765 if (get_valid_blocks(sbi, segno, true) == 0) 766 clear_bit(GET_SEC_FROM_SEG(sbi, segno), 767 dirty_i->victim_secmap); 768 } 769 } 770 771 /* 772 * Should not occur error such as -ENOMEM. 773 * Adding dirty entry into seglist is not critical operation. 774 * If a given segment is one of current working segments, it won't be added. 775 */ 776 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) 777 { 778 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 779 unsigned short valid_blocks; 780 781 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) 782 return; 783 784 mutex_lock(&dirty_i->seglist_lock); 785 786 valid_blocks = get_valid_blocks(sbi, segno, false); 787 788 if (valid_blocks == 0) { 789 __locate_dirty_segment(sbi, segno, PRE); 790 __remove_dirty_segment(sbi, segno, DIRTY); 791 } else if (valid_blocks < sbi->blocks_per_seg) { 792 __locate_dirty_segment(sbi, segno, DIRTY); 793 } else { 794 /* Recovery routine with SSR needs this */ 795 __remove_dirty_segment(sbi, segno, DIRTY); 796 } 797 798 mutex_unlock(&dirty_i->seglist_lock); 799 } 800 801 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi, 802 struct block_device *bdev, block_t lstart, 803 block_t start, block_t len) 804 { 805 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 806 struct list_head *pend_list; 807 struct discard_cmd *dc; 808 809 f2fs_bug_on(sbi, !len); 810 811 pend_list = &dcc->pend_list[plist_idx(len)]; 812 813 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS); 814 INIT_LIST_HEAD(&dc->list); 815 dc->bdev = bdev; 816 dc->lstart = lstart; 817 dc->start = start; 818 dc->len = len; 819 dc->ref = 0; 820 dc->state = D_PREP; 821 dc->error = 0; 822 init_completion(&dc->wait); 823 list_add_tail(&dc->list, pend_list); 824 atomic_inc(&dcc->discard_cmd_cnt); 825 dcc->undiscard_blks += len; 826 827 return dc; 828 } 829 830 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi, 831 struct block_device *bdev, block_t lstart, 832 block_t start, block_t len, 833 struct rb_node *parent, struct rb_node **p) 834 { 835 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 836 struct discard_cmd *dc; 837 838 dc = __create_discard_cmd(sbi, bdev, lstart, start, len); 839 840 rb_link_node(&dc->rb_node, parent, p); 841 rb_insert_color(&dc->rb_node, &dcc->root); 842 843 return dc; 844 } 845 846 static void __detach_discard_cmd(struct discard_cmd_control *dcc, 847 struct discard_cmd *dc) 848 { 849 if (dc->state == D_DONE) 850 atomic_dec(&dcc->issing_discard); 851 852 list_del(&dc->list); 853 rb_erase(&dc->rb_node, &dcc->root); 854 dcc->undiscard_blks -= dc->len; 855 856 kmem_cache_free(discard_cmd_slab, dc); 857 858 atomic_dec(&dcc->discard_cmd_cnt); 859 } 860 861 static void __remove_discard_cmd(struct f2fs_sb_info *sbi, 862 struct discard_cmd *dc) 863 { 864 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 865 866 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len); 867 868 f2fs_bug_on(sbi, dc->ref); 869 870 if (dc->error == -EOPNOTSUPP) 871 dc->error = 0; 872 873 if (dc->error) 874 f2fs_msg(sbi->sb, KERN_INFO, 875 "Issue discard(%u, %u, %u) failed, ret: %d", 876 dc->lstart, dc->start, dc->len, dc->error); 877 __detach_discard_cmd(dcc, dc); 878 } 879 880 static void f2fs_submit_discard_endio(struct bio *bio) 881 { 882 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private; 883 884 dc->error = blk_status_to_errno(bio->bi_status); 885 dc->state = D_DONE; 886 complete_all(&dc->wait); 887 bio_put(bio); 888 } 889 890 static void __check_sit_bitmap(struct f2fs_sb_info *sbi, 891 block_t start, block_t end) 892 { 893 #ifdef CONFIG_F2FS_CHECK_FS 894 struct seg_entry *sentry; 895 unsigned int segno; 896 block_t blk = start; 897 unsigned long offset, size, max_blocks = sbi->blocks_per_seg; 898 unsigned long *map; 899 900 while (blk < end) { 901 segno = GET_SEGNO(sbi, blk); 902 sentry = get_seg_entry(sbi, segno); 903 offset = GET_BLKOFF_FROM_SEG0(sbi, blk); 904 905 if (end < START_BLOCK(sbi, segno + 1)) 906 size = GET_BLKOFF_FROM_SEG0(sbi, end); 907 else 908 size = max_blocks; 909 map = (unsigned long *)(sentry->cur_valid_map); 910 offset = __find_rev_next_bit(map, size, offset); 911 f2fs_bug_on(sbi, offset != size); 912 blk = START_BLOCK(sbi, segno + 1); 913 } 914 #endif 915 } 916 917 static void __init_discard_policy(struct f2fs_sb_info *sbi, 918 struct discard_policy *dpolicy, 919 int discard_type, unsigned int granularity) 920 { 921 /* common policy */ 922 dpolicy->type = discard_type; 923 dpolicy->sync = true; 924 dpolicy->granularity = granularity; 925 926 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST; 927 dpolicy->io_aware_gran = MAX_PLIST_NUM; 928 929 if (discard_type == DPOLICY_BG) { 930 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME; 931 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME; 932 dpolicy->io_aware = true; 933 dpolicy->sync = false; 934 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) { 935 dpolicy->granularity = 1; 936 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME; 937 } 938 } else if (discard_type == DPOLICY_FORCE) { 939 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME; 940 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME; 941 dpolicy->io_aware = false; 942 } else if (discard_type == DPOLICY_FSTRIM) { 943 dpolicy->io_aware = false; 944 } else if (discard_type == DPOLICY_UMOUNT) { 945 dpolicy->max_requests = UINT_MAX; 946 dpolicy->io_aware = false; 947 } 948 } 949 950 951 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */ 952 static void __submit_discard_cmd(struct f2fs_sb_info *sbi, 953 struct discard_policy *dpolicy, 954 struct discard_cmd *dc) 955 { 956 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 957 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 958 &(dcc->fstrim_list) : &(dcc->wait_list); 959 struct bio *bio = NULL; 960 int flag = dpolicy->sync ? REQ_SYNC : 0; 961 962 if (dc->state != D_PREP) 963 return; 964 965 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 966 return; 967 968 trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len); 969 970 dc->error = __blkdev_issue_discard(dc->bdev, 971 SECTOR_FROM_BLOCK(dc->start), 972 SECTOR_FROM_BLOCK(dc->len), 973 GFP_NOFS, 0, &bio); 974 if (!dc->error) { 975 /* should keep before submission to avoid D_DONE right away */ 976 dc->state = D_SUBMIT; 977 atomic_inc(&dcc->issued_discard); 978 atomic_inc(&dcc->issing_discard); 979 if (bio) { 980 bio->bi_private = dc; 981 bio->bi_end_io = f2fs_submit_discard_endio; 982 bio->bi_opf |= flag; 983 submit_bio(bio); 984 list_move_tail(&dc->list, wait_list); 985 __check_sit_bitmap(sbi, dc->start, dc->start + dc->len); 986 987 f2fs_update_iostat(sbi, FS_DISCARD, 1); 988 } 989 } else { 990 __remove_discard_cmd(sbi, dc); 991 } 992 } 993 994 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi, 995 struct block_device *bdev, block_t lstart, 996 block_t start, block_t len, 997 struct rb_node **insert_p, 998 struct rb_node *insert_parent) 999 { 1000 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1001 struct rb_node **p; 1002 struct rb_node *parent = NULL; 1003 struct discard_cmd *dc = NULL; 1004 1005 if (insert_p && insert_parent) { 1006 parent = insert_parent; 1007 p = insert_p; 1008 goto do_insert; 1009 } 1010 1011 p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart); 1012 do_insert: 1013 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p); 1014 if (!dc) 1015 return NULL; 1016 1017 return dc; 1018 } 1019 1020 static void __relocate_discard_cmd(struct discard_cmd_control *dcc, 1021 struct discard_cmd *dc) 1022 { 1023 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]); 1024 } 1025 1026 static void __punch_discard_cmd(struct f2fs_sb_info *sbi, 1027 struct discard_cmd *dc, block_t blkaddr) 1028 { 1029 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1030 struct discard_info di = dc->di; 1031 bool modified = false; 1032 1033 if (dc->state == D_DONE || dc->len == 1) { 1034 __remove_discard_cmd(sbi, dc); 1035 return; 1036 } 1037 1038 dcc->undiscard_blks -= di.len; 1039 1040 if (blkaddr > di.lstart) { 1041 dc->len = blkaddr - dc->lstart; 1042 dcc->undiscard_blks += dc->len; 1043 __relocate_discard_cmd(dcc, dc); 1044 modified = true; 1045 } 1046 1047 if (blkaddr < di.lstart + di.len - 1) { 1048 if (modified) { 1049 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1, 1050 di.start + blkaddr + 1 - di.lstart, 1051 di.lstart + di.len - 1 - blkaddr, 1052 NULL, NULL); 1053 } else { 1054 dc->lstart++; 1055 dc->len--; 1056 dc->start++; 1057 dcc->undiscard_blks += dc->len; 1058 __relocate_discard_cmd(dcc, dc); 1059 } 1060 } 1061 } 1062 1063 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1064 struct block_device *bdev, block_t lstart, 1065 block_t start, block_t len) 1066 { 1067 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1068 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1069 struct discard_cmd *dc; 1070 struct discard_info di = {0}; 1071 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1072 block_t end = lstart + len; 1073 1074 mutex_lock(&dcc->cmd_lock); 1075 1076 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root, 1077 NULL, lstart, 1078 (struct rb_entry **)&prev_dc, 1079 (struct rb_entry **)&next_dc, 1080 &insert_p, &insert_parent, true); 1081 if (dc) 1082 prev_dc = dc; 1083 1084 if (!prev_dc) { 1085 di.lstart = lstart; 1086 di.len = next_dc ? next_dc->lstart - lstart : len; 1087 di.len = min(di.len, len); 1088 di.start = start; 1089 } 1090 1091 while (1) { 1092 struct rb_node *node; 1093 bool merged = false; 1094 struct discard_cmd *tdc = NULL; 1095 1096 if (prev_dc) { 1097 di.lstart = prev_dc->lstart + prev_dc->len; 1098 if (di.lstart < lstart) 1099 di.lstart = lstart; 1100 if (di.lstart >= end) 1101 break; 1102 1103 if (!next_dc || next_dc->lstart > end) 1104 di.len = end - di.lstart; 1105 else 1106 di.len = next_dc->lstart - di.lstart; 1107 di.start = start + di.lstart - lstart; 1108 } 1109 1110 if (!di.len) 1111 goto next; 1112 1113 if (prev_dc && prev_dc->state == D_PREP && 1114 prev_dc->bdev == bdev && 1115 __is_discard_back_mergeable(&di, &prev_dc->di)) { 1116 prev_dc->di.len += di.len; 1117 dcc->undiscard_blks += di.len; 1118 __relocate_discard_cmd(dcc, prev_dc); 1119 di = prev_dc->di; 1120 tdc = prev_dc; 1121 merged = true; 1122 } 1123 1124 if (next_dc && next_dc->state == D_PREP && 1125 next_dc->bdev == bdev && 1126 __is_discard_front_mergeable(&di, &next_dc->di)) { 1127 next_dc->di.lstart = di.lstart; 1128 next_dc->di.len += di.len; 1129 next_dc->di.start = di.start; 1130 dcc->undiscard_blks += di.len; 1131 __relocate_discard_cmd(dcc, next_dc); 1132 if (tdc) 1133 __remove_discard_cmd(sbi, tdc); 1134 merged = true; 1135 } 1136 1137 if (!merged) { 1138 __insert_discard_tree(sbi, bdev, di.lstart, di.start, 1139 di.len, NULL, NULL); 1140 } 1141 next: 1142 prev_dc = next_dc; 1143 if (!prev_dc) 1144 break; 1145 1146 node = rb_next(&prev_dc->rb_node); 1147 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1148 } 1149 1150 mutex_unlock(&dcc->cmd_lock); 1151 } 1152 1153 static int __queue_discard_cmd(struct f2fs_sb_info *sbi, 1154 struct block_device *bdev, block_t blkstart, block_t blklen) 1155 { 1156 block_t lblkstart = blkstart; 1157 1158 trace_f2fs_queue_discard(bdev, blkstart, blklen); 1159 1160 if (sbi->s_ndevs) { 1161 int devi = f2fs_target_device_index(sbi, blkstart); 1162 1163 blkstart -= FDEV(devi).start_blk; 1164 } 1165 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen); 1166 return 0; 1167 } 1168 1169 static int __issue_discard_cmd(struct f2fs_sb_info *sbi, 1170 struct discard_policy *dpolicy) 1171 { 1172 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1173 struct list_head *pend_list; 1174 struct discard_cmd *dc, *tmp; 1175 struct blk_plug plug; 1176 int i, iter = 0, issued = 0; 1177 bool io_interrupted = false; 1178 1179 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1180 if (i + 1 < dpolicy->granularity) 1181 break; 1182 pend_list = &dcc->pend_list[i]; 1183 1184 mutex_lock(&dcc->cmd_lock); 1185 if (list_empty(pend_list)) 1186 goto next; 1187 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root)); 1188 blk_start_plug(&plug); 1189 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1190 f2fs_bug_on(sbi, dc->state != D_PREP); 1191 1192 if (dpolicy->io_aware && i < dpolicy->io_aware_gran && 1193 !is_idle(sbi)) { 1194 io_interrupted = true; 1195 goto skip; 1196 } 1197 1198 __submit_discard_cmd(sbi, dpolicy, dc); 1199 issued++; 1200 skip: 1201 if (++iter >= dpolicy->max_requests) 1202 break; 1203 } 1204 blk_finish_plug(&plug); 1205 next: 1206 mutex_unlock(&dcc->cmd_lock); 1207 1208 if (iter >= dpolicy->max_requests) 1209 break; 1210 } 1211 1212 if (!issued && io_interrupted) 1213 issued = -1; 1214 1215 return issued; 1216 } 1217 1218 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi) 1219 { 1220 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1221 struct list_head *pend_list; 1222 struct discard_cmd *dc, *tmp; 1223 int i; 1224 bool dropped = false; 1225 1226 mutex_lock(&dcc->cmd_lock); 1227 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1228 pend_list = &dcc->pend_list[i]; 1229 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1230 f2fs_bug_on(sbi, dc->state != D_PREP); 1231 __remove_discard_cmd(sbi, dc); 1232 dropped = true; 1233 } 1234 } 1235 mutex_unlock(&dcc->cmd_lock); 1236 1237 return dropped; 1238 } 1239 1240 void drop_discard_cmd(struct f2fs_sb_info *sbi) 1241 { 1242 __drop_discard_cmd(sbi); 1243 } 1244 1245 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi, 1246 struct discard_cmd *dc) 1247 { 1248 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1249 unsigned int len = 0; 1250 1251 wait_for_completion_io(&dc->wait); 1252 mutex_lock(&dcc->cmd_lock); 1253 f2fs_bug_on(sbi, dc->state != D_DONE); 1254 dc->ref--; 1255 if (!dc->ref) { 1256 if (!dc->error) 1257 len = dc->len; 1258 __remove_discard_cmd(sbi, dc); 1259 } 1260 mutex_unlock(&dcc->cmd_lock); 1261 1262 return len; 1263 } 1264 1265 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi, 1266 struct discard_policy *dpolicy, 1267 block_t start, block_t end) 1268 { 1269 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1270 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1271 &(dcc->fstrim_list) : &(dcc->wait_list); 1272 struct discard_cmd *dc, *tmp; 1273 bool need_wait; 1274 unsigned int trimmed = 0; 1275 1276 next: 1277 need_wait = false; 1278 1279 mutex_lock(&dcc->cmd_lock); 1280 list_for_each_entry_safe(dc, tmp, wait_list, list) { 1281 if (dc->lstart + dc->len <= start || end <= dc->lstart) 1282 continue; 1283 if (dc->len < dpolicy->granularity) 1284 continue; 1285 if (dc->state == D_DONE && !dc->ref) { 1286 wait_for_completion_io(&dc->wait); 1287 if (!dc->error) 1288 trimmed += dc->len; 1289 __remove_discard_cmd(sbi, dc); 1290 } else { 1291 dc->ref++; 1292 need_wait = true; 1293 break; 1294 } 1295 } 1296 mutex_unlock(&dcc->cmd_lock); 1297 1298 if (need_wait) { 1299 trimmed += __wait_one_discard_bio(sbi, dc); 1300 goto next; 1301 } 1302 1303 return trimmed; 1304 } 1305 1306 static void __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1307 struct discard_policy *dpolicy) 1308 { 1309 struct discard_policy dp; 1310 1311 if (dpolicy) { 1312 __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX); 1313 return; 1314 } 1315 1316 /* wait all */ 1317 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1); 1318 __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1319 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1); 1320 __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1321 } 1322 1323 /* This should be covered by global mutex, &sit_i->sentry_lock */ 1324 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr) 1325 { 1326 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1327 struct discard_cmd *dc; 1328 bool need_wait = false; 1329 1330 mutex_lock(&dcc->cmd_lock); 1331 dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr); 1332 if (dc) { 1333 if (dc->state == D_PREP) { 1334 __punch_discard_cmd(sbi, dc, blkaddr); 1335 } else { 1336 dc->ref++; 1337 need_wait = true; 1338 } 1339 } 1340 mutex_unlock(&dcc->cmd_lock); 1341 1342 if (need_wait) 1343 __wait_one_discard_bio(sbi, dc); 1344 } 1345 1346 void stop_discard_thread(struct f2fs_sb_info *sbi) 1347 { 1348 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1349 1350 if (dcc && dcc->f2fs_issue_discard) { 1351 struct task_struct *discard_thread = dcc->f2fs_issue_discard; 1352 1353 dcc->f2fs_issue_discard = NULL; 1354 kthread_stop(discard_thread); 1355 } 1356 } 1357 1358 /* This comes from f2fs_put_super */ 1359 bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi) 1360 { 1361 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1362 struct discard_policy dpolicy; 1363 bool dropped; 1364 1365 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT, 1366 dcc->discard_granularity); 1367 __issue_discard_cmd(sbi, &dpolicy); 1368 dropped = __drop_discard_cmd(sbi); 1369 1370 /* just to make sure there is no pending discard commands */ 1371 __wait_all_discard_cmd(sbi, NULL); 1372 return dropped; 1373 } 1374 1375 static int issue_discard_thread(void *data) 1376 { 1377 struct f2fs_sb_info *sbi = data; 1378 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1379 wait_queue_head_t *q = &dcc->discard_wait_queue; 1380 struct discard_policy dpolicy; 1381 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME; 1382 int issued; 1383 1384 set_freezable(); 1385 1386 do { 1387 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG, 1388 dcc->discard_granularity); 1389 1390 wait_event_interruptible_timeout(*q, 1391 kthread_should_stop() || freezing(current) || 1392 dcc->discard_wake, 1393 msecs_to_jiffies(wait_ms)); 1394 if (try_to_freeze()) 1395 continue; 1396 if (f2fs_readonly(sbi->sb)) 1397 continue; 1398 if (kthread_should_stop()) 1399 return 0; 1400 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 1401 wait_ms = dpolicy.max_interval; 1402 continue; 1403 } 1404 1405 if (dcc->discard_wake) 1406 dcc->discard_wake = 0; 1407 1408 if (sbi->gc_mode == GC_URGENT) 1409 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1); 1410 1411 sb_start_intwrite(sbi->sb); 1412 1413 issued = __issue_discard_cmd(sbi, &dpolicy); 1414 if (issued) { 1415 __wait_all_discard_cmd(sbi, &dpolicy); 1416 wait_ms = dpolicy.min_interval; 1417 } else { 1418 wait_ms = dpolicy.max_interval; 1419 } 1420 1421 sb_end_intwrite(sbi->sb); 1422 1423 } while (!kthread_should_stop()); 1424 return 0; 1425 } 1426 1427 #ifdef CONFIG_BLK_DEV_ZONED 1428 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi, 1429 struct block_device *bdev, block_t blkstart, block_t blklen) 1430 { 1431 sector_t sector, nr_sects; 1432 block_t lblkstart = blkstart; 1433 int devi = 0; 1434 1435 if (sbi->s_ndevs) { 1436 devi = f2fs_target_device_index(sbi, blkstart); 1437 blkstart -= FDEV(devi).start_blk; 1438 } 1439 1440 /* 1441 * We need to know the type of the zone: for conventional zones, 1442 * use regular discard if the drive supports it. For sequential 1443 * zones, reset the zone write pointer. 1444 */ 1445 switch (get_blkz_type(sbi, bdev, blkstart)) { 1446 1447 case BLK_ZONE_TYPE_CONVENTIONAL: 1448 if (!blk_queue_discard(bdev_get_queue(bdev))) 1449 return 0; 1450 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen); 1451 case BLK_ZONE_TYPE_SEQWRITE_REQ: 1452 case BLK_ZONE_TYPE_SEQWRITE_PREF: 1453 sector = SECTOR_FROM_BLOCK(blkstart); 1454 nr_sects = SECTOR_FROM_BLOCK(blklen); 1455 1456 if (sector & (bdev_zone_sectors(bdev) - 1) || 1457 nr_sects != bdev_zone_sectors(bdev)) { 1458 f2fs_msg(sbi->sb, KERN_INFO, 1459 "(%d) %s: Unaligned discard attempted (block %x + %x)", 1460 devi, sbi->s_ndevs ? FDEV(devi).path: "", 1461 blkstart, blklen); 1462 return -EIO; 1463 } 1464 trace_f2fs_issue_reset_zone(bdev, blkstart); 1465 return blkdev_reset_zones(bdev, sector, 1466 nr_sects, GFP_NOFS); 1467 default: 1468 /* Unknown zone type: broken device ? */ 1469 return -EIO; 1470 } 1471 } 1472 #endif 1473 1474 static int __issue_discard_async(struct f2fs_sb_info *sbi, 1475 struct block_device *bdev, block_t blkstart, block_t blklen) 1476 { 1477 #ifdef CONFIG_BLK_DEV_ZONED 1478 if (f2fs_sb_has_blkzoned(sbi->sb) && 1479 bdev_zoned_model(bdev) != BLK_ZONED_NONE) 1480 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen); 1481 #endif 1482 return __queue_discard_cmd(sbi, bdev, blkstart, blklen); 1483 } 1484 1485 static int f2fs_issue_discard(struct f2fs_sb_info *sbi, 1486 block_t blkstart, block_t blklen) 1487 { 1488 sector_t start = blkstart, len = 0; 1489 struct block_device *bdev; 1490 struct seg_entry *se; 1491 unsigned int offset; 1492 block_t i; 1493 int err = 0; 1494 1495 bdev = f2fs_target_device(sbi, blkstart, NULL); 1496 1497 for (i = blkstart; i < blkstart + blklen; i++, len++) { 1498 if (i != start) { 1499 struct block_device *bdev2 = 1500 f2fs_target_device(sbi, i, NULL); 1501 1502 if (bdev2 != bdev) { 1503 err = __issue_discard_async(sbi, bdev, 1504 start, len); 1505 if (err) 1506 return err; 1507 bdev = bdev2; 1508 start = i; 1509 len = 0; 1510 } 1511 } 1512 1513 se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); 1514 offset = GET_BLKOFF_FROM_SEG0(sbi, i); 1515 1516 if (!f2fs_test_and_set_bit(offset, se->discard_map)) 1517 sbi->discard_blks--; 1518 } 1519 1520 if (len) 1521 err = __issue_discard_async(sbi, bdev, start, len); 1522 return err; 1523 } 1524 1525 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc, 1526 bool check_only) 1527 { 1528 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 1529 int max_blocks = sbi->blocks_per_seg; 1530 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); 1531 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 1532 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 1533 unsigned long *discard_map = (unsigned long *)se->discard_map; 1534 unsigned long *dmap = SIT_I(sbi)->tmp_map; 1535 unsigned int start = 0, end = -1; 1536 bool force = (cpc->reason & CP_DISCARD); 1537 struct discard_entry *de = NULL; 1538 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list; 1539 int i; 1540 1541 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi)) 1542 return false; 1543 1544 if (!force) { 1545 if (!test_opt(sbi, DISCARD) || !se->valid_blocks || 1546 SM_I(sbi)->dcc_info->nr_discards >= 1547 SM_I(sbi)->dcc_info->max_discards) 1548 return false; 1549 } 1550 1551 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ 1552 for (i = 0; i < entries; i++) 1553 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : 1554 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; 1555 1556 while (force || SM_I(sbi)->dcc_info->nr_discards <= 1557 SM_I(sbi)->dcc_info->max_discards) { 1558 start = __find_rev_next_bit(dmap, max_blocks, end + 1); 1559 if (start >= max_blocks) 1560 break; 1561 1562 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); 1563 if (force && start && end != max_blocks 1564 && (end - start) < cpc->trim_minlen) 1565 continue; 1566 1567 if (check_only) 1568 return true; 1569 1570 if (!de) { 1571 de = f2fs_kmem_cache_alloc(discard_entry_slab, 1572 GFP_F2FS_ZERO); 1573 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start); 1574 list_add_tail(&de->list, head); 1575 } 1576 1577 for (i = start; i < end; i++) 1578 __set_bit_le(i, (void *)de->discard_map); 1579 1580 SM_I(sbi)->dcc_info->nr_discards += end - start; 1581 } 1582 return false; 1583 } 1584 1585 static void release_discard_addr(struct discard_entry *entry) 1586 { 1587 list_del(&entry->list); 1588 kmem_cache_free(discard_entry_slab, entry); 1589 } 1590 1591 void release_discard_addrs(struct f2fs_sb_info *sbi) 1592 { 1593 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); 1594 struct discard_entry *entry, *this; 1595 1596 /* drop caches */ 1597 list_for_each_entry_safe(entry, this, head, list) 1598 release_discard_addr(entry); 1599 } 1600 1601 /* 1602 * Should call clear_prefree_segments after checkpoint is done. 1603 */ 1604 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 1605 { 1606 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1607 unsigned int segno; 1608 1609 mutex_lock(&dirty_i->seglist_lock); 1610 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) 1611 __set_test_and_free(sbi, segno); 1612 mutex_unlock(&dirty_i->seglist_lock); 1613 } 1614 1615 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1616 { 1617 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1618 struct list_head *head = &dcc->entry_list; 1619 struct discard_entry *entry, *this; 1620 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1621 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; 1622 unsigned int start = 0, end = -1; 1623 unsigned int secno, start_segno; 1624 bool force = (cpc->reason & CP_DISCARD); 1625 1626 mutex_lock(&dirty_i->seglist_lock); 1627 1628 while (1) { 1629 int i; 1630 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); 1631 if (start >= MAIN_SEGS(sbi)) 1632 break; 1633 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), 1634 start + 1); 1635 1636 for (i = start; i < end; i++) 1637 clear_bit(i, prefree_map); 1638 1639 dirty_i->nr_dirty[PRE] -= end - start; 1640 1641 if (!test_opt(sbi, DISCARD)) 1642 continue; 1643 1644 if (force && start >= cpc->trim_start && 1645 (end - 1) <= cpc->trim_end) 1646 continue; 1647 1648 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) { 1649 f2fs_issue_discard(sbi, START_BLOCK(sbi, start), 1650 (end - start) << sbi->log_blocks_per_seg); 1651 continue; 1652 } 1653 next: 1654 secno = GET_SEC_FROM_SEG(sbi, start); 1655 start_segno = GET_SEG_FROM_SEC(sbi, secno); 1656 if (!IS_CURSEC(sbi, secno) && 1657 !get_valid_blocks(sbi, start, true)) 1658 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno), 1659 sbi->segs_per_sec << sbi->log_blocks_per_seg); 1660 1661 start = start_segno + sbi->segs_per_sec; 1662 if (start < end) 1663 goto next; 1664 else 1665 end = start - 1; 1666 } 1667 mutex_unlock(&dirty_i->seglist_lock); 1668 1669 /* send small discards */ 1670 list_for_each_entry_safe(entry, this, head, list) { 1671 unsigned int cur_pos = 0, next_pos, len, total_len = 0; 1672 bool is_valid = test_bit_le(0, entry->discard_map); 1673 1674 find_next: 1675 if (is_valid) { 1676 next_pos = find_next_zero_bit_le(entry->discard_map, 1677 sbi->blocks_per_seg, cur_pos); 1678 len = next_pos - cur_pos; 1679 1680 if (f2fs_sb_has_blkzoned(sbi->sb) || 1681 (force && len < cpc->trim_minlen)) 1682 goto skip; 1683 1684 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos, 1685 len); 1686 total_len += len; 1687 } else { 1688 next_pos = find_next_bit_le(entry->discard_map, 1689 sbi->blocks_per_seg, cur_pos); 1690 } 1691 skip: 1692 cur_pos = next_pos; 1693 is_valid = !is_valid; 1694 1695 if (cur_pos < sbi->blocks_per_seg) 1696 goto find_next; 1697 1698 release_discard_addr(entry); 1699 dcc->nr_discards -= total_len; 1700 } 1701 1702 wake_up_discard_thread(sbi, false); 1703 } 1704 1705 static int create_discard_cmd_control(struct f2fs_sb_info *sbi) 1706 { 1707 dev_t dev = sbi->sb->s_bdev->bd_dev; 1708 struct discard_cmd_control *dcc; 1709 int err = 0, i; 1710 1711 if (SM_I(sbi)->dcc_info) { 1712 dcc = SM_I(sbi)->dcc_info; 1713 goto init_thread; 1714 } 1715 1716 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL); 1717 if (!dcc) 1718 return -ENOMEM; 1719 1720 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY; 1721 INIT_LIST_HEAD(&dcc->entry_list); 1722 for (i = 0; i < MAX_PLIST_NUM; i++) 1723 INIT_LIST_HEAD(&dcc->pend_list[i]); 1724 INIT_LIST_HEAD(&dcc->wait_list); 1725 INIT_LIST_HEAD(&dcc->fstrim_list); 1726 mutex_init(&dcc->cmd_lock); 1727 atomic_set(&dcc->issued_discard, 0); 1728 atomic_set(&dcc->issing_discard, 0); 1729 atomic_set(&dcc->discard_cmd_cnt, 0); 1730 dcc->nr_discards = 0; 1731 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg; 1732 dcc->undiscard_blks = 0; 1733 dcc->root = RB_ROOT; 1734 1735 init_waitqueue_head(&dcc->discard_wait_queue); 1736 SM_I(sbi)->dcc_info = dcc; 1737 init_thread: 1738 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi, 1739 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev)); 1740 if (IS_ERR(dcc->f2fs_issue_discard)) { 1741 err = PTR_ERR(dcc->f2fs_issue_discard); 1742 kfree(dcc); 1743 SM_I(sbi)->dcc_info = NULL; 1744 return err; 1745 } 1746 1747 return err; 1748 } 1749 1750 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi) 1751 { 1752 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1753 1754 if (!dcc) 1755 return; 1756 1757 stop_discard_thread(sbi); 1758 1759 kfree(dcc); 1760 SM_I(sbi)->dcc_info = NULL; 1761 } 1762 1763 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 1764 { 1765 struct sit_info *sit_i = SIT_I(sbi); 1766 1767 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { 1768 sit_i->dirty_sentries++; 1769 return false; 1770 } 1771 1772 return true; 1773 } 1774 1775 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 1776 unsigned int segno, int modified) 1777 { 1778 struct seg_entry *se = get_seg_entry(sbi, segno); 1779 se->type = type; 1780 if (modified) 1781 __mark_sit_entry_dirty(sbi, segno); 1782 } 1783 1784 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 1785 { 1786 struct seg_entry *se; 1787 unsigned int segno, offset; 1788 long int new_vblocks; 1789 bool exist; 1790 #ifdef CONFIG_F2FS_CHECK_FS 1791 bool mir_exist; 1792 #endif 1793 1794 segno = GET_SEGNO(sbi, blkaddr); 1795 1796 se = get_seg_entry(sbi, segno); 1797 new_vblocks = se->valid_blocks + del; 1798 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 1799 1800 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) || 1801 (new_vblocks > sbi->blocks_per_seg))); 1802 1803 se->valid_blocks = new_vblocks; 1804 se->mtime = get_mtime(sbi); 1805 SIT_I(sbi)->max_mtime = se->mtime; 1806 1807 /* Update valid block bitmap */ 1808 if (del > 0) { 1809 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map); 1810 #ifdef CONFIG_F2FS_CHECK_FS 1811 mir_exist = f2fs_test_and_set_bit(offset, 1812 se->cur_valid_map_mir); 1813 if (unlikely(exist != mir_exist)) { 1814 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error " 1815 "when setting bitmap, blk:%u, old bit:%d", 1816 blkaddr, exist); 1817 f2fs_bug_on(sbi, 1); 1818 } 1819 #endif 1820 if (unlikely(exist)) { 1821 f2fs_msg(sbi->sb, KERN_ERR, 1822 "Bitmap was wrongly set, blk:%u", blkaddr); 1823 f2fs_bug_on(sbi, 1); 1824 se->valid_blocks--; 1825 del = 0; 1826 } 1827 1828 if (f2fs_discard_en(sbi) && 1829 !f2fs_test_and_set_bit(offset, se->discard_map)) 1830 sbi->discard_blks--; 1831 1832 /* don't overwrite by SSR to keep node chain */ 1833 if (IS_NODESEG(se->type)) { 1834 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map)) 1835 se->ckpt_valid_blocks++; 1836 } 1837 } else { 1838 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map); 1839 #ifdef CONFIG_F2FS_CHECK_FS 1840 mir_exist = f2fs_test_and_clear_bit(offset, 1841 se->cur_valid_map_mir); 1842 if (unlikely(exist != mir_exist)) { 1843 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error " 1844 "when clearing bitmap, blk:%u, old bit:%d", 1845 blkaddr, exist); 1846 f2fs_bug_on(sbi, 1); 1847 } 1848 #endif 1849 if (unlikely(!exist)) { 1850 f2fs_msg(sbi->sb, KERN_ERR, 1851 "Bitmap was wrongly cleared, blk:%u", blkaddr); 1852 f2fs_bug_on(sbi, 1); 1853 se->valid_blocks++; 1854 del = 0; 1855 } 1856 1857 if (f2fs_discard_en(sbi) && 1858 f2fs_test_and_clear_bit(offset, se->discard_map)) 1859 sbi->discard_blks++; 1860 } 1861 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) 1862 se->ckpt_valid_blocks += del; 1863 1864 __mark_sit_entry_dirty(sbi, segno); 1865 1866 /* update total number of valid blocks to be written in ckpt area */ 1867 SIT_I(sbi)->written_valid_blocks += del; 1868 1869 if (sbi->segs_per_sec > 1) 1870 get_sec_entry(sbi, segno)->valid_blocks += del; 1871 } 1872 1873 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) 1874 { 1875 unsigned int segno = GET_SEGNO(sbi, addr); 1876 struct sit_info *sit_i = SIT_I(sbi); 1877 1878 f2fs_bug_on(sbi, addr == NULL_ADDR); 1879 if (addr == NEW_ADDR) 1880 return; 1881 1882 /* add it into sit main buffer */ 1883 down_write(&sit_i->sentry_lock); 1884 1885 update_sit_entry(sbi, addr, -1); 1886 1887 /* add it into dirty seglist */ 1888 locate_dirty_segment(sbi, segno); 1889 1890 up_write(&sit_i->sentry_lock); 1891 } 1892 1893 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr) 1894 { 1895 struct sit_info *sit_i = SIT_I(sbi); 1896 unsigned int segno, offset; 1897 struct seg_entry *se; 1898 bool is_cp = false; 1899 1900 if (!is_valid_blkaddr(blkaddr)) 1901 return true; 1902 1903 down_read(&sit_i->sentry_lock); 1904 1905 segno = GET_SEGNO(sbi, blkaddr); 1906 se = get_seg_entry(sbi, segno); 1907 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 1908 1909 if (f2fs_test_bit(offset, se->ckpt_valid_map)) 1910 is_cp = true; 1911 1912 up_read(&sit_i->sentry_lock); 1913 1914 return is_cp; 1915 } 1916 1917 /* 1918 * This function should be resided under the curseg_mutex lock 1919 */ 1920 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, 1921 struct f2fs_summary *sum) 1922 { 1923 struct curseg_info *curseg = CURSEG_I(sbi, type); 1924 void *addr = curseg->sum_blk; 1925 addr += curseg->next_blkoff * sizeof(struct f2fs_summary); 1926 memcpy(addr, sum, sizeof(struct f2fs_summary)); 1927 } 1928 1929 /* 1930 * Calculate the number of current summary pages for writing 1931 */ 1932 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) 1933 { 1934 int valid_sum_count = 0; 1935 int i, sum_in_page; 1936 1937 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 1938 if (sbi->ckpt->alloc_type[i] == SSR) 1939 valid_sum_count += sbi->blocks_per_seg; 1940 else { 1941 if (for_ra) 1942 valid_sum_count += le16_to_cpu( 1943 F2FS_CKPT(sbi)->cur_data_blkoff[i]); 1944 else 1945 valid_sum_count += curseg_blkoff(sbi, i); 1946 } 1947 } 1948 1949 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE - 1950 SUM_FOOTER_SIZE) / SUMMARY_SIZE; 1951 if (valid_sum_count <= sum_in_page) 1952 return 1; 1953 else if ((valid_sum_count - sum_in_page) <= 1954 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) 1955 return 2; 1956 return 3; 1957 } 1958 1959 /* 1960 * Caller should put this summary page 1961 */ 1962 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) 1963 { 1964 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno)); 1965 } 1966 1967 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr) 1968 { 1969 struct page *page = grab_meta_page(sbi, blk_addr); 1970 1971 memcpy(page_address(page), src, PAGE_SIZE); 1972 set_page_dirty(page); 1973 f2fs_put_page(page, 1); 1974 } 1975 1976 static void write_sum_page(struct f2fs_sb_info *sbi, 1977 struct f2fs_summary_block *sum_blk, block_t blk_addr) 1978 { 1979 update_meta_page(sbi, (void *)sum_blk, blk_addr); 1980 } 1981 1982 static void write_current_sum_page(struct f2fs_sb_info *sbi, 1983 int type, block_t blk_addr) 1984 { 1985 struct curseg_info *curseg = CURSEG_I(sbi, type); 1986 struct page *page = grab_meta_page(sbi, blk_addr); 1987 struct f2fs_summary_block *src = curseg->sum_blk; 1988 struct f2fs_summary_block *dst; 1989 1990 dst = (struct f2fs_summary_block *)page_address(page); 1991 memset(dst, 0, PAGE_SIZE); 1992 1993 mutex_lock(&curseg->curseg_mutex); 1994 1995 down_read(&curseg->journal_rwsem); 1996 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE); 1997 up_read(&curseg->journal_rwsem); 1998 1999 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE); 2000 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE); 2001 2002 mutex_unlock(&curseg->curseg_mutex); 2003 2004 set_page_dirty(page); 2005 f2fs_put_page(page, 1); 2006 } 2007 2008 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type) 2009 { 2010 struct curseg_info *curseg = CURSEG_I(sbi, type); 2011 unsigned int segno = curseg->segno + 1; 2012 struct free_segmap_info *free_i = FREE_I(sbi); 2013 2014 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec) 2015 return !test_bit(segno, free_i->free_segmap); 2016 return 0; 2017 } 2018 2019 /* 2020 * Find a new segment from the free segments bitmap to right order 2021 * This function should be returned with success, otherwise BUG 2022 */ 2023 static void get_new_segment(struct f2fs_sb_info *sbi, 2024 unsigned int *newseg, bool new_sec, int dir) 2025 { 2026 struct free_segmap_info *free_i = FREE_I(sbi); 2027 unsigned int segno, secno, zoneno; 2028 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; 2029 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg); 2030 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg); 2031 unsigned int left_start = hint; 2032 bool init = true; 2033 int go_left = 0; 2034 int i; 2035 2036 spin_lock(&free_i->segmap_lock); 2037 2038 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { 2039 segno = find_next_zero_bit(free_i->free_segmap, 2040 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1); 2041 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1)) 2042 goto got_it; 2043 } 2044 find_other_zone: 2045 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 2046 if (secno >= MAIN_SECS(sbi)) { 2047 if (dir == ALLOC_RIGHT) { 2048 secno = find_next_zero_bit(free_i->free_secmap, 2049 MAIN_SECS(sbi), 0); 2050 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi)); 2051 } else { 2052 go_left = 1; 2053 left_start = hint - 1; 2054 } 2055 } 2056 if (go_left == 0) 2057 goto skip_left; 2058 2059 while (test_bit(left_start, free_i->free_secmap)) { 2060 if (left_start > 0) { 2061 left_start--; 2062 continue; 2063 } 2064 left_start = find_next_zero_bit(free_i->free_secmap, 2065 MAIN_SECS(sbi), 0); 2066 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi)); 2067 break; 2068 } 2069 secno = left_start; 2070 skip_left: 2071 segno = GET_SEG_FROM_SEC(sbi, secno); 2072 zoneno = GET_ZONE_FROM_SEC(sbi, secno); 2073 2074 /* give up on finding another zone */ 2075 if (!init) 2076 goto got_it; 2077 if (sbi->secs_per_zone == 1) 2078 goto got_it; 2079 if (zoneno == old_zoneno) 2080 goto got_it; 2081 if (dir == ALLOC_LEFT) { 2082 if (!go_left && zoneno + 1 >= total_zones) 2083 goto got_it; 2084 if (go_left && zoneno == 0) 2085 goto got_it; 2086 } 2087 for (i = 0; i < NR_CURSEG_TYPE; i++) 2088 if (CURSEG_I(sbi, i)->zone == zoneno) 2089 break; 2090 2091 if (i < NR_CURSEG_TYPE) { 2092 /* zone is in user, try another */ 2093 if (go_left) 2094 hint = zoneno * sbi->secs_per_zone - 1; 2095 else if (zoneno + 1 >= total_zones) 2096 hint = 0; 2097 else 2098 hint = (zoneno + 1) * sbi->secs_per_zone; 2099 init = false; 2100 goto find_other_zone; 2101 } 2102 got_it: 2103 /* set it as dirty segment in free segmap */ 2104 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); 2105 __set_inuse(sbi, segno); 2106 *newseg = segno; 2107 spin_unlock(&free_i->segmap_lock); 2108 } 2109 2110 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 2111 { 2112 struct curseg_info *curseg = CURSEG_I(sbi, type); 2113 struct summary_footer *sum_footer; 2114 2115 curseg->segno = curseg->next_segno; 2116 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno); 2117 curseg->next_blkoff = 0; 2118 curseg->next_segno = NULL_SEGNO; 2119 2120 sum_footer = &(curseg->sum_blk->footer); 2121 memset(sum_footer, 0, sizeof(struct summary_footer)); 2122 if (IS_DATASEG(type)) 2123 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 2124 if (IS_NODESEG(type)) 2125 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 2126 __set_sit_entry_type(sbi, type, curseg->segno, modified); 2127 } 2128 2129 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type) 2130 { 2131 /* if segs_per_sec is large than 1, we need to keep original policy. */ 2132 if (sbi->segs_per_sec != 1) 2133 return CURSEG_I(sbi, type)->segno; 2134 2135 if (test_opt(sbi, NOHEAP) && 2136 (type == CURSEG_HOT_DATA || IS_NODESEG(type))) 2137 return 0; 2138 2139 if (SIT_I(sbi)->last_victim[ALLOC_NEXT]) 2140 return SIT_I(sbi)->last_victim[ALLOC_NEXT]; 2141 2142 /* find segments from 0 to reuse freed segments */ 2143 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE) 2144 return 0; 2145 2146 return CURSEG_I(sbi, type)->segno; 2147 } 2148 2149 /* 2150 * Allocate a current working segment. 2151 * This function always allocates a free segment in LFS manner. 2152 */ 2153 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 2154 { 2155 struct curseg_info *curseg = CURSEG_I(sbi, type); 2156 unsigned int segno = curseg->segno; 2157 int dir = ALLOC_LEFT; 2158 2159 write_sum_page(sbi, curseg->sum_blk, 2160 GET_SUM_BLOCK(sbi, segno)); 2161 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA) 2162 dir = ALLOC_RIGHT; 2163 2164 if (test_opt(sbi, NOHEAP)) 2165 dir = ALLOC_RIGHT; 2166 2167 segno = __get_next_segno(sbi, type); 2168 get_new_segment(sbi, &segno, new_sec, dir); 2169 curseg->next_segno = segno; 2170 reset_curseg(sbi, type, 1); 2171 curseg->alloc_type = LFS; 2172 } 2173 2174 static void __next_free_blkoff(struct f2fs_sb_info *sbi, 2175 struct curseg_info *seg, block_t start) 2176 { 2177 struct seg_entry *se = get_seg_entry(sbi, seg->segno); 2178 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 2179 unsigned long *target_map = SIT_I(sbi)->tmp_map; 2180 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 2181 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 2182 int i, pos; 2183 2184 for (i = 0; i < entries; i++) 2185 target_map[i] = ckpt_map[i] | cur_map[i]; 2186 2187 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); 2188 2189 seg->next_blkoff = pos; 2190 } 2191 2192 /* 2193 * If a segment is written by LFS manner, next block offset is just obtained 2194 * by increasing the current block offset. However, if a segment is written by 2195 * SSR manner, next block offset obtained by calling __next_free_blkoff 2196 */ 2197 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, 2198 struct curseg_info *seg) 2199 { 2200 if (seg->alloc_type == SSR) 2201 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1); 2202 else 2203 seg->next_blkoff++; 2204 } 2205 2206 /* 2207 * This function always allocates a used segment(from dirty seglist) by SSR 2208 * manner, so it should recover the existing segment information of valid blocks 2209 */ 2210 static void change_curseg(struct f2fs_sb_info *sbi, int type) 2211 { 2212 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2213 struct curseg_info *curseg = CURSEG_I(sbi, type); 2214 unsigned int new_segno = curseg->next_segno; 2215 struct f2fs_summary_block *sum_node; 2216 struct page *sum_page; 2217 2218 write_sum_page(sbi, curseg->sum_blk, 2219 GET_SUM_BLOCK(sbi, curseg->segno)); 2220 __set_test_and_inuse(sbi, new_segno); 2221 2222 mutex_lock(&dirty_i->seglist_lock); 2223 __remove_dirty_segment(sbi, new_segno, PRE); 2224 __remove_dirty_segment(sbi, new_segno, DIRTY); 2225 mutex_unlock(&dirty_i->seglist_lock); 2226 2227 reset_curseg(sbi, type, 1); 2228 curseg->alloc_type = SSR; 2229 __next_free_blkoff(sbi, curseg, 0); 2230 2231 sum_page = get_sum_page(sbi, new_segno); 2232 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 2233 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); 2234 f2fs_put_page(sum_page, 1); 2235 } 2236 2237 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type) 2238 { 2239 struct curseg_info *curseg = CURSEG_I(sbi, type); 2240 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; 2241 unsigned segno = NULL_SEGNO; 2242 int i, cnt; 2243 bool reversed = false; 2244 2245 /* need_SSR() already forces to do this */ 2246 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) { 2247 curseg->next_segno = segno; 2248 return 1; 2249 } 2250 2251 /* For node segments, let's do SSR more intensively */ 2252 if (IS_NODESEG(type)) { 2253 if (type >= CURSEG_WARM_NODE) { 2254 reversed = true; 2255 i = CURSEG_COLD_NODE; 2256 } else { 2257 i = CURSEG_HOT_NODE; 2258 } 2259 cnt = NR_CURSEG_NODE_TYPE; 2260 } else { 2261 if (type >= CURSEG_WARM_DATA) { 2262 reversed = true; 2263 i = CURSEG_COLD_DATA; 2264 } else { 2265 i = CURSEG_HOT_DATA; 2266 } 2267 cnt = NR_CURSEG_DATA_TYPE; 2268 } 2269 2270 for (; cnt-- > 0; reversed ? i-- : i++) { 2271 if (i == type) 2272 continue; 2273 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) { 2274 curseg->next_segno = segno; 2275 return 1; 2276 } 2277 } 2278 return 0; 2279 } 2280 2281 /* 2282 * flush out current segment and replace it with new segment 2283 * This function should be returned with success, otherwise BUG 2284 */ 2285 static void allocate_segment_by_default(struct f2fs_sb_info *sbi, 2286 int type, bool force) 2287 { 2288 struct curseg_info *curseg = CURSEG_I(sbi, type); 2289 2290 if (force) 2291 new_curseg(sbi, type, true); 2292 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) && 2293 type == CURSEG_WARM_NODE) 2294 new_curseg(sbi, type, false); 2295 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type)) 2296 new_curseg(sbi, type, false); 2297 else if (need_SSR(sbi) && get_ssr_segment(sbi, type)) 2298 change_curseg(sbi, type); 2299 else 2300 new_curseg(sbi, type, false); 2301 2302 stat_inc_seg_type(sbi, curseg); 2303 } 2304 2305 void allocate_new_segments(struct f2fs_sb_info *sbi) 2306 { 2307 struct curseg_info *curseg; 2308 unsigned int old_segno; 2309 int i; 2310 2311 down_write(&SIT_I(sbi)->sentry_lock); 2312 2313 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2314 curseg = CURSEG_I(sbi, i); 2315 old_segno = curseg->segno; 2316 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true); 2317 locate_dirty_segment(sbi, old_segno); 2318 } 2319 2320 up_write(&SIT_I(sbi)->sentry_lock); 2321 } 2322 2323 static const struct segment_allocation default_salloc_ops = { 2324 .allocate_segment = allocate_segment_by_default, 2325 }; 2326 2327 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc) 2328 { 2329 __u64 trim_start = cpc->trim_start; 2330 bool has_candidate = false; 2331 2332 down_write(&SIT_I(sbi)->sentry_lock); 2333 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) { 2334 if (add_discard_addrs(sbi, cpc, true)) { 2335 has_candidate = true; 2336 break; 2337 } 2338 } 2339 up_write(&SIT_I(sbi)->sentry_lock); 2340 2341 cpc->trim_start = trim_start; 2342 return has_candidate; 2343 } 2344 2345 static void __issue_discard_cmd_range(struct f2fs_sb_info *sbi, 2346 struct discard_policy *dpolicy, 2347 unsigned int start, unsigned int end) 2348 { 2349 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2350 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 2351 struct rb_node **insert_p = NULL, *insert_parent = NULL; 2352 struct discard_cmd *dc; 2353 struct blk_plug plug; 2354 int issued; 2355 2356 next: 2357 issued = 0; 2358 2359 mutex_lock(&dcc->cmd_lock); 2360 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root)); 2361 2362 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root, 2363 NULL, start, 2364 (struct rb_entry **)&prev_dc, 2365 (struct rb_entry **)&next_dc, 2366 &insert_p, &insert_parent, true); 2367 if (!dc) 2368 dc = next_dc; 2369 2370 blk_start_plug(&plug); 2371 2372 while (dc && dc->lstart <= end) { 2373 struct rb_node *node; 2374 2375 if (dc->len < dpolicy->granularity) 2376 goto skip; 2377 2378 if (dc->state != D_PREP) { 2379 list_move_tail(&dc->list, &dcc->fstrim_list); 2380 goto skip; 2381 } 2382 2383 __submit_discard_cmd(sbi, dpolicy, dc); 2384 2385 if (++issued >= dpolicy->max_requests) { 2386 start = dc->lstart + dc->len; 2387 2388 blk_finish_plug(&plug); 2389 mutex_unlock(&dcc->cmd_lock); 2390 __wait_all_discard_cmd(sbi, NULL); 2391 congestion_wait(BLK_RW_ASYNC, HZ/50); 2392 goto next; 2393 } 2394 skip: 2395 node = rb_next(&dc->rb_node); 2396 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 2397 2398 if (fatal_signal_pending(current)) 2399 break; 2400 } 2401 2402 blk_finish_plug(&plug); 2403 mutex_unlock(&dcc->cmd_lock); 2404 } 2405 2406 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) 2407 { 2408 __u64 start = F2FS_BYTES_TO_BLK(range->start); 2409 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; 2410 unsigned int start_segno, end_segno; 2411 block_t start_block, end_block; 2412 struct cp_control cpc; 2413 struct discard_policy dpolicy; 2414 unsigned long long trimmed = 0; 2415 int err = 0; 2416 2417 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize) 2418 return -EINVAL; 2419 2420 if (end <= MAIN_BLKADDR(sbi)) 2421 return -EINVAL; 2422 2423 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 2424 f2fs_msg(sbi->sb, KERN_WARNING, 2425 "Found FS corruption, run fsck to fix."); 2426 return -EIO; 2427 } 2428 2429 /* start/end segment number in main_area */ 2430 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); 2431 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : 2432 GET_SEGNO(sbi, end); 2433 2434 cpc.reason = CP_DISCARD; 2435 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen)); 2436 cpc.trim_start = start_segno; 2437 cpc.trim_end = end_segno; 2438 2439 if (sbi->discard_blks == 0) 2440 goto out; 2441 2442 mutex_lock(&sbi->gc_mutex); 2443 err = write_checkpoint(sbi, &cpc); 2444 mutex_unlock(&sbi->gc_mutex); 2445 if (err) 2446 goto out; 2447 2448 start_block = START_BLOCK(sbi, start_segno); 2449 end_block = START_BLOCK(sbi, end_segno + 1); 2450 2451 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen); 2452 __issue_discard_cmd_range(sbi, &dpolicy, start_block, end_block); 2453 2454 /* 2455 * We filed discard candidates, but actually we don't need to wait for 2456 * all of them, since they'll be issued in idle time along with runtime 2457 * discard option. User configuration looks like using runtime discard 2458 * or periodic fstrim instead of it. 2459 */ 2460 if (!test_opt(sbi, DISCARD)) { 2461 trimmed = __wait_discard_cmd_range(sbi, &dpolicy, 2462 start_block, end_block); 2463 range->len = F2FS_BLK_TO_BYTES(trimmed); 2464 } 2465 out: 2466 return err; 2467 } 2468 2469 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type) 2470 { 2471 struct curseg_info *curseg = CURSEG_I(sbi, type); 2472 if (curseg->next_blkoff < sbi->blocks_per_seg) 2473 return true; 2474 return false; 2475 } 2476 2477 int rw_hint_to_seg_type(enum rw_hint hint) 2478 { 2479 switch (hint) { 2480 case WRITE_LIFE_SHORT: 2481 return CURSEG_HOT_DATA; 2482 case WRITE_LIFE_EXTREME: 2483 return CURSEG_COLD_DATA; 2484 default: 2485 return CURSEG_WARM_DATA; 2486 } 2487 } 2488 2489 /* This returns write hints for each segment type. This hints will be 2490 * passed down to block layer. There are mapping tables which depend on 2491 * the mount option 'whint_mode'. 2492 * 2493 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET. 2494 * 2495 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users. 2496 * 2497 * User F2FS Block 2498 * ---- ---- ----- 2499 * META WRITE_LIFE_NOT_SET 2500 * HOT_NODE " 2501 * WARM_NODE " 2502 * COLD_NODE " 2503 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME 2504 * extension list " " 2505 * 2506 * -- buffered io 2507 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 2508 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 2509 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 2510 * WRITE_LIFE_NONE " " 2511 * WRITE_LIFE_MEDIUM " " 2512 * WRITE_LIFE_LONG " " 2513 * 2514 * -- direct io 2515 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 2516 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 2517 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 2518 * WRITE_LIFE_NONE " WRITE_LIFE_NONE 2519 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM 2520 * WRITE_LIFE_LONG " WRITE_LIFE_LONG 2521 * 2522 * 3) whint_mode=fs-based. F2FS passes down hints with its policy. 2523 * 2524 * User F2FS Block 2525 * ---- ---- ----- 2526 * META WRITE_LIFE_MEDIUM; 2527 * HOT_NODE WRITE_LIFE_NOT_SET 2528 * WARM_NODE " 2529 * COLD_NODE WRITE_LIFE_NONE 2530 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME 2531 * extension list " " 2532 * 2533 * -- buffered io 2534 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 2535 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 2536 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG 2537 * WRITE_LIFE_NONE " " 2538 * WRITE_LIFE_MEDIUM " " 2539 * WRITE_LIFE_LONG " " 2540 * 2541 * -- direct io 2542 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 2543 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 2544 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 2545 * WRITE_LIFE_NONE " WRITE_LIFE_NONE 2546 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM 2547 * WRITE_LIFE_LONG " WRITE_LIFE_LONG 2548 */ 2549 2550 enum rw_hint io_type_to_rw_hint(struct f2fs_sb_info *sbi, 2551 enum page_type type, enum temp_type temp) 2552 { 2553 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) { 2554 if (type == DATA) { 2555 if (temp == WARM) 2556 return WRITE_LIFE_NOT_SET; 2557 else if (temp == HOT) 2558 return WRITE_LIFE_SHORT; 2559 else if (temp == COLD) 2560 return WRITE_LIFE_EXTREME; 2561 } else { 2562 return WRITE_LIFE_NOT_SET; 2563 } 2564 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) { 2565 if (type == DATA) { 2566 if (temp == WARM) 2567 return WRITE_LIFE_LONG; 2568 else if (temp == HOT) 2569 return WRITE_LIFE_SHORT; 2570 else if (temp == COLD) 2571 return WRITE_LIFE_EXTREME; 2572 } else if (type == NODE) { 2573 if (temp == WARM || temp == HOT) 2574 return WRITE_LIFE_NOT_SET; 2575 else if (temp == COLD) 2576 return WRITE_LIFE_NONE; 2577 } else if (type == META) { 2578 return WRITE_LIFE_MEDIUM; 2579 } 2580 } 2581 return WRITE_LIFE_NOT_SET; 2582 } 2583 2584 static int __get_segment_type_2(struct f2fs_io_info *fio) 2585 { 2586 if (fio->type == DATA) 2587 return CURSEG_HOT_DATA; 2588 else 2589 return CURSEG_HOT_NODE; 2590 } 2591 2592 static int __get_segment_type_4(struct f2fs_io_info *fio) 2593 { 2594 if (fio->type == DATA) { 2595 struct inode *inode = fio->page->mapping->host; 2596 2597 if (S_ISDIR(inode->i_mode)) 2598 return CURSEG_HOT_DATA; 2599 else 2600 return CURSEG_COLD_DATA; 2601 } else { 2602 if (IS_DNODE(fio->page) && is_cold_node(fio->page)) 2603 return CURSEG_WARM_NODE; 2604 else 2605 return CURSEG_COLD_NODE; 2606 } 2607 } 2608 2609 static int __get_segment_type_6(struct f2fs_io_info *fio) 2610 { 2611 if (fio->type == DATA) { 2612 struct inode *inode = fio->page->mapping->host; 2613 2614 if (is_cold_data(fio->page) || file_is_cold(inode)) 2615 return CURSEG_COLD_DATA; 2616 if (file_is_hot(inode) || 2617 is_inode_flag_set(inode, FI_HOT_DATA) || 2618 is_inode_flag_set(inode, FI_ATOMIC_FILE) || 2619 is_inode_flag_set(inode, FI_VOLATILE_FILE)) 2620 return CURSEG_HOT_DATA; 2621 return rw_hint_to_seg_type(inode->i_write_hint); 2622 } else { 2623 if (IS_DNODE(fio->page)) 2624 return is_cold_node(fio->page) ? CURSEG_WARM_NODE : 2625 CURSEG_HOT_NODE; 2626 return CURSEG_COLD_NODE; 2627 } 2628 } 2629 2630 static int __get_segment_type(struct f2fs_io_info *fio) 2631 { 2632 int type = 0; 2633 2634 switch (F2FS_OPTION(fio->sbi).active_logs) { 2635 case 2: 2636 type = __get_segment_type_2(fio); 2637 break; 2638 case 4: 2639 type = __get_segment_type_4(fio); 2640 break; 2641 case 6: 2642 type = __get_segment_type_6(fio); 2643 break; 2644 default: 2645 f2fs_bug_on(fio->sbi, true); 2646 } 2647 2648 if (IS_HOT(type)) 2649 fio->temp = HOT; 2650 else if (IS_WARM(type)) 2651 fio->temp = WARM; 2652 else 2653 fio->temp = COLD; 2654 return type; 2655 } 2656 2657 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, 2658 block_t old_blkaddr, block_t *new_blkaddr, 2659 struct f2fs_summary *sum, int type, 2660 struct f2fs_io_info *fio, bool add_list) 2661 { 2662 struct sit_info *sit_i = SIT_I(sbi); 2663 struct curseg_info *curseg = CURSEG_I(sbi, type); 2664 2665 down_read(&SM_I(sbi)->curseg_lock); 2666 2667 mutex_lock(&curseg->curseg_mutex); 2668 down_write(&sit_i->sentry_lock); 2669 2670 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 2671 2672 f2fs_wait_discard_bio(sbi, *new_blkaddr); 2673 2674 /* 2675 * __add_sum_entry should be resided under the curseg_mutex 2676 * because, this function updates a summary entry in the 2677 * current summary block. 2678 */ 2679 __add_sum_entry(sbi, type, sum); 2680 2681 __refresh_next_blkoff(sbi, curseg); 2682 2683 stat_inc_block_count(sbi, curseg); 2684 2685 /* 2686 * SIT information should be updated before segment allocation, 2687 * since SSR needs latest valid block information. 2688 */ 2689 update_sit_entry(sbi, *new_blkaddr, 1); 2690 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 2691 update_sit_entry(sbi, old_blkaddr, -1); 2692 2693 if (!__has_curseg_space(sbi, type)) 2694 sit_i->s_ops->allocate_segment(sbi, type, false); 2695 2696 /* 2697 * segment dirty status should be updated after segment allocation, 2698 * so we just need to update status only one time after previous 2699 * segment being closed. 2700 */ 2701 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 2702 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr)); 2703 2704 up_write(&sit_i->sentry_lock); 2705 2706 if (page && IS_NODESEG(type)) { 2707 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); 2708 2709 f2fs_inode_chksum_set(sbi, page); 2710 } 2711 2712 if (add_list) { 2713 struct f2fs_bio_info *io; 2714 2715 INIT_LIST_HEAD(&fio->list); 2716 fio->in_list = true; 2717 io = sbi->write_io[fio->type] + fio->temp; 2718 spin_lock(&io->io_lock); 2719 list_add_tail(&fio->list, &io->io_list); 2720 spin_unlock(&io->io_lock); 2721 } 2722 2723 mutex_unlock(&curseg->curseg_mutex); 2724 2725 up_read(&SM_I(sbi)->curseg_lock); 2726 } 2727 2728 static void update_device_state(struct f2fs_io_info *fio) 2729 { 2730 struct f2fs_sb_info *sbi = fio->sbi; 2731 unsigned int devidx; 2732 2733 if (!sbi->s_ndevs) 2734 return; 2735 2736 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr); 2737 2738 /* update device state for fsync */ 2739 set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO); 2740 2741 /* update device state for checkpoint */ 2742 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) { 2743 spin_lock(&sbi->dev_lock); 2744 f2fs_set_bit(devidx, (char *)&sbi->dirty_device); 2745 spin_unlock(&sbi->dev_lock); 2746 } 2747 } 2748 2749 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) 2750 { 2751 int type = __get_segment_type(fio); 2752 int err; 2753 bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA); 2754 2755 if (keep_order) 2756 down_read(&fio->sbi->io_order_lock); 2757 reallocate: 2758 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr, 2759 &fio->new_blkaddr, sum, type, fio, true); 2760 2761 /* writeout dirty page into bdev */ 2762 err = f2fs_submit_page_write(fio); 2763 if (err == -EAGAIN) { 2764 fio->old_blkaddr = fio->new_blkaddr; 2765 goto reallocate; 2766 } else if (!err) { 2767 update_device_state(fio); 2768 } 2769 if (keep_order) 2770 up_read(&fio->sbi->io_order_lock); 2771 } 2772 2773 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page, 2774 enum iostat_type io_type) 2775 { 2776 struct f2fs_io_info fio = { 2777 .sbi = sbi, 2778 .type = META, 2779 .temp = HOT, 2780 .op = REQ_OP_WRITE, 2781 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO, 2782 .old_blkaddr = page->index, 2783 .new_blkaddr = page->index, 2784 .page = page, 2785 .encrypted_page = NULL, 2786 .in_list = false, 2787 }; 2788 2789 if (unlikely(page->index >= MAIN_BLKADDR(sbi))) 2790 fio.op_flags &= ~REQ_META; 2791 2792 set_page_writeback(page); 2793 ClearPageError(page); 2794 f2fs_submit_page_write(&fio); 2795 2796 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE); 2797 } 2798 2799 void write_node_page(unsigned int nid, struct f2fs_io_info *fio) 2800 { 2801 struct f2fs_summary sum; 2802 2803 set_summary(&sum, nid, 0, 0); 2804 do_write_page(&sum, fio); 2805 2806 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE); 2807 } 2808 2809 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio) 2810 { 2811 struct f2fs_sb_info *sbi = fio->sbi; 2812 struct f2fs_summary sum; 2813 struct node_info ni; 2814 2815 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); 2816 get_node_info(sbi, dn->nid, &ni); 2817 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); 2818 do_write_page(&sum, fio); 2819 f2fs_update_data_blkaddr(dn, fio->new_blkaddr); 2820 2821 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE); 2822 } 2823 2824 int rewrite_data_page(struct f2fs_io_info *fio) 2825 { 2826 int err; 2827 struct f2fs_sb_info *sbi = fio->sbi; 2828 2829 fio->new_blkaddr = fio->old_blkaddr; 2830 /* i/o temperature is needed for passing down write hints */ 2831 __get_segment_type(fio); 2832 2833 f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi, 2834 GET_SEGNO(sbi, fio->new_blkaddr))->type)); 2835 2836 stat_inc_inplace_blocks(fio->sbi); 2837 2838 err = f2fs_submit_page_bio(fio); 2839 if (!err) 2840 update_device_state(fio); 2841 2842 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE); 2843 2844 return err; 2845 } 2846 2847 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi, 2848 unsigned int segno) 2849 { 2850 int i; 2851 2852 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 2853 if (CURSEG_I(sbi, i)->segno == segno) 2854 break; 2855 } 2856 return i; 2857 } 2858 2859 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 2860 block_t old_blkaddr, block_t new_blkaddr, 2861 bool recover_curseg, bool recover_newaddr) 2862 { 2863 struct sit_info *sit_i = SIT_I(sbi); 2864 struct curseg_info *curseg; 2865 unsigned int segno, old_cursegno; 2866 struct seg_entry *se; 2867 int type; 2868 unsigned short old_blkoff; 2869 2870 segno = GET_SEGNO(sbi, new_blkaddr); 2871 se = get_seg_entry(sbi, segno); 2872 type = se->type; 2873 2874 down_write(&SM_I(sbi)->curseg_lock); 2875 2876 if (!recover_curseg) { 2877 /* for recovery flow */ 2878 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { 2879 if (old_blkaddr == NULL_ADDR) 2880 type = CURSEG_COLD_DATA; 2881 else 2882 type = CURSEG_WARM_DATA; 2883 } 2884 } else { 2885 if (IS_CURSEG(sbi, segno)) { 2886 /* se->type is volatile as SSR allocation */ 2887 type = __f2fs_get_curseg(sbi, segno); 2888 f2fs_bug_on(sbi, type == NO_CHECK_TYPE); 2889 } else { 2890 type = CURSEG_WARM_DATA; 2891 } 2892 } 2893 2894 f2fs_bug_on(sbi, !IS_DATASEG(type)); 2895 curseg = CURSEG_I(sbi, type); 2896 2897 mutex_lock(&curseg->curseg_mutex); 2898 down_write(&sit_i->sentry_lock); 2899 2900 old_cursegno = curseg->segno; 2901 old_blkoff = curseg->next_blkoff; 2902 2903 /* change the current segment */ 2904 if (segno != curseg->segno) { 2905 curseg->next_segno = segno; 2906 change_curseg(sbi, type); 2907 } 2908 2909 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); 2910 __add_sum_entry(sbi, type, sum); 2911 2912 if (!recover_curseg || recover_newaddr) 2913 update_sit_entry(sbi, new_blkaddr, 1); 2914 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 2915 update_sit_entry(sbi, old_blkaddr, -1); 2916 2917 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 2918 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); 2919 2920 locate_dirty_segment(sbi, old_cursegno); 2921 2922 if (recover_curseg) { 2923 if (old_cursegno != curseg->segno) { 2924 curseg->next_segno = old_cursegno; 2925 change_curseg(sbi, type); 2926 } 2927 curseg->next_blkoff = old_blkoff; 2928 } 2929 2930 up_write(&sit_i->sentry_lock); 2931 mutex_unlock(&curseg->curseg_mutex); 2932 up_write(&SM_I(sbi)->curseg_lock); 2933 } 2934 2935 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, 2936 block_t old_addr, block_t new_addr, 2937 unsigned char version, bool recover_curseg, 2938 bool recover_newaddr) 2939 { 2940 struct f2fs_summary sum; 2941 2942 set_summary(&sum, dn->nid, dn->ofs_in_node, version); 2943 2944 __f2fs_replace_block(sbi, &sum, old_addr, new_addr, 2945 recover_curseg, recover_newaddr); 2946 2947 f2fs_update_data_blkaddr(dn, new_addr); 2948 } 2949 2950 void f2fs_wait_on_page_writeback(struct page *page, 2951 enum page_type type, bool ordered) 2952 { 2953 if (PageWriteback(page)) { 2954 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 2955 2956 f2fs_submit_merged_write_cond(sbi, page->mapping->host, 2957 0, page->index, type); 2958 if (ordered) 2959 wait_on_page_writeback(page); 2960 else 2961 wait_for_stable_page(page); 2962 } 2963 } 2964 2965 void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr) 2966 { 2967 struct page *cpage; 2968 2969 if (!is_valid_blkaddr(blkaddr)) 2970 return; 2971 2972 cpage = find_lock_page(META_MAPPING(sbi), blkaddr); 2973 if (cpage) { 2974 f2fs_wait_on_page_writeback(cpage, DATA, true); 2975 f2fs_put_page(cpage, 1); 2976 } 2977 } 2978 2979 static void read_compacted_summaries(struct f2fs_sb_info *sbi) 2980 { 2981 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 2982 struct curseg_info *seg_i; 2983 unsigned char *kaddr; 2984 struct page *page; 2985 block_t start; 2986 int i, j, offset; 2987 2988 start = start_sum_block(sbi); 2989 2990 page = get_meta_page(sbi, start++); 2991 kaddr = (unsigned char *)page_address(page); 2992 2993 /* Step 1: restore nat cache */ 2994 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 2995 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE); 2996 2997 /* Step 2: restore sit cache */ 2998 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 2999 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE); 3000 offset = 2 * SUM_JOURNAL_SIZE; 3001 3002 /* Step 3: restore summary entries */ 3003 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 3004 unsigned short blk_off; 3005 unsigned int segno; 3006 3007 seg_i = CURSEG_I(sbi, i); 3008 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 3009 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 3010 seg_i->next_segno = segno; 3011 reset_curseg(sbi, i, 0); 3012 seg_i->alloc_type = ckpt->alloc_type[i]; 3013 seg_i->next_blkoff = blk_off; 3014 3015 if (seg_i->alloc_type == SSR) 3016 blk_off = sbi->blocks_per_seg; 3017 3018 for (j = 0; j < blk_off; j++) { 3019 struct f2fs_summary *s; 3020 s = (struct f2fs_summary *)(kaddr + offset); 3021 seg_i->sum_blk->entries[j] = *s; 3022 offset += SUMMARY_SIZE; 3023 if (offset + SUMMARY_SIZE <= PAGE_SIZE - 3024 SUM_FOOTER_SIZE) 3025 continue; 3026 3027 f2fs_put_page(page, 1); 3028 page = NULL; 3029 3030 page = get_meta_page(sbi, start++); 3031 kaddr = (unsigned char *)page_address(page); 3032 offset = 0; 3033 } 3034 } 3035 f2fs_put_page(page, 1); 3036 } 3037 3038 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 3039 { 3040 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3041 struct f2fs_summary_block *sum; 3042 struct curseg_info *curseg; 3043 struct page *new; 3044 unsigned short blk_off; 3045 unsigned int segno = 0; 3046 block_t blk_addr = 0; 3047 3048 /* get segment number and block addr */ 3049 if (IS_DATASEG(type)) { 3050 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 3051 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 3052 CURSEG_HOT_DATA]); 3053 if (__exist_node_summaries(sbi)) 3054 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type); 3055 else 3056 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 3057 } else { 3058 segno = le32_to_cpu(ckpt->cur_node_segno[type - 3059 CURSEG_HOT_NODE]); 3060 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 3061 CURSEG_HOT_NODE]); 3062 if (__exist_node_summaries(sbi)) 3063 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 3064 type - CURSEG_HOT_NODE); 3065 else 3066 blk_addr = GET_SUM_BLOCK(sbi, segno); 3067 } 3068 3069 new = get_meta_page(sbi, blk_addr); 3070 sum = (struct f2fs_summary_block *)page_address(new); 3071 3072 if (IS_NODESEG(type)) { 3073 if (__exist_node_summaries(sbi)) { 3074 struct f2fs_summary *ns = &sum->entries[0]; 3075 int i; 3076 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { 3077 ns->version = 0; 3078 ns->ofs_in_node = 0; 3079 } 3080 } else { 3081 restore_node_summary(sbi, segno, sum); 3082 } 3083 } 3084 3085 /* set uncompleted segment to curseg */ 3086 curseg = CURSEG_I(sbi, type); 3087 mutex_lock(&curseg->curseg_mutex); 3088 3089 /* update journal info */ 3090 down_write(&curseg->journal_rwsem); 3091 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE); 3092 up_write(&curseg->journal_rwsem); 3093 3094 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE); 3095 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE); 3096 curseg->next_segno = segno; 3097 reset_curseg(sbi, type, 0); 3098 curseg->alloc_type = ckpt->alloc_type[type]; 3099 curseg->next_blkoff = blk_off; 3100 mutex_unlock(&curseg->curseg_mutex); 3101 f2fs_put_page(new, 1); 3102 return 0; 3103 } 3104 3105 static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 3106 { 3107 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal; 3108 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal; 3109 int type = CURSEG_HOT_DATA; 3110 int err; 3111 3112 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) { 3113 int npages = npages_for_summary_flush(sbi, true); 3114 3115 if (npages >= 2) 3116 ra_meta_pages(sbi, start_sum_block(sbi), npages, 3117 META_CP, true); 3118 3119 /* restore for compacted data summary */ 3120 read_compacted_summaries(sbi); 3121 type = CURSEG_HOT_NODE; 3122 } 3123 3124 if (__exist_node_summaries(sbi)) 3125 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type), 3126 NR_CURSEG_TYPE - type, META_CP, true); 3127 3128 for (; type <= CURSEG_COLD_NODE; type++) { 3129 err = read_normal_summaries(sbi, type); 3130 if (err) 3131 return err; 3132 } 3133 3134 /* sanity check for summary blocks */ 3135 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES || 3136 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) 3137 return -EINVAL; 3138 3139 return 0; 3140 } 3141 3142 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 3143 { 3144 struct page *page; 3145 unsigned char *kaddr; 3146 struct f2fs_summary *summary; 3147 struct curseg_info *seg_i; 3148 int written_size = 0; 3149 int i, j; 3150 3151 page = grab_meta_page(sbi, blkaddr++); 3152 kaddr = (unsigned char *)page_address(page); 3153 memset(kaddr, 0, PAGE_SIZE); 3154 3155 /* Step 1: write nat cache */ 3156 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 3157 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE); 3158 written_size += SUM_JOURNAL_SIZE; 3159 3160 /* Step 2: write sit cache */ 3161 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 3162 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE); 3163 written_size += SUM_JOURNAL_SIZE; 3164 3165 /* Step 3: write summary entries */ 3166 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 3167 unsigned short blkoff; 3168 seg_i = CURSEG_I(sbi, i); 3169 if (sbi->ckpt->alloc_type[i] == SSR) 3170 blkoff = sbi->blocks_per_seg; 3171 else 3172 blkoff = curseg_blkoff(sbi, i); 3173 3174 for (j = 0; j < blkoff; j++) { 3175 if (!page) { 3176 page = grab_meta_page(sbi, blkaddr++); 3177 kaddr = (unsigned char *)page_address(page); 3178 memset(kaddr, 0, PAGE_SIZE); 3179 written_size = 0; 3180 } 3181 summary = (struct f2fs_summary *)(kaddr + written_size); 3182 *summary = seg_i->sum_blk->entries[j]; 3183 written_size += SUMMARY_SIZE; 3184 3185 if (written_size + SUMMARY_SIZE <= PAGE_SIZE - 3186 SUM_FOOTER_SIZE) 3187 continue; 3188 3189 set_page_dirty(page); 3190 f2fs_put_page(page, 1); 3191 page = NULL; 3192 } 3193 } 3194 if (page) { 3195 set_page_dirty(page); 3196 f2fs_put_page(page, 1); 3197 } 3198 } 3199 3200 static void write_normal_summaries(struct f2fs_sb_info *sbi, 3201 block_t blkaddr, int type) 3202 { 3203 int i, end; 3204 if (IS_DATASEG(type)) 3205 end = type + NR_CURSEG_DATA_TYPE; 3206 else 3207 end = type + NR_CURSEG_NODE_TYPE; 3208 3209 for (i = type; i < end; i++) 3210 write_current_sum_page(sbi, i, blkaddr + (i - type)); 3211 } 3212 3213 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 3214 { 3215 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) 3216 write_compacted_summaries(sbi, start_blk); 3217 else 3218 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 3219 } 3220 3221 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 3222 { 3223 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 3224 } 3225 3226 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type, 3227 unsigned int val, int alloc) 3228 { 3229 int i; 3230 3231 if (type == NAT_JOURNAL) { 3232 for (i = 0; i < nats_in_cursum(journal); i++) { 3233 if (le32_to_cpu(nid_in_journal(journal, i)) == val) 3234 return i; 3235 } 3236 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL)) 3237 return update_nats_in_cursum(journal, 1); 3238 } else if (type == SIT_JOURNAL) { 3239 for (i = 0; i < sits_in_cursum(journal); i++) 3240 if (le32_to_cpu(segno_in_journal(journal, i)) == val) 3241 return i; 3242 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL)) 3243 return update_sits_in_cursum(journal, 1); 3244 } 3245 return -1; 3246 } 3247 3248 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, 3249 unsigned int segno) 3250 { 3251 return get_meta_page(sbi, current_sit_addr(sbi, segno)); 3252 } 3253 3254 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, 3255 unsigned int start) 3256 { 3257 struct sit_info *sit_i = SIT_I(sbi); 3258 struct page *page; 3259 pgoff_t src_off, dst_off; 3260 3261 src_off = current_sit_addr(sbi, start); 3262 dst_off = next_sit_addr(sbi, src_off); 3263 3264 page = grab_meta_page(sbi, dst_off); 3265 seg_info_to_sit_page(sbi, page, start); 3266 3267 set_page_dirty(page); 3268 set_to_next_sit(sit_i, start); 3269 3270 return page; 3271 } 3272 3273 static struct sit_entry_set *grab_sit_entry_set(void) 3274 { 3275 struct sit_entry_set *ses = 3276 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS); 3277 3278 ses->entry_cnt = 0; 3279 INIT_LIST_HEAD(&ses->set_list); 3280 return ses; 3281 } 3282 3283 static void release_sit_entry_set(struct sit_entry_set *ses) 3284 { 3285 list_del(&ses->set_list); 3286 kmem_cache_free(sit_entry_set_slab, ses); 3287 } 3288 3289 static void adjust_sit_entry_set(struct sit_entry_set *ses, 3290 struct list_head *head) 3291 { 3292 struct sit_entry_set *next = ses; 3293 3294 if (list_is_last(&ses->set_list, head)) 3295 return; 3296 3297 list_for_each_entry_continue(next, head, set_list) 3298 if (ses->entry_cnt <= next->entry_cnt) 3299 break; 3300 3301 list_move_tail(&ses->set_list, &next->set_list); 3302 } 3303 3304 static void add_sit_entry(unsigned int segno, struct list_head *head) 3305 { 3306 struct sit_entry_set *ses; 3307 unsigned int start_segno = START_SEGNO(segno); 3308 3309 list_for_each_entry(ses, head, set_list) { 3310 if (ses->start_segno == start_segno) { 3311 ses->entry_cnt++; 3312 adjust_sit_entry_set(ses, head); 3313 return; 3314 } 3315 } 3316 3317 ses = grab_sit_entry_set(); 3318 3319 ses->start_segno = start_segno; 3320 ses->entry_cnt++; 3321 list_add(&ses->set_list, head); 3322 } 3323 3324 static void add_sits_in_set(struct f2fs_sb_info *sbi) 3325 { 3326 struct f2fs_sm_info *sm_info = SM_I(sbi); 3327 struct list_head *set_list = &sm_info->sit_entry_set; 3328 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; 3329 unsigned int segno; 3330 3331 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) 3332 add_sit_entry(segno, set_list); 3333 } 3334 3335 static void remove_sits_in_journal(struct f2fs_sb_info *sbi) 3336 { 3337 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 3338 struct f2fs_journal *journal = curseg->journal; 3339 int i; 3340 3341 down_write(&curseg->journal_rwsem); 3342 for (i = 0; i < sits_in_cursum(journal); i++) { 3343 unsigned int segno; 3344 bool dirtied; 3345 3346 segno = le32_to_cpu(segno_in_journal(journal, i)); 3347 dirtied = __mark_sit_entry_dirty(sbi, segno); 3348 3349 if (!dirtied) 3350 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); 3351 } 3352 update_sits_in_cursum(journal, -i); 3353 up_write(&curseg->journal_rwsem); 3354 } 3355 3356 /* 3357 * CP calls this function, which flushes SIT entries including sit_journal, 3358 * and moves prefree segs to free segs. 3359 */ 3360 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 3361 { 3362 struct sit_info *sit_i = SIT_I(sbi); 3363 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 3364 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 3365 struct f2fs_journal *journal = curseg->journal; 3366 struct sit_entry_set *ses, *tmp; 3367 struct list_head *head = &SM_I(sbi)->sit_entry_set; 3368 bool to_journal = true; 3369 struct seg_entry *se; 3370 3371 down_write(&sit_i->sentry_lock); 3372 3373 if (!sit_i->dirty_sentries) 3374 goto out; 3375 3376 /* 3377 * add and account sit entries of dirty bitmap in sit entry 3378 * set temporarily 3379 */ 3380 add_sits_in_set(sbi); 3381 3382 /* 3383 * if there are no enough space in journal to store dirty sit 3384 * entries, remove all entries from journal and add and account 3385 * them in sit entry set. 3386 */ 3387 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL)) 3388 remove_sits_in_journal(sbi); 3389 3390 /* 3391 * there are two steps to flush sit entries: 3392 * #1, flush sit entries to journal in current cold data summary block. 3393 * #2, flush sit entries to sit page. 3394 */ 3395 list_for_each_entry_safe(ses, tmp, head, set_list) { 3396 struct page *page = NULL; 3397 struct f2fs_sit_block *raw_sit = NULL; 3398 unsigned int start_segno = ses->start_segno; 3399 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, 3400 (unsigned long)MAIN_SEGS(sbi)); 3401 unsigned int segno = start_segno; 3402 3403 if (to_journal && 3404 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL)) 3405 to_journal = false; 3406 3407 if (to_journal) { 3408 down_write(&curseg->journal_rwsem); 3409 } else { 3410 page = get_next_sit_page(sbi, start_segno); 3411 raw_sit = page_address(page); 3412 } 3413 3414 /* flush dirty sit entries in region of current sit set */ 3415 for_each_set_bit_from(segno, bitmap, end) { 3416 int offset, sit_offset; 3417 3418 se = get_seg_entry(sbi, segno); 3419 #ifdef CONFIG_F2FS_CHECK_FS 3420 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir, 3421 SIT_VBLOCK_MAP_SIZE)) 3422 f2fs_bug_on(sbi, 1); 3423 #endif 3424 3425 /* add discard candidates */ 3426 if (!(cpc->reason & CP_DISCARD)) { 3427 cpc->trim_start = segno; 3428 add_discard_addrs(sbi, cpc, false); 3429 } 3430 3431 if (to_journal) { 3432 offset = lookup_journal_in_cursum(journal, 3433 SIT_JOURNAL, segno, 1); 3434 f2fs_bug_on(sbi, offset < 0); 3435 segno_in_journal(journal, offset) = 3436 cpu_to_le32(segno); 3437 seg_info_to_raw_sit(se, 3438 &sit_in_journal(journal, offset)); 3439 check_block_count(sbi, segno, 3440 &sit_in_journal(journal, offset)); 3441 } else { 3442 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 3443 seg_info_to_raw_sit(se, 3444 &raw_sit->entries[sit_offset]); 3445 check_block_count(sbi, segno, 3446 &raw_sit->entries[sit_offset]); 3447 } 3448 3449 __clear_bit(segno, bitmap); 3450 sit_i->dirty_sentries--; 3451 ses->entry_cnt--; 3452 } 3453 3454 if (to_journal) 3455 up_write(&curseg->journal_rwsem); 3456 else 3457 f2fs_put_page(page, 1); 3458 3459 f2fs_bug_on(sbi, ses->entry_cnt); 3460 release_sit_entry_set(ses); 3461 } 3462 3463 f2fs_bug_on(sbi, !list_empty(head)); 3464 f2fs_bug_on(sbi, sit_i->dirty_sentries); 3465 out: 3466 if (cpc->reason & CP_DISCARD) { 3467 __u64 trim_start = cpc->trim_start; 3468 3469 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) 3470 add_discard_addrs(sbi, cpc, false); 3471 3472 cpc->trim_start = trim_start; 3473 } 3474 up_write(&sit_i->sentry_lock); 3475 3476 set_prefree_as_free_segments(sbi); 3477 } 3478 3479 static int build_sit_info(struct f2fs_sb_info *sbi) 3480 { 3481 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 3482 struct sit_info *sit_i; 3483 unsigned int sit_segs, start; 3484 char *src_bitmap; 3485 unsigned int bitmap_size; 3486 3487 /* allocate memory for SIT information */ 3488 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL); 3489 if (!sit_i) 3490 return -ENOMEM; 3491 3492 SM_I(sbi)->sit_info = sit_i; 3493 3494 sit_i->sentries = f2fs_kvzalloc(sbi, MAIN_SEGS(sbi) * 3495 sizeof(struct seg_entry), GFP_KERNEL); 3496 if (!sit_i->sentries) 3497 return -ENOMEM; 3498 3499 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 3500 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size, 3501 GFP_KERNEL); 3502 if (!sit_i->dirty_sentries_bitmap) 3503 return -ENOMEM; 3504 3505 for (start = 0; start < MAIN_SEGS(sbi); start++) { 3506 sit_i->sentries[start].cur_valid_map 3507 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 3508 sit_i->sentries[start].ckpt_valid_map 3509 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 3510 if (!sit_i->sentries[start].cur_valid_map || 3511 !sit_i->sentries[start].ckpt_valid_map) 3512 return -ENOMEM; 3513 3514 #ifdef CONFIG_F2FS_CHECK_FS 3515 sit_i->sentries[start].cur_valid_map_mir 3516 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 3517 if (!sit_i->sentries[start].cur_valid_map_mir) 3518 return -ENOMEM; 3519 #endif 3520 3521 if (f2fs_discard_en(sbi)) { 3522 sit_i->sentries[start].discard_map 3523 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, 3524 GFP_KERNEL); 3525 if (!sit_i->sentries[start].discard_map) 3526 return -ENOMEM; 3527 } 3528 } 3529 3530 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 3531 if (!sit_i->tmp_map) 3532 return -ENOMEM; 3533 3534 if (sbi->segs_per_sec > 1) { 3535 sit_i->sec_entries = f2fs_kvzalloc(sbi, MAIN_SECS(sbi) * 3536 sizeof(struct sec_entry), GFP_KERNEL); 3537 if (!sit_i->sec_entries) 3538 return -ENOMEM; 3539 } 3540 3541 /* get information related with SIT */ 3542 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 3543 3544 /* setup SIT bitmap from ckeckpoint pack */ 3545 bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 3546 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 3547 3548 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); 3549 if (!sit_i->sit_bitmap) 3550 return -ENOMEM; 3551 3552 #ifdef CONFIG_F2FS_CHECK_FS 3553 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); 3554 if (!sit_i->sit_bitmap_mir) 3555 return -ENOMEM; 3556 #endif 3557 3558 /* init SIT information */ 3559 sit_i->s_ops = &default_salloc_ops; 3560 3561 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 3562 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; 3563 sit_i->written_valid_blocks = 0; 3564 sit_i->bitmap_size = bitmap_size; 3565 sit_i->dirty_sentries = 0; 3566 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 3567 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 3568 sit_i->mounted_time = ktime_get_real_seconds(); 3569 init_rwsem(&sit_i->sentry_lock); 3570 return 0; 3571 } 3572 3573 static int build_free_segmap(struct f2fs_sb_info *sbi) 3574 { 3575 struct free_segmap_info *free_i; 3576 unsigned int bitmap_size, sec_bitmap_size; 3577 3578 /* allocate memory for free segmap information */ 3579 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL); 3580 if (!free_i) 3581 return -ENOMEM; 3582 3583 SM_I(sbi)->free_info = free_i; 3584 3585 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 3586 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL); 3587 if (!free_i->free_segmap) 3588 return -ENOMEM; 3589 3590 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 3591 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL); 3592 if (!free_i->free_secmap) 3593 return -ENOMEM; 3594 3595 /* set all segments as dirty temporarily */ 3596 memset(free_i->free_segmap, 0xff, bitmap_size); 3597 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 3598 3599 /* init free segmap information */ 3600 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); 3601 free_i->free_segments = 0; 3602 free_i->free_sections = 0; 3603 spin_lock_init(&free_i->segmap_lock); 3604 return 0; 3605 } 3606 3607 static int build_curseg(struct f2fs_sb_info *sbi) 3608 { 3609 struct curseg_info *array; 3610 int i; 3611 3612 array = f2fs_kzalloc(sbi, sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL); 3613 if (!array) 3614 return -ENOMEM; 3615 3616 SM_I(sbi)->curseg_array = array; 3617 3618 for (i = 0; i < NR_CURSEG_TYPE; i++) { 3619 mutex_init(&array[i].curseg_mutex); 3620 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL); 3621 if (!array[i].sum_blk) 3622 return -ENOMEM; 3623 init_rwsem(&array[i].journal_rwsem); 3624 array[i].journal = f2fs_kzalloc(sbi, 3625 sizeof(struct f2fs_journal), GFP_KERNEL); 3626 if (!array[i].journal) 3627 return -ENOMEM; 3628 array[i].segno = NULL_SEGNO; 3629 array[i].next_blkoff = 0; 3630 } 3631 return restore_curseg_summaries(sbi); 3632 } 3633 3634 static int build_sit_entries(struct f2fs_sb_info *sbi) 3635 { 3636 struct sit_info *sit_i = SIT_I(sbi); 3637 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 3638 struct f2fs_journal *journal = curseg->journal; 3639 struct seg_entry *se; 3640 struct f2fs_sit_entry sit; 3641 int sit_blk_cnt = SIT_BLK_CNT(sbi); 3642 unsigned int i, start, end; 3643 unsigned int readed, start_blk = 0; 3644 int err = 0; 3645 block_t total_node_blocks = 0; 3646 3647 do { 3648 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES, 3649 META_SIT, true); 3650 3651 start = start_blk * sit_i->sents_per_block; 3652 end = (start_blk + readed) * sit_i->sents_per_block; 3653 3654 for (; start < end && start < MAIN_SEGS(sbi); start++) { 3655 struct f2fs_sit_block *sit_blk; 3656 struct page *page; 3657 3658 se = &sit_i->sentries[start]; 3659 page = get_current_sit_page(sbi, start); 3660 sit_blk = (struct f2fs_sit_block *)page_address(page); 3661 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 3662 f2fs_put_page(page, 1); 3663 3664 err = check_block_count(sbi, start, &sit); 3665 if (err) 3666 return err; 3667 seg_info_from_raw_sit(se, &sit); 3668 if (IS_NODESEG(se->type)) 3669 total_node_blocks += se->valid_blocks; 3670 3671 /* build discard map only one time */ 3672 if (f2fs_discard_en(sbi)) { 3673 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 3674 memset(se->discard_map, 0xff, 3675 SIT_VBLOCK_MAP_SIZE); 3676 } else { 3677 memcpy(se->discard_map, 3678 se->cur_valid_map, 3679 SIT_VBLOCK_MAP_SIZE); 3680 sbi->discard_blks += 3681 sbi->blocks_per_seg - 3682 se->valid_blocks; 3683 } 3684 } 3685 3686 if (sbi->segs_per_sec > 1) 3687 get_sec_entry(sbi, start)->valid_blocks += 3688 se->valid_blocks; 3689 } 3690 start_blk += readed; 3691 } while (start_blk < sit_blk_cnt); 3692 3693 down_read(&curseg->journal_rwsem); 3694 for (i = 0; i < sits_in_cursum(journal); i++) { 3695 unsigned int old_valid_blocks; 3696 3697 start = le32_to_cpu(segno_in_journal(journal, i)); 3698 if (start >= MAIN_SEGS(sbi)) { 3699 f2fs_msg(sbi->sb, KERN_ERR, 3700 "Wrong journal entry on segno %u", 3701 start); 3702 set_sbi_flag(sbi, SBI_NEED_FSCK); 3703 err = -EINVAL; 3704 break; 3705 } 3706 3707 se = &sit_i->sentries[start]; 3708 sit = sit_in_journal(journal, i); 3709 3710 old_valid_blocks = se->valid_blocks; 3711 if (IS_NODESEG(se->type)) 3712 total_node_blocks -= old_valid_blocks; 3713 3714 err = check_block_count(sbi, start, &sit); 3715 if (err) 3716 break; 3717 seg_info_from_raw_sit(se, &sit); 3718 if (IS_NODESEG(se->type)) 3719 total_node_blocks += se->valid_blocks; 3720 3721 if (f2fs_discard_en(sbi)) { 3722 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 3723 memset(se->discard_map, 0xff, 3724 SIT_VBLOCK_MAP_SIZE); 3725 } else { 3726 memcpy(se->discard_map, se->cur_valid_map, 3727 SIT_VBLOCK_MAP_SIZE); 3728 sbi->discard_blks += old_valid_blocks; 3729 sbi->discard_blks -= se->valid_blocks; 3730 } 3731 } 3732 3733 if (sbi->segs_per_sec > 1) { 3734 get_sec_entry(sbi, start)->valid_blocks += 3735 se->valid_blocks; 3736 get_sec_entry(sbi, start)->valid_blocks -= 3737 old_valid_blocks; 3738 } 3739 } 3740 up_read(&curseg->journal_rwsem); 3741 3742 if (!err && total_node_blocks != valid_node_count(sbi)) { 3743 f2fs_msg(sbi->sb, KERN_ERR, 3744 "SIT is corrupted node# %u vs %u", 3745 total_node_blocks, valid_node_count(sbi)); 3746 set_sbi_flag(sbi, SBI_NEED_FSCK); 3747 err = -EINVAL; 3748 } 3749 3750 return err; 3751 } 3752 3753 static void init_free_segmap(struct f2fs_sb_info *sbi) 3754 { 3755 unsigned int start; 3756 int type; 3757 3758 for (start = 0; start < MAIN_SEGS(sbi); start++) { 3759 struct seg_entry *sentry = get_seg_entry(sbi, start); 3760 if (!sentry->valid_blocks) 3761 __set_free(sbi, start); 3762 else 3763 SIT_I(sbi)->written_valid_blocks += 3764 sentry->valid_blocks; 3765 } 3766 3767 /* set use the current segments */ 3768 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 3769 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 3770 __set_test_and_inuse(sbi, curseg_t->segno); 3771 } 3772 } 3773 3774 static void init_dirty_segmap(struct f2fs_sb_info *sbi) 3775 { 3776 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3777 struct free_segmap_info *free_i = FREE_I(sbi); 3778 unsigned int segno = 0, offset = 0; 3779 unsigned short valid_blocks; 3780 3781 while (1) { 3782 /* find dirty segment based on free segmap */ 3783 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); 3784 if (segno >= MAIN_SEGS(sbi)) 3785 break; 3786 offset = segno + 1; 3787 valid_blocks = get_valid_blocks(sbi, segno, false); 3788 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks) 3789 continue; 3790 if (valid_blocks > sbi->blocks_per_seg) { 3791 f2fs_bug_on(sbi, 1); 3792 continue; 3793 } 3794 mutex_lock(&dirty_i->seglist_lock); 3795 __locate_dirty_segment(sbi, segno, DIRTY); 3796 mutex_unlock(&dirty_i->seglist_lock); 3797 } 3798 } 3799 3800 static int init_victim_secmap(struct f2fs_sb_info *sbi) 3801 { 3802 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3803 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 3804 3805 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 3806 if (!dirty_i->victim_secmap) 3807 return -ENOMEM; 3808 return 0; 3809 } 3810 3811 static int build_dirty_segmap(struct f2fs_sb_info *sbi) 3812 { 3813 struct dirty_seglist_info *dirty_i; 3814 unsigned int bitmap_size, i; 3815 3816 /* allocate memory for dirty segments list information */ 3817 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info), 3818 GFP_KERNEL); 3819 if (!dirty_i) 3820 return -ENOMEM; 3821 3822 SM_I(sbi)->dirty_info = dirty_i; 3823 mutex_init(&dirty_i->seglist_lock); 3824 3825 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 3826 3827 for (i = 0; i < NR_DIRTY_TYPE; i++) { 3828 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size, 3829 GFP_KERNEL); 3830 if (!dirty_i->dirty_segmap[i]) 3831 return -ENOMEM; 3832 } 3833 3834 init_dirty_segmap(sbi); 3835 return init_victim_secmap(sbi); 3836 } 3837 3838 /* 3839 * Update min, max modified time for cost-benefit GC algorithm 3840 */ 3841 static void init_min_max_mtime(struct f2fs_sb_info *sbi) 3842 { 3843 struct sit_info *sit_i = SIT_I(sbi); 3844 unsigned int segno; 3845 3846 down_write(&sit_i->sentry_lock); 3847 3848 sit_i->min_mtime = ULLONG_MAX; 3849 3850 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 3851 unsigned int i; 3852 unsigned long long mtime = 0; 3853 3854 for (i = 0; i < sbi->segs_per_sec; i++) 3855 mtime += get_seg_entry(sbi, segno + i)->mtime; 3856 3857 mtime = div_u64(mtime, sbi->segs_per_sec); 3858 3859 if (sit_i->min_mtime > mtime) 3860 sit_i->min_mtime = mtime; 3861 } 3862 sit_i->max_mtime = get_mtime(sbi); 3863 up_write(&sit_i->sentry_lock); 3864 } 3865 3866 int build_segment_manager(struct f2fs_sb_info *sbi) 3867 { 3868 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 3869 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3870 struct f2fs_sm_info *sm_info; 3871 int err; 3872 3873 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL); 3874 if (!sm_info) 3875 return -ENOMEM; 3876 3877 /* init sm info */ 3878 sbi->sm_info = sm_info; 3879 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 3880 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 3881 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 3882 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 3883 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 3884 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 3885 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 3886 sm_info->rec_prefree_segments = sm_info->main_segments * 3887 DEF_RECLAIM_PREFREE_SEGMENTS / 100; 3888 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS) 3889 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS; 3890 3891 if (!test_opt(sbi, LFS)) 3892 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC; 3893 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; 3894 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; 3895 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS; 3896 sm_info->min_ssr_sections = reserved_sections(sbi); 3897 3898 INIT_LIST_HEAD(&sm_info->sit_entry_set); 3899 3900 init_rwsem(&sm_info->curseg_lock); 3901 3902 if (!f2fs_readonly(sbi->sb)) { 3903 err = create_flush_cmd_control(sbi); 3904 if (err) 3905 return err; 3906 } 3907 3908 err = create_discard_cmd_control(sbi); 3909 if (err) 3910 return err; 3911 3912 err = build_sit_info(sbi); 3913 if (err) 3914 return err; 3915 err = build_free_segmap(sbi); 3916 if (err) 3917 return err; 3918 err = build_curseg(sbi); 3919 if (err) 3920 return err; 3921 3922 /* reinit free segmap based on SIT */ 3923 err = build_sit_entries(sbi); 3924 if (err) 3925 return err; 3926 3927 init_free_segmap(sbi); 3928 err = build_dirty_segmap(sbi); 3929 if (err) 3930 return err; 3931 3932 init_min_max_mtime(sbi); 3933 return 0; 3934 } 3935 3936 static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 3937 enum dirty_type dirty_type) 3938 { 3939 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3940 3941 mutex_lock(&dirty_i->seglist_lock); 3942 kvfree(dirty_i->dirty_segmap[dirty_type]); 3943 dirty_i->nr_dirty[dirty_type] = 0; 3944 mutex_unlock(&dirty_i->seglist_lock); 3945 } 3946 3947 static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 3948 { 3949 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3950 kvfree(dirty_i->victim_secmap); 3951 } 3952 3953 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 3954 { 3955 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3956 int i; 3957 3958 if (!dirty_i) 3959 return; 3960 3961 /* discard pre-free/dirty segments list */ 3962 for (i = 0; i < NR_DIRTY_TYPE; i++) 3963 discard_dirty_segmap(sbi, i); 3964 3965 destroy_victim_secmap(sbi); 3966 SM_I(sbi)->dirty_info = NULL; 3967 kfree(dirty_i); 3968 } 3969 3970 static void destroy_curseg(struct f2fs_sb_info *sbi) 3971 { 3972 struct curseg_info *array = SM_I(sbi)->curseg_array; 3973 int i; 3974 3975 if (!array) 3976 return; 3977 SM_I(sbi)->curseg_array = NULL; 3978 for (i = 0; i < NR_CURSEG_TYPE; i++) { 3979 kfree(array[i].sum_blk); 3980 kfree(array[i].journal); 3981 } 3982 kfree(array); 3983 } 3984 3985 static void destroy_free_segmap(struct f2fs_sb_info *sbi) 3986 { 3987 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 3988 if (!free_i) 3989 return; 3990 SM_I(sbi)->free_info = NULL; 3991 kvfree(free_i->free_segmap); 3992 kvfree(free_i->free_secmap); 3993 kfree(free_i); 3994 } 3995 3996 static void destroy_sit_info(struct f2fs_sb_info *sbi) 3997 { 3998 struct sit_info *sit_i = SIT_I(sbi); 3999 unsigned int start; 4000 4001 if (!sit_i) 4002 return; 4003 4004 if (sit_i->sentries) { 4005 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4006 kfree(sit_i->sentries[start].cur_valid_map); 4007 #ifdef CONFIG_F2FS_CHECK_FS 4008 kfree(sit_i->sentries[start].cur_valid_map_mir); 4009 #endif 4010 kfree(sit_i->sentries[start].ckpt_valid_map); 4011 kfree(sit_i->sentries[start].discard_map); 4012 } 4013 } 4014 kfree(sit_i->tmp_map); 4015 4016 kvfree(sit_i->sentries); 4017 kvfree(sit_i->sec_entries); 4018 kvfree(sit_i->dirty_sentries_bitmap); 4019 4020 SM_I(sbi)->sit_info = NULL; 4021 kfree(sit_i->sit_bitmap); 4022 #ifdef CONFIG_F2FS_CHECK_FS 4023 kfree(sit_i->sit_bitmap_mir); 4024 #endif 4025 kfree(sit_i); 4026 } 4027 4028 void destroy_segment_manager(struct f2fs_sb_info *sbi) 4029 { 4030 struct f2fs_sm_info *sm_info = SM_I(sbi); 4031 4032 if (!sm_info) 4033 return; 4034 destroy_flush_cmd_control(sbi, true); 4035 destroy_discard_cmd_control(sbi); 4036 destroy_dirty_segmap(sbi); 4037 destroy_curseg(sbi); 4038 destroy_free_segmap(sbi); 4039 destroy_sit_info(sbi); 4040 sbi->sm_info = NULL; 4041 kfree(sm_info); 4042 } 4043 4044 int __init create_segment_manager_caches(void) 4045 { 4046 discard_entry_slab = f2fs_kmem_cache_create("discard_entry", 4047 sizeof(struct discard_entry)); 4048 if (!discard_entry_slab) 4049 goto fail; 4050 4051 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd", 4052 sizeof(struct discard_cmd)); 4053 if (!discard_cmd_slab) 4054 goto destroy_discard_entry; 4055 4056 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set", 4057 sizeof(struct sit_entry_set)); 4058 if (!sit_entry_set_slab) 4059 goto destroy_discard_cmd; 4060 4061 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry", 4062 sizeof(struct inmem_pages)); 4063 if (!inmem_entry_slab) 4064 goto destroy_sit_entry_set; 4065 return 0; 4066 4067 destroy_sit_entry_set: 4068 kmem_cache_destroy(sit_entry_set_slab); 4069 destroy_discard_cmd: 4070 kmem_cache_destroy(discard_cmd_slab); 4071 destroy_discard_entry: 4072 kmem_cache_destroy(discard_entry_slab); 4073 fail: 4074 return -ENOMEM; 4075 } 4076 4077 void destroy_segment_manager_caches(void) 4078 { 4079 kmem_cache_destroy(sit_entry_set_slab); 4080 kmem_cache_destroy(discard_cmd_slab); 4081 kmem_cache_destroy(discard_entry_slab); 4082 kmem_cache_destroy(inmem_entry_slab); 4083 } 4084