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 f2fs_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 f2fs_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 = f2fs_get_dnode_of_data(&dn, page->index, 243 LOOKUP_NODE); 244 if (err) { 245 if (err == -ENOMEM) { 246 congestion_wait(BLK_RW_ASYNC, HZ/50); 247 cond_resched(); 248 goto retry; 249 } 250 err = -EAGAIN; 251 goto next; 252 } 253 f2fs_get_node_info(sbi, dn.nid, &ni); 254 if (cur->old_addr == NEW_ADDR) { 255 f2fs_invalidate_blocks(sbi, dn.data_blkaddr); 256 f2fs_update_data_blkaddr(&dn, NEW_ADDR); 257 } else 258 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, 259 cur->old_addr, ni.version, true, true); 260 f2fs_put_dnode(&dn); 261 } 262 next: 263 /* we don't need to invalidate this in the sccessful status */ 264 if (drop || recover) 265 ClearPageUptodate(page); 266 set_page_private(page, 0); 267 ClearPagePrivate(page); 268 f2fs_put_page(page, 1); 269 270 list_del(&cur->list); 271 kmem_cache_free(inmem_entry_slab, cur); 272 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES); 273 } 274 return err; 275 } 276 277 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure) 278 { 279 struct list_head *head = &sbi->inode_list[ATOMIC_FILE]; 280 struct inode *inode; 281 struct f2fs_inode_info *fi; 282 next: 283 spin_lock(&sbi->inode_lock[ATOMIC_FILE]); 284 if (list_empty(head)) { 285 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 286 return; 287 } 288 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist); 289 inode = igrab(&fi->vfs_inode); 290 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 291 292 if (inode) { 293 if (gc_failure) { 294 if (fi->i_gc_failures[GC_FAILURE_ATOMIC]) 295 goto drop; 296 goto skip; 297 } 298 drop: 299 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST); 300 f2fs_drop_inmem_pages(inode); 301 iput(inode); 302 } 303 skip: 304 congestion_wait(BLK_RW_ASYNC, HZ/50); 305 cond_resched(); 306 goto next; 307 } 308 309 void f2fs_drop_inmem_pages(struct inode *inode) 310 { 311 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 312 struct f2fs_inode_info *fi = F2FS_I(inode); 313 314 mutex_lock(&fi->inmem_lock); 315 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false); 316 spin_lock(&sbi->inode_lock[ATOMIC_FILE]); 317 if (!list_empty(&fi->inmem_ilist)) 318 list_del_init(&fi->inmem_ilist); 319 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 320 mutex_unlock(&fi->inmem_lock); 321 322 clear_inode_flag(inode, FI_ATOMIC_FILE); 323 fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0; 324 stat_dec_atomic_write(inode); 325 } 326 327 void f2fs_drop_inmem_page(struct inode *inode, struct page *page) 328 { 329 struct f2fs_inode_info *fi = F2FS_I(inode); 330 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 331 struct list_head *head = &fi->inmem_pages; 332 struct inmem_pages *cur = NULL; 333 334 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page)); 335 336 mutex_lock(&fi->inmem_lock); 337 list_for_each_entry(cur, head, list) { 338 if (cur->page == page) 339 break; 340 } 341 342 f2fs_bug_on(sbi, list_empty(head) || cur->page != page); 343 list_del(&cur->list); 344 mutex_unlock(&fi->inmem_lock); 345 346 dec_page_count(sbi, F2FS_INMEM_PAGES); 347 kmem_cache_free(inmem_entry_slab, cur); 348 349 ClearPageUptodate(page); 350 set_page_private(page, 0); 351 ClearPagePrivate(page); 352 f2fs_put_page(page, 0); 353 354 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE); 355 } 356 357 static int __f2fs_commit_inmem_pages(struct inode *inode) 358 { 359 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 360 struct f2fs_inode_info *fi = F2FS_I(inode); 361 struct inmem_pages *cur, *tmp; 362 struct f2fs_io_info fio = { 363 .sbi = sbi, 364 .ino = inode->i_ino, 365 .type = DATA, 366 .op = REQ_OP_WRITE, 367 .op_flags = REQ_SYNC | REQ_PRIO, 368 .io_type = FS_DATA_IO, 369 }; 370 struct list_head revoke_list; 371 pgoff_t last_idx = ULONG_MAX; 372 int err = 0; 373 374 INIT_LIST_HEAD(&revoke_list); 375 376 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) { 377 struct page *page = cur->page; 378 379 lock_page(page); 380 if (page->mapping == inode->i_mapping) { 381 trace_f2fs_commit_inmem_page(page, INMEM); 382 383 set_page_dirty(page); 384 f2fs_wait_on_page_writeback(page, DATA, true); 385 if (clear_page_dirty_for_io(page)) { 386 inode_dec_dirty_pages(inode); 387 f2fs_remove_dirty_inode(inode); 388 } 389 retry: 390 fio.page = page; 391 fio.old_blkaddr = NULL_ADDR; 392 fio.encrypted_page = NULL; 393 fio.need_lock = LOCK_DONE; 394 err = f2fs_do_write_data_page(&fio); 395 if (err) { 396 if (err == -ENOMEM) { 397 congestion_wait(BLK_RW_ASYNC, HZ/50); 398 cond_resched(); 399 goto retry; 400 } 401 unlock_page(page); 402 break; 403 } 404 /* record old blkaddr for revoking */ 405 cur->old_addr = fio.old_blkaddr; 406 last_idx = page->index; 407 } 408 unlock_page(page); 409 list_move_tail(&cur->list, &revoke_list); 410 } 411 412 if (last_idx != ULONG_MAX) 413 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA); 414 415 if (err) { 416 /* 417 * try to revoke all committed pages, but still we could fail 418 * due to no memory or other reason, if that happened, EAGAIN 419 * will be returned, which means in such case, transaction is 420 * already not integrity, caller should use journal to do the 421 * recovery or rewrite & commit last transaction. For other 422 * error number, revoking was done by filesystem itself. 423 */ 424 err = __revoke_inmem_pages(inode, &revoke_list, false, true); 425 426 /* drop all uncommitted pages */ 427 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false); 428 } else { 429 __revoke_inmem_pages(inode, &revoke_list, false, false); 430 } 431 432 return err; 433 } 434 435 int f2fs_commit_inmem_pages(struct inode *inode) 436 { 437 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 438 struct f2fs_inode_info *fi = F2FS_I(inode); 439 int err; 440 441 f2fs_balance_fs(sbi, true); 442 f2fs_lock_op(sbi); 443 444 set_inode_flag(inode, FI_ATOMIC_COMMIT); 445 446 mutex_lock(&fi->inmem_lock); 447 err = __f2fs_commit_inmem_pages(inode); 448 449 spin_lock(&sbi->inode_lock[ATOMIC_FILE]); 450 if (!list_empty(&fi->inmem_ilist)) 451 list_del_init(&fi->inmem_ilist); 452 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]); 453 mutex_unlock(&fi->inmem_lock); 454 455 clear_inode_flag(inode, FI_ATOMIC_COMMIT); 456 457 f2fs_unlock_op(sbi); 458 return err; 459 } 460 461 /* 462 * This function balances dirty node and dentry pages. 463 * In addition, it controls garbage collection. 464 */ 465 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need) 466 { 467 #ifdef CONFIG_F2FS_FAULT_INJECTION 468 if (time_to_inject(sbi, FAULT_CHECKPOINT)) { 469 f2fs_show_injection_info(FAULT_CHECKPOINT); 470 f2fs_stop_checkpoint(sbi, false); 471 } 472 #endif 473 474 /* balance_fs_bg is able to be pending */ 475 if (need && excess_cached_nats(sbi)) 476 f2fs_balance_fs_bg(sbi); 477 478 /* 479 * We should do GC or end up with checkpoint, if there are so many dirty 480 * dir/node pages without enough free segments. 481 */ 482 if (has_not_enough_free_secs(sbi, 0, 0)) { 483 mutex_lock(&sbi->gc_mutex); 484 f2fs_gc(sbi, false, false, NULL_SEGNO); 485 } 486 } 487 488 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi) 489 { 490 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 491 return; 492 493 /* try to shrink extent cache when there is no enough memory */ 494 if (!f2fs_available_free_memory(sbi, EXTENT_CACHE)) 495 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER); 496 497 /* check the # of cached NAT entries */ 498 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES)) 499 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK); 500 501 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) 502 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS); 503 else 504 f2fs_build_free_nids(sbi, false, false); 505 506 if (!is_idle(sbi) && !excess_dirty_nats(sbi)) 507 return; 508 509 /* checkpoint is the only way to shrink partial cached entries */ 510 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) || 511 !f2fs_available_free_memory(sbi, INO_ENTRIES) || 512 excess_prefree_segs(sbi) || 513 excess_dirty_nats(sbi) || 514 f2fs_time_over(sbi, CP_TIME)) { 515 if (test_opt(sbi, DATA_FLUSH)) { 516 struct blk_plug plug; 517 518 blk_start_plug(&plug); 519 f2fs_sync_dirty_inodes(sbi, FILE_INODE); 520 blk_finish_plug(&plug); 521 } 522 f2fs_sync_fs(sbi->sb, true); 523 stat_inc_bg_cp_count(sbi->stat_info); 524 } 525 } 526 527 static int __submit_flush_wait(struct f2fs_sb_info *sbi, 528 struct block_device *bdev) 529 { 530 struct bio *bio = f2fs_bio_alloc(sbi, 0, true); 531 int ret; 532 533 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH; 534 bio_set_dev(bio, bdev); 535 ret = submit_bio_wait(bio); 536 bio_put(bio); 537 538 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER), 539 test_opt(sbi, FLUSH_MERGE), ret); 540 return ret; 541 } 542 543 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino) 544 { 545 int ret = 0; 546 int i; 547 548 if (!sbi->s_ndevs) 549 return __submit_flush_wait(sbi, sbi->sb->s_bdev); 550 551 for (i = 0; i < sbi->s_ndevs; i++) { 552 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO)) 553 continue; 554 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 555 if (ret) 556 break; 557 } 558 return ret; 559 } 560 561 static int issue_flush_thread(void *data) 562 { 563 struct f2fs_sb_info *sbi = data; 564 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 565 wait_queue_head_t *q = &fcc->flush_wait_queue; 566 repeat: 567 if (kthread_should_stop()) 568 return 0; 569 570 sb_start_intwrite(sbi->sb); 571 572 if (!llist_empty(&fcc->issue_list)) { 573 struct flush_cmd *cmd, *next; 574 int ret; 575 576 fcc->dispatch_list = llist_del_all(&fcc->issue_list); 577 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); 578 579 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode); 580 581 ret = submit_flush_wait(sbi, cmd->ino); 582 atomic_inc(&fcc->issued_flush); 583 584 llist_for_each_entry_safe(cmd, next, 585 fcc->dispatch_list, llnode) { 586 cmd->ret = ret; 587 complete(&cmd->wait); 588 } 589 fcc->dispatch_list = NULL; 590 } 591 592 sb_end_intwrite(sbi->sb); 593 594 wait_event_interruptible(*q, 595 kthread_should_stop() || !llist_empty(&fcc->issue_list)); 596 goto repeat; 597 } 598 599 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino) 600 { 601 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 602 struct flush_cmd cmd; 603 int ret; 604 605 if (test_opt(sbi, NOBARRIER)) 606 return 0; 607 608 if (!test_opt(sbi, FLUSH_MERGE)) { 609 ret = submit_flush_wait(sbi, ino); 610 atomic_inc(&fcc->issued_flush); 611 return ret; 612 } 613 614 if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) { 615 ret = submit_flush_wait(sbi, ino); 616 atomic_dec(&fcc->issing_flush); 617 618 atomic_inc(&fcc->issued_flush); 619 return ret; 620 } 621 622 cmd.ino = ino; 623 init_completion(&cmd.wait); 624 625 llist_add(&cmd.llnode, &fcc->issue_list); 626 627 /* update issue_list before we wake up issue_flush thread */ 628 smp_mb(); 629 630 if (waitqueue_active(&fcc->flush_wait_queue)) 631 wake_up(&fcc->flush_wait_queue); 632 633 if (fcc->f2fs_issue_flush) { 634 wait_for_completion(&cmd.wait); 635 atomic_dec(&fcc->issing_flush); 636 } else { 637 struct llist_node *list; 638 639 list = llist_del_all(&fcc->issue_list); 640 if (!list) { 641 wait_for_completion(&cmd.wait); 642 atomic_dec(&fcc->issing_flush); 643 } else { 644 struct flush_cmd *tmp, *next; 645 646 ret = submit_flush_wait(sbi, ino); 647 648 llist_for_each_entry_safe(tmp, next, list, llnode) { 649 if (tmp == &cmd) { 650 cmd.ret = ret; 651 atomic_dec(&fcc->issing_flush); 652 continue; 653 } 654 tmp->ret = ret; 655 complete(&tmp->wait); 656 } 657 } 658 } 659 660 return cmd.ret; 661 } 662 663 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi) 664 { 665 dev_t dev = sbi->sb->s_bdev->bd_dev; 666 struct flush_cmd_control *fcc; 667 int err = 0; 668 669 if (SM_I(sbi)->fcc_info) { 670 fcc = SM_I(sbi)->fcc_info; 671 if (fcc->f2fs_issue_flush) 672 return err; 673 goto init_thread; 674 } 675 676 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL); 677 if (!fcc) 678 return -ENOMEM; 679 atomic_set(&fcc->issued_flush, 0); 680 atomic_set(&fcc->issing_flush, 0); 681 init_waitqueue_head(&fcc->flush_wait_queue); 682 init_llist_head(&fcc->issue_list); 683 SM_I(sbi)->fcc_info = fcc; 684 if (!test_opt(sbi, FLUSH_MERGE)) 685 return err; 686 687 init_thread: 688 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, 689 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); 690 if (IS_ERR(fcc->f2fs_issue_flush)) { 691 err = PTR_ERR(fcc->f2fs_issue_flush); 692 kfree(fcc); 693 SM_I(sbi)->fcc_info = NULL; 694 return err; 695 } 696 697 return err; 698 } 699 700 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free) 701 { 702 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 703 704 if (fcc && fcc->f2fs_issue_flush) { 705 struct task_struct *flush_thread = fcc->f2fs_issue_flush; 706 707 fcc->f2fs_issue_flush = NULL; 708 kthread_stop(flush_thread); 709 } 710 if (free) { 711 kfree(fcc); 712 SM_I(sbi)->fcc_info = NULL; 713 } 714 } 715 716 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi) 717 { 718 int ret = 0, i; 719 720 if (!sbi->s_ndevs) 721 return 0; 722 723 for (i = 1; i < sbi->s_ndevs; i++) { 724 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device)) 725 continue; 726 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 727 if (ret) 728 break; 729 730 spin_lock(&sbi->dev_lock); 731 f2fs_clear_bit(i, (char *)&sbi->dirty_device); 732 spin_unlock(&sbi->dev_lock); 733 } 734 735 return ret; 736 } 737 738 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 739 enum dirty_type dirty_type) 740 { 741 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 742 743 /* need not be added */ 744 if (IS_CURSEG(sbi, segno)) 745 return; 746 747 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 748 dirty_i->nr_dirty[dirty_type]++; 749 750 if (dirty_type == DIRTY) { 751 struct seg_entry *sentry = get_seg_entry(sbi, segno); 752 enum dirty_type t = sentry->type; 753 754 if (unlikely(t >= DIRTY)) { 755 f2fs_bug_on(sbi, 1); 756 return; 757 } 758 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) 759 dirty_i->nr_dirty[t]++; 760 } 761 } 762 763 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 764 enum dirty_type dirty_type) 765 { 766 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 767 768 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) 769 dirty_i->nr_dirty[dirty_type]--; 770 771 if (dirty_type == DIRTY) { 772 struct seg_entry *sentry = get_seg_entry(sbi, segno); 773 enum dirty_type t = sentry->type; 774 775 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) 776 dirty_i->nr_dirty[t]--; 777 778 if (get_valid_blocks(sbi, segno, true) == 0) 779 clear_bit(GET_SEC_FROM_SEG(sbi, segno), 780 dirty_i->victim_secmap); 781 } 782 } 783 784 /* 785 * Should not occur error such as -ENOMEM. 786 * Adding dirty entry into seglist is not critical operation. 787 * If a given segment is one of current working segments, it won't be added. 788 */ 789 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) 790 { 791 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 792 unsigned short valid_blocks; 793 794 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) 795 return; 796 797 mutex_lock(&dirty_i->seglist_lock); 798 799 valid_blocks = get_valid_blocks(sbi, segno, false); 800 801 if (valid_blocks == 0) { 802 __locate_dirty_segment(sbi, segno, PRE); 803 __remove_dirty_segment(sbi, segno, DIRTY); 804 } else if (valid_blocks < sbi->blocks_per_seg) { 805 __locate_dirty_segment(sbi, segno, DIRTY); 806 } else { 807 /* Recovery routine with SSR needs this */ 808 __remove_dirty_segment(sbi, segno, DIRTY); 809 } 810 811 mutex_unlock(&dirty_i->seglist_lock); 812 } 813 814 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi, 815 struct block_device *bdev, block_t lstart, 816 block_t start, block_t len) 817 { 818 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 819 struct list_head *pend_list; 820 struct discard_cmd *dc; 821 822 f2fs_bug_on(sbi, !len); 823 824 pend_list = &dcc->pend_list[plist_idx(len)]; 825 826 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS); 827 INIT_LIST_HEAD(&dc->list); 828 dc->bdev = bdev; 829 dc->lstart = lstart; 830 dc->start = start; 831 dc->len = len; 832 dc->ref = 0; 833 dc->state = D_PREP; 834 dc->error = 0; 835 init_completion(&dc->wait); 836 list_add_tail(&dc->list, pend_list); 837 atomic_inc(&dcc->discard_cmd_cnt); 838 dcc->undiscard_blks += len; 839 840 return dc; 841 } 842 843 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi, 844 struct block_device *bdev, block_t lstart, 845 block_t start, block_t len, 846 struct rb_node *parent, struct rb_node **p) 847 { 848 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 849 struct discard_cmd *dc; 850 851 dc = __create_discard_cmd(sbi, bdev, lstart, start, len); 852 853 rb_link_node(&dc->rb_node, parent, p); 854 rb_insert_color(&dc->rb_node, &dcc->root); 855 856 return dc; 857 } 858 859 static void __detach_discard_cmd(struct discard_cmd_control *dcc, 860 struct discard_cmd *dc) 861 { 862 if (dc->state == D_DONE) 863 atomic_dec(&dcc->issing_discard); 864 865 list_del(&dc->list); 866 rb_erase(&dc->rb_node, &dcc->root); 867 dcc->undiscard_blks -= dc->len; 868 869 kmem_cache_free(discard_cmd_slab, dc); 870 871 atomic_dec(&dcc->discard_cmd_cnt); 872 } 873 874 static void __remove_discard_cmd(struct f2fs_sb_info *sbi, 875 struct discard_cmd *dc) 876 { 877 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 878 879 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len); 880 881 f2fs_bug_on(sbi, dc->ref); 882 883 if (dc->error == -EOPNOTSUPP) 884 dc->error = 0; 885 886 if (dc->error) 887 f2fs_msg(sbi->sb, KERN_INFO, 888 "Issue discard(%u, %u, %u) failed, ret: %d", 889 dc->lstart, dc->start, dc->len, dc->error); 890 __detach_discard_cmd(dcc, dc); 891 } 892 893 static void f2fs_submit_discard_endio(struct bio *bio) 894 { 895 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private; 896 897 dc->error = blk_status_to_errno(bio->bi_status); 898 dc->state = D_DONE; 899 complete_all(&dc->wait); 900 bio_put(bio); 901 } 902 903 static void __check_sit_bitmap(struct f2fs_sb_info *sbi, 904 block_t start, block_t end) 905 { 906 #ifdef CONFIG_F2FS_CHECK_FS 907 struct seg_entry *sentry; 908 unsigned int segno; 909 block_t blk = start; 910 unsigned long offset, size, max_blocks = sbi->blocks_per_seg; 911 unsigned long *map; 912 913 while (blk < end) { 914 segno = GET_SEGNO(sbi, blk); 915 sentry = get_seg_entry(sbi, segno); 916 offset = GET_BLKOFF_FROM_SEG0(sbi, blk); 917 918 if (end < START_BLOCK(sbi, segno + 1)) 919 size = GET_BLKOFF_FROM_SEG0(sbi, end); 920 else 921 size = max_blocks; 922 map = (unsigned long *)(sentry->cur_valid_map); 923 offset = __find_rev_next_bit(map, size, offset); 924 f2fs_bug_on(sbi, offset != size); 925 blk = START_BLOCK(sbi, segno + 1); 926 } 927 #endif 928 } 929 930 static void __init_discard_policy(struct f2fs_sb_info *sbi, 931 struct discard_policy *dpolicy, 932 int discard_type, unsigned int granularity) 933 { 934 /* common policy */ 935 dpolicy->type = discard_type; 936 dpolicy->sync = true; 937 dpolicy->granularity = granularity; 938 939 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST; 940 dpolicy->io_aware_gran = MAX_PLIST_NUM; 941 942 if (discard_type == DPOLICY_BG) { 943 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME; 944 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME; 945 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME; 946 dpolicy->io_aware = true; 947 dpolicy->sync = false; 948 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) { 949 dpolicy->granularity = 1; 950 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME; 951 } 952 } else if (discard_type == DPOLICY_FORCE) { 953 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME; 954 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME; 955 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME; 956 dpolicy->io_aware = false; 957 } else if (discard_type == DPOLICY_FSTRIM) { 958 dpolicy->io_aware = false; 959 } else if (discard_type == DPOLICY_UMOUNT) { 960 dpolicy->max_requests = UINT_MAX; 961 dpolicy->io_aware = false; 962 } 963 } 964 965 966 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */ 967 static void __submit_discard_cmd(struct f2fs_sb_info *sbi, 968 struct discard_policy *dpolicy, 969 struct discard_cmd *dc) 970 { 971 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 972 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 973 &(dcc->fstrim_list) : &(dcc->wait_list); 974 struct bio *bio = NULL; 975 int flag = dpolicy->sync ? REQ_SYNC : 0; 976 977 if (dc->state != D_PREP) 978 return; 979 980 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 981 return; 982 983 trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len); 984 985 dc->error = __blkdev_issue_discard(dc->bdev, 986 SECTOR_FROM_BLOCK(dc->start), 987 SECTOR_FROM_BLOCK(dc->len), 988 GFP_NOFS, 0, &bio); 989 if (!dc->error) { 990 /* should keep before submission to avoid D_DONE right away */ 991 dc->state = D_SUBMIT; 992 atomic_inc(&dcc->issued_discard); 993 atomic_inc(&dcc->issing_discard); 994 if (bio) { 995 bio->bi_private = dc; 996 bio->bi_end_io = f2fs_submit_discard_endio; 997 bio->bi_opf |= flag; 998 submit_bio(bio); 999 list_move_tail(&dc->list, wait_list); 1000 __check_sit_bitmap(sbi, dc->start, dc->start + dc->len); 1001 1002 f2fs_update_iostat(sbi, FS_DISCARD, 1); 1003 } 1004 } else { 1005 __remove_discard_cmd(sbi, dc); 1006 } 1007 } 1008 1009 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi, 1010 struct block_device *bdev, block_t lstart, 1011 block_t start, block_t len, 1012 struct rb_node **insert_p, 1013 struct rb_node *insert_parent) 1014 { 1015 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1016 struct rb_node **p; 1017 struct rb_node *parent = NULL; 1018 struct discard_cmd *dc = NULL; 1019 1020 if (insert_p && insert_parent) { 1021 parent = insert_parent; 1022 p = insert_p; 1023 goto do_insert; 1024 } 1025 1026 p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart); 1027 do_insert: 1028 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p); 1029 if (!dc) 1030 return NULL; 1031 1032 return dc; 1033 } 1034 1035 static void __relocate_discard_cmd(struct discard_cmd_control *dcc, 1036 struct discard_cmd *dc) 1037 { 1038 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]); 1039 } 1040 1041 static void __punch_discard_cmd(struct f2fs_sb_info *sbi, 1042 struct discard_cmd *dc, block_t blkaddr) 1043 { 1044 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1045 struct discard_info di = dc->di; 1046 bool modified = false; 1047 1048 if (dc->state == D_DONE || dc->len == 1) { 1049 __remove_discard_cmd(sbi, dc); 1050 return; 1051 } 1052 1053 dcc->undiscard_blks -= di.len; 1054 1055 if (blkaddr > di.lstart) { 1056 dc->len = blkaddr - dc->lstart; 1057 dcc->undiscard_blks += dc->len; 1058 __relocate_discard_cmd(dcc, dc); 1059 modified = true; 1060 } 1061 1062 if (blkaddr < di.lstart + di.len - 1) { 1063 if (modified) { 1064 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1, 1065 di.start + blkaddr + 1 - di.lstart, 1066 di.lstart + di.len - 1 - blkaddr, 1067 NULL, NULL); 1068 } else { 1069 dc->lstart++; 1070 dc->len--; 1071 dc->start++; 1072 dcc->undiscard_blks += dc->len; 1073 __relocate_discard_cmd(dcc, dc); 1074 } 1075 } 1076 } 1077 1078 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1079 struct block_device *bdev, block_t lstart, 1080 block_t start, block_t len) 1081 { 1082 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1083 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1084 struct discard_cmd *dc; 1085 struct discard_info di = {0}; 1086 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1087 block_t end = lstart + len; 1088 1089 mutex_lock(&dcc->cmd_lock); 1090 1091 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root, 1092 NULL, lstart, 1093 (struct rb_entry **)&prev_dc, 1094 (struct rb_entry **)&next_dc, 1095 &insert_p, &insert_parent, true); 1096 if (dc) 1097 prev_dc = dc; 1098 1099 if (!prev_dc) { 1100 di.lstart = lstart; 1101 di.len = next_dc ? next_dc->lstart - lstart : len; 1102 di.len = min(di.len, len); 1103 di.start = start; 1104 } 1105 1106 while (1) { 1107 struct rb_node *node; 1108 bool merged = false; 1109 struct discard_cmd *tdc = NULL; 1110 1111 if (prev_dc) { 1112 di.lstart = prev_dc->lstart + prev_dc->len; 1113 if (di.lstart < lstart) 1114 di.lstart = lstart; 1115 if (di.lstart >= end) 1116 break; 1117 1118 if (!next_dc || next_dc->lstart > end) 1119 di.len = end - di.lstart; 1120 else 1121 di.len = next_dc->lstart - di.lstart; 1122 di.start = start + di.lstart - lstart; 1123 } 1124 1125 if (!di.len) 1126 goto next; 1127 1128 if (prev_dc && prev_dc->state == D_PREP && 1129 prev_dc->bdev == bdev && 1130 __is_discard_back_mergeable(&di, &prev_dc->di)) { 1131 prev_dc->di.len += di.len; 1132 dcc->undiscard_blks += di.len; 1133 __relocate_discard_cmd(dcc, prev_dc); 1134 di = prev_dc->di; 1135 tdc = prev_dc; 1136 merged = true; 1137 } 1138 1139 if (next_dc && next_dc->state == D_PREP && 1140 next_dc->bdev == bdev && 1141 __is_discard_front_mergeable(&di, &next_dc->di)) { 1142 next_dc->di.lstart = di.lstart; 1143 next_dc->di.len += di.len; 1144 next_dc->di.start = di.start; 1145 dcc->undiscard_blks += di.len; 1146 __relocate_discard_cmd(dcc, next_dc); 1147 if (tdc) 1148 __remove_discard_cmd(sbi, tdc); 1149 merged = true; 1150 } 1151 1152 if (!merged) { 1153 __insert_discard_tree(sbi, bdev, di.lstart, di.start, 1154 di.len, NULL, NULL); 1155 } 1156 next: 1157 prev_dc = next_dc; 1158 if (!prev_dc) 1159 break; 1160 1161 node = rb_next(&prev_dc->rb_node); 1162 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1163 } 1164 1165 mutex_unlock(&dcc->cmd_lock); 1166 } 1167 1168 static int __queue_discard_cmd(struct f2fs_sb_info *sbi, 1169 struct block_device *bdev, block_t blkstart, block_t blklen) 1170 { 1171 block_t lblkstart = blkstart; 1172 1173 trace_f2fs_queue_discard(bdev, blkstart, blklen); 1174 1175 if (sbi->s_ndevs) { 1176 int devi = f2fs_target_device_index(sbi, blkstart); 1177 1178 blkstart -= FDEV(devi).start_blk; 1179 } 1180 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen); 1181 return 0; 1182 } 1183 1184 static int __issue_discard_cmd(struct f2fs_sb_info *sbi, 1185 struct discard_policy *dpolicy) 1186 { 1187 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1188 struct list_head *pend_list; 1189 struct discard_cmd *dc, *tmp; 1190 struct blk_plug plug; 1191 int i, iter = 0, issued = 0; 1192 bool io_interrupted = false; 1193 1194 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1195 if (i + 1 < dpolicy->granularity) 1196 break; 1197 pend_list = &dcc->pend_list[i]; 1198 1199 mutex_lock(&dcc->cmd_lock); 1200 if (list_empty(pend_list)) 1201 goto next; 1202 if (unlikely(dcc->rbtree_check)) 1203 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi, 1204 &dcc->root)); 1205 blk_start_plug(&plug); 1206 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1207 f2fs_bug_on(sbi, dc->state != D_PREP); 1208 1209 if (dpolicy->io_aware && i < dpolicy->io_aware_gran && 1210 !is_idle(sbi)) { 1211 io_interrupted = true; 1212 goto skip; 1213 } 1214 1215 __submit_discard_cmd(sbi, dpolicy, dc); 1216 issued++; 1217 skip: 1218 if (++iter >= dpolicy->max_requests) 1219 break; 1220 } 1221 blk_finish_plug(&plug); 1222 next: 1223 mutex_unlock(&dcc->cmd_lock); 1224 1225 if (iter >= dpolicy->max_requests) 1226 break; 1227 } 1228 1229 if (!issued && io_interrupted) 1230 issued = -1; 1231 1232 return issued; 1233 } 1234 1235 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi) 1236 { 1237 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1238 struct list_head *pend_list; 1239 struct discard_cmd *dc, *tmp; 1240 int i; 1241 bool dropped = false; 1242 1243 mutex_lock(&dcc->cmd_lock); 1244 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1245 pend_list = &dcc->pend_list[i]; 1246 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1247 f2fs_bug_on(sbi, dc->state != D_PREP); 1248 __remove_discard_cmd(sbi, dc); 1249 dropped = true; 1250 } 1251 } 1252 mutex_unlock(&dcc->cmd_lock); 1253 1254 return dropped; 1255 } 1256 1257 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi) 1258 { 1259 __drop_discard_cmd(sbi); 1260 } 1261 1262 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi, 1263 struct discard_cmd *dc) 1264 { 1265 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1266 unsigned int len = 0; 1267 1268 wait_for_completion_io(&dc->wait); 1269 mutex_lock(&dcc->cmd_lock); 1270 f2fs_bug_on(sbi, dc->state != D_DONE); 1271 dc->ref--; 1272 if (!dc->ref) { 1273 if (!dc->error) 1274 len = dc->len; 1275 __remove_discard_cmd(sbi, dc); 1276 } 1277 mutex_unlock(&dcc->cmd_lock); 1278 1279 return len; 1280 } 1281 1282 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi, 1283 struct discard_policy *dpolicy, 1284 block_t start, block_t end) 1285 { 1286 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1287 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1288 &(dcc->fstrim_list) : &(dcc->wait_list); 1289 struct discard_cmd *dc, *tmp; 1290 bool need_wait; 1291 unsigned int trimmed = 0; 1292 1293 next: 1294 need_wait = false; 1295 1296 mutex_lock(&dcc->cmd_lock); 1297 list_for_each_entry_safe(dc, tmp, wait_list, list) { 1298 if (dc->lstart + dc->len <= start || end <= dc->lstart) 1299 continue; 1300 if (dc->len < dpolicy->granularity) 1301 continue; 1302 if (dc->state == D_DONE && !dc->ref) { 1303 wait_for_completion_io(&dc->wait); 1304 if (!dc->error) 1305 trimmed += dc->len; 1306 __remove_discard_cmd(sbi, dc); 1307 } else { 1308 dc->ref++; 1309 need_wait = true; 1310 break; 1311 } 1312 } 1313 mutex_unlock(&dcc->cmd_lock); 1314 1315 if (need_wait) { 1316 trimmed += __wait_one_discard_bio(sbi, dc); 1317 goto next; 1318 } 1319 1320 return trimmed; 1321 } 1322 1323 static void __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1324 struct discard_policy *dpolicy) 1325 { 1326 struct discard_policy dp; 1327 1328 if (dpolicy) { 1329 __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX); 1330 return; 1331 } 1332 1333 /* wait all */ 1334 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1); 1335 __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1336 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1); 1337 __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1338 } 1339 1340 /* This should be covered by global mutex, &sit_i->sentry_lock */ 1341 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr) 1342 { 1343 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1344 struct discard_cmd *dc; 1345 bool need_wait = false; 1346 1347 mutex_lock(&dcc->cmd_lock); 1348 dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root, 1349 NULL, blkaddr); 1350 if (dc) { 1351 if (dc->state == D_PREP) { 1352 __punch_discard_cmd(sbi, dc, blkaddr); 1353 } else { 1354 dc->ref++; 1355 need_wait = true; 1356 } 1357 } 1358 mutex_unlock(&dcc->cmd_lock); 1359 1360 if (need_wait) 1361 __wait_one_discard_bio(sbi, dc); 1362 } 1363 1364 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi) 1365 { 1366 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1367 1368 if (dcc && dcc->f2fs_issue_discard) { 1369 struct task_struct *discard_thread = dcc->f2fs_issue_discard; 1370 1371 dcc->f2fs_issue_discard = NULL; 1372 kthread_stop(discard_thread); 1373 } 1374 } 1375 1376 /* This comes from f2fs_put_super */ 1377 bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi) 1378 { 1379 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1380 struct discard_policy dpolicy; 1381 bool dropped; 1382 1383 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT, 1384 dcc->discard_granularity); 1385 __issue_discard_cmd(sbi, &dpolicy); 1386 dropped = __drop_discard_cmd(sbi); 1387 1388 /* just to make sure there is no pending discard commands */ 1389 __wait_all_discard_cmd(sbi, NULL); 1390 return dropped; 1391 } 1392 1393 static int issue_discard_thread(void *data) 1394 { 1395 struct f2fs_sb_info *sbi = data; 1396 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1397 wait_queue_head_t *q = &dcc->discard_wait_queue; 1398 struct discard_policy dpolicy; 1399 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME; 1400 int issued; 1401 1402 set_freezable(); 1403 1404 do { 1405 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG, 1406 dcc->discard_granularity); 1407 1408 wait_event_interruptible_timeout(*q, 1409 kthread_should_stop() || freezing(current) || 1410 dcc->discard_wake, 1411 msecs_to_jiffies(wait_ms)); 1412 1413 if (dcc->discard_wake) 1414 dcc->discard_wake = 0; 1415 1416 if (try_to_freeze()) 1417 continue; 1418 if (f2fs_readonly(sbi->sb)) 1419 continue; 1420 if (kthread_should_stop()) 1421 return 0; 1422 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 1423 wait_ms = dpolicy.max_interval; 1424 continue; 1425 } 1426 1427 if (sbi->gc_mode == GC_URGENT) 1428 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1); 1429 1430 sb_start_intwrite(sbi->sb); 1431 1432 issued = __issue_discard_cmd(sbi, &dpolicy); 1433 if (issued > 0) { 1434 __wait_all_discard_cmd(sbi, &dpolicy); 1435 wait_ms = dpolicy.min_interval; 1436 } else if (issued == -1){ 1437 wait_ms = dpolicy.mid_interval; 1438 } else { 1439 wait_ms = dpolicy.max_interval; 1440 } 1441 1442 sb_end_intwrite(sbi->sb); 1443 1444 } while (!kthread_should_stop()); 1445 return 0; 1446 } 1447 1448 #ifdef CONFIG_BLK_DEV_ZONED 1449 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi, 1450 struct block_device *bdev, block_t blkstart, block_t blklen) 1451 { 1452 sector_t sector, nr_sects; 1453 block_t lblkstart = blkstart; 1454 int devi = 0; 1455 1456 if (sbi->s_ndevs) { 1457 devi = f2fs_target_device_index(sbi, blkstart); 1458 blkstart -= FDEV(devi).start_blk; 1459 } 1460 1461 /* 1462 * We need to know the type of the zone: for conventional zones, 1463 * use regular discard if the drive supports it. For sequential 1464 * zones, reset the zone write pointer. 1465 */ 1466 switch (get_blkz_type(sbi, bdev, blkstart)) { 1467 1468 case BLK_ZONE_TYPE_CONVENTIONAL: 1469 if (!blk_queue_discard(bdev_get_queue(bdev))) 1470 return 0; 1471 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen); 1472 case BLK_ZONE_TYPE_SEQWRITE_REQ: 1473 case BLK_ZONE_TYPE_SEQWRITE_PREF: 1474 sector = SECTOR_FROM_BLOCK(blkstart); 1475 nr_sects = SECTOR_FROM_BLOCK(blklen); 1476 1477 if (sector & (bdev_zone_sectors(bdev) - 1) || 1478 nr_sects != bdev_zone_sectors(bdev)) { 1479 f2fs_msg(sbi->sb, KERN_INFO, 1480 "(%d) %s: Unaligned discard attempted (block %x + %x)", 1481 devi, sbi->s_ndevs ? FDEV(devi).path: "", 1482 blkstart, blklen); 1483 return -EIO; 1484 } 1485 trace_f2fs_issue_reset_zone(bdev, blkstart); 1486 return blkdev_reset_zones(bdev, sector, 1487 nr_sects, GFP_NOFS); 1488 default: 1489 /* Unknown zone type: broken device ? */ 1490 return -EIO; 1491 } 1492 } 1493 #endif 1494 1495 static int __issue_discard_async(struct f2fs_sb_info *sbi, 1496 struct block_device *bdev, block_t blkstart, block_t blklen) 1497 { 1498 #ifdef CONFIG_BLK_DEV_ZONED 1499 if (f2fs_sb_has_blkzoned(sbi->sb) && 1500 bdev_zoned_model(bdev) != BLK_ZONED_NONE) 1501 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen); 1502 #endif 1503 return __queue_discard_cmd(sbi, bdev, blkstart, blklen); 1504 } 1505 1506 static int f2fs_issue_discard(struct f2fs_sb_info *sbi, 1507 block_t blkstart, block_t blklen) 1508 { 1509 sector_t start = blkstart, len = 0; 1510 struct block_device *bdev; 1511 struct seg_entry *se; 1512 unsigned int offset; 1513 block_t i; 1514 int err = 0; 1515 1516 bdev = f2fs_target_device(sbi, blkstart, NULL); 1517 1518 for (i = blkstart; i < blkstart + blklen; i++, len++) { 1519 if (i != start) { 1520 struct block_device *bdev2 = 1521 f2fs_target_device(sbi, i, NULL); 1522 1523 if (bdev2 != bdev) { 1524 err = __issue_discard_async(sbi, bdev, 1525 start, len); 1526 if (err) 1527 return err; 1528 bdev = bdev2; 1529 start = i; 1530 len = 0; 1531 } 1532 } 1533 1534 se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); 1535 offset = GET_BLKOFF_FROM_SEG0(sbi, i); 1536 1537 if (!f2fs_test_and_set_bit(offset, se->discard_map)) 1538 sbi->discard_blks--; 1539 } 1540 1541 if (len) 1542 err = __issue_discard_async(sbi, bdev, start, len); 1543 return err; 1544 } 1545 1546 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc, 1547 bool check_only) 1548 { 1549 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 1550 int max_blocks = sbi->blocks_per_seg; 1551 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); 1552 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 1553 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 1554 unsigned long *discard_map = (unsigned long *)se->discard_map; 1555 unsigned long *dmap = SIT_I(sbi)->tmp_map; 1556 unsigned int start = 0, end = -1; 1557 bool force = (cpc->reason & CP_DISCARD); 1558 struct discard_entry *de = NULL; 1559 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list; 1560 int i; 1561 1562 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi)) 1563 return false; 1564 1565 if (!force) { 1566 if (!test_opt(sbi, DISCARD) || !se->valid_blocks || 1567 SM_I(sbi)->dcc_info->nr_discards >= 1568 SM_I(sbi)->dcc_info->max_discards) 1569 return false; 1570 } 1571 1572 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ 1573 for (i = 0; i < entries; i++) 1574 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : 1575 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; 1576 1577 while (force || SM_I(sbi)->dcc_info->nr_discards <= 1578 SM_I(sbi)->dcc_info->max_discards) { 1579 start = __find_rev_next_bit(dmap, max_blocks, end + 1); 1580 if (start >= max_blocks) 1581 break; 1582 1583 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); 1584 if (force && start && end != max_blocks 1585 && (end - start) < cpc->trim_minlen) 1586 continue; 1587 1588 if (check_only) 1589 return true; 1590 1591 if (!de) { 1592 de = f2fs_kmem_cache_alloc(discard_entry_slab, 1593 GFP_F2FS_ZERO); 1594 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start); 1595 list_add_tail(&de->list, head); 1596 } 1597 1598 for (i = start; i < end; i++) 1599 __set_bit_le(i, (void *)de->discard_map); 1600 1601 SM_I(sbi)->dcc_info->nr_discards += end - start; 1602 } 1603 return false; 1604 } 1605 1606 static void release_discard_addr(struct discard_entry *entry) 1607 { 1608 list_del(&entry->list); 1609 kmem_cache_free(discard_entry_slab, entry); 1610 } 1611 1612 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi) 1613 { 1614 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); 1615 struct discard_entry *entry, *this; 1616 1617 /* drop caches */ 1618 list_for_each_entry_safe(entry, this, head, list) 1619 release_discard_addr(entry); 1620 } 1621 1622 /* 1623 * Should call f2fs_clear_prefree_segments after checkpoint is done. 1624 */ 1625 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 1626 { 1627 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1628 unsigned int segno; 1629 1630 mutex_lock(&dirty_i->seglist_lock); 1631 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) 1632 __set_test_and_free(sbi, segno); 1633 mutex_unlock(&dirty_i->seglist_lock); 1634 } 1635 1636 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi, 1637 struct cp_control *cpc) 1638 { 1639 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1640 struct list_head *head = &dcc->entry_list; 1641 struct discard_entry *entry, *this; 1642 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1643 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; 1644 unsigned int start = 0, end = -1; 1645 unsigned int secno, start_segno; 1646 bool force = (cpc->reason & CP_DISCARD); 1647 1648 mutex_lock(&dirty_i->seglist_lock); 1649 1650 while (1) { 1651 int i; 1652 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); 1653 if (start >= MAIN_SEGS(sbi)) 1654 break; 1655 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), 1656 start + 1); 1657 1658 for (i = start; i < end; i++) 1659 clear_bit(i, prefree_map); 1660 1661 dirty_i->nr_dirty[PRE] -= end - start; 1662 1663 if (!test_opt(sbi, DISCARD)) 1664 continue; 1665 1666 if (force && start >= cpc->trim_start && 1667 (end - 1) <= cpc->trim_end) 1668 continue; 1669 1670 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) { 1671 f2fs_issue_discard(sbi, START_BLOCK(sbi, start), 1672 (end - start) << sbi->log_blocks_per_seg); 1673 continue; 1674 } 1675 next: 1676 secno = GET_SEC_FROM_SEG(sbi, start); 1677 start_segno = GET_SEG_FROM_SEC(sbi, secno); 1678 if (!IS_CURSEC(sbi, secno) && 1679 !get_valid_blocks(sbi, start, true)) 1680 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno), 1681 sbi->segs_per_sec << sbi->log_blocks_per_seg); 1682 1683 start = start_segno + sbi->segs_per_sec; 1684 if (start < end) 1685 goto next; 1686 else 1687 end = start - 1; 1688 } 1689 mutex_unlock(&dirty_i->seglist_lock); 1690 1691 /* send small discards */ 1692 list_for_each_entry_safe(entry, this, head, list) { 1693 unsigned int cur_pos = 0, next_pos, len, total_len = 0; 1694 bool is_valid = test_bit_le(0, entry->discard_map); 1695 1696 find_next: 1697 if (is_valid) { 1698 next_pos = find_next_zero_bit_le(entry->discard_map, 1699 sbi->blocks_per_seg, cur_pos); 1700 len = next_pos - cur_pos; 1701 1702 if (f2fs_sb_has_blkzoned(sbi->sb) || 1703 (force && len < cpc->trim_minlen)) 1704 goto skip; 1705 1706 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos, 1707 len); 1708 total_len += len; 1709 } else { 1710 next_pos = find_next_bit_le(entry->discard_map, 1711 sbi->blocks_per_seg, cur_pos); 1712 } 1713 skip: 1714 cur_pos = next_pos; 1715 is_valid = !is_valid; 1716 1717 if (cur_pos < sbi->blocks_per_seg) 1718 goto find_next; 1719 1720 release_discard_addr(entry); 1721 dcc->nr_discards -= total_len; 1722 } 1723 1724 wake_up_discard_thread(sbi, false); 1725 } 1726 1727 static int create_discard_cmd_control(struct f2fs_sb_info *sbi) 1728 { 1729 dev_t dev = sbi->sb->s_bdev->bd_dev; 1730 struct discard_cmd_control *dcc; 1731 int err = 0, i; 1732 1733 if (SM_I(sbi)->dcc_info) { 1734 dcc = SM_I(sbi)->dcc_info; 1735 goto init_thread; 1736 } 1737 1738 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL); 1739 if (!dcc) 1740 return -ENOMEM; 1741 1742 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY; 1743 INIT_LIST_HEAD(&dcc->entry_list); 1744 for (i = 0; i < MAX_PLIST_NUM; i++) 1745 INIT_LIST_HEAD(&dcc->pend_list[i]); 1746 INIT_LIST_HEAD(&dcc->wait_list); 1747 INIT_LIST_HEAD(&dcc->fstrim_list); 1748 mutex_init(&dcc->cmd_lock); 1749 atomic_set(&dcc->issued_discard, 0); 1750 atomic_set(&dcc->issing_discard, 0); 1751 atomic_set(&dcc->discard_cmd_cnt, 0); 1752 dcc->nr_discards = 0; 1753 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg; 1754 dcc->undiscard_blks = 0; 1755 dcc->root = RB_ROOT; 1756 dcc->rbtree_check = false; 1757 1758 init_waitqueue_head(&dcc->discard_wait_queue); 1759 SM_I(sbi)->dcc_info = dcc; 1760 init_thread: 1761 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi, 1762 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev)); 1763 if (IS_ERR(dcc->f2fs_issue_discard)) { 1764 err = PTR_ERR(dcc->f2fs_issue_discard); 1765 kfree(dcc); 1766 SM_I(sbi)->dcc_info = NULL; 1767 return err; 1768 } 1769 1770 return err; 1771 } 1772 1773 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi) 1774 { 1775 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1776 1777 if (!dcc) 1778 return; 1779 1780 f2fs_stop_discard_thread(sbi); 1781 1782 kfree(dcc); 1783 SM_I(sbi)->dcc_info = NULL; 1784 } 1785 1786 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 1787 { 1788 struct sit_info *sit_i = SIT_I(sbi); 1789 1790 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { 1791 sit_i->dirty_sentries++; 1792 return false; 1793 } 1794 1795 return true; 1796 } 1797 1798 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 1799 unsigned int segno, int modified) 1800 { 1801 struct seg_entry *se = get_seg_entry(sbi, segno); 1802 se->type = type; 1803 if (modified) 1804 __mark_sit_entry_dirty(sbi, segno); 1805 } 1806 1807 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 1808 { 1809 struct seg_entry *se; 1810 unsigned int segno, offset; 1811 long int new_vblocks; 1812 bool exist; 1813 #ifdef CONFIG_F2FS_CHECK_FS 1814 bool mir_exist; 1815 #endif 1816 1817 segno = GET_SEGNO(sbi, blkaddr); 1818 1819 se = get_seg_entry(sbi, segno); 1820 new_vblocks = se->valid_blocks + del; 1821 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 1822 1823 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) || 1824 (new_vblocks > sbi->blocks_per_seg))); 1825 1826 se->valid_blocks = new_vblocks; 1827 se->mtime = get_mtime(sbi, false); 1828 if (se->mtime > SIT_I(sbi)->max_mtime) 1829 SIT_I(sbi)->max_mtime = se->mtime; 1830 1831 /* Update valid block bitmap */ 1832 if (del > 0) { 1833 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map); 1834 #ifdef CONFIG_F2FS_CHECK_FS 1835 mir_exist = f2fs_test_and_set_bit(offset, 1836 se->cur_valid_map_mir); 1837 if (unlikely(exist != mir_exist)) { 1838 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error " 1839 "when setting bitmap, blk:%u, old bit:%d", 1840 blkaddr, exist); 1841 f2fs_bug_on(sbi, 1); 1842 } 1843 #endif 1844 if (unlikely(exist)) { 1845 f2fs_msg(sbi->sb, KERN_ERR, 1846 "Bitmap was wrongly set, blk:%u", blkaddr); 1847 f2fs_bug_on(sbi, 1); 1848 se->valid_blocks--; 1849 del = 0; 1850 } 1851 1852 if (f2fs_discard_en(sbi) && 1853 !f2fs_test_and_set_bit(offset, se->discard_map)) 1854 sbi->discard_blks--; 1855 1856 /* don't overwrite by SSR to keep node chain */ 1857 if (IS_NODESEG(se->type)) { 1858 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map)) 1859 se->ckpt_valid_blocks++; 1860 } 1861 } else { 1862 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map); 1863 #ifdef CONFIG_F2FS_CHECK_FS 1864 mir_exist = f2fs_test_and_clear_bit(offset, 1865 se->cur_valid_map_mir); 1866 if (unlikely(exist != mir_exist)) { 1867 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error " 1868 "when clearing bitmap, blk:%u, old bit:%d", 1869 blkaddr, exist); 1870 f2fs_bug_on(sbi, 1); 1871 } 1872 #endif 1873 if (unlikely(!exist)) { 1874 f2fs_msg(sbi->sb, KERN_ERR, 1875 "Bitmap was wrongly cleared, blk:%u", blkaddr); 1876 f2fs_bug_on(sbi, 1); 1877 se->valid_blocks++; 1878 del = 0; 1879 } 1880 1881 if (f2fs_discard_en(sbi) && 1882 f2fs_test_and_clear_bit(offset, se->discard_map)) 1883 sbi->discard_blks++; 1884 } 1885 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) 1886 se->ckpt_valid_blocks += del; 1887 1888 __mark_sit_entry_dirty(sbi, segno); 1889 1890 /* update total number of valid blocks to be written in ckpt area */ 1891 SIT_I(sbi)->written_valid_blocks += del; 1892 1893 if (sbi->segs_per_sec > 1) 1894 get_sec_entry(sbi, segno)->valid_blocks += del; 1895 } 1896 1897 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) 1898 { 1899 unsigned int segno = GET_SEGNO(sbi, addr); 1900 struct sit_info *sit_i = SIT_I(sbi); 1901 1902 f2fs_bug_on(sbi, addr == NULL_ADDR); 1903 if (addr == NEW_ADDR) 1904 return; 1905 1906 /* add it into sit main buffer */ 1907 down_write(&sit_i->sentry_lock); 1908 1909 update_sit_entry(sbi, addr, -1); 1910 1911 /* add it into dirty seglist */ 1912 locate_dirty_segment(sbi, segno); 1913 1914 up_write(&sit_i->sentry_lock); 1915 } 1916 1917 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr) 1918 { 1919 struct sit_info *sit_i = SIT_I(sbi); 1920 unsigned int segno, offset; 1921 struct seg_entry *se; 1922 bool is_cp = false; 1923 1924 if (!is_valid_data_blkaddr(sbi, blkaddr)) 1925 return true; 1926 1927 down_read(&sit_i->sentry_lock); 1928 1929 segno = GET_SEGNO(sbi, blkaddr); 1930 se = get_seg_entry(sbi, segno); 1931 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 1932 1933 if (f2fs_test_bit(offset, se->ckpt_valid_map)) 1934 is_cp = true; 1935 1936 up_read(&sit_i->sentry_lock); 1937 1938 return is_cp; 1939 } 1940 1941 /* 1942 * This function should be resided under the curseg_mutex lock 1943 */ 1944 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, 1945 struct f2fs_summary *sum) 1946 { 1947 struct curseg_info *curseg = CURSEG_I(sbi, type); 1948 void *addr = curseg->sum_blk; 1949 addr += curseg->next_blkoff * sizeof(struct f2fs_summary); 1950 memcpy(addr, sum, sizeof(struct f2fs_summary)); 1951 } 1952 1953 /* 1954 * Calculate the number of current summary pages for writing 1955 */ 1956 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) 1957 { 1958 int valid_sum_count = 0; 1959 int i, sum_in_page; 1960 1961 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 1962 if (sbi->ckpt->alloc_type[i] == SSR) 1963 valid_sum_count += sbi->blocks_per_seg; 1964 else { 1965 if (for_ra) 1966 valid_sum_count += le16_to_cpu( 1967 F2FS_CKPT(sbi)->cur_data_blkoff[i]); 1968 else 1969 valid_sum_count += curseg_blkoff(sbi, i); 1970 } 1971 } 1972 1973 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE - 1974 SUM_FOOTER_SIZE) / SUMMARY_SIZE; 1975 if (valid_sum_count <= sum_in_page) 1976 return 1; 1977 else if ((valid_sum_count - sum_in_page) <= 1978 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) 1979 return 2; 1980 return 3; 1981 } 1982 1983 /* 1984 * Caller should put this summary page 1985 */ 1986 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) 1987 { 1988 return f2fs_get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno)); 1989 } 1990 1991 void f2fs_update_meta_page(struct f2fs_sb_info *sbi, 1992 void *src, block_t blk_addr) 1993 { 1994 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 1995 1996 memcpy(page_address(page), src, PAGE_SIZE); 1997 set_page_dirty(page); 1998 f2fs_put_page(page, 1); 1999 } 2000 2001 static void write_sum_page(struct f2fs_sb_info *sbi, 2002 struct f2fs_summary_block *sum_blk, block_t blk_addr) 2003 { 2004 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr); 2005 } 2006 2007 static void write_current_sum_page(struct f2fs_sb_info *sbi, 2008 int type, block_t blk_addr) 2009 { 2010 struct curseg_info *curseg = CURSEG_I(sbi, type); 2011 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2012 struct f2fs_summary_block *src = curseg->sum_blk; 2013 struct f2fs_summary_block *dst; 2014 2015 dst = (struct f2fs_summary_block *)page_address(page); 2016 memset(dst, 0, PAGE_SIZE); 2017 2018 mutex_lock(&curseg->curseg_mutex); 2019 2020 down_read(&curseg->journal_rwsem); 2021 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE); 2022 up_read(&curseg->journal_rwsem); 2023 2024 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE); 2025 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE); 2026 2027 mutex_unlock(&curseg->curseg_mutex); 2028 2029 set_page_dirty(page); 2030 f2fs_put_page(page, 1); 2031 } 2032 2033 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type) 2034 { 2035 struct curseg_info *curseg = CURSEG_I(sbi, type); 2036 unsigned int segno = curseg->segno + 1; 2037 struct free_segmap_info *free_i = FREE_I(sbi); 2038 2039 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec) 2040 return !test_bit(segno, free_i->free_segmap); 2041 return 0; 2042 } 2043 2044 /* 2045 * Find a new segment from the free segments bitmap to right order 2046 * This function should be returned with success, otherwise BUG 2047 */ 2048 static void get_new_segment(struct f2fs_sb_info *sbi, 2049 unsigned int *newseg, bool new_sec, int dir) 2050 { 2051 struct free_segmap_info *free_i = FREE_I(sbi); 2052 unsigned int segno, secno, zoneno; 2053 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; 2054 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg); 2055 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg); 2056 unsigned int left_start = hint; 2057 bool init = true; 2058 int go_left = 0; 2059 int i; 2060 2061 spin_lock(&free_i->segmap_lock); 2062 2063 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { 2064 segno = find_next_zero_bit(free_i->free_segmap, 2065 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1); 2066 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1)) 2067 goto got_it; 2068 } 2069 find_other_zone: 2070 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 2071 if (secno >= MAIN_SECS(sbi)) { 2072 if (dir == ALLOC_RIGHT) { 2073 secno = find_next_zero_bit(free_i->free_secmap, 2074 MAIN_SECS(sbi), 0); 2075 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi)); 2076 } else { 2077 go_left = 1; 2078 left_start = hint - 1; 2079 } 2080 } 2081 if (go_left == 0) 2082 goto skip_left; 2083 2084 while (test_bit(left_start, free_i->free_secmap)) { 2085 if (left_start > 0) { 2086 left_start--; 2087 continue; 2088 } 2089 left_start = find_next_zero_bit(free_i->free_secmap, 2090 MAIN_SECS(sbi), 0); 2091 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi)); 2092 break; 2093 } 2094 secno = left_start; 2095 skip_left: 2096 segno = GET_SEG_FROM_SEC(sbi, secno); 2097 zoneno = GET_ZONE_FROM_SEC(sbi, secno); 2098 2099 /* give up on finding another zone */ 2100 if (!init) 2101 goto got_it; 2102 if (sbi->secs_per_zone == 1) 2103 goto got_it; 2104 if (zoneno == old_zoneno) 2105 goto got_it; 2106 if (dir == ALLOC_LEFT) { 2107 if (!go_left && zoneno + 1 >= total_zones) 2108 goto got_it; 2109 if (go_left && zoneno == 0) 2110 goto got_it; 2111 } 2112 for (i = 0; i < NR_CURSEG_TYPE; i++) 2113 if (CURSEG_I(sbi, i)->zone == zoneno) 2114 break; 2115 2116 if (i < NR_CURSEG_TYPE) { 2117 /* zone is in user, try another */ 2118 if (go_left) 2119 hint = zoneno * sbi->secs_per_zone - 1; 2120 else if (zoneno + 1 >= total_zones) 2121 hint = 0; 2122 else 2123 hint = (zoneno + 1) * sbi->secs_per_zone; 2124 init = false; 2125 goto find_other_zone; 2126 } 2127 got_it: 2128 /* set it as dirty segment in free segmap */ 2129 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); 2130 __set_inuse(sbi, segno); 2131 *newseg = segno; 2132 spin_unlock(&free_i->segmap_lock); 2133 } 2134 2135 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 2136 { 2137 struct curseg_info *curseg = CURSEG_I(sbi, type); 2138 struct summary_footer *sum_footer; 2139 2140 curseg->segno = curseg->next_segno; 2141 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno); 2142 curseg->next_blkoff = 0; 2143 curseg->next_segno = NULL_SEGNO; 2144 2145 sum_footer = &(curseg->sum_blk->footer); 2146 memset(sum_footer, 0, sizeof(struct summary_footer)); 2147 if (IS_DATASEG(type)) 2148 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 2149 if (IS_NODESEG(type)) 2150 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 2151 __set_sit_entry_type(sbi, type, curseg->segno, modified); 2152 } 2153 2154 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type) 2155 { 2156 /* if segs_per_sec is large than 1, we need to keep original policy. */ 2157 if (sbi->segs_per_sec != 1) 2158 return CURSEG_I(sbi, type)->segno; 2159 2160 if (test_opt(sbi, NOHEAP) && 2161 (type == CURSEG_HOT_DATA || IS_NODESEG(type))) 2162 return 0; 2163 2164 if (SIT_I(sbi)->last_victim[ALLOC_NEXT]) 2165 return SIT_I(sbi)->last_victim[ALLOC_NEXT]; 2166 2167 /* find segments from 0 to reuse freed segments */ 2168 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE) 2169 return 0; 2170 2171 return CURSEG_I(sbi, type)->segno; 2172 } 2173 2174 /* 2175 * Allocate a current working segment. 2176 * This function always allocates a free segment in LFS manner. 2177 */ 2178 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 2179 { 2180 struct curseg_info *curseg = CURSEG_I(sbi, type); 2181 unsigned int segno = curseg->segno; 2182 int dir = ALLOC_LEFT; 2183 2184 write_sum_page(sbi, curseg->sum_blk, 2185 GET_SUM_BLOCK(sbi, segno)); 2186 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA) 2187 dir = ALLOC_RIGHT; 2188 2189 if (test_opt(sbi, NOHEAP)) 2190 dir = ALLOC_RIGHT; 2191 2192 segno = __get_next_segno(sbi, type); 2193 get_new_segment(sbi, &segno, new_sec, dir); 2194 curseg->next_segno = segno; 2195 reset_curseg(sbi, type, 1); 2196 curseg->alloc_type = LFS; 2197 } 2198 2199 static void __next_free_blkoff(struct f2fs_sb_info *sbi, 2200 struct curseg_info *seg, block_t start) 2201 { 2202 struct seg_entry *se = get_seg_entry(sbi, seg->segno); 2203 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 2204 unsigned long *target_map = SIT_I(sbi)->tmp_map; 2205 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 2206 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 2207 int i, pos; 2208 2209 for (i = 0; i < entries; i++) 2210 target_map[i] = ckpt_map[i] | cur_map[i]; 2211 2212 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); 2213 2214 seg->next_blkoff = pos; 2215 } 2216 2217 /* 2218 * If a segment is written by LFS manner, next block offset is just obtained 2219 * by increasing the current block offset. However, if a segment is written by 2220 * SSR manner, next block offset obtained by calling __next_free_blkoff 2221 */ 2222 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, 2223 struct curseg_info *seg) 2224 { 2225 if (seg->alloc_type == SSR) 2226 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1); 2227 else 2228 seg->next_blkoff++; 2229 } 2230 2231 /* 2232 * This function always allocates a used segment(from dirty seglist) by SSR 2233 * manner, so it should recover the existing segment information of valid blocks 2234 */ 2235 static void change_curseg(struct f2fs_sb_info *sbi, int type) 2236 { 2237 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2238 struct curseg_info *curseg = CURSEG_I(sbi, type); 2239 unsigned int new_segno = curseg->next_segno; 2240 struct f2fs_summary_block *sum_node; 2241 struct page *sum_page; 2242 2243 write_sum_page(sbi, curseg->sum_blk, 2244 GET_SUM_BLOCK(sbi, curseg->segno)); 2245 __set_test_and_inuse(sbi, new_segno); 2246 2247 mutex_lock(&dirty_i->seglist_lock); 2248 __remove_dirty_segment(sbi, new_segno, PRE); 2249 __remove_dirty_segment(sbi, new_segno, DIRTY); 2250 mutex_unlock(&dirty_i->seglist_lock); 2251 2252 reset_curseg(sbi, type, 1); 2253 curseg->alloc_type = SSR; 2254 __next_free_blkoff(sbi, curseg, 0); 2255 2256 sum_page = f2fs_get_sum_page(sbi, new_segno); 2257 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 2258 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); 2259 f2fs_put_page(sum_page, 1); 2260 } 2261 2262 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type) 2263 { 2264 struct curseg_info *curseg = CURSEG_I(sbi, type); 2265 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; 2266 unsigned segno = NULL_SEGNO; 2267 int i, cnt; 2268 bool reversed = false; 2269 2270 /* f2fs_need_SSR() already forces to do this */ 2271 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) { 2272 curseg->next_segno = segno; 2273 return 1; 2274 } 2275 2276 /* For node segments, let's do SSR more intensively */ 2277 if (IS_NODESEG(type)) { 2278 if (type >= CURSEG_WARM_NODE) { 2279 reversed = true; 2280 i = CURSEG_COLD_NODE; 2281 } else { 2282 i = CURSEG_HOT_NODE; 2283 } 2284 cnt = NR_CURSEG_NODE_TYPE; 2285 } else { 2286 if (type >= CURSEG_WARM_DATA) { 2287 reversed = true; 2288 i = CURSEG_COLD_DATA; 2289 } else { 2290 i = CURSEG_HOT_DATA; 2291 } 2292 cnt = NR_CURSEG_DATA_TYPE; 2293 } 2294 2295 for (; cnt-- > 0; reversed ? i-- : i++) { 2296 if (i == type) 2297 continue; 2298 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) { 2299 curseg->next_segno = segno; 2300 return 1; 2301 } 2302 } 2303 return 0; 2304 } 2305 2306 /* 2307 * flush out current segment and replace it with new segment 2308 * This function should be returned with success, otherwise BUG 2309 */ 2310 static void allocate_segment_by_default(struct f2fs_sb_info *sbi, 2311 int type, bool force) 2312 { 2313 struct curseg_info *curseg = CURSEG_I(sbi, type); 2314 2315 if (force) 2316 new_curseg(sbi, type, true); 2317 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) && 2318 type == CURSEG_WARM_NODE) 2319 new_curseg(sbi, type, false); 2320 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type)) 2321 new_curseg(sbi, type, false); 2322 else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type)) 2323 change_curseg(sbi, type); 2324 else 2325 new_curseg(sbi, type, false); 2326 2327 stat_inc_seg_type(sbi, curseg); 2328 } 2329 2330 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi) 2331 { 2332 struct curseg_info *curseg; 2333 unsigned int old_segno; 2334 int i; 2335 2336 down_write(&SIT_I(sbi)->sentry_lock); 2337 2338 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2339 curseg = CURSEG_I(sbi, i); 2340 old_segno = curseg->segno; 2341 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true); 2342 locate_dirty_segment(sbi, old_segno); 2343 } 2344 2345 up_write(&SIT_I(sbi)->sentry_lock); 2346 } 2347 2348 static const struct segment_allocation default_salloc_ops = { 2349 .allocate_segment = allocate_segment_by_default, 2350 }; 2351 2352 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi, 2353 struct cp_control *cpc) 2354 { 2355 __u64 trim_start = cpc->trim_start; 2356 bool has_candidate = false; 2357 2358 down_write(&SIT_I(sbi)->sentry_lock); 2359 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) { 2360 if (add_discard_addrs(sbi, cpc, true)) { 2361 has_candidate = true; 2362 break; 2363 } 2364 } 2365 up_write(&SIT_I(sbi)->sentry_lock); 2366 2367 cpc->trim_start = trim_start; 2368 return has_candidate; 2369 } 2370 2371 static void __issue_discard_cmd_range(struct f2fs_sb_info *sbi, 2372 struct discard_policy *dpolicy, 2373 unsigned int start, unsigned int end) 2374 { 2375 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2376 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 2377 struct rb_node **insert_p = NULL, *insert_parent = NULL; 2378 struct discard_cmd *dc; 2379 struct blk_plug plug; 2380 int issued; 2381 2382 next: 2383 issued = 0; 2384 2385 mutex_lock(&dcc->cmd_lock); 2386 if (unlikely(dcc->rbtree_check)) 2387 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi, 2388 &dcc->root)); 2389 2390 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root, 2391 NULL, start, 2392 (struct rb_entry **)&prev_dc, 2393 (struct rb_entry **)&next_dc, 2394 &insert_p, &insert_parent, true); 2395 if (!dc) 2396 dc = next_dc; 2397 2398 blk_start_plug(&plug); 2399 2400 while (dc && dc->lstart <= end) { 2401 struct rb_node *node; 2402 2403 if (dc->len < dpolicy->granularity) 2404 goto skip; 2405 2406 if (dc->state != D_PREP) { 2407 list_move_tail(&dc->list, &dcc->fstrim_list); 2408 goto skip; 2409 } 2410 2411 __submit_discard_cmd(sbi, dpolicy, dc); 2412 2413 if (++issued >= dpolicy->max_requests) { 2414 start = dc->lstart + dc->len; 2415 2416 blk_finish_plug(&plug); 2417 mutex_unlock(&dcc->cmd_lock); 2418 __wait_all_discard_cmd(sbi, NULL); 2419 congestion_wait(BLK_RW_ASYNC, HZ/50); 2420 goto next; 2421 } 2422 skip: 2423 node = rb_next(&dc->rb_node); 2424 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 2425 2426 if (fatal_signal_pending(current)) 2427 break; 2428 } 2429 2430 blk_finish_plug(&plug); 2431 mutex_unlock(&dcc->cmd_lock); 2432 } 2433 2434 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) 2435 { 2436 __u64 start = F2FS_BYTES_TO_BLK(range->start); 2437 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; 2438 unsigned int start_segno, end_segno; 2439 block_t start_block, end_block; 2440 struct cp_control cpc; 2441 struct discard_policy dpolicy; 2442 unsigned long long trimmed = 0; 2443 int err = 0; 2444 2445 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize) 2446 return -EINVAL; 2447 2448 if (end <= MAIN_BLKADDR(sbi)) 2449 return -EINVAL; 2450 2451 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 2452 f2fs_msg(sbi->sb, KERN_WARNING, 2453 "Found FS corruption, run fsck to fix."); 2454 return -EIO; 2455 } 2456 2457 /* start/end segment number in main_area */ 2458 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); 2459 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : 2460 GET_SEGNO(sbi, end); 2461 2462 cpc.reason = CP_DISCARD; 2463 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen)); 2464 cpc.trim_start = start_segno; 2465 cpc.trim_end = end_segno; 2466 2467 if (sbi->discard_blks == 0) 2468 goto out; 2469 2470 mutex_lock(&sbi->gc_mutex); 2471 err = f2fs_write_checkpoint(sbi, &cpc); 2472 mutex_unlock(&sbi->gc_mutex); 2473 if (err) 2474 goto out; 2475 2476 /* 2477 * We filed discard candidates, but actually we don't need to wait for 2478 * all of them, since they'll be issued in idle time along with runtime 2479 * discard option. User configuration looks like using runtime discard 2480 * or periodic fstrim instead of it. 2481 */ 2482 if (test_opt(sbi, DISCARD)) 2483 goto out; 2484 2485 start_block = START_BLOCK(sbi, start_segno); 2486 end_block = START_BLOCK(sbi, end_segno + 1); 2487 2488 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen); 2489 __issue_discard_cmd_range(sbi, &dpolicy, start_block, end_block); 2490 2491 trimmed = __wait_discard_cmd_range(sbi, &dpolicy, 2492 start_block, end_block); 2493 range->len = F2FS_BLK_TO_BYTES(trimmed); 2494 out: 2495 return err; 2496 } 2497 2498 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type) 2499 { 2500 struct curseg_info *curseg = CURSEG_I(sbi, type); 2501 if (curseg->next_blkoff < sbi->blocks_per_seg) 2502 return true; 2503 return false; 2504 } 2505 2506 int f2fs_rw_hint_to_seg_type(enum rw_hint hint) 2507 { 2508 switch (hint) { 2509 case WRITE_LIFE_SHORT: 2510 return CURSEG_HOT_DATA; 2511 case WRITE_LIFE_EXTREME: 2512 return CURSEG_COLD_DATA; 2513 default: 2514 return CURSEG_WARM_DATA; 2515 } 2516 } 2517 2518 /* This returns write hints for each segment type. This hints will be 2519 * passed down to block layer. There are mapping tables which depend on 2520 * the mount option 'whint_mode'. 2521 * 2522 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET. 2523 * 2524 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users. 2525 * 2526 * User F2FS Block 2527 * ---- ---- ----- 2528 * META WRITE_LIFE_NOT_SET 2529 * HOT_NODE " 2530 * WARM_NODE " 2531 * COLD_NODE " 2532 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME 2533 * extension list " " 2534 * 2535 * -- buffered io 2536 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 2537 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 2538 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 2539 * WRITE_LIFE_NONE " " 2540 * WRITE_LIFE_MEDIUM " " 2541 * WRITE_LIFE_LONG " " 2542 * 2543 * -- direct io 2544 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 2545 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 2546 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 2547 * WRITE_LIFE_NONE " WRITE_LIFE_NONE 2548 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM 2549 * WRITE_LIFE_LONG " WRITE_LIFE_LONG 2550 * 2551 * 3) whint_mode=fs-based. F2FS passes down hints with its policy. 2552 * 2553 * User F2FS Block 2554 * ---- ---- ----- 2555 * META WRITE_LIFE_MEDIUM; 2556 * HOT_NODE WRITE_LIFE_NOT_SET 2557 * WARM_NODE " 2558 * COLD_NODE WRITE_LIFE_NONE 2559 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME 2560 * extension list " " 2561 * 2562 * -- buffered io 2563 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 2564 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 2565 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG 2566 * WRITE_LIFE_NONE " " 2567 * WRITE_LIFE_MEDIUM " " 2568 * WRITE_LIFE_LONG " " 2569 * 2570 * -- direct io 2571 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 2572 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 2573 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 2574 * WRITE_LIFE_NONE " WRITE_LIFE_NONE 2575 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM 2576 * WRITE_LIFE_LONG " WRITE_LIFE_LONG 2577 */ 2578 2579 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi, 2580 enum page_type type, enum temp_type temp) 2581 { 2582 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) { 2583 if (type == DATA) { 2584 if (temp == WARM) 2585 return WRITE_LIFE_NOT_SET; 2586 else if (temp == HOT) 2587 return WRITE_LIFE_SHORT; 2588 else if (temp == COLD) 2589 return WRITE_LIFE_EXTREME; 2590 } else { 2591 return WRITE_LIFE_NOT_SET; 2592 } 2593 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) { 2594 if (type == DATA) { 2595 if (temp == WARM) 2596 return WRITE_LIFE_LONG; 2597 else if (temp == HOT) 2598 return WRITE_LIFE_SHORT; 2599 else if (temp == COLD) 2600 return WRITE_LIFE_EXTREME; 2601 } else if (type == NODE) { 2602 if (temp == WARM || temp == HOT) 2603 return WRITE_LIFE_NOT_SET; 2604 else if (temp == COLD) 2605 return WRITE_LIFE_NONE; 2606 } else if (type == META) { 2607 return WRITE_LIFE_MEDIUM; 2608 } 2609 } 2610 return WRITE_LIFE_NOT_SET; 2611 } 2612 2613 static int __get_segment_type_2(struct f2fs_io_info *fio) 2614 { 2615 if (fio->type == DATA) 2616 return CURSEG_HOT_DATA; 2617 else 2618 return CURSEG_HOT_NODE; 2619 } 2620 2621 static int __get_segment_type_4(struct f2fs_io_info *fio) 2622 { 2623 if (fio->type == DATA) { 2624 struct inode *inode = fio->page->mapping->host; 2625 2626 if (S_ISDIR(inode->i_mode)) 2627 return CURSEG_HOT_DATA; 2628 else 2629 return CURSEG_COLD_DATA; 2630 } else { 2631 if (IS_DNODE(fio->page) && is_cold_node(fio->page)) 2632 return CURSEG_WARM_NODE; 2633 else 2634 return CURSEG_COLD_NODE; 2635 } 2636 } 2637 2638 static int __get_segment_type_6(struct f2fs_io_info *fio) 2639 { 2640 if (fio->type == DATA) { 2641 struct inode *inode = fio->page->mapping->host; 2642 2643 if (is_cold_data(fio->page) || file_is_cold(inode)) 2644 return CURSEG_COLD_DATA; 2645 if (file_is_hot(inode) || 2646 is_inode_flag_set(inode, FI_HOT_DATA) || 2647 is_inode_flag_set(inode, FI_ATOMIC_FILE) || 2648 is_inode_flag_set(inode, FI_VOLATILE_FILE)) 2649 return CURSEG_HOT_DATA; 2650 return f2fs_rw_hint_to_seg_type(inode->i_write_hint); 2651 } else { 2652 if (IS_DNODE(fio->page)) 2653 return is_cold_node(fio->page) ? CURSEG_WARM_NODE : 2654 CURSEG_HOT_NODE; 2655 return CURSEG_COLD_NODE; 2656 } 2657 } 2658 2659 static int __get_segment_type(struct f2fs_io_info *fio) 2660 { 2661 int type = 0; 2662 2663 switch (F2FS_OPTION(fio->sbi).active_logs) { 2664 case 2: 2665 type = __get_segment_type_2(fio); 2666 break; 2667 case 4: 2668 type = __get_segment_type_4(fio); 2669 break; 2670 case 6: 2671 type = __get_segment_type_6(fio); 2672 break; 2673 default: 2674 f2fs_bug_on(fio->sbi, true); 2675 } 2676 2677 if (IS_HOT(type)) 2678 fio->temp = HOT; 2679 else if (IS_WARM(type)) 2680 fio->temp = WARM; 2681 else 2682 fio->temp = COLD; 2683 return type; 2684 } 2685 2686 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, 2687 block_t old_blkaddr, block_t *new_blkaddr, 2688 struct f2fs_summary *sum, int type, 2689 struct f2fs_io_info *fio, bool add_list) 2690 { 2691 struct sit_info *sit_i = SIT_I(sbi); 2692 struct curseg_info *curseg = CURSEG_I(sbi, type); 2693 2694 down_read(&SM_I(sbi)->curseg_lock); 2695 2696 mutex_lock(&curseg->curseg_mutex); 2697 down_write(&sit_i->sentry_lock); 2698 2699 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 2700 2701 f2fs_wait_discard_bio(sbi, *new_blkaddr); 2702 2703 /* 2704 * __add_sum_entry should be resided under the curseg_mutex 2705 * because, this function updates a summary entry in the 2706 * current summary block. 2707 */ 2708 __add_sum_entry(sbi, type, sum); 2709 2710 __refresh_next_blkoff(sbi, curseg); 2711 2712 stat_inc_block_count(sbi, curseg); 2713 2714 /* 2715 * SIT information should be updated before segment allocation, 2716 * since SSR needs latest valid block information. 2717 */ 2718 update_sit_entry(sbi, *new_blkaddr, 1); 2719 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 2720 update_sit_entry(sbi, old_blkaddr, -1); 2721 2722 if (!__has_curseg_space(sbi, type)) 2723 sit_i->s_ops->allocate_segment(sbi, type, false); 2724 2725 /* 2726 * segment dirty status should be updated after segment allocation, 2727 * so we just need to update status only one time after previous 2728 * segment being closed. 2729 */ 2730 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 2731 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr)); 2732 2733 up_write(&sit_i->sentry_lock); 2734 2735 if (page && IS_NODESEG(type)) { 2736 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); 2737 2738 f2fs_inode_chksum_set(sbi, page); 2739 } 2740 2741 if (add_list) { 2742 struct f2fs_bio_info *io; 2743 2744 INIT_LIST_HEAD(&fio->list); 2745 fio->in_list = true; 2746 fio->retry = false; 2747 io = sbi->write_io[fio->type] + fio->temp; 2748 spin_lock(&io->io_lock); 2749 list_add_tail(&fio->list, &io->io_list); 2750 spin_unlock(&io->io_lock); 2751 } 2752 2753 mutex_unlock(&curseg->curseg_mutex); 2754 2755 up_read(&SM_I(sbi)->curseg_lock); 2756 } 2757 2758 static void update_device_state(struct f2fs_io_info *fio) 2759 { 2760 struct f2fs_sb_info *sbi = fio->sbi; 2761 unsigned int devidx; 2762 2763 if (!sbi->s_ndevs) 2764 return; 2765 2766 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr); 2767 2768 /* update device state for fsync */ 2769 f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO); 2770 2771 /* update device state for checkpoint */ 2772 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) { 2773 spin_lock(&sbi->dev_lock); 2774 f2fs_set_bit(devidx, (char *)&sbi->dirty_device); 2775 spin_unlock(&sbi->dev_lock); 2776 } 2777 } 2778 2779 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) 2780 { 2781 int type = __get_segment_type(fio); 2782 bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA); 2783 2784 if (keep_order) 2785 down_read(&fio->sbi->io_order_lock); 2786 reallocate: 2787 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr, 2788 &fio->new_blkaddr, sum, type, fio, true); 2789 2790 /* writeout dirty page into bdev */ 2791 f2fs_submit_page_write(fio); 2792 if (fio->retry) { 2793 fio->old_blkaddr = fio->new_blkaddr; 2794 goto reallocate; 2795 } 2796 2797 update_device_state(fio); 2798 2799 if (keep_order) 2800 up_read(&fio->sbi->io_order_lock); 2801 } 2802 2803 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page, 2804 enum iostat_type io_type) 2805 { 2806 struct f2fs_io_info fio = { 2807 .sbi = sbi, 2808 .type = META, 2809 .temp = HOT, 2810 .op = REQ_OP_WRITE, 2811 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO, 2812 .old_blkaddr = page->index, 2813 .new_blkaddr = page->index, 2814 .page = page, 2815 .encrypted_page = NULL, 2816 .in_list = false, 2817 }; 2818 2819 if (unlikely(page->index >= MAIN_BLKADDR(sbi))) 2820 fio.op_flags &= ~REQ_META; 2821 2822 set_page_writeback(page); 2823 ClearPageError(page); 2824 f2fs_submit_page_write(&fio); 2825 2826 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE); 2827 } 2828 2829 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio) 2830 { 2831 struct f2fs_summary sum; 2832 2833 set_summary(&sum, nid, 0, 0); 2834 do_write_page(&sum, fio); 2835 2836 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE); 2837 } 2838 2839 void f2fs_outplace_write_data(struct dnode_of_data *dn, 2840 struct f2fs_io_info *fio) 2841 { 2842 struct f2fs_sb_info *sbi = fio->sbi; 2843 struct f2fs_summary sum; 2844 struct node_info ni; 2845 2846 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); 2847 f2fs_get_node_info(sbi, dn->nid, &ni); 2848 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); 2849 do_write_page(&sum, fio); 2850 f2fs_update_data_blkaddr(dn, fio->new_blkaddr); 2851 2852 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE); 2853 } 2854 2855 int f2fs_inplace_write_data(struct f2fs_io_info *fio) 2856 { 2857 int err; 2858 struct f2fs_sb_info *sbi = fio->sbi; 2859 2860 fio->new_blkaddr = fio->old_blkaddr; 2861 /* i/o temperature is needed for passing down write hints */ 2862 __get_segment_type(fio); 2863 2864 f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi, 2865 GET_SEGNO(sbi, fio->new_blkaddr))->type)); 2866 2867 stat_inc_inplace_blocks(fio->sbi); 2868 2869 err = f2fs_submit_page_bio(fio); 2870 if (!err) 2871 update_device_state(fio); 2872 2873 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE); 2874 2875 return err; 2876 } 2877 2878 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi, 2879 unsigned int segno) 2880 { 2881 int i; 2882 2883 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 2884 if (CURSEG_I(sbi, i)->segno == segno) 2885 break; 2886 } 2887 return i; 2888 } 2889 2890 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 2891 block_t old_blkaddr, block_t new_blkaddr, 2892 bool recover_curseg, bool recover_newaddr) 2893 { 2894 struct sit_info *sit_i = SIT_I(sbi); 2895 struct curseg_info *curseg; 2896 unsigned int segno, old_cursegno; 2897 struct seg_entry *se; 2898 int type; 2899 unsigned short old_blkoff; 2900 2901 segno = GET_SEGNO(sbi, new_blkaddr); 2902 se = get_seg_entry(sbi, segno); 2903 type = se->type; 2904 2905 down_write(&SM_I(sbi)->curseg_lock); 2906 2907 if (!recover_curseg) { 2908 /* for recovery flow */ 2909 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { 2910 if (old_blkaddr == NULL_ADDR) 2911 type = CURSEG_COLD_DATA; 2912 else 2913 type = CURSEG_WARM_DATA; 2914 } 2915 } else { 2916 if (IS_CURSEG(sbi, segno)) { 2917 /* se->type is volatile as SSR allocation */ 2918 type = __f2fs_get_curseg(sbi, segno); 2919 f2fs_bug_on(sbi, type == NO_CHECK_TYPE); 2920 } else { 2921 type = CURSEG_WARM_DATA; 2922 } 2923 } 2924 2925 f2fs_bug_on(sbi, !IS_DATASEG(type)); 2926 curseg = CURSEG_I(sbi, type); 2927 2928 mutex_lock(&curseg->curseg_mutex); 2929 down_write(&sit_i->sentry_lock); 2930 2931 old_cursegno = curseg->segno; 2932 old_blkoff = curseg->next_blkoff; 2933 2934 /* change the current segment */ 2935 if (segno != curseg->segno) { 2936 curseg->next_segno = segno; 2937 change_curseg(sbi, type); 2938 } 2939 2940 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); 2941 __add_sum_entry(sbi, type, sum); 2942 2943 if (!recover_curseg || recover_newaddr) 2944 update_sit_entry(sbi, new_blkaddr, 1); 2945 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 2946 update_sit_entry(sbi, old_blkaddr, -1); 2947 2948 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 2949 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); 2950 2951 locate_dirty_segment(sbi, old_cursegno); 2952 2953 if (recover_curseg) { 2954 if (old_cursegno != curseg->segno) { 2955 curseg->next_segno = old_cursegno; 2956 change_curseg(sbi, type); 2957 } 2958 curseg->next_blkoff = old_blkoff; 2959 } 2960 2961 up_write(&sit_i->sentry_lock); 2962 mutex_unlock(&curseg->curseg_mutex); 2963 up_write(&SM_I(sbi)->curseg_lock); 2964 } 2965 2966 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, 2967 block_t old_addr, block_t new_addr, 2968 unsigned char version, bool recover_curseg, 2969 bool recover_newaddr) 2970 { 2971 struct f2fs_summary sum; 2972 2973 set_summary(&sum, dn->nid, dn->ofs_in_node, version); 2974 2975 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr, 2976 recover_curseg, recover_newaddr); 2977 2978 f2fs_update_data_blkaddr(dn, new_addr); 2979 } 2980 2981 void f2fs_wait_on_page_writeback(struct page *page, 2982 enum page_type type, bool ordered) 2983 { 2984 if (PageWriteback(page)) { 2985 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 2986 2987 f2fs_submit_merged_write_cond(sbi, page->mapping->host, 2988 0, page->index, type); 2989 if (ordered) 2990 wait_on_page_writeback(page); 2991 else 2992 wait_for_stable_page(page); 2993 } 2994 } 2995 2996 void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr) 2997 { 2998 struct page *cpage; 2999 3000 if (!is_valid_data_blkaddr(sbi, blkaddr)) 3001 return; 3002 3003 cpage = find_lock_page(META_MAPPING(sbi), blkaddr); 3004 if (cpage) { 3005 f2fs_wait_on_page_writeback(cpage, DATA, true); 3006 f2fs_put_page(cpage, 1); 3007 } 3008 } 3009 3010 static void read_compacted_summaries(struct f2fs_sb_info *sbi) 3011 { 3012 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3013 struct curseg_info *seg_i; 3014 unsigned char *kaddr; 3015 struct page *page; 3016 block_t start; 3017 int i, j, offset; 3018 3019 start = start_sum_block(sbi); 3020 3021 page = f2fs_get_meta_page(sbi, start++); 3022 kaddr = (unsigned char *)page_address(page); 3023 3024 /* Step 1: restore nat cache */ 3025 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 3026 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE); 3027 3028 /* Step 2: restore sit cache */ 3029 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 3030 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE); 3031 offset = 2 * SUM_JOURNAL_SIZE; 3032 3033 /* Step 3: restore summary entries */ 3034 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 3035 unsigned short blk_off; 3036 unsigned int segno; 3037 3038 seg_i = CURSEG_I(sbi, i); 3039 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 3040 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 3041 seg_i->next_segno = segno; 3042 reset_curseg(sbi, i, 0); 3043 seg_i->alloc_type = ckpt->alloc_type[i]; 3044 seg_i->next_blkoff = blk_off; 3045 3046 if (seg_i->alloc_type == SSR) 3047 blk_off = sbi->blocks_per_seg; 3048 3049 for (j = 0; j < blk_off; j++) { 3050 struct f2fs_summary *s; 3051 s = (struct f2fs_summary *)(kaddr + offset); 3052 seg_i->sum_blk->entries[j] = *s; 3053 offset += SUMMARY_SIZE; 3054 if (offset + SUMMARY_SIZE <= PAGE_SIZE - 3055 SUM_FOOTER_SIZE) 3056 continue; 3057 3058 f2fs_put_page(page, 1); 3059 page = NULL; 3060 3061 page = f2fs_get_meta_page(sbi, start++); 3062 kaddr = (unsigned char *)page_address(page); 3063 offset = 0; 3064 } 3065 } 3066 f2fs_put_page(page, 1); 3067 } 3068 3069 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 3070 { 3071 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3072 struct f2fs_summary_block *sum; 3073 struct curseg_info *curseg; 3074 struct page *new; 3075 unsigned short blk_off; 3076 unsigned int segno = 0; 3077 block_t blk_addr = 0; 3078 3079 /* get segment number and block addr */ 3080 if (IS_DATASEG(type)) { 3081 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 3082 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 3083 CURSEG_HOT_DATA]); 3084 if (__exist_node_summaries(sbi)) 3085 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type); 3086 else 3087 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 3088 } else { 3089 segno = le32_to_cpu(ckpt->cur_node_segno[type - 3090 CURSEG_HOT_NODE]); 3091 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 3092 CURSEG_HOT_NODE]); 3093 if (__exist_node_summaries(sbi)) 3094 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 3095 type - CURSEG_HOT_NODE); 3096 else 3097 blk_addr = GET_SUM_BLOCK(sbi, segno); 3098 } 3099 3100 new = f2fs_get_meta_page(sbi, blk_addr); 3101 sum = (struct f2fs_summary_block *)page_address(new); 3102 3103 if (IS_NODESEG(type)) { 3104 if (__exist_node_summaries(sbi)) { 3105 struct f2fs_summary *ns = &sum->entries[0]; 3106 int i; 3107 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { 3108 ns->version = 0; 3109 ns->ofs_in_node = 0; 3110 } 3111 } else { 3112 f2fs_restore_node_summary(sbi, segno, sum); 3113 } 3114 } 3115 3116 /* set uncompleted segment to curseg */ 3117 curseg = CURSEG_I(sbi, type); 3118 mutex_lock(&curseg->curseg_mutex); 3119 3120 /* update journal info */ 3121 down_write(&curseg->journal_rwsem); 3122 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE); 3123 up_write(&curseg->journal_rwsem); 3124 3125 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE); 3126 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE); 3127 curseg->next_segno = segno; 3128 reset_curseg(sbi, type, 0); 3129 curseg->alloc_type = ckpt->alloc_type[type]; 3130 curseg->next_blkoff = blk_off; 3131 mutex_unlock(&curseg->curseg_mutex); 3132 f2fs_put_page(new, 1); 3133 return 0; 3134 } 3135 3136 static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 3137 { 3138 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal; 3139 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal; 3140 int type = CURSEG_HOT_DATA; 3141 int err; 3142 3143 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) { 3144 int npages = f2fs_npages_for_summary_flush(sbi, true); 3145 3146 if (npages >= 2) 3147 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages, 3148 META_CP, true); 3149 3150 /* restore for compacted data summary */ 3151 read_compacted_summaries(sbi); 3152 type = CURSEG_HOT_NODE; 3153 } 3154 3155 if (__exist_node_summaries(sbi)) 3156 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type), 3157 NR_CURSEG_TYPE - type, META_CP, true); 3158 3159 for (; type <= CURSEG_COLD_NODE; type++) { 3160 err = read_normal_summaries(sbi, type); 3161 if (err) 3162 return err; 3163 } 3164 3165 /* sanity check for summary blocks */ 3166 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES || 3167 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) 3168 return -EINVAL; 3169 3170 return 0; 3171 } 3172 3173 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 3174 { 3175 struct page *page; 3176 unsigned char *kaddr; 3177 struct f2fs_summary *summary; 3178 struct curseg_info *seg_i; 3179 int written_size = 0; 3180 int i, j; 3181 3182 page = f2fs_grab_meta_page(sbi, blkaddr++); 3183 kaddr = (unsigned char *)page_address(page); 3184 memset(kaddr, 0, PAGE_SIZE); 3185 3186 /* Step 1: write nat cache */ 3187 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 3188 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE); 3189 written_size += SUM_JOURNAL_SIZE; 3190 3191 /* Step 2: write sit cache */ 3192 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 3193 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE); 3194 written_size += SUM_JOURNAL_SIZE; 3195 3196 /* Step 3: write summary entries */ 3197 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 3198 unsigned short blkoff; 3199 seg_i = CURSEG_I(sbi, i); 3200 if (sbi->ckpt->alloc_type[i] == SSR) 3201 blkoff = sbi->blocks_per_seg; 3202 else 3203 blkoff = curseg_blkoff(sbi, i); 3204 3205 for (j = 0; j < blkoff; j++) { 3206 if (!page) { 3207 page = f2fs_grab_meta_page(sbi, blkaddr++); 3208 kaddr = (unsigned char *)page_address(page); 3209 memset(kaddr, 0, PAGE_SIZE); 3210 written_size = 0; 3211 } 3212 summary = (struct f2fs_summary *)(kaddr + written_size); 3213 *summary = seg_i->sum_blk->entries[j]; 3214 written_size += SUMMARY_SIZE; 3215 3216 if (written_size + SUMMARY_SIZE <= PAGE_SIZE - 3217 SUM_FOOTER_SIZE) 3218 continue; 3219 3220 set_page_dirty(page); 3221 f2fs_put_page(page, 1); 3222 page = NULL; 3223 } 3224 } 3225 if (page) { 3226 set_page_dirty(page); 3227 f2fs_put_page(page, 1); 3228 } 3229 } 3230 3231 static void write_normal_summaries(struct f2fs_sb_info *sbi, 3232 block_t blkaddr, int type) 3233 { 3234 int i, end; 3235 if (IS_DATASEG(type)) 3236 end = type + NR_CURSEG_DATA_TYPE; 3237 else 3238 end = type + NR_CURSEG_NODE_TYPE; 3239 3240 for (i = type; i < end; i++) 3241 write_current_sum_page(sbi, i, blkaddr + (i - type)); 3242 } 3243 3244 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 3245 { 3246 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) 3247 write_compacted_summaries(sbi, start_blk); 3248 else 3249 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 3250 } 3251 3252 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 3253 { 3254 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 3255 } 3256 3257 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type, 3258 unsigned int val, int alloc) 3259 { 3260 int i; 3261 3262 if (type == NAT_JOURNAL) { 3263 for (i = 0; i < nats_in_cursum(journal); i++) { 3264 if (le32_to_cpu(nid_in_journal(journal, i)) == val) 3265 return i; 3266 } 3267 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL)) 3268 return update_nats_in_cursum(journal, 1); 3269 } else if (type == SIT_JOURNAL) { 3270 for (i = 0; i < sits_in_cursum(journal); i++) 3271 if (le32_to_cpu(segno_in_journal(journal, i)) == val) 3272 return i; 3273 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL)) 3274 return update_sits_in_cursum(journal, 1); 3275 } 3276 return -1; 3277 } 3278 3279 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, 3280 unsigned int segno) 3281 { 3282 return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno)); 3283 } 3284 3285 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, 3286 unsigned int start) 3287 { 3288 struct sit_info *sit_i = SIT_I(sbi); 3289 struct page *page; 3290 pgoff_t src_off, dst_off; 3291 3292 src_off = current_sit_addr(sbi, start); 3293 dst_off = next_sit_addr(sbi, src_off); 3294 3295 page = f2fs_grab_meta_page(sbi, dst_off); 3296 seg_info_to_sit_page(sbi, page, start); 3297 3298 set_page_dirty(page); 3299 set_to_next_sit(sit_i, start); 3300 3301 return page; 3302 } 3303 3304 static struct sit_entry_set *grab_sit_entry_set(void) 3305 { 3306 struct sit_entry_set *ses = 3307 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS); 3308 3309 ses->entry_cnt = 0; 3310 INIT_LIST_HEAD(&ses->set_list); 3311 return ses; 3312 } 3313 3314 static void release_sit_entry_set(struct sit_entry_set *ses) 3315 { 3316 list_del(&ses->set_list); 3317 kmem_cache_free(sit_entry_set_slab, ses); 3318 } 3319 3320 static void adjust_sit_entry_set(struct sit_entry_set *ses, 3321 struct list_head *head) 3322 { 3323 struct sit_entry_set *next = ses; 3324 3325 if (list_is_last(&ses->set_list, head)) 3326 return; 3327 3328 list_for_each_entry_continue(next, head, set_list) 3329 if (ses->entry_cnt <= next->entry_cnt) 3330 break; 3331 3332 list_move_tail(&ses->set_list, &next->set_list); 3333 } 3334 3335 static void add_sit_entry(unsigned int segno, struct list_head *head) 3336 { 3337 struct sit_entry_set *ses; 3338 unsigned int start_segno = START_SEGNO(segno); 3339 3340 list_for_each_entry(ses, head, set_list) { 3341 if (ses->start_segno == start_segno) { 3342 ses->entry_cnt++; 3343 adjust_sit_entry_set(ses, head); 3344 return; 3345 } 3346 } 3347 3348 ses = grab_sit_entry_set(); 3349 3350 ses->start_segno = start_segno; 3351 ses->entry_cnt++; 3352 list_add(&ses->set_list, head); 3353 } 3354 3355 static void add_sits_in_set(struct f2fs_sb_info *sbi) 3356 { 3357 struct f2fs_sm_info *sm_info = SM_I(sbi); 3358 struct list_head *set_list = &sm_info->sit_entry_set; 3359 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; 3360 unsigned int segno; 3361 3362 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) 3363 add_sit_entry(segno, set_list); 3364 } 3365 3366 static void remove_sits_in_journal(struct f2fs_sb_info *sbi) 3367 { 3368 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 3369 struct f2fs_journal *journal = curseg->journal; 3370 int i; 3371 3372 down_write(&curseg->journal_rwsem); 3373 for (i = 0; i < sits_in_cursum(journal); i++) { 3374 unsigned int segno; 3375 bool dirtied; 3376 3377 segno = le32_to_cpu(segno_in_journal(journal, i)); 3378 dirtied = __mark_sit_entry_dirty(sbi, segno); 3379 3380 if (!dirtied) 3381 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); 3382 } 3383 update_sits_in_cursum(journal, -i); 3384 up_write(&curseg->journal_rwsem); 3385 } 3386 3387 /* 3388 * CP calls this function, which flushes SIT entries including sit_journal, 3389 * and moves prefree segs to free segs. 3390 */ 3391 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 3392 { 3393 struct sit_info *sit_i = SIT_I(sbi); 3394 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 3395 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 3396 struct f2fs_journal *journal = curseg->journal; 3397 struct sit_entry_set *ses, *tmp; 3398 struct list_head *head = &SM_I(sbi)->sit_entry_set; 3399 bool to_journal = true; 3400 struct seg_entry *se; 3401 3402 down_write(&sit_i->sentry_lock); 3403 3404 if (!sit_i->dirty_sentries) 3405 goto out; 3406 3407 /* 3408 * add and account sit entries of dirty bitmap in sit entry 3409 * set temporarily 3410 */ 3411 add_sits_in_set(sbi); 3412 3413 /* 3414 * if there are no enough space in journal to store dirty sit 3415 * entries, remove all entries from journal and add and account 3416 * them in sit entry set. 3417 */ 3418 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL)) 3419 remove_sits_in_journal(sbi); 3420 3421 /* 3422 * there are two steps to flush sit entries: 3423 * #1, flush sit entries to journal in current cold data summary block. 3424 * #2, flush sit entries to sit page. 3425 */ 3426 list_for_each_entry_safe(ses, tmp, head, set_list) { 3427 struct page *page = NULL; 3428 struct f2fs_sit_block *raw_sit = NULL; 3429 unsigned int start_segno = ses->start_segno; 3430 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, 3431 (unsigned long)MAIN_SEGS(sbi)); 3432 unsigned int segno = start_segno; 3433 3434 if (to_journal && 3435 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL)) 3436 to_journal = false; 3437 3438 if (to_journal) { 3439 down_write(&curseg->journal_rwsem); 3440 } else { 3441 page = get_next_sit_page(sbi, start_segno); 3442 raw_sit = page_address(page); 3443 } 3444 3445 /* flush dirty sit entries in region of current sit set */ 3446 for_each_set_bit_from(segno, bitmap, end) { 3447 int offset, sit_offset; 3448 3449 se = get_seg_entry(sbi, segno); 3450 #ifdef CONFIG_F2FS_CHECK_FS 3451 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir, 3452 SIT_VBLOCK_MAP_SIZE)) 3453 f2fs_bug_on(sbi, 1); 3454 #endif 3455 3456 /* add discard candidates */ 3457 if (!(cpc->reason & CP_DISCARD)) { 3458 cpc->trim_start = segno; 3459 add_discard_addrs(sbi, cpc, false); 3460 } 3461 3462 if (to_journal) { 3463 offset = f2fs_lookup_journal_in_cursum(journal, 3464 SIT_JOURNAL, segno, 1); 3465 f2fs_bug_on(sbi, offset < 0); 3466 segno_in_journal(journal, offset) = 3467 cpu_to_le32(segno); 3468 seg_info_to_raw_sit(se, 3469 &sit_in_journal(journal, offset)); 3470 check_block_count(sbi, segno, 3471 &sit_in_journal(journal, offset)); 3472 } else { 3473 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 3474 seg_info_to_raw_sit(se, 3475 &raw_sit->entries[sit_offset]); 3476 check_block_count(sbi, segno, 3477 &raw_sit->entries[sit_offset]); 3478 } 3479 3480 __clear_bit(segno, bitmap); 3481 sit_i->dirty_sentries--; 3482 ses->entry_cnt--; 3483 } 3484 3485 if (to_journal) 3486 up_write(&curseg->journal_rwsem); 3487 else 3488 f2fs_put_page(page, 1); 3489 3490 f2fs_bug_on(sbi, ses->entry_cnt); 3491 release_sit_entry_set(ses); 3492 } 3493 3494 f2fs_bug_on(sbi, !list_empty(head)); 3495 f2fs_bug_on(sbi, sit_i->dirty_sentries); 3496 out: 3497 if (cpc->reason & CP_DISCARD) { 3498 __u64 trim_start = cpc->trim_start; 3499 3500 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) 3501 add_discard_addrs(sbi, cpc, false); 3502 3503 cpc->trim_start = trim_start; 3504 } 3505 up_write(&sit_i->sentry_lock); 3506 3507 set_prefree_as_free_segments(sbi); 3508 } 3509 3510 static int build_sit_info(struct f2fs_sb_info *sbi) 3511 { 3512 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 3513 struct sit_info *sit_i; 3514 unsigned int sit_segs, start; 3515 char *src_bitmap; 3516 unsigned int bitmap_size; 3517 3518 /* allocate memory for SIT information */ 3519 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL); 3520 if (!sit_i) 3521 return -ENOMEM; 3522 3523 SM_I(sbi)->sit_info = sit_i; 3524 3525 sit_i->sentries = 3526 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry), 3527 MAIN_SEGS(sbi)), 3528 GFP_KERNEL); 3529 if (!sit_i->sentries) 3530 return -ENOMEM; 3531 3532 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 3533 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size, 3534 GFP_KERNEL); 3535 if (!sit_i->dirty_sentries_bitmap) 3536 return -ENOMEM; 3537 3538 for (start = 0; start < MAIN_SEGS(sbi); start++) { 3539 sit_i->sentries[start].cur_valid_map 3540 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 3541 sit_i->sentries[start].ckpt_valid_map 3542 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 3543 if (!sit_i->sentries[start].cur_valid_map || 3544 !sit_i->sentries[start].ckpt_valid_map) 3545 return -ENOMEM; 3546 3547 #ifdef CONFIG_F2FS_CHECK_FS 3548 sit_i->sentries[start].cur_valid_map_mir 3549 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 3550 if (!sit_i->sentries[start].cur_valid_map_mir) 3551 return -ENOMEM; 3552 #endif 3553 3554 if (f2fs_discard_en(sbi)) { 3555 sit_i->sentries[start].discard_map 3556 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, 3557 GFP_KERNEL); 3558 if (!sit_i->sentries[start].discard_map) 3559 return -ENOMEM; 3560 } 3561 } 3562 3563 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 3564 if (!sit_i->tmp_map) 3565 return -ENOMEM; 3566 3567 if (sbi->segs_per_sec > 1) { 3568 sit_i->sec_entries = 3569 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry), 3570 MAIN_SECS(sbi)), 3571 GFP_KERNEL); 3572 if (!sit_i->sec_entries) 3573 return -ENOMEM; 3574 } 3575 3576 /* get information related with SIT */ 3577 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 3578 3579 /* setup SIT bitmap from ckeckpoint pack */ 3580 bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 3581 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 3582 3583 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); 3584 if (!sit_i->sit_bitmap) 3585 return -ENOMEM; 3586 3587 #ifdef CONFIG_F2FS_CHECK_FS 3588 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); 3589 if (!sit_i->sit_bitmap_mir) 3590 return -ENOMEM; 3591 #endif 3592 3593 /* init SIT information */ 3594 sit_i->s_ops = &default_salloc_ops; 3595 3596 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 3597 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; 3598 sit_i->written_valid_blocks = 0; 3599 sit_i->bitmap_size = bitmap_size; 3600 sit_i->dirty_sentries = 0; 3601 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 3602 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 3603 sit_i->mounted_time = ktime_get_real_seconds(); 3604 init_rwsem(&sit_i->sentry_lock); 3605 return 0; 3606 } 3607 3608 static int build_free_segmap(struct f2fs_sb_info *sbi) 3609 { 3610 struct free_segmap_info *free_i; 3611 unsigned int bitmap_size, sec_bitmap_size; 3612 3613 /* allocate memory for free segmap information */ 3614 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL); 3615 if (!free_i) 3616 return -ENOMEM; 3617 3618 SM_I(sbi)->free_info = free_i; 3619 3620 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 3621 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL); 3622 if (!free_i->free_segmap) 3623 return -ENOMEM; 3624 3625 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 3626 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL); 3627 if (!free_i->free_secmap) 3628 return -ENOMEM; 3629 3630 /* set all segments as dirty temporarily */ 3631 memset(free_i->free_segmap, 0xff, bitmap_size); 3632 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 3633 3634 /* init free segmap information */ 3635 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); 3636 free_i->free_segments = 0; 3637 free_i->free_sections = 0; 3638 spin_lock_init(&free_i->segmap_lock); 3639 return 0; 3640 } 3641 3642 static int build_curseg(struct f2fs_sb_info *sbi) 3643 { 3644 struct curseg_info *array; 3645 int i; 3646 3647 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)), 3648 GFP_KERNEL); 3649 if (!array) 3650 return -ENOMEM; 3651 3652 SM_I(sbi)->curseg_array = array; 3653 3654 for (i = 0; i < NR_CURSEG_TYPE; i++) { 3655 mutex_init(&array[i].curseg_mutex); 3656 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL); 3657 if (!array[i].sum_blk) 3658 return -ENOMEM; 3659 init_rwsem(&array[i].journal_rwsem); 3660 array[i].journal = f2fs_kzalloc(sbi, 3661 sizeof(struct f2fs_journal), GFP_KERNEL); 3662 if (!array[i].journal) 3663 return -ENOMEM; 3664 array[i].segno = NULL_SEGNO; 3665 array[i].next_blkoff = 0; 3666 } 3667 return restore_curseg_summaries(sbi); 3668 } 3669 3670 static int build_sit_entries(struct f2fs_sb_info *sbi) 3671 { 3672 struct sit_info *sit_i = SIT_I(sbi); 3673 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 3674 struct f2fs_journal *journal = curseg->journal; 3675 struct seg_entry *se; 3676 struct f2fs_sit_entry sit; 3677 int sit_blk_cnt = SIT_BLK_CNT(sbi); 3678 unsigned int i, start, end; 3679 unsigned int readed, start_blk = 0; 3680 int err = 0; 3681 block_t total_node_blocks = 0; 3682 3683 do { 3684 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES, 3685 META_SIT, true); 3686 3687 start = start_blk * sit_i->sents_per_block; 3688 end = (start_blk + readed) * sit_i->sents_per_block; 3689 3690 for (; start < end && start < MAIN_SEGS(sbi); start++) { 3691 struct f2fs_sit_block *sit_blk; 3692 struct page *page; 3693 3694 se = &sit_i->sentries[start]; 3695 page = get_current_sit_page(sbi, start); 3696 sit_blk = (struct f2fs_sit_block *)page_address(page); 3697 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 3698 f2fs_put_page(page, 1); 3699 3700 err = check_block_count(sbi, start, &sit); 3701 if (err) 3702 return err; 3703 seg_info_from_raw_sit(se, &sit); 3704 if (IS_NODESEG(se->type)) 3705 total_node_blocks += se->valid_blocks; 3706 3707 /* build discard map only one time */ 3708 if (f2fs_discard_en(sbi)) { 3709 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 3710 memset(se->discard_map, 0xff, 3711 SIT_VBLOCK_MAP_SIZE); 3712 } else { 3713 memcpy(se->discard_map, 3714 se->cur_valid_map, 3715 SIT_VBLOCK_MAP_SIZE); 3716 sbi->discard_blks += 3717 sbi->blocks_per_seg - 3718 se->valid_blocks; 3719 } 3720 } 3721 3722 if (sbi->segs_per_sec > 1) 3723 get_sec_entry(sbi, start)->valid_blocks += 3724 se->valid_blocks; 3725 } 3726 start_blk += readed; 3727 } while (start_blk < sit_blk_cnt); 3728 3729 down_read(&curseg->journal_rwsem); 3730 for (i = 0; i < sits_in_cursum(journal); i++) { 3731 unsigned int old_valid_blocks; 3732 3733 start = le32_to_cpu(segno_in_journal(journal, i)); 3734 if (start >= MAIN_SEGS(sbi)) { 3735 f2fs_msg(sbi->sb, KERN_ERR, 3736 "Wrong journal entry on segno %u", 3737 start); 3738 set_sbi_flag(sbi, SBI_NEED_FSCK); 3739 err = -EINVAL; 3740 break; 3741 } 3742 3743 se = &sit_i->sentries[start]; 3744 sit = sit_in_journal(journal, i); 3745 3746 old_valid_blocks = se->valid_blocks; 3747 if (IS_NODESEG(se->type)) 3748 total_node_blocks -= old_valid_blocks; 3749 3750 err = check_block_count(sbi, start, &sit); 3751 if (err) 3752 break; 3753 seg_info_from_raw_sit(se, &sit); 3754 if (IS_NODESEG(se->type)) 3755 total_node_blocks += se->valid_blocks; 3756 3757 if (f2fs_discard_en(sbi)) { 3758 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 3759 memset(se->discard_map, 0xff, 3760 SIT_VBLOCK_MAP_SIZE); 3761 } else { 3762 memcpy(se->discard_map, se->cur_valid_map, 3763 SIT_VBLOCK_MAP_SIZE); 3764 sbi->discard_blks += old_valid_blocks; 3765 sbi->discard_blks -= se->valid_blocks; 3766 } 3767 } 3768 3769 if (sbi->segs_per_sec > 1) { 3770 get_sec_entry(sbi, start)->valid_blocks += 3771 se->valid_blocks; 3772 get_sec_entry(sbi, start)->valid_blocks -= 3773 old_valid_blocks; 3774 } 3775 } 3776 up_read(&curseg->journal_rwsem); 3777 3778 if (!err && total_node_blocks != valid_node_count(sbi)) { 3779 f2fs_msg(sbi->sb, KERN_ERR, 3780 "SIT is corrupted node# %u vs %u", 3781 total_node_blocks, valid_node_count(sbi)); 3782 set_sbi_flag(sbi, SBI_NEED_FSCK); 3783 err = -EINVAL; 3784 } 3785 3786 return err; 3787 } 3788 3789 static void init_free_segmap(struct f2fs_sb_info *sbi) 3790 { 3791 unsigned int start; 3792 int type; 3793 3794 for (start = 0; start < MAIN_SEGS(sbi); start++) { 3795 struct seg_entry *sentry = get_seg_entry(sbi, start); 3796 if (!sentry->valid_blocks) 3797 __set_free(sbi, start); 3798 else 3799 SIT_I(sbi)->written_valid_blocks += 3800 sentry->valid_blocks; 3801 } 3802 3803 /* set use the current segments */ 3804 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 3805 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 3806 __set_test_and_inuse(sbi, curseg_t->segno); 3807 } 3808 } 3809 3810 static void init_dirty_segmap(struct f2fs_sb_info *sbi) 3811 { 3812 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3813 struct free_segmap_info *free_i = FREE_I(sbi); 3814 unsigned int segno = 0, offset = 0; 3815 unsigned short valid_blocks; 3816 3817 while (1) { 3818 /* find dirty segment based on free segmap */ 3819 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); 3820 if (segno >= MAIN_SEGS(sbi)) 3821 break; 3822 offset = segno + 1; 3823 valid_blocks = get_valid_blocks(sbi, segno, false); 3824 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks) 3825 continue; 3826 if (valid_blocks > sbi->blocks_per_seg) { 3827 f2fs_bug_on(sbi, 1); 3828 continue; 3829 } 3830 mutex_lock(&dirty_i->seglist_lock); 3831 __locate_dirty_segment(sbi, segno, DIRTY); 3832 mutex_unlock(&dirty_i->seglist_lock); 3833 } 3834 } 3835 3836 static int init_victim_secmap(struct f2fs_sb_info *sbi) 3837 { 3838 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3839 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 3840 3841 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 3842 if (!dirty_i->victim_secmap) 3843 return -ENOMEM; 3844 return 0; 3845 } 3846 3847 static int build_dirty_segmap(struct f2fs_sb_info *sbi) 3848 { 3849 struct dirty_seglist_info *dirty_i; 3850 unsigned int bitmap_size, i; 3851 3852 /* allocate memory for dirty segments list information */ 3853 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info), 3854 GFP_KERNEL); 3855 if (!dirty_i) 3856 return -ENOMEM; 3857 3858 SM_I(sbi)->dirty_info = dirty_i; 3859 mutex_init(&dirty_i->seglist_lock); 3860 3861 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 3862 3863 for (i = 0; i < NR_DIRTY_TYPE; i++) { 3864 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size, 3865 GFP_KERNEL); 3866 if (!dirty_i->dirty_segmap[i]) 3867 return -ENOMEM; 3868 } 3869 3870 init_dirty_segmap(sbi); 3871 return init_victim_secmap(sbi); 3872 } 3873 3874 /* 3875 * Update min, max modified time for cost-benefit GC algorithm 3876 */ 3877 static void init_min_max_mtime(struct f2fs_sb_info *sbi) 3878 { 3879 struct sit_info *sit_i = SIT_I(sbi); 3880 unsigned int segno; 3881 3882 down_write(&sit_i->sentry_lock); 3883 3884 sit_i->min_mtime = ULLONG_MAX; 3885 3886 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 3887 unsigned int i; 3888 unsigned long long mtime = 0; 3889 3890 for (i = 0; i < sbi->segs_per_sec; i++) 3891 mtime += get_seg_entry(sbi, segno + i)->mtime; 3892 3893 mtime = div_u64(mtime, sbi->segs_per_sec); 3894 3895 if (sit_i->min_mtime > mtime) 3896 sit_i->min_mtime = mtime; 3897 } 3898 sit_i->max_mtime = get_mtime(sbi, false); 3899 up_write(&sit_i->sentry_lock); 3900 } 3901 3902 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi) 3903 { 3904 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 3905 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3906 struct f2fs_sm_info *sm_info; 3907 int err; 3908 3909 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL); 3910 if (!sm_info) 3911 return -ENOMEM; 3912 3913 /* init sm info */ 3914 sbi->sm_info = sm_info; 3915 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 3916 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 3917 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 3918 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 3919 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 3920 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 3921 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 3922 sm_info->rec_prefree_segments = sm_info->main_segments * 3923 DEF_RECLAIM_PREFREE_SEGMENTS / 100; 3924 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS) 3925 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS; 3926 3927 if (!test_opt(sbi, LFS)) 3928 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC; 3929 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; 3930 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; 3931 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS; 3932 sm_info->min_ssr_sections = reserved_sections(sbi); 3933 3934 INIT_LIST_HEAD(&sm_info->sit_entry_set); 3935 3936 init_rwsem(&sm_info->curseg_lock); 3937 3938 if (!f2fs_readonly(sbi->sb)) { 3939 err = f2fs_create_flush_cmd_control(sbi); 3940 if (err) 3941 return err; 3942 } 3943 3944 err = create_discard_cmd_control(sbi); 3945 if (err) 3946 return err; 3947 3948 err = build_sit_info(sbi); 3949 if (err) 3950 return err; 3951 err = build_free_segmap(sbi); 3952 if (err) 3953 return err; 3954 err = build_curseg(sbi); 3955 if (err) 3956 return err; 3957 3958 /* reinit free segmap based on SIT */ 3959 err = build_sit_entries(sbi); 3960 if (err) 3961 return err; 3962 3963 init_free_segmap(sbi); 3964 err = build_dirty_segmap(sbi); 3965 if (err) 3966 return err; 3967 3968 init_min_max_mtime(sbi); 3969 return 0; 3970 } 3971 3972 static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 3973 enum dirty_type dirty_type) 3974 { 3975 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3976 3977 mutex_lock(&dirty_i->seglist_lock); 3978 kvfree(dirty_i->dirty_segmap[dirty_type]); 3979 dirty_i->nr_dirty[dirty_type] = 0; 3980 mutex_unlock(&dirty_i->seglist_lock); 3981 } 3982 3983 static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 3984 { 3985 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3986 kvfree(dirty_i->victim_secmap); 3987 } 3988 3989 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 3990 { 3991 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 3992 int i; 3993 3994 if (!dirty_i) 3995 return; 3996 3997 /* discard pre-free/dirty segments list */ 3998 for (i = 0; i < NR_DIRTY_TYPE; i++) 3999 discard_dirty_segmap(sbi, i); 4000 4001 destroy_victim_secmap(sbi); 4002 SM_I(sbi)->dirty_info = NULL; 4003 kfree(dirty_i); 4004 } 4005 4006 static void destroy_curseg(struct f2fs_sb_info *sbi) 4007 { 4008 struct curseg_info *array = SM_I(sbi)->curseg_array; 4009 int i; 4010 4011 if (!array) 4012 return; 4013 SM_I(sbi)->curseg_array = NULL; 4014 for (i = 0; i < NR_CURSEG_TYPE; i++) { 4015 kfree(array[i].sum_blk); 4016 kfree(array[i].journal); 4017 } 4018 kfree(array); 4019 } 4020 4021 static void destroy_free_segmap(struct f2fs_sb_info *sbi) 4022 { 4023 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 4024 if (!free_i) 4025 return; 4026 SM_I(sbi)->free_info = NULL; 4027 kvfree(free_i->free_segmap); 4028 kvfree(free_i->free_secmap); 4029 kfree(free_i); 4030 } 4031 4032 static void destroy_sit_info(struct f2fs_sb_info *sbi) 4033 { 4034 struct sit_info *sit_i = SIT_I(sbi); 4035 unsigned int start; 4036 4037 if (!sit_i) 4038 return; 4039 4040 if (sit_i->sentries) { 4041 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4042 kfree(sit_i->sentries[start].cur_valid_map); 4043 #ifdef CONFIG_F2FS_CHECK_FS 4044 kfree(sit_i->sentries[start].cur_valid_map_mir); 4045 #endif 4046 kfree(sit_i->sentries[start].ckpt_valid_map); 4047 kfree(sit_i->sentries[start].discard_map); 4048 } 4049 } 4050 kfree(sit_i->tmp_map); 4051 4052 kvfree(sit_i->sentries); 4053 kvfree(sit_i->sec_entries); 4054 kvfree(sit_i->dirty_sentries_bitmap); 4055 4056 SM_I(sbi)->sit_info = NULL; 4057 kfree(sit_i->sit_bitmap); 4058 #ifdef CONFIG_F2FS_CHECK_FS 4059 kfree(sit_i->sit_bitmap_mir); 4060 #endif 4061 kfree(sit_i); 4062 } 4063 4064 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi) 4065 { 4066 struct f2fs_sm_info *sm_info = SM_I(sbi); 4067 4068 if (!sm_info) 4069 return; 4070 f2fs_destroy_flush_cmd_control(sbi, true); 4071 destroy_discard_cmd_control(sbi); 4072 destroy_dirty_segmap(sbi); 4073 destroy_curseg(sbi); 4074 destroy_free_segmap(sbi); 4075 destroy_sit_info(sbi); 4076 sbi->sm_info = NULL; 4077 kfree(sm_info); 4078 } 4079 4080 int __init f2fs_create_segment_manager_caches(void) 4081 { 4082 discard_entry_slab = f2fs_kmem_cache_create("discard_entry", 4083 sizeof(struct discard_entry)); 4084 if (!discard_entry_slab) 4085 goto fail; 4086 4087 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd", 4088 sizeof(struct discard_cmd)); 4089 if (!discard_cmd_slab) 4090 goto destroy_discard_entry; 4091 4092 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set", 4093 sizeof(struct sit_entry_set)); 4094 if (!sit_entry_set_slab) 4095 goto destroy_discard_cmd; 4096 4097 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry", 4098 sizeof(struct inmem_pages)); 4099 if (!inmem_entry_slab) 4100 goto destroy_sit_entry_set; 4101 return 0; 4102 4103 destroy_sit_entry_set: 4104 kmem_cache_destroy(sit_entry_set_slab); 4105 destroy_discard_cmd: 4106 kmem_cache_destroy(discard_cmd_slab); 4107 destroy_discard_entry: 4108 kmem_cache_destroy(discard_entry_slab); 4109 fail: 4110 return -ENOMEM; 4111 } 4112 4113 void f2fs_destroy_segment_manager_caches(void) 4114 { 4115 kmem_cache_destroy(sit_entry_set_slab); 4116 kmem_cache_destroy(discard_cmd_slab); 4117 kmem_cache_destroy(discard_entry_slab); 4118 kmem_cache_destroy(inmem_entry_slab); 4119 } 4120