1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * fs/f2fs/segment.c 4 * 5 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 6 * http://www.samsung.com/ 7 */ 8 #include <linux/fs.h> 9 #include <linux/f2fs_fs.h> 10 #include <linux/bio.h> 11 #include <linux/blkdev.h> 12 #include <linux/sched/mm.h> 13 #include <linux/prefetch.h> 14 #include <linux/kthread.h> 15 #include <linux/swap.h> 16 #include <linux/timer.h> 17 #include <linux/freezer.h> 18 #include <linux/sched/signal.h> 19 #include <linux/random.h> 20 21 #include "f2fs.h" 22 #include "segment.h" 23 #include "node.h" 24 #include "gc.h" 25 #include "iostat.h" 26 #include <trace/events/f2fs.h> 27 28 #define __reverse_ffz(x) __reverse_ffs(~(x)) 29 30 static struct kmem_cache *discard_entry_slab; 31 static struct kmem_cache *discard_cmd_slab; 32 static struct kmem_cache *sit_entry_set_slab; 33 static struct kmem_cache *revoke_entry_slab; 34 35 static unsigned long __reverse_ulong(unsigned char *str) 36 { 37 unsigned long tmp = 0; 38 int shift = 24, idx = 0; 39 40 #if BITS_PER_LONG == 64 41 shift = 56; 42 #endif 43 while (shift >= 0) { 44 tmp |= (unsigned long)str[idx++] << shift; 45 shift -= BITS_PER_BYTE; 46 } 47 return tmp; 48 } 49 50 /* 51 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since 52 * MSB and LSB are reversed in a byte by f2fs_set_bit. 53 */ 54 static inline unsigned long __reverse_ffs(unsigned long word) 55 { 56 int num = 0; 57 58 #if BITS_PER_LONG == 64 59 if ((word & 0xffffffff00000000UL) == 0) 60 num += 32; 61 else 62 word >>= 32; 63 #endif 64 if ((word & 0xffff0000) == 0) 65 num += 16; 66 else 67 word >>= 16; 68 69 if ((word & 0xff00) == 0) 70 num += 8; 71 else 72 word >>= 8; 73 74 if ((word & 0xf0) == 0) 75 num += 4; 76 else 77 word >>= 4; 78 79 if ((word & 0xc) == 0) 80 num += 2; 81 else 82 word >>= 2; 83 84 if ((word & 0x2) == 0) 85 num += 1; 86 return num; 87 } 88 89 /* 90 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because 91 * f2fs_set_bit makes MSB and LSB reversed in a byte. 92 * @size must be integral times of unsigned long. 93 * Example: 94 * MSB <--> LSB 95 * f2fs_set_bit(0, bitmap) => 1000 0000 96 * f2fs_set_bit(7, bitmap) => 0000 0001 97 */ 98 static unsigned long __find_rev_next_bit(const unsigned long *addr, 99 unsigned long size, unsigned long offset) 100 { 101 const unsigned long *p = addr + BIT_WORD(offset); 102 unsigned long result = size; 103 unsigned long tmp; 104 105 if (offset >= size) 106 return size; 107 108 size -= (offset & ~(BITS_PER_LONG - 1)); 109 offset %= BITS_PER_LONG; 110 111 while (1) { 112 if (*p == 0) 113 goto pass; 114 115 tmp = __reverse_ulong((unsigned char *)p); 116 117 tmp &= ~0UL >> offset; 118 if (size < BITS_PER_LONG) 119 tmp &= (~0UL << (BITS_PER_LONG - size)); 120 if (tmp) 121 goto found; 122 pass: 123 if (size <= BITS_PER_LONG) 124 break; 125 size -= BITS_PER_LONG; 126 offset = 0; 127 p++; 128 } 129 return result; 130 found: 131 return result - size + __reverse_ffs(tmp); 132 } 133 134 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr, 135 unsigned long size, unsigned long offset) 136 { 137 const unsigned long *p = addr + BIT_WORD(offset); 138 unsigned long result = size; 139 unsigned long tmp; 140 141 if (offset >= size) 142 return size; 143 144 size -= (offset & ~(BITS_PER_LONG - 1)); 145 offset %= BITS_PER_LONG; 146 147 while (1) { 148 if (*p == ~0UL) 149 goto pass; 150 151 tmp = __reverse_ulong((unsigned char *)p); 152 153 if (offset) 154 tmp |= ~0UL << (BITS_PER_LONG - offset); 155 if (size < BITS_PER_LONG) 156 tmp |= ~0UL >> size; 157 if (tmp != ~0UL) 158 goto found; 159 pass: 160 if (size <= BITS_PER_LONG) 161 break; 162 size -= BITS_PER_LONG; 163 offset = 0; 164 p++; 165 } 166 return result; 167 found: 168 return result - size + __reverse_ffz(tmp); 169 } 170 171 bool f2fs_need_SSR(struct f2fs_sb_info *sbi) 172 { 173 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES); 174 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS); 175 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA); 176 177 if (f2fs_lfs_mode(sbi)) 178 return false; 179 if (sbi->gc_mode == GC_URGENT_HIGH) 180 return true; 181 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 182 return true; 183 184 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs + 185 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi)); 186 } 187 188 void f2fs_abort_atomic_write(struct inode *inode, bool clean) 189 { 190 struct f2fs_inode_info *fi = F2FS_I(inode); 191 192 if (!f2fs_is_atomic_file(inode)) 193 return; 194 195 release_atomic_write_cnt(inode); 196 clear_inode_flag(inode, FI_ATOMIC_COMMITTED); 197 clear_inode_flag(inode, FI_ATOMIC_REPLACE); 198 clear_inode_flag(inode, FI_ATOMIC_FILE); 199 stat_dec_atomic_inode(inode); 200 201 F2FS_I(inode)->atomic_write_task = NULL; 202 203 if (clean) { 204 truncate_inode_pages_final(inode->i_mapping); 205 f2fs_i_size_write(inode, fi->original_i_size); 206 fi->original_i_size = 0; 207 } 208 /* avoid stale dirty inode during eviction */ 209 sync_inode_metadata(inode, 0); 210 } 211 212 static int __replace_atomic_write_block(struct inode *inode, pgoff_t index, 213 block_t new_addr, block_t *old_addr, bool recover) 214 { 215 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 216 struct dnode_of_data dn; 217 struct node_info ni; 218 int err; 219 220 retry: 221 set_new_dnode(&dn, inode, NULL, NULL, 0); 222 err = f2fs_get_dnode_of_data(&dn, index, ALLOC_NODE); 223 if (err) { 224 if (err == -ENOMEM) { 225 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 226 goto retry; 227 } 228 return err; 229 } 230 231 err = f2fs_get_node_info(sbi, dn.nid, &ni, false); 232 if (err) { 233 f2fs_put_dnode(&dn); 234 return err; 235 } 236 237 if (recover) { 238 /* dn.data_blkaddr is always valid */ 239 if (!__is_valid_data_blkaddr(new_addr)) { 240 if (new_addr == NULL_ADDR) 241 dec_valid_block_count(sbi, inode, 1); 242 f2fs_invalidate_blocks(sbi, dn.data_blkaddr); 243 f2fs_update_data_blkaddr(&dn, new_addr); 244 } else { 245 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, 246 new_addr, ni.version, true, true); 247 } 248 } else { 249 blkcnt_t count = 1; 250 251 err = inc_valid_block_count(sbi, inode, &count); 252 if (err) { 253 f2fs_put_dnode(&dn); 254 return err; 255 } 256 257 *old_addr = dn.data_blkaddr; 258 f2fs_truncate_data_blocks_range(&dn, 1); 259 dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count); 260 261 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr, 262 ni.version, true, false); 263 } 264 265 f2fs_put_dnode(&dn); 266 267 trace_f2fs_replace_atomic_write_block(inode, F2FS_I(inode)->cow_inode, 268 index, old_addr ? *old_addr : 0, new_addr, recover); 269 return 0; 270 } 271 272 static void __complete_revoke_list(struct inode *inode, struct list_head *head, 273 bool revoke) 274 { 275 struct revoke_entry *cur, *tmp; 276 pgoff_t start_index = 0; 277 bool truncate = is_inode_flag_set(inode, FI_ATOMIC_REPLACE); 278 279 list_for_each_entry_safe(cur, tmp, head, list) { 280 if (revoke) { 281 __replace_atomic_write_block(inode, cur->index, 282 cur->old_addr, NULL, true); 283 } else if (truncate) { 284 f2fs_truncate_hole(inode, start_index, cur->index); 285 start_index = cur->index + 1; 286 } 287 288 list_del(&cur->list); 289 kmem_cache_free(revoke_entry_slab, cur); 290 } 291 292 if (!revoke && truncate) 293 f2fs_do_truncate_blocks(inode, start_index * PAGE_SIZE, false); 294 } 295 296 static int __f2fs_commit_atomic_write(struct inode *inode) 297 { 298 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 299 struct f2fs_inode_info *fi = F2FS_I(inode); 300 struct inode *cow_inode = fi->cow_inode; 301 struct revoke_entry *new; 302 struct list_head revoke_list; 303 block_t blkaddr; 304 struct dnode_of_data dn; 305 pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); 306 pgoff_t off = 0, blen, index; 307 int ret = 0, i; 308 309 INIT_LIST_HEAD(&revoke_list); 310 311 while (len) { 312 blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len); 313 314 set_new_dnode(&dn, cow_inode, NULL, NULL, 0); 315 ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA); 316 if (ret && ret != -ENOENT) { 317 goto out; 318 } else if (ret == -ENOENT) { 319 ret = 0; 320 if (dn.max_level == 0) 321 goto out; 322 goto next; 323 } 324 325 blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_page, cow_inode), 326 len); 327 index = off; 328 for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) { 329 blkaddr = f2fs_data_blkaddr(&dn); 330 331 if (!__is_valid_data_blkaddr(blkaddr)) { 332 continue; 333 } else if (!f2fs_is_valid_blkaddr(sbi, blkaddr, 334 DATA_GENERIC_ENHANCE)) { 335 f2fs_put_dnode(&dn); 336 ret = -EFSCORRUPTED; 337 f2fs_handle_error(sbi, 338 ERROR_INVALID_BLKADDR); 339 goto out; 340 } 341 342 new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS, 343 true, NULL); 344 345 ret = __replace_atomic_write_block(inode, index, blkaddr, 346 &new->old_addr, false); 347 if (ret) { 348 f2fs_put_dnode(&dn); 349 kmem_cache_free(revoke_entry_slab, new); 350 goto out; 351 } 352 353 f2fs_update_data_blkaddr(&dn, NULL_ADDR); 354 new->index = index; 355 list_add_tail(&new->list, &revoke_list); 356 } 357 f2fs_put_dnode(&dn); 358 next: 359 off += blen; 360 len -= blen; 361 } 362 363 out: 364 if (ret) { 365 sbi->revoked_atomic_block += fi->atomic_write_cnt; 366 } else { 367 sbi->committed_atomic_block += fi->atomic_write_cnt; 368 set_inode_flag(inode, FI_ATOMIC_COMMITTED); 369 } 370 371 __complete_revoke_list(inode, &revoke_list, ret ? true : false); 372 373 return ret; 374 } 375 376 int f2fs_commit_atomic_write(struct inode *inode) 377 { 378 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 379 struct f2fs_inode_info *fi = F2FS_I(inode); 380 int err; 381 382 err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX); 383 if (err) 384 return err; 385 386 f2fs_down_write(&fi->i_gc_rwsem[WRITE]); 387 f2fs_lock_op(sbi); 388 389 err = __f2fs_commit_atomic_write(inode); 390 391 f2fs_unlock_op(sbi); 392 f2fs_up_write(&fi->i_gc_rwsem[WRITE]); 393 394 return err; 395 } 396 397 /* 398 * This function balances dirty node and dentry pages. 399 * In addition, it controls garbage collection. 400 */ 401 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need) 402 { 403 if (time_to_inject(sbi, FAULT_CHECKPOINT)) 404 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT); 405 406 /* balance_fs_bg is able to be pending */ 407 if (need && excess_cached_nats(sbi)) 408 f2fs_balance_fs_bg(sbi, false); 409 410 if (!f2fs_is_checkpoint_ready(sbi)) 411 return; 412 413 /* 414 * We should do GC or end up with checkpoint, if there are so many dirty 415 * dir/node pages without enough free segments. 416 */ 417 if (has_enough_free_secs(sbi, 0, 0)) 418 return; 419 420 if (test_opt(sbi, GC_MERGE) && sbi->gc_thread && 421 sbi->gc_thread->f2fs_gc_task) { 422 DEFINE_WAIT(wait); 423 424 prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait, 425 TASK_UNINTERRUPTIBLE); 426 wake_up(&sbi->gc_thread->gc_wait_queue_head); 427 io_schedule(); 428 finish_wait(&sbi->gc_thread->fggc_wq, &wait); 429 } else { 430 struct f2fs_gc_control gc_control = { 431 .victim_segno = NULL_SEGNO, 432 .init_gc_type = BG_GC, 433 .no_bg_gc = true, 434 .should_migrate_blocks = false, 435 .err_gc_skipped = false, 436 .nr_free_secs = 1 }; 437 f2fs_down_write(&sbi->gc_lock); 438 f2fs_gc(sbi, &gc_control); 439 } 440 } 441 442 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi) 443 { 444 int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2; 445 unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS); 446 unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA); 447 unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES); 448 unsigned int meta = get_pages(sbi, F2FS_DIRTY_META); 449 unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA); 450 unsigned int threshold = sbi->blocks_per_seg * factor * 451 DEFAULT_DIRTY_THRESHOLD; 452 unsigned int global_threshold = threshold * 3 / 2; 453 454 if (dents >= threshold || qdata >= threshold || 455 nodes >= threshold || meta >= threshold || 456 imeta >= threshold) 457 return true; 458 return dents + qdata + nodes + meta + imeta > global_threshold; 459 } 460 461 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg) 462 { 463 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 464 return; 465 466 /* try to shrink extent cache when there is no enough memory */ 467 if (!f2fs_available_free_memory(sbi, READ_EXTENT_CACHE)) 468 f2fs_shrink_read_extent_tree(sbi, 469 READ_EXTENT_CACHE_SHRINK_NUMBER); 470 471 /* try to shrink age extent cache when there is no enough memory */ 472 if (!f2fs_available_free_memory(sbi, AGE_EXTENT_CACHE)) 473 f2fs_shrink_age_extent_tree(sbi, 474 AGE_EXTENT_CACHE_SHRINK_NUMBER); 475 476 /* check the # of cached NAT entries */ 477 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES)) 478 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK); 479 480 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) 481 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS); 482 else 483 f2fs_build_free_nids(sbi, false, false); 484 485 if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) || 486 excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi)) 487 goto do_sync; 488 489 /* there is background inflight IO or foreground operation recently */ 490 if (is_inflight_io(sbi, REQ_TIME) || 491 (!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem))) 492 return; 493 494 /* exceed periodical checkpoint timeout threshold */ 495 if (f2fs_time_over(sbi, CP_TIME)) 496 goto do_sync; 497 498 /* checkpoint is the only way to shrink partial cached entries */ 499 if (f2fs_available_free_memory(sbi, NAT_ENTRIES) && 500 f2fs_available_free_memory(sbi, INO_ENTRIES)) 501 return; 502 503 do_sync: 504 if (test_opt(sbi, DATA_FLUSH) && from_bg) { 505 struct blk_plug plug; 506 507 mutex_lock(&sbi->flush_lock); 508 509 blk_start_plug(&plug); 510 f2fs_sync_dirty_inodes(sbi, FILE_INODE, false); 511 blk_finish_plug(&plug); 512 513 mutex_unlock(&sbi->flush_lock); 514 } 515 f2fs_sync_fs(sbi->sb, 1); 516 stat_inc_bg_cp_count(sbi->stat_info); 517 } 518 519 static int __submit_flush_wait(struct f2fs_sb_info *sbi, 520 struct block_device *bdev) 521 { 522 int ret = blkdev_issue_flush(bdev); 523 524 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER), 525 test_opt(sbi, FLUSH_MERGE), ret); 526 if (!ret) 527 f2fs_update_iostat(sbi, NULL, FS_FLUSH_IO, 0); 528 return ret; 529 } 530 531 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino) 532 { 533 int ret = 0; 534 int i; 535 536 if (!f2fs_is_multi_device(sbi)) 537 return __submit_flush_wait(sbi, sbi->sb->s_bdev); 538 539 for (i = 0; i < sbi->s_ndevs; i++) { 540 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO)) 541 continue; 542 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 543 if (ret) 544 break; 545 } 546 return ret; 547 } 548 549 static int issue_flush_thread(void *data) 550 { 551 struct f2fs_sb_info *sbi = data; 552 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 553 wait_queue_head_t *q = &fcc->flush_wait_queue; 554 repeat: 555 if (kthread_should_stop()) 556 return 0; 557 558 if (!llist_empty(&fcc->issue_list)) { 559 struct flush_cmd *cmd, *next; 560 int ret; 561 562 fcc->dispatch_list = llist_del_all(&fcc->issue_list); 563 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); 564 565 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode); 566 567 ret = submit_flush_wait(sbi, cmd->ino); 568 atomic_inc(&fcc->issued_flush); 569 570 llist_for_each_entry_safe(cmd, next, 571 fcc->dispatch_list, llnode) { 572 cmd->ret = ret; 573 complete(&cmd->wait); 574 } 575 fcc->dispatch_list = NULL; 576 } 577 578 wait_event_interruptible(*q, 579 kthread_should_stop() || !llist_empty(&fcc->issue_list)); 580 goto repeat; 581 } 582 583 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino) 584 { 585 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 586 struct flush_cmd cmd; 587 int ret; 588 589 if (test_opt(sbi, NOBARRIER)) 590 return 0; 591 592 if (!test_opt(sbi, FLUSH_MERGE)) { 593 atomic_inc(&fcc->queued_flush); 594 ret = submit_flush_wait(sbi, ino); 595 atomic_dec(&fcc->queued_flush); 596 atomic_inc(&fcc->issued_flush); 597 return ret; 598 } 599 600 if (atomic_inc_return(&fcc->queued_flush) == 1 || 601 f2fs_is_multi_device(sbi)) { 602 ret = submit_flush_wait(sbi, ino); 603 atomic_dec(&fcc->queued_flush); 604 605 atomic_inc(&fcc->issued_flush); 606 return ret; 607 } 608 609 cmd.ino = ino; 610 init_completion(&cmd.wait); 611 612 llist_add(&cmd.llnode, &fcc->issue_list); 613 614 /* 615 * update issue_list before we wake up issue_flush thread, this 616 * smp_mb() pairs with another barrier in ___wait_event(), see 617 * more details in comments of waitqueue_active(). 618 */ 619 smp_mb(); 620 621 if (waitqueue_active(&fcc->flush_wait_queue)) 622 wake_up(&fcc->flush_wait_queue); 623 624 if (fcc->f2fs_issue_flush) { 625 wait_for_completion(&cmd.wait); 626 atomic_dec(&fcc->queued_flush); 627 } else { 628 struct llist_node *list; 629 630 list = llist_del_all(&fcc->issue_list); 631 if (!list) { 632 wait_for_completion(&cmd.wait); 633 atomic_dec(&fcc->queued_flush); 634 } else { 635 struct flush_cmd *tmp, *next; 636 637 ret = submit_flush_wait(sbi, ino); 638 639 llist_for_each_entry_safe(tmp, next, list, llnode) { 640 if (tmp == &cmd) { 641 cmd.ret = ret; 642 atomic_dec(&fcc->queued_flush); 643 continue; 644 } 645 tmp->ret = ret; 646 complete(&tmp->wait); 647 } 648 } 649 } 650 651 return cmd.ret; 652 } 653 654 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi) 655 { 656 dev_t dev = sbi->sb->s_bdev->bd_dev; 657 struct flush_cmd_control *fcc; 658 659 if (SM_I(sbi)->fcc_info) { 660 fcc = SM_I(sbi)->fcc_info; 661 if (fcc->f2fs_issue_flush) 662 return 0; 663 goto init_thread; 664 } 665 666 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL); 667 if (!fcc) 668 return -ENOMEM; 669 atomic_set(&fcc->issued_flush, 0); 670 atomic_set(&fcc->queued_flush, 0); 671 init_waitqueue_head(&fcc->flush_wait_queue); 672 init_llist_head(&fcc->issue_list); 673 SM_I(sbi)->fcc_info = fcc; 674 if (!test_opt(sbi, FLUSH_MERGE)) 675 return 0; 676 677 init_thread: 678 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, 679 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); 680 if (IS_ERR(fcc->f2fs_issue_flush)) { 681 int err = PTR_ERR(fcc->f2fs_issue_flush); 682 683 fcc->f2fs_issue_flush = NULL; 684 return err; 685 } 686 687 return 0; 688 } 689 690 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free) 691 { 692 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 693 694 if (fcc && fcc->f2fs_issue_flush) { 695 struct task_struct *flush_thread = fcc->f2fs_issue_flush; 696 697 fcc->f2fs_issue_flush = NULL; 698 kthread_stop(flush_thread); 699 } 700 if (free) { 701 kfree(fcc); 702 SM_I(sbi)->fcc_info = NULL; 703 } 704 } 705 706 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi) 707 { 708 int ret = 0, i; 709 710 if (!f2fs_is_multi_device(sbi)) 711 return 0; 712 713 if (test_opt(sbi, NOBARRIER)) 714 return 0; 715 716 for (i = 1; i < sbi->s_ndevs; i++) { 717 int count = DEFAULT_RETRY_IO_COUNT; 718 719 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device)) 720 continue; 721 722 do { 723 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 724 if (ret) 725 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 726 } while (ret && --count); 727 728 if (ret) { 729 f2fs_stop_checkpoint(sbi, false, 730 STOP_CP_REASON_FLUSH_FAIL); 731 break; 732 } 733 734 spin_lock(&sbi->dev_lock); 735 f2fs_clear_bit(i, (char *)&sbi->dirty_device); 736 spin_unlock(&sbi->dev_lock); 737 } 738 739 return ret; 740 } 741 742 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 743 enum dirty_type dirty_type) 744 { 745 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 746 747 /* need not be added */ 748 if (IS_CURSEG(sbi, segno)) 749 return; 750 751 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 752 dirty_i->nr_dirty[dirty_type]++; 753 754 if (dirty_type == DIRTY) { 755 struct seg_entry *sentry = get_seg_entry(sbi, segno); 756 enum dirty_type t = sentry->type; 757 758 if (unlikely(t >= DIRTY)) { 759 f2fs_bug_on(sbi, 1); 760 return; 761 } 762 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) 763 dirty_i->nr_dirty[t]++; 764 765 if (__is_large_section(sbi)) { 766 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 767 block_t valid_blocks = 768 get_valid_blocks(sbi, segno, true); 769 770 f2fs_bug_on(sbi, unlikely(!valid_blocks || 771 valid_blocks == CAP_BLKS_PER_SEC(sbi))); 772 773 if (!IS_CURSEC(sbi, secno)) 774 set_bit(secno, dirty_i->dirty_secmap); 775 } 776 } 777 } 778 779 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 780 enum dirty_type dirty_type) 781 { 782 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 783 block_t valid_blocks; 784 785 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) 786 dirty_i->nr_dirty[dirty_type]--; 787 788 if (dirty_type == DIRTY) { 789 struct seg_entry *sentry = get_seg_entry(sbi, segno); 790 enum dirty_type t = sentry->type; 791 792 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) 793 dirty_i->nr_dirty[t]--; 794 795 valid_blocks = get_valid_blocks(sbi, segno, true); 796 if (valid_blocks == 0) { 797 clear_bit(GET_SEC_FROM_SEG(sbi, segno), 798 dirty_i->victim_secmap); 799 #ifdef CONFIG_F2FS_CHECK_FS 800 clear_bit(segno, SIT_I(sbi)->invalid_segmap); 801 #endif 802 } 803 if (__is_large_section(sbi)) { 804 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 805 806 if (!valid_blocks || 807 valid_blocks == CAP_BLKS_PER_SEC(sbi)) { 808 clear_bit(secno, dirty_i->dirty_secmap); 809 return; 810 } 811 812 if (!IS_CURSEC(sbi, secno)) 813 set_bit(secno, dirty_i->dirty_secmap); 814 } 815 } 816 } 817 818 /* 819 * Should not occur error such as -ENOMEM. 820 * Adding dirty entry into seglist is not critical operation. 821 * If a given segment is one of current working segments, it won't be added. 822 */ 823 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) 824 { 825 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 826 unsigned short valid_blocks, ckpt_valid_blocks; 827 unsigned int usable_blocks; 828 829 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) 830 return; 831 832 usable_blocks = f2fs_usable_blks_in_seg(sbi, segno); 833 mutex_lock(&dirty_i->seglist_lock); 834 835 valid_blocks = get_valid_blocks(sbi, segno, false); 836 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false); 837 838 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) || 839 ckpt_valid_blocks == usable_blocks)) { 840 __locate_dirty_segment(sbi, segno, PRE); 841 __remove_dirty_segment(sbi, segno, DIRTY); 842 } else if (valid_blocks < usable_blocks) { 843 __locate_dirty_segment(sbi, segno, DIRTY); 844 } else { 845 /* Recovery routine with SSR needs this */ 846 __remove_dirty_segment(sbi, segno, DIRTY); 847 } 848 849 mutex_unlock(&dirty_i->seglist_lock); 850 } 851 852 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */ 853 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi) 854 { 855 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 856 unsigned int segno; 857 858 mutex_lock(&dirty_i->seglist_lock); 859 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 860 if (get_valid_blocks(sbi, segno, false)) 861 continue; 862 if (IS_CURSEG(sbi, segno)) 863 continue; 864 __locate_dirty_segment(sbi, segno, PRE); 865 __remove_dirty_segment(sbi, segno, DIRTY); 866 } 867 mutex_unlock(&dirty_i->seglist_lock); 868 } 869 870 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi) 871 { 872 int ovp_hole_segs = 873 (overprovision_segments(sbi) - reserved_segments(sbi)); 874 block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg; 875 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 876 block_t holes[2] = {0, 0}; /* DATA and NODE */ 877 block_t unusable; 878 struct seg_entry *se; 879 unsigned int segno; 880 881 mutex_lock(&dirty_i->seglist_lock); 882 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 883 se = get_seg_entry(sbi, segno); 884 if (IS_NODESEG(se->type)) 885 holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) - 886 se->valid_blocks; 887 else 888 holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) - 889 se->valid_blocks; 890 } 891 mutex_unlock(&dirty_i->seglist_lock); 892 893 unusable = max(holes[DATA], holes[NODE]); 894 if (unusable > ovp_holes) 895 return unusable - ovp_holes; 896 return 0; 897 } 898 899 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable) 900 { 901 int ovp_hole_segs = 902 (overprovision_segments(sbi) - reserved_segments(sbi)); 903 if (unusable > F2FS_OPTION(sbi).unusable_cap) 904 return -EAGAIN; 905 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) && 906 dirty_segments(sbi) > ovp_hole_segs) 907 return -EAGAIN; 908 return 0; 909 } 910 911 /* This is only used by SBI_CP_DISABLED */ 912 static unsigned int get_free_segment(struct f2fs_sb_info *sbi) 913 { 914 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 915 unsigned int segno = 0; 916 917 mutex_lock(&dirty_i->seglist_lock); 918 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 919 if (get_valid_blocks(sbi, segno, false)) 920 continue; 921 if (get_ckpt_valid_blocks(sbi, segno, false)) 922 continue; 923 mutex_unlock(&dirty_i->seglist_lock); 924 return segno; 925 } 926 mutex_unlock(&dirty_i->seglist_lock); 927 return NULL_SEGNO; 928 } 929 930 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi, 931 struct block_device *bdev, block_t lstart, 932 block_t start, block_t len) 933 { 934 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 935 struct list_head *pend_list; 936 struct discard_cmd *dc; 937 938 f2fs_bug_on(sbi, !len); 939 940 pend_list = &dcc->pend_list[plist_idx(len)]; 941 942 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL); 943 INIT_LIST_HEAD(&dc->list); 944 dc->bdev = bdev; 945 dc->di.lstart = lstart; 946 dc->di.start = start; 947 dc->di.len = len; 948 dc->ref = 0; 949 dc->state = D_PREP; 950 dc->queued = 0; 951 dc->error = 0; 952 init_completion(&dc->wait); 953 list_add_tail(&dc->list, pend_list); 954 spin_lock_init(&dc->lock); 955 dc->bio_ref = 0; 956 atomic_inc(&dcc->discard_cmd_cnt); 957 dcc->undiscard_blks += len; 958 959 return dc; 960 } 961 962 static bool f2fs_check_discard_tree(struct f2fs_sb_info *sbi) 963 { 964 #ifdef CONFIG_F2FS_CHECK_FS 965 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 966 struct rb_node *cur = rb_first_cached(&dcc->root), *next; 967 struct discard_cmd *cur_dc, *next_dc; 968 969 while (cur) { 970 next = rb_next(cur); 971 if (!next) 972 return true; 973 974 cur_dc = rb_entry(cur, struct discard_cmd, rb_node); 975 next_dc = rb_entry(next, struct discard_cmd, rb_node); 976 977 if (cur_dc->di.lstart + cur_dc->di.len > next_dc->di.lstart) { 978 f2fs_info(sbi, "broken discard_rbtree, " 979 "cur(%u, %u) next(%u, %u)", 980 cur_dc->di.lstart, cur_dc->di.len, 981 next_dc->di.lstart, next_dc->di.len); 982 return false; 983 } 984 cur = next; 985 } 986 #endif 987 return true; 988 } 989 990 static struct discard_cmd *__lookup_discard_cmd(struct f2fs_sb_info *sbi, 991 block_t blkaddr) 992 { 993 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 994 struct rb_node *node = dcc->root.rb_root.rb_node; 995 struct discard_cmd *dc; 996 997 while (node) { 998 dc = rb_entry(node, struct discard_cmd, rb_node); 999 1000 if (blkaddr < dc->di.lstart) 1001 node = node->rb_left; 1002 else if (blkaddr >= dc->di.lstart + dc->di.len) 1003 node = node->rb_right; 1004 else 1005 return dc; 1006 } 1007 return NULL; 1008 } 1009 1010 static struct discard_cmd *__lookup_discard_cmd_ret(struct rb_root_cached *root, 1011 block_t blkaddr, 1012 struct discard_cmd **prev_entry, 1013 struct discard_cmd **next_entry, 1014 struct rb_node ***insert_p, 1015 struct rb_node **insert_parent) 1016 { 1017 struct rb_node **pnode = &root->rb_root.rb_node; 1018 struct rb_node *parent = NULL, *tmp_node; 1019 struct discard_cmd *dc; 1020 1021 *insert_p = NULL; 1022 *insert_parent = NULL; 1023 *prev_entry = NULL; 1024 *next_entry = NULL; 1025 1026 if (RB_EMPTY_ROOT(&root->rb_root)) 1027 return NULL; 1028 1029 while (*pnode) { 1030 parent = *pnode; 1031 dc = rb_entry(*pnode, struct discard_cmd, rb_node); 1032 1033 if (blkaddr < dc->di.lstart) 1034 pnode = &(*pnode)->rb_left; 1035 else if (blkaddr >= dc->di.lstart + dc->di.len) 1036 pnode = &(*pnode)->rb_right; 1037 else 1038 goto lookup_neighbors; 1039 } 1040 1041 *insert_p = pnode; 1042 *insert_parent = parent; 1043 1044 dc = rb_entry(parent, struct discard_cmd, rb_node); 1045 tmp_node = parent; 1046 if (parent && blkaddr > dc->di.lstart) 1047 tmp_node = rb_next(parent); 1048 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1049 1050 tmp_node = parent; 1051 if (parent && blkaddr < dc->di.lstart) 1052 tmp_node = rb_prev(parent); 1053 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1054 return NULL; 1055 1056 lookup_neighbors: 1057 /* lookup prev node for merging backward later */ 1058 tmp_node = rb_prev(&dc->rb_node); 1059 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1060 1061 /* lookup next node for merging frontward later */ 1062 tmp_node = rb_next(&dc->rb_node); 1063 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1064 return dc; 1065 } 1066 1067 static void __detach_discard_cmd(struct discard_cmd_control *dcc, 1068 struct discard_cmd *dc) 1069 { 1070 if (dc->state == D_DONE) 1071 atomic_sub(dc->queued, &dcc->queued_discard); 1072 1073 list_del(&dc->list); 1074 rb_erase_cached(&dc->rb_node, &dcc->root); 1075 dcc->undiscard_blks -= dc->di.len; 1076 1077 kmem_cache_free(discard_cmd_slab, dc); 1078 1079 atomic_dec(&dcc->discard_cmd_cnt); 1080 } 1081 1082 static void __remove_discard_cmd(struct f2fs_sb_info *sbi, 1083 struct discard_cmd *dc) 1084 { 1085 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1086 unsigned long flags; 1087 1088 trace_f2fs_remove_discard(dc->bdev, dc->di.start, dc->di.len); 1089 1090 spin_lock_irqsave(&dc->lock, flags); 1091 if (dc->bio_ref) { 1092 spin_unlock_irqrestore(&dc->lock, flags); 1093 return; 1094 } 1095 spin_unlock_irqrestore(&dc->lock, flags); 1096 1097 f2fs_bug_on(sbi, dc->ref); 1098 1099 if (dc->error == -EOPNOTSUPP) 1100 dc->error = 0; 1101 1102 if (dc->error) 1103 printk_ratelimited( 1104 "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d", 1105 KERN_INFO, sbi->sb->s_id, 1106 dc->di.lstart, dc->di.start, dc->di.len, dc->error); 1107 __detach_discard_cmd(dcc, dc); 1108 } 1109 1110 static void f2fs_submit_discard_endio(struct bio *bio) 1111 { 1112 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private; 1113 unsigned long flags; 1114 1115 spin_lock_irqsave(&dc->lock, flags); 1116 if (!dc->error) 1117 dc->error = blk_status_to_errno(bio->bi_status); 1118 dc->bio_ref--; 1119 if (!dc->bio_ref && dc->state == D_SUBMIT) { 1120 dc->state = D_DONE; 1121 complete_all(&dc->wait); 1122 } 1123 spin_unlock_irqrestore(&dc->lock, flags); 1124 bio_put(bio); 1125 } 1126 1127 static void __check_sit_bitmap(struct f2fs_sb_info *sbi, 1128 block_t start, block_t end) 1129 { 1130 #ifdef CONFIG_F2FS_CHECK_FS 1131 struct seg_entry *sentry; 1132 unsigned int segno; 1133 block_t blk = start; 1134 unsigned long offset, size, max_blocks = sbi->blocks_per_seg; 1135 unsigned long *map; 1136 1137 while (blk < end) { 1138 segno = GET_SEGNO(sbi, blk); 1139 sentry = get_seg_entry(sbi, segno); 1140 offset = GET_BLKOFF_FROM_SEG0(sbi, blk); 1141 1142 if (end < START_BLOCK(sbi, segno + 1)) 1143 size = GET_BLKOFF_FROM_SEG0(sbi, end); 1144 else 1145 size = max_blocks; 1146 map = (unsigned long *)(sentry->cur_valid_map); 1147 offset = __find_rev_next_bit(map, size, offset); 1148 f2fs_bug_on(sbi, offset != size); 1149 blk = START_BLOCK(sbi, segno + 1); 1150 } 1151 #endif 1152 } 1153 1154 static void __init_discard_policy(struct f2fs_sb_info *sbi, 1155 struct discard_policy *dpolicy, 1156 int discard_type, unsigned int granularity) 1157 { 1158 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1159 1160 /* common policy */ 1161 dpolicy->type = discard_type; 1162 dpolicy->sync = true; 1163 dpolicy->ordered = false; 1164 dpolicy->granularity = granularity; 1165 1166 dpolicy->max_requests = dcc->max_discard_request; 1167 dpolicy->io_aware_gran = dcc->discard_io_aware_gran; 1168 dpolicy->timeout = false; 1169 1170 if (discard_type == DPOLICY_BG) { 1171 dpolicy->min_interval = dcc->min_discard_issue_time; 1172 dpolicy->mid_interval = dcc->mid_discard_issue_time; 1173 dpolicy->max_interval = dcc->max_discard_issue_time; 1174 dpolicy->io_aware = true; 1175 dpolicy->sync = false; 1176 dpolicy->ordered = true; 1177 if (utilization(sbi) > dcc->discard_urgent_util) { 1178 dpolicy->granularity = MIN_DISCARD_GRANULARITY; 1179 if (atomic_read(&dcc->discard_cmd_cnt)) 1180 dpolicy->max_interval = 1181 dcc->min_discard_issue_time; 1182 } 1183 } else if (discard_type == DPOLICY_FORCE) { 1184 dpolicy->min_interval = dcc->min_discard_issue_time; 1185 dpolicy->mid_interval = dcc->mid_discard_issue_time; 1186 dpolicy->max_interval = dcc->max_discard_issue_time; 1187 dpolicy->io_aware = false; 1188 } else if (discard_type == DPOLICY_FSTRIM) { 1189 dpolicy->io_aware = false; 1190 } else if (discard_type == DPOLICY_UMOUNT) { 1191 dpolicy->io_aware = false; 1192 /* we need to issue all to keep CP_TRIMMED_FLAG */ 1193 dpolicy->granularity = MIN_DISCARD_GRANULARITY; 1194 dpolicy->timeout = true; 1195 } 1196 } 1197 1198 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1199 struct block_device *bdev, block_t lstart, 1200 block_t start, block_t len); 1201 1202 #ifdef CONFIG_BLK_DEV_ZONED 1203 static void __submit_zone_reset_cmd(struct f2fs_sb_info *sbi, 1204 struct discard_cmd *dc, blk_opf_t flag, 1205 struct list_head *wait_list, 1206 unsigned int *issued) 1207 { 1208 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1209 struct block_device *bdev = dc->bdev; 1210 struct bio *bio = bio_alloc(bdev, 0, REQ_OP_ZONE_RESET | flag, GFP_NOFS); 1211 unsigned long flags; 1212 1213 trace_f2fs_issue_reset_zone(bdev, dc->di.start); 1214 1215 spin_lock_irqsave(&dc->lock, flags); 1216 dc->state = D_SUBMIT; 1217 dc->bio_ref++; 1218 spin_unlock_irqrestore(&dc->lock, flags); 1219 1220 if (issued) 1221 (*issued)++; 1222 1223 atomic_inc(&dcc->queued_discard); 1224 dc->queued++; 1225 list_move_tail(&dc->list, wait_list); 1226 1227 /* sanity check on discard range */ 1228 __check_sit_bitmap(sbi, dc->di.lstart, dc->di.lstart + dc->di.len); 1229 1230 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(dc->di.start); 1231 bio->bi_private = dc; 1232 bio->bi_end_io = f2fs_submit_discard_endio; 1233 submit_bio(bio); 1234 1235 atomic_inc(&dcc->issued_discard); 1236 f2fs_update_iostat(sbi, NULL, FS_ZONE_RESET_IO, dc->di.len * F2FS_BLKSIZE); 1237 } 1238 #endif 1239 1240 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */ 1241 static int __submit_discard_cmd(struct f2fs_sb_info *sbi, 1242 struct discard_policy *dpolicy, 1243 struct discard_cmd *dc, int *issued) 1244 { 1245 struct block_device *bdev = dc->bdev; 1246 unsigned int max_discard_blocks = 1247 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev)); 1248 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1249 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1250 &(dcc->fstrim_list) : &(dcc->wait_list); 1251 blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0; 1252 block_t lstart, start, len, total_len; 1253 int err = 0; 1254 1255 if (dc->state != D_PREP) 1256 return 0; 1257 1258 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 1259 return 0; 1260 1261 #ifdef CONFIG_BLK_DEV_ZONED 1262 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) { 1263 int devi = f2fs_bdev_index(sbi, bdev); 1264 1265 if (devi < 0) 1266 return -EINVAL; 1267 1268 if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) { 1269 __submit_zone_reset_cmd(sbi, dc, flag, 1270 wait_list, issued); 1271 return 0; 1272 } 1273 } 1274 #endif 1275 1276 trace_f2fs_issue_discard(bdev, dc->di.start, dc->di.len); 1277 1278 lstart = dc->di.lstart; 1279 start = dc->di.start; 1280 len = dc->di.len; 1281 total_len = len; 1282 1283 dc->di.len = 0; 1284 1285 while (total_len && *issued < dpolicy->max_requests && !err) { 1286 struct bio *bio = NULL; 1287 unsigned long flags; 1288 bool last = true; 1289 1290 if (len > max_discard_blocks) { 1291 len = max_discard_blocks; 1292 last = false; 1293 } 1294 1295 (*issued)++; 1296 if (*issued == dpolicy->max_requests) 1297 last = true; 1298 1299 dc->di.len += len; 1300 1301 if (time_to_inject(sbi, FAULT_DISCARD)) { 1302 err = -EIO; 1303 } else { 1304 err = __blkdev_issue_discard(bdev, 1305 SECTOR_FROM_BLOCK(start), 1306 SECTOR_FROM_BLOCK(len), 1307 GFP_NOFS, &bio); 1308 } 1309 if (err) { 1310 spin_lock_irqsave(&dc->lock, flags); 1311 if (dc->state == D_PARTIAL) 1312 dc->state = D_SUBMIT; 1313 spin_unlock_irqrestore(&dc->lock, flags); 1314 1315 break; 1316 } 1317 1318 f2fs_bug_on(sbi, !bio); 1319 1320 /* 1321 * should keep before submission to avoid D_DONE 1322 * right away 1323 */ 1324 spin_lock_irqsave(&dc->lock, flags); 1325 if (last) 1326 dc->state = D_SUBMIT; 1327 else 1328 dc->state = D_PARTIAL; 1329 dc->bio_ref++; 1330 spin_unlock_irqrestore(&dc->lock, flags); 1331 1332 atomic_inc(&dcc->queued_discard); 1333 dc->queued++; 1334 list_move_tail(&dc->list, wait_list); 1335 1336 /* sanity check on discard range */ 1337 __check_sit_bitmap(sbi, lstart, lstart + len); 1338 1339 bio->bi_private = dc; 1340 bio->bi_end_io = f2fs_submit_discard_endio; 1341 bio->bi_opf |= flag; 1342 submit_bio(bio); 1343 1344 atomic_inc(&dcc->issued_discard); 1345 1346 f2fs_update_iostat(sbi, NULL, FS_DISCARD_IO, len * F2FS_BLKSIZE); 1347 1348 lstart += len; 1349 start += len; 1350 total_len -= len; 1351 len = total_len; 1352 } 1353 1354 if (!err && len) { 1355 dcc->undiscard_blks -= len; 1356 __update_discard_tree_range(sbi, bdev, lstart, start, len); 1357 } 1358 return err; 1359 } 1360 1361 static void __insert_discard_cmd(struct f2fs_sb_info *sbi, 1362 struct block_device *bdev, block_t lstart, 1363 block_t start, block_t len) 1364 { 1365 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1366 struct rb_node **p = &dcc->root.rb_root.rb_node; 1367 struct rb_node *parent = NULL; 1368 struct discard_cmd *dc; 1369 bool leftmost = true; 1370 1371 /* look up rb tree to find parent node */ 1372 while (*p) { 1373 parent = *p; 1374 dc = rb_entry(parent, struct discard_cmd, rb_node); 1375 1376 if (lstart < dc->di.lstart) { 1377 p = &(*p)->rb_left; 1378 } else if (lstart >= dc->di.lstart + dc->di.len) { 1379 p = &(*p)->rb_right; 1380 leftmost = false; 1381 } else { 1382 f2fs_bug_on(sbi, 1); 1383 } 1384 } 1385 1386 dc = __create_discard_cmd(sbi, bdev, lstart, start, len); 1387 1388 rb_link_node(&dc->rb_node, parent, p); 1389 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost); 1390 } 1391 1392 static void __relocate_discard_cmd(struct discard_cmd_control *dcc, 1393 struct discard_cmd *dc) 1394 { 1395 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->di.len)]); 1396 } 1397 1398 static void __punch_discard_cmd(struct f2fs_sb_info *sbi, 1399 struct discard_cmd *dc, block_t blkaddr) 1400 { 1401 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1402 struct discard_info di = dc->di; 1403 bool modified = false; 1404 1405 if (dc->state == D_DONE || dc->di.len == 1) { 1406 __remove_discard_cmd(sbi, dc); 1407 return; 1408 } 1409 1410 dcc->undiscard_blks -= di.len; 1411 1412 if (blkaddr > di.lstart) { 1413 dc->di.len = blkaddr - dc->di.lstart; 1414 dcc->undiscard_blks += dc->di.len; 1415 __relocate_discard_cmd(dcc, dc); 1416 modified = true; 1417 } 1418 1419 if (blkaddr < di.lstart + di.len - 1) { 1420 if (modified) { 1421 __insert_discard_cmd(sbi, dc->bdev, blkaddr + 1, 1422 di.start + blkaddr + 1 - di.lstart, 1423 di.lstart + di.len - 1 - blkaddr); 1424 } else { 1425 dc->di.lstart++; 1426 dc->di.len--; 1427 dc->di.start++; 1428 dcc->undiscard_blks += dc->di.len; 1429 __relocate_discard_cmd(dcc, dc); 1430 } 1431 } 1432 } 1433 1434 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1435 struct block_device *bdev, block_t lstart, 1436 block_t start, block_t len) 1437 { 1438 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1439 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1440 struct discard_cmd *dc; 1441 struct discard_info di = {0}; 1442 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1443 unsigned int max_discard_blocks = 1444 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev)); 1445 block_t end = lstart + len; 1446 1447 dc = __lookup_discard_cmd_ret(&dcc->root, lstart, 1448 &prev_dc, &next_dc, &insert_p, &insert_parent); 1449 if (dc) 1450 prev_dc = dc; 1451 1452 if (!prev_dc) { 1453 di.lstart = lstart; 1454 di.len = next_dc ? next_dc->di.lstart - lstart : len; 1455 di.len = min(di.len, len); 1456 di.start = start; 1457 } 1458 1459 while (1) { 1460 struct rb_node *node; 1461 bool merged = false; 1462 struct discard_cmd *tdc = NULL; 1463 1464 if (prev_dc) { 1465 di.lstart = prev_dc->di.lstart + prev_dc->di.len; 1466 if (di.lstart < lstart) 1467 di.lstart = lstart; 1468 if (di.lstart >= end) 1469 break; 1470 1471 if (!next_dc || next_dc->di.lstart > end) 1472 di.len = end - di.lstart; 1473 else 1474 di.len = next_dc->di.lstart - di.lstart; 1475 di.start = start + di.lstart - lstart; 1476 } 1477 1478 if (!di.len) 1479 goto next; 1480 1481 if (prev_dc && prev_dc->state == D_PREP && 1482 prev_dc->bdev == bdev && 1483 __is_discard_back_mergeable(&di, &prev_dc->di, 1484 max_discard_blocks)) { 1485 prev_dc->di.len += di.len; 1486 dcc->undiscard_blks += di.len; 1487 __relocate_discard_cmd(dcc, prev_dc); 1488 di = prev_dc->di; 1489 tdc = prev_dc; 1490 merged = true; 1491 } 1492 1493 if (next_dc && next_dc->state == D_PREP && 1494 next_dc->bdev == bdev && 1495 __is_discard_front_mergeable(&di, &next_dc->di, 1496 max_discard_blocks)) { 1497 next_dc->di.lstart = di.lstart; 1498 next_dc->di.len += di.len; 1499 next_dc->di.start = di.start; 1500 dcc->undiscard_blks += di.len; 1501 __relocate_discard_cmd(dcc, next_dc); 1502 if (tdc) 1503 __remove_discard_cmd(sbi, tdc); 1504 merged = true; 1505 } 1506 1507 if (!merged) 1508 __insert_discard_cmd(sbi, bdev, 1509 di.lstart, di.start, di.len); 1510 next: 1511 prev_dc = next_dc; 1512 if (!prev_dc) 1513 break; 1514 1515 node = rb_next(&prev_dc->rb_node); 1516 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1517 } 1518 } 1519 1520 #ifdef CONFIG_BLK_DEV_ZONED 1521 static void __queue_zone_reset_cmd(struct f2fs_sb_info *sbi, 1522 struct block_device *bdev, block_t blkstart, block_t lblkstart, 1523 block_t blklen) 1524 { 1525 trace_f2fs_queue_reset_zone(bdev, blkstart); 1526 1527 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock); 1528 __insert_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen); 1529 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock); 1530 } 1531 #endif 1532 1533 static void __queue_discard_cmd(struct f2fs_sb_info *sbi, 1534 struct block_device *bdev, block_t blkstart, block_t blklen) 1535 { 1536 block_t lblkstart = blkstart; 1537 1538 if (!f2fs_bdev_support_discard(bdev)) 1539 return; 1540 1541 trace_f2fs_queue_discard(bdev, blkstart, blklen); 1542 1543 if (f2fs_is_multi_device(sbi)) { 1544 int devi = f2fs_target_device_index(sbi, blkstart); 1545 1546 blkstart -= FDEV(devi).start_blk; 1547 } 1548 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock); 1549 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen); 1550 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock); 1551 } 1552 1553 static void __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi, 1554 struct discard_policy *dpolicy, int *issued) 1555 { 1556 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1557 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1558 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1559 struct discard_cmd *dc; 1560 struct blk_plug plug; 1561 bool io_interrupted = false; 1562 1563 mutex_lock(&dcc->cmd_lock); 1564 dc = __lookup_discard_cmd_ret(&dcc->root, dcc->next_pos, 1565 &prev_dc, &next_dc, &insert_p, &insert_parent); 1566 if (!dc) 1567 dc = next_dc; 1568 1569 blk_start_plug(&plug); 1570 1571 while (dc) { 1572 struct rb_node *node; 1573 int err = 0; 1574 1575 if (dc->state != D_PREP) 1576 goto next; 1577 1578 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) { 1579 io_interrupted = true; 1580 break; 1581 } 1582 1583 dcc->next_pos = dc->di.lstart + dc->di.len; 1584 err = __submit_discard_cmd(sbi, dpolicy, dc, issued); 1585 1586 if (*issued >= dpolicy->max_requests) 1587 break; 1588 next: 1589 node = rb_next(&dc->rb_node); 1590 if (err) 1591 __remove_discard_cmd(sbi, dc); 1592 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1593 } 1594 1595 blk_finish_plug(&plug); 1596 1597 if (!dc) 1598 dcc->next_pos = 0; 1599 1600 mutex_unlock(&dcc->cmd_lock); 1601 1602 if (!(*issued) && io_interrupted) 1603 *issued = -1; 1604 } 1605 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1606 struct discard_policy *dpolicy); 1607 1608 static int __issue_discard_cmd(struct f2fs_sb_info *sbi, 1609 struct discard_policy *dpolicy) 1610 { 1611 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1612 struct list_head *pend_list; 1613 struct discard_cmd *dc, *tmp; 1614 struct blk_plug plug; 1615 int i, issued; 1616 bool io_interrupted = false; 1617 1618 if (dpolicy->timeout) 1619 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT); 1620 1621 retry: 1622 issued = 0; 1623 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1624 if (dpolicy->timeout && 1625 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1626 break; 1627 1628 if (i + 1 < dpolicy->granularity) 1629 break; 1630 1631 if (i + 1 < dcc->max_ordered_discard && dpolicy->ordered) { 1632 __issue_discard_cmd_orderly(sbi, dpolicy, &issued); 1633 return issued; 1634 } 1635 1636 pend_list = &dcc->pend_list[i]; 1637 1638 mutex_lock(&dcc->cmd_lock); 1639 if (list_empty(pend_list)) 1640 goto next; 1641 if (unlikely(dcc->rbtree_check)) 1642 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi)); 1643 blk_start_plug(&plug); 1644 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1645 f2fs_bug_on(sbi, dc->state != D_PREP); 1646 1647 if (dpolicy->timeout && 1648 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1649 break; 1650 1651 if (dpolicy->io_aware && i < dpolicy->io_aware_gran && 1652 !is_idle(sbi, DISCARD_TIME)) { 1653 io_interrupted = true; 1654 break; 1655 } 1656 1657 __submit_discard_cmd(sbi, dpolicy, dc, &issued); 1658 1659 if (issued >= dpolicy->max_requests) 1660 break; 1661 } 1662 blk_finish_plug(&plug); 1663 next: 1664 mutex_unlock(&dcc->cmd_lock); 1665 1666 if (issued >= dpolicy->max_requests || io_interrupted) 1667 break; 1668 } 1669 1670 if (dpolicy->type == DPOLICY_UMOUNT && issued) { 1671 __wait_all_discard_cmd(sbi, dpolicy); 1672 goto retry; 1673 } 1674 1675 if (!issued && io_interrupted) 1676 issued = -1; 1677 1678 return issued; 1679 } 1680 1681 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi) 1682 { 1683 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1684 struct list_head *pend_list; 1685 struct discard_cmd *dc, *tmp; 1686 int i; 1687 bool dropped = false; 1688 1689 mutex_lock(&dcc->cmd_lock); 1690 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1691 pend_list = &dcc->pend_list[i]; 1692 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1693 f2fs_bug_on(sbi, dc->state != D_PREP); 1694 __remove_discard_cmd(sbi, dc); 1695 dropped = true; 1696 } 1697 } 1698 mutex_unlock(&dcc->cmd_lock); 1699 1700 return dropped; 1701 } 1702 1703 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi) 1704 { 1705 __drop_discard_cmd(sbi); 1706 } 1707 1708 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi, 1709 struct discard_cmd *dc) 1710 { 1711 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1712 unsigned int len = 0; 1713 1714 wait_for_completion_io(&dc->wait); 1715 mutex_lock(&dcc->cmd_lock); 1716 f2fs_bug_on(sbi, dc->state != D_DONE); 1717 dc->ref--; 1718 if (!dc->ref) { 1719 if (!dc->error) 1720 len = dc->di.len; 1721 __remove_discard_cmd(sbi, dc); 1722 } 1723 mutex_unlock(&dcc->cmd_lock); 1724 1725 return len; 1726 } 1727 1728 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi, 1729 struct discard_policy *dpolicy, 1730 block_t start, block_t end) 1731 { 1732 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1733 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1734 &(dcc->fstrim_list) : &(dcc->wait_list); 1735 struct discard_cmd *dc = NULL, *iter, *tmp; 1736 unsigned int trimmed = 0; 1737 1738 next: 1739 dc = NULL; 1740 1741 mutex_lock(&dcc->cmd_lock); 1742 list_for_each_entry_safe(iter, tmp, wait_list, list) { 1743 if (iter->di.lstart + iter->di.len <= start || 1744 end <= iter->di.lstart) 1745 continue; 1746 if (iter->di.len < dpolicy->granularity) 1747 continue; 1748 if (iter->state == D_DONE && !iter->ref) { 1749 wait_for_completion_io(&iter->wait); 1750 if (!iter->error) 1751 trimmed += iter->di.len; 1752 __remove_discard_cmd(sbi, iter); 1753 } else { 1754 iter->ref++; 1755 dc = iter; 1756 break; 1757 } 1758 } 1759 mutex_unlock(&dcc->cmd_lock); 1760 1761 if (dc) { 1762 trimmed += __wait_one_discard_bio(sbi, dc); 1763 goto next; 1764 } 1765 1766 return trimmed; 1767 } 1768 1769 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1770 struct discard_policy *dpolicy) 1771 { 1772 struct discard_policy dp; 1773 unsigned int discard_blks; 1774 1775 if (dpolicy) 1776 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX); 1777 1778 /* wait all */ 1779 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, MIN_DISCARD_GRANULARITY); 1780 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1781 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, MIN_DISCARD_GRANULARITY); 1782 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1783 1784 return discard_blks; 1785 } 1786 1787 /* This should be covered by global mutex, &sit_i->sentry_lock */ 1788 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr) 1789 { 1790 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1791 struct discard_cmd *dc; 1792 bool need_wait = false; 1793 1794 mutex_lock(&dcc->cmd_lock); 1795 dc = __lookup_discard_cmd(sbi, blkaddr); 1796 #ifdef CONFIG_BLK_DEV_ZONED 1797 if (dc && f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(dc->bdev)) { 1798 int devi = f2fs_bdev_index(sbi, dc->bdev); 1799 1800 if (devi < 0) { 1801 mutex_unlock(&dcc->cmd_lock); 1802 return; 1803 } 1804 1805 if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) { 1806 /* force submit zone reset */ 1807 if (dc->state == D_PREP) 1808 __submit_zone_reset_cmd(sbi, dc, REQ_SYNC, 1809 &dcc->wait_list, NULL); 1810 dc->ref++; 1811 mutex_unlock(&dcc->cmd_lock); 1812 /* wait zone reset */ 1813 __wait_one_discard_bio(sbi, dc); 1814 return; 1815 } 1816 } 1817 #endif 1818 if (dc) { 1819 if (dc->state == D_PREP) { 1820 __punch_discard_cmd(sbi, dc, blkaddr); 1821 } else { 1822 dc->ref++; 1823 need_wait = true; 1824 } 1825 } 1826 mutex_unlock(&dcc->cmd_lock); 1827 1828 if (need_wait) 1829 __wait_one_discard_bio(sbi, dc); 1830 } 1831 1832 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi) 1833 { 1834 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1835 1836 if (dcc && dcc->f2fs_issue_discard) { 1837 struct task_struct *discard_thread = dcc->f2fs_issue_discard; 1838 1839 dcc->f2fs_issue_discard = NULL; 1840 kthread_stop(discard_thread); 1841 } 1842 } 1843 1844 /** 1845 * f2fs_issue_discard_timeout() - Issue all discard cmd within UMOUNT_DISCARD_TIMEOUT 1846 * @sbi: the f2fs_sb_info data for discard cmd to issue 1847 * 1848 * When UMOUNT_DISCARD_TIMEOUT is exceeded, all remaining discard commands will be dropped 1849 * 1850 * Return true if issued all discard cmd or no discard cmd need issue, otherwise return false. 1851 */ 1852 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi) 1853 { 1854 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1855 struct discard_policy dpolicy; 1856 bool dropped; 1857 1858 if (!atomic_read(&dcc->discard_cmd_cnt)) 1859 return true; 1860 1861 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT, 1862 dcc->discard_granularity); 1863 __issue_discard_cmd(sbi, &dpolicy); 1864 dropped = __drop_discard_cmd(sbi); 1865 1866 /* just to make sure there is no pending discard commands */ 1867 __wait_all_discard_cmd(sbi, NULL); 1868 1869 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt)); 1870 return !dropped; 1871 } 1872 1873 static int issue_discard_thread(void *data) 1874 { 1875 struct f2fs_sb_info *sbi = data; 1876 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1877 wait_queue_head_t *q = &dcc->discard_wait_queue; 1878 struct discard_policy dpolicy; 1879 unsigned int wait_ms = dcc->min_discard_issue_time; 1880 int issued; 1881 1882 set_freezable(); 1883 1884 do { 1885 wait_event_interruptible_timeout(*q, 1886 kthread_should_stop() || freezing(current) || 1887 dcc->discard_wake, 1888 msecs_to_jiffies(wait_ms)); 1889 1890 if (sbi->gc_mode == GC_URGENT_HIGH || 1891 !f2fs_available_free_memory(sbi, DISCARD_CACHE)) 1892 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1893 MIN_DISCARD_GRANULARITY); 1894 else 1895 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG, 1896 dcc->discard_granularity); 1897 1898 if (dcc->discard_wake) 1899 dcc->discard_wake = false; 1900 1901 /* clean up pending candidates before going to sleep */ 1902 if (atomic_read(&dcc->queued_discard)) 1903 __wait_all_discard_cmd(sbi, NULL); 1904 1905 if (try_to_freeze()) 1906 continue; 1907 if (f2fs_readonly(sbi->sb)) 1908 continue; 1909 if (kthread_should_stop()) 1910 return 0; 1911 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) || 1912 !atomic_read(&dcc->discard_cmd_cnt)) { 1913 wait_ms = dpolicy.max_interval; 1914 continue; 1915 } 1916 1917 sb_start_intwrite(sbi->sb); 1918 1919 issued = __issue_discard_cmd(sbi, &dpolicy); 1920 if (issued > 0) { 1921 __wait_all_discard_cmd(sbi, &dpolicy); 1922 wait_ms = dpolicy.min_interval; 1923 } else if (issued == -1) { 1924 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME); 1925 if (!wait_ms) 1926 wait_ms = dpolicy.mid_interval; 1927 } else { 1928 wait_ms = dpolicy.max_interval; 1929 } 1930 if (!atomic_read(&dcc->discard_cmd_cnt)) 1931 wait_ms = dpolicy.max_interval; 1932 1933 sb_end_intwrite(sbi->sb); 1934 1935 } while (!kthread_should_stop()); 1936 return 0; 1937 } 1938 1939 #ifdef CONFIG_BLK_DEV_ZONED 1940 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi, 1941 struct block_device *bdev, block_t blkstart, block_t blklen) 1942 { 1943 sector_t sector, nr_sects; 1944 block_t lblkstart = blkstart; 1945 int devi = 0; 1946 u64 remainder = 0; 1947 1948 if (f2fs_is_multi_device(sbi)) { 1949 devi = f2fs_target_device_index(sbi, blkstart); 1950 if (blkstart < FDEV(devi).start_blk || 1951 blkstart > FDEV(devi).end_blk) { 1952 f2fs_err(sbi, "Invalid block %x", blkstart); 1953 return -EIO; 1954 } 1955 blkstart -= FDEV(devi).start_blk; 1956 } 1957 1958 /* For sequential zones, reset the zone write pointer */ 1959 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) { 1960 sector = SECTOR_FROM_BLOCK(blkstart); 1961 nr_sects = SECTOR_FROM_BLOCK(blklen); 1962 div64_u64_rem(sector, bdev_zone_sectors(bdev), &remainder); 1963 1964 if (remainder || nr_sects != bdev_zone_sectors(bdev)) { 1965 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)", 1966 devi, sbi->s_ndevs ? FDEV(devi).path : "", 1967 blkstart, blklen); 1968 return -EIO; 1969 } 1970 1971 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) { 1972 trace_f2fs_issue_reset_zone(bdev, blkstart); 1973 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, 1974 sector, nr_sects, GFP_NOFS); 1975 } 1976 1977 __queue_zone_reset_cmd(sbi, bdev, blkstart, lblkstart, blklen); 1978 return 0; 1979 } 1980 1981 /* For conventional zones, use regular discard if supported */ 1982 __queue_discard_cmd(sbi, bdev, lblkstart, blklen); 1983 return 0; 1984 } 1985 #endif 1986 1987 static int __issue_discard_async(struct f2fs_sb_info *sbi, 1988 struct block_device *bdev, block_t blkstart, block_t blklen) 1989 { 1990 #ifdef CONFIG_BLK_DEV_ZONED 1991 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) 1992 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen); 1993 #endif 1994 __queue_discard_cmd(sbi, bdev, blkstart, blklen); 1995 return 0; 1996 } 1997 1998 static int f2fs_issue_discard(struct f2fs_sb_info *sbi, 1999 block_t blkstart, block_t blklen) 2000 { 2001 sector_t start = blkstart, len = 0; 2002 struct block_device *bdev; 2003 struct seg_entry *se; 2004 unsigned int offset; 2005 block_t i; 2006 int err = 0; 2007 2008 bdev = f2fs_target_device(sbi, blkstart, NULL); 2009 2010 for (i = blkstart; i < blkstart + blklen; i++, len++) { 2011 if (i != start) { 2012 struct block_device *bdev2 = 2013 f2fs_target_device(sbi, i, NULL); 2014 2015 if (bdev2 != bdev) { 2016 err = __issue_discard_async(sbi, bdev, 2017 start, len); 2018 if (err) 2019 return err; 2020 bdev = bdev2; 2021 start = i; 2022 len = 0; 2023 } 2024 } 2025 2026 se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); 2027 offset = GET_BLKOFF_FROM_SEG0(sbi, i); 2028 2029 if (f2fs_block_unit_discard(sbi) && 2030 !f2fs_test_and_set_bit(offset, se->discard_map)) 2031 sbi->discard_blks--; 2032 } 2033 2034 if (len) 2035 err = __issue_discard_async(sbi, bdev, start, len); 2036 return err; 2037 } 2038 2039 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc, 2040 bool check_only) 2041 { 2042 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 2043 int max_blocks = sbi->blocks_per_seg; 2044 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); 2045 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 2046 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 2047 unsigned long *discard_map = (unsigned long *)se->discard_map; 2048 unsigned long *dmap = SIT_I(sbi)->tmp_map; 2049 unsigned int start = 0, end = -1; 2050 bool force = (cpc->reason & CP_DISCARD); 2051 struct discard_entry *de = NULL; 2052 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list; 2053 int i; 2054 2055 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi) || 2056 !f2fs_block_unit_discard(sbi)) 2057 return false; 2058 2059 if (!force) { 2060 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks || 2061 SM_I(sbi)->dcc_info->nr_discards >= 2062 SM_I(sbi)->dcc_info->max_discards) 2063 return false; 2064 } 2065 2066 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ 2067 for (i = 0; i < entries; i++) 2068 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : 2069 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; 2070 2071 while (force || SM_I(sbi)->dcc_info->nr_discards <= 2072 SM_I(sbi)->dcc_info->max_discards) { 2073 start = __find_rev_next_bit(dmap, max_blocks, end + 1); 2074 if (start >= max_blocks) 2075 break; 2076 2077 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); 2078 if (force && start && end != max_blocks 2079 && (end - start) < cpc->trim_minlen) 2080 continue; 2081 2082 if (check_only) 2083 return true; 2084 2085 if (!de) { 2086 de = f2fs_kmem_cache_alloc(discard_entry_slab, 2087 GFP_F2FS_ZERO, true, NULL); 2088 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start); 2089 list_add_tail(&de->list, head); 2090 } 2091 2092 for (i = start; i < end; i++) 2093 __set_bit_le(i, (void *)de->discard_map); 2094 2095 SM_I(sbi)->dcc_info->nr_discards += end - start; 2096 } 2097 return false; 2098 } 2099 2100 static void release_discard_addr(struct discard_entry *entry) 2101 { 2102 list_del(&entry->list); 2103 kmem_cache_free(discard_entry_slab, entry); 2104 } 2105 2106 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi) 2107 { 2108 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); 2109 struct discard_entry *entry, *this; 2110 2111 /* drop caches */ 2112 list_for_each_entry_safe(entry, this, head, list) 2113 release_discard_addr(entry); 2114 } 2115 2116 /* 2117 * Should call f2fs_clear_prefree_segments after checkpoint is done. 2118 */ 2119 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 2120 { 2121 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2122 unsigned int segno; 2123 2124 mutex_lock(&dirty_i->seglist_lock); 2125 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) 2126 __set_test_and_free(sbi, segno, false); 2127 mutex_unlock(&dirty_i->seglist_lock); 2128 } 2129 2130 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi, 2131 struct cp_control *cpc) 2132 { 2133 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2134 struct list_head *head = &dcc->entry_list; 2135 struct discard_entry *entry, *this; 2136 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2137 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; 2138 unsigned int start = 0, end = -1; 2139 unsigned int secno, start_segno; 2140 bool force = (cpc->reason & CP_DISCARD); 2141 bool section_alignment = F2FS_OPTION(sbi).discard_unit == 2142 DISCARD_UNIT_SECTION; 2143 2144 if (f2fs_lfs_mode(sbi) && __is_large_section(sbi)) 2145 section_alignment = true; 2146 2147 mutex_lock(&dirty_i->seglist_lock); 2148 2149 while (1) { 2150 int i; 2151 2152 if (section_alignment && end != -1) 2153 end--; 2154 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); 2155 if (start >= MAIN_SEGS(sbi)) 2156 break; 2157 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), 2158 start + 1); 2159 2160 if (section_alignment) { 2161 start = rounddown(start, sbi->segs_per_sec); 2162 end = roundup(end, sbi->segs_per_sec); 2163 } 2164 2165 for (i = start; i < end; i++) { 2166 if (test_and_clear_bit(i, prefree_map)) 2167 dirty_i->nr_dirty[PRE]--; 2168 } 2169 2170 if (!f2fs_realtime_discard_enable(sbi)) 2171 continue; 2172 2173 if (force && start >= cpc->trim_start && 2174 (end - 1) <= cpc->trim_end) 2175 continue; 2176 2177 /* Should cover 2MB zoned device for zone-based reset */ 2178 if (!f2fs_sb_has_blkzoned(sbi) && 2179 (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi))) { 2180 f2fs_issue_discard(sbi, START_BLOCK(sbi, start), 2181 (end - start) << sbi->log_blocks_per_seg); 2182 continue; 2183 } 2184 next: 2185 secno = GET_SEC_FROM_SEG(sbi, start); 2186 start_segno = GET_SEG_FROM_SEC(sbi, secno); 2187 if (!IS_CURSEC(sbi, secno) && 2188 !get_valid_blocks(sbi, start, true)) 2189 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno), 2190 sbi->segs_per_sec << sbi->log_blocks_per_seg); 2191 2192 start = start_segno + sbi->segs_per_sec; 2193 if (start < end) 2194 goto next; 2195 else 2196 end = start - 1; 2197 } 2198 mutex_unlock(&dirty_i->seglist_lock); 2199 2200 if (!f2fs_block_unit_discard(sbi)) 2201 goto wakeup; 2202 2203 /* send small discards */ 2204 list_for_each_entry_safe(entry, this, head, list) { 2205 unsigned int cur_pos = 0, next_pos, len, total_len = 0; 2206 bool is_valid = test_bit_le(0, entry->discard_map); 2207 2208 find_next: 2209 if (is_valid) { 2210 next_pos = find_next_zero_bit_le(entry->discard_map, 2211 sbi->blocks_per_seg, cur_pos); 2212 len = next_pos - cur_pos; 2213 2214 if (f2fs_sb_has_blkzoned(sbi) || 2215 !force || len < cpc->trim_minlen) 2216 goto skip; 2217 2218 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos, 2219 len); 2220 total_len += len; 2221 } else { 2222 next_pos = find_next_bit_le(entry->discard_map, 2223 sbi->blocks_per_seg, cur_pos); 2224 } 2225 skip: 2226 cur_pos = next_pos; 2227 is_valid = !is_valid; 2228 2229 if (cur_pos < sbi->blocks_per_seg) 2230 goto find_next; 2231 2232 release_discard_addr(entry); 2233 dcc->nr_discards -= total_len; 2234 } 2235 2236 wakeup: 2237 wake_up_discard_thread(sbi, false); 2238 } 2239 2240 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi) 2241 { 2242 dev_t dev = sbi->sb->s_bdev->bd_dev; 2243 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2244 int err = 0; 2245 2246 if (!f2fs_realtime_discard_enable(sbi)) 2247 return 0; 2248 2249 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi, 2250 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev)); 2251 if (IS_ERR(dcc->f2fs_issue_discard)) { 2252 err = PTR_ERR(dcc->f2fs_issue_discard); 2253 dcc->f2fs_issue_discard = NULL; 2254 } 2255 2256 return err; 2257 } 2258 2259 static int create_discard_cmd_control(struct f2fs_sb_info *sbi) 2260 { 2261 struct discard_cmd_control *dcc; 2262 int err = 0, i; 2263 2264 if (SM_I(sbi)->dcc_info) { 2265 dcc = SM_I(sbi)->dcc_info; 2266 goto init_thread; 2267 } 2268 2269 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL); 2270 if (!dcc) 2271 return -ENOMEM; 2272 2273 dcc->discard_io_aware_gran = MAX_PLIST_NUM; 2274 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY; 2275 dcc->max_ordered_discard = DEFAULT_MAX_ORDERED_DISCARD_GRANULARITY; 2276 if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT) 2277 dcc->discard_granularity = sbi->blocks_per_seg; 2278 else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION) 2279 dcc->discard_granularity = BLKS_PER_SEC(sbi); 2280 2281 INIT_LIST_HEAD(&dcc->entry_list); 2282 for (i = 0; i < MAX_PLIST_NUM; i++) 2283 INIT_LIST_HEAD(&dcc->pend_list[i]); 2284 INIT_LIST_HEAD(&dcc->wait_list); 2285 INIT_LIST_HEAD(&dcc->fstrim_list); 2286 mutex_init(&dcc->cmd_lock); 2287 atomic_set(&dcc->issued_discard, 0); 2288 atomic_set(&dcc->queued_discard, 0); 2289 atomic_set(&dcc->discard_cmd_cnt, 0); 2290 dcc->nr_discards = 0; 2291 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg; 2292 dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST; 2293 dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME; 2294 dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME; 2295 dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME; 2296 dcc->discard_urgent_util = DEF_DISCARD_URGENT_UTIL; 2297 dcc->undiscard_blks = 0; 2298 dcc->next_pos = 0; 2299 dcc->root = RB_ROOT_CACHED; 2300 dcc->rbtree_check = false; 2301 2302 init_waitqueue_head(&dcc->discard_wait_queue); 2303 SM_I(sbi)->dcc_info = dcc; 2304 init_thread: 2305 err = f2fs_start_discard_thread(sbi); 2306 if (err) { 2307 kfree(dcc); 2308 SM_I(sbi)->dcc_info = NULL; 2309 } 2310 2311 return err; 2312 } 2313 2314 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi) 2315 { 2316 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2317 2318 if (!dcc) 2319 return; 2320 2321 f2fs_stop_discard_thread(sbi); 2322 2323 /* 2324 * Recovery can cache discard commands, so in error path of 2325 * fill_super(), it needs to give a chance to handle them. 2326 */ 2327 f2fs_issue_discard_timeout(sbi); 2328 2329 kfree(dcc); 2330 SM_I(sbi)->dcc_info = NULL; 2331 } 2332 2333 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 2334 { 2335 struct sit_info *sit_i = SIT_I(sbi); 2336 2337 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { 2338 sit_i->dirty_sentries++; 2339 return false; 2340 } 2341 2342 return true; 2343 } 2344 2345 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 2346 unsigned int segno, int modified) 2347 { 2348 struct seg_entry *se = get_seg_entry(sbi, segno); 2349 2350 se->type = type; 2351 if (modified) 2352 __mark_sit_entry_dirty(sbi, segno); 2353 } 2354 2355 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi, 2356 block_t blkaddr) 2357 { 2358 unsigned int segno = GET_SEGNO(sbi, blkaddr); 2359 2360 if (segno == NULL_SEGNO) 2361 return 0; 2362 return get_seg_entry(sbi, segno)->mtime; 2363 } 2364 2365 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr, 2366 unsigned long long old_mtime) 2367 { 2368 struct seg_entry *se; 2369 unsigned int segno = GET_SEGNO(sbi, blkaddr); 2370 unsigned long long ctime = get_mtime(sbi, false); 2371 unsigned long long mtime = old_mtime ? old_mtime : ctime; 2372 2373 if (segno == NULL_SEGNO) 2374 return; 2375 2376 se = get_seg_entry(sbi, segno); 2377 2378 if (!se->mtime) 2379 se->mtime = mtime; 2380 else 2381 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime, 2382 se->valid_blocks + 1); 2383 2384 if (ctime > SIT_I(sbi)->max_mtime) 2385 SIT_I(sbi)->max_mtime = ctime; 2386 } 2387 2388 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 2389 { 2390 struct seg_entry *se; 2391 unsigned int segno, offset; 2392 long int new_vblocks; 2393 bool exist; 2394 #ifdef CONFIG_F2FS_CHECK_FS 2395 bool mir_exist; 2396 #endif 2397 2398 segno = GET_SEGNO(sbi, blkaddr); 2399 2400 se = get_seg_entry(sbi, segno); 2401 new_vblocks = se->valid_blocks + del; 2402 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2403 2404 f2fs_bug_on(sbi, (new_vblocks < 0 || 2405 (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno)))); 2406 2407 se->valid_blocks = new_vblocks; 2408 2409 /* Update valid block bitmap */ 2410 if (del > 0) { 2411 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map); 2412 #ifdef CONFIG_F2FS_CHECK_FS 2413 mir_exist = f2fs_test_and_set_bit(offset, 2414 se->cur_valid_map_mir); 2415 if (unlikely(exist != mir_exist)) { 2416 f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d", 2417 blkaddr, exist); 2418 f2fs_bug_on(sbi, 1); 2419 } 2420 #endif 2421 if (unlikely(exist)) { 2422 f2fs_err(sbi, "Bitmap was wrongly set, blk:%u", 2423 blkaddr); 2424 f2fs_bug_on(sbi, 1); 2425 se->valid_blocks--; 2426 del = 0; 2427 } 2428 2429 if (f2fs_block_unit_discard(sbi) && 2430 !f2fs_test_and_set_bit(offset, se->discard_map)) 2431 sbi->discard_blks--; 2432 2433 /* 2434 * SSR should never reuse block which is checkpointed 2435 * or newly invalidated. 2436 */ 2437 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) { 2438 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map)) 2439 se->ckpt_valid_blocks++; 2440 } 2441 } else { 2442 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map); 2443 #ifdef CONFIG_F2FS_CHECK_FS 2444 mir_exist = f2fs_test_and_clear_bit(offset, 2445 se->cur_valid_map_mir); 2446 if (unlikely(exist != mir_exist)) { 2447 f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d", 2448 blkaddr, exist); 2449 f2fs_bug_on(sbi, 1); 2450 } 2451 #endif 2452 if (unlikely(!exist)) { 2453 f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u", 2454 blkaddr); 2455 f2fs_bug_on(sbi, 1); 2456 se->valid_blocks++; 2457 del = 0; 2458 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 2459 /* 2460 * If checkpoints are off, we must not reuse data that 2461 * was used in the previous checkpoint. If it was used 2462 * before, we must track that to know how much space we 2463 * really have. 2464 */ 2465 if (f2fs_test_bit(offset, se->ckpt_valid_map)) { 2466 spin_lock(&sbi->stat_lock); 2467 sbi->unusable_block_count++; 2468 spin_unlock(&sbi->stat_lock); 2469 } 2470 } 2471 2472 if (f2fs_block_unit_discard(sbi) && 2473 f2fs_test_and_clear_bit(offset, se->discard_map)) 2474 sbi->discard_blks++; 2475 } 2476 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) 2477 se->ckpt_valid_blocks += del; 2478 2479 __mark_sit_entry_dirty(sbi, segno); 2480 2481 /* update total number of valid blocks to be written in ckpt area */ 2482 SIT_I(sbi)->written_valid_blocks += del; 2483 2484 if (__is_large_section(sbi)) 2485 get_sec_entry(sbi, segno)->valid_blocks += del; 2486 } 2487 2488 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) 2489 { 2490 unsigned int segno = GET_SEGNO(sbi, addr); 2491 struct sit_info *sit_i = SIT_I(sbi); 2492 2493 f2fs_bug_on(sbi, addr == NULL_ADDR); 2494 if (addr == NEW_ADDR || addr == COMPRESS_ADDR) 2495 return; 2496 2497 invalidate_mapping_pages(META_MAPPING(sbi), addr, addr); 2498 f2fs_invalidate_compress_page(sbi, addr); 2499 2500 /* add it into sit main buffer */ 2501 down_write(&sit_i->sentry_lock); 2502 2503 update_segment_mtime(sbi, addr, 0); 2504 update_sit_entry(sbi, addr, -1); 2505 2506 /* add it into dirty seglist */ 2507 locate_dirty_segment(sbi, segno); 2508 2509 up_write(&sit_i->sentry_lock); 2510 } 2511 2512 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr) 2513 { 2514 struct sit_info *sit_i = SIT_I(sbi); 2515 unsigned int segno, offset; 2516 struct seg_entry *se; 2517 bool is_cp = false; 2518 2519 if (!__is_valid_data_blkaddr(blkaddr)) 2520 return true; 2521 2522 down_read(&sit_i->sentry_lock); 2523 2524 segno = GET_SEGNO(sbi, blkaddr); 2525 se = get_seg_entry(sbi, segno); 2526 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2527 2528 if (f2fs_test_bit(offset, se->ckpt_valid_map)) 2529 is_cp = true; 2530 2531 up_read(&sit_i->sentry_lock); 2532 2533 return is_cp; 2534 } 2535 2536 static unsigned short f2fs_curseg_valid_blocks(struct f2fs_sb_info *sbi, int type) 2537 { 2538 struct curseg_info *curseg = CURSEG_I(sbi, type); 2539 2540 if (sbi->ckpt->alloc_type[type] == SSR) 2541 return sbi->blocks_per_seg; 2542 return curseg->next_blkoff; 2543 } 2544 2545 /* 2546 * Calculate the number of current summary pages for writing 2547 */ 2548 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) 2549 { 2550 int valid_sum_count = 0; 2551 int i, sum_in_page; 2552 2553 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2554 if (sbi->ckpt->alloc_type[i] != SSR && for_ra) 2555 valid_sum_count += 2556 le16_to_cpu(F2FS_CKPT(sbi)->cur_data_blkoff[i]); 2557 else 2558 valid_sum_count += f2fs_curseg_valid_blocks(sbi, i); 2559 } 2560 2561 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE - 2562 SUM_FOOTER_SIZE) / SUMMARY_SIZE; 2563 if (valid_sum_count <= sum_in_page) 2564 return 1; 2565 else if ((valid_sum_count - sum_in_page) <= 2566 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) 2567 return 2; 2568 return 3; 2569 } 2570 2571 /* 2572 * Caller should put this summary page 2573 */ 2574 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) 2575 { 2576 if (unlikely(f2fs_cp_error(sbi))) 2577 return ERR_PTR(-EIO); 2578 return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno)); 2579 } 2580 2581 void f2fs_update_meta_page(struct f2fs_sb_info *sbi, 2582 void *src, block_t blk_addr) 2583 { 2584 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2585 2586 memcpy(page_address(page), src, PAGE_SIZE); 2587 set_page_dirty(page); 2588 f2fs_put_page(page, 1); 2589 } 2590 2591 static void write_sum_page(struct f2fs_sb_info *sbi, 2592 struct f2fs_summary_block *sum_blk, block_t blk_addr) 2593 { 2594 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr); 2595 } 2596 2597 static void write_current_sum_page(struct f2fs_sb_info *sbi, 2598 int type, block_t blk_addr) 2599 { 2600 struct curseg_info *curseg = CURSEG_I(sbi, type); 2601 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2602 struct f2fs_summary_block *src = curseg->sum_blk; 2603 struct f2fs_summary_block *dst; 2604 2605 dst = (struct f2fs_summary_block *)page_address(page); 2606 memset(dst, 0, PAGE_SIZE); 2607 2608 mutex_lock(&curseg->curseg_mutex); 2609 2610 down_read(&curseg->journal_rwsem); 2611 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE); 2612 up_read(&curseg->journal_rwsem); 2613 2614 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE); 2615 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE); 2616 2617 mutex_unlock(&curseg->curseg_mutex); 2618 2619 set_page_dirty(page); 2620 f2fs_put_page(page, 1); 2621 } 2622 2623 static int is_next_segment_free(struct f2fs_sb_info *sbi, 2624 struct curseg_info *curseg, int type) 2625 { 2626 unsigned int segno = curseg->segno + 1; 2627 struct free_segmap_info *free_i = FREE_I(sbi); 2628 2629 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec) 2630 return !test_bit(segno, free_i->free_segmap); 2631 return 0; 2632 } 2633 2634 /* 2635 * Find a new segment from the free segments bitmap to right order 2636 * This function should be returned with success, otherwise BUG 2637 */ 2638 static void get_new_segment(struct f2fs_sb_info *sbi, 2639 unsigned int *newseg, bool new_sec, int dir) 2640 { 2641 struct free_segmap_info *free_i = FREE_I(sbi); 2642 unsigned int segno, secno, zoneno; 2643 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; 2644 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg); 2645 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg); 2646 unsigned int left_start = hint; 2647 bool init = true; 2648 int go_left = 0; 2649 int i; 2650 2651 spin_lock(&free_i->segmap_lock); 2652 2653 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { 2654 segno = find_next_zero_bit(free_i->free_segmap, 2655 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1); 2656 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1)) 2657 goto got_it; 2658 } 2659 find_other_zone: 2660 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 2661 if (secno >= MAIN_SECS(sbi)) { 2662 if (dir == ALLOC_RIGHT) { 2663 secno = find_first_zero_bit(free_i->free_secmap, 2664 MAIN_SECS(sbi)); 2665 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi)); 2666 } else { 2667 go_left = 1; 2668 left_start = hint - 1; 2669 } 2670 } 2671 if (go_left == 0) 2672 goto skip_left; 2673 2674 while (test_bit(left_start, free_i->free_secmap)) { 2675 if (left_start > 0) { 2676 left_start--; 2677 continue; 2678 } 2679 left_start = find_first_zero_bit(free_i->free_secmap, 2680 MAIN_SECS(sbi)); 2681 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi)); 2682 break; 2683 } 2684 secno = left_start; 2685 skip_left: 2686 segno = GET_SEG_FROM_SEC(sbi, secno); 2687 zoneno = GET_ZONE_FROM_SEC(sbi, secno); 2688 2689 /* give up on finding another zone */ 2690 if (!init) 2691 goto got_it; 2692 if (sbi->secs_per_zone == 1) 2693 goto got_it; 2694 if (zoneno == old_zoneno) 2695 goto got_it; 2696 if (dir == ALLOC_LEFT) { 2697 if (!go_left && zoneno + 1 >= total_zones) 2698 goto got_it; 2699 if (go_left && zoneno == 0) 2700 goto got_it; 2701 } 2702 for (i = 0; i < NR_CURSEG_TYPE; i++) 2703 if (CURSEG_I(sbi, i)->zone == zoneno) 2704 break; 2705 2706 if (i < NR_CURSEG_TYPE) { 2707 /* zone is in user, try another */ 2708 if (go_left) 2709 hint = zoneno * sbi->secs_per_zone - 1; 2710 else if (zoneno + 1 >= total_zones) 2711 hint = 0; 2712 else 2713 hint = (zoneno + 1) * sbi->secs_per_zone; 2714 init = false; 2715 goto find_other_zone; 2716 } 2717 got_it: 2718 /* set it as dirty segment in free segmap */ 2719 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); 2720 __set_inuse(sbi, segno); 2721 *newseg = segno; 2722 spin_unlock(&free_i->segmap_lock); 2723 } 2724 2725 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 2726 { 2727 struct curseg_info *curseg = CURSEG_I(sbi, type); 2728 struct summary_footer *sum_footer; 2729 unsigned short seg_type = curseg->seg_type; 2730 2731 curseg->inited = true; 2732 curseg->segno = curseg->next_segno; 2733 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno); 2734 curseg->next_blkoff = 0; 2735 curseg->next_segno = NULL_SEGNO; 2736 2737 sum_footer = &(curseg->sum_blk->footer); 2738 memset(sum_footer, 0, sizeof(struct summary_footer)); 2739 2740 sanity_check_seg_type(sbi, seg_type); 2741 2742 if (IS_DATASEG(seg_type)) 2743 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 2744 if (IS_NODESEG(seg_type)) 2745 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 2746 __set_sit_entry_type(sbi, seg_type, curseg->segno, modified); 2747 } 2748 2749 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type) 2750 { 2751 struct curseg_info *curseg = CURSEG_I(sbi, type); 2752 unsigned short seg_type = curseg->seg_type; 2753 2754 sanity_check_seg_type(sbi, seg_type); 2755 if (f2fs_need_rand_seg(sbi)) 2756 return get_random_u32_below(MAIN_SECS(sbi) * sbi->segs_per_sec); 2757 2758 /* if segs_per_sec is large than 1, we need to keep original policy. */ 2759 if (__is_large_section(sbi)) 2760 return curseg->segno; 2761 2762 /* inmem log may not locate on any segment after mount */ 2763 if (!curseg->inited) 2764 return 0; 2765 2766 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 2767 return 0; 2768 2769 if (test_opt(sbi, NOHEAP) && 2770 (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type))) 2771 return 0; 2772 2773 if (SIT_I(sbi)->last_victim[ALLOC_NEXT]) 2774 return SIT_I(sbi)->last_victim[ALLOC_NEXT]; 2775 2776 /* find segments from 0 to reuse freed segments */ 2777 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE) 2778 return 0; 2779 2780 return curseg->segno; 2781 } 2782 2783 /* 2784 * Allocate a current working segment. 2785 * This function always allocates a free segment in LFS manner. 2786 */ 2787 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 2788 { 2789 struct curseg_info *curseg = CURSEG_I(sbi, type); 2790 unsigned short seg_type = curseg->seg_type; 2791 unsigned int segno = curseg->segno; 2792 int dir = ALLOC_LEFT; 2793 2794 if (curseg->inited) 2795 write_sum_page(sbi, curseg->sum_blk, 2796 GET_SUM_BLOCK(sbi, segno)); 2797 if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA) 2798 dir = ALLOC_RIGHT; 2799 2800 if (test_opt(sbi, NOHEAP)) 2801 dir = ALLOC_RIGHT; 2802 2803 segno = __get_next_segno(sbi, type); 2804 get_new_segment(sbi, &segno, new_sec, dir); 2805 curseg->next_segno = segno; 2806 reset_curseg(sbi, type, 1); 2807 curseg->alloc_type = LFS; 2808 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) 2809 curseg->fragment_remained_chunk = 2810 get_random_u32_inclusive(1, sbi->max_fragment_chunk); 2811 } 2812 2813 static int __next_free_blkoff(struct f2fs_sb_info *sbi, 2814 int segno, block_t start) 2815 { 2816 struct seg_entry *se = get_seg_entry(sbi, segno); 2817 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 2818 unsigned long *target_map = SIT_I(sbi)->tmp_map; 2819 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 2820 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 2821 int i; 2822 2823 for (i = 0; i < entries; i++) 2824 target_map[i] = ckpt_map[i] | cur_map[i]; 2825 2826 return __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); 2827 } 2828 2829 static int f2fs_find_next_ssr_block(struct f2fs_sb_info *sbi, 2830 struct curseg_info *seg) 2831 { 2832 return __next_free_blkoff(sbi, seg->segno, seg->next_blkoff + 1); 2833 } 2834 2835 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno) 2836 { 2837 return __next_free_blkoff(sbi, segno, 0) < sbi->blocks_per_seg; 2838 } 2839 2840 /* 2841 * This function always allocates a used segment(from dirty seglist) by SSR 2842 * manner, so it should recover the existing segment information of valid blocks 2843 */ 2844 static void change_curseg(struct f2fs_sb_info *sbi, int type) 2845 { 2846 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2847 struct curseg_info *curseg = CURSEG_I(sbi, type); 2848 unsigned int new_segno = curseg->next_segno; 2849 struct f2fs_summary_block *sum_node; 2850 struct page *sum_page; 2851 2852 write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, curseg->segno)); 2853 2854 __set_test_and_inuse(sbi, new_segno); 2855 2856 mutex_lock(&dirty_i->seglist_lock); 2857 __remove_dirty_segment(sbi, new_segno, PRE); 2858 __remove_dirty_segment(sbi, new_segno, DIRTY); 2859 mutex_unlock(&dirty_i->seglist_lock); 2860 2861 reset_curseg(sbi, type, 1); 2862 curseg->alloc_type = SSR; 2863 curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0); 2864 2865 sum_page = f2fs_get_sum_page(sbi, new_segno); 2866 if (IS_ERR(sum_page)) { 2867 /* GC won't be able to use stale summary pages by cp_error */ 2868 memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE); 2869 return; 2870 } 2871 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 2872 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); 2873 f2fs_put_page(sum_page, 1); 2874 } 2875 2876 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type, 2877 int alloc_mode, unsigned long long age); 2878 2879 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type, 2880 int target_type, int alloc_mode, 2881 unsigned long long age) 2882 { 2883 struct curseg_info *curseg = CURSEG_I(sbi, type); 2884 2885 curseg->seg_type = target_type; 2886 2887 if (get_ssr_segment(sbi, type, alloc_mode, age)) { 2888 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno); 2889 2890 curseg->seg_type = se->type; 2891 change_curseg(sbi, type); 2892 } else { 2893 /* allocate cold segment by default */ 2894 curseg->seg_type = CURSEG_COLD_DATA; 2895 new_curseg(sbi, type, true); 2896 } 2897 stat_inc_seg_type(sbi, curseg); 2898 } 2899 2900 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi) 2901 { 2902 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC); 2903 2904 if (!sbi->am.atgc_enabled) 2905 return; 2906 2907 f2fs_down_read(&SM_I(sbi)->curseg_lock); 2908 2909 mutex_lock(&curseg->curseg_mutex); 2910 down_write(&SIT_I(sbi)->sentry_lock); 2911 2912 get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0); 2913 2914 up_write(&SIT_I(sbi)->sentry_lock); 2915 mutex_unlock(&curseg->curseg_mutex); 2916 2917 f2fs_up_read(&SM_I(sbi)->curseg_lock); 2918 2919 } 2920 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi) 2921 { 2922 __f2fs_init_atgc_curseg(sbi); 2923 } 2924 2925 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type) 2926 { 2927 struct curseg_info *curseg = CURSEG_I(sbi, type); 2928 2929 mutex_lock(&curseg->curseg_mutex); 2930 if (!curseg->inited) 2931 goto out; 2932 2933 if (get_valid_blocks(sbi, curseg->segno, false)) { 2934 write_sum_page(sbi, curseg->sum_blk, 2935 GET_SUM_BLOCK(sbi, curseg->segno)); 2936 } else { 2937 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 2938 __set_test_and_free(sbi, curseg->segno, true); 2939 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 2940 } 2941 out: 2942 mutex_unlock(&curseg->curseg_mutex); 2943 } 2944 2945 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi) 2946 { 2947 __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED); 2948 2949 if (sbi->am.atgc_enabled) 2950 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC); 2951 } 2952 2953 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type) 2954 { 2955 struct curseg_info *curseg = CURSEG_I(sbi, type); 2956 2957 mutex_lock(&curseg->curseg_mutex); 2958 if (!curseg->inited) 2959 goto out; 2960 if (get_valid_blocks(sbi, curseg->segno, false)) 2961 goto out; 2962 2963 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 2964 __set_test_and_inuse(sbi, curseg->segno); 2965 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 2966 out: 2967 mutex_unlock(&curseg->curseg_mutex); 2968 } 2969 2970 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi) 2971 { 2972 __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED); 2973 2974 if (sbi->am.atgc_enabled) 2975 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC); 2976 } 2977 2978 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type, 2979 int alloc_mode, unsigned long long age) 2980 { 2981 struct curseg_info *curseg = CURSEG_I(sbi, type); 2982 unsigned segno = NULL_SEGNO; 2983 unsigned short seg_type = curseg->seg_type; 2984 int i, cnt; 2985 bool reversed = false; 2986 2987 sanity_check_seg_type(sbi, seg_type); 2988 2989 /* f2fs_need_SSR() already forces to do this */ 2990 if (!f2fs_get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) { 2991 curseg->next_segno = segno; 2992 return 1; 2993 } 2994 2995 /* For node segments, let's do SSR more intensively */ 2996 if (IS_NODESEG(seg_type)) { 2997 if (seg_type >= CURSEG_WARM_NODE) { 2998 reversed = true; 2999 i = CURSEG_COLD_NODE; 3000 } else { 3001 i = CURSEG_HOT_NODE; 3002 } 3003 cnt = NR_CURSEG_NODE_TYPE; 3004 } else { 3005 if (seg_type >= CURSEG_WARM_DATA) { 3006 reversed = true; 3007 i = CURSEG_COLD_DATA; 3008 } else { 3009 i = CURSEG_HOT_DATA; 3010 } 3011 cnt = NR_CURSEG_DATA_TYPE; 3012 } 3013 3014 for (; cnt-- > 0; reversed ? i-- : i++) { 3015 if (i == seg_type) 3016 continue; 3017 if (!f2fs_get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) { 3018 curseg->next_segno = segno; 3019 return 1; 3020 } 3021 } 3022 3023 /* find valid_blocks=0 in dirty list */ 3024 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 3025 segno = get_free_segment(sbi); 3026 if (segno != NULL_SEGNO) { 3027 curseg->next_segno = segno; 3028 return 1; 3029 } 3030 } 3031 return 0; 3032 } 3033 3034 static bool need_new_seg(struct f2fs_sb_info *sbi, int type) 3035 { 3036 struct curseg_info *curseg = CURSEG_I(sbi, type); 3037 3038 if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) && 3039 curseg->seg_type == CURSEG_WARM_NODE) 3040 return true; 3041 if (curseg->alloc_type == LFS && 3042 is_next_segment_free(sbi, curseg, type) && 3043 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 3044 return true; 3045 if (!f2fs_need_SSR(sbi) || !get_ssr_segment(sbi, type, SSR, 0)) 3046 return true; 3047 return false; 3048 } 3049 3050 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type, 3051 unsigned int start, unsigned int end) 3052 { 3053 struct curseg_info *curseg = CURSEG_I(sbi, type); 3054 unsigned int segno; 3055 3056 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3057 mutex_lock(&curseg->curseg_mutex); 3058 down_write(&SIT_I(sbi)->sentry_lock); 3059 3060 segno = CURSEG_I(sbi, type)->segno; 3061 if (segno < start || segno > end) 3062 goto unlock; 3063 3064 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0)) 3065 change_curseg(sbi, type); 3066 else 3067 new_curseg(sbi, type, true); 3068 3069 stat_inc_seg_type(sbi, curseg); 3070 3071 locate_dirty_segment(sbi, segno); 3072 unlock: 3073 up_write(&SIT_I(sbi)->sentry_lock); 3074 3075 if (segno != curseg->segno) 3076 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u", 3077 type, segno, curseg->segno); 3078 3079 mutex_unlock(&curseg->curseg_mutex); 3080 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3081 } 3082 3083 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type, 3084 bool new_sec, bool force) 3085 { 3086 struct curseg_info *curseg = CURSEG_I(sbi, type); 3087 unsigned int old_segno; 3088 3089 if (!force && curseg->inited && 3090 !curseg->next_blkoff && 3091 !get_valid_blocks(sbi, curseg->segno, new_sec) && 3092 !get_ckpt_valid_blocks(sbi, curseg->segno, new_sec)) 3093 return; 3094 3095 old_segno = curseg->segno; 3096 new_curseg(sbi, type, true); 3097 stat_inc_seg_type(sbi, curseg); 3098 locate_dirty_segment(sbi, old_segno); 3099 } 3100 3101 void f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force) 3102 { 3103 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3104 down_write(&SIT_I(sbi)->sentry_lock); 3105 __allocate_new_segment(sbi, type, true, force); 3106 up_write(&SIT_I(sbi)->sentry_lock); 3107 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3108 } 3109 3110 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi) 3111 { 3112 int i; 3113 3114 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3115 down_write(&SIT_I(sbi)->sentry_lock); 3116 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) 3117 __allocate_new_segment(sbi, i, false, false); 3118 up_write(&SIT_I(sbi)->sentry_lock); 3119 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3120 } 3121 3122 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi, 3123 struct cp_control *cpc) 3124 { 3125 __u64 trim_start = cpc->trim_start; 3126 bool has_candidate = false; 3127 3128 down_write(&SIT_I(sbi)->sentry_lock); 3129 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) { 3130 if (add_discard_addrs(sbi, cpc, true)) { 3131 has_candidate = true; 3132 break; 3133 } 3134 } 3135 up_write(&SIT_I(sbi)->sentry_lock); 3136 3137 cpc->trim_start = trim_start; 3138 return has_candidate; 3139 } 3140 3141 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi, 3142 struct discard_policy *dpolicy, 3143 unsigned int start, unsigned int end) 3144 { 3145 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 3146 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 3147 struct rb_node **insert_p = NULL, *insert_parent = NULL; 3148 struct discard_cmd *dc; 3149 struct blk_plug plug; 3150 int issued; 3151 unsigned int trimmed = 0; 3152 3153 next: 3154 issued = 0; 3155 3156 mutex_lock(&dcc->cmd_lock); 3157 if (unlikely(dcc->rbtree_check)) 3158 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi)); 3159 3160 dc = __lookup_discard_cmd_ret(&dcc->root, start, 3161 &prev_dc, &next_dc, &insert_p, &insert_parent); 3162 if (!dc) 3163 dc = next_dc; 3164 3165 blk_start_plug(&plug); 3166 3167 while (dc && dc->di.lstart <= end) { 3168 struct rb_node *node; 3169 int err = 0; 3170 3171 if (dc->di.len < dpolicy->granularity) 3172 goto skip; 3173 3174 if (dc->state != D_PREP) { 3175 list_move_tail(&dc->list, &dcc->fstrim_list); 3176 goto skip; 3177 } 3178 3179 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued); 3180 3181 if (issued >= dpolicy->max_requests) { 3182 start = dc->di.lstart + dc->di.len; 3183 3184 if (err) 3185 __remove_discard_cmd(sbi, dc); 3186 3187 blk_finish_plug(&plug); 3188 mutex_unlock(&dcc->cmd_lock); 3189 trimmed += __wait_all_discard_cmd(sbi, NULL); 3190 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 3191 goto next; 3192 } 3193 skip: 3194 node = rb_next(&dc->rb_node); 3195 if (err) 3196 __remove_discard_cmd(sbi, dc); 3197 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 3198 3199 if (fatal_signal_pending(current)) 3200 break; 3201 } 3202 3203 blk_finish_plug(&plug); 3204 mutex_unlock(&dcc->cmd_lock); 3205 3206 return trimmed; 3207 } 3208 3209 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) 3210 { 3211 __u64 start = F2FS_BYTES_TO_BLK(range->start); 3212 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; 3213 unsigned int start_segno, end_segno; 3214 block_t start_block, end_block; 3215 struct cp_control cpc; 3216 struct discard_policy dpolicy; 3217 unsigned long long trimmed = 0; 3218 int err = 0; 3219 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi); 3220 3221 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize) 3222 return -EINVAL; 3223 3224 if (end < MAIN_BLKADDR(sbi)) 3225 goto out; 3226 3227 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 3228 f2fs_warn(sbi, "Found FS corruption, run fsck to fix."); 3229 return -EFSCORRUPTED; 3230 } 3231 3232 /* start/end segment number in main_area */ 3233 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); 3234 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : 3235 GET_SEGNO(sbi, end); 3236 if (need_align) { 3237 start_segno = rounddown(start_segno, sbi->segs_per_sec); 3238 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1; 3239 } 3240 3241 cpc.reason = CP_DISCARD; 3242 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen)); 3243 cpc.trim_start = start_segno; 3244 cpc.trim_end = end_segno; 3245 3246 if (sbi->discard_blks == 0) 3247 goto out; 3248 3249 f2fs_down_write(&sbi->gc_lock); 3250 err = f2fs_write_checkpoint(sbi, &cpc); 3251 f2fs_up_write(&sbi->gc_lock); 3252 if (err) 3253 goto out; 3254 3255 /* 3256 * We filed discard candidates, but actually we don't need to wait for 3257 * all of them, since they'll be issued in idle time along with runtime 3258 * discard option. User configuration looks like using runtime discard 3259 * or periodic fstrim instead of it. 3260 */ 3261 if (f2fs_realtime_discard_enable(sbi)) 3262 goto out; 3263 3264 start_block = START_BLOCK(sbi, start_segno); 3265 end_block = START_BLOCK(sbi, end_segno + 1); 3266 3267 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen); 3268 trimmed = __issue_discard_cmd_range(sbi, &dpolicy, 3269 start_block, end_block); 3270 3271 trimmed += __wait_discard_cmd_range(sbi, &dpolicy, 3272 start_block, end_block); 3273 out: 3274 if (!err) 3275 range->len = F2FS_BLK_TO_BYTES(trimmed); 3276 return err; 3277 } 3278 3279 int f2fs_rw_hint_to_seg_type(enum rw_hint hint) 3280 { 3281 switch (hint) { 3282 case WRITE_LIFE_SHORT: 3283 return CURSEG_HOT_DATA; 3284 case WRITE_LIFE_EXTREME: 3285 return CURSEG_COLD_DATA; 3286 default: 3287 return CURSEG_WARM_DATA; 3288 } 3289 } 3290 3291 static int __get_segment_type_2(struct f2fs_io_info *fio) 3292 { 3293 if (fio->type == DATA) 3294 return CURSEG_HOT_DATA; 3295 else 3296 return CURSEG_HOT_NODE; 3297 } 3298 3299 static int __get_segment_type_4(struct f2fs_io_info *fio) 3300 { 3301 if (fio->type == DATA) { 3302 struct inode *inode = fio->page->mapping->host; 3303 3304 if (S_ISDIR(inode->i_mode)) 3305 return CURSEG_HOT_DATA; 3306 else 3307 return CURSEG_COLD_DATA; 3308 } else { 3309 if (IS_DNODE(fio->page) && is_cold_node(fio->page)) 3310 return CURSEG_WARM_NODE; 3311 else 3312 return CURSEG_COLD_NODE; 3313 } 3314 } 3315 3316 static int __get_age_segment_type(struct inode *inode, pgoff_t pgofs) 3317 { 3318 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3319 struct extent_info ei = {}; 3320 3321 if (f2fs_lookup_age_extent_cache(inode, pgofs, &ei)) { 3322 if (!ei.age) 3323 return NO_CHECK_TYPE; 3324 if (ei.age <= sbi->hot_data_age_threshold) 3325 return CURSEG_HOT_DATA; 3326 if (ei.age <= sbi->warm_data_age_threshold) 3327 return CURSEG_WARM_DATA; 3328 return CURSEG_COLD_DATA; 3329 } 3330 return NO_CHECK_TYPE; 3331 } 3332 3333 static int __get_segment_type_6(struct f2fs_io_info *fio) 3334 { 3335 if (fio->type == DATA) { 3336 struct inode *inode = fio->page->mapping->host; 3337 int type; 3338 3339 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE)) 3340 return CURSEG_COLD_DATA_PINNED; 3341 3342 if (page_private_gcing(fio->page)) { 3343 if (fio->sbi->am.atgc_enabled && 3344 (fio->io_type == FS_DATA_IO) && 3345 (fio->sbi->gc_mode != GC_URGENT_HIGH)) 3346 return CURSEG_ALL_DATA_ATGC; 3347 else 3348 return CURSEG_COLD_DATA; 3349 } 3350 if (file_is_cold(inode) || f2fs_need_compress_data(inode)) 3351 return CURSEG_COLD_DATA; 3352 3353 type = __get_age_segment_type(inode, fio->page->index); 3354 if (type != NO_CHECK_TYPE) 3355 return type; 3356 3357 if (file_is_hot(inode) || 3358 is_inode_flag_set(inode, FI_HOT_DATA) || 3359 f2fs_is_cow_file(inode)) 3360 return CURSEG_HOT_DATA; 3361 return f2fs_rw_hint_to_seg_type(inode->i_write_hint); 3362 } else { 3363 if (IS_DNODE(fio->page)) 3364 return is_cold_node(fio->page) ? CURSEG_WARM_NODE : 3365 CURSEG_HOT_NODE; 3366 return CURSEG_COLD_NODE; 3367 } 3368 } 3369 3370 static int __get_segment_type(struct f2fs_io_info *fio) 3371 { 3372 int type = 0; 3373 3374 switch (F2FS_OPTION(fio->sbi).active_logs) { 3375 case 2: 3376 type = __get_segment_type_2(fio); 3377 break; 3378 case 4: 3379 type = __get_segment_type_4(fio); 3380 break; 3381 case 6: 3382 type = __get_segment_type_6(fio); 3383 break; 3384 default: 3385 f2fs_bug_on(fio->sbi, true); 3386 } 3387 3388 if (IS_HOT(type)) 3389 fio->temp = HOT; 3390 else if (IS_WARM(type)) 3391 fio->temp = WARM; 3392 else 3393 fio->temp = COLD; 3394 return type; 3395 } 3396 3397 static void f2fs_randomize_chunk(struct f2fs_sb_info *sbi, 3398 struct curseg_info *seg) 3399 { 3400 /* To allocate block chunks in different sizes, use random number */ 3401 if (--seg->fragment_remained_chunk > 0) 3402 return; 3403 3404 seg->fragment_remained_chunk = 3405 get_random_u32_inclusive(1, sbi->max_fragment_chunk); 3406 seg->next_blkoff += 3407 get_random_u32_inclusive(1, sbi->max_fragment_hole); 3408 } 3409 3410 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, 3411 block_t old_blkaddr, block_t *new_blkaddr, 3412 struct f2fs_summary *sum, int type, 3413 struct f2fs_io_info *fio) 3414 { 3415 struct sit_info *sit_i = SIT_I(sbi); 3416 struct curseg_info *curseg = CURSEG_I(sbi, type); 3417 unsigned long long old_mtime; 3418 bool from_gc = (type == CURSEG_ALL_DATA_ATGC); 3419 struct seg_entry *se = NULL; 3420 bool segment_full = false; 3421 3422 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3423 3424 mutex_lock(&curseg->curseg_mutex); 3425 down_write(&sit_i->sentry_lock); 3426 3427 if (from_gc) { 3428 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO); 3429 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr)); 3430 sanity_check_seg_type(sbi, se->type); 3431 f2fs_bug_on(sbi, IS_NODESEG(se->type)); 3432 } 3433 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 3434 3435 f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg); 3436 3437 f2fs_wait_discard_bio(sbi, *new_blkaddr); 3438 3439 curseg->sum_blk->entries[curseg->next_blkoff] = *sum; 3440 if (curseg->alloc_type == SSR) { 3441 curseg->next_blkoff = f2fs_find_next_ssr_block(sbi, curseg); 3442 } else { 3443 curseg->next_blkoff++; 3444 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) 3445 f2fs_randomize_chunk(sbi, curseg); 3446 } 3447 if (curseg->next_blkoff >= f2fs_usable_blks_in_seg(sbi, curseg->segno)) 3448 segment_full = true; 3449 stat_inc_block_count(sbi, curseg); 3450 3451 if (from_gc) { 3452 old_mtime = get_segment_mtime(sbi, old_blkaddr); 3453 } else { 3454 update_segment_mtime(sbi, old_blkaddr, 0); 3455 old_mtime = 0; 3456 } 3457 update_segment_mtime(sbi, *new_blkaddr, old_mtime); 3458 3459 /* 3460 * SIT information should be updated before segment allocation, 3461 * since SSR needs latest valid block information. 3462 */ 3463 update_sit_entry(sbi, *new_blkaddr, 1); 3464 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 3465 update_sit_entry(sbi, old_blkaddr, -1); 3466 3467 /* 3468 * If the current segment is full, flush it out and replace it with a 3469 * new segment. 3470 */ 3471 if (segment_full) { 3472 if (from_gc) { 3473 get_atssr_segment(sbi, type, se->type, 3474 AT_SSR, se->mtime); 3475 } else { 3476 if (need_new_seg(sbi, type)) 3477 new_curseg(sbi, type, false); 3478 else 3479 change_curseg(sbi, type); 3480 stat_inc_seg_type(sbi, curseg); 3481 } 3482 } 3483 /* 3484 * segment dirty status should be updated after segment allocation, 3485 * so we just need to update status only one time after previous 3486 * segment being closed. 3487 */ 3488 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3489 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr)); 3490 3491 if (IS_DATASEG(type)) 3492 atomic64_inc(&sbi->allocated_data_blocks); 3493 3494 up_write(&sit_i->sentry_lock); 3495 3496 if (page && IS_NODESEG(type)) { 3497 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); 3498 3499 f2fs_inode_chksum_set(sbi, page); 3500 } 3501 3502 if (fio) { 3503 struct f2fs_bio_info *io; 3504 3505 if (F2FS_IO_ALIGNED(sbi)) 3506 fio->retry = 0; 3507 3508 INIT_LIST_HEAD(&fio->list); 3509 fio->in_list = 1; 3510 io = sbi->write_io[fio->type] + fio->temp; 3511 spin_lock(&io->io_lock); 3512 list_add_tail(&fio->list, &io->io_list); 3513 spin_unlock(&io->io_lock); 3514 } 3515 3516 mutex_unlock(&curseg->curseg_mutex); 3517 3518 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3519 } 3520 3521 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino, 3522 block_t blkaddr, unsigned int blkcnt) 3523 { 3524 if (!f2fs_is_multi_device(sbi)) 3525 return; 3526 3527 while (1) { 3528 unsigned int devidx = f2fs_target_device_index(sbi, blkaddr); 3529 unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1; 3530 3531 /* update device state for fsync */ 3532 f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO); 3533 3534 /* update device state for checkpoint */ 3535 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) { 3536 spin_lock(&sbi->dev_lock); 3537 f2fs_set_bit(devidx, (char *)&sbi->dirty_device); 3538 spin_unlock(&sbi->dev_lock); 3539 } 3540 3541 if (blkcnt <= blks) 3542 break; 3543 blkcnt -= blks; 3544 blkaddr += blks; 3545 } 3546 } 3547 3548 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) 3549 { 3550 int type = __get_segment_type(fio); 3551 bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA); 3552 3553 if (keep_order) 3554 f2fs_down_read(&fio->sbi->io_order_lock); 3555 reallocate: 3556 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr, 3557 &fio->new_blkaddr, sum, type, fio); 3558 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO) { 3559 invalidate_mapping_pages(META_MAPPING(fio->sbi), 3560 fio->old_blkaddr, fio->old_blkaddr); 3561 f2fs_invalidate_compress_page(fio->sbi, fio->old_blkaddr); 3562 } 3563 3564 /* writeout dirty page into bdev */ 3565 f2fs_submit_page_write(fio); 3566 if (fio->retry) { 3567 fio->old_blkaddr = fio->new_blkaddr; 3568 goto reallocate; 3569 } 3570 3571 f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1); 3572 3573 if (keep_order) 3574 f2fs_up_read(&fio->sbi->io_order_lock); 3575 } 3576 3577 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page, 3578 enum iostat_type io_type) 3579 { 3580 struct f2fs_io_info fio = { 3581 .sbi = sbi, 3582 .type = META, 3583 .temp = HOT, 3584 .op = REQ_OP_WRITE, 3585 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO, 3586 .old_blkaddr = page->index, 3587 .new_blkaddr = page->index, 3588 .page = page, 3589 .encrypted_page = NULL, 3590 .in_list = 0, 3591 }; 3592 3593 if (unlikely(page->index >= MAIN_BLKADDR(sbi))) 3594 fio.op_flags &= ~REQ_META; 3595 3596 set_page_writeback(page); 3597 f2fs_submit_page_write(&fio); 3598 3599 stat_inc_meta_count(sbi, page->index); 3600 f2fs_update_iostat(sbi, NULL, io_type, F2FS_BLKSIZE); 3601 } 3602 3603 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio) 3604 { 3605 struct f2fs_summary sum; 3606 3607 set_summary(&sum, nid, 0, 0); 3608 do_write_page(&sum, fio); 3609 3610 f2fs_update_iostat(fio->sbi, NULL, fio->io_type, F2FS_BLKSIZE); 3611 } 3612 3613 void f2fs_outplace_write_data(struct dnode_of_data *dn, 3614 struct f2fs_io_info *fio) 3615 { 3616 struct f2fs_sb_info *sbi = fio->sbi; 3617 struct f2fs_summary sum; 3618 3619 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); 3620 if (fio->io_type == FS_DATA_IO || fio->io_type == FS_CP_DATA_IO) 3621 f2fs_update_age_extent_cache(dn); 3622 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version); 3623 do_write_page(&sum, fio); 3624 f2fs_update_data_blkaddr(dn, fio->new_blkaddr); 3625 3626 f2fs_update_iostat(sbi, dn->inode, fio->io_type, F2FS_BLKSIZE); 3627 } 3628 3629 int f2fs_inplace_write_data(struct f2fs_io_info *fio) 3630 { 3631 int err; 3632 struct f2fs_sb_info *sbi = fio->sbi; 3633 unsigned int segno; 3634 3635 fio->new_blkaddr = fio->old_blkaddr; 3636 /* i/o temperature is needed for passing down write hints */ 3637 __get_segment_type(fio); 3638 3639 segno = GET_SEGNO(sbi, fio->new_blkaddr); 3640 3641 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) { 3642 set_sbi_flag(sbi, SBI_NEED_FSCK); 3643 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.", 3644 __func__, segno); 3645 err = -EFSCORRUPTED; 3646 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE); 3647 goto drop_bio; 3648 } 3649 3650 if (f2fs_cp_error(sbi)) { 3651 err = -EIO; 3652 goto drop_bio; 3653 } 3654 3655 if (fio->post_read) 3656 invalidate_mapping_pages(META_MAPPING(sbi), 3657 fio->new_blkaddr, fio->new_blkaddr); 3658 3659 stat_inc_inplace_blocks(fio->sbi); 3660 3661 if (fio->bio && !IS_F2FS_IPU_NOCACHE(sbi)) 3662 err = f2fs_merge_page_bio(fio); 3663 else 3664 err = f2fs_submit_page_bio(fio); 3665 if (!err) { 3666 f2fs_update_device_state(fio->sbi, fio->ino, 3667 fio->new_blkaddr, 1); 3668 f2fs_update_iostat(fio->sbi, fio->page->mapping->host, 3669 fio->io_type, F2FS_BLKSIZE); 3670 } 3671 3672 return err; 3673 drop_bio: 3674 if (fio->bio && *(fio->bio)) { 3675 struct bio *bio = *(fio->bio); 3676 3677 bio->bi_status = BLK_STS_IOERR; 3678 bio_endio(bio); 3679 *(fio->bio) = NULL; 3680 } 3681 return err; 3682 } 3683 3684 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi, 3685 unsigned int segno) 3686 { 3687 int i; 3688 3689 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 3690 if (CURSEG_I(sbi, i)->segno == segno) 3691 break; 3692 } 3693 return i; 3694 } 3695 3696 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 3697 block_t old_blkaddr, block_t new_blkaddr, 3698 bool recover_curseg, bool recover_newaddr, 3699 bool from_gc) 3700 { 3701 struct sit_info *sit_i = SIT_I(sbi); 3702 struct curseg_info *curseg; 3703 unsigned int segno, old_cursegno; 3704 struct seg_entry *se; 3705 int type; 3706 unsigned short old_blkoff; 3707 unsigned char old_alloc_type; 3708 3709 segno = GET_SEGNO(sbi, new_blkaddr); 3710 se = get_seg_entry(sbi, segno); 3711 type = se->type; 3712 3713 f2fs_down_write(&SM_I(sbi)->curseg_lock); 3714 3715 if (!recover_curseg) { 3716 /* for recovery flow */ 3717 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { 3718 if (old_blkaddr == NULL_ADDR) 3719 type = CURSEG_COLD_DATA; 3720 else 3721 type = CURSEG_WARM_DATA; 3722 } 3723 } else { 3724 if (IS_CURSEG(sbi, segno)) { 3725 /* se->type is volatile as SSR allocation */ 3726 type = __f2fs_get_curseg(sbi, segno); 3727 f2fs_bug_on(sbi, type == NO_CHECK_TYPE); 3728 } else { 3729 type = CURSEG_WARM_DATA; 3730 } 3731 } 3732 3733 f2fs_bug_on(sbi, !IS_DATASEG(type)); 3734 curseg = CURSEG_I(sbi, type); 3735 3736 mutex_lock(&curseg->curseg_mutex); 3737 down_write(&sit_i->sentry_lock); 3738 3739 old_cursegno = curseg->segno; 3740 old_blkoff = curseg->next_blkoff; 3741 old_alloc_type = curseg->alloc_type; 3742 3743 /* change the current segment */ 3744 if (segno != curseg->segno) { 3745 curseg->next_segno = segno; 3746 change_curseg(sbi, type); 3747 } 3748 3749 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); 3750 curseg->sum_blk->entries[curseg->next_blkoff] = *sum; 3751 3752 if (!recover_curseg || recover_newaddr) { 3753 if (!from_gc) 3754 update_segment_mtime(sbi, new_blkaddr, 0); 3755 update_sit_entry(sbi, new_blkaddr, 1); 3756 } 3757 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) { 3758 invalidate_mapping_pages(META_MAPPING(sbi), 3759 old_blkaddr, old_blkaddr); 3760 f2fs_invalidate_compress_page(sbi, old_blkaddr); 3761 if (!from_gc) 3762 update_segment_mtime(sbi, old_blkaddr, 0); 3763 update_sit_entry(sbi, old_blkaddr, -1); 3764 } 3765 3766 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3767 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); 3768 3769 locate_dirty_segment(sbi, old_cursegno); 3770 3771 if (recover_curseg) { 3772 if (old_cursegno != curseg->segno) { 3773 curseg->next_segno = old_cursegno; 3774 change_curseg(sbi, type); 3775 } 3776 curseg->next_blkoff = old_blkoff; 3777 curseg->alloc_type = old_alloc_type; 3778 } 3779 3780 up_write(&sit_i->sentry_lock); 3781 mutex_unlock(&curseg->curseg_mutex); 3782 f2fs_up_write(&SM_I(sbi)->curseg_lock); 3783 } 3784 3785 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, 3786 block_t old_addr, block_t new_addr, 3787 unsigned char version, bool recover_curseg, 3788 bool recover_newaddr) 3789 { 3790 struct f2fs_summary sum; 3791 3792 set_summary(&sum, dn->nid, dn->ofs_in_node, version); 3793 3794 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr, 3795 recover_curseg, recover_newaddr, false); 3796 3797 f2fs_update_data_blkaddr(dn, new_addr); 3798 } 3799 3800 void f2fs_wait_on_page_writeback(struct page *page, 3801 enum page_type type, bool ordered, bool locked) 3802 { 3803 if (PageWriteback(page)) { 3804 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 3805 3806 /* submit cached LFS IO */ 3807 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type); 3808 /* submit cached IPU IO */ 3809 f2fs_submit_merged_ipu_write(sbi, NULL, page); 3810 if (ordered) { 3811 wait_on_page_writeback(page); 3812 f2fs_bug_on(sbi, locked && PageWriteback(page)); 3813 } else { 3814 wait_for_stable_page(page); 3815 } 3816 } 3817 } 3818 3819 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr) 3820 { 3821 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3822 struct page *cpage; 3823 3824 if (!f2fs_post_read_required(inode)) 3825 return; 3826 3827 if (!__is_valid_data_blkaddr(blkaddr)) 3828 return; 3829 3830 cpage = find_lock_page(META_MAPPING(sbi), blkaddr); 3831 if (cpage) { 3832 f2fs_wait_on_page_writeback(cpage, DATA, true, true); 3833 f2fs_put_page(cpage, 1); 3834 } 3835 } 3836 3837 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr, 3838 block_t len) 3839 { 3840 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3841 block_t i; 3842 3843 if (!f2fs_post_read_required(inode)) 3844 return; 3845 3846 for (i = 0; i < len; i++) 3847 f2fs_wait_on_block_writeback(inode, blkaddr + i); 3848 3849 invalidate_mapping_pages(META_MAPPING(sbi), blkaddr, blkaddr + len - 1); 3850 } 3851 3852 static int read_compacted_summaries(struct f2fs_sb_info *sbi) 3853 { 3854 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3855 struct curseg_info *seg_i; 3856 unsigned char *kaddr; 3857 struct page *page; 3858 block_t start; 3859 int i, j, offset; 3860 3861 start = start_sum_block(sbi); 3862 3863 page = f2fs_get_meta_page(sbi, start++); 3864 if (IS_ERR(page)) 3865 return PTR_ERR(page); 3866 kaddr = (unsigned char *)page_address(page); 3867 3868 /* Step 1: restore nat cache */ 3869 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 3870 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE); 3871 3872 /* Step 2: restore sit cache */ 3873 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 3874 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE); 3875 offset = 2 * SUM_JOURNAL_SIZE; 3876 3877 /* Step 3: restore summary entries */ 3878 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 3879 unsigned short blk_off; 3880 unsigned int segno; 3881 3882 seg_i = CURSEG_I(sbi, i); 3883 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 3884 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 3885 seg_i->next_segno = segno; 3886 reset_curseg(sbi, i, 0); 3887 seg_i->alloc_type = ckpt->alloc_type[i]; 3888 seg_i->next_blkoff = blk_off; 3889 3890 if (seg_i->alloc_type == SSR) 3891 blk_off = sbi->blocks_per_seg; 3892 3893 for (j = 0; j < blk_off; j++) { 3894 struct f2fs_summary *s; 3895 3896 s = (struct f2fs_summary *)(kaddr + offset); 3897 seg_i->sum_blk->entries[j] = *s; 3898 offset += SUMMARY_SIZE; 3899 if (offset + SUMMARY_SIZE <= PAGE_SIZE - 3900 SUM_FOOTER_SIZE) 3901 continue; 3902 3903 f2fs_put_page(page, 1); 3904 page = NULL; 3905 3906 page = f2fs_get_meta_page(sbi, start++); 3907 if (IS_ERR(page)) 3908 return PTR_ERR(page); 3909 kaddr = (unsigned char *)page_address(page); 3910 offset = 0; 3911 } 3912 } 3913 f2fs_put_page(page, 1); 3914 return 0; 3915 } 3916 3917 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 3918 { 3919 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3920 struct f2fs_summary_block *sum; 3921 struct curseg_info *curseg; 3922 struct page *new; 3923 unsigned short blk_off; 3924 unsigned int segno = 0; 3925 block_t blk_addr = 0; 3926 int err = 0; 3927 3928 /* get segment number and block addr */ 3929 if (IS_DATASEG(type)) { 3930 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 3931 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 3932 CURSEG_HOT_DATA]); 3933 if (__exist_node_summaries(sbi)) 3934 blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type); 3935 else 3936 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 3937 } else { 3938 segno = le32_to_cpu(ckpt->cur_node_segno[type - 3939 CURSEG_HOT_NODE]); 3940 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 3941 CURSEG_HOT_NODE]); 3942 if (__exist_node_summaries(sbi)) 3943 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 3944 type - CURSEG_HOT_NODE); 3945 else 3946 blk_addr = GET_SUM_BLOCK(sbi, segno); 3947 } 3948 3949 new = f2fs_get_meta_page(sbi, blk_addr); 3950 if (IS_ERR(new)) 3951 return PTR_ERR(new); 3952 sum = (struct f2fs_summary_block *)page_address(new); 3953 3954 if (IS_NODESEG(type)) { 3955 if (__exist_node_summaries(sbi)) { 3956 struct f2fs_summary *ns = &sum->entries[0]; 3957 int i; 3958 3959 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { 3960 ns->version = 0; 3961 ns->ofs_in_node = 0; 3962 } 3963 } else { 3964 err = f2fs_restore_node_summary(sbi, segno, sum); 3965 if (err) 3966 goto out; 3967 } 3968 } 3969 3970 /* set uncompleted segment to curseg */ 3971 curseg = CURSEG_I(sbi, type); 3972 mutex_lock(&curseg->curseg_mutex); 3973 3974 /* update journal info */ 3975 down_write(&curseg->journal_rwsem); 3976 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE); 3977 up_write(&curseg->journal_rwsem); 3978 3979 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE); 3980 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE); 3981 curseg->next_segno = segno; 3982 reset_curseg(sbi, type, 0); 3983 curseg->alloc_type = ckpt->alloc_type[type]; 3984 curseg->next_blkoff = blk_off; 3985 mutex_unlock(&curseg->curseg_mutex); 3986 out: 3987 f2fs_put_page(new, 1); 3988 return err; 3989 } 3990 3991 static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 3992 { 3993 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal; 3994 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal; 3995 int type = CURSEG_HOT_DATA; 3996 int err; 3997 3998 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) { 3999 int npages = f2fs_npages_for_summary_flush(sbi, true); 4000 4001 if (npages >= 2) 4002 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages, 4003 META_CP, true); 4004 4005 /* restore for compacted data summary */ 4006 err = read_compacted_summaries(sbi); 4007 if (err) 4008 return err; 4009 type = CURSEG_HOT_NODE; 4010 } 4011 4012 if (__exist_node_summaries(sbi)) 4013 f2fs_ra_meta_pages(sbi, 4014 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type), 4015 NR_CURSEG_PERSIST_TYPE - type, META_CP, true); 4016 4017 for (; type <= CURSEG_COLD_NODE; type++) { 4018 err = read_normal_summaries(sbi, type); 4019 if (err) 4020 return err; 4021 } 4022 4023 /* sanity check for summary blocks */ 4024 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES || 4025 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) { 4026 f2fs_err(sbi, "invalid journal entries nats %u sits %u", 4027 nats_in_cursum(nat_j), sits_in_cursum(sit_j)); 4028 return -EINVAL; 4029 } 4030 4031 return 0; 4032 } 4033 4034 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 4035 { 4036 struct page *page; 4037 unsigned char *kaddr; 4038 struct f2fs_summary *summary; 4039 struct curseg_info *seg_i; 4040 int written_size = 0; 4041 int i, j; 4042 4043 page = f2fs_grab_meta_page(sbi, blkaddr++); 4044 kaddr = (unsigned char *)page_address(page); 4045 memset(kaddr, 0, PAGE_SIZE); 4046 4047 /* Step 1: write nat cache */ 4048 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 4049 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE); 4050 written_size += SUM_JOURNAL_SIZE; 4051 4052 /* Step 2: write sit cache */ 4053 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 4054 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE); 4055 written_size += SUM_JOURNAL_SIZE; 4056 4057 /* Step 3: write summary entries */ 4058 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 4059 seg_i = CURSEG_I(sbi, i); 4060 for (j = 0; j < f2fs_curseg_valid_blocks(sbi, i); j++) { 4061 if (!page) { 4062 page = f2fs_grab_meta_page(sbi, blkaddr++); 4063 kaddr = (unsigned char *)page_address(page); 4064 memset(kaddr, 0, PAGE_SIZE); 4065 written_size = 0; 4066 } 4067 summary = (struct f2fs_summary *)(kaddr + written_size); 4068 *summary = seg_i->sum_blk->entries[j]; 4069 written_size += SUMMARY_SIZE; 4070 4071 if (written_size + SUMMARY_SIZE <= PAGE_SIZE - 4072 SUM_FOOTER_SIZE) 4073 continue; 4074 4075 set_page_dirty(page); 4076 f2fs_put_page(page, 1); 4077 page = NULL; 4078 } 4079 } 4080 if (page) { 4081 set_page_dirty(page); 4082 f2fs_put_page(page, 1); 4083 } 4084 } 4085 4086 static void write_normal_summaries(struct f2fs_sb_info *sbi, 4087 block_t blkaddr, int type) 4088 { 4089 int i, end; 4090 4091 if (IS_DATASEG(type)) 4092 end = type + NR_CURSEG_DATA_TYPE; 4093 else 4094 end = type + NR_CURSEG_NODE_TYPE; 4095 4096 for (i = type; i < end; i++) 4097 write_current_sum_page(sbi, i, blkaddr + (i - type)); 4098 } 4099 4100 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 4101 { 4102 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) 4103 write_compacted_summaries(sbi, start_blk); 4104 else 4105 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 4106 } 4107 4108 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 4109 { 4110 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 4111 } 4112 4113 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type, 4114 unsigned int val, int alloc) 4115 { 4116 int i; 4117 4118 if (type == NAT_JOURNAL) { 4119 for (i = 0; i < nats_in_cursum(journal); i++) { 4120 if (le32_to_cpu(nid_in_journal(journal, i)) == val) 4121 return i; 4122 } 4123 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL)) 4124 return update_nats_in_cursum(journal, 1); 4125 } else if (type == SIT_JOURNAL) { 4126 for (i = 0; i < sits_in_cursum(journal); i++) 4127 if (le32_to_cpu(segno_in_journal(journal, i)) == val) 4128 return i; 4129 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL)) 4130 return update_sits_in_cursum(journal, 1); 4131 } 4132 return -1; 4133 } 4134 4135 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, 4136 unsigned int segno) 4137 { 4138 return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno)); 4139 } 4140 4141 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, 4142 unsigned int start) 4143 { 4144 struct sit_info *sit_i = SIT_I(sbi); 4145 struct page *page; 4146 pgoff_t src_off, dst_off; 4147 4148 src_off = current_sit_addr(sbi, start); 4149 dst_off = next_sit_addr(sbi, src_off); 4150 4151 page = f2fs_grab_meta_page(sbi, dst_off); 4152 seg_info_to_sit_page(sbi, page, start); 4153 4154 set_page_dirty(page); 4155 set_to_next_sit(sit_i, start); 4156 4157 return page; 4158 } 4159 4160 static struct sit_entry_set *grab_sit_entry_set(void) 4161 { 4162 struct sit_entry_set *ses = 4163 f2fs_kmem_cache_alloc(sit_entry_set_slab, 4164 GFP_NOFS, true, NULL); 4165 4166 ses->entry_cnt = 0; 4167 INIT_LIST_HEAD(&ses->set_list); 4168 return ses; 4169 } 4170 4171 static void release_sit_entry_set(struct sit_entry_set *ses) 4172 { 4173 list_del(&ses->set_list); 4174 kmem_cache_free(sit_entry_set_slab, ses); 4175 } 4176 4177 static void adjust_sit_entry_set(struct sit_entry_set *ses, 4178 struct list_head *head) 4179 { 4180 struct sit_entry_set *next = ses; 4181 4182 if (list_is_last(&ses->set_list, head)) 4183 return; 4184 4185 list_for_each_entry_continue(next, head, set_list) 4186 if (ses->entry_cnt <= next->entry_cnt) { 4187 list_move_tail(&ses->set_list, &next->set_list); 4188 return; 4189 } 4190 4191 list_move_tail(&ses->set_list, head); 4192 } 4193 4194 static void add_sit_entry(unsigned int segno, struct list_head *head) 4195 { 4196 struct sit_entry_set *ses; 4197 unsigned int start_segno = START_SEGNO(segno); 4198 4199 list_for_each_entry(ses, head, set_list) { 4200 if (ses->start_segno == start_segno) { 4201 ses->entry_cnt++; 4202 adjust_sit_entry_set(ses, head); 4203 return; 4204 } 4205 } 4206 4207 ses = grab_sit_entry_set(); 4208 4209 ses->start_segno = start_segno; 4210 ses->entry_cnt++; 4211 list_add(&ses->set_list, head); 4212 } 4213 4214 static void add_sits_in_set(struct f2fs_sb_info *sbi) 4215 { 4216 struct f2fs_sm_info *sm_info = SM_I(sbi); 4217 struct list_head *set_list = &sm_info->sit_entry_set; 4218 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; 4219 unsigned int segno; 4220 4221 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) 4222 add_sit_entry(segno, set_list); 4223 } 4224 4225 static void remove_sits_in_journal(struct f2fs_sb_info *sbi) 4226 { 4227 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4228 struct f2fs_journal *journal = curseg->journal; 4229 int i; 4230 4231 down_write(&curseg->journal_rwsem); 4232 for (i = 0; i < sits_in_cursum(journal); i++) { 4233 unsigned int segno; 4234 bool dirtied; 4235 4236 segno = le32_to_cpu(segno_in_journal(journal, i)); 4237 dirtied = __mark_sit_entry_dirty(sbi, segno); 4238 4239 if (!dirtied) 4240 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); 4241 } 4242 update_sits_in_cursum(journal, -i); 4243 up_write(&curseg->journal_rwsem); 4244 } 4245 4246 /* 4247 * CP calls this function, which flushes SIT entries including sit_journal, 4248 * and moves prefree segs to free segs. 4249 */ 4250 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 4251 { 4252 struct sit_info *sit_i = SIT_I(sbi); 4253 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 4254 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4255 struct f2fs_journal *journal = curseg->journal; 4256 struct sit_entry_set *ses, *tmp; 4257 struct list_head *head = &SM_I(sbi)->sit_entry_set; 4258 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS); 4259 struct seg_entry *se; 4260 4261 down_write(&sit_i->sentry_lock); 4262 4263 if (!sit_i->dirty_sentries) 4264 goto out; 4265 4266 /* 4267 * add and account sit entries of dirty bitmap in sit entry 4268 * set temporarily 4269 */ 4270 add_sits_in_set(sbi); 4271 4272 /* 4273 * if there are no enough space in journal to store dirty sit 4274 * entries, remove all entries from journal and add and account 4275 * them in sit entry set. 4276 */ 4277 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) || 4278 !to_journal) 4279 remove_sits_in_journal(sbi); 4280 4281 /* 4282 * there are two steps to flush sit entries: 4283 * #1, flush sit entries to journal in current cold data summary block. 4284 * #2, flush sit entries to sit page. 4285 */ 4286 list_for_each_entry_safe(ses, tmp, head, set_list) { 4287 struct page *page = NULL; 4288 struct f2fs_sit_block *raw_sit = NULL; 4289 unsigned int start_segno = ses->start_segno; 4290 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, 4291 (unsigned long)MAIN_SEGS(sbi)); 4292 unsigned int segno = start_segno; 4293 4294 if (to_journal && 4295 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL)) 4296 to_journal = false; 4297 4298 if (to_journal) { 4299 down_write(&curseg->journal_rwsem); 4300 } else { 4301 page = get_next_sit_page(sbi, start_segno); 4302 raw_sit = page_address(page); 4303 } 4304 4305 /* flush dirty sit entries in region of current sit set */ 4306 for_each_set_bit_from(segno, bitmap, end) { 4307 int offset, sit_offset; 4308 4309 se = get_seg_entry(sbi, segno); 4310 #ifdef CONFIG_F2FS_CHECK_FS 4311 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir, 4312 SIT_VBLOCK_MAP_SIZE)) 4313 f2fs_bug_on(sbi, 1); 4314 #endif 4315 4316 /* add discard candidates */ 4317 if (!(cpc->reason & CP_DISCARD)) { 4318 cpc->trim_start = segno; 4319 add_discard_addrs(sbi, cpc, false); 4320 } 4321 4322 if (to_journal) { 4323 offset = f2fs_lookup_journal_in_cursum(journal, 4324 SIT_JOURNAL, segno, 1); 4325 f2fs_bug_on(sbi, offset < 0); 4326 segno_in_journal(journal, offset) = 4327 cpu_to_le32(segno); 4328 seg_info_to_raw_sit(se, 4329 &sit_in_journal(journal, offset)); 4330 check_block_count(sbi, segno, 4331 &sit_in_journal(journal, offset)); 4332 } else { 4333 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 4334 seg_info_to_raw_sit(se, 4335 &raw_sit->entries[sit_offset]); 4336 check_block_count(sbi, segno, 4337 &raw_sit->entries[sit_offset]); 4338 } 4339 4340 __clear_bit(segno, bitmap); 4341 sit_i->dirty_sentries--; 4342 ses->entry_cnt--; 4343 } 4344 4345 if (to_journal) 4346 up_write(&curseg->journal_rwsem); 4347 else 4348 f2fs_put_page(page, 1); 4349 4350 f2fs_bug_on(sbi, ses->entry_cnt); 4351 release_sit_entry_set(ses); 4352 } 4353 4354 f2fs_bug_on(sbi, !list_empty(head)); 4355 f2fs_bug_on(sbi, sit_i->dirty_sentries); 4356 out: 4357 if (cpc->reason & CP_DISCARD) { 4358 __u64 trim_start = cpc->trim_start; 4359 4360 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) 4361 add_discard_addrs(sbi, cpc, false); 4362 4363 cpc->trim_start = trim_start; 4364 } 4365 up_write(&sit_i->sentry_lock); 4366 4367 set_prefree_as_free_segments(sbi); 4368 } 4369 4370 static int build_sit_info(struct f2fs_sb_info *sbi) 4371 { 4372 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 4373 struct sit_info *sit_i; 4374 unsigned int sit_segs, start; 4375 char *src_bitmap, *bitmap; 4376 unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size; 4377 unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0; 4378 4379 /* allocate memory for SIT information */ 4380 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL); 4381 if (!sit_i) 4382 return -ENOMEM; 4383 4384 SM_I(sbi)->sit_info = sit_i; 4385 4386 sit_i->sentries = 4387 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry), 4388 MAIN_SEGS(sbi)), 4389 GFP_KERNEL); 4390 if (!sit_i->sentries) 4391 return -ENOMEM; 4392 4393 main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4394 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size, 4395 GFP_KERNEL); 4396 if (!sit_i->dirty_sentries_bitmap) 4397 return -ENOMEM; 4398 4399 #ifdef CONFIG_F2FS_CHECK_FS 4400 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map); 4401 #else 4402 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map); 4403 #endif 4404 sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4405 if (!sit_i->bitmap) 4406 return -ENOMEM; 4407 4408 bitmap = sit_i->bitmap; 4409 4410 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4411 sit_i->sentries[start].cur_valid_map = bitmap; 4412 bitmap += SIT_VBLOCK_MAP_SIZE; 4413 4414 sit_i->sentries[start].ckpt_valid_map = bitmap; 4415 bitmap += SIT_VBLOCK_MAP_SIZE; 4416 4417 #ifdef CONFIG_F2FS_CHECK_FS 4418 sit_i->sentries[start].cur_valid_map_mir = bitmap; 4419 bitmap += SIT_VBLOCK_MAP_SIZE; 4420 #endif 4421 4422 if (discard_map) { 4423 sit_i->sentries[start].discard_map = bitmap; 4424 bitmap += SIT_VBLOCK_MAP_SIZE; 4425 } 4426 } 4427 4428 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 4429 if (!sit_i->tmp_map) 4430 return -ENOMEM; 4431 4432 if (__is_large_section(sbi)) { 4433 sit_i->sec_entries = 4434 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry), 4435 MAIN_SECS(sbi)), 4436 GFP_KERNEL); 4437 if (!sit_i->sec_entries) 4438 return -ENOMEM; 4439 } 4440 4441 /* get information related with SIT */ 4442 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 4443 4444 /* setup SIT bitmap from ckeckpoint pack */ 4445 sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 4446 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 4447 4448 sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL); 4449 if (!sit_i->sit_bitmap) 4450 return -ENOMEM; 4451 4452 #ifdef CONFIG_F2FS_CHECK_FS 4453 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, 4454 sit_bitmap_size, GFP_KERNEL); 4455 if (!sit_i->sit_bitmap_mir) 4456 return -ENOMEM; 4457 4458 sit_i->invalid_segmap = f2fs_kvzalloc(sbi, 4459 main_bitmap_size, GFP_KERNEL); 4460 if (!sit_i->invalid_segmap) 4461 return -ENOMEM; 4462 #endif 4463 4464 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 4465 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; 4466 sit_i->written_valid_blocks = 0; 4467 sit_i->bitmap_size = sit_bitmap_size; 4468 sit_i->dirty_sentries = 0; 4469 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 4470 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 4471 sit_i->mounted_time = ktime_get_boottime_seconds(); 4472 init_rwsem(&sit_i->sentry_lock); 4473 return 0; 4474 } 4475 4476 static int build_free_segmap(struct f2fs_sb_info *sbi) 4477 { 4478 struct free_segmap_info *free_i; 4479 unsigned int bitmap_size, sec_bitmap_size; 4480 4481 /* allocate memory for free segmap information */ 4482 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL); 4483 if (!free_i) 4484 return -ENOMEM; 4485 4486 SM_I(sbi)->free_info = free_i; 4487 4488 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4489 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL); 4490 if (!free_i->free_segmap) 4491 return -ENOMEM; 4492 4493 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4494 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL); 4495 if (!free_i->free_secmap) 4496 return -ENOMEM; 4497 4498 /* set all segments as dirty temporarily */ 4499 memset(free_i->free_segmap, 0xff, bitmap_size); 4500 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 4501 4502 /* init free segmap information */ 4503 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); 4504 free_i->free_segments = 0; 4505 free_i->free_sections = 0; 4506 spin_lock_init(&free_i->segmap_lock); 4507 return 0; 4508 } 4509 4510 static int build_curseg(struct f2fs_sb_info *sbi) 4511 { 4512 struct curseg_info *array; 4513 int i; 4514 4515 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, 4516 sizeof(*array)), GFP_KERNEL); 4517 if (!array) 4518 return -ENOMEM; 4519 4520 SM_I(sbi)->curseg_array = array; 4521 4522 for (i = 0; i < NO_CHECK_TYPE; i++) { 4523 mutex_init(&array[i].curseg_mutex); 4524 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL); 4525 if (!array[i].sum_blk) 4526 return -ENOMEM; 4527 init_rwsem(&array[i].journal_rwsem); 4528 array[i].journal = f2fs_kzalloc(sbi, 4529 sizeof(struct f2fs_journal), GFP_KERNEL); 4530 if (!array[i].journal) 4531 return -ENOMEM; 4532 if (i < NR_PERSISTENT_LOG) 4533 array[i].seg_type = CURSEG_HOT_DATA + i; 4534 else if (i == CURSEG_COLD_DATA_PINNED) 4535 array[i].seg_type = CURSEG_COLD_DATA; 4536 else if (i == CURSEG_ALL_DATA_ATGC) 4537 array[i].seg_type = CURSEG_COLD_DATA; 4538 array[i].segno = NULL_SEGNO; 4539 array[i].next_blkoff = 0; 4540 array[i].inited = false; 4541 } 4542 return restore_curseg_summaries(sbi); 4543 } 4544 4545 static int build_sit_entries(struct f2fs_sb_info *sbi) 4546 { 4547 struct sit_info *sit_i = SIT_I(sbi); 4548 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4549 struct f2fs_journal *journal = curseg->journal; 4550 struct seg_entry *se; 4551 struct f2fs_sit_entry sit; 4552 int sit_blk_cnt = SIT_BLK_CNT(sbi); 4553 unsigned int i, start, end; 4554 unsigned int readed, start_blk = 0; 4555 int err = 0; 4556 block_t sit_valid_blocks[2] = {0, 0}; 4557 4558 do { 4559 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS, 4560 META_SIT, true); 4561 4562 start = start_blk * sit_i->sents_per_block; 4563 end = (start_blk + readed) * sit_i->sents_per_block; 4564 4565 for (; start < end && start < MAIN_SEGS(sbi); start++) { 4566 struct f2fs_sit_block *sit_blk; 4567 struct page *page; 4568 4569 se = &sit_i->sentries[start]; 4570 page = get_current_sit_page(sbi, start); 4571 if (IS_ERR(page)) 4572 return PTR_ERR(page); 4573 sit_blk = (struct f2fs_sit_block *)page_address(page); 4574 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 4575 f2fs_put_page(page, 1); 4576 4577 err = check_block_count(sbi, start, &sit); 4578 if (err) 4579 return err; 4580 seg_info_from_raw_sit(se, &sit); 4581 4582 if (se->type >= NR_PERSISTENT_LOG) { 4583 f2fs_err(sbi, "Invalid segment type: %u, segno: %u", 4584 se->type, start); 4585 f2fs_handle_error(sbi, 4586 ERROR_INCONSISTENT_SUM_TYPE); 4587 return -EFSCORRUPTED; 4588 } 4589 4590 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks; 4591 4592 if (f2fs_block_unit_discard(sbi)) { 4593 /* build discard map only one time */ 4594 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4595 memset(se->discard_map, 0xff, 4596 SIT_VBLOCK_MAP_SIZE); 4597 } else { 4598 memcpy(se->discard_map, 4599 se->cur_valid_map, 4600 SIT_VBLOCK_MAP_SIZE); 4601 sbi->discard_blks += 4602 sbi->blocks_per_seg - 4603 se->valid_blocks; 4604 } 4605 } 4606 4607 if (__is_large_section(sbi)) 4608 get_sec_entry(sbi, start)->valid_blocks += 4609 se->valid_blocks; 4610 } 4611 start_blk += readed; 4612 } while (start_blk < sit_blk_cnt); 4613 4614 down_read(&curseg->journal_rwsem); 4615 for (i = 0; i < sits_in_cursum(journal); i++) { 4616 unsigned int old_valid_blocks; 4617 4618 start = le32_to_cpu(segno_in_journal(journal, i)); 4619 if (start >= MAIN_SEGS(sbi)) { 4620 f2fs_err(sbi, "Wrong journal entry on segno %u", 4621 start); 4622 err = -EFSCORRUPTED; 4623 f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL); 4624 break; 4625 } 4626 4627 se = &sit_i->sentries[start]; 4628 sit = sit_in_journal(journal, i); 4629 4630 old_valid_blocks = se->valid_blocks; 4631 4632 sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks; 4633 4634 err = check_block_count(sbi, start, &sit); 4635 if (err) 4636 break; 4637 seg_info_from_raw_sit(se, &sit); 4638 4639 if (se->type >= NR_PERSISTENT_LOG) { 4640 f2fs_err(sbi, "Invalid segment type: %u, segno: %u", 4641 se->type, start); 4642 err = -EFSCORRUPTED; 4643 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE); 4644 break; 4645 } 4646 4647 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks; 4648 4649 if (f2fs_block_unit_discard(sbi)) { 4650 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4651 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE); 4652 } else { 4653 memcpy(se->discard_map, se->cur_valid_map, 4654 SIT_VBLOCK_MAP_SIZE); 4655 sbi->discard_blks += old_valid_blocks; 4656 sbi->discard_blks -= se->valid_blocks; 4657 } 4658 } 4659 4660 if (__is_large_section(sbi)) { 4661 get_sec_entry(sbi, start)->valid_blocks += 4662 se->valid_blocks; 4663 get_sec_entry(sbi, start)->valid_blocks -= 4664 old_valid_blocks; 4665 } 4666 } 4667 up_read(&curseg->journal_rwsem); 4668 4669 if (err) 4670 return err; 4671 4672 if (sit_valid_blocks[NODE] != valid_node_count(sbi)) { 4673 f2fs_err(sbi, "SIT is corrupted node# %u vs %u", 4674 sit_valid_blocks[NODE], valid_node_count(sbi)); 4675 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT); 4676 return -EFSCORRUPTED; 4677 } 4678 4679 if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] > 4680 valid_user_blocks(sbi)) { 4681 f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u", 4682 sit_valid_blocks[DATA], sit_valid_blocks[NODE], 4683 valid_user_blocks(sbi)); 4684 f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT); 4685 return -EFSCORRUPTED; 4686 } 4687 4688 return 0; 4689 } 4690 4691 static void init_free_segmap(struct f2fs_sb_info *sbi) 4692 { 4693 unsigned int start; 4694 int type; 4695 struct seg_entry *sentry; 4696 4697 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4698 if (f2fs_usable_blks_in_seg(sbi, start) == 0) 4699 continue; 4700 sentry = get_seg_entry(sbi, start); 4701 if (!sentry->valid_blocks) 4702 __set_free(sbi, start); 4703 else 4704 SIT_I(sbi)->written_valid_blocks += 4705 sentry->valid_blocks; 4706 } 4707 4708 /* set use the current segments */ 4709 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 4710 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 4711 4712 __set_test_and_inuse(sbi, curseg_t->segno); 4713 } 4714 } 4715 4716 static void init_dirty_segmap(struct f2fs_sb_info *sbi) 4717 { 4718 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4719 struct free_segmap_info *free_i = FREE_I(sbi); 4720 unsigned int segno = 0, offset = 0, secno; 4721 block_t valid_blocks, usable_blks_in_seg; 4722 4723 while (1) { 4724 /* find dirty segment based on free segmap */ 4725 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); 4726 if (segno >= MAIN_SEGS(sbi)) 4727 break; 4728 offset = segno + 1; 4729 valid_blocks = get_valid_blocks(sbi, segno, false); 4730 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); 4731 if (valid_blocks == usable_blks_in_seg || !valid_blocks) 4732 continue; 4733 if (valid_blocks > usable_blks_in_seg) { 4734 f2fs_bug_on(sbi, 1); 4735 continue; 4736 } 4737 mutex_lock(&dirty_i->seglist_lock); 4738 __locate_dirty_segment(sbi, segno, DIRTY); 4739 mutex_unlock(&dirty_i->seglist_lock); 4740 } 4741 4742 if (!__is_large_section(sbi)) 4743 return; 4744 4745 mutex_lock(&dirty_i->seglist_lock); 4746 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 4747 valid_blocks = get_valid_blocks(sbi, segno, true); 4748 secno = GET_SEC_FROM_SEG(sbi, segno); 4749 4750 if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi)) 4751 continue; 4752 if (IS_CURSEC(sbi, secno)) 4753 continue; 4754 set_bit(secno, dirty_i->dirty_secmap); 4755 } 4756 mutex_unlock(&dirty_i->seglist_lock); 4757 } 4758 4759 static int init_victim_secmap(struct f2fs_sb_info *sbi) 4760 { 4761 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4762 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4763 4764 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4765 if (!dirty_i->victim_secmap) 4766 return -ENOMEM; 4767 4768 dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4769 if (!dirty_i->pinned_secmap) 4770 return -ENOMEM; 4771 4772 dirty_i->pinned_secmap_cnt = 0; 4773 dirty_i->enable_pin_section = true; 4774 return 0; 4775 } 4776 4777 static int build_dirty_segmap(struct f2fs_sb_info *sbi) 4778 { 4779 struct dirty_seglist_info *dirty_i; 4780 unsigned int bitmap_size, i; 4781 4782 /* allocate memory for dirty segments list information */ 4783 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info), 4784 GFP_KERNEL); 4785 if (!dirty_i) 4786 return -ENOMEM; 4787 4788 SM_I(sbi)->dirty_info = dirty_i; 4789 mutex_init(&dirty_i->seglist_lock); 4790 4791 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4792 4793 for (i = 0; i < NR_DIRTY_TYPE; i++) { 4794 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size, 4795 GFP_KERNEL); 4796 if (!dirty_i->dirty_segmap[i]) 4797 return -ENOMEM; 4798 } 4799 4800 if (__is_large_section(sbi)) { 4801 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4802 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi, 4803 bitmap_size, GFP_KERNEL); 4804 if (!dirty_i->dirty_secmap) 4805 return -ENOMEM; 4806 } 4807 4808 init_dirty_segmap(sbi); 4809 return init_victim_secmap(sbi); 4810 } 4811 4812 static int sanity_check_curseg(struct f2fs_sb_info *sbi) 4813 { 4814 int i; 4815 4816 /* 4817 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr; 4818 * In LFS curseg, all blkaddr after .next_blkoff should be unused. 4819 */ 4820 for (i = 0; i < NR_PERSISTENT_LOG; i++) { 4821 struct curseg_info *curseg = CURSEG_I(sbi, i); 4822 struct seg_entry *se = get_seg_entry(sbi, curseg->segno); 4823 unsigned int blkofs = curseg->next_blkoff; 4824 4825 if (f2fs_sb_has_readonly(sbi) && 4826 i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE) 4827 continue; 4828 4829 sanity_check_seg_type(sbi, curseg->seg_type); 4830 4831 if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) { 4832 f2fs_err(sbi, 4833 "Current segment has invalid alloc_type:%d", 4834 curseg->alloc_type); 4835 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG); 4836 return -EFSCORRUPTED; 4837 } 4838 4839 if (f2fs_test_bit(blkofs, se->cur_valid_map)) 4840 goto out; 4841 4842 if (curseg->alloc_type == SSR) 4843 continue; 4844 4845 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) { 4846 if (!f2fs_test_bit(blkofs, se->cur_valid_map)) 4847 continue; 4848 out: 4849 f2fs_err(sbi, 4850 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u", 4851 i, curseg->segno, curseg->alloc_type, 4852 curseg->next_blkoff, blkofs); 4853 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG); 4854 return -EFSCORRUPTED; 4855 } 4856 } 4857 return 0; 4858 } 4859 4860 #ifdef CONFIG_BLK_DEV_ZONED 4861 4862 static int check_zone_write_pointer(struct f2fs_sb_info *sbi, 4863 struct f2fs_dev_info *fdev, 4864 struct blk_zone *zone) 4865 { 4866 unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno; 4867 block_t zone_block, wp_block, last_valid_block; 4868 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 4869 int i, s, b, ret; 4870 struct seg_entry *se; 4871 4872 if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4873 return 0; 4874 4875 wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block); 4876 wp_segno = GET_SEGNO(sbi, wp_block); 4877 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno); 4878 zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block); 4879 zone_segno = GET_SEGNO(sbi, zone_block); 4880 zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno); 4881 4882 if (zone_segno >= MAIN_SEGS(sbi)) 4883 return 0; 4884 4885 /* 4886 * Skip check of zones cursegs point to, since 4887 * fix_curseg_write_pointer() checks them. 4888 */ 4889 for (i = 0; i < NO_CHECK_TYPE; i++) 4890 if (zone_secno == GET_SEC_FROM_SEG(sbi, 4891 CURSEG_I(sbi, i)->segno)) 4892 return 0; 4893 4894 /* 4895 * Get last valid block of the zone. 4896 */ 4897 last_valid_block = zone_block - 1; 4898 for (s = sbi->segs_per_sec - 1; s >= 0; s--) { 4899 segno = zone_segno + s; 4900 se = get_seg_entry(sbi, segno); 4901 for (b = sbi->blocks_per_seg - 1; b >= 0; b--) 4902 if (f2fs_test_bit(b, se->cur_valid_map)) { 4903 last_valid_block = START_BLOCK(sbi, segno) + b; 4904 break; 4905 } 4906 if (last_valid_block >= zone_block) 4907 break; 4908 } 4909 4910 /* 4911 * The write pointer matches with the valid blocks or 4912 * already points to the end of the zone. 4913 */ 4914 if ((last_valid_block + 1 == wp_block) || 4915 (zone->wp == zone->start + zone->len)) 4916 return 0; 4917 4918 if (last_valid_block + 1 == zone_block) { 4919 /* 4920 * If there is no valid block in the zone and if write pointer 4921 * is not at zone start, reset the write pointer. 4922 */ 4923 f2fs_notice(sbi, 4924 "Zone without valid block has non-zero write " 4925 "pointer. Reset the write pointer: wp[0x%x,0x%x]", 4926 wp_segno, wp_blkoff); 4927 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block, 4928 zone->len >> log_sectors_per_block); 4929 if (ret) 4930 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 4931 fdev->path, ret); 4932 4933 return ret; 4934 } 4935 4936 /* 4937 * If there are valid blocks and the write pointer doesn't 4938 * match with them, we need to report the inconsistency and 4939 * fill the zone till the end to close the zone. This inconsistency 4940 * does not cause write error because the zone will not be selected 4941 * for write operation until it get discarded. 4942 */ 4943 f2fs_notice(sbi, "Valid blocks are not aligned with write pointer: " 4944 "valid block[0x%x,0x%x] wp[0x%x,0x%x]", 4945 GET_SEGNO(sbi, last_valid_block), 4946 GET_BLKOFF_FROM_SEG0(sbi, last_valid_block), 4947 wp_segno, wp_blkoff); 4948 4949 ret = blkdev_issue_zeroout(fdev->bdev, zone->wp, 4950 zone->len - (zone->wp - zone->start), 4951 GFP_NOFS, 0); 4952 if (ret) 4953 f2fs_err(sbi, "Fill up zone failed: %s (errno=%d)", 4954 fdev->path, ret); 4955 4956 return ret; 4957 } 4958 4959 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi, 4960 block_t zone_blkaddr) 4961 { 4962 int i; 4963 4964 for (i = 0; i < sbi->s_ndevs; i++) { 4965 if (!bdev_is_zoned(FDEV(i).bdev)) 4966 continue; 4967 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr && 4968 zone_blkaddr <= FDEV(i).end_blk)) 4969 return &FDEV(i); 4970 } 4971 4972 return NULL; 4973 } 4974 4975 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx, 4976 void *data) 4977 { 4978 memcpy(data, zone, sizeof(struct blk_zone)); 4979 return 0; 4980 } 4981 4982 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type) 4983 { 4984 struct curseg_info *cs = CURSEG_I(sbi, type); 4985 struct f2fs_dev_info *zbd; 4986 struct blk_zone zone; 4987 unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off; 4988 block_t cs_zone_block, wp_block; 4989 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 4990 sector_t zone_sector; 4991 int err; 4992 4993 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 4994 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 4995 4996 zbd = get_target_zoned_dev(sbi, cs_zone_block); 4997 if (!zbd) 4998 return 0; 4999 5000 /* report zone for the sector the curseg points to */ 5001 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 5002 << log_sectors_per_block; 5003 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 5004 report_one_zone_cb, &zone); 5005 if (err != 1) { 5006 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 5007 zbd->path, err); 5008 return err; 5009 } 5010 5011 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 5012 return 0; 5013 5014 wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block); 5015 wp_segno = GET_SEGNO(sbi, wp_block); 5016 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno); 5017 wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0); 5018 5019 if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff && 5020 wp_sector_off == 0) 5021 return 0; 5022 5023 f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: " 5024 "curseg[0x%x,0x%x] wp[0x%x,0x%x]", 5025 type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff); 5026 5027 f2fs_notice(sbi, "Assign new section to curseg[%d]: " 5028 "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff); 5029 5030 f2fs_allocate_new_section(sbi, type, true); 5031 5032 /* check consistency of the zone curseg pointed to */ 5033 if (check_zone_write_pointer(sbi, zbd, &zone)) 5034 return -EIO; 5035 5036 /* check newly assigned zone */ 5037 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 5038 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 5039 5040 zbd = get_target_zoned_dev(sbi, cs_zone_block); 5041 if (!zbd) 5042 return 0; 5043 5044 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 5045 << log_sectors_per_block; 5046 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 5047 report_one_zone_cb, &zone); 5048 if (err != 1) { 5049 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 5050 zbd->path, err); 5051 return err; 5052 } 5053 5054 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 5055 return 0; 5056 5057 if (zone.wp != zone.start) { 5058 f2fs_notice(sbi, 5059 "New zone for curseg[%d] is not yet discarded. " 5060 "Reset the zone: curseg[0x%x,0x%x]", 5061 type, cs->segno, cs->next_blkoff); 5062 err = __f2fs_issue_discard_zone(sbi, zbd->bdev, cs_zone_block, 5063 zone.len >> log_sectors_per_block); 5064 if (err) { 5065 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 5066 zbd->path, err); 5067 return err; 5068 } 5069 } 5070 5071 return 0; 5072 } 5073 5074 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 5075 { 5076 int i, ret; 5077 5078 for (i = 0; i < NR_PERSISTENT_LOG; i++) { 5079 ret = fix_curseg_write_pointer(sbi, i); 5080 if (ret) 5081 return ret; 5082 } 5083 5084 return 0; 5085 } 5086 5087 struct check_zone_write_pointer_args { 5088 struct f2fs_sb_info *sbi; 5089 struct f2fs_dev_info *fdev; 5090 }; 5091 5092 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx, 5093 void *data) 5094 { 5095 struct check_zone_write_pointer_args *args; 5096 5097 args = (struct check_zone_write_pointer_args *)data; 5098 5099 return check_zone_write_pointer(args->sbi, args->fdev, zone); 5100 } 5101 5102 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi) 5103 { 5104 int i, ret; 5105 struct check_zone_write_pointer_args args; 5106 5107 for (i = 0; i < sbi->s_ndevs; i++) { 5108 if (!bdev_is_zoned(FDEV(i).bdev)) 5109 continue; 5110 5111 args.sbi = sbi; 5112 args.fdev = &FDEV(i); 5113 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES, 5114 check_zone_write_pointer_cb, &args); 5115 if (ret < 0) 5116 return ret; 5117 } 5118 5119 return 0; 5120 } 5121 5122 /* 5123 * Return the number of usable blocks in a segment. The number of blocks 5124 * returned is always equal to the number of blocks in a segment for 5125 * segments fully contained within a sequential zone capacity or a 5126 * conventional zone. For segments partially contained in a sequential 5127 * zone capacity, the number of usable blocks up to the zone capacity 5128 * is returned. 0 is returned in all other cases. 5129 */ 5130 static inline unsigned int f2fs_usable_zone_blks_in_seg( 5131 struct f2fs_sb_info *sbi, unsigned int segno) 5132 { 5133 block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr; 5134 unsigned int secno; 5135 5136 if (!sbi->unusable_blocks_per_sec) 5137 return sbi->blocks_per_seg; 5138 5139 secno = GET_SEC_FROM_SEG(sbi, segno); 5140 seg_start = START_BLOCK(sbi, segno); 5141 sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno)); 5142 sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi); 5143 5144 /* 5145 * If segment starts before zone capacity and spans beyond 5146 * zone capacity, then usable blocks are from seg start to 5147 * zone capacity. If the segment starts after the zone capacity, 5148 * then there are no usable blocks. 5149 */ 5150 if (seg_start >= sec_cap_blkaddr) 5151 return 0; 5152 if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr) 5153 return sec_cap_blkaddr - seg_start; 5154 5155 return sbi->blocks_per_seg; 5156 } 5157 #else 5158 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 5159 { 5160 return 0; 5161 } 5162 5163 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi) 5164 { 5165 return 0; 5166 } 5167 5168 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi, 5169 unsigned int segno) 5170 { 5171 return 0; 5172 } 5173 5174 #endif 5175 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi, 5176 unsigned int segno) 5177 { 5178 if (f2fs_sb_has_blkzoned(sbi)) 5179 return f2fs_usable_zone_blks_in_seg(sbi, segno); 5180 5181 return sbi->blocks_per_seg; 5182 } 5183 5184 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi, 5185 unsigned int segno) 5186 { 5187 if (f2fs_sb_has_blkzoned(sbi)) 5188 return CAP_SEGS_PER_SEC(sbi); 5189 5190 return sbi->segs_per_sec; 5191 } 5192 5193 /* 5194 * Update min, max modified time for cost-benefit GC algorithm 5195 */ 5196 static void init_min_max_mtime(struct f2fs_sb_info *sbi) 5197 { 5198 struct sit_info *sit_i = SIT_I(sbi); 5199 unsigned int segno; 5200 5201 down_write(&sit_i->sentry_lock); 5202 5203 sit_i->min_mtime = ULLONG_MAX; 5204 5205 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 5206 unsigned int i; 5207 unsigned long long mtime = 0; 5208 5209 for (i = 0; i < sbi->segs_per_sec; i++) 5210 mtime += get_seg_entry(sbi, segno + i)->mtime; 5211 5212 mtime = div_u64(mtime, sbi->segs_per_sec); 5213 5214 if (sit_i->min_mtime > mtime) 5215 sit_i->min_mtime = mtime; 5216 } 5217 sit_i->max_mtime = get_mtime(sbi, false); 5218 sit_i->dirty_max_mtime = 0; 5219 up_write(&sit_i->sentry_lock); 5220 } 5221 5222 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi) 5223 { 5224 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 5225 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 5226 struct f2fs_sm_info *sm_info; 5227 int err; 5228 5229 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL); 5230 if (!sm_info) 5231 return -ENOMEM; 5232 5233 /* init sm info */ 5234 sbi->sm_info = sm_info; 5235 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 5236 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 5237 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 5238 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 5239 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 5240 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 5241 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 5242 sm_info->rec_prefree_segments = sm_info->main_segments * 5243 DEF_RECLAIM_PREFREE_SEGMENTS / 100; 5244 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS) 5245 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS; 5246 5247 if (!f2fs_lfs_mode(sbi)) 5248 sm_info->ipu_policy = BIT(F2FS_IPU_FSYNC); 5249 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; 5250 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; 5251 sm_info->min_seq_blocks = sbi->blocks_per_seg; 5252 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS; 5253 sm_info->min_ssr_sections = reserved_sections(sbi); 5254 5255 INIT_LIST_HEAD(&sm_info->sit_entry_set); 5256 5257 init_f2fs_rwsem(&sm_info->curseg_lock); 5258 5259 err = f2fs_create_flush_cmd_control(sbi); 5260 if (err) 5261 return err; 5262 5263 err = create_discard_cmd_control(sbi); 5264 if (err) 5265 return err; 5266 5267 err = build_sit_info(sbi); 5268 if (err) 5269 return err; 5270 err = build_free_segmap(sbi); 5271 if (err) 5272 return err; 5273 err = build_curseg(sbi); 5274 if (err) 5275 return err; 5276 5277 /* reinit free segmap based on SIT */ 5278 err = build_sit_entries(sbi); 5279 if (err) 5280 return err; 5281 5282 init_free_segmap(sbi); 5283 err = build_dirty_segmap(sbi); 5284 if (err) 5285 return err; 5286 5287 err = sanity_check_curseg(sbi); 5288 if (err) 5289 return err; 5290 5291 init_min_max_mtime(sbi); 5292 return 0; 5293 } 5294 5295 static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 5296 enum dirty_type dirty_type) 5297 { 5298 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5299 5300 mutex_lock(&dirty_i->seglist_lock); 5301 kvfree(dirty_i->dirty_segmap[dirty_type]); 5302 dirty_i->nr_dirty[dirty_type] = 0; 5303 mutex_unlock(&dirty_i->seglist_lock); 5304 } 5305 5306 static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 5307 { 5308 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5309 5310 kvfree(dirty_i->pinned_secmap); 5311 kvfree(dirty_i->victim_secmap); 5312 } 5313 5314 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 5315 { 5316 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5317 int i; 5318 5319 if (!dirty_i) 5320 return; 5321 5322 /* discard pre-free/dirty segments list */ 5323 for (i = 0; i < NR_DIRTY_TYPE; i++) 5324 discard_dirty_segmap(sbi, i); 5325 5326 if (__is_large_section(sbi)) { 5327 mutex_lock(&dirty_i->seglist_lock); 5328 kvfree(dirty_i->dirty_secmap); 5329 mutex_unlock(&dirty_i->seglist_lock); 5330 } 5331 5332 destroy_victim_secmap(sbi); 5333 SM_I(sbi)->dirty_info = NULL; 5334 kfree(dirty_i); 5335 } 5336 5337 static void destroy_curseg(struct f2fs_sb_info *sbi) 5338 { 5339 struct curseg_info *array = SM_I(sbi)->curseg_array; 5340 int i; 5341 5342 if (!array) 5343 return; 5344 SM_I(sbi)->curseg_array = NULL; 5345 for (i = 0; i < NR_CURSEG_TYPE; i++) { 5346 kfree(array[i].sum_blk); 5347 kfree(array[i].journal); 5348 } 5349 kfree(array); 5350 } 5351 5352 static void destroy_free_segmap(struct f2fs_sb_info *sbi) 5353 { 5354 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 5355 5356 if (!free_i) 5357 return; 5358 SM_I(sbi)->free_info = NULL; 5359 kvfree(free_i->free_segmap); 5360 kvfree(free_i->free_secmap); 5361 kfree(free_i); 5362 } 5363 5364 static void destroy_sit_info(struct f2fs_sb_info *sbi) 5365 { 5366 struct sit_info *sit_i = SIT_I(sbi); 5367 5368 if (!sit_i) 5369 return; 5370 5371 if (sit_i->sentries) 5372 kvfree(sit_i->bitmap); 5373 kfree(sit_i->tmp_map); 5374 5375 kvfree(sit_i->sentries); 5376 kvfree(sit_i->sec_entries); 5377 kvfree(sit_i->dirty_sentries_bitmap); 5378 5379 SM_I(sbi)->sit_info = NULL; 5380 kvfree(sit_i->sit_bitmap); 5381 #ifdef CONFIG_F2FS_CHECK_FS 5382 kvfree(sit_i->sit_bitmap_mir); 5383 kvfree(sit_i->invalid_segmap); 5384 #endif 5385 kfree(sit_i); 5386 } 5387 5388 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi) 5389 { 5390 struct f2fs_sm_info *sm_info = SM_I(sbi); 5391 5392 if (!sm_info) 5393 return; 5394 f2fs_destroy_flush_cmd_control(sbi, true); 5395 destroy_discard_cmd_control(sbi); 5396 destroy_dirty_segmap(sbi); 5397 destroy_curseg(sbi); 5398 destroy_free_segmap(sbi); 5399 destroy_sit_info(sbi); 5400 sbi->sm_info = NULL; 5401 kfree(sm_info); 5402 } 5403 5404 int __init f2fs_create_segment_manager_caches(void) 5405 { 5406 discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry", 5407 sizeof(struct discard_entry)); 5408 if (!discard_entry_slab) 5409 goto fail; 5410 5411 discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd", 5412 sizeof(struct discard_cmd)); 5413 if (!discard_cmd_slab) 5414 goto destroy_discard_entry; 5415 5416 sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set", 5417 sizeof(struct sit_entry_set)); 5418 if (!sit_entry_set_slab) 5419 goto destroy_discard_cmd; 5420 5421 revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry", 5422 sizeof(struct revoke_entry)); 5423 if (!revoke_entry_slab) 5424 goto destroy_sit_entry_set; 5425 return 0; 5426 5427 destroy_sit_entry_set: 5428 kmem_cache_destroy(sit_entry_set_slab); 5429 destroy_discard_cmd: 5430 kmem_cache_destroy(discard_cmd_slab); 5431 destroy_discard_entry: 5432 kmem_cache_destroy(discard_entry_slab); 5433 fail: 5434 return -ENOMEM; 5435 } 5436 5437 void f2fs_destroy_segment_manager_caches(void) 5438 { 5439 kmem_cache_destroy(sit_entry_set_slab); 5440 kmem_cache_destroy(discard_cmd_slab); 5441 kmem_cache_destroy(discard_entry_slab); 5442 kmem_cache_destroy(revoke_entry_slab); 5443 } 5444