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 if (clean) 196 truncate_inode_pages_final(inode->i_mapping); 197 198 release_atomic_write_cnt(inode); 199 clear_inode_flag(inode, FI_ATOMIC_COMMITTED); 200 clear_inode_flag(inode, FI_ATOMIC_REPLACE); 201 clear_inode_flag(inode, FI_ATOMIC_FILE); 202 if (is_inode_flag_set(inode, FI_ATOMIC_DIRTIED)) { 203 clear_inode_flag(inode, FI_ATOMIC_DIRTIED); 204 /* 205 * The vfs inode keeps clean during commit, but the f2fs inode 206 * doesn't. So clear the dirty state after commit and let 207 * f2fs_mark_inode_dirty_sync ensure a consistent dirty state. 208 */ 209 f2fs_inode_synced(inode); 210 f2fs_mark_inode_dirty_sync(inode, true); 211 } 212 stat_dec_atomic_inode(inode); 213 214 F2FS_I(inode)->atomic_write_task = NULL; 215 216 if (clean) { 217 f2fs_i_size_write(inode, fi->original_i_size); 218 fi->original_i_size = 0; 219 } 220 /* avoid stale dirty inode during eviction */ 221 sync_inode_metadata(inode, 0); 222 } 223 224 static int __replace_atomic_write_block(struct inode *inode, pgoff_t index, 225 block_t new_addr, block_t *old_addr, bool recover) 226 { 227 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 228 struct dnode_of_data dn; 229 struct node_info ni; 230 int err; 231 232 retry: 233 set_new_dnode(&dn, inode, NULL, NULL, 0); 234 err = f2fs_get_dnode_of_data(&dn, index, ALLOC_NODE); 235 if (err) { 236 if (err == -ENOMEM) { 237 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 238 goto retry; 239 } 240 return err; 241 } 242 243 err = f2fs_get_node_info(sbi, dn.nid, &ni, false); 244 if (err) { 245 f2fs_put_dnode(&dn); 246 return err; 247 } 248 249 if (recover) { 250 /* dn.data_blkaddr is always valid */ 251 if (!__is_valid_data_blkaddr(new_addr)) { 252 if (new_addr == NULL_ADDR) 253 dec_valid_block_count(sbi, inode, 1); 254 f2fs_invalidate_blocks(sbi, dn.data_blkaddr, 1); 255 f2fs_update_data_blkaddr(&dn, new_addr); 256 } else { 257 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, 258 new_addr, ni.version, true, true); 259 } 260 } else { 261 blkcnt_t count = 1; 262 263 err = inc_valid_block_count(sbi, inode, &count, true); 264 if (err) { 265 f2fs_put_dnode(&dn); 266 return err; 267 } 268 269 *old_addr = dn.data_blkaddr; 270 f2fs_truncate_data_blocks_range(&dn, 1); 271 dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count); 272 273 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr, 274 ni.version, true, false); 275 } 276 277 f2fs_put_dnode(&dn); 278 279 trace_f2fs_replace_atomic_write_block(inode, F2FS_I(inode)->cow_inode, 280 index, old_addr ? *old_addr : 0, new_addr, recover); 281 return 0; 282 } 283 284 static void __complete_revoke_list(struct inode *inode, struct list_head *head, 285 bool revoke) 286 { 287 struct revoke_entry *cur, *tmp; 288 pgoff_t start_index = 0; 289 bool truncate = is_inode_flag_set(inode, FI_ATOMIC_REPLACE); 290 291 list_for_each_entry_safe(cur, tmp, head, list) { 292 if (revoke) { 293 __replace_atomic_write_block(inode, cur->index, 294 cur->old_addr, NULL, true); 295 } else if (truncate) { 296 f2fs_truncate_hole(inode, start_index, cur->index); 297 start_index = cur->index + 1; 298 } 299 300 list_del(&cur->list); 301 kmem_cache_free(revoke_entry_slab, cur); 302 } 303 304 if (!revoke && truncate) 305 f2fs_do_truncate_blocks(inode, start_index * PAGE_SIZE, false); 306 } 307 308 static int __f2fs_commit_atomic_write(struct inode *inode) 309 { 310 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 311 struct f2fs_inode_info *fi = F2FS_I(inode); 312 struct inode *cow_inode = fi->cow_inode; 313 struct revoke_entry *new; 314 struct list_head revoke_list; 315 block_t blkaddr; 316 struct dnode_of_data dn; 317 pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); 318 pgoff_t off = 0, blen, index; 319 int ret = 0, i; 320 321 INIT_LIST_HEAD(&revoke_list); 322 323 while (len) { 324 blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len); 325 326 set_new_dnode(&dn, cow_inode, NULL, NULL, 0); 327 ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA); 328 if (ret && ret != -ENOENT) { 329 goto out; 330 } else if (ret == -ENOENT) { 331 ret = 0; 332 if (dn.max_level == 0) 333 goto out; 334 goto next; 335 } 336 337 blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_page, cow_inode), 338 len); 339 index = off; 340 for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) { 341 blkaddr = f2fs_data_blkaddr(&dn); 342 343 if (!__is_valid_data_blkaddr(blkaddr)) { 344 continue; 345 } else if (!f2fs_is_valid_blkaddr(sbi, blkaddr, 346 DATA_GENERIC_ENHANCE)) { 347 f2fs_put_dnode(&dn); 348 ret = -EFSCORRUPTED; 349 goto out; 350 } 351 352 new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS, 353 true, NULL); 354 355 ret = __replace_atomic_write_block(inode, index, blkaddr, 356 &new->old_addr, false); 357 if (ret) { 358 f2fs_put_dnode(&dn); 359 kmem_cache_free(revoke_entry_slab, new); 360 goto out; 361 } 362 363 f2fs_update_data_blkaddr(&dn, NULL_ADDR); 364 new->index = index; 365 list_add_tail(&new->list, &revoke_list); 366 } 367 f2fs_put_dnode(&dn); 368 next: 369 off += blen; 370 len -= blen; 371 } 372 373 out: 374 if (ret) { 375 sbi->revoked_atomic_block += fi->atomic_write_cnt; 376 } else { 377 sbi->committed_atomic_block += fi->atomic_write_cnt; 378 set_inode_flag(inode, FI_ATOMIC_COMMITTED); 379 if (is_inode_flag_set(inode, FI_ATOMIC_DIRTIED)) { 380 clear_inode_flag(inode, FI_ATOMIC_DIRTIED); 381 f2fs_mark_inode_dirty_sync(inode, true); 382 } 383 } 384 385 __complete_revoke_list(inode, &revoke_list, ret ? true : false); 386 387 return ret; 388 } 389 390 int f2fs_commit_atomic_write(struct inode *inode) 391 { 392 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 393 struct f2fs_inode_info *fi = F2FS_I(inode); 394 int err; 395 396 err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX); 397 if (err) 398 return err; 399 400 f2fs_down_write(&fi->i_gc_rwsem[WRITE]); 401 f2fs_lock_op(sbi); 402 403 err = __f2fs_commit_atomic_write(inode); 404 405 f2fs_unlock_op(sbi); 406 f2fs_up_write(&fi->i_gc_rwsem[WRITE]); 407 408 return err; 409 } 410 411 /* 412 * This function balances dirty node and dentry pages. 413 * In addition, it controls garbage collection. 414 */ 415 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need) 416 { 417 if (f2fs_cp_error(sbi)) 418 return; 419 420 if (time_to_inject(sbi, FAULT_CHECKPOINT)) 421 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT); 422 423 /* balance_fs_bg is able to be pending */ 424 if (need && excess_cached_nats(sbi)) 425 f2fs_balance_fs_bg(sbi, false); 426 427 if (!f2fs_is_checkpoint_ready(sbi)) 428 return; 429 430 /* 431 * We should do GC or end up with checkpoint, if there are so many dirty 432 * dir/node pages without enough free segments. 433 */ 434 if (has_enough_free_secs(sbi, 0, 0)) 435 return; 436 437 if (test_opt(sbi, GC_MERGE) && sbi->gc_thread && 438 sbi->gc_thread->f2fs_gc_task) { 439 DEFINE_WAIT(wait); 440 441 prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait, 442 TASK_UNINTERRUPTIBLE); 443 wake_up(&sbi->gc_thread->gc_wait_queue_head); 444 io_schedule(); 445 finish_wait(&sbi->gc_thread->fggc_wq, &wait); 446 } else { 447 struct f2fs_gc_control gc_control = { 448 .victim_segno = NULL_SEGNO, 449 .init_gc_type = BG_GC, 450 .no_bg_gc = true, 451 .should_migrate_blocks = false, 452 .err_gc_skipped = false, 453 .nr_free_secs = 1 }; 454 f2fs_down_write(&sbi->gc_lock); 455 stat_inc_gc_call_count(sbi, FOREGROUND); 456 f2fs_gc(sbi, &gc_control); 457 } 458 } 459 460 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi) 461 { 462 int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2; 463 unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS); 464 unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA); 465 unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES); 466 unsigned int meta = get_pages(sbi, F2FS_DIRTY_META); 467 unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA); 468 unsigned int threshold = 469 SEGS_TO_BLKS(sbi, (factor * DEFAULT_DIRTY_THRESHOLD)); 470 unsigned int global_threshold = threshold * 3 / 2; 471 472 if (dents >= threshold || qdata >= threshold || 473 nodes >= threshold || meta >= threshold || 474 imeta >= threshold) 475 return true; 476 return dents + qdata + nodes + meta + imeta > global_threshold; 477 } 478 479 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg) 480 { 481 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 482 return; 483 484 /* try to shrink extent cache when there is no enough memory */ 485 if (!f2fs_available_free_memory(sbi, READ_EXTENT_CACHE)) 486 f2fs_shrink_read_extent_tree(sbi, 487 READ_EXTENT_CACHE_SHRINK_NUMBER); 488 489 /* try to shrink age extent cache when there is no enough memory */ 490 if (!f2fs_available_free_memory(sbi, AGE_EXTENT_CACHE)) 491 f2fs_shrink_age_extent_tree(sbi, 492 AGE_EXTENT_CACHE_SHRINK_NUMBER); 493 494 /* check the # of cached NAT entries */ 495 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES)) 496 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK); 497 498 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) 499 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS); 500 else 501 f2fs_build_free_nids(sbi, false, false); 502 503 if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) || 504 excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi)) 505 goto do_sync; 506 507 /* there is background inflight IO or foreground operation recently */ 508 if (is_inflight_io(sbi, REQ_TIME) || 509 (!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem))) 510 return; 511 512 /* exceed periodical checkpoint timeout threshold */ 513 if (f2fs_time_over(sbi, CP_TIME)) 514 goto do_sync; 515 516 /* checkpoint is the only way to shrink partial cached entries */ 517 if (f2fs_available_free_memory(sbi, NAT_ENTRIES) && 518 f2fs_available_free_memory(sbi, INO_ENTRIES)) 519 return; 520 521 do_sync: 522 if (test_opt(sbi, DATA_FLUSH) && from_bg) { 523 struct blk_plug plug; 524 525 mutex_lock(&sbi->flush_lock); 526 527 blk_start_plug(&plug); 528 f2fs_sync_dirty_inodes(sbi, FILE_INODE, false); 529 blk_finish_plug(&plug); 530 531 mutex_unlock(&sbi->flush_lock); 532 } 533 stat_inc_cp_call_count(sbi, BACKGROUND); 534 f2fs_sync_fs(sbi->sb, 1); 535 } 536 537 static int __submit_flush_wait(struct f2fs_sb_info *sbi, 538 struct block_device *bdev) 539 { 540 int ret = blkdev_issue_flush(bdev); 541 542 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER), 543 test_opt(sbi, FLUSH_MERGE), ret); 544 if (!ret) 545 f2fs_update_iostat(sbi, NULL, FS_FLUSH_IO, 0); 546 return ret; 547 } 548 549 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino) 550 { 551 int ret = 0; 552 int i; 553 554 if (!f2fs_is_multi_device(sbi)) 555 return __submit_flush_wait(sbi, sbi->sb->s_bdev); 556 557 for (i = 0; i < sbi->s_ndevs; i++) { 558 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO)) 559 continue; 560 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 561 if (ret) 562 break; 563 } 564 return ret; 565 } 566 567 static int issue_flush_thread(void *data) 568 { 569 struct f2fs_sb_info *sbi = data; 570 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 571 wait_queue_head_t *q = &fcc->flush_wait_queue; 572 repeat: 573 if (kthread_should_stop()) 574 return 0; 575 576 if (!llist_empty(&fcc->issue_list)) { 577 struct flush_cmd *cmd, *next; 578 int ret; 579 580 fcc->dispatch_list = llist_del_all(&fcc->issue_list); 581 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); 582 583 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode); 584 585 ret = submit_flush_wait(sbi, cmd->ino); 586 atomic_inc(&fcc->issued_flush); 587 588 llist_for_each_entry_safe(cmd, next, 589 fcc->dispatch_list, llnode) { 590 cmd->ret = ret; 591 complete(&cmd->wait); 592 } 593 fcc->dispatch_list = NULL; 594 } 595 596 wait_event_interruptible(*q, 597 kthread_should_stop() || !llist_empty(&fcc->issue_list)); 598 goto repeat; 599 } 600 601 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino) 602 { 603 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 604 struct flush_cmd cmd; 605 int ret; 606 607 if (test_opt(sbi, NOBARRIER)) 608 return 0; 609 610 if (!test_opt(sbi, FLUSH_MERGE)) { 611 atomic_inc(&fcc->queued_flush); 612 ret = submit_flush_wait(sbi, ino); 613 atomic_dec(&fcc->queued_flush); 614 atomic_inc(&fcc->issued_flush); 615 return ret; 616 } 617 618 if (atomic_inc_return(&fcc->queued_flush) == 1 || 619 f2fs_is_multi_device(sbi)) { 620 ret = submit_flush_wait(sbi, ino); 621 atomic_dec(&fcc->queued_flush); 622 623 atomic_inc(&fcc->issued_flush); 624 return ret; 625 } 626 627 cmd.ino = ino; 628 init_completion(&cmd.wait); 629 630 llist_add(&cmd.llnode, &fcc->issue_list); 631 632 /* 633 * update issue_list before we wake up issue_flush thread, this 634 * smp_mb() pairs with another barrier in ___wait_event(), see 635 * more details in comments of waitqueue_active(). 636 */ 637 smp_mb(); 638 639 if (waitqueue_active(&fcc->flush_wait_queue)) 640 wake_up(&fcc->flush_wait_queue); 641 642 if (fcc->f2fs_issue_flush) { 643 wait_for_completion(&cmd.wait); 644 atomic_dec(&fcc->queued_flush); 645 } else { 646 struct llist_node *list; 647 648 list = llist_del_all(&fcc->issue_list); 649 if (!list) { 650 wait_for_completion(&cmd.wait); 651 atomic_dec(&fcc->queued_flush); 652 } else { 653 struct flush_cmd *tmp, *next; 654 655 ret = submit_flush_wait(sbi, ino); 656 657 llist_for_each_entry_safe(tmp, next, list, llnode) { 658 if (tmp == &cmd) { 659 cmd.ret = ret; 660 atomic_dec(&fcc->queued_flush); 661 continue; 662 } 663 tmp->ret = ret; 664 complete(&tmp->wait); 665 } 666 } 667 } 668 669 return cmd.ret; 670 } 671 672 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi) 673 { 674 dev_t dev = sbi->sb->s_bdev->bd_dev; 675 struct flush_cmd_control *fcc; 676 677 if (SM_I(sbi)->fcc_info) { 678 fcc = SM_I(sbi)->fcc_info; 679 if (fcc->f2fs_issue_flush) 680 return 0; 681 goto init_thread; 682 } 683 684 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL); 685 if (!fcc) 686 return -ENOMEM; 687 atomic_set(&fcc->issued_flush, 0); 688 atomic_set(&fcc->queued_flush, 0); 689 init_waitqueue_head(&fcc->flush_wait_queue); 690 init_llist_head(&fcc->issue_list); 691 SM_I(sbi)->fcc_info = fcc; 692 if (!test_opt(sbi, FLUSH_MERGE)) 693 return 0; 694 695 init_thread: 696 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, 697 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); 698 if (IS_ERR(fcc->f2fs_issue_flush)) { 699 int err = PTR_ERR(fcc->f2fs_issue_flush); 700 701 fcc->f2fs_issue_flush = NULL; 702 return err; 703 } 704 705 return 0; 706 } 707 708 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free) 709 { 710 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 711 712 if (fcc && fcc->f2fs_issue_flush) { 713 struct task_struct *flush_thread = fcc->f2fs_issue_flush; 714 715 fcc->f2fs_issue_flush = NULL; 716 kthread_stop(flush_thread); 717 } 718 if (free) { 719 kfree(fcc); 720 SM_I(sbi)->fcc_info = NULL; 721 } 722 } 723 724 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi) 725 { 726 int ret = 0, i; 727 728 if (!f2fs_is_multi_device(sbi)) 729 return 0; 730 731 if (test_opt(sbi, NOBARRIER)) 732 return 0; 733 734 for (i = 1; i < sbi->s_ndevs; i++) { 735 int count = DEFAULT_RETRY_IO_COUNT; 736 737 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device)) 738 continue; 739 740 do { 741 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 742 if (ret) 743 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 744 } while (ret && --count); 745 746 if (ret) { 747 f2fs_stop_checkpoint(sbi, false, 748 STOP_CP_REASON_FLUSH_FAIL); 749 break; 750 } 751 752 spin_lock(&sbi->dev_lock); 753 f2fs_clear_bit(i, (char *)&sbi->dirty_device); 754 spin_unlock(&sbi->dev_lock); 755 } 756 757 return ret; 758 } 759 760 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 761 enum dirty_type dirty_type) 762 { 763 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 764 765 /* need not be added */ 766 if (IS_CURSEG(sbi, segno)) 767 return; 768 769 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 770 dirty_i->nr_dirty[dirty_type]++; 771 772 if (dirty_type == DIRTY) { 773 struct seg_entry *sentry = get_seg_entry(sbi, segno); 774 enum dirty_type t = sentry->type; 775 776 if (unlikely(t >= DIRTY)) { 777 f2fs_bug_on(sbi, 1); 778 return; 779 } 780 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) 781 dirty_i->nr_dirty[t]++; 782 783 if (__is_large_section(sbi)) { 784 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 785 block_t valid_blocks = 786 get_valid_blocks(sbi, segno, true); 787 788 f2fs_bug_on(sbi, 789 (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 790 !valid_blocks) || 791 valid_blocks == CAP_BLKS_PER_SEC(sbi)); 792 793 if (!IS_CURSEC(sbi, secno)) 794 set_bit(secno, dirty_i->dirty_secmap); 795 } 796 } 797 } 798 799 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 800 enum dirty_type dirty_type) 801 { 802 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 803 block_t valid_blocks; 804 805 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) 806 dirty_i->nr_dirty[dirty_type]--; 807 808 if (dirty_type == DIRTY) { 809 struct seg_entry *sentry = get_seg_entry(sbi, segno); 810 enum dirty_type t = sentry->type; 811 812 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) 813 dirty_i->nr_dirty[t]--; 814 815 valid_blocks = get_valid_blocks(sbi, segno, true); 816 if (valid_blocks == 0) { 817 clear_bit(GET_SEC_FROM_SEG(sbi, segno), 818 dirty_i->victim_secmap); 819 #ifdef CONFIG_F2FS_CHECK_FS 820 clear_bit(segno, SIT_I(sbi)->invalid_segmap); 821 #endif 822 } 823 if (__is_large_section(sbi)) { 824 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 825 826 if (!valid_blocks || 827 valid_blocks == CAP_BLKS_PER_SEC(sbi)) { 828 clear_bit(secno, dirty_i->dirty_secmap); 829 return; 830 } 831 832 if (!IS_CURSEC(sbi, secno)) 833 set_bit(secno, dirty_i->dirty_secmap); 834 } 835 } 836 } 837 838 /* 839 * Should not occur error such as -ENOMEM. 840 * Adding dirty entry into seglist is not critical operation. 841 * If a given segment is one of current working segments, it won't be added. 842 */ 843 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) 844 { 845 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 846 unsigned short valid_blocks, ckpt_valid_blocks; 847 unsigned int usable_blocks; 848 849 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) 850 return; 851 852 usable_blocks = f2fs_usable_blks_in_seg(sbi, segno); 853 mutex_lock(&dirty_i->seglist_lock); 854 855 valid_blocks = get_valid_blocks(sbi, segno, false); 856 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false); 857 858 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) || 859 ckpt_valid_blocks == usable_blocks)) { 860 __locate_dirty_segment(sbi, segno, PRE); 861 __remove_dirty_segment(sbi, segno, DIRTY); 862 } else if (valid_blocks < usable_blocks) { 863 __locate_dirty_segment(sbi, segno, DIRTY); 864 } else { 865 /* Recovery routine with SSR needs this */ 866 __remove_dirty_segment(sbi, segno, DIRTY); 867 } 868 869 mutex_unlock(&dirty_i->seglist_lock); 870 } 871 872 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */ 873 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi) 874 { 875 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 876 unsigned int segno; 877 878 mutex_lock(&dirty_i->seglist_lock); 879 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 880 if (get_valid_blocks(sbi, segno, false)) 881 continue; 882 if (IS_CURSEG(sbi, segno)) 883 continue; 884 __locate_dirty_segment(sbi, segno, PRE); 885 __remove_dirty_segment(sbi, segno, DIRTY); 886 } 887 mutex_unlock(&dirty_i->seglist_lock); 888 } 889 890 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi) 891 { 892 int ovp_hole_segs = 893 (overprovision_segments(sbi) - reserved_segments(sbi)); 894 block_t ovp_holes = SEGS_TO_BLKS(sbi, ovp_hole_segs); 895 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 896 block_t holes[2] = {0, 0}; /* DATA and NODE */ 897 block_t unusable; 898 struct seg_entry *se; 899 unsigned int segno; 900 901 mutex_lock(&dirty_i->seglist_lock); 902 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 903 se = get_seg_entry(sbi, segno); 904 if (IS_NODESEG(se->type)) 905 holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) - 906 se->valid_blocks; 907 else 908 holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) - 909 se->valid_blocks; 910 } 911 mutex_unlock(&dirty_i->seglist_lock); 912 913 unusable = max(holes[DATA], holes[NODE]); 914 if (unusable > ovp_holes) 915 return unusable - ovp_holes; 916 return 0; 917 } 918 919 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable) 920 { 921 int ovp_hole_segs = 922 (overprovision_segments(sbi) - reserved_segments(sbi)); 923 924 if (F2FS_OPTION(sbi).unusable_cap_perc == 100) 925 return 0; 926 if (unusable > F2FS_OPTION(sbi).unusable_cap) 927 return -EAGAIN; 928 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) && 929 dirty_segments(sbi) > ovp_hole_segs) 930 return -EAGAIN; 931 if (has_not_enough_free_secs(sbi, 0, 0)) 932 return -EAGAIN; 933 return 0; 934 } 935 936 /* This is only used by SBI_CP_DISABLED */ 937 static unsigned int get_free_segment(struct f2fs_sb_info *sbi) 938 { 939 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 940 unsigned int segno = 0; 941 942 mutex_lock(&dirty_i->seglist_lock); 943 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 944 if (get_valid_blocks(sbi, segno, false)) 945 continue; 946 if (get_ckpt_valid_blocks(sbi, segno, false)) 947 continue; 948 mutex_unlock(&dirty_i->seglist_lock); 949 return segno; 950 } 951 mutex_unlock(&dirty_i->seglist_lock); 952 return NULL_SEGNO; 953 } 954 955 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi, 956 struct block_device *bdev, block_t lstart, 957 block_t start, block_t len) 958 { 959 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 960 struct list_head *pend_list; 961 struct discard_cmd *dc; 962 963 f2fs_bug_on(sbi, !len); 964 965 pend_list = &dcc->pend_list[plist_idx(len)]; 966 967 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL); 968 INIT_LIST_HEAD(&dc->list); 969 dc->bdev = bdev; 970 dc->di.lstart = lstart; 971 dc->di.start = start; 972 dc->di.len = len; 973 dc->ref = 0; 974 dc->state = D_PREP; 975 dc->queued = 0; 976 dc->error = 0; 977 init_completion(&dc->wait); 978 list_add_tail(&dc->list, pend_list); 979 spin_lock_init(&dc->lock); 980 dc->bio_ref = 0; 981 atomic_inc(&dcc->discard_cmd_cnt); 982 dcc->undiscard_blks += len; 983 984 return dc; 985 } 986 987 static bool f2fs_check_discard_tree(struct f2fs_sb_info *sbi) 988 { 989 #ifdef CONFIG_F2FS_CHECK_FS 990 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 991 struct rb_node *cur = rb_first_cached(&dcc->root), *next; 992 struct discard_cmd *cur_dc, *next_dc; 993 994 while (cur) { 995 next = rb_next(cur); 996 if (!next) 997 return true; 998 999 cur_dc = rb_entry(cur, struct discard_cmd, rb_node); 1000 next_dc = rb_entry(next, struct discard_cmd, rb_node); 1001 1002 if (cur_dc->di.lstart + cur_dc->di.len > next_dc->di.lstart) { 1003 f2fs_info(sbi, "broken discard_rbtree, " 1004 "cur(%u, %u) next(%u, %u)", 1005 cur_dc->di.lstart, cur_dc->di.len, 1006 next_dc->di.lstart, next_dc->di.len); 1007 return false; 1008 } 1009 cur = next; 1010 } 1011 #endif 1012 return true; 1013 } 1014 1015 static struct discard_cmd *__lookup_discard_cmd(struct f2fs_sb_info *sbi, 1016 block_t blkaddr) 1017 { 1018 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1019 struct rb_node *node = dcc->root.rb_root.rb_node; 1020 struct discard_cmd *dc; 1021 1022 while (node) { 1023 dc = rb_entry(node, struct discard_cmd, rb_node); 1024 1025 if (blkaddr < dc->di.lstart) 1026 node = node->rb_left; 1027 else if (blkaddr >= dc->di.lstart + dc->di.len) 1028 node = node->rb_right; 1029 else 1030 return dc; 1031 } 1032 return NULL; 1033 } 1034 1035 static struct discard_cmd *__lookup_discard_cmd_ret(struct rb_root_cached *root, 1036 block_t blkaddr, 1037 struct discard_cmd **prev_entry, 1038 struct discard_cmd **next_entry, 1039 struct rb_node ***insert_p, 1040 struct rb_node **insert_parent) 1041 { 1042 struct rb_node **pnode = &root->rb_root.rb_node; 1043 struct rb_node *parent = NULL, *tmp_node; 1044 struct discard_cmd *dc; 1045 1046 *insert_p = NULL; 1047 *insert_parent = NULL; 1048 *prev_entry = NULL; 1049 *next_entry = NULL; 1050 1051 if (RB_EMPTY_ROOT(&root->rb_root)) 1052 return NULL; 1053 1054 while (*pnode) { 1055 parent = *pnode; 1056 dc = rb_entry(*pnode, struct discard_cmd, rb_node); 1057 1058 if (blkaddr < dc->di.lstart) 1059 pnode = &(*pnode)->rb_left; 1060 else if (blkaddr >= dc->di.lstart + dc->di.len) 1061 pnode = &(*pnode)->rb_right; 1062 else 1063 goto lookup_neighbors; 1064 } 1065 1066 *insert_p = pnode; 1067 *insert_parent = parent; 1068 1069 dc = rb_entry(parent, struct discard_cmd, rb_node); 1070 tmp_node = parent; 1071 if (parent && blkaddr > dc->di.lstart) 1072 tmp_node = rb_next(parent); 1073 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1074 1075 tmp_node = parent; 1076 if (parent && blkaddr < dc->di.lstart) 1077 tmp_node = rb_prev(parent); 1078 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1079 return NULL; 1080 1081 lookup_neighbors: 1082 /* lookup prev node for merging backward later */ 1083 tmp_node = rb_prev(&dc->rb_node); 1084 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1085 1086 /* lookup next node for merging frontward later */ 1087 tmp_node = rb_next(&dc->rb_node); 1088 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1089 return dc; 1090 } 1091 1092 static void __detach_discard_cmd(struct discard_cmd_control *dcc, 1093 struct discard_cmd *dc) 1094 { 1095 if (dc->state == D_DONE) 1096 atomic_sub(dc->queued, &dcc->queued_discard); 1097 1098 list_del(&dc->list); 1099 rb_erase_cached(&dc->rb_node, &dcc->root); 1100 dcc->undiscard_blks -= dc->di.len; 1101 1102 kmem_cache_free(discard_cmd_slab, dc); 1103 1104 atomic_dec(&dcc->discard_cmd_cnt); 1105 } 1106 1107 static void __remove_discard_cmd(struct f2fs_sb_info *sbi, 1108 struct discard_cmd *dc) 1109 { 1110 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1111 unsigned long flags; 1112 1113 trace_f2fs_remove_discard(dc->bdev, dc->di.start, dc->di.len); 1114 1115 spin_lock_irqsave(&dc->lock, flags); 1116 if (dc->bio_ref) { 1117 spin_unlock_irqrestore(&dc->lock, flags); 1118 return; 1119 } 1120 spin_unlock_irqrestore(&dc->lock, flags); 1121 1122 f2fs_bug_on(sbi, dc->ref); 1123 1124 if (dc->error == -EOPNOTSUPP) 1125 dc->error = 0; 1126 1127 if (dc->error) 1128 f2fs_info_ratelimited(sbi, 1129 "Issue discard(%u, %u, %u) failed, ret: %d", 1130 dc->di.lstart, dc->di.start, dc->di.len, dc->error); 1131 __detach_discard_cmd(dcc, dc); 1132 } 1133 1134 static void f2fs_submit_discard_endio(struct bio *bio) 1135 { 1136 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private; 1137 unsigned long flags; 1138 1139 spin_lock_irqsave(&dc->lock, flags); 1140 if (!dc->error) 1141 dc->error = blk_status_to_errno(bio->bi_status); 1142 dc->bio_ref--; 1143 if (!dc->bio_ref && dc->state == D_SUBMIT) { 1144 dc->state = D_DONE; 1145 complete_all(&dc->wait); 1146 } 1147 spin_unlock_irqrestore(&dc->lock, flags); 1148 bio_put(bio); 1149 } 1150 1151 static void __check_sit_bitmap(struct f2fs_sb_info *sbi, 1152 block_t start, block_t end) 1153 { 1154 #ifdef CONFIG_F2FS_CHECK_FS 1155 struct seg_entry *sentry; 1156 unsigned int segno; 1157 block_t blk = start; 1158 unsigned long offset, size, *map; 1159 1160 while (blk < end) { 1161 segno = GET_SEGNO(sbi, blk); 1162 sentry = get_seg_entry(sbi, segno); 1163 offset = GET_BLKOFF_FROM_SEG0(sbi, blk); 1164 1165 if (end < START_BLOCK(sbi, segno + 1)) 1166 size = GET_BLKOFF_FROM_SEG0(sbi, end); 1167 else 1168 size = BLKS_PER_SEG(sbi); 1169 map = (unsigned long *)(sentry->cur_valid_map); 1170 offset = __find_rev_next_bit(map, size, offset); 1171 f2fs_bug_on(sbi, offset != size); 1172 blk = START_BLOCK(sbi, segno + 1); 1173 } 1174 #endif 1175 } 1176 1177 static void __init_discard_policy(struct f2fs_sb_info *sbi, 1178 struct discard_policy *dpolicy, 1179 int discard_type, unsigned int granularity) 1180 { 1181 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1182 1183 /* common policy */ 1184 dpolicy->type = discard_type; 1185 dpolicy->sync = true; 1186 dpolicy->ordered = false; 1187 dpolicy->granularity = granularity; 1188 1189 dpolicy->max_requests = dcc->max_discard_request; 1190 dpolicy->io_aware_gran = dcc->discard_io_aware_gran; 1191 dpolicy->timeout = false; 1192 1193 if (discard_type == DPOLICY_BG) { 1194 dpolicy->min_interval = dcc->min_discard_issue_time; 1195 dpolicy->mid_interval = dcc->mid_discard_issue_time; 1196 dpolicy->max_interval = dcc->max_discard_issue_time; 1197 if (dcc->discard_io_aware == DPOLICY_IO_AWARE_ENABLE) 1198 dpolicy->io_aware = true; 1199 else if (dcc->discard_io_aware == DPOLICY_IO_AWARE_DISABLE) 1200 dpolicy->io_aware = false; 1201 dpolicy->sync = false; 1202 dpolicy->ordered = true; 1203 if (utilization(sbi) > dcc->discard_urgent_util) { 1204 dpolicy->granularity = MIN_DISCARD_GRANULARITY; 1205 if (atomic_read(&dcc->discard_cmd_cnt)) 1206 dpolicy->max_interval = 1207 dcc->min_discard_issue_time; 1208 } 1209 } else if (discard_type == DPOLICY_FORCE) { 1210 dpolicy->min_interval = dcc->min_discard_issue_time; 1211 dpolicy->mid_interval = dcc->mid_discard_issue_time; 1212 dpolicy->max_interval = dcc->max_discard_issue_time; 1213 dpolicy->io_aware = false; 1214 } else if (discard_type == DPOLICY_FSTRIM) { 1215 dpolicy->io_aware = false; 1216 } else if (discard_type == DPOLICY_UMOUNT) { 1217 dpolicy->io_aware = false; 1218 /* we need to issue all to keep CP_TRIMMED_FLAG */ 1219 dpolicy->granularity = MIN_DISCARD_GRANULARITY; 1220 dpolicy->timeout = true; 1221 } 1222 } 1223 1224 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1225 struct block_device *bdev, block_t lstart, 1226 block_t start, block_t len); 1227 1228 #ifdef CONFIG_BLK_DEV_ZONED 1229 static void __submit_zone_reset_cmd(struct f2fs_sb_info *sbi, 1230 struct discard_cmd *dc, blk_opf_t flag, 1231 struct list_head *wait_list, 1232 unsigned int *issued) 1233 { 1234 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1235 struct block_device *bdev = dc->bdev; 1236 struct bio *bio = bio_alloc(bdev, 0, REQ_OP_ZONE_RESET | flag, GFP_NOFS); 1237 unsigned long flags; 1238 1239 trace_f2fs_issue_reset_zone(bdev, dc->di.start); 1240 1241 spin_lock_irqsave(&dc->lock, flags); 1242 dc->state = D_SUBMIT; 1243 dc->bio_ref++; 1244 spin_unlock_irqrestore(&dc->lock, flags); 1245 1246 if (issued) 1247 (*issued)++; 1248 1249 atomic_inc(&dcc->queued_discard); 1250 dc->queued++; 1251 list_move_tail(&dc->list, wait_list); 1252 1253 /* sanity check on discard range */ 1254 __check_sit_bitmap(sbi, dc->di.lstart, dc->di.lstart + dc->di.len); 1255 1256 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(dc->di.start); 1257 bio->bi_private = dc; 1258 bio->bi_end_io = f2fs_submit_discard_endio; 1259 submit_bio(bio); 1260 1261 atomic_inc(&dcc->issued_discard); 1262 f2fs_update_iostat(sbi, NULL, FS_ZONE_RESET_IO, dc->di.len * F2FS_BLKSIZE); 1263 } 1264 #endif 1265 1266 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */ 1267 static int __submit_discard_cmd(struct f2fs_sb_info *sbi, 1268 struct discard_policy *dpolicy, 1269 struct discard_cmd *dc, int *issued) 1270 { 1271 struct block_device *bdev = dc->bdev; 1272 unsigned int max_discard_blocks = 1273 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev)); 1274 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1275 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1276 &(dcc->fstrim_list) : &(dcc->wait_list); 1277 blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0; 1278 block_t lstart, start, len, total_len; 1279 int err = 0; 1280 1281 if (dc->state != D_PREP) 1282 return 0; 1283 1284 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 1285 return 0; 1286 1287 #ifdef CONFIG_BLK_DEV_ZONED 1288 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) { 1289 int devi = f2fs_bdev_index(sbi, bdev); 1290 1291 if (devi < 0) 1292 return -EINVAL; 1293 1294 if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) { 1295 __submit_zone_reset_cmd(sbi, dc, flag, 1296 wait_list, issued); 1297 return 0; 1298 } 1299 } 1300 #endif 1301 1302 /* 1303 * stop issuing discard for any of below cases: 1304 * 1. device is conventional zone, but it doesn't support discard. 1305 * 2. device is regulare device, after snapshot it doesn't support 1306 * discard. 1307 */ 1308 if (!bdev_max_discard_sectors(bdev)) 1309 return -EOPNOTSUPP; 1310 1311 trace_f2fs_issue_discard(bdev, dc->di.start, dc->di.len); 1312 1313 lstart = dc->di.lstart; 1314 start = dc->di.start; 1315 len = dc->di.len; 1316 total_len = len; 1317 1318 dc->di.len = 0; 1319 1320 while (total_len && *issued < dpolicy->max_requests && !err) { 1321 struct bio *bio = NULL; 1322 unsigned long flags; 1323 bool last = true; 1324 1325 if (len > max_discard_blocks) { 1326 len = max_discard_blocks; 1327 last = false; 1328 } 1329 1330 (*issued)++; 1331 if (*issued == dpolicy->max_requests) 1332 last = true; 1333 1334 dc->di.len += len; 1335 1336 if (time_to_inject(sbi, FAULT_DISCARD)) { 1337 err = -EIO; 1338 } else { 1339 err = __blkdev_issue_discard(bdev, 1340 SECTOR_FROM_BLOCK(start), 1341 SECTOR_FROM_BLOCK(len), 1342 GFP_NOFS, &bio); 1343 } 1344 if (err) { 1345 spin_lock_irqsave(&dc->lock, flags); 1346 if (dc->state == D_PARTIAL) 1347 dc->state = D_SUBMIT; 1348 spin_unlock_irqrestore(&dc->lock, flags); 1349 1350 break; 1351 } 1352 1353 f2fs_bug_on(sbi, !bio); 1354 1355 /* 1356 * should keep before submission to avoid D_DONE 1357 * right away 1358 */ 1359 spin_lock_irqsave(&dc->lock, flags); 1360 if (last) 1361 dc->state = D_SUBMIT; 1362 else 1363 dc->state = D_PARTIAL; 1364 dc->bio_ref++; 1365 spin_unlock_irqrestore(&dc->lock, flags); 1366 1367 atomic_inc(&dcc->queued_discard); 1368 dc->queued++; 1369 list_move_tail(&dc->list, wait_list); 1370 1371 /* sanity check on discard range */ 1372 __check_sit_bitmap(sbi, lstart, lstart + len); 1373 1374 bio->bi_private = dc; 1375 bio->bi_end_io = f2fs_submit_discard_endio; 1376 bio->bi_opf |= flag; 1377 submit_bio(bio); 1378 1379 atomic_inc(&dcc->issued_discard); 1380 1381 f2fs_update_iostat(sbi, NULL, FS_DISCARD_IO, len * F2FS_BLKSIZE); 1382 1383 lstart += len; 1384 start += len; 1385 total_len -= len; 1386 len = total_len; 1387 } 1388 1389 if (!err && len) { 1390 dcc->undiscard_blks -= len; 1391 __update_discard_tree_range(sbi, bdev, lstart, start, len); 1392 } 1393 return err; 1394 } 1395 1396 static void __insert_discard_cmd(struct f2fs_sb_info *sbi, 1397 struct block_device *bdev, block_t lstart, 1398 block_t start, block_t len) 1399 { 1400 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1401 struct rb_node **p = &dcc->root.rb_root.rb_node; 1402 struct rb_node *parent = NULL; 1403 struct discard_cmd *dc; 1404 bool leftmost = true; 1405 1406 /* look up rb tree to find parent node */ 1407 while (*p) { 1408 parent = *p; 1409 dc = rb_entry(parent, struct discard_cmd, rb_node); 1410 1411 if (lstart < dc->di.lstart) { 1412 p = &(*p)->rb_left; 1413 } else if (lstart >= dc->di.lstart + dc->di.len) { 1414 p = &(*p)->rb_right; 1415 leftmost = false; 1416 } else { 1417 /* Let's skip to add, if exists */ 1418 return; 1419 } 1420 } 1421 1422 dc = __create_discard_cmd(sbi, bdev, lstart, start, len); 1423 1424 rb_link_node(&dc->rb_node, parent, p); 1425 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost); 1426 } 1427 1428 static void __relocate_discard_cmd(struct discard_cmd_control *dcc, 1429 struct discard_cmd *dc) 1430 { 1431 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->di.len)]); 1432 } 1433 1434 static void __punch_discard_cmd(struct f2fs_sb_info *sbi, 1435 struct discard_cmd *dc, block_t blkaddr) 1436 { 1437 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1438 struct discard_info di = dc->di; 1439 bool modified = false; 1440 1441 if (dc->state == D_DONE || dc->di.len == 1) { 1442 __remove_discard_cmd(sbi, dc); 1443 return; 1444 } 1445 1446 dcc->undiscard_blks -= di.len; 1447 1448 if (blkaddr > di.lstart) { 1449 dc->di.len = blkaddr - dc->di.lstart; 1450 dcc->undiscard_blks += dc->di.len; 1451 __relocate_discard_cmd(dcc, dc); 1452 modified = true; 1453 } 1454 1455 if (blkaddr < di.lstart + di.len - 1) { 1456 if (modified) { 1457 __insert_discard_cmd(sbi, dc->bdev, blkaddr + 1, 1458 di.start + blkaddr + 1 - di.lstart, 1459 di.lstart + di.len - 1 - blkaddr); 1460 } else { 1461 dc->di.lstart++; 1462 dc->di.len--; 1463 dc->di.start++; 1464 dcc->undiscard_blks += dc->di.len; 1465 __relocate_discard_cmd(dcc, dc); 1466 } 1467 } 1468 } 1469 1470 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1471 struct block_device *bdev, block_t lstart, 1472 block_t start, block_t len) 1473 { 1474 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1475 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1476 struct discard_cmd *dc; 1477 struct discard_info di = {0}; 1478 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1479 unsigned int max_discard_blocks = 1480 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev)); 1481 block_t end = lstart + len; 1482 1483 dc = __lookup_discard_cmd_ret(&dcc->root, lstart, 1484 &prev_dc, &next_dc, &insert_p, &insert_parent); 1485 if (dc) 1486 prev_dc = dc; 1487 1488 if (!prev_dc) { 1489 di.lstart = lstart; 1490 di.len = next_dc ? next_dc->di.lstart - lstart : len; 1491 di.len = min(di.len, len); 1492 di.start = start; 1493 } 1494 1495 while (1) { 1496 struct rb_node *node; 1497 bool merged = false; 1498 struct discard_cmd *tdc = NULL; 1499 1500 if (prev_dc) { 1501 di.lstart = prev_dc->di.lstart + prev_dc->di.len; 1502 if (di.lstart < lstart) 1503 di.lstart = lstart; 1504 if (di.lstart >= end) 1505 break; 1506 1507 if (!next_dc || next_dc->di.lstart > end) 1508 di.len = end - di.lstart; 1509 else 1510 di.len = next_dc->di.lstart - di.lstart; 1511 di.start = start + di.lstart - lstart; 1512 } 1513 1514 if (!di.len) 1515 goto next; 1516 1517 if (prev_dc && prev_dc->state == D_PREP && 1518 prev_dc->bdev == bdev && 1519 __is_discard_back_mergeable(&di, &prev_dc->di, 1520 max_discard_blocks)) { 1521 prev_dc->di.len += di.len; 1522 dcc->undiscard_blks += di.len; 1523 __relocate_discard_cmd(dcc, prev_dc); 1524 di = prev_dc->di; 1525 tdc = prev_dc; 1526 merged = true; 1527 } 1528 1529 if (next_dc && next_dc->state == D_PREP && 1530 next_dc->bdev == bdev && 1531 __is_discard_front_mergeable(&di, &next_dc->di, 1532 max_discard_blocks)) { 1533 next_dc->di.lstart = di.lstart; 1534 next_dc->di.len += di.len; 1535 next_dc->di.start = di.start; 1536 dcc->undiscard_blks += di.len; 1537 __relocate_discard_cmd(dcc, next_dc); 1538 if (tdc) 1539 __remove_discard_cmd(sbi, tdc); 1540 merged = true; 1541 } 1542 1543 if (!merged) 1544 __insert_discard_cmd(sbi, bdev, 1545 di.lstart, di.start, di.len); 1546 next: 1547 prev_dc = next_dc; 1548 if (!prev_dc) 1549 break; 1550 1551 node = rb_next(&prev_dc->rb_node); 1552 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1553 } 1554 } 1555 1556 #ifdef CONFIG_BLK_DEV_ZONED 1557 static void __queue_zone_reset_cmd(struct f2fs_sb_info *sbi, 1558 struct block_device *bdev, block_t blkstart, block_t lblkstart, 1559 block_t blklen) 1560 { 1561 trace_f2fs_queue_reset_zone(bdev, blkstart); 1562 1563 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock); 1564 __insert_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen); 1565 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock); 1566 } 1567 #endif 1568 1569 static void __queue_discard_cmd(struct f2fs_sb_info *sbi, 1570 struct block_device *bdev, block_t blkstart, block_t blklen) 1571 { 1572 block_t lblkstart = blkstart; 1573 1574 if (!f2fs_bdev_support_discard(bdev)) 1575 return; 1576 1577 trace_f2fs_queue_discard(bdev, blkstart, blklen); 1578 1579 if (f2fs_is_multi_device(sbi)) { 1580 int devi = f2fs_target_device_index(sbi, blkstart); 1581 1582 blkstart -= FDEV(devi).start_blk; 1583 } 1584 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock); 1585 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen); 1586 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock); 1587 } 1588 1589 static void __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi, 1590 struct discard_policy *dpolicy, int *issued) 1591 { 1592 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1593 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1594 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1595 struct discard_cmd *dc; 1596 struct blk_plug plug; 1597 bool io_interrupted = false; 1598 1599 mutex_lock(&dcc->cmd_lock); 1600 dc = __lookup_discard_cmd_ret(&dcc->root, dcc->next_pos, 1601 &prev_dc, &next_dc, &insert_p, &insert_parent); 1602 if (!dc) 1603 dc = next_dc; 1604 1605 blk_start_plug(&plug); 1606 1607 while (dc) { 1608 struct rb_node *node; 1609 int err = 0; 1610 1611 if (dc->state != D_PREP) 1612 goto next; 1613 1614 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) { 1615 io_interrupted = true; 1616 break; 1617 } 1618 1619 dcc->next_pos = dc->di.lstart + dc->di.len; 1620 err = __submit_discard_cmd(sbi, dpolicy, dc, issued); 1621 1622 if (*issued >= dpolicy->max_requests) 1623 break; 1624 next: 1625 node = rb_next(&dc->rb_node); 1626 if (err) 1627 __remove_discard_cmd(sbi, dc); 1628 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1629 } 1630 1631 blk_finish_plug(&plug); 1632 1633 if (!dc) 1634 dcc->next_pos = 0; 1635 1636 mutex_unlock(&dcc->cmd_lock); 1637 1638 if (!(*issued) && io_interrupted) 1639 *issued = -1; 1640 } 1641 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1642 struct discard_policy *dpolicy); 1643 1644 static int __issue_discard_cmd(struct f2fs_sb_info *sbi, 1645 struct discard_policy *dpolicy) 1646 { 1647 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1648 struct list_head *pend_list; 1649 struct discard_cmd *dc, *tmp; 1650 struct blk_plug plug; 1651 int i, issued; 1652 bool io_interrupted = false; 1653 1654 if (dpolicy->timeout) 1655 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT); 1656 1657 retry: 1658 issued = 0; 1659 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1660 if (dpolicy->timeout && 1661 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1662 break; 1663 1664 if (i + 1 < dpolicy->granularity) 1665 break; 1666 1667 if (i + 1 < dcc->max_ordered_discard && dpolicy->ordered) { 1668 __issue_discard_cmd_orderly(sbi, dpolicy, &issued); 1669 return issued; 1670 } 1671 1672 pend_list = &dcc->pend_list[i]; 1673 1674 mutex_lock(&dcc->cmd_lock); 1675 if (list_empty(pend_list)) 1676 goto next; 1677 if (unlikely(dcc->rbtree_check)) 1678 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi)); 1679 blk_start_plug(&plug); 1680 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1681 f2fs_bug_on(sbi, dc->state != D_PREP); 1682 1683 if (dpolicy->timeout && 1684 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1685 break; 1686 1687 if (dpolicy->io_aware && i < dpolicy->io_aware_gran && 1688 !is_idle(sbi, DISCARD_TIME)) { 1689 io_interrupted = true; 1690 break; 1691 } 1692 1693 __submit_discard_cmd(sbi, dpolicy, dc, &issued); 1694 1695 if (issued >= dpolicy->max_requests) 1696 break; 1697 } 1698 blk_finish_plug(&plug); 1699 next: 1700 mutex_unlock(&dcc->cmd_lock); 1701 1702 if (issued >= dpolicy->max_requests || io_interrupted) 1703 break; 1704 } 1705 1706 if (dpolicy->type == DPOLICY_UMOUNT && issued) { 1707 __wait_all_discard_cmd(sbi, dpolicy); 1708 goto retry; 1709 } 1710 1711 if (!issued && io_interrupted) 1712 issued = -1; 1713 1714 return issued; 1715 } 1716 1717 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi) 1718 { 1719 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1720 struct list_head *pend_list; 1721 struct discard_cmd *dc, *tmp; 1722 int i; 1723 bool dropped = false; 1724 1725 mutex_lock(&dcc->cmd_lock); 1726 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1727 pend_list = &dcc->pend_list[i]; 1728 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1729 f2fs_bug_on(sbi, dc->state != D_PREP); 1730 __remove_discard_cmd(sbi, dc); 1731 dropped = true; 1732 } 1733 } 1734 mutex_unlock(&dcc->cmd_lock); 1735 1736 return dropped; 1737 } 1738 1739 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi) 1740 { 1741 __drop_discard_cmd(sbi); 1742 } 1743 1744 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi, 1745 struct discard_cmd *dc) 1746 { 1747 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1748 unsigned int len = 0; 1749 1750 wait_for_completion_io(&dc->wait); 1751 mutex_lock(&dcc->cmd_lock); 1752 f2fs_bug_on(sbi, dc->state != D_DONE); 1753 dc->ref--; 1754 if (!dc->ref) { 1755 if (!dc->error) 1756 len = dc->di.len; 1757 __remove_discard_cmd(sbi, dc); 1758 } 1759 mutex_unlock(&dcc->cmd_lock); 1760 1761 return len; 1762 } 1763 1764 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi, 1765 struct discard_policy *dpolicy, 1766 block_t start, block_t end) 1767 { 1768 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1769 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1770 &(dcc->fstrim_list) : &(dcc->wait_list); 1771 struct discard_cmd *dc = NULL, *iter, *tmp; 1772 unsigned int trimmed = 0; 1773 1774 next: 1775 dc = NULL; 1776 1777 mutex_lock(&dcc->cmd_lock); 1778 list_for_each_entry_safe(iter, tmp, wait_list, list) { 1779 if (iter->di.lstart + iter->di.len <= start || 1780 end <= iter->di.lstart) 1781 continue; 1782 if (iter->di.len < dpolicy->granularity) 1783 continue; 1784 if (iter->state == D_DONE && !iter->ref) { 1785 wait_for_completion_io(&iter->wait); 1786 if (!iter->error) 1787 trimmed += iter->di.len; 1788 __remove_discard_cmd(sbi, iter); 1789 } else { 1790 iter->ref++; 1791 dc = iter; 1792 break; 1793 } 1794 } 1795 mutex_unlock(&dcc->cmd_lock); 1796 1797 if (dc) { 1798 trimmed += __wait_one_discard_bio(sbi, dc); 1799 goto next; 1800 } 1801 1802 return trimmed; 1803 } 1804 1805 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1806 struct discard_policy *dpolicy) 1807 { 1808 struct discard_policy dp; 1809 unsigned int discard_blks; 1810 1811 if (dpolicy) 1812 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX); 1813 1814 /* wait all */ 1815 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, MIN_DISCARD_GRANULARITY); 1816 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1817 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, MIN_DISCARD_GRANULARITY); 1818 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1819 1820 return discard_blks; 1821 } 1822 1823 /* This should be covered by global mutex, &sit_i->sentry_lock */ 1824 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr) 1825 { 1826 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1827 struct discard_cmd *dc; 1828 bool need_wait = false; 1829 1830 mutex_lock(&dcc->cmd_lock); 1831 dc = __lookup_discard_cmd(sbi, blkaddr); 1832 #ifdef CONFIG_BLK_DEV_ZONED 1833 if (dc && f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(dc->bdev)) { 1834 int devi = f2fs_bdev_index(sbi, dc->bdev); 1835 1836 if (devi < 0) { 1837 mutex_unlock(&dcc->cmd_lock); 1838 return; 1839 } 1840 1841 if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) { 1842 /* force submit zone reset */ 1843 if (dc->state == D_PREP) 1844 __submit_zone_reset_cmd(sbi, dc, REQ_SYNC, 1845 &dcc->wait_list, NULL); 1846 dc->ref++; 1847 mutex_unlock(&dcc->cmd_lock); 1848 /* wait zone reset */ 1849 __wait_one_discard_bio(sbi, dc); 1850 return; 1851 } 1852 } 1853 #endif 1854 if (dc) { 1855 if (dc->state == D_PREP) { 1856 __punch_discard_cmd(sbi, dc, blkaddr); 1857 } else { 1858 dc->ref++; 1859 need_wait = true; 1860 } 1861 } 1862 mutex_unlock(&dcc->cmd_lock); 1863 1864 if (need_wait) 1865 __wait_one_discard_bio(sbi, dc); 1866 } 1867 1868 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi) 1869 { 1870 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1871 1872 if (dcc && dcc->f2fs_issue_discard) { 1873 struct task_struct *discard_thread = dcc->f2fs_issue_discard; 1874 1875 dcc->f2fs_issue_discard = NULL; 1876 kthread_stop(discard_thread); 1877 } 1878 } 1879 1880 /** 1881 * f2fs_issue_discard_timeout() - Issue all discard cmd within UMOUNT_DISCARD_TIMEOUT 1882 * @sbi: the f2fs_sb_info data for discard cmd to issue 1883 * 1884 * When UMOUNT_DISCARD_TIMEOUT is exceeded, all remaining discard commands will be dropped 1885 * 1886 * Return true if issued all discard cmd or no discard cmd need issue, otherwise return false. 1887 */ 1888 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi) 1889 { 1890 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1891 struct discard_policy dpolicy; 1892 bool dropped; 1893 1894 if (!atomic_read(&dcc->discard_cmd_cnt)) 1895 return true; 1896 1897 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT, 1898 dcc->discard_granularity); 1899 __issue_discard_cmd(sbi, &dpolicy); 1900 dropped = __drop_discard_cmd(sbi); 1901 1902 /* just to make sure there is no pending discard commands */ 1903 __wait_all_discard_cmd(sbi, NULL); 1904 1905 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt)); 1906 return !dropped; 1907 } 1908 1909 static int issue_discard_thread(void *data) 1910 { 1911 struct f2fs_sb_info *sbi = data; 1912 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1913 wait_queue_head_t *q = &dcc->discard_wait_queue; 1914 struct discard_policy dpolicy; 1915 unsigned int wait_ms = dcc->min_discard_issue_time; 1916 int issued; 1917 1918 set_freezable(); 1919 1920 do { 1921 wait_event_freezable_timeout(*q, 1922 kthread_should_stop() || dcc->discard_wake, 1923 msecs_to_jiffies(wait_ms)); 1924 1925 if (sbi->gc_mode == GC_URGENT_HIGH || 1926 !f2fs_available_free_memory(sbi, DISCARD_CACHE)) 1927 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1928 MIN_DISCARD_GRANULARITY); 1929 else 1930 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG, 1931 dcc->discard_granularity); 1932 1933 if (dcc->discard_wake) 1934 dcc->discard_wake = false; 1935 1936 /* clean up pending candidates before going to sleep */ 1937 if (atomic_read(&dcc->queued_discard)) 1938 __wait_all_discard_cmd(sbi, NULL); 1939 1940 if (f2fs_readonly(sbi->sb)) 1941 continue; 1942 if (kthread_should_stop()) 1943 return 0; 1944 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) || 1945 !atomic_read(&dcc->discard_cmd_cnt)) { 1946 wait_ms = dpolicy.max_interval; 1947 continue; 1948 } 1949 1950 sb_start_intwrite(sbi->sb); 1951 1952 issued = __issue_discard_cmd(sbi, &dpolicy); 1953 if (issued > 0) { 1954 __wait_all_discard_cmd(sbi, &dpolicy); 1955 wait_ms = dpolicy.min_interval; 1956 } else if (issued == -1) { 1957 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME); 1958 if (!wait_ms) 1959 wait_ms = dpolicy.mid_interval; 1960 } else { 1961 wait_ms = dpolicy.max_interval; 1962 } 1963 if (!atomic_read(&dcc->discard_cmd_cnt)) 1964 wait_ms = dpolicy.max_interval; 1965 1966 sb_end_intwrite(sbi->sb); 1967 1968 } while (!kthread_should_stop()); 1969 return 0; 1970 } 1971 1972 #ifdef CONFIG_BLK_DEV_ZONED 1973 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi, 1974 struct block_device *bdev, block_t blkstart, block_t blklen) 1975 { 1976 sector_t sector, nr_sects; 1977 block_t lblkstart = blkstart; 1978 int devi = 0; 1979 u64 remainder = 0; 1980 1981 if (f2fs_is_multi_device(sbi)) { 1982 devi = f2fs_target_device_index(sbi, blkstart); 1983 if (blkstart < FDEV(devi).start_blk || 1984 blkstart > FDEV(devi).end_blk) { 1985 f2fs_err(sbi, "Invalid block %x", blkstart); 1986 return -EIO; 1987 } 1988 blkstart -= FDEV(devi).start_blk; 1989 } 1990 1991 /* For sequential zones, reset the zone write pointer */ 1992 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) { 1993 sector = SECTOR_FROM_BLOCK(blkstart); 1994 nr_sects = SECTOR_FROM_BLOCK(blklen); 1995 div64_u64_rem(sector, bdev_zone_sectors(bdev), &remainder); 1996 1997 if (remainder || nr_sects != bdev_zone_sectors(bdev)) { 1998 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)", 1999 devi, sbi->s_ndevs ? FDEV(devi).path : "", 2000 blkstart, blklen); 2001 return -EIO; 2002 } 2003 2004 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) { 2005 unsigned int nofs_flags; 2006 int ret; 2007 2008 trace_f2fs_issue_reset_zone(bdev, blkstart); 2009 nofs_flags = memalloc_nofs_save(); 2010 ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, 2011 sector, nr_sects); 2012 memalloc_nofs_restore(nofs_flags); 2013 return ret; 2014 } 2015 2016 __queue_zone_reset_cmd(sbi, bdev, blkstart, lblkstart, blklen); 2017 return 0; 2018 } 2019 2020 /* For conventional zones, use regular discard if supported */ 2021 __queue_discard_cmd(sbi, bdev, lblkstart, blklen); 2022 return 0; 2023 } 2024 #endif 2025 2026 static int __issue_discard_async(struct f2fs_sb_info *sbi, 2027 struct block_device *bdev, block_t blkstart, block_t blklen) 2028 { 2029 #ifdef CONFIG_BLK_DEV_ZONED 2030 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) 2031 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen); 2032 #endif 2033 __queue_discard_cmd(sbi, bdev, blkstart, blklen); 2034 return 0; 2035 } 2036 2037 static int f2fs_issue_discard(struct f2fs_sb_info *sbi, 2038 block_t blkstart, block_t blklen) 2039 { 2040 sector_t start = blkstart, len = 0; 2041 struct block_device *bdev; 2042 struct seg_entry *se; 2043 unsigned int offset; 2044 block_t i; 2045 int err = 0; 2046 2047 bdev = f2fs_target_device(sbi, blkstart, NULL); 2048 2049 for (i = blkstart; i < blkstart + blklen; i++, len++) { 2050 if (i != start) { 2051 struct block_device *bdev2 = 2052 f2fs_target_device(sbi, i, NULL); 2053 2054 if (bdev2 != bdev) { 2055 err = __issue_discard_async(sbi, bdev, 2056 start, len); 2057 if (err) 2058 return err; 2059 bdev = bdev2; 2060 start = i; 2061 len = 0; 2062 } 2063 } 2064 2065 se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); 2066 offset = GET_BLKOFF_FROM_SEG0(sbi, i); 2067 2068 if (f2fs_block_unit_discard(sbi) && 2069 !f2fs_test_and_set_bit(offset, se->discard_map)) 2070 sbi->discard_blks--; 2071 } 2072 2073 if (len) 2074 err = __issue_discard_async(sbi, bdev, start, len); 2075 return err; 2076 } 2077 2078 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc, 2079 bool check_only) 2080 { 2081 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 2082 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); 2083 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 2084 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 2085 unsigned long *discard_map = (unsigned long *)se->discard_map; 2086 unsigned long *dmap = SIT_I(sbi)->tmp_map; 2087 unsigned int start = 0, end = -1; 2088 bool force = (cpc->reason & CP_DISCARD); 2089 struct discard_entry *de = NULL; 2090 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list; 2091 int i; 2092 2093 if (se->valid_blocks == BLKS_PER_SEG(sbi) || 2094 !f2fs_hw_support_discard(sbi) || 2095 !f2fs_block_unit_discard(sbi)) 2096 return false; 2097 2098 if (!force) { 2099 if (!f2fs_realtime_discard_enable(sbi) || 2100 (!se->valid_blocks && 2101 !IS_CURSEG(sbi, cpc->trim_start)) || 2102 SM_I(sbi)->dcc_info->nr_discards >= 2103 SM_I(sbi)->dcc_info->max_discards) 2104 return false; 2105 } 2106 2107 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ 2108 for (i = 0; i < entries; i++) 2109 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : 2110 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; 2111 2112 while (force || SM_I(sbi)->dcc_info->nr_discards <= 2113 SM_I(sbi)->dcc_info->max_discards) { 2114 start = __find_rev_next_bit(dmap, BLKS_PER_SEG(sbi), end + 1); 2115 if (start >= BLKS_PER_SEG(sbi)) 2116 break; 2117 2118 end = __find_rev_next_zero_bit(dmap, 2119 BLKS_PER_SEG(sbi), start + 1); 2120 if (force && start && end != BLKS_PER_SEG(sbi) && 2121 (end - start) < cpc->trim_minlen) 2122 continue; 2123 2124 if (check_only) 2125 return true; 2126 2127 if (!de) { 2128 de = f2fs_kmem_cache_alloc(discard_entry_slab, 2129 GFP_F2FS_ZERO, true, NULL); 2130 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start); 2131 list_add_tail(&de->list, head); 2132 } 2133 2134 for (i = start; i < end; i++) 2135 __set_bit_le(i, (void *)de->discard_map); 2136 2137 SM_I(sbi)->dcc_info->nr_discards += end - start; 2138 } 2139 return false; 2140 } 2141 2142 static void release_discard_addr(struct discard_entry *entry) 2143 { 2144 list_del(&entry->list); 2145 kmem_cache_free(discard_entry_slab, entry); 2146 } 2147 2148 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi) 2149 { 2150 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); 2151 struct discard_entry *entry, *this; 2152 2153 /* drop caches */ 2154 list_for_each_entry_safe(entry, this, head, list) 2155 release_discard_addr(entry); 2156 } 2157 2158 /* 2159 * Should call f2fs_clear_prefree_segments after checkpoint is done. 2160 */ 2161 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 2162 { 2163 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2164 unsigned int segno; 2165 2166 mutex_lock(&dirty_i->seglist_lock); 2167 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) 2168 __set_test_and_free(sbi, segno, false); 2169 mutex_unlock(&dirty_i->seglist_lock); 2170 } 2171 2172 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi, 2173 struct cp_control *cpc) 2174 { 2175 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2176 struct list_head *head = &dcc->entry_list; 2177 struct discard_entry *entry, *this; 2178 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2179 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; 2180 unsigned int start = 0, end = -1; 2181 unsigned int secno, start_segno; 2182 bool force = (cpc->reason & CP_DISCARD); 2183 bool section_alignment = F2FS_OPTION(sbi).discard_unit == 2184 DISCARD_UNIT_SECTION; 2185 2186 if (f2fs_lfs_mode(sbi) && __is_large_section(sbi)) 2187 section_alignment = true; 2188 2189 mutex_lock(&dirty_i->seglist_lock); 2190 2191 while (1) { 2192 int i; 2193 2194 if (section_alignment && end != -1) 2195 end--; 2196 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); 2197 if (start >= MAIN_SEGS(sbi)) 2198 break; 2199 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), 2200 start + 1); 2201 2202 if (section_alignment) { 2203 start = rounddown(start, SEGS_PER_SEC(sbi)); 2204 end = roundup(end, SEGS_PER_SEC(sbi)); 2205 } 2206 2207 for (i = start; i < end; i++) { 2208 if (test_and_clear_bit(i, prefree_map)) 2209 dirty_i->nr_dirty[PRE]--; 2210 } 2211 2212 if (!f2fs_realtime_discard_enable(sbi)) 2213 continue; 2214 2215 if (force && start >= cpc->trim_start && 2216 (end - 1) <= cpc->trim_end) 2217 continue; 2218 2219 /* Should cover 2MB zoned device for zone-based reset */ 2220 if (!f2fs_sb_has_blkzoned(sbi) && 2221 (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi))) { 2222 f2fs_issue_discard(sbi, START_BLOCK(sbi, start), 2223 SEGS_TO_BLKS(sbi, end - start)); 2224 continue; 2225 } 2226 next: 2227 secno = GET_SEC_FROM_SEG(sbi, start); 2228 start_segno = GET_SEG_FROM_SEC(sbi, secno); 2229 if (!IS_CURSEC(sbi, secno) && 2230 !get_valid_blocks(sbi, start, true)) 2231 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno), 2232 BLKS_PER_SEC(sbi)); 2233 2234 start = start_segno + SEGS_PER_SEC(sbi); 2235 if (start < end) 2236 goto next; 2237 else 2238 end = start - 1; 2239 } 2240 mutex_unlock(&dirty_i->seglist_lock); 2241 2242 if (!f2fs_block_unit_discard(sbi)) 2243 goto wakeup; 2244 2245 /* send small discards */ 2246 list_for_each_entry_safe(entry, this, head, list) { 2247 unsigned int cur_pos = 0, next_pos, len, total_len = 0; 2248 bool is_valid = test_bit_le(0, entry->discard_map); 2249 2250 find_next: 2251 if (is_valid) { 2252 next_pos = find_next_zero_bit_le(entry->discard_map, 2253 BLKS_PER_SEG(sbi), cur_pos); 2254 len = next_pos - cur_pos; 2255 2256 if (f2fs_sb_has_blkzoned(sbi) || 2257 (force && len < cpc->trim_minlen)) 2258 goto skip; 2259 2260 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos, 2261 len); 2262 total_len += len; 2263 } else { 2264 next_pos = find_next_bit_le(entry->discard_map, 2265 BLKS_PER_SEG(sbi), cur_pos); 2266 } 2267 skip: 2268 cur_pos = next_pos; 2269 is_valid = !is_valid; 2270 2271 if (cur_pos < BLKS_PER_SEG(sbi)) 2272 goto find_next; 2273 2274 release_discard_addr(entry); 2275 dcc->nr_discards -= total_len; 2276 } 2277 2278 wakeup: 2279 wake_up_discard_thread(sbi, false); 2280 } 2281 2282 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi) 2283 { 2284 dev_t dev = sbi->sb->s_bdev->bd_dev; 2285 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2286 int err = 0; 2287 2288 if (f2fs_sb_has_readonly(sbi)) { 2289 f2fs_info(sbi, 2290 "Skip to start discard thread for readonly image"); 2291 return 0; 2292 } 2293 2294 if (!f2fs_realtime_discard_enable(sbi)) 2295 return 0; 2296 2297 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi, 2298 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev)); 2299 if (IS_ERR(dcc->f2fs_issue_discard)) { 2300 err = PTR_ERR(dcc->f2fs_issue_discard); 2301 dcc->f2fs_issue_discard = NULL; 2302 } 2303 2304 return err; 2305 } 2306 2307 static int create_discard_cmd_control(struct f2fs_sb_info *sbi) 2308 { 2309 struct discard_cmd_control *dcc; 2310 int err = 0, i; 2311 2312 if (SM_I(sbi)->dcc_info) { 2313 dcc = SM_I(sbi)->dcc_info; 2314 goto init_thread; 2315 } 2316 2317 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL); 2318 if (!dcc) 2319 return -ENOMEM; 2320 2321 dcc->discard_io_aware_gran = MAX_PLIST_NUM; 2322 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY; 2323 dcc->max_ordered_discard = DEFAULT_MAX_ORDERED_DISCARD_GRANULARITY; 2324 dcc->discard_io_aware = DPOLICY_IO_AWARE_ENABLE; 2325 if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT || 2326 F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION) 2327 dcc->discard_granularity = BLKS_PER_SEG(sbi); 2328 2329 INIT_LIST_HEAD(&dcc->entry_list); 2330 for (i = 0; i < MAX_PLIST_NUM; i++) 2331 INIT_LIST_HEAD(&dcc->pend_list[i]); 2332 INIT_LIST_HEAD(&dcc->wait_list); 2333 INIT_LIST_HEAD(&dcc->fstrim_list); 2334 mutex_init(&dcc->cmd_lock); 2335 atomic_set(&dcc->issued_discard, 0); 2336 atomic_set(&dcc->queued_discard, 0); 2337 atomic_set(&dcc->discard_cmd_cnt, 0); 2338 dcc->nr_discards = 0; 2339 dcc->max_discards = SEGS_TO_BLKS(sbi, MAIN_SEGS(sbi)); 2340 dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST; 2341 dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME; 2342 dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME; 2343 dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME; 2344 dcc->discard_urgent_util = DEF_DISCARD_URGENT_UTIL; 2345 dcc->undiscard_blks = 0; 2346 dcc->next_pos = 0; 2347 dcc->root = RB_ROOT_CACHED; 2348 dcc->rbtree_check = false; 2349 2350 init_waitqueue_head(&dcc->discard_wait_queue); 2351 SM_I(sbi)->dcc_info = dcc; 2352 init_thread: 2353 err = f2fs_start_discard_thread(sbi); 2354 if (err) { 2355 kfree(dcc); 2356 SM_I(sbi)->dcc_info = NULL; 2357 } 2358 2359 return err; 2360 } 2361 2362 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi) 2363 { 2364 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2365 2366 if (!dcc) 2367 return; 2368 2369 f2fs_stop_discard_thread(sbi); 2370 2371 /* 2372 * Recovery can cache discard commands, so in error path of 2373 * fill_super(), it needs to give a chance to handle them. 2374 */ 2375 f2fs_issue_discard_timeout(sbi); 2376 2377 kfree(dcc); 2378 SM_I(sbi)->dcc_info = NULL; 2379 } 2380 2381 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 2382 { 2383 struct sit_info *sit_i = SIT_I(sbi); 2384 2385 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { 2386 sit_i->dirty_sentries++; 2387 return false; 2388 } 2389 2390 return true; 2391 } 2392 2393 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 2394 unsigned int segno, int modified) 2395 { 2396 struct seg_entry *se = get_seg_entry(sbi, segno); 2397 2398 se->type = type; 2399 if (modified) 2400 __mark_sit_entry_dirty(sbi, segno); 2401 } 2402 2403 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi, 2404 block_t blkaddr) 2405 { 2406 unsigned int segno = GET_SEGNO(sbi, blkaddr); 2407 2408 if (segno == NULL_SEGNO) 2409 return 0; 2410 return get_seg_entry(sbi, segno)->mtime; 2411 } 2412 2413 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr, 2414 unsigned long long old_mtime) 2415 { 2416 struct seg_entry *se; 2417 unsigned int segno = GET_SEGNO(sbi, blkaddr); 2418 unsigned long long ctime = get_mtime(sbi, false); 2419 unsigned long long mtime = old_mtime ? old_mtime : ctime; 2420 2421 if (segno == NULL_SEGNO) 2422 return; 2423 2424 se = get_seg_entry(sbi, segno); 2425 2426 if (!se->mtime) 2427 se->mtime = mtime; 2428 else 2429 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime, 2430 se->valid_blocks + 1); 2431 2432 if (ctime > SIT_I(sbi)->max_mtime) 2433 SIT_I(sbi)->max_mtime = ctime; 2434 } 2435 2436 /* 2437 * NOTE: when updating multiple blocks at the same time, please ensure 2438 * that the consecutive input blocks belong to the same segment. 2439 */ 2440 static int update_sit_entry_for_release(struct f2fs_sb_info *sbi, struct seg_entry *se, 2441 block_t blkaddr, unsigned int offset, int del) 2442 { 2443 bool exist; 2444 #ifdef CONFIG_F2FS_CHECK_FS 2445 bool mir_exist; 2446 #endif 2447 int i; 2448 int del_count = -del; 2449 2450 f2fs_bug_on(sbi, GET_SEGNO(sbi, blkaddr) != GET_SEGNO(sbi, blkaddr + del_count - 1)); 2451 2452 for (i = 0; i < del_count; i++) { 2453 exist = f2fs_test_and_clear_bit(offset + i, se->cur_valid_map); 2454 #ifdef CONFIG_F2FS_CHECK_FS 2455 mir_exist = f2fs_test_and_clear_bit(offset + i, 2456 se->cur_valid_map_mir); 2457 if (unlikely(exist != mir_exist)) { 2458 f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d", 2459 blkaddr + i, exist); 2460 f2fs_bug_on(sbi, 1); 2461 } 2462 #endif 2463 if (unlikely(!exist)) { 2464 f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u", blkaddr + i); 2465 f2fs_bug_on(sbi, 1); 2466 se->valid_blocks++; 2467 del += 1; 2468 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 2469 /* 2470 * If checkpoints are off, we must not reuse data that 2471 * was used in the previous checkpoint. If it was used 2472 * before, we must track that to know how much space we 2473 * really have. 2474 */ 2475 if (f2fs_test_bit(offset + i, se->ckpt_valid_map)) { 2476 spin_lock(&sbi->stat_lock); 2477 sbi->unusable_block_count++; 2478 spin_unlock(&sbi->stat_lock); 2479 } 2480 } 2481 2482 if (f2fs_block_unit_discard(sbi) && 2483 f2fs_test_and_clear_bit(offset + i, se->discard_map)) 2484 sbi->discard_blks++; 2485 2486 if (!f2fs_test_bit(offset + i, se->ckpt_valid_map)) 2487 se->ckpt_valid_blocks -= 1; 2488 } 2489 2490 return del; 2491 } 2492 2493 static int update_sit_entry_for_alloc(struct f2fs_sb_info *sbi, struct seg_entry *se, 2494 block_t blkaddr, unsigned int offset, int del) 2495 { 2496 bool exist; 2497 #ifdef CONFIG_F2FS_CHECK_FS 2498 bool mir_exist; 2499 #endif 2500 2501 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map); 2502 #ifdef CONFIG_F2FS_CHECK_FS 2503 mir_exist = f2fs_test_and_set_bit(offset, 2504 se->cur_valid_map_mir); 2505 if (unlikely(exist != mir_exist)) { 2506 f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d", 2507 blkaddr, exist); 2508 f2fs_bug_on(sbi, 1); 2509 } 2510 #endif 2511 if (unlikely(exist)) { 2512 f2fs_err(sbi, "Bitmap was wrongly set, blk:%u", blkaddr); 2513 f2fs_bug_on(sbi, 1); 2514 se->valid_blocks--; 2515 del = 0; 2516 } 2517 2518 if (f2fs_block_unit_discard(sbi) && 2519 !f2fs_test_and_set_bit(offset, se->discard_map)) 2520 sbi->discard_blks--; 2521 2522 /* 2523 * SSR should never reuse block which is checkpointed 2524 * or newly invalidated. 2525 */ 2526 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) { 2527 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map)) 2528 se->ckpt_valid_blocks++; 2529 } 2530 2531 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) 2532 se->ckpt_valid_blocks += del; 2533 2534 return del; 2535 } 2536 2537 /* 2538 * If releasing blocks, this function supports updating multiple consecutive blocks 2539 * at one time, but please note that these consecutive blocks need to belong to the 2540 * same segment. 2541 */ 2542 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 2543 { 2544 struct seg_entry *se; 2545 unsigned int segno, offset; 2546 long int new_vblocks; 2547 2548 segno = GET_SEGNO(sbi, blkaddr); 2549 if (segno == NULL_SEGNO) 2550 return; 2551 2552 se = get_seg_entry(sbi, segno); 2553 new_vblocks = se->valid_blocks + del; 2554 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2555 2556 f2fs_bug_on(sbi, (new_vblocks < 0 || 2557 (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno)))); 2558 2559 se->valid_blocks = new_vblocks; 2560 2561 /* Update valid block bitmap */ 2562 if (del > 0) { 2563 del = update_sit_entry_for_alloc(sbi, se, blkaddr, offset, del); 2564 } else { 2565 del = update_sit_entry_for_release(sbi, se, blkaddr, offset, del); 2566 } 2567 2568 __mark_sit_entry_dirty(sbi, segno); 2569 2570 /* update total number of valid blocks to be written in ckpt area */ 2571 SIT_I(sbi)->written_valid_blocks += del; 2572 2573 if (__is_large_section(sbi)) 2574 get_sec_entry(sbi, segno)->valid_blocks += del; 2575 } 2576 2577 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr, 2578 unsigned int len) 2579 { 2580 unsigned int segno = GET_SEGNO(sbi, addr); 2581 struct sit_info *sit_i = SIT_I(sbi); 2582 block_t addr_start = addr, addr_end = addr + len - 1; 2583 unsigned int seg_num = GET_SEGNO(sbi, addr_end) - segno + 1; 2584 unsigned int i = 1, max_blocks = sbi->blocks_per_seg, cnt; 2585 2586 f2fs_bug_on(sbi, addr == NULL_ADDR); 2587 if (addr == NEW_ADDR || addr == COMPRESS_ADDR) 2588 return; 2589 2590 f2fs_invalidate_internal_cache(sbi, addr, len); 2591 2592 /* add it into sit main buffer */ 2593 down_write(&sit_i->sentry_lock); 2594 2595 if (seg_num == 1) 2596 cnt = len; 2597 else 2598 cnt = max_blocks - GET_BLKOFF_FROM_SEG0(sbi, addr); 2599 2600 do { 2601 update_segment_mtime(sbi, addr_start, 0); 2602 update_sit_entry(sbi, addr_start, -cnt); 2603 2604 /* add it into dirty seglist */ 2605 locate_dirty_segment(sbi, segno); 2606 2607 /* update @addr_start and @cnt and @segno */ 2608 addr_start = START_BLOCK(sbi, ++segno); 2609 if (++i == seg_num) 2610 cnt = GET_BLKOFF_FROM_SEG0(sbi, addr_end) + 1; 2611 else 2612 cnt = max_blocks; 2613 } while (i <= seg_num); 2614 2615 up_write(&sit_i->sentry_lock); 2616 } 2617 2618 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr) 2619 { 2620 struct sit_info *sit_i = SIT_I(sbi); 2621 unsigned int segno, offset; 2622 struct seg_entry *se; 2623 bool is_cp = false; 2624 2625 if (!__is_valid_data_blkaddr(blkaddr)) 2626 return true; 2627 2628 down_read(&sit_i->sentry_lock); 2629 2630 segno = GET_SEGNO(sbi, blkaddr); 2631 se = get_seg_entry(sbi, segno); 2632 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2633 2634 if (f2fs_test_bit(offset, se->ckpt_valid_map)) 2635 is_cp = true; 2636 2637 up_read(&sit_i->sentry_lock); 2638 2639 return is_cp; 2640 } 2641 2642 static unsigned short f2fs_curseg_valid_blocks(struct f2fs_sb_info *sbi, int type) 2643 { 2644 struct curseg_info *curseg = CURSEG_I(sbi, type); 2645 2646 if (sbi->ckpt->alloc_type[type] == SSR) 2647 return BLKS_PER_SEG(sbi); 2648 return curseg->next_blkoff; 2649 } 2650 2651 /* 2652 * Calculate the number of current summary pages for writing 2653 */ 2654 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) 2655 { 2656 int valid_sum_count = 0; 2657 int i, sum_in_page; 2658 2659 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2660 if (sbi->ckpt->alloc_type[i] != SSR && for_ra) 2661 valid_sum_count += 2662 le16_to_cpu(F2FS_CKPT(sbi)->cur_data_blkoff[i]); 2663 else 2664 valid_sum_count += f2fs_curseg_valid_blocks(sbi, i); 2665 } 2666 2667 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE - 2668 SUM_FOOTER_SIZE) / SUMMARY_SIZE; 2669 if (valid_sum_count <= sum_in_page) 2670 return 1; 2671 else if ((valid_sum_count - sum_in_page) <= 2672 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) 2673 return 2; 2674 return 3; 2675 } 2676 2677 /* 2678 * Caller should put this summary page 2679 */ 2680 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) 2681 { 2682 if (unlikely(f2fs_cp_error(sbi))) 2683 return ERR_PTR(-EIO); 2684 return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno)); 2685 } 2686 2687 void f2fs_update_meta_page(struct f2fs_sb_info *sbi, 2688 void *src, block_t blk_addr) 2689 { 2690 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2691 2692 memcpy(page_address(page), src, PAGE_SIZE); 2693 set_page_dirty(page); 2694 f2fs_put_page(page, 1); 2695 } 2696 2697 static void write_sum_page(struct f2fs_sb_info *sbi, 2698 struct f2fs_summary_block *sum_blk, block_t blk_addr) 2699 { 2700 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr); 2701 } 2702 2703 static void write_current_sum_page(struct f2fs_sb_info *sbi, 2704 int type, block_t blk_addr) 2705 { 2706 struct curseg_info *curseg = CURSEG_I(sbi, type); 2707 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2708 struct f2fs_summary_block *src = curseg->sum_blk; 2709 struct f2fs_summary_block *dst; 2710 2711 dst = (struct f2fs_summary_block *)page_address(page); 2712 memset(dst, 0, PAGE_SIZE); 2713 2714 mutex_lock(&curseg->curseg_mutex); 2715 2716 down_read(&curseg->journal_rwsem); 2717 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE); 2718 up_read(&curseg->journal_rwsem); 2719 2720 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE); 2721 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE); 2722 2723 mutex_unlock(&curseg->curseg_mutex); 2724 2725 set_page_dirty(page); 2726 f2fs_put_page(page, 1); 2727 } 2728 2729 static int is_next_segment_free(struct f2fs_sb_info *sbi, 2730 struct curseg_info *curseg) 2731 { 2732 unsigned int segno = curseg->segno + 1; 2733 struct free_segmap_info *free_i = FREE_I(sbi); 2734 2735 if (segno < MAIN_SEGS(sbi) && segno % SEGS_PER_SEC(sbi)) 2736 return !test_bit(segno, free_i->free_segmap); 2737 return 0; 2738 } 2739 2740 /* 2741 * Find a new segment from the free segments bitmap to right order 2742 * This function should be returned with success, otherwise BUG 2743 */ 2744 static int get_new_segment(struct f2fs_sb_info *sbi, 2745 unsigned int *newseg, bool new_sec, bool pinning) 2746 { 2747 struct free_segmap_info *free_i = FREE_I(sbi); 2748 unsigned int segno, secno, zoneno; 2749 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; 2750 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg); 2751 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg); 2752 bool init = true; 2753 int i; 2754 int ret = 0; 2755 2756 spin_lock(&free_i->segmap_lock); 2757 2758 if (time_to_inject(sbi, FAULT_NO_SEGMENT)) { 2759 ret = -ENOSPC; 2760 goto out_unlock; 2761 } 2762 2763 if (!new_sec && ((*newseg + 1) % SEGS_PER_SEC(sbi))) { 2764 segno = find_next_zero_bit(free_i->free_segmap, 2765 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1); 2766 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1)) 2767 goto got_it; 2768 } 2769 2770 #ifdef CONFIG_BLK_DEV_ZONED 2771 /* 2772 * If we format f2fs on zoned storage, let's try to get pinned sections 2773 * from beginning of the storage, which should be a conventional one. 2774 */ 2775 if (f2fs_sb_has_blkzoned(sbi)) { 2776 /* Prioritize writing to conventional zones */ 2777 if (sbi->blkzone_alloc_policy == BLKZONE_ALLOC_PRIOR_CONV || pinning) 2778 segno = 0; 2779 else 2780 segno = max(sbi->first_zoned_segno, *newseg); 2781 hint = GET_SEC_FROM_SEG(sbi, segno); 2782 } 2783 #endif 2784 2785 find_other_zone: 2786 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 2787 2788 #ifdef CONFIG_BLK_DEV_ZONED 2789 if (secno >= MAIN_SECS(sbi) && f2fs_sb_has_blkzoned(sbi)) { 2790 /* Write only to sequential zones */ 2791 if (sbi->blkzone_alloc_policy == BLKZONE_ALLOC_ONLY_SEQ) { 2792 hint = GET_SEC_FROM_SEG(sbi, sbi->first_zoned_segno); 2793 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 2794 } else 2795 secno = find_first_zero_bit(free_i->free_secmap, 2796 MAIN_SECS(sbi)); 2797 if (secno >= MAIN_SECS(sbi)) { 2798 ret = -ENOSPC; 2799 f2fs_bug_on(sbi, 1); 2800 goto out_unlock; 2801 } 2802 } 2803 #endif 2804 2805 if (secno >= MAIN_SECS(sbi)) { 2806 secno = find_first_zero_bit(free_i->free_secmap, 2807 MAIN_SECS(sbi)); 2808 if (secno >= MAIN_SECS(sbi)) { 2809 ret = -ENOSPC; 2810 f2fs_bug_on(sbi, !pinning); 2811 goto out_unlock; 2812 } 2813 } 2814 segno = GET_SEG_FROM_SEC(sbi, secno); 2815 zoneno = GET_ZONE_FROM_SEC(sbi, secno); 2816 2817 /* give up on finding another zone */ 2818 if (!init) 2819 goto got_it; 2820 if (sbi->secs_per_zone == 1) 2821 goto got_it; 2822 if (zoneno == old_zoneno) 2823 goto got_it; 2824 for (i = 0; i < NR_CURSEG_TYPE; i++) 2825 if (CURSEG_I(sbi, i)->zone == zoneno) 2826 break; 2827 2828 if (i < NR_CURSEG_TYPE) { 2829 /* zone is in user, try another */ 2830 if (zoneno + 1 >= total_zones) 2831 hint = 0; 2832 else 2833 hint = (zoneno + 1) * sbi->secs_per_zone; 2834 init = false; 2835 goto find_other_zone; 2836 } 2837 got_it: 2838 /* set it as dirty segment in free segmap */ 2839 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); 2840 2841 /* no free section in conventional zone */ 2842 if (new_sec && pinning && 2843 !f2fs_valid_pinned_area(sbi, START_BLOCK(sbi, segno))) { 2844 ret = -EAGAIN; 2845 goto out_unlock; 2846 } 2847 __set_inuse(sbi, segno); 2848 *newseg = segno; 2849 out_unlock: 2850 spin_unlock(&free_i->segmap_lock); 2851 2852 if (ret == -ENOSPC && !pinning) 2853 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_NO_SEGMENT); 2854 return ret; 2855 } 2856 2857 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 2858 { 2859 struct curseg_info *curseg = CURSEG_I(sbi, type); 2860 struct summary_footer *sum_footer; 2861 unsigned short seg_type = curseg->seg_type; 2862 2863 /* only happen when get_new_segment() fails */ 2864 if (curseg->next_segno == NULL_SEGNO) 2865 return; 2866 2867 curseg->inited = true; 2868 curseg->segno = curseg->next_segno; 2869 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno); 2870 curseg->next_blkoff = 0; 2871 curseg->next_segno = NULL_SEGNO; 2872 2873 sum_footer = &(curseg->sum_blk->footer); 2874 memset(sum_footer, 0, sizeof(struct summary_footer)); 2875 2876 sanity_check_seg_type(sbi, seg_type); 2877 2878 if (IS_DATASEG(seg_type)) 2879 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 2880 if (IS_NODESEG(seg_type)) 2881 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 2882 __set_sit_entry_type(sbi, seg_type, curseg->segno, modified); 2883 } 2884 2885 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type) 2886 { 2887 struct curseg_info *curseg = CURSEG_I(sbi, type); 2888 unsigned short seg_type = curseg->seg_type; 2889 2890 sanity_check_seg_type(sbi, seg_type); 2891 if (__is_large_section(sbi)) { 2892 if (f2fs_need_rand_seg(sbi)) { 2893 unsigned int hint = GET_SEC_FROM_SEG(sbi, curseg->segno); 2894 2895 if (GET_SEC_FROM_SEG(sbi, curseg->segno + 1) != hint) 2896 return curseg->segno; 2897 return get_random_u32_inclusive(curseg->segno + 1, 2898 GET_SEG_FROM_SEC(sbi, hint + 1) - 1); 2899 } 2900 return curseg->segno; 2901 } else if (f2fs_need_rand_seg(sbi)) { 2902 return get_random_u32_below(MAIN_SECS(sbi) * SEGS_PER_SEC(sbi)); 2903 } 2904 2905 /* inmem log may not locate on any segment after mount */ 2906 if (!curseg->inited) 2907 return 0; 2908 2909 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 2910 return 0; 2911 2912 if (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type)) 2913 return 0; 2914 2915 if (SIT_I(sbi)->last_victim[ALLOC_NEXT]) 2916 return SIT_I(sbi)->last_victim[ALLOC_NEXT]; 2917 2918 /* find segments from 0 to reuse freed segments */ 2919 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE) 2920 return 0; 2921 2922 return curseg->segno; 2923 } 2924 2925 static void reset_curseg_fields(struct curseg_info *curseg) 2926 { 2927 curseg->inited = false; 2928 curseg->segno = NULL_SEGNO; 2929 curseg->next_segno = 0; 2930 } 2931 2932 /* 2933 * Allocate a current working segment. 2934 * This function always allocates a free segment in LFS manner. 2935 */ 2936 static int new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 2937 { 2938 struct curseg_info *curseg = CURSEG_I(sbi, type); 2939 unsigned int segno = curseg->segno; 2940 bool pinning = type == CURSEG_COLD_DATA_PINNED; 2941 int ret; 2942 2943 if (curseg->inited) 2944 write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, segno)); 2945 2946 segno = __get_next_segno(sbi, type); 2947 ret = get_new_segment(sbi, &segno, new_sec, pinning); 2948 if (ret) { 2949 if (ret == -ENOSPC) 2950 reset_curseg_fields(curseg); 2951 return ret; 2952 } 2953 2954 curseg->next_segno = segno; 2955 reset_curseg(sbi, type, 1); 2956 curseg->alloc_type = LFS; 2957 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) 2958 curseg->fragment_remained_chunk = 2959 get_random_u32_inclusive(1, sbi->max_fragment_chunk); 2960 return 0; 2961 } 2962 2963 static int __next_free_blkoff(struct f2fs_sb_info *sbi, 2964 int segno, block_t start) 2965 { 2966 struct seg_entry *se = get_seg_entry(sbi, segno); 2967 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 2968 unsigned long *target_map = SIT_I(sbi)->tmp_map; 2969 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 2970 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 2971 int i; 2972 2973 for (i = 0; i < entries; i++) 2974 target_map[i] = ckpt_map[i] | cur_map[i]; 2975 2976 return __find_rev_next_zero_bit(target_map, BLKS_PER_SEG(sbi), start); 2977 } 2978 2979 static int f2fs_find_next_ssr_block(struct f2fs_sb_info *sbi, 2980 struct curseg_info *seg) 2981 { 2982 return __next_free_blkoff(sbi, seg->segno, seg->next_blkoff + 1); 2983 } 2984 2985 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno) 2986 { 2987 return __next_free_blkoff(sbi, segno, 0) < BLKS_PER_SEG(sbi); 2988 } 2989 2990 /* 2991 * This function always allocates a used segment(from dirty seglist) by SSR 2992 * manner, so it should recover the existing segment information of valid blocks 2993 */ 2994 static int change_curseg(struct f2fs_sb_info *sbi, int type) 2995 { 2996 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2997 struct curseg_info *curseg = CURSEG_I(sbi, type); 2998 unsigned int new_segno = curseg->next_segno; 2999 struct f2fs_summary_block *sum_node; 3000 struct page *sum_page; 3001 3002 if (curseg->inited) 3003 write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, curseg->segno)); 3004 3005 __set_test_and_inuse(sbi, new_segno); 3006 3007 mutex_lock(&dirty_i->seglist_lock); 3008 __remove_dirty_segment(sbi, new_segno, PRE); 3009 __remove_dirty_segment(sbi, new_segno, DIRTY); 3010 mutex_unlock(&dirty_i->seglist_lock); 3011 3012 reset_curseg(sbi, type, 1); 3013 curseg->alloc_type = SSR; 3014 curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0); 3015 3016 sum_page = f2fs_get_sum_page(sbi, new_segno); 3017 if (IS_ERR(sum_page)) { 3018 /* GC won't be able to use stale summary pages by cp_error */ 3019 memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE); 3020 return PTR_ERR(sum_page); 3021 } 3022 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 3023 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); 3024 f2fs_put_page(sum_page, 1); 3025 return 0; 3026 } 3027 3028 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type, 3029 int alloc_mode, unsigned long long age); 3030 3031 static int get_atssr_segment(struct f2fs_sb_info *sbi, int type, 3032 int target_type, int alloc_mode, 3033 unsigned long long age) 3034 { 3035 struct curseg_info *curseg = CURSEG_I(sbi, type); 3036 int ret = 0; 3037 3038 curseg->seg_type = target_type; 3039 3040 if (get_ssr_segment(sbi, type, alloc_mode, age)) { 3041 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno); 3042 3043 curseg->seg_type = se->type; 3044 ret = change_curseg(sbi, type); 3045 } else { 3046 /* allocate cold segment by default */ 3047 curseg->seg_type = CURSEG_COLD_DATA; 3048 ret = new_curseg(sbi, type, true); 3049 } 3050 stat_inc_seg_type(sbi, curseg); 3051 return ret; 3052 } 3053 3054 static int __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi, bool force) 3055 { 3056 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC); 3057 int ret = 0; 3058 3059 if (!sbi->am.atgc_enabled && !force) 3060 return 0; 3061 3062 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3063 3064 mutex_lock(&curseg->curseg_mutex); 3065 down_write(&SIT_I(sbi)->sentry_lock); 3066 3067 ret = get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, 3068 CURSEG_COLD_DATA, SSR, 0); 3069 3070 up_write(&SIT_I(sbi)->sentry_lock); 3071 mutex_unlock(&curseg->curseg_mutex); 3072 3073 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3074 return ret; 3075 } 3076 3077 int f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi) 3078 { 3079 return __f2fs_init_atgc_curseg(sbi, false); 3080 } 3081 3082 int f2fs_reinit_atgc_curseg(struct f2fs_sb_info *sbi) 3083 { 3084 int ret; 3085 3086 if (!test_opt(sbi, ATGC)) 3087 return 0; 3088 if (sbi->am.atgc_enabled) 3089 return 0; 3090 if (le64_to_cpu(F2FS_CKPT(sbi)->elapsed_time) < 3091 sbi->am.age_threshold) 3092 return 0; 3093 3094 ret = __f2fs_init_atgc_curseg(sbi, true); 3095 if (!ret) { 3096 sbi->am.atgc_enabled = true; 3097 f2fs_info(sbi, "reenabled age threshold GC"); 3098 } 3099 return ret; 3100 } 3101 3102 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type) 3103 { 3104 struct curseg_info *curseg = CURSEG_I(sbi, type); 3105 3106 mutex_lock(&curseg->curseg_mutex); 3107 if (!curseg->inited) 3108 goto out; 3109 3110 if (get_valid_blocks(sbi, curseg->segno, false)) { 3111 write_sum_page(sbi, curseg->sum_blk, 3112 GET_SUM_BLOCK(sbi, curseg->segno)); 3113 } else { 3114 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 3115 __set_test_and_free(sbi, curseg->segno, true); 3116 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 3117 } 3118 out: 3119 mutex_unlock(&curseg->curseg_mutex); 3120 } 3121 3122 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi) 3123 { 3124 __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED); 3125 3126 if (sbi->am.atgc_enabled) 3127 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC); 3128 } 3129 3130 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type) 3131 { 3132 struct curseg_info *curseg = CURSEG_I(sbi, type); 3133 3134 mutex_lock(&curseg->curseg_mutex); 3135 if (!curseg->inited) 3136 goto out; 3137 if (get_valid_blocks(sbi, curseg->segno, false)) 3138 goto out; 3139 3140 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 3141 __set_test_and_inuse(sbi, curseg->segno); 3142 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 3143 out: 3144 mutex_unlock(&curseg->curseg_mutex); 3145 } 3146 3147 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi) 3148 { 3149 __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED); 3150 3151 if (sbi->am.atgc_enabled) 3152 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC); 3153 } 3154 3155 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type, 3156 int alloc_mode, unsigned long long age) 3157 { 3158 struct curseg_info *curseg = CURSEG_I(sbi, type); 3159 unsigned segno = NULL_SEGNO; 3160 unsigned short seg_type = curseg->seg_type; 3161 int i, cnt; 3162 bool reversed = false; 3163 3164 sanity_check_seg_type(sbi, seg_type); 3165 3166 /* f2fs_need_SSR() already forces to do this */ 3167 if (!f2fs_get_victim(sbi, &segno, BG_GC, seg_type, 3168 alloc_mode, age, false)) { 3169 curseg->next_segno = segno; 3170 return 1; 3171 } 3172 3173 /* For node segments, let's do SSR more intensively */ 3174 if (IS_NODESEG(seg_type)) { 3175 if (seg_type >= CURSEG_WARM_NODE) { 3176 reversed = true; 3177 i = CURSEG_COLD_NODE; 3178 } else { 3179 i = CURSEG_HOT_NODE; 3180 } 3181 cnt = NR_CURSEG_NODE_TYPE; 3182 } else { 3183 if (seg_type >= CURSEG_WARM_DATA) { 3184 reversed = true; 3185 i = CURSEG_COLD_DATA; 3186 } else { 3187 i = CURSEG_HOT_DATA; 3188 } 3189 cnt = NR_CURSEG_DATA_TYPE; 3190 } 3191 3192 for (; cnt-- > 0; reversed ? i-- : i++) { 3193 if (i == seg_type) 3194 continue; 3195 if (!f2fs_get_victim(sbi, &segno, BG_GC, i, 3196 alloc_mode, age, false)) { 3197 curseg->next_segno = segno; 3198 return 1; 3199 } 3200 } 3201 3202 /* find valid_blocks=0 in dirty list */ 3203 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 3204 segno = get_free_segment(sbi); 3205 if (segno != NULL_SEGNO) { 3206 curseg->next_segno = segno; 3207 return 1; 3208 } 3209 } 3210 return 0; 3211 } 3212 3213 static bool need_new_seg(struct f2fs_sb_info *sbi, int type) 3214 { 3215 struct curseg_info *curseg = CURSEG_I(sbi, type); 3216 3217 if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) && 3218 curseg->seg_type == CURSEG_WARM_NODE) 3219 return true; 3220 if (curseg->alloc_type == LFS && is_next_segment_free(sbi, curseg) && 3221 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 3222 return true; 3223 if (!f2fs_need_SSR(sbi) || !get_ssr_segment(sbi, type, SSR, 0)) 3224 return true; 3225 return false; 3226 } 3227 3228 int f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type, 3229 unsigned int start, unsigned int end) 3230 { 3231 struct curseg_info *curseg = CURSEG_I(sbi, type); 3232 unsigned int segno; 3233 int ret = 0; 3234 3235 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3236 mutex_lock(&curseg->curseg_mutex); 3237 down_write(&SIT_I(sbi)->sentry_lock); 3238 3239 segno = CURSEG_I(sbi, type)->segno; 3240 if (segno < start || segno > end) 3241 goto unlock; 3242 3243 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0)) 3244 ret = change_curseg(sbi, type); 3245 else 3246 ret = new_curseg(sbi, type, true); 3247 3248 stat_inc_seg_type(sbi, curseg); 3249 3250 locate_dirty_segment(sbi, segno); 3251 unlock: 3252 up_write(&SIT_I(sbi)->sentry_lock); 3253 3254 if (segno != curseg->segno) 3255 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u", 3256 type, segno, curseg->segno); 3257 3258 mutex_unlock(&curseg->curseg_mutex); 3259 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3260 return ret; 3261 } 3262 3263 static int __allocate_new_segment(struct f2fs_sb_info *sbi, int type, 3264 bool new_sec, bool force) 3265 { 3266 struct curseg_info *curseg = CURSEG_I(sbi, type); 3267 unsigned int old_segno; 3268 int err = 0; 3269 3270 if (type == CURSEG_COLD_DATA_PINNED && !curseg->inited) 3271 goto allocate; 3272 3273 if (!force && curseg->inited && 3274 !curseg->next_blkoff && 3275 !get_valid_blocks(sbi, curseg->segno, new_sec) && 3276 !get_ckpt_valid_blocks(sbi, curseg->segno, new_sec)) 3277 return 0; 3278 3279 allocate: 3280 old_segno = curseg->segno; 3281 err = new_curseg(sbi, type, true); 3282 if (err) 3283 return err; 3284 stat_inc_seg_type(sbi, curseg); 3285 locate_dirty_segment(sbi, old_segno); 3286 return 0; 3287 } 3288 3289 int f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force) 3290 { 3291 int ret; 3292 3293 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3294 down_write(&SIT_I(sbi)->sentry_lock); 3295 ret = __allocate_new_segment(sbi, type, true, force); 3296 up_write(&SIT_I(sbi)->sentry_lock); 3297 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3298 3299 return ret; 3300 } 3301 3302 int f2fs_allocate_pinning_section(struct f2fs_sb_info *sbi) 3303 { 3304 int err; 3305 bool gc_required = true; 3306 3307 retry: 3308 f2fs_lock_op(sbi); 3309 err = f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false); 3310 f2fs_unlock_op(sbi); 3311 3312 if (f2fs_sb_has_blkzoned(sbi) && err == -EAGAIN && gc_required) { 3313 f2fs_down_write(&sbi->gc_lock); 3314 err = f2fs_gc_range(sbi, 0, GET_SEGNO(sbi, FDEV(0).end_blk), 3315 true, ZONED_PIN_SEC_REQUIRED_COUNT); 3316 f2fs_up_write(&sbi->gc_lock); 3317 3318 gc_required = false; 3319 if (!err) 3320 goto retry; 3321 } 3322 3323 return err; 3324 } 3325 3326 int f2fs_allocate_new_segments(struct f2fs_sb_info *sbi) 3327 { 3328 int i; 3329 int err = 0; 3330 3331 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3332 down_write(&SIT_I(sbi)->sentry_lock); 3333 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) 3334 err += __allocate_new_segment(sbi, i, false, false); 3335 up_write(&SIT_I(sbi)->sentry_lock); 3336 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3337 3338 return err; 3339 } 3340 3341 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi, 3342 struct cp_control *cpc) 3343 { 3344 __u64 trim_start = cpc->trim_start; 3345 bool has_candidate = false; 3346 3347 down_write(&SIT_I(sbi)->sentry_lock); 3348 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) { 3349 if (add_discard_addrs(sbi, cpc, true)) { 3350 has_candidate = true; 3351 break; 3352 } 3353 } 3354 up_write(&SIT_I(sbi)->sentry_lock); 3355 3356 cpc->trim_start = trim_start; 3357 return has_candidate; 3358 } 3359 3360 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi, 3361 struct discard_policy *dpolicy, 3362 unsigned int start, unsigned int end) 3363 { 3364 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 3365 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 3366 struct rb_node **insert_p = NULL, *insert_parent = NULL; 3367 struct discard_cmd *dc; 3368 struct blk_plug plug; 3369 int issued; 3370 unsigned int trimmed = 0; 3371 3372 next: 3373 issued = 0; 3374 3375 mutex_lock(&dcc->cmd_lock); 3376 if (unlikely(dcc->rbtree_check)) 3377 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi)); 3378 3379 dc = __lookup_discard_cmd_ret(&dcc->root, start, 3380 &prev_dc, &next_dc, &insert_p, &insert_parent); 3381 if (!dc) 3382 dc = next_dc; 3383 3384 blk_start_plug(&plug); 3385 3386 while (dc && dc->di.lstart <= end) { 3387 struct rb_node *node; 3388 int err = 0; 3389 3390 if (dc->di.len < dpolicy->granularity) 3391 goto skip; 3392 3393 if (dc->state != D_PREP) { 3394 list_move_tail(&dc->list, &dcc->fstrim_list); 3395 goto skip; 3396 } 3397 3398 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued); 3399 3400 if (issued >= dpolicy->max_requests) { 3401 start = dc->di.lstart + dc->di.len; 3402 3403 if (err) 3404 __remove_discard_cmd(sbi, dc); 3405 3406 blk_finish_plug(&plug); 3407 mutex_unlock(&dcc->cmd_lock); 3408 trimmed += __wait_all_discard_cmd(sbi, NULL); 3409 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 3410 goto next; 3411 } 3412 skip: 3413 node = rb_next(&dc->rb_node); 3414 if (err) 3415 __remove_discard_cmd(sbi, dc); 3416 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 3417 3418 if (fatal_signal_pending(current)) 3419 break; 3420 } 3421 3422 blk_finish_plug(&plug); 3423 mutex_unlock(&dcc->cmd_lock); 3424 3425 return trimmed; 3426 } 3427 3428 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) 3429 { 3430 __u64 start = F2FS_BYTES_TO_BLK(range->start); 3431 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; 3432 unsigned int start_segno, end_segno; 3433 block_t start_block, end_block; 3434 struct cp_control cpc; 3435 struct discard_policy dpolicy; 3436 unsigned long long trimmed = 0; 3437 int err = 0; 3438 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi); 3439 3440 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize) 3441 return -EINVAL; 3442 3443 if (end < MAIN_BLKADDR(sbi)) 3444 goto out; 3445 3446 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 3447 f2fs_warn(sbi, "Found FS corruption, run fsck to fix."); 3448 return -EFSCORRUPTED; 3449 } 3450 3451 /* start/end segment number in main_area */ 3452 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); 3453 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : 3454 GET_SEGNO(sbi, end); 3455 if (need_align) { 3456 start_segno = rounddown(start_segno, SEGS_PER_SEC(sbi)); 3457 end_segno = roundup(end_segno + 1, SEGS_PER_SEC(sbi)) - 1; 3458 } 3459 3460 cpc.reason = CP_DISCARD; 3461 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen)); 3462 cpc.trim_start = start_segno; 3463 cpc.trim_end = end_segno; 3464 3465 if (sbi->discard_blks == 0) 3466 goto out; 3467 3468 f2fs_down_write(&sbi->gc_lock); 3469 stat_inc_cp_call_count(sbi, TOTAL_CALL); 3470 err = f2fs_write_checkpoint(sbi, &cpc); 3471 f2fs_up_write(&sbi->gc_lock); 3472 if (err) 3473 goto out; 3474 3475 /* 3476 * We filed discard candidates, but actually we don't need to wait for 3477 * all of them, since they'll be issued in idle time along with runtime 3478 * discard option. User configuration looks like using runtime discard 3479 * or periodic fstrim instead of it. 3480 */ 3481 if (f2fs_realtime_discard_enable(sbi)) 3482 goto out; 3483 3484 start_block = START_BLOCK(sbi, start_segno); 3485 end_block = START_BLOCK(sbi, end_segno + 1); 3486 3487 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen); 3488 trimmed = __issue_discard_cmd_range(sbi, &dpolicy, 3489 start_block, end_block); 3490 3491 trimmed += __wait_discard_cmd_range(sbi, &dpolicy, 3492 start_block, end_block); 3493 out: 3494 if (!err) 3495 range->len = F2FS_BLK_TO_BYTES(trimmed); 3496 return err; 3497 } 3498 3499 int f2fs_rw_hint_to_seg_type(struct f2fs_sb_info *sbi, enum rw_hint hint) 3500 { 3501 if (F2FS_OPTION(sbi).active_logs == 2) 3502 return CURSEG_HOT_DATA; 3503 else if (F2FS_OPTION(sbi).active_logs == 4) 3504 return CURSEG_COLD_DATA; 3505 3506 /* active_log == 6 */ 3507 switch (hint) { 3508 case WRITE_LIFE_SHORT: 3509 return CURSEG_HOT_DATA; 3510 case WRITE_LIFE_EXTREME: 3511 return CURSEG_COLD_DATA; 3512 default: 3513 return CURSEG_WARM_DATA; 3514 } 3515 } 3516 3517 /* 3518 * This returns write hints for each segment type. This hints will be 3519 * passed down to block layer as below by default. 3520 * 3521 * User F2FS Block 3522 * ---- ---- ----- 3523 * META WRITE_LIFE_NONE|REQ_META 3524 * HOT_NODE WRITE_LIFE_NONE 3525 * WARM_NODE WRITE_LIFE_MEDIUM 3526 * COLD_NODE WRITE_LIFE_LONG 3527 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME 3528 * extension list " " 3529 * 3530 * -- buffered io 3531 * COLD_DATA WRITE_LIFE_EXTREME 3532 * HOT_DATA WRITE_LIFE_SHORT 3533 * WARM_DATA WRITE_LIFE_NOT_SET 3534 * 3535 * -- direct io 3536 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 3537 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 3538 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 3539 * WRITE_LIFE_NONE " WRITE_LIFE_NONE 3540 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM 3541 * WRITE_LIFE_LONG " WRITE_LIFE_LONG 3542 */ 3543 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi, 3544 enum page_type type, enum temp_type temp) 3545 { 3546 switch (type) { 3547 case DATA: 3548 switch (temp) { 3549 case WARM: 3550 return WRITE_LIFE_NOT_SET; 3551 case HOT: 3552 return WRITE_LIFE_SHORT; 3553 case COLD: 3554 return WRITE_LIFE_EXTREME; 3555 default: 3556 return WRITE_LIFE_NONE; 3557 } 3558 case NODE: 3559 switch (temp) { 3560 case WARM: 3561 return WRITE_LIFE_MEDIUM; 3562 case HOT: 3563 return WRITE_LIFE_NONE; 3564 case COLD: 3565 return WRITE_LIFE_LONG; 3566 default: 3567 return WRITE_LIFE_NONE; 3568 } 3569 case META: 3570 return WRITE_LIFE_NONE; 3571 default: 3572 return WRITE_LIFE_NONE; 3573 } 3574 } 3575 3576 static int __get_segment_type_2(struct f2fs_io_info *fio) 3577 { 3578 if (fio->type == DATA) 3579 return CURSEG_HOT_DATA; 3580 else 3581 return CURSEG_HOT_NODE; 3582 } 3583 3584 static int __get_segment_type_4(struct f2fs_io_info *fio) 3585 { 3586 if (fio->type == DATA) { 3587 struct inode *inode = fio->page->mapping->host; 3588 3589 if (S_ISDIR(inode->i_mode)) 3590 return CURSEG_HOT_DATA; 3591 else 3592 return CURSEG_COLD_DATA; 3593 } else { 3594 if (IS_DNODE(fio->page) && is_cold_node(fio->page)) 3595 return CURSEG_WARM_NODE; 3596 else 3597 return CURSEG_COLD_NODE; 3598 } 3599 } 3600 3601 static int __get_age_segment_type(struct inode *inode, pgoff_t pgofs) 3602 { 3603 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3604 struct extent_info ei = {}; 3605 3606 if (f2fs_lookup_age_extent_cache(inode, pgofs, &ei)) { 3607 if (!ei.age) 3608 return NO_CHECK_TYPE; 3609 if (ei.age <= sbi->hot_data_age_threshold) 3610 return CURSEG_HOT_DATA; 3611 if (ei.age <= sbi->warm_data_age_threshold) 3612 return CURSEG_WARM_DATA; 3613 return CURSEG_COLD_DATA; 3614 } 3615 return NO_CHECK_TYPE; 3616 } 3617 3618 static int __get_segment_type_6(struct f2fs_io_info *fio) 3619 { 3620 if (fio->type == DATA) { 3621 struct inode *inode = fio->page->mapping->host; 3622 int type; 3623 3624 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE)) 3625 return CURSEG_COLD_DATA_PINNED; 3626 3627 if (page_private_gcing(fio->page)) { 3628 if (fio->sbi->am.atgc_enabled && 3629 (fio->io_type == FS_DATA_IO) && 3630 (fio->sbi->gc_mode != GC_URGENT_HIGH) && 3631 __is_valid_data_blkaddr(fio->old_blkaddr) && 3632 !is_inode_flag_set(inode, FI_OPU_WRITE)) 3633 return CURSEG_ALL_DATA_ATGC; 3634 else 3635 return CURSEG_COLD_DATA; 3636 } 3637 if (file_is_cold(inode) || f2fs_need_compress_data(inode)) 3638 return CURSEG_COLD_DATA; 3639 3640 type = __get_age_segment_type(inode, 3641 page_folio(fio->page)->index); 3642 if (type != NO_CHECK_TYPE) 3643 return type; 3644 3645 if (file_is_hot(inode) || 3646 is_inode_flag_set(inode, FI_HOT_DATA) || 3647 f2fs_is_cow_file(inode)) 3648 return CURSEG_HOT_DATA; 3649 return f2fs_rw_hint_to_seg_type(F2FS_I_SB(inode), 3650 inode->i_write_hint); 3651 } else { 3652 if (IS_DNODE(fio->page)) 3653 return is_cold_node(fio->page) ? CURSEG_WARM_NODE : 3654 CURSEG_HOT_NODE; 3655 return CURSEG_COLD_NODE; 3656 } 3657 } 3658 3659 enum temp_type f2fs_get_segment_temp(struct f2fs_sb_info *sbi, 3660 enum log_type type) 3661 { 3662 struct curseg_info *curseg = CURSEG_I(sbi, type); 3663 enum temp_type temp = COLD; 3664 3665 switch (curseg->seg_type) { 3666 case CURSEG_HOT_NODE: 3667 case CURSEG_HOT_DATA: 3668 temp = HOT; 3669 break; 3670 case CURSEG_WARM_NODE: 3671 case CURSEG_WARM_DATA: 3672 temp = WARM; 3673 break; 3674 case CURSEG_COLD_NODE: 3675 case CURSEG_COLD_DATA: 3676 temp = COLD; 3677 break; 3678 default: 3679 f2fs_bug_on(sbi, 1); 3680 } 3681 3682 return temp; 3683 } 3684 3685 static int __get_segment_type(struct f2fs_io_info *fio) 3686 { 3687 enum log_type type = CURSEG_HOT_DATA; 3688 3689 switch (F2FS_OPTION(fio->sbi).active_logs) { 3690 case 2: 3691 type = __get_segment_type_2(fio); 3692 break; 3693 case 4: 3694 type = __get_segment_type_4(fio); 3695 break; 3696 case 6: 3697 type = __get_segment_type_6(fio); 3698 break; 3699 default: 3700 f2fs_bug_on(fio->sbi, true); 3701 } 3702 3703 fio->temp = f2fs_get_segment_temp(fio->sbi, type); 3704 3705 return type; 3706 } 3707 3708 static void f2fs_randomize_chunk(struct f2fs_sb_info *sbi, 3709 struct curseg_info *seg) 3710 { 3711 /* To allocate block chunks in different sizes, use random number */ 3712 if (--seg->fragment_remained_chunk > 0) 3713 return; 3714 3715 seg->fragment_remained_chunk = 3716 get_random_u32_inclusive(1, sbi->max_fragment_chunk); 3717 seg->next_blkoff += 3718 get_random_u32_inclusive(1, sbi->max_fragment_hole); 3719 } 3720 3721 int f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, 3722 block_t old_blkaddr, block_t *new_blkaddr, 3723 struct f2fs_summary *sum, int type, 3724 struct f2fs_io_info *fio) 3725 { 3726 struct sit_info *sit_i = SIT_I(sbi); 3727 struct curseg_info *curseg = CURSEG_I(sbi, type); 3728 unsigned long long old_mtime; 3729 bool from_gc = (type == CURSEG_ALL_DATA_ATGC); 3730 struct seg_entry *se = NULL; 3731 bool segment_full = false; 3732 int ret = 0; 3733 3734 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3735 3736 mutex_lock(&curseg->curseg_mutex); 3737 down_write(&sit_i->sentry_lock); 3738 3739 if (curseg->segno == NULL_SEGNO) { 3740 ret = -ENOSPC; 3741 goto out_err; 3742 } 3743 3744 if (from_gc) { 3745 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO); 3746 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr)); 3747 sanity_check_seg_type(sbi, se->type); 3748 f2fs_bug_on(sbi, IS_NODESEG(se->type)); 3749 } 3750 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 3751 3752 f2fs_bug_on(sbi, curseg->next_blkoff >= BLKS_PER_SEG(sbi)); 3753 3754 f2fs_wait_discard_bio(sbi, *new_blkaddr); 3755 3756 curseg->sum_blk->entries[curseg->next_blkoff] = *sum; 3757 if (curseg->alloc_type == SSR) { 3758 curseg->next_blkoff = f2fs_find_next_ssr_block(sbi, curseg); 3759 } else { 3760 curseg->next_blkoff++; 3761 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) 3762 f2fs_randomize_chunk(sbi, curseg); 3763 } 3764 if (curseg->next_blkoff >= f2fs_usable_blks_in_seg(sbi, curseg->segno)) 3765 segment_full = true; 3766 stat_inc_block_count(sbi, curseg); 3767 3768 if (from_gc) { 3769 old_mtime = get_segment_mtime(sbi, old_blkaddr); 3770 } else { 3771 update_segment_mtime(sbi, old_blkaddr, 0); 3772 old_mtime = 0; 3773 } 3774 update_segment_mtime(sbi, *new_blkaddr, old_mtime); 3775 3776 /* 3777 * SIT information should be updated before segment allocation, 3778 * since SSR needs latest valid block information. 3779 */ 3780 update_sit_entry(sbi, *new_blkaddr, 1); 3781 update_sit_entry(sbi, old_blkaddr, -1); 3782 3783 /* 3784 * If the current segment is full, flush it out and replace it with a 3785 * new segment. 3786 */ 3787 if (segment_full) { 3788 if (type == CURSEG_COLD_DATA_PINNED && 3789 !((curseg->segno + 1) % sbi->segs_per_sec)) { 3790 write_sum_page(sbi, curseg->sum_blk, 3791 GET_SUM_BLOCK(sbi, curseg->segno)); 3792 reset_curseg_fields(curseg); 3793 goto skip_new_segment; 3794 } 3795 3796 if (from_gc) { 3797 ret = get_atssr_segment(sbi, type, se->type, 3798 AT_SSR, se->mtime); 3799 } else { 3800 if (need_new_seg(sbi, type)) 3801 ret = new_curseg(sbi, type, false); 3802 else 3803 ret = change_curseg(sbi, type); 3804 stat_inc_seg_type(sbi, curseg); 3805 } 3806 3807 if (ret) 3808 goto out_err; 3809 } 3810 3811 skip_new_segment: 3812 /* 3813 * segment dirty status should be updated after segment allocation, 3814 * so we just need to update status only one time after previous 3815 * segment being closed. 3816 */ 3817 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3818 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr)); 3819 3820 if (IS_DATASEG(curseg->seg_type)) 3821 atomic64_inc(&sbi->allocated_data_blocks); 3822 3823 up_write(&sit_i->sentry_lock); 3824 3825 if (page && IS_NODESEG(curseg->seg_type)) { 3826 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); 3827 3828 f2fs_inode_chksum_set(sbi, page); 3829 } 3830 3831 if (fio) { 3832 struct f2fs_bio_info *io; 3833 3834 INIT_LIST_HEAD(&fio->list); 3835 fio->in_list = 1; 3836 io = sbi->write_io[fio->type] + fio->temp; 3837 spin_lock(&io->io_lock); 3838 list_add_tail(&fio->list, &io->io_list); 3839 spin_unlock(&io->io_lock); 3840 } 3841 3842 mutex_unlock(&curseg->curseg_mutex); 3843 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3844 return 0; 3845 3846 out_err: 3847 *new_blkaddr = NULL_ADDR; 3848 up_write(&sit_i->sentry_lock); 3849 mutex_unlock(&curseg->curseg_mutex); 3850 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3851 return ret; 3852 } 3853 3854 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino, 3855 block_t blkaddr, unsigned int blkcnt) 3856 { 3857 if (!f2fs_is_multi_device(sbi)) 3858 return; 3859 3860 while (1) { 3861 unsigned int devidx = f2fs_target_device_index(sbi, blkaddr); 3862 unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1; 3863 3864 /* update device state for fsync */ 3865 f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO); 3866 3867 /* update device state for checkpoint */ 3868 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) { 3869 spin_lock(&sbi->dev_lock); 3870 f2fs_set_bit(devidx, (char *)&sbi->dirty_device); 3871 spin_unlock(&sbi->dev_lock); 3872 } 3873 3874 if (blkcnt <= blks) 3875 break; 3876 blkcnt -= blks; 3877 blkaddr += blks; 3878 } 3879 } 3880 3881 static int log_type_to_seg_type(enum log_type type) 3882 { 3883 int seg_type = CURSEG_COLD_DATA; 3884 3885 switch (type) { 3886 case CURSEG_HOT_DATA: 3887 case CURSEG_WARM_DATA: 3888 case CURSEG_COLD_DATA: 3889 case CURSEG_HOT_NODE: 3890 case CURSEG_WARM_NODE: 3891 case CURSEG_COLD_NODE: 3892 seg_type = (int)type; 3893 break; 3894 case CURSEG_COLD_DATA_PINNED: 3895 case CURSEG_ALL_DATA_ATGC: 3896 seg_type = CURSEG_COLD_DATA; 3897 break; 3898 default: 3899 break; 3900 } 3901 return seg_type; 3902 } 3903 3904 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) 3905 { 3906 struct folio *folio = page_folio(fio->page); 3907 enum log_type type = __get_segment_type(fio); 3908 int seg_type = log_type_to_seg_type(type); 3909 bool keep_order = (f2fs_lfs_mode(fio->sbi) && 3910 seg_type == CURSEG_COLD_DATA); 3911 3912 if (keep_order) 3913 f2fs_down_read(&fio->sbi->io_order_lock); 3914 3915 if (f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr, 3916 &fio->new_blkaddr, sum, type, fio)) { 3917 if (fscrypt_inode_uses_fs_layer_crypto(folio->mapping->host)) 3918 fscrypt_finalize_bounce_page(&fio->encrypted_page); 3919 folio_end_writeback(folio); 3920 if (f2fs_in_warm_node_list(fio->sbi, folio)) 3921 f2fs_del_fsync_node_entry(fio->sbi, fio->page); 3922 goto out; 3923 } 3924 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO) 3925 f2fs_invalidate_internal_cache(fio->sbi, fio->old_blkaddr, 1); 3926 3927 /* writeout dirty page into bdev */ 3928 f2fs_submit_page_write(fio); 3929 3930 f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1); 3931 out: 3932 if (keep_order) 3933 f2fs_up_read(&fio->sbi->io_order_lock); 3934 } 3935 3936 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct folio *folio, 3937 enum iostat_type io_type) 3938 { 3939 struct f2fs_io_info fio = { 3940 .sbi = sbi, 3941 .type = META, 3942 .temp = HOT, 3943 .op = REQ_OP_WRITE, 3944 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO, 3945 .old_blkaddr = folio->index, 3946 .new_blkaddr = folio->index, 3947 .page = folio_page(folio, 0), 3948 .encrypted_page = NULL, 3949 .in_list = 0, 3950 }; 3951 3952 if (unlikely(folio->index >= MAIN_BLKADDR(sbi))) 3953 fio.op_flags &= ~REQ_META; 3954 3955 folio_start_writeback(folio); 3956 f2fs_submit_page_write(&fio); 3957 3958 stat_inc_meta_count(sbi, folio->index); 3959 f2fs_update_iostat(sbi, NULL, io_type, F2FS_BLKSIZE); 3960 } 3961 3962 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio) 3963 { 3964 struct f2fs_summary sum; 3965 3966 set_summary(&sum, nid, 0, 0); 3967 do_write_page(&sum, fio); 3968 3969 f2fs_update_iostat(fio->sbi, NULL, fio->io_type, F2FS_BLKSIZE); 3970 } 3971 3972 void f2fs_outplace_write_data(struct dnode_of_data *dn, 3973 struct f2fs_io_info *fio) 3974 { 3975 struct f2fs_sb_info *sbi = fio->sbi; 3976 struct f2fs_summary sum; 3977 3978 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); 3979 if (fio->io_type == FS_DATA_IO || fio->io_type == FS_CP_DATA_IO) 3980 f2fs_update_age_extent_cache(dn); 3981 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version); 3982 do_write_page(&sum, fio); 3983 f2fs_update_data_blkaddr(dn, fio->new_blkaddr); 3984 3985 f2fs_update_iostat(sbi, dn->inode, fio->io_type, F2FS_BLKSIZE); 3986 } 3987 3988 int f2fs_inplace_write_data(struct f2fs_io_info *fio) 3989 { 3990 int err; 3991 struct f2fs_sb_info *sbi = fio->sbi; 3992 unsigned int segno; 3993 3994 fio->new_blkaddr = fio->old_blkaddr; 3995 /* i/o temperature is needed for passing down write hints */ 3996 __get_segment_type(fio); 3997 3998 segno = GET_SEGNO(sbi, fio->new_blkaddr); 3999 4000 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) { 4001 set_sbi_flag(sbi, SBI_NEED_FSCK); 4002 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.", 4003 __func__, segno); 4004 err = -EFSCORRUPTED; 4005 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE); 4006 goto drop_bio; 4007 } 4008 4009 if (f2fs_cp_error(sbi)) { 4010 err = -EIO; 4011 goto drop_bio; 4012 } 4013 4014 if (fio->meta_gc) 4015 f2fs_truncate_meta_inode_pages(sbi, fio->new_blkaddr, 1); 4016 4017 stat_inc_inplace_blocks(fio->sbi); 4018 4019 if (fio->bio && !IS_F2FS_IPU_NOCACHE(sbi)) 4020 err = f2fs_merge_page_bio(fio); 4021 else 4022 err = f2fs_submit_page_bio(fio); 4023 if (!err) { 4024 f2fs_update_device_state(fio->sbi, fio->ino, 4025 fio->new_blkaddr, 1); 4026 f2fs_update_iostat(fio->sbi, fio->page->mapping->host, 4027 fio->io_type, F2FS_BLKSIZE); 4028 } 4029 4030 return err; 4031 drop_bio: 4032 if (fio->bio && *(fio->bio)) { 4033 struct bio *bio = *(fio->bio); 4034 4035 bio->bi_status = BLK_STS_IOERR; 4036 bio_endio(bio); 4037 *(fio->bio) = NULL; 4038 } 4039 return err; 4040 } 4041 4042 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi, 4043 unsigned int segno) 4044 { 4045 int i; 4046 4047 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 4048 if (CURSEG_I(sbi, i)->segno == segno) 4049 break; 4050 } 4051 return i; 4052 } 4053 4054 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 4055 block_t old_blkaddr, block_t new_blkaddr, 4056 bool recover_curseg, bool recover_newaddr, 4057 bool from_gc) 4058 { 4059 struct sit_info *sit_i = SIT_I(sbi); 4060 struct curseg_info *curseg; 4061 unsigned int segno, old_cursegno; 4062 struct seg_entry *se; 4063 int type; 4064 unsigned short old_blkoff; 4065 unsigned char old_alloc_type; 4066 4067 segno = GET_SEGNO(sbi, new_blkaddr); 4068 se = get_seg_entry(sbi, segno); 4069 type = se->type; 4070 4071 f2fs_down_write(&SM_I(sbi)->curseg_lock); 4072 4073 if (!recover_curseg) { 4074 /* for recovery flow */ 4075 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { 4076 if (old_blkaddr == NULL_ADDR) 4077 type = CURSEG_COLD_DATA; 4078 else 4079 type = CURSEG_WARM_DATA; 4080 } 4081 } else { 4082 if (IS_CURSEG(sbi, segno)) { 4083 /* se->type is volatile as SSR allocation */ 4084 type = __f2fs_get_curseg(sbi, segno); 4085 f2fs_bug_on(sbi, type == NO_CHECK_TYPE); 4086 } else { 4087 type = CURSEG_WARM_DATA; 4088 } 4089 } 4090 4091 curseg = CURSEG_I(sbi, type); 4092 f2fs_bug_on(sbi, !IS_DATASEG(curseg->seg_type)); 4093 4094 mutex_lock(&curseg->curseg_mutex); 4095 down_write(&sit_i->sentry_lock); 4096 4097 old_cursegno = curseg->segno; 4098 old_blkoff = curseg->next_blkoff; 4099 old_alloc_type = curseg->alloc_type; 4100 4101 /* change the current segment */ 4102 if (segno != curseg->segno) { 4103 curseg->next_segno = segno; 4104 if (change_curseg(sbi, type)) 4105 goto out_unlock; 4106 } 4107 4108 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); 4109 curseg->sum_blk->entries[curseg->next_blkoff] = *sum; 4110 4111 if (!recover_curseg || recover_newaddr) { 4112 if (!from_gc) 4113 update_segment_mtime(sbi, new_blkaddr, 0); 4114 update_sit_entry(sbi, new_blkaddr, 1); 4115 } 4116 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) { 4117 f2fs_invalidate_internal_cache(sbi, old_blkaddr, 1); 4118 if (!from_gc) 4119 update_segment_mtime(sbi, old_blkaddr, 0); 4120 update_sit_entry(sbi, old_blkaddr, -1); 4121 } 4122 4123 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 4124 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); 4125 4126 locate_dirty_segment(sbi, old_cursegno); 4127 4128 if (recover_curseg) { 4129 if (old_cursegno != curseg->segno) { 4130 curseg->next_segno = old_cursegno; 4131 if (change_curseg(sbi, type)) 4132 goto out_unlock; 4133 } 4134 curseg->next_blkoff = old_blkoff; 4135 curseg->alloc_type = old_alloc_type; 4136 } 4137 4138 out_unlock: 4139 up_write(&sit_i->sentry_lock); 4140 mutex_unlock(&curseg->curseg_mutex); 4141 f2fs_up_write(&SM_I(sbi)->curseg_lock); 4142 } 4143 4144 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, 4145 block_t old_addr, block_t new_addr, 4146 unsigned char version, bool recover_curseg, 4147 bool recover_newaddr) 4148 { 4149 struct f2fs_summary sum; 4150 4151 set_summary(&sum, dn->nid, dn->ofs_in_node, version); 4152 4153 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr, 4154 recover_curseg, recover_newaddr, false); 4155 4156 f2fs_update_data_blkaddr(dn, new_addr); 4157 } 4158 4159 void f2fs_folio_wait_writeback(struct folio *folio, enum page_type type, 4160 bool ordered, bool locked) 4161 { 4162 if (folio_test_writeback(folio)) { 4163 struct f2fs_sb_info *sbi = F2FS_F_SB(folio); 4164 4165 /* submit cached LFS IO */ 4166 f2fs_submit_merged_write_cond(sbi, NULL, &folio->page, 0, type); 4167 /* submit cached IPU IO */ 4168 f2fs_submit_merged_ipu_write(sbi, NULL, &folio->page); 4169 if (ordered) { 4170 folio_wait_writeback(folio); 4171 f2fs_bug_on(sbi, locked && folio_test_writeback(folio)); 4172 } else { 4173 folio_wait_stable(folio); 4174 } 4175 } 4176 } 4177 4178 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr) 4179 { 4180 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 4181 struct page *cpage; 4182 4183 if (!f2fs_meta_inode_gc_required(inode)) 4184 return; 4185 4186 if (!__is_valid_data_blkaddr(blkaddr)) 4187 return; 4188 4189 cpage = find_lock_page(META_MAPPING(sbi), blkaddr); 4190 if (cpage) { 4191 f2fs_wait_on_page_writeback(cpage, DATA, true, true); 4192 f2fs_put_page(cpage, 1); 4193 } 4194 } 4195 4196 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr, 4197 block_t len) 4198 { 4199 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 4200 block_t i; 4201 4202 if (!f2fs_meta_inode_gc_required(inode)) 4203 return; 4204 4205 for (i = 0; i < len; i++) 4206 f2fs_wait_on_block_writeback(inode, blkaddr + i); 4207 4208 f2fs_truncate_meta_inode_pages(sbi, blkaddr, len); 4209 } 4210 4211 static int read_compacted_summaries(struct f2fs_sb_info *sbi) 4212 { 4213 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 4214 struct curseg_info *seg_i; 4215 unsigned char *kaddr; 4216 struct page *page; 4217 block_t start; 4218 int i, j, offset; 4219 4220 start = start_sum_block(sbi); 4221 4222 page = f2fs_get_meta_page(sbi, start++); 4223 if (IS_ERR(page)) 4224 return PTR_ERR(page); 4225 kaddr = (unsigned char *)page_address(page); 4226 4227 /* Step 1: restore nat cache */ 4228 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 4229 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE); 4230 4231 /* Step 2: restore sit cache */ 4232 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 4233 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE); 4234 offset = 2 * SUM_JOURNAL_SIZE; 4235 4236 /* Step 3: restore summary entries */ 4237 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 4238 unsigned short blk_off; 4239 unsigned int segno; 4240 4241 seg_i = CURSEG_I(sbi, i); 4242 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 4243 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 4244 seg_i->next_segno = segno; 4245 reset_curseg(sbi, i, 0); 4246 seg_i->alloc_type = ckpt->alloc_type[i]; 4247 seg_i->next_blkoff = blk_off; 4248 4249 if (seg_i->alloc_type == SSR) 4250 blk_off = BLKS_PER_SEG(sbi); 4251 4252 for (j = 0; j < blk_off; j++) { 4253 struct f2fs_summary *s; 4254 4255 s = (struct f2fs_summary *)(kaddr + offset); 4256 seg_i->sum_blk->entries[j] = *s; 4257 offset += SUMMARY_SIZE; 4258 if (offset + SUMMARY_SIZE <= PAGE_SIZE - 4259 SUM_FOOTER_SIZE) 4260 continue; 4261 4262 f2fs_put_page(page, 1); 4263 page = NULL; 4264 4265 page = f2fs_get_meta_page(sbi, start++); 4266 if (IS_ERR(page)) 4267 return PTR_ERR(page); 4268 kaddr = (unsigned char *)page_address(page); 4269 offset = 0; 4270 } 4271 } 4272 f2fs_put_page(page, 1); 4273 return 0; 4274 } 4275 4276 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 4277 { 4278 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 4279 struct f2fs_summary_block *sum; 4280 struct curseg_info *curseg; 4281 struct page *new; 4282 unsigned short blk_off; 4283 unsigned int segno = 0; 4284 block_t blk_addr = 0; 4285 int err = 0; 4286 4287 /* get segment number and block addr */ 4288 if (IS_DATASEG(type)) { 4289 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 4290 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 4291 CURSEG_HOT_DATA]); 4292 if (__exist_node_summaries(sbi)) 4293 blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type); 4294 else 4295 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 4296 } else { 4297 segno = le32_to_cpu(ckpt->cur_node_segno[type - 4298 CURSEG_HOT_NODE]); 4299 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 4300 CURSEG_HOT_NODE]); 4301 if (__exist_node_summaries(sbi)) 4302 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 4303 type - CURSEG_HOT_NODE); 4304 else 4305 blk_addr = GET_SUM_BLOCK(sbi, segno); 4306 } 4307 4308 new = f2fs_get_meta_page(sbi, blk_addr); 4309 if (IS_ERR(new)) 4310 return PTR_ERR(new); 4311 sum = (struct f2fs_summary_block *)page_address(new); 4312 4313 if (IS_NODESEG(type)) { 4314 if (__exist_node_summaries(sbi)) { 4315 struct f2fs_summary *ns = &sum->entries[0]; 4316 int i; 4317 4318 for (i = 0; i < BLKS_PER_SEG(sbi); i++, ns++) { 4319 ns->version = 0; 4320 ns->ofs_in_node = 0; 4321 } 4322 } else { 4323 err = f2fs_restore_node_summary(sbi, segno, sum); 4324 if (err) 4325 goto out; 4326 } 4327 } 4328 4329 /* set uncompleted segment to curseg */ 4330 curseg = CURSEG_I(sbi, type); 4331 mutex_lock(&curseg->curseg_mutex); 4332 4333 /* update journal info */ 4334 down_write(&curseg->journal_rwsem); 4335 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE); 4336 up_write(&curseg->journal_rwsem); 4337 4338 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE); 4339 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE); 4340 curseg->next_segno = segno; 4341 reset_curseg(sbi, type, 0); 4342 curseg->alloc_type = ckpt->alloc_type[type]; 4343 curseg->next_blkoff = blk_off; 4344 mutex_unlock(&curseg->curseg_mutex); 4345 out: 4346 f2fs_put_page(new, 1); 4347 return err; 4348 } 4349 4350 static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 4351 { 4352 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal; 4353 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal; 4354 int type = CURSEG_HOT_DATA; 4355 int err; 4356 4357 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) { 4358 int npages = f2fs_npages_for_summary_flush(sbi, true); 4359 4360 if (npages >= 2) 4361 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages, 4362 META_CP, true); 4363 4364 /* restore for compacted data summary */ 4365 err = read_compacted_summaries(sbi); 4366 if (err) 4367 return err; 4368 type = CURSEG_HOT_NODE; 4369 } 4370 4371 if (__exist_node_summaries(sbi)) 4372 f2fs_ra_meta_pages(sbi, 4373 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type), 4374 NR_CURSEG_PERSIST_TYPE - type, META_CP, true); 4375 4376 for (; type <= CURSEG_COLD_NODE; type++) { 4377 err = read_normal_summaries(sbi, type); 4378 if (err) 4379 return err; 4380 } 4381 4382 /* sanity check for summary blocks */ 4383 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES || 4384 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) { 4385 f2fs_err(sbi, "invalid journal entries nats %u sits %u", 4386 nats_in_cursum(nat_j), sits_in_cursum(sit_j)); 4387 return -EINVAL; 4388 } 4389 4390 return 0; 4391 } 4392 4393 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 4394 { 4395 struct page *page; 4396 unsigned char *kaddr; 4397 struct f2fs_summary *summary; 4398 struct curseg_info *seg_i; 4399 int written_size = 0; 4400 int i, j; 4401 4402 page = f2fs_grab_meta_page(sbi, blkaddr++); 4403 kaddr = (unsigned char *)page_address(page); 4404 memset(kaddr, 0, PAGE_SIZE); 4405 4406 /* Step 1: write nat cache */ 4407 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 4408 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE); 4409 written_size += SUM_JOURNAL_SIZE; 4410 4411 /* Step 2: write sit cache */ 4412 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 4413 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE); 4414 written_size += SUM_JOURNAL_SIZE; 4415 4416 /* Step 3: write summary entries */ 4417 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 4418 seg_i = CURSEG_I(sbi, i); 4419 for (j = 0; j < f2fs_curseg_valid_blocks(sbi, i); j++) { 4420 if (!page) { 4421 page = f2fs_grab_meta_page(sbi, blkaddr++); 4422 kaddr = (unsigned char *)page_address(page); 4423 memset(kaddr, 0, PAGE_SIZE); 4424 written_size = 0; 4425 } 4426 summary = (struct f2fs_summary *)(kaddr + written_size); 4427 *summary = seg_i->sum_blk->entries[j]; 4428 written_size += SUMMARY_SIZE; 4429 4430 if (written_size + SUMMARY_SIZE <= PAGE_SIZE - 4431 SUM_FOOTER_SIZE) 4432 continue; 4433 4434 set_page_dirty(page); 4435 f2fs_put_page(page, 1); 4436 page = NULL; 4437 } 4438 } 4439 if (page) { 4440 set_page_dirty(page); 4441 f2fs_put_page(page, 1); 4442 } 4443 } 4444 4445 static void write_normal_summaries(struct f2fs_sb_info *sbi, 4446 block_t blkaddr, int type) 4447 { 4448 int i, end; 4449 4450 if (IS_DATASEG(type)) 4451 end = type + NR_CURSEG_DATA_TYPE; 4452 else 4453 end = type + NR_CURSEG_NODE_TYPE; 4454 4455 for (i = type; i < end; i++) 4456 write_current_sum_page(sbi, i, blkaddr + (i - type)); 4457 } 4458 4459 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 4460 { 4461 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) 4462 write_compacted_summaries(sbi, start_blk); 4463 else 4464 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 4465 } 4466 4467 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 4468 { 4469 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 4470 } 4471 4472 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type, 4473 unsigned int val, int alloc) 4474 { 4475 int i; 4476 4477 if (type == NAT_JOURNAL) { 4478 for (i = 0; i < nats_in_cursum(journal); i++) { 4479 if (le32_to_cpu(nid_in_journal(journal, i)) == val) 4480 return i; 4481 } 4482 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL)) 4483 return update_nats_in_cursum(journal, 1); 4484 } else if (type == SIT_JOURNAL) { 4485 for (i = 0; i < sits_in_cursum(journal); i++) 4486 if (le32_to_cpu(segno_in_journal(journal, i)) == val) 4487 return i; 4488 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL)) 4489 return update_sits_in_cursum(journal, 1); 4490 } 4491 return -1; 4492 } 4493 4494 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, 4495 unsigned int segno) 4496 { 4497 return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno)); 4498 } 4499 4500 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, 4501 unsigned int start) 4502 { 4503 struct sit_info *sit_i = SIT_I(sbi); 4504 struct page *page; 4505 pgoff_t src_off, dst_off; 4506 4507 src_off = current_sit_addr(sbi, start); 4508 dst_off = next_sit_addr(sbi, src_off); 4509 4510 page = f2fs_grab_meta_page(sbi, dst_off); 4511 seg_info_to_sit_page(sbi, page, start); 4512 4513 set_page_dirty(page); 4514 set_to_next_sit(sit_i, start); 4515 4516 return page; 4517 } 4518 4519 static struct sit_entry_set *grab_sit_entry_set(void) 4520 { 4521 struct sit_entry_set *ses = 4522 f2fs_kmem_cache_alloc(sit_entry_set_slab, 4523 GFP_NOFS, true, NULL); 4524 4525 ses->entry_cnt = 0; 4526 INIT_LIST_HEAD(&ses->set_list); 4527 return ses; 4528 } 4529 4530 static void release_sit_entry_set(struct sit_entry_set *ses) 4531 { 4532 list_del(&ses->set_list); 4533 kmem_cache_free(sit_entry_set_slab, ses); 4534 } 4535 4536 static void adjust_sit_entry_set(struct sit_entry_set *ses, 4537 struct list_head *head) 4538 { 4539 struct sit_entry_set *next = ses; 4540 4541 if (list_is_last(&ses->set_list, head)) 4542 return; 4543 4544 list_for_each_entry_continue(next, head, set_list) 4545 if (ses->entry_cnt <= next->entry_cnt) { 4546 list_move_tail(&ses->set_list, &next->set_list); 4547 return; 4548 } 4549 4550 list_move_tail(&ses->set_list, head); 4551 } 4552 4553 static void add_sit_entry(unsigned int segno, struct list_head *head) 4554 { 4555 struct sit_entry_set *ses; 4556 unsigned int start_segno = START_SEGNO(segno); 4557 4558 list_for_each_entry(ses, head, set_list) { 4559 if (ses->start_segno == start_segno) { 4560 ses->entry_cnt++; 4561 adjust_sit_entry_set(ses, head); 4562 return; 4563 } 4564 } 4565 4566 ses = grab_sit_entry_set(); 4567 4568 ses->start_segno = start_segno; 4569 ses->entry_cnt++; 4570 list_add(&ses->set_list, head); 4571 } 4572 4573 static void add_sits_in_set(struct f2fs_sb_info *sbi) 4574 { 4575 struct f2fs_sm_info *sm_info = SM_I(sbi); 4576 struct list_head *set_list = &sm_info->sit_entry_set; 4577 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; 4578 unsigned int segno; 4579 4580 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) 4581 add_sit_entry(segno, set_list); 4582 } 4583 4584 static void remove_sits_in_journal(struct f2fs_sb_info *sbi) 4585 { 4586 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4587 struct f2fs_journal *journal = curseg->journal; 4588 int i; 4589 4590 down_write(&curseg->journal_rwsem); 4591 for (i = 0; i < sits_in_cursum(journal); i++) { 4592 unsigned int segno; 4593 bool dirtied; 4594 4595 segno = le32_to_cpu(segno_in_journal(journal, i)); 4596 dirtied = __mark_sit_entry_dirty(sbi, segno); 4597 4598 if (!dirtied) 4599 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); 4600 } 4601 update_sits_in_cursum(journal, -i); 4602 up_write(&curseg->journal_rwsem); 4603 } 4604 4605 /* 4606 * CP calls this function, which flushes SIT entries including sit_journal, 4607 * and moves prefree segs to free segs. 4608 */ 4609 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 4610 { 4611 struct sit_info *sit_i = SIT_I(sbi); 4612 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 4613 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4614 struct f2fs_journal *journal = curseg->journal; 4615 struct sit_entry_set *ses, *tmp; 4616 struct list_head *head = &SM_I(sbi)->sit_entry_set; 4617 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS); 4618 struct seg_entry *se; 4619 4620 down_write(&sit_i->sentry_lock); 4621 4622 if (!sit_i->dirty_sentries) 4623 goto out; 4624 4625 /* 4626 * add and account sit entries of dirty bitmap in sit entry 4627 * set temporarily 4628 */ 4629 add_sits_in_set(sbi); 4630 4631 /* 4632 * if there are no enough space in journal to store dirty sit 4633 * entries, remove all entries from journal and add and account 4634 * them in sit entry set. 4635 */ 4636 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) || 4637 !to_journal) 4638 remove_sits_in_journal(sbi); 4639 4640 /* 4641 * there are two steps to flush sit entries: 4642 * #1, flush sit entries to journal in current cold data summary block. 4643 * #2, flush sit entries to sit page. 4644 */ 4645 list_for_each_entry_safe(ses, tmp, head, set_list) { 4646 struct page *page = NULL; 4647 struct f2fs_sit_block *raw_sit = NULL; 4648 unsigned int start_segno = ses->start_segno; 4649 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, 4650 (unsigned long)MAIN_SEGS(sbi)); 4651 unsigned int segno = start_segno; 4652 4653 if (to_journal && 4654 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL)) 4655 to_journal = false; 4656 4657 if (to_journal) { 4658 down_write(&curseg->journal_rwsem); 4659 } else { 4660 page = get_next_sit_page(sbi, start_segno); 4661 raw_sit = page_address(page); 4662 } 4663 4664 /* flush dirty sit entries in region of current sit set */ 4665 for_each_set_bit_from(segno, bitmap, end) { 4666 int offset, sit_offset; 4667 4668 se = get_seg_entry(sbi, segno); 4669 #ifdef CONFIG_F2FS_CHECK_FS 4670 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir, 4671 SIT_VBLOCK_MAP_SIZE)) 4672 f2fs_bug_on(sbi, 1); 4673 #endif 4674 4675 /* add discard candidates */ 4676 if (!(cpc->reason & CP_DISCARD)) { 4677 cpc->trim_start = segno; 4678 add_discard_addrs(sbi, cpc, false); 4679 } 4680 4681 if (to_journal) { 4682 offset = f2fs_lookup_journal_in_cursum(journal, 4683 SIT_JOURNAL, segno, 1); 4684 f2fs_bug_on(sbi, offset < 0); 4685 segno_in_journal(journal, offset) = 4686 cpu_to_le32(segno); 4687 seg_info_to_raw_sit(se, 4688 &sit_in_journal(journal, offset)); 4689 check_block_count(sbi, segno, 4690 &sit_in_journal(journal, offset)); 4691 } else { 4692 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 4693 seg_info_to_raw_sit(se, 4694 &raw_sit->entries[sit_offset]); 4695 check_block_count(sbi, segno, 4696 &raw_sit->entries[sit_offset]); 4697 } 4698 4699 __clear_bit(segno, bitmap); 4700 sit_i->dirty_sentries--; 4701 ses->entry_cnt--; 4702 } 4703 4704 if (to_journal) 4705 up_write(&curseg->journal_rwsem); 4706 else 4707 f2fs_put_page(page, 1); 4708 4709 f2fs_bug_on(sbi, ses->entry_cnt); 4710 release_sit_entry_set(ses); 4711 } 4712 4713 f2fs_bug_on(sbi, !list_empty(head)); 4714 f2fs_bug_on(sbi, sit_i->dirty_sentries); 4715 out: 4716 if (cpc->reason & CP_DISCARD) { 4717 __u64 trim_start = cpc->trim_start; 4718 4719 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) 4720 add_discard_addrs(sbi, cpc, false); 4721 4722 cpc->trim_start = trim_start; 4723 } 4724 up_write(&sit_i->sentry_lock); 4725 4726 set_prefree_as_free_segments(sbi); 4727 } 4728 4729 static int build_sit_info(struct f2fs_sb_info *sbi) 4730 { 4731 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 4732 struct sit_info *sit_i; 4733 unsigned int sit_segs, start; 4734 char *src_bitmap, *bitmap; 4735 unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size; 4736 unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0; 4737 4738 /* allocate memory for SIT information */ 4739 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL); 4740 if (!sit_i) 4741 return -ENOMEM; 4742 4743 SM_I(sbi)->sit_info = sit_i; 4744 4745 sit_i->sentries = 4746 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry), 4747 MAIN_SEGS(sbi)), 4748 GFP_KERNEL); 4749 if (!sit_i->sentries) 4750 return -ENOMEM; 4751 4752 main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4753 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size, 4754 GFP_KERNEL); 4755 if (!sit_i->dirty_sentries_bitmap) 4756 return -ENOMEM; 4757 4758 #ifdef CONFIG_F2FS_CHECK_FS 4759 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map); 4760 #else 4761 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map); 4762 #endif 4763 sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4764 if (!sit_i->bitmap) 4765 return -ENOMEM; 4766 4767 bitmap = sit_i->bitmap; 4768 4769 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4770 sit_i->sentries[start].cur_valid_map = bitmap; 4771 bitmap += SIT_VBLOCK_MAP_SIZE; 4772 4773 sit_i->sentries[start].ckpt_valid_map = bitmap; 4774 bitmap += SIT_VBLOCK_MAP_SIZE; 4775 4776 #ifdef CONFIG_F2FS_CHECK_FS 4777 sit_i->sentries[start].cur_valid_map_mir = bitmap; 4778 bitmap += SIT_VBLOCK_MAP_SIZE; 4779 #endif 4780 4781 if (discard_map) { 4782 sit_i->sentries[start].discard_map = bitmap; 4783 bitmap += SIT_VBLOCK_MAP_SIZE; 4784 } 4785 } 4786 4787 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 4788 if (!sit_i->tmp_map) 4789 return -ENOMEM; 4790 4791 if (__is_large_section(sbi)) { 4792 sit_i->sec_entries = 4793 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry), 4794 MAIN_SECS(sbi)), 4795 GFP_KERNEL); 4796 if (!sit_i->sec_entries) 4797 return -ENOMEM; 4798 } 4799 4800 /* get information related with SIT */ 4801 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 4802 4803 /* setup SIT bitmap from ckeckpoint pack */ 4804 sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 4805 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 4806 4807 sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL); 4808 if (!sit_i->sit_bitmap) 4809 return -ENOMEM; 4810 4811 #ifdef CONFIG_F2FS_CHECK_FS 4812 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, 4813 sit_bitmap_size, GFP_KERNEL); 4814 if (!sit_i->sit_bitmap_mir) 4815 return -ENOMEM; 4816 4817 sit_i->invalid_segmap = f2fs_kvzalloc(sbi, 4818 main_bitmap_size, GFP_KERNEL); 4819 if (!sit_i->invalid_segmap) 4820 return -ENOMEM; 4821 #endif 4822 4823 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 4824 sit_i->sit_blocks = SEGS_TO_BLKS(sbi, sit_segs); 4825 sit_i->written_valid_blocks = 0; 4826 sit_i->bitmap_size = sit_bitmap_size; 4827 sit_i->dirty_sentries = 0; 4828 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 4829 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 4830 sit_i->mounted_time = ktime_get_boottime_seconds(); 4831 init_rwsem(&sit_i->sentry_lock); 4832 return 0; 4833 } 4834 4835 static int build_free_segmap(struct f2fs_sb_info *sbi) 4836 { 4837 struct free_segmap_info *free_i; 4838 unsigned int bitmap_size, sec_bitmap_size; 4839 4840 /* allocate memory for free segmap information */ 4841 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL); 4842 if (!free_i) 4843 return -ENOMEM; 4844 4845 SM_I(sbi)->free_info = free_i; 4846 4847 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4848 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL); 4849 if (!free_i->free_segmap) 4850 return -ENOMEM; 4851 4852 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4853 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL); 4854 if (!free_i->free_secmap) 4855 return -ENOMEM; 4856 4857 /* set all segments as dirty temporarily */ 4858 memset(free_i->free_segmap, 0xff, bitmap_size); 4859 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 4860 4861 /* init free segmap information */ 4862 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); 4863 free_i->free_segments = 0; 4864 free_i->free_sections = 0; 4865 spin_lock_init(&free_i->segmap_lock); 4866 return 0; 4867 } 4868 4869 static int build_curseg(struct f2fs_sb_info *sbi) 4870 { 4871 struct curseg_info *array; 4872 int i; 4873 4874 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, 4875 sizeof(*array)), GFP_KERNEL); 4876 if (!array) 4877 return -ENOMEM; 4878 4879 SM_I(sbi)->curseg_array = array; 4880 4881 for (i = 0; i < NO_CHECK_TYPE; i++) { 4882 mutex_init(&array[i].curseg_mutex); 4883 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL); 4884 if (!array[i].sum_blk) 4885 return -ENOMEM; 4886 init_rwsem(&array[i].journal_rwsem); 4887 array[i].journal = f2fs_kzalloc(sbi, 4888 sizeof(struct f2fs_journal), GFP_KERNEL); 4889 if (!array[i].journal) 4890 return -ENOMEM; 4891 array[i].seg_type = log_type_to_seg_type(i); 4892 reset_curseg_fields(&array[i]); 4893 } 4894 return restore_curseg_summaries(sbi); 4895 } 4896 4897 static int build_sit_entries(struct f2fs_sb_info *sbi) 4898 { 4899 struct sit_info *sit_i = SIT_I(sbi); 4900 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4901 struct f2fs_journal *journal = curseg->journal; 4902 struct seg_entry *se; 4903 struct f2fs_sit_entry sit; 4904 int sit_blk_cnt = SIT_BLK_CNT(sbi); 4905 unsigned int i, start, end; 4906 unsigned int readed, start_blk = 0; 4907 int err = 0; 4908 block_t sit_valid_blocks[2] = {0, 0}; 4909 4910 do { 4911 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS, 4912 META_SIT, true); 4913 4914 start = start_blk * sit_i->sents_per_block; 4915 end = (start_blk + readed) * sit_i->sents_per_block; 4916 4917 for (; start < end && start < MAIN_SEGS(sbi); start++) { 4918 struct f2fs_sit_block *sit_blk; 4919 struct page *page; 4920 4921 se = &sit_i->sentries[start]; 4922 page = get_current_sit_page(sbi, start); 4923 if (IS_ERR(page)) 4924 return PTR_ERR(page); 4925 sit_blk = (struct f2fs_sit_block *)page_address(page); 4926 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 4927 f2fs_put_page(page, 1); 4928 4929 err = check_block_count(sbi, start, &sit); 4930 if (err) 4931 return err; 4932 seg_info_from_raw_sit(se, &sit); 4933 4934 if (se->type >= NR_PERSISTENT_LOG) { 4935 f2fs_err(sbi, "Invalid segment type: %u, segno: %u", 4936 se->type, start); 4937 f2fs_handle_error(sbi, 4938 ERROR_INCONSISTENT_SUM_TYPE); 4939 return -EFSCORRUPTED; 4940 } 4941 4942 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks; 4943 4944 if (!f2fs_block_unit_discard(sbi)) 4945 goto init_discard_map_done; 4946 4947 /* build discard map only one time */ 4948 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4949 memset(se->discard_map, 0xff, 4950 SIT_VBLOCK_MAP_SIZE); 4951 goto init_discard_map_done; 4952 } 4953 memcpy(se->discard_map, se->cur_valid_map, 4954 SIT_VBLOCK_MAP_SIZE); 4955 sbi->discard_blks += BLKS_PER_SEG(sbi) - 4956 se->valid_blocks; 4957 init_discard_map_done: 4958 if (__is_large_section(sbi)) 4959 get_sec_entry(sbi, start)->valid_blocks += 4960 se->valid_blocks; 4961 } 4962 start_blk += readed; 4963 } while (start_blk < sit_blk_cnt); 4964 4965 down_read(&curseg->journal_rwsem); 4966 for (i = 0; i < sits_in_cursum(journal); i++) { 4967 unsigned int old_valid_blocks; 4968 4969 start = le32_to_cpu(segno_in_journal(journal, i)); 4970 if (start >= MAIN_SEGS(sbi)) { 4971 f2fs_err(sbi, "Wrong journal entry on segno %u", 4972 start); 4973 err = -EFSCORRUPTED; 4974 f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL); 4975 break; 4976 } 4977 4978 se = &sit_i->sentries[start]; 4979 sit = sit_in_journal(journal, i); 4980 4981 old_valid_blocks = se->valid_blocks; 4982 4983 sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks; 4984 4985 err = check_block_count(sbi, start, &sit); 4986 if (err) 4987 break; 4988 seg_info_from_raw_sit(se, &sit); 4989 4990 if (se->type >= NR_PERSISTENT_LOG) { 4991 f2fs_err(sbi, "Invalid segment type: %u, segno: %u", 4992 se->type, start); 4993 err = -EFSCORRUPTED; 4994 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE); 4995 break; 4996 } 4997 4998 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks; 4999 5000 if (f2fs_block_unit_discard(sbi)) { 5001 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 5002 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE); 5003 } else { 5004 memcpy(se->discard_map, se->cur_valid_map, 5005 SIT_VBLOCK_MAP_SIZE); 5006 sbi->discard_blks += old_valid_blocks; 5007 sbi->discard_blks -= se->valid_blocks; 5008 } 5009 } 5010 5011 if (__is_large_section(sbi)) { 5012 get_sec_entry(sbi, start)->valid_blocks += 5013 se->valid_blocks; 5014 get_sec_entry(sbi, start)->valid_blocks -= 5015 old_valid_blocks; 5016 } 5017 } 5018 up_read(&curseg->journal_rwsem); 5019 5020 if (err) 5021 return err; 5022 5023 if (sit_valid_blocks[NODE] != valid_node_count(sbi)) { 5024 f2fs_err(sbi, "SIT is corrupted node# %u vs %u", 5025 sit_valid_blocks[NODE], valid_node_count(sbi)); 5026 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT); 5027 return -EFSCORRUPTED; 5028 } 5029 5030 if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] > 5031 valid_user_blocks(sbi)) { 5032 f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u", 5033 sit_valid_blocks[DATA], sit_valid_blocks[NODE], 5034 valid_user_blocks(sbi)); 5035 f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT); 5036 return -EFSCORRUPTED; 5037 } 5038 5039 return 0; 5040 } 5041 5042 static void init_free_segmap(struct f2fs_sb_info *sbi) 5043 { 5044 unsigned int start; 5045 int type; 5046 struct seg_entry *sentry; 5047 5048 for (start = 0; start < MAIN_SEGS(sbi); start++) { 5049 if (f2fs_usable_blks_in_seg(sbi, start) == 0) 5050 continue; 5051 sentry = get_seg_entry(sbi, start); 5052 if (!sentry->valid_blocks) 5053 __set_free(sbi, start); 5054 else 5055 SIT_I(sbi)->written_valid_blocks += 5056 sentry->valid_blocks; 5057 } 5058 5059 /* set use the current segments */ 5060 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 5061 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 5062 5063 __set_test_and_inuse(sbi, curseg_t->segno); 5064 } 5065 } 5066 5067 static void init_dirty_segmap(struct f2fs_sb_info *sbi) 5068 { 5069 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5070 struct free_segmap_info *free_i = FREE_I(sbi); 5071 unsigned int segno = 0, offset = 0, secno; 5072 block_t valid_blocks, usable_blks_in_seg; 5073 5074 while (1) { 5075 /* find dirty segment based on free segmap */ 5076 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); 5077 if (segno >= MAIN_SEGS(sbi)) 5078 break; 5079 offset = segno + 1; 5080 valid_blocks = get_valid_blocks(sbi, segno, false); 5081 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); 5082 if (valid_blocks == usable_blks_in_seg || !valid_blocks) 5083 continue; 5084 if (valid_blocks > usable_blks_in_seg) { 5085 f2fs_bug_on(sbi, 1); 5086 continue; 5087 } 5088 mutex_lock(&dirty_i->seglist_lock); 5089 __locate_dirty_segment(sbi, segno, DIRTY); 5090 mutex_unlock(&dirty_i->seglist_lock); 5091 } 5092 5093 if (!__is_large_section(sbi)) 5094 return; 5095 5096 mutex_lock(&dirty_i->seglist_lock); 5097 for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) { 5098 valid_blocks = get_valid_blocks(sbi, segno, true); 5099 secno = GET_SEC_FROM_SEG(sbi, segno); 5100 5101 if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi)) 5102 continue; 5103 if (IS_CURSEC(sbi, secno)) 5104 continue; 5105 set_bit(secno, dirty_i->dirty_secmap); 5106 } 5107 mutex_unlock(&dirty_i->seglist_lock); 5108 } 5109 5110 static int init_victim_secmap(struct f2fs_sb_info *sbi) 5111 { 5112 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5113 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 5114 5115 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 5116 if (!dirty_i->victim_secmap) 5117 return -ENOMEM; 5118 5119 dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 5120 if (!dirty_i->pinned_secmap) 5121 return -ENOMEM; 5122 5123 dirty_i->pinned_secmap_cnt = 0; 5124 dirty_i->enable_pin_section = true; 5125 return 0; 5126 } 5127 5128 static int build_dirty_segmap(struct f2fs_sb_info *sbi) 5129 { 5130 struct dirty_seglist_info *dirty_i; 5131 unsigned int bitmap_size, i; 5132 5133 /* allocate memory for dirty segments list information */ 5134 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info), 5135 GFP_KERNEL); 5136 if (!dirty_i) 5137 return -ENOMEM; 5138 5139 SM_I(sbi)->dirty_info = dirty_i; 5140 mutex_init(&dirty_i->seglist_lock); 5141 5142 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 5143 5144 for (i = 0; i < NR_DIRTY_TYPE; i++) { 5145 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size, 5146 GFP_KERNEL); 5147 if (!dirty_i->dirty_segmap[i]) 5148 return -ENOMEM; 5149 } 5150 5151 if (__is_large_section(sbi)) { 5152 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 5153 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi, 5154 bitmap_size, GFP_KERNEL); 5155 if (!dirty_i->dirty_secmap) 5156 return -ENOMEM; 5157 } 5158 5159 init_dirty_segmap(sbi); 5160 return init_victim_secmap(sbi); 5161 } 5162 5163 static int sanity_check_curseg(struct f2fs_sb_info *sbi) 5164 { 5165 int i; 5166 5167 /* 5168 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr; 5169 * In LFS curseg, all blkaddr after .next_blkoff should be unused. 5170 */ 5171 for (i = 0; i < NR_PERSISTENT_LOG; i++) { 5172 struct curseg_info *curseg = CURSEG_I(sbi, i); 5173 struct seg_entry *se = get_seg_entry(sbi, curseg->segno); 5174 unsigned int blkofs = curseg->next_blkoff; 5175 5176 if (f2fs_sb_has_readonly(sbi) && 5177 i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE) 5178 continue; 5179 5180 sanity_check_seg_type(sbi, curseg->seg_type); 5181 5182 if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) { 5183 f2fs_err(sbi, 5184 "Current segment has invalid alloc_type:%d", 5185 curseg->alloc_type); 5186 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG); 5187 return -EFSCORRUPTED; 5188 } 5189 5190 if (f2fs_test_bit(blkofs, se->cur_valid_map)) 5191 goto out; 5192 5193 if (curseg->alloc_type == SSR) 5194 continue; 5195 5196 for (blkofs += 1; blkofs < BLKS_PER_SEG(sbi); blkofs++) { 5197 if (!f2fs_test_bit(blkofs, se->cur_valid_map)) 5198 continue; 5199 out: 5200 f2fs_err(sbi, 5201 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u", 5202 i, curseg->segno, curseg->alloc_type, 5203 curseg->next_blkoff, blkofs); 5204 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG); 5205 return -EFSCORRUPTED; 5206 } 5207 } 5208 return 0; 5209 } 5210 5211 #ifdef CONFIG_BLK_DEV_ZONED 5212 static int check_zone_write_pointer(struct f2fs_sb_info *sbi, 5213 struct f2fs_dev_info *fdev, 5214 struct blk_zone *zone) 5215 { 5216 unsigned int zone_segno; 5217 block_t zone_block, valid_block_cnt; 5218 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 5219 int ret; 5220 unsigned int nofs_flags; 5221 5222 if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ) 5223 return 0; 5224 5225 zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block); 5226 zone_segno = GET_SEGNO(sbi, zone_block); 5227 5228 /* 5229 * Skip check of zones cursegs point to, since 5230 * fix_curseg_write_pointer() checks them. 5231 */ 5232 if (zone_segno >= MAIN_SEGS(sbi)) 5233 return 0; 5234 5235 /* 5236 * Get # of valid block of the zone. 5237 */ 5238 valid_block_cnt = get_valid_blocks(sbi, zone_segno, true); 5239 if (IS_CURSEC(sbi, GET_SEC_FROM_SEG(sbi, zone_segno))) { 5240 f2fs_notice(sbi, "Open zones: valid block[0x%x,0x%x] cond[%s]", 5241 zone_segno, valid_block_cnt, 5242 blk_zone_cond_str(zone->cond)); 5243 return 0; 5244 } 5245 5246 if ((!valid_block_cnt && zone->cond == BLK_ZONE_COND_EMPTY) || 5247 (valid_block_cnt && zone->cond == BLK_ZONE_COND_FULL)) 5248 return 0; 5249 5250 if (!valid_block_cnt) { 5251 f2fs_notice(sbi, "Zone without valid block has non-zero write " 5252 "pointer. Reset the write pointer: cond[%s]", 5253 blk_zone_cond_str(zone->cond)); 5254 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block, 5255 zone->len >> log_sectors_per_block); 5256 if (ret) 5257 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 5258 fdev->path, ret); 5259 return ret; 5260 } 5261 5262 /* 5263 * If there are valid blocks and the write pointer doesn't match 5264 * with them, we need to report the inconsistency and fill 5265 * the zone till the end to close the zone. This inconsistency 5266 * does not cause write error because the zone will not be 5267 * selected for write operation until it get discarded. 5268 */ 5269 f2fs_notice(sbi, "Valid blocks are not aligned with write " 5270 "pointer: valid block[0x%x,0x%x] cond[%s]", 5271 zone_segno, valid_block_cnt, blk_zone_cond_str(zone->cond)); 5272 5273 nofs_flags = memalloc_nofs_save(); 5274 ret = blkdev_zone_mgmt(fdev->bdev, REQ_OP_ZONE_FINISH, 5275 zone->start, zone->len); 5276 memalloc_nofs_restore(nofs_flags); 5277 if (ret == -EOPNOTSUPP) { 5278 ret = blkdev_issue_zeroout(fdev->bdev, zone->wp, 5279 zone->len - (zone->wp - zone->start), 5280 GFP_NOFS, 0); 5281 if (ret) 5282 f2fs_err(sbi, "Fill up zone failed: %s (errno=%d)", 5283 fdev->path, ret); 5284 } else if (ret) { 5285 f2fs_err(sbi, "Finishing zone failed: %s (errno=%d)", 5286 fdev->path, ret); 5287 } 5288 5289 return ret; 5290 } 5291 5292 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi, 5293 block_t zone_blkaddr) 5294 { 5295 int i; 5296 5297 for (i = 0; i < sbi->s_ndevs; i++) { 5298 if (!bdev_is_zoned(FDEV(i).bdev)) 5299 continue; 5300 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr && 5301 zone_blkaddr <= FDEV(i).end_blk)) 5302 return &FDEV(i); 5303 } 5304 5305 return NULL; 5306 } 5307 5308 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx, 5309 void *data) 5310 { 5311 memcpy(data, zone, sizeof(struct blk_zone)); 5312 return 0; 5313 } 5314 5315 static int do_fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type) 5316 { 5317 struct curseg_info *cs = CURSEG_I(sbi, type); 5318 struct f2fs_dev_info *zbd; 5319 struct blk_zone zone; 5320 unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off; 5321 block_t cs_zone_block, wp_block; 5322 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 5323 sector_t zone_sector; 5324 int err; 5325 5326 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 5327 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 5328 5329 zbd = get_target_zoned_dev(sbi, cs_zone_block); 5330 if (!zbd) 5331 return 0; 5332 5333 /* report zone for the sector the curseg points to */ 5334 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 5335 << log_sectors_per_block; 5336 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 5337 report_one_zone_cb, &zone); 5338 if (err != 1) { 5339 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 5340 zbd->path, err); 5341 return err; 5342 } 5343 5344 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 5345 return 0; 5346 5347 /* 5348 * When safely unmounted in the previous mount, we could use current 5349 * segments. Otherwise, allocate new sections. 5350 */ 5351 if (is_set_ckpt_flags(sbi, CP_UMOUNT_FLAG)) { 5352 wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block); 5353 wp_segno = GET_SEGNO(sbi, wp_block); 5354 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno); 5355 wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0); 5356 5357 if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff && 5358 wp_sector_off == 0) 5359 return 0; 5360 5361 f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: " 5362 "curseg[0x%x,0x%x] wp[0x%x,0x%x]", type, cs->segno, 5363 cs->next_blkoff, wp_segno, wp_blkoff); 5364 } 5365 5366 /* Allocate a new section if it's not new. */ 5367 if (cs->next_blkoff || 5368 cs->segno != GET_SEG_FROM_SEC(sbi, GET_ZONE_FROM_SEC(sbi, cs_section))) { 5369 unsigned int old_segno = cs->segno, old_blkoff = cs->next_blkoff; 5370 5371 f2fs_allocate_new_section(sbi, type, true); 5372 f2fs_notice(sbi, "Assign new section to curseg[%d]: " 5373 "[0x%x,0x%x] -> [0x%x,0x%x]", 5374 type, old_segno, old_blkoff, 5375 cs->segno, cs->next_blkoff); 5376 } 5377 5378 /* check consistency of the zone curseg pointed to */ 5379 if (check_zone_write_pointer(sbi, zbd, &zone)) 5380 return -EIO; 5381 5382 /* check newly assigned zone */ 5383 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 5384 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 5385 5386 zbd = get_target_zoned_dev(sbi, cs_zone_block); 5387 if (!zbd) 5388 return 0; 5389 5390 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 5391 << log_sectors_per_block; 5392 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 5393 report_one_zone_cb, &zone); 5394 if (err != 1) { 5395 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 5396 zbd->path, err); 5397 return err; 5398 } 5399 5400 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 5401 return 0; 5402 5403 if (zone.wp != zone.start) { 5404 f2fs_notice(sbi, 5405 "New zone for curseg[%d] is not yet discarded. " 5406 "Reset the zone: curseg[0x%x,0x%x]", 5407 type, cs->segno, cs->next_blkoff); 5408 err = __f2fs_issue_discard_zone(sbi, zbd->bdev, cs_zone_block, 5409 zone.len >> log_sectors_per_block); 5410 if (err) { 5411 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 5412 zbd->path, err); 5413 return err; 5414 } 5415 } 5416 5417 return 0; 5418 } 5419 5420 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 5421 { 5422 int i, ret; 5423 5424 for (i = 0; i < NR_PERSISTENT_LOG; i++) { 5425 ret = do_fix_curseg_write_pointer(sbi, i); 5426 if (ret) 5427 return ret; 5428 } 5429 5430 return 0; 5431 } 5432 5433 struct check_zone_write_pointer_args { 5434 struct f2fs_sb_info *sbi; 5435 struct f2fs_dev_info *fdev; 5436 }; 5437 5438 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx, 5439 void *data) 5440 { 5441 struct check_zone_write_pointer_args *args; 5442 5443 args = (struct check_zone_write_pointer_args *)data; 5444 5445 return check_zone_write_pointer(args->sbi, args->fdev, zone); 5446 } 5447 5448 static int check_write_pointer(struct f2fs_sb_info *sbi) 5449 { 5450 int i, ret; 5451 struct check_zone_write_pointer_args args; 5452 5453 for (i = 0; i < sbi->s_ndevs; i++) { 5454 if (!bdev_is_zoned(FDEV(i).bdev)) 5455 continue; 5456 5457 args.sbi = sbi; 5458 args.fdev = &FDEV(i); 5459 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES, 5460 check_zone_write_pointer_cb, &args); 5461 if (ret < 0) 5462 return ret; 5463 } 5464 5465 return 0; 5466 } 5467 5468 int f2fs_check_and_fix_write_pointer(struct f2fs_sb_info *sbi) 5469 { 5470 int ret; 5471 5472 if (!f2fs_sb_has_blkzoned(sbi) || f2fs_readonly(sbi->sb) || 5473 f2fs_hw_is_readonly(sbi)) 5474 return 0; 5475 5476 f2fs_notice(sbi, "Checking entire write pointers"); 5477 ret = fix_curseg_write_pointer(sbi); 5478 if (!ret) 5479 ret = check_write_pointer(sbi); 5480 return ret; 5481 } 5482 5483 /* 5484 * Return the number of usable blocks in a segment. The number of blocks 5485 * returned is always equal to the number of blocks in a segment for 5486 * segments fully contained within a sequential zone capacity or a 5487 * conventional zone. For segments partially contained in a sequential 5488 * zone capacity, the number of usable blocks up to the zone capacity 5489 * is returned. 0 is returned in all other cases. 5490 */ 5491 static inline unsigned int f2fs_usable_zone_blks_in_seg( 5492 struct f2fs_sb_info *sbi, unsigned int segno) 5493 { 5494 block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr; 5495 unsigned int secno; 5496 5497 if (!sbi->unusable_blocks_per_sec) 5498 return BLKS_PER_SEG(sbi); 5499 5500 secno = GET_SEC_FROM_SEG(sbi, segno); 5501 seg_start = START_BLOCK(sbi, segno); 5502 sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno)); 5503 sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi); 5504 5505 /* 5506 * If segment starts before zone capacity and spans beyond 5507 * zone capacity, then usable blocks are from seg start to 5508 * zone capacity. If the segment starts after the zone capacity, 5509 * then there are no usable blocks. 5510 */ 5511 if (seg_start >= sec_cap_blkaddr) 5512 return 0; 5513 if (seg_start + BLKS_PER_SEG(sbi) > sec_cap_blkaddr) 5514 return sec_cap_blkaddr - seg_start; 5515 5516 return BLKS_PER_SEG(sbi); 5517 } 5518 #else 5519 int f2fs_check_and_fix_write_pointer(struct f2fs_sb_info *sbi) 5520 { 5521 return 0; 5522 } 5523 5524 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi, 5525 unsigned int segno) 5526 { 5527 return 0; 5528 } 5529 5530 #endif 5531 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi, 5532 unsigned int segno) 5533 { 5534 if (f2fs_sb_has_blkzoned(sbi)) 5535 return f2fs_usable_zone_blks_in_seg(sbi, segno); 5536 5537 return BLKS_PER_SEG(sbi); 5538 } 5539 5540 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi) 5541 { 5542 if (f2fs_sb_has_blkzoned(sbi)) 5543 return CAP_SEGS_PER_SEC(sbi); 5544 5545 return SEGS_PER_SEC(sbi); 5546 } 5547 5548 unsigned long long f2fs_get_section_mtime(struct f2fs_sb_info *sbi, 5549 unsigned int segno) 5550 { 5551 unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi); 5552 unsigned int secno = 0, start = 0; 5553 unsigned int total_valid_blocks = 0; 5554 unsigned long long mtime = 0; 5555 unsigned int i = 0; 5556 5557 secno = GET_SEC_FROM_SEG(sbi, segno); 5558 start = GET_SEG_FROM_SEC(sbi, secno); 5559 5560 if (!__is_large_section(sbi)) { 5561 mtime = get_seg_entry(sbi, start + i)->mtime; 5562 goto out; 5563 } 5564 5565 for (i = 0; i < usable_segs_per_sec; i++) { 5566 /* for large section, only check the mtime of valid segments */ 5567 struct seg_entry *se = get_seg_entry(sbi, start+i); 5568 5569 mtime += se->mtime * se->valid_blocks; 5570 total_valid_blocks += se->valid_blocks; 5571 } 5572 5573 if (total_valid_blocks == 0) 5574 return INVALID_MTIME; 5575 5576 mtime = div_u64(mtime, total_valid_blocks); 5577 out: 5578 if (unlikely(mtime == INVALID_MTIME)) 5579 mtime -= 1; 5580 return mtime; 5581 } 5582 5583 /* 5584 * Update min, max modified time for cost-benefit GC algorithm 5585 */ 5586 static void init_min_max_mtime(struct f2fs_sb_info *sbi) 5587 { 5588 struct sit_info *sit_i = SIT_I(sbi); 5589 unsigned int segno; 5590 5591 down_write(&sit_i->sentry_lock); 5592 5593 sit_i->min_mtime = ULLONG_MAX; 5594 5595 for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) { 5596 unsigned long long mtime = 0; 5597 5598 mtime = f2fs_get_section_mtime(sbi, segno); 5599 5600 if (sit_i->min_mtime > mtime) 5601 sit_i->min_mtime = mtime; 5602 } 5603 sit_i->max_mtime = get_mtime(sbi, false); 5604 sit_i->dirty_max_mtime = 0; 5605 up_write(&sit_i->sentry_lock); 5606 } 5607 5608 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi) 5609 { 5610 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 5611 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 5612 struct f2fs_sm_info *sm_info; 5613 int err; 5614 5615 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL); 5616 if (!sm_info) 5617 return -ENOMEM; 5618 5619 /* init sm info */ 5620 sbi->sm_info = sm_info; 5621 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 5622 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 5623 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 5624 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 5625 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 5626 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 5627 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 5628 sm_info->rec_prefree_segments = sm_info->main_segments * 5629 DEF_RECLAIM_PREFREE_SEGMENTS / 100; 5630 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS) 5631 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS; 5632 5633 if (!f2fs_lfs_mode(sbi)) 5634 sm_info->ipu_policy = BIT(F2FS_IPU_FSYNC); 5635 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; 5636 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; 5637 sm_info->min_seq_blocks = BLKS_PER_SEG(sbi); 5638 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS; 5639 sm_info->min_ssr_sections = reserved_sections(sbi); 5640 5641 INIT_LIST_HEAD(&sm_info->sit_entry_set); 5642 5643 init_f2fs_rwsem(&sm_info->curseg_lock); 5644 5645 err = f2fs_create_flush_cmd_control(sbi); 5646 if (err) 5647 return err; 5648 5649 err = create_discard_cmd_control(sbi); 5650 if (err) 5651 return err; 5652 5653 err = build_sit_info(sbi); 5654 if (err) 5655 return err; 5656 err = build_free_segmap(sbi); 5657 if (err) 5658 return err; 5659 err = build_curseg(sbi); 5660 if (err) 5661 return err; 5662 5663 /* reinit free segmap based on SIT */ 5664 err = build_sit_entries(sbi); 5665 if (err) 5666 return err; 5667 5668 init_free_segmap(sbi); 5669 err = build_dirty_segmap(sbi); 5670 if (err) 5671 return err; 5672 5673 err = sanity_check_curseg(sbi); 5674 if (err) 5675 return err; 5676 5677 init_min_max_mtime(sbi); 5678 return 0; 5679 } 5680 5681 static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 5682 enum dirty_type dirty_type) 5683 { 5684 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5685 5686 mutex_lock(&dirty_i->seglist_lock); 5687 kvfree(dirty_i->dirty_segmap[dirty_type]); 5688 dirty_i->nr_dirty[dirty_type] = 0; 5689 mutex_unlock(&dirty_i->seglist_lock); 5690 } 5691 5692 static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 5693 { 5694 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5695 5696 kvfree(dirty_i->pinned_secmap); 5697 kvfree(dirty_i->victim_secmap); 5698 } 5699 5700 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 5701 { 5702 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5703 int i; 5704 5705 if (!dirty_i) 5706 return; 5707 5708 /* discard pre-free/dirty segments list */ 5709 for (i = 0; i < NR_DIRTY_TYPE; i++) 5710 discard_dirty_segmap(sbi, i); 5711 5712 if (__is_large_section(sbi)) { 5713 mutex_lock(&dirty_i->seglist_lock); 5714 kvfree(dirty_i->dirty_secmap); 5715 mutex_unlock(&dirty_i->seglist_lock); 5716 } 5717 5718 destroy_victim_secmap(sbi); 5719 SM_I(sbi)->dirty_info = NULL; 5720 kfree(dirty_i); 5721 } 5722 5723 static void destroy_curseg(struct f2fs_sb_info *sbi) 5724 { 5725 struct curseg_info *array = SM_I(sbi)->curseg_array; 5726 int i; 5727 5728 if (!array) 5729 return; 5730 SM_I(sbi)->curseg_array = NULL; 5731 for (i = 0; i < NR_CURSEG_TYPE; i++) { 5732 kfree(array[i].sum_blk); 5733 kfree(array[i].journal); 5734 } 5735 kfree(array); 5736 } 5737 5738 static void destroy_free_segmap(struct f2fs_sb_info *sbi) 5739 { 5740 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 5741 5742 if (!free_i) 5743 return; 5744 SM_I(sbi)->free_info = NULL; 5745 kvfree(free_i->free_segmap); 5746 kvfree(free_i->free_secmap); 5747 kfree(free_i); 5748 } 5749 5750 static void destroy_sit_info(struct f2fs_sb_info *sbi) 5751 { 5752 struct sit_info *sit_i = SIT_I(sbi); 5753 5754 if (!sit_i) 5755 return; 5756 5757 if (sit_i->sentries) 5758 kvfree(sit_i->bitmap); 5759 kfree(sit_i->tmp_map); 5760 5761 kvfree(sit_i->sentries); 5762 kvfree(sit_i->sec_entries); 5763 kvfree(sit_i->dirty_sentries_bitmap); 5764 5765 SM_I(sbi)->sit_info = NULL; 5766 kvfree(sit_i->sit_bitmap); 5767 #ifdef CONFIG_F2FS_CHECK_FS 5768 kvfree(sit_i->sit_bitmap_mir); 5769 kvfree(sit_i->invalid_segmap); 5770 #endif 5771 kfree(sit_i); 5772 } 5773 5774 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi) 5775 { 5776 struct f2fs_sm_info *sm_info = SM_I(sbi); 5777 5778 if (!sm_info) 5779 return; 5780 f2fs_destroy_flush_cmd_control(sbi, true); 5781 destroy_discard_cmd_control(sbi); 5782 destroy_dirty_segmap(sbi); 5783 destroy_curseg(sbi); 5784 destroy_free_segmap(sbi); 5785 destroy_sit_info(sbi); 5786 sbi->sm_info = NULL; 5787 kfree(sm_info); 5788 } 5789 5790 int __init f2fs_create_segment_manager_caches(void) 5791 { 5792 discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry", 5793 sizeof(struct discard_entry)); 5794 if (!discard_entry_slab) 5795 goto fail; 5796 5797 discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd", 5798 sizeof(struct discard_cmd)); 5799 if (!discard_cmd_slab) 5800 goto destroy_discard_entry; 5801 5802 sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set", 5803 sizeof(struct sit_entry_set)); 5804 if (!sit_entry_set_slab) 5805 goto destroy_discard_cmd; 5806 5807 revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry", 5808 sizeof(struct revoke_entry)); 5809 if (!revoke_entry_slab) 5810 goto destroy_sit_entry_set; 5811 return 0; 5812 5813 destroy_sit_entry_set: 5814 kmem_cache_destroy(sit_entry_set_slab); 5815 destroy_discard_cmd: 5816 kmem_cache_destroy(discard_cmd_slab); 5817 destroy_discard_entry: 5818 kmem_cache_destroy(discard_entry_slab); 5819 fail: 5820 return -ENOMEM; 5821 } 5822 5823 void f2fs_destroy_segment_manager_caches(void) 5824 { 5825 kmem_cache_destroy(sit_entry_set_slab); 5826 kmem_cache_destroy(discard_cmd_slab); 5827 kmem_cache_destroy(discard_entry_slab); 5828 kmem_cache_destroy(revoke_entry_slab); 5829 } 5830