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 stat_dec_atomic_inode(inode); 203 204 F2FS_I(inode)->atomic_write_task = NULL; 205 206 if (clean) { 207 f2fs_i_size_write(inode, fi->original_i_size); 208 fi->original_i_size = 0; 209 } 210 /* avoid stale dirty inode during eviction */ 211 sync_inode_metadata(inode, 0); 212 } 213 214 static int __replace_atomic_write_block(struct inode *inode, pgoff_t index, 215 block_t new_addr, block_t *old_addr, bool recover) 216 { 217 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 218 struct dnode_of_data dn; 219 struct node_info ni; 220 int err; 221 222 retry: 223 set_new_dnode(&dn, inode, NULL, NULL, 0); 224 err = f2fs_get_dnode_of_data(&dn, index, ALLOC_NODE); 225 if (err) { 226 if (err == -ENOMEM) { 227 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 228 goto retry; 229 } 230 return err; 231 } 232 233 err = f2fs_get_node_info(sbi, dn.nid, &ni, false); 234 if (err) { 235 f2fs_put_dnode(&dn); 236 return err; 237 } 238 239 if (recover) { 240 /* dn.data_blkaddr is always valid */ 241 if (!__is_valid_data_blkaddr(new_addr)) { 242 if (new_addr == NULL_ADDR) 243 dec_valid_block_count(sbi, inode, 1); 244 f2fs_invalidate_blocks(sbi, dn.data_blkaddr); 245 f2fs_update_data_blkaddr(&dn, new_addr); 246 } else { 247 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, 248 new_addr, ni.version, true, true); 249 } 250 } else { 251 blkcnt_t count = 1; 252 253 err = inc_valid_block_count(sbi, inode, &count, true); 254 if (err) { 255 f2fs_put_dnode(&dn); 256 return err; 257 } 258 259 *old_addr = dn.data_blkaddr; 260 f2fs_truncate_data_blocks_range(&dn, 1); 261 dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count); 262 263 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr, 264 ni.version, true, false); 265 } 266 267 f2fs_put_dnode(&dn); 268 269 trace_f2fs_replace_atomic_write_block(inode, F2FS_I(inode)->cow_inode, 270 index, old_addr ? *old_addr : 0, new_addr, recover); 271 return 0; 272 } 273 274 static void __complete_revoke_list(struct inode *inode, struct list_head *head, 275 bool revoke) 276 { 277 struct revoke_entry *cur, *tmp; 278 pgoff_t start_index = 0; 279 bool truncate = is_inode_flag_set(inode, FI_ATOMIC_REPLACE); 280 281 list_for_each_entry_safe(cur, tmp, head, list) { 282 if (revoke) { 283 __replace_atomic_write_block(inode, cur->index, 284 cur->old_addr, NULL, true); 285 } else if (truncate) { 286 f2fs_truncate_hole(inode, start_index, cur->index); 287 start_index = cur->index + 1; 288 } 289 290 list_del(&cur->list); 291 kmem_cache_free(revoke_entry_slab, cur); 292 } 293 294 if (!revoke && truncate) 295 f2fs_do_truncate_blocks(inode, start_index * PAGE_SIZE, false); 296 } 297 298 static int __f2fs_commit_atomic_write(struct inode *inode) 299 { 300 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 301 struct f2fs_inode_info *fi = F2FS_I(inode); 302 struct inode *cow_inode = fi->cow_inode; 303 struct revoke_entry *new; 304 struct list_head revoke_list; 305 block_t blkaddr; 306 struct dnode_of_data dn; 307 pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); 308 pgoff_t off = 0, blen, index; 309 int ret = 0, i; 310 311 INIT_LIST_HEAD(&revoke_list); 312 313 while (len) { 314 blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len); 315 316 set_new_dnode(&dn, cow_inode, NULL, NULL, 0); 317 ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA); 318 if (ret && ret != -ENOENT) { 319 goto out; 320 } else if (ret == -ENOENT) { 321 ret = 0; 322 if (dn.max_level == 0) 323 goto out; 324 goto next; 325 } 326 327 blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_page, cow_inode), 328 len); 329 index = off; 330 for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) { 331 blkaddr = f2fs_data_blkaddr(&dn); 332 333 if (!__is_valid_data_blkaddr(blkaddr)) { 334 continue; 335 } else if (!f2fs_is_valid_blkaddr(sbi, blkaddr, 336 DATA_GENERIC_ENHANCE)) { 337 f2fs_put_dnode(&dn); 338 ret = -EFSCORRUPTED; 339 f2fs_handle_error(sbi, 340 ERROR_INVALID_BLKADDR); 341 goto out; 342 } 343 344 new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS, 345 true, NULL); 346 347 ret = __replace_atomic_write_block(inode, index, blkaddr, 348 &new->old_addr, false); 349 if (ret) { 350 f2fs_put_dnode(&dn); 351 kmem_cache_free(revoke_entry_slab, new); 352 goto out; 353 } 354 355 f2fs_update_data_blkaddr(&dn, NULL_ADDR); 356 new->index = index; 357 list_add_tail(&new->list, &revoke_list); 358 } 359 f2fs_put_dnode(&dn); 360 next: 361 off += blen; 362 len -= blen; 363 } 364 365 out: 366 if (ret) { 367 sbi->revoked_atomic_block += fi->atomic_write_cnt; 368 } else { 369 sbi->committed_atomic_block += fi->atomic_write_cnt; 370 set_inode_flag(inode, FI_ATOMIC_COMMITTED); 371 } 372 373 __complete_revoke_list(inode, &revoke_list, ret ? true : false); 374 375 return ret; 376 } 377 378 int f2fs_commit_atomic_write(struct inode *inode) 379 { 380 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 381 struct f2fs_inode_info *fi = F2FS_I(inode); 382 int err; 383 384 err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX); 385 if (err) 386 return err; 387 388 f2fs_down_write(&fi->i_gc_rwsem[WRITE]); 389 f2fs_lock_op(sbi); 390 391 err = __f2fs_commit_atomic_write(inode); 392 393 f2fs_unlock_op(sbi); 394 f2fs_up_write(&fi->i_gc_rwsem[WRITE]); 395 396 return err; 397 } 398 399 /* 400 * This function balances dirty node and dentry pages. 401 * In addition, it controls garbage collection. 402 */ 403 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need) 404 { 405 if (time_to_inject(sbi, FAULT_CHECKPOINT)) 406 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT); 407 408 /* balance_fs_bg is able to be pending */ 409 if (need && excess_cached_nats(sbi)) 410 f2fs_balance_fs_bg(sbi, false); 411 412 if (!f2fs_is_checkpoint_ready(sbi)) 413 return; 414 415 /* 416 * We should do GC or end up with checkpoint, if there are so many dirty 417 * dir/node pages without enough free segments. 418 */ 419 if (has_enough_free_secs(sbi, 0, 0)) 420 return; 421 422 if (test_opt(sbi, GC_MERGE) && sbi->gc_thread && 423 sbi->gc_thread->f2fs_gc_task) { 424 DEFINE_WAIT(wait); 425 426 prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait, 427 TASK_UNINTERRUPTIBLE); 428 wake_up(&sbi->gc_thread->gc_wait_queue_head); 429 io_schedule(); 430 finish_wait(&sbi->gc_thread->fggc_wq, &wait); 431 } else { 432 struct f2fs_gc_control gc_control = { 433 .victim_segno = NULL_SEGNO, 434 .init_gc_type = BG_GC, 435 .no_bg_gc = true, 436 .should_migrate_blocks = false, 437 .err_gc_skipped = false, 438 .nr_free_secs = 1 }; 439 f2fs_down_write(&sbi->gc_lock); 440 stat_inc_gc_call_count(sbi, FOREGROUND); 441 f2fs_gc(sbi, &gc_control); 442 } 443 } 444 445 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi) 446 { 447 int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2; 448 unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS); 449 unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA); 450 unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES); 451 unsigned int meta = get_pages(sbi, F2FS_DIRTY_META); 452 unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA); 453 unsigned int threshold = (factor * DEFAULT_DIRTY_THRESHOLD) << 454 sbi->log_blocks_per_seg; 455 unsigned int global_threshold = threshold * 3 / 2; 456 457 if (dents >= threshold || qdata >= threshold || 458 nodes >= threshold || meta >= threshold || 459 imeta >= threshold) 460 return true; 461 return dents + qdata + nodes + meta + imeta > global_threshold; 462 } 463 464 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg) 465 { 466 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 467 return; 468 469 /* try to shrink extent cache when there is no enough memory */ 470 if (!f2fs_available_free_memory(sbi, READ_EXTENT_CACHE)) 471 f2fs_shrink_read_extent_tree(sbi, 472 READ_EXTENT_CACHE_SHRINK_NUMBER); 473 474 /* try to shrink age extent cache when there is no enough memory */ 475 if (!f2fs_available_free_memory(sbi, AGE_EXTENT_CACHE)) 476 f2fs_shrink_age_extent_tree(sbi, 477 AGE_EXTENT_CACHE_SHRINK_NUMBER); 478 479 /* check the # of cached NAT entries */ 480 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES)) 481 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK); 482 483 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) 484 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS); 485 else 486 f2fs_build_free_nids(sbi, false, false); 487 488 if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) || 489 excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi)) 490 goto do_sync; 491 492 /* there is background inflight IO or foreground operation recently */ 493 if (is_inflight_io(sbi, REQ_TIME) || 494 (!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem))) 495 return; 496 497 /* exceed periodical checkpoint timeout threshold */ 498 if (f2fs_time_over(sbi, CP_TIME)) 499 goto do_sync; 500 501 /* checkpoint is the only way to shrink partial cached entries */ 502 if (f2fs_available_free_memory(sbi, NAT_ENTRIES) && 503 f2fs_available_free_memory(sbi, INO_ENTRIES)) 504 return; 505 506 do_sync: 507 if (test_opt(sbi, DATA_FLUSH) && from_bg) { 508 struct blk_plug plug; 509 510 mutex_lock(&sbi->flush_lock); 511 512 blk_start_plug(&plug); 513 f2fs_sync_dirty_inodes(sbi, FILE_INODE, false); 514 blk_finish_plug(&plug); 515 516 mutex_unlock(&sbi->flush_lock); 517 } 518 stat_inc_cp_call_count(sbi, BACKGROUND); 519 f2fs_sync_fs(sbi->sb, 1); 520 } 521 522 static int __submit_flush_wait(struct f2fs_sb_info *sbi, 523 struct block_device *bdev) 524 { 525 int ret = blkdev_issue_flush(bdev); 526 527 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER), 528 test_opt(sbi, FLUSH_MERGE), ret); 529 if (!ret) 530 f2fs_update_iostat(sbi, NULL, FS_FLUSH_IO, 0); 531 return ret; 532 } 533 534 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino) 535 { 536 int ret = 0; 537 int i; 538 539 if (!f2fs_is_multi_device(sbi)) 540 return __submit_flush_wait(sbi, sbi->sb->s_bdev); 541 542 for (i = 0; i < sbi->s_ndevs; i++) { 543 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO)) 544 continue; 545 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 546 if (ret) 547 break; 548 } 549 return ret; 550 } 551 552 static int issue_flush_thread(void *data) 553 { 554 struct f2fs_sb_info *sbi = data; 555 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 556 wait_queue_head_t *q = &fcc->flush_wait_queue; 557 repeat: 558 if (kthread_should_stop()) 559 return 0; 560 561 if (!llist_empty(&fcc->issue_list)) { 562 struct flush_cmd *cmd, *next; 563 int ret; 564 565 fcc->dispatch_list = llist_del_all(&fcc->issue_list); 566 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); 567 568 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode); 569 570 ret = submit_flush_wait(sbi, cmd->ino); 571 atomic_inc(&fcc->issued_flush); 572 573 llist_for_each_entry_safe(cmd, next, 574 fcc->dispatch_list, llnode) { 575 cmd->ret = ret; 576 complete(&cmd->wait); 577 } 578 fcc->dispatch_list = NULL; 579 } 580 581 wait_event_interruptible(*q, 582 kthread_should_stop() || !llist_empty(&fcc->issue_list)); 583 goto repeat; 584 } 585 586 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino) 587 { 588 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 589 struct flush_cmd cmd; 590 int ret; 591 592 if (test_opt(sbi, NOBARRIER)) 593 return 0; 594 595 if (!test_opt(sbi, FLUSH_MERGE)) { 596 atomic_inc(&fcc->queued_flush); 597 ret = submit_flush_wait(sbi, ino); 598 atomic_dec(&fcc->queued_flush); 599 atomic_inc(&fcc->issued_flush); 600 return ret; 601 } 602 603 if (atomic_inc_return(&fcc->queued_flush) == 1 || 604 f2fs_is_multi_device(sbi)) { 605 ret = submit_flush_wait(sbi, ino); 606 atomic_dec(&fcc->queued_flush); 607 608 atomic_inc(&fcc->issued_flush); 609 return ret; 610 } 611 612 cmd.ino = ino; 613 init_completion(&cmd.wait); 614 615 llist_add(&cmd.llnode, &fcc->issue_list); 616 617 /* 618 * update issue_list before we wake up issue_flush thread, this 619 * smp_mb() pairs with another barrier in ___wait_event(), see 620 * more details in comments of waitqueue_active(). 621 */ 622 smp_mb(); 623 624 if (waitqueue_active(&fcc->flush_wait_queue)) 625 wake_up(&fcc->flush_wait_queue); 626 627 if (fcc->f2fs_issue_flush) { 628 wait_for_completion(&cmd.wait); 629 atomic_dec(&fcc->queued_flush); 630 } else { 631 struct llist_node *list; 632 633 list = llist_del_all(&fcc->issue_list); 634 if (!list) { 635 wait_for_completion(&cmd.wait); 636 atomic_dec(&fcc->queued_flush); 637 } else { 638 struct flush_cmd *tmp, *next; 639 640 ret = submit_flush_wait(sbi, ino); 641 642 llist_for_each_entry_safe(tmp, next, list, llnode) { 643 if (tmp == &cmd) { 644 cmd.ret = ret; 645 atomic_dec(&fcc->queued_flush); 646 continue; 647 } 648 tmp->ret = ret; 649 complete(&tmp->wait); 650 } 651 } 652 } 653 654 return cmd.ret; 655 } 656 657 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi) 658 { 659 dev_t dev = sbi->sb->s_bdev->bd_dev; 660 struct flush_cmd_control *fcc; 661 662 if (SM_I(sbi)->fcc_info) { 663 fcc = SM_I(sbi)->fcc_info; 664 if (fcc->f2fs_issue_flush) 665 return 0; 666 goto init_thread; 667 } 668 669 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL); 670 if (!fcc) 671 return -ENOMEM; 672 atomic_set(&fcc->issued_flush, 0); 673 atomic_set(&fcc->queued_flush, 0); 674 init_waitqueue_head(&fcc->flush_wait_queue); 675 init_llist_head(&fcc->issue_list); 676 SM_I(sbi)->fcc_info = fcc; 677 if (!test_opt(sbi, FLUSH_MERGE)) 678 return 0; 679 680 init_thread: 681 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, 682 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); 683 if (IS_ERR(fcc->f2fs_issue_flush)) { 684 int err = PTR_ERR(fcc->f2fs_issue_flush); 685 686 fcc->f2fs_issue_flush = NULL; 687 return err; 688 } 689 690 return 0; 691 } 692 693 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free) 694 { 695 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 696 697 if (fcc && fcc->f2fs_issue_flush) { 698 struct task_struct *flush_thread = fcc->f2fs_issue_flush; 699 700 fcc->f2fs_issue_flush = NULL; 701 kthread_stop(flush_thread); 702 } 703 if (free) { 704 kfree(fcc); 705 SM_I(sbi)->fcc_info = NULL; 706 } 707 } 708 709 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi) 710 { 711 int ret = 0, i; 712 713 if (!f2fs_is_multi_device(sbi)) 714 return 0; 715 716 if (test_opt(sbi, NOBARRIER)) 717 return 0; 718 719 for (i = 1; i < sbi->s_ndevs; i++) { 720 int count = DEFAULT_RETRY_IO_COUNT; 721 722 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device)) 723 continue; 724 725 do { 726 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 727 if (ret) 728 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 729 } while (ret && --count); 730 731 if (ret) { 732 f2fs_stop_checkpoint(sbi, false, 733 STOP_CP_REASON_FLUSH_FAIL); 734 break; 735 } 736 737 spin_lock(&sbi->dev_lock); 738 f2fs_clear_bit(i, (char *)&sbi->dirty_device); 739 spin_unlock(&sbi->dev_lock); 740 } 741 742 return ret; 743 } 744 745 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 746 enum dirty_type dirty_type) 747 { 748 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 749 750 /* need not be added */ 751 if (IS_CURSEG(sbi, segno)) 752 return; 753 754 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 755 dirty_i->nr_dirty[dirty_type]++; 756 757 if (dirty_type == DIRTY) { 758 struct seg_entry *sentry = get_seg_entry(sbi, segno); 759 enum dirty_type t = sentry->type; 760 761 if (unlikely(t >= DIRTY)) { 762 f2fs_bug_on(sbi, 1); 763 return; 764 } 765 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) 766 dirty_i->nr_dirty[t]++; 767 768 if (__is_large_section(sbi)) { 769 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 770 block_t valid_blocks = 771 get_valid_blocks(sbi, segno, true); 772 773 f2fs_bug_on(sbi, unlikely(!valid_blocks || 774 valid_blocks == CAP_BLKS_PER_SEC(sbi))); 775 776 if (!IS_CURSEC(sbi, secno)) 777 set_bit(secno, dirty_i->dirty_secmap); 778 } 779 } 780 } 781 782 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 783 enum dirty_type dirty_type) 784 { 785 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 786 block_t valid_blocks; 787 788 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) 789 dirty_i->nr_dirty[dirty_type]--; 790 791 if (dirty_type == DIRTY) { 792 struct seg_entry *sentry = get_seg_entry(sbi, segno); 793 enum dirty_type t = sentry->type; 794 795 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) 796 dirty_i->nr_dirty[t]--; 797 798 valid_blocks = get_valid_blocks(sbi, segno, true); 799 if (valid_blocks == 0) { 800 clear_bit(GET_SEC_FROM_SEG(sbi, segno), 801 dirty_i->victim_secmap); 802 #ifdef CONFIG_F2FS_CHECK_FS 803 clear_bit(segno, SIT_I(sbi)->invalid_segmap); 804 #endif 805 } 806 if (__is_large_section(sbi)) { 807 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 808 809 if (!valid_blocks || 810 valid_blocks == CAP_BLKS_PER_SEC(sbi)) { 811 clear_bit(secno, dirty_i->dirty_secmap); 812 return; 813 } 814 815 if (!IS_CURSEC(sbi, secno)) 816 set_bit(secno, dirty_i->dirty_secmap); 817 } 818 } 819 } 820 821 /* 822 * Should not occur error such as -ENOMEM. 823 * Adding dirty entry into seglist is not critical operation. 824 * If a given segment is one of current working segments, it won't be added. 825 */ 826 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) 827 { 828 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 829 unsigned short valid_blocks, ckpt_valid_blocks; 830 unsigned int usable_blocks; 831 832 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) 833 return; 834 835 usable_blocks = f2fs_usable_blks_in_seg(sbi, segno); 836 mutex_lock(&dirty_i->seglist_lock); 837 838 valid_blocks = get_valid_blocks(sbi, segno, false); 839 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false); 840 841 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) || 842 ckpt_valid_blocks == usable_blocks)) { 843 __locate_dirty_segment(sbi, segno, PRE); 844 __remove_dirty_segment(sbi, segno, DIRTY); 845 } else if (valid_blocks < usable_blocks) { 846 __locate_dirty_segment(sbi, segno, DIRTY); 847 } else { 848 /* Recovery routine with SSR needs this */ 849 __remove_dirty_segment(sbi, segno, DIRTY); 850 } 851 852 mutex_unlock(&dirty_i->seglist_lock); 853 } 854 855 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */ 856 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi) 857 { 858 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 859 unsigned int segno; 860 861 mutex_lock(&dirty_i->seglist_lock); 862 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 863 if (get_valid_blocks(sbi, segno, false)) 864 continue; 865 if (IS_CURSEG(sbi, segno)) 866 continue; 867 __locate_dirty_segment(sbi, segno, PRE); 868 __remove_dirty_segment(sbi, segno, DIRTY); 869 } 870 mutex_unlock(&dirty_i->seglist_lock); 871 } 872 873 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi) 874 { 875 int ovp_hole_segs = 876 (overprovision_segments(sbi) - reserved_segments(sbi)); 877 block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg; 878 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 879 block_t holes[2] = {0, 0}; /* DATA and NODE */ 880 block_t unusable; 881 struct seg_entry *se; 882 unsigned int segno; 883 884 mutex_lock(&dirty_i->seglist_lock); 885 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 886 se = get_seg_entry(sbi, segno); 887 if (IS_NODESEG(se->type)) 888 holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) - 889 se->valid_blocks; 890 else 891 holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) - 892 se->valid_blocks; 893 } 894 mutex_unlock(&dirty_i->seglist_lock); 895 896 unusable = max(holes[DATA], holes[NODE]); 897 if (unusable > ovp_holes) 898 return unusable - ovp_holes; 899 return 0; 900 } 901 902 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable) 903 { 904 int ovp_hole_segs = 905 (overprovision_segments(sbi) - reserved_segments(sbi)); 906 if (unusable > F2FS_OPTION(sbi).unusable_cap) 907 return -EAGAIN; 908 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) && 909 dirty_segments(sbi) > ovp_hole_segs) 910 return -EAGAIN; 911 return 0; 912 } 913 914 /* This is only used by SBI_CP_DISABLED */ 915 static unsigned int get_free_segment(struct f2fs_sb_info *sbi) 916 { 917 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 918 unsigned int segno = 0; 919 920 mutex_lock(&dirty_i->seglist_lock); 921 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 922 if (get_valid_blocks(sbi, segno, false)) 923 continue; 924 if (get_ckpt_valid_blocks(sbi, segno, false)) 925 continue; 926 mutex_unlock(&dirty_i->seglist_lock); 927 return segno; 928 } 929 mutex_unlock(&dirty_i->seglist_lock); 930 return NULL_SEGNO; 931 } 932 933 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi, 934 struct block_device *bdev, block_t lstart, 935 block_t start, block_t len) 936 { 937 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 938 struct list_head *pend_list; 939 struct discard_cmd *dc; 940 941 f2fs_bug_on(sbi, !len); 942 943 pend_list = &dcc->pend_list[plist_idx(len)]; 944 945 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL); 946 INIT_LIST_HEAD(&dc->list); 947 dc->bdev = bdev; 948 dc->di.lstart = lstart; 949 dc->di.start = start; 950 dc->di.len = len; 951 dc->ref = 0; 952 dc->state = D_PREP; 953 dc->queued = 0; 954 dc->error = 0; 955 init_completion(&dc->wait); 956 list_add_tail(&dc->list, pend_list); 957 spin_lock_init(&dc->lock); 958 dc->bio_ref = 0; 959 atomic_inc(&dcc->discard_cmd_cnt); 960 dcc->undiscard_blks += len; 961 962 return dc; 963 } 964 965 static bool f2fs_check_discard_tree(struct f2fs_sb_info *sbi) 966 { 967 #ifdef CONFIG_F2FS_CHECK_FS 968 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 969 struct rb_node *cur = rb_first_cached(&dcc->root), *next; 970 struct discard_cmd *cur_dc, *next_dc; 971 972 while (cur) { 973 next = rb_next(cur); 974 if (!next) 975 return true; 976 977 cur_dc = rb_entry(cur, struct discard_cmd, rb_node); 978 next_dc = rb_entry(next, struct discard_cmd, rb_node); 979 980 if (cur_dc->di.lstart + cur_dc->di.len > next_dc->di.lstart) { 981 f2fs_info(sbi, "broken discard_rbtree, " 982 "cur(%u, %u) next(%u, %u)", 983 cur_dc->di.lstart, cur_dc->di.len, 984 next_dc->di.lstart, next_dc->di.len); 985 return false; 986 } 987 cur = next; 988 } 989 #endif 990 return true; 991 } 992 993 static struct discard_cmd *__lookup_discard_cmd(struct f2fs_sb_info *sbi, 994 block_t blkaddr) 995 { 996 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 997 struct rb_node *node = dcc->root.rb_root.rb_node; 998 struct discard_cmd *dc; 999 1000 while (node) { 1001 dc = rb_entry(node, struct discard_cmd, rb_node); 1002 1003 if (blkaddr < dc->di.lstart) 1004 node = node->rb_left; 1005 else if (blkaddr >= dc->di.lstart + dc->di.len) 1006 node = node->rb_right; 1007 else 1008 return dc; 1009 } 1010 return NULL; 1011 } 1012 1013 static struct discard_cmd *__lookup_discard_cmd_ret(struct rb_root_cached *root, 1014 block_t blkaddr, 1015 struct discard_cmd **prev_entry, 1016 struct discard_cmd **next_entry, 1017 struct rb_node ***insert_p, 1018 struct rb_node **insert_parent) 1019 { 1020 struct rb_node **pnode = &root->rb_root.rb_node; 1021 struct rb_node *parent = NULL, *tmp_node; 1022 struct discard_cmd *dc; 1023 1024 *insert_p = NULL; 1025 *insert_parent = NULL; 1026 *prev_entry = NULL; 1027 *next_entry = NULL; 1028 1029 if (RB_EMPTY_ROOT(&root->rb_root)) 1030 return NULL; 1031 1032 while (*pnode) { 1033 parent = *pnode; 1034 dc = rb_entry(*pnode, struct discard_cmd, rb_node); 1035 1036 if (blkaddr < dc->di.lstart) 1037 pnode = &(*pnode)->rb_left; 1038 else if (blkaddr >= dc->di.lstart + dc->di.len) 1039 pnode = &(*pnode)->rb_right; 1040 else 1041 goto lookup_neighbors; 1042 } 1043 1044 *insert_p = pnode; 1045 *insert_parent = parent; 1046 1047 dc = rb_entry(parent, struct discard_cmd, rb_node); 1048 tmp_node = parent; 1049 if (parent && blkaddr > dc->di.lstart) 1050 tmp_node = rb_next(parent); 1051 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1052 1053 tmp_node = parent; 1054 if (parent && blkaddr < dc->di.lstart) 1055 tmp_node = rb_prev(parent); 1056 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1057 return NULL; 1058 1059 lookup_neighbors: 1060 /* lookup prev node for merging backward later */ 1061 tmp_node = rb_prev(&dc->rb_node); 1062 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1063 1064 /* lookup next node for merging frontward later */ 1065 tmp_node = rb_next(&dc->rb_node); 1066 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node); 1067 return dc; 1068 } 1069 1070 static void __detach_discard_cmd(struct discard_cmd_control *dcc, 1071 struct discard_cmd *dc) 1072 { 1073 if (dc->state == D_DONE) 1074 atomic_sub(dc->queued, &dcc->queued_discard); 1075 1076 list_del(&dc->list); 1077 rb_erase_cached(&dc->rb_node, &dcc->root); 1078 dcc->undiscard_blks -= dc->di.len; 1079 1080 kmem_cache_free(discard_cmd_slab, dc); 1081 1082 atomic_dec(&dcc->discard_cmd_cnt); 1083 } 1084 1085 static void __remove_discard_cmd(struct f2fs_sb_info *sbi, 1086 struct discard_cmd *dc) 1087 { 1088 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1089 unsigned long flags; 1090 1091 trace_f2fs_remove_discard(dc->bdev, dc->di.start, dc->di.len); 1092 1093 spin_lock_irqsave(&dc->lock, flags); 1094 if (dc->bio_ref) { 1095 spin_unlock_irqrestore(&dc->lock, flags); 1096 return; 1097 } 1098 spin_unlock_irqrestore(&dc->lock, flags); 1099 1100 f2fs_bug_on(sbi, dc->ref); 1101 1102 if (dc->error == -EOPNOTSUPP) 1103 dc->error = 0; 1104 1105 if (dc->error) 1106 f2fs_info_ratelimited(sbi, 1107 "Issue discard(%u, %u, %u) failed, ret: %d", 1108 dc->di.lstart, dc->di.start, dc->di.len, dc->error); 1109 __detach_discard_cmd(dcc, dc); 1110 } 1111 1112 static void f2fs_submit_discard_endio(struct bio *bio) 1113 { 1114 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private; 1115 unsigned long flags; 1116 1117 spin_lock_irqsave(&dc->lock, flags); 1118 if (!dc->error) 1119 dc->error = blk_status_to_errno(bio->bi_status); 1120 dc->bio_ref--; 1121 if (!dc->bio_ref && dc->state == D_SUBMIT) { 1122 dc->state = D_DONE; 1123 complete_all(&dc->wait); 1124 } 1125 spin_unlock_irqrestore(&dc->lock, flags); 1126 bio_put(bio); 1127 } 1128 1129 static void __check_sit_bitmap(struct f2fs_sb_info *sbi, 1130 block_t start, block_t end) 1131 { 1132 #ifdef CONFIG_F2FS_CHECK_FS 1133 struct seg_entry *sentry; 1134 unsigned int segno; 1135 block_t blk = start; 1136 unsigned long offset, size, *map; 1137 1138 while (blk < end) { 1139 segno = GET_SEGNO(sbi, blk); 1140 sentry = get_seg_entry(sbi, segno); 1141 offset = GET_BLKOFF_FROM_SEG0(sbi, blk); 1142 1143 if (end < START_BLOCK(sbi, segno + 1)) 1144 size = GET_BLKOFF_FROM_SEG0(sbi, end); 1145 else 1146 size = BLKS_PER_SEG(sbi); 1147 map = (unsigned long *)(sentry->cur_valid_map); 1148 offset = __find_rev_next_bit(map, size, offset); 1149 f2fs_bug_on(sbi, offset != size); 1150 blk = START_BLOCK(sbi, segno + 1); 1151 } 1152 #endif 1153 } 1154 1155 static void __init_discard_policy(struct f2fs_sb_info *sbi, 1156 struct discard_policy *dpolicy, 1157 int discard_type, unsigned int granularity) 1158 { 1159 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1160 1161 /* common policy */ 1162 dpolicy->type = discard_type; 1163 dpolicy->sync = true; 1164 dpolicy->ordered = false; 1165 dpolicy->granularity = granularity; 1166 1167 dpolicy->max_requests = dcc->max_discard_request; 1168 dpolicy->io_aware_gran = dcc->discard_io_aware_gran; 1169 dpolicy->timeout = false; 1170 1171 if (discard_type == DPOLICY_BG) { 1172 dpolicy->min_interval = dcc->min_discard_issue_time; 1173 dpolicy->mid_interval = dcc->mid_discard_issue_time; 1174 dpolicy->max_interval = dcc->max_discard_issue_time; 1175 dpolicy->io_aware = true; 1176 dpolicy->sync = false; 1177 dpolicy->ordered = true; 1178 if (utilization(sbi) > dcc->discard_urgent_util) { 1179 dpolicy->granularity = MIN_DISCARD_GRANULARITY; 1180 if (atomic_read(&dcc->discard_cmd_cnt)) 1181 dpolicy->max_interval = 1182 dcc->min_discard_issue_time; 1183 } 1184 } else if (discard_type == DPOLICY_FORCE) { 1185 dpolicy->min_interval = dcc->min_discard_issue_time; 1186 dpolicy->mid_interval = dcc->mid_discard_issue_time; 1187 dpolicy->max_interval = dcc->max_discard_issue_time; 1188 dpolicy->io_aware = false; 1189 } else if (discard_type == DPOLICY_FSTRIM) { 1190 dpolicy->io_aware = false; 1191 } else if (discard_type == DPOLICY_UMOUNT) { 1192 dpolicy->io_aware = false; 1193 /* we need to issue all to keep CP_TRIMMED_FLAG */ 1194 dpolicy->granularity = MIN_DISCARD_GRANULARITY; 1195 dpolicy->timeout = true; 1196 } 1197 } 1198 1199 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1200 struct block_device *bdev, block_t lstart, 1201 block_t start, block_t len); 1202 1203 #ifdef CONFIG_BLK_DEV_ZONED 1204 static void __submit_zone_reset_cmd(struct f2fs_sb_info *sbi, 1205 struct discard_cmd *dc, blk_opf_t flag, 1206 struct list_head *wait_list, 1207 unsigned int *issued) 1208 { 1209 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1210 struct block_device *bdev = dc->bdev; 1211 struct bio *bio = bio_alloc(bdev, 0, REQ_OP_ZONE_RESET | flag, GFP_NOFS); 1212 unsigned long flags; 1213 1214 trace_f2fs_issue_reset_zone(bdev, dc->di.start); 1215 1216 spin_lock_irqsave(&dc->lock, flags); 1217 dc->state = D_SUBMIT; 1218 dc->bio_ref++; 1219 spin_unlock_irqrestore(&dc->lock, flags); 1220 1221 if (issued) 1222 (*issued)++; 1223 1224 atomic_inc(&dcc->queued_discard); 1225 dc->queued++; 1226 list_move_tail(&dc->list, wait_list); 1227 1228 /* sanity check on discard range */ 1229 __check_sit_bitmap(sbi, dc->di.lstart, dc->di.lstart + dc->di.len); 1230 1231 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(dc->di.start); 1232 bio->bi_private = dc; 1233 bio->bi_end_io = f2fs_submit_discard_endio; 1234 submit_bio(bio); 1235 1236 atomic_inc(&dcc->issued_discard); 1237 f2fs_update_iostat(sbi, NULL, FS_ZONE_RESET_IO, dc->di.len * F2FS_BLKSIZE); 1238 } 1239 #endif 1240 1241 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */ 1242 static int __submit_discard_cmd(struct f2fs_sb_info *sbi, 1243 struct discard_policy *dpolicy, 1244 struct discard_cmd *dc, int *issued) 1245 { 1246 struct block_device *bdev = dc->bdev; 1247 unsigned int max_discard_blocks = 1248 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev)); 1249 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1250 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1251 &(dcc->fstrim_list) : &(dcc->wait_list); 1252 blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0; 1253 block_t lstart, start, len, total_len; 1254 int err = 0; 1255 1256 if (dc->state != D_PREP) 1257 return 0; 1258 1259 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 1260 return 0; 1261 1262 #ifdef CONFIG_BLK_DEV_ZONED 1263 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) { 1264 int devi = f2fs_bdev_index(sbi, bdev); 1265 1266 if (devi < 0) 1267 return -EINVAL; 1268 1269 if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) { 1270 __submit_zone_reset_cmd(sbi, dc, flag, 1271 wait_list, issued); 1272 return 0; 1273 } 1274 } 1275 #endif 1276 1277 trace_f2fs_issue_discard(bdev, dc->di.start, dc->di.len); 1278 1279 lstart = dc->di.lstart; 1280 start = dc->di.start; 1281 len = dc->di.len; 1282 total_len = len; 1283 1284 dc->di.len = 0; 1285 1286 while (total_len && *issued < dpolicy->max_requests && !err) { 1287 struct bio *bio = NULL; 1288 unsigned long flags; 1289 bool last = true; 1290 1291 if (len > max_discard_blocks) { 1292 len = max_discard_blocks; 1293 last = false; 1294 } 1295 1296 (*issued)++; 1297 if (*issued == dpolicy->max_requests) 1298 last = true; 1299 1300 dc->di.len += len; 1301 1302 if (time_to_inject(sbi, FAULT_DISCARD)) { 1303 err = -EIO; 1304 } else { 1305 err = __blkdev_issue_discard(bdev, 1306 SECTOR_FROM_BLOCK(start), 1307 SECTOR_FROM_BLOCK(len), 1308 GFP_NOFS, &bio); 1309 } 1310 if (err) { 1311 spin_lock_irqsave(&dc->lock, flags); 1312 if (dc->state == D_PARTIAL) 1313 dc->state = D_SUBMIT; 1314 spin_unlock_irqrestore(&dc->lock, flags); 1315 1316 break; 1317 } 1318 1319 f2fs_bug_on(sbi, !bio); 1320 1321 /* 1322 * should keep before submission to avoid D_DONE 1323 * right away 1324 */ 1325 spin_lock_irqsave(&dc->lock, flags); 1326 if (last) 1327 dc->state = D_SUBMIT; 1328 else 1329 dc->state = D_PARTIAL; 1330 dc->bio_ref++; 1331 spin_unlock_irqrestore(&dc->lock, flags); 1332 1333 atomic_inc(&dcc->queued_discard); 1334 dc->queued++; 1335 list_move_tail(&dc->list, wait_list); 1336 1337 /* sanity check on discard range */ 1338 __check_sit_bitmap(sbi, lstart, lstart + len); 1339 1340 bio->bi_private = dc; 1341 bio->bi_end_io = f2fs_submit_discard_endio; 1342 bio->bi_opf |= flag; 1343 submit_bio(bio); 1344 1345 atomic_inc(&dcc->issued_discard); 1346 1347 f2fs_update_iostat(sbi, NULL, FS_DISCARD_IO, len * F2FS_BLKSIZE); 1348 1349 lstart += len; 1350 start += len; 1351 total_len -= len; 1352 len = total_len; 1353 } 1354 1355 if (!err && len) { 1356 dcc->undiscard_blks -= len; 1357 __update_discard_tree_range(sbi, bdev, lstart, start, len); 1358 } 1359 return err; 1360 } 1361 1362 static void __insert_discard_cmd(struct f2fs_sb_info *sbi, 1363 struct block_device *bdev, block_t lstart, 1364 block_t start, block_t len) 1365 { 1366 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1367 struct rb_node **p = &dcc->root.rb_root.rb_node; 1368 struct rb_node *parent = NULL; 1369 struct discard_cmd *dc; 1370 bool leftmost = true; 1371 1372 /* look up rb tree to find parent node */ 1373 while (*p) { 1374 parent = *p; 1375 dc = rb_entry(parent, struct discard_cmd, rb_node); 1376 1377 if (lstart < dc->di.lstart) { 1378 p = &(*p)->rb_left; 1379 } else if (lstart >= dc->di.lstart + dc->di.len) { 1380 p = &(*p)->rb_right; 1381 leftmost = false; 1382 } else { 1383 f2fs_bug_on(sbi, 1); 1384 } 1385 } 1386 1387 dc = __create_discard_cmd(sbi, bdev, lstart, start, len); 1388 1389 rb_link_node(&dc->rb_node, parent, p); 1390 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost); 1391 } 1392 1393 static void __relocate_discard_cmd(struct discard_cmd_control *dcc, 1394 struct discard_cmd *dc) 1395 { 1396 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->di.len)]); 1397 } 1398 1399 static void __punch_discard_cmd(struct f2fs_sb_info *sbi, 1400 struct discard_cmd *dc, block_t blkaddr) 1401 { 1402 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1403 struct discard_info di = dc->di; 1404 bool modified = false; 1405 1406 if (dc->state == D_DONE || dc->di.len == 1) { 1407 __remove_discard_cmd(sbi, dc); 1408 return; 1409 } 1410 1411 dcc->undiscard_blks -= di.len; 1412 1413 if (blkaddr > di.lstart) { 1414 dc->di.len = blkaddr - dc->di.lstart; 1415 dcc->undiscard_blks += dc->di.len; 1416 __relocate_discard_cmd(dcc, dc); 1417 modified = true; 1418 } 1419 1420 if (blkaddr < di.lstart + di.len - 1) { 1421 if (modified) { 1422 __insert_discard_cmd(sbi, dc->bdev, blkaddr + 1, 1423 di.start + blkaddr + 1 - di.lstart, 1424 di.lstart + di.len - 1 - blkaddr); 1425 } else { 1426 dc->di.lstart++; 1427 dc->di.len--; 1428 dc->di.start++; 1429 dcc->undiscard_blks += dc->di.len; 1430 __relocate_discard_cmd(dcc, dc); 1431 } 1432 } 1433 } 1434 1435 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1436 struct block_device *bdev, block_t lstart, 1437 block_t start, block_t len) 1438 { 1439 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1440 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1441 struct discard_cmd *dc; 1442 struct discard_info di = {0}; 1443 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1444 unsigned int max_discard_blocks = 1445 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev)); 1446 block_t end = lstart + len; 1447 1448 dc = __lookup_discard_cmd_ret(&dcc->root, lstart, 1449 &prev_dc, &next_dc, &insert_p, &insert_parent); 1450 if (dc) 1451 prev_dc = dc; 1452 1453 if (!prev_dc) { 1454 di.lstart = lstart; 1455 di.len = next_dc ? next_dc->di.lstart - lstart : len; 1456 di.len = min(di.len, len); 1457 di.start = start; 1458 } 1459 1460 while (1) { 1461 struct rb_node *node; 1462 bool merged = false; 1463 struct discard_cmd *tdc = NULL; 1464 1465 if (prev_dc) { 1466 di.lstart = prev_dc->di.lstart + prev_dc->di.len; 1467 if (di.lstart < lstart) 1468 di.lstart = lstart; 1469 if (di.lstart >= end) 1470 break; 1471 1472 if (!next_dc || next_dc->di.lstart > end) 1473 di.len = end - di.lstart; 1474 else 1475 di.len = next_dc->di.lstart - di.lstart; 1476 di.start = start + di.lstart - lstart; 1477 } 1478 1479 if (!di.len) 1480 goto next; 1481 1482 if (prev_dc && prev_dc->state == D_PREP && 1483 prev_dc->bdev == bdev && 1484 __is_discard_back_mergeable(&di, &prev_dc->di, 1485 max_discard_blocks)) { 1486 prev_dc->di.len += di.len; 1487 dcc->undiscard_blks += di.len; 1488 __relocate_discard_cmd(dcc, prev_dc); 1489 di = prev_dc->di; 1490 tdc = prev_dc; 1491 merged = true; 1492 } 1493 1494 if (next_dc && next_dc->state == D_PREP && 1495 next_dc->bdev == bdev && 1496 __is_discard_front_mergeable(&di, &next_dc->di, 1497 max_discard_blocks)) { 1498 next_dc->di.lstart = di.lstart; 1499 next_dc->di.len += di.len; 1500 next_dc->di.start = di.start; 1501 dcc->undiscard_blks += di.len; 1502 __relocate_discard_cmd(dcc, next_dc); 1503 if (tdc) 1504 __remove_discard_cmd(sbi, tdc); 1505 merged = true; 1506 } 1507 1508 if (!merged) 1509 __insert_discard_cmd(sbi, bdev, 1510 di.lstart, di.start, di.len); 1511 next: 1512 prev_dc = next_dc; 1513 if (!prev_dc) 1514 break; 1515 1516 node = rb_next(&prev_dc->rb_node); 1517 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1518 } 1519 } 1520 1521 #ifdef CONFIG_BLK_DEV_ZONED 1522 static void __queue_zone_reset_cmd(struct f2fs_sb_info *sbi, 1523 struct block_device *bdev, block_t blkstart, block_t lblkstart, 1524 block_t blklen) 1525 { 1526 trace_f2fs_queue_reset_zone(bdev, blkstart); 1527 1528 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock); 1529 __insert_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen); 1530 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock); 1531 } 1532 #endif 1533 1534 static void __queue_discard_cmd(struct f2fs_sb_info *sbi, 1535 struct block_device *bdev, block_t blkstart, block_t blklen) 1536 { 1537 block_t lblkstart = blkstart; 1538 1539 if (!f2fs_bdev_support_discard(bdev)) 1540 return; 1541 1542 trace_f2fs_queue_discard(bdev, blkstart, blklen); 1543 1544 if (f2fs_is_multi_device(sbi)) { 1545 int devi = f2fs_target_device_index(sbi, blkstart); 1546 1547 blkstart -= FDEV(devi).start_blk; 1548 } 1549 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock); 1550 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen); 1551 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock); 1552 } 1553 1554 static void __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi, 1555 struct discard_policy *dpolicy, int *issued) 1556 { 1557 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1558 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1559 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1560 struct discard_cmd *dc; 1561 struct blk_plug plug; 1562 bool io_interrupted = false; 1563 1564 mutex_lock(&dcc->cmd_lock); 1565 dc = __lookup_discard_cmd_ret(&dcc->root, dcc->next_pos, 1566 &prev_dc, &next_dc, &insert_p, &insert_parent); 1567 if (!dc) 1568 dc = next_dc; 1569 1570 blk_start_plug(&plug); 1571 1572 while (dc) { 1573 struct rb_node *node; 1574 int err = 0; 1575 1576 if (dc->state != D_PREP) 1577 goto next; 1578 1579 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) { 1580 io_interrupted = true; 1581 break; 1582 } 1583 1584 dcc->next_pos = dc->di.lstart + dc->di.len; 1585 err = __submit_discard_cmd(sbi, dpolicy, dc, issued); 1586 1587 if (*issued >= dpolicy->max_requests) 1588 break; 1589 next: 1590 node = rb_next(&dc->rb_node); 1591 if (err) 1592 __remove_discard_cmd(sbi, dc); 1593 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1594 } 1595 1596 blk_finish_plug(&plug); 1597 1598 if (!dc) 1599 dcc->next_pos = 0; 1600 1601 mutex_unlock(&dcc->cmd_lock); 1602 1603 if (!(*issued) && io_interrupted) 1604 *issued = -1; 1605 } 1606 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1607 struct discard_policy *dpolicy); 1608 1609 static int __issue_discard_cmd(struct f2fs_sb_info *sbi, 1610 struct discard_policy *dpolicy) 1611 { 1612 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1613 struct list_head *pend_list; 1614 struct discard_cmd *dc, *tmp; 1615 struct blk_plug plug; 1616 int i, issued; 1617 bool io_interrupted = false; 1618 1619 if (dpolicy->timeout) 1620 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT); 1621 1622 retry: 1623 issued = 0; 1624 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1625 if (dpolicy->timeout && 1626 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1627 break; 1628 1629 if (i + 1 < dpolicy->granularity) 1630 break; 1631 1632 if (i + 1 < dcc->max_ordered_discard && dpolicy->ordered) { 1633 __issue_discard_cmd_orderly(sbi, dpolicy, &issued); 1634 return issued; 1635 } 1636 1637 pend_list = &dcc->pend_list[i]; 1638 1639 mutex_lock(&dcc->cmd_lock); 1640 if (list_empty(pend_list)) 1641 goto next; 1642 if (unlikely(dcc->rbtree_check)) 1643 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi)); 1644 blk_start_plug(&plug); 1645 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1646 f2fs_bug_on(sbi, dc->state != D_PREP); 1647 1648 if (dpolicy->timeout && 1649 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1650 break; 1651 1652 if (dpolicy->io_aware && i < dpolicy->io_aware_gran && 1653 !is_idle(sbi, DISCARD_TIME)) { 1654 io_interrupted = true; 1655 break; 1656 } 1657 1658 __submit_discard_cmd(sbi, dpolicy, dc, &issued); 1659 1660 if (issued >= dpolicy->max_requests) 1661 break; 1662 } 1663 blk_finish_plug(&plug); 1664 next: 1665 mutex_unlock(&dcc->cmd_lock); 1666 1667 if (issued >= dpolicy->max_requests || io_interrupted) 1668 break; 1669 } 1670 1671 if (dpolicy->type == DPOLICY_UMOUNT && issued) { 1672 __wait_all_discard_cmd(sbi, dpolicy); 1673 goto retry; 1674 } 1675 1676 if (!issued && io_interrupted) 1677 issued = -1; 1678 1679 return issued; 1680 } 1681 1682 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi) 1683 { 1684 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1685 struct list_head *pend_list; 1686 struct discard_cmd *dc, *tmp; 1687 int i; 1688 bool dropped = false; 1689 1690 mutex_lock(&dcc->cmd_lock); 1691 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1692 pend_list = &dcc->pend_list[i]; 1693 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1694 f2fs_bug_on(sbi, dc->state != D_PREP); 1695 __remove_discard_cmd(sbi, dc); 1696 dropped = true; 1697 } 1698 } 1699 mutex_unlock(&dcc->cmd_lock); 1700 1701 return dropped; 1702 } 1703 1704 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi) 1705 { 1706 __drop_discard_cmd(sbi); 1707 } 1708 1709 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi, 1710 struct discard_cmd *dc) 1711 { 1712 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1713 unsigned int len = 0; 1714 1715 wait_for_completion_io(&dc->wait); 1716 mutex_lock(&dcc->cmd_lock); 1717 f2fs_bug_on(sbi, dc->state != D_DONE); 1718 dc->ref--; 1719 if (!dc->ref) { 1720 if (!dc->error) 1721 len = dc->di.len; 1722 __remove_discard_cmd(sbi, dc); 1723 } 1724 mutex_unlock(&dcc->cmd_lock); 1725 1726 return len; 1727 } 1728 1729 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi, 1730 struct discard_policy *dpolicy, 1731 block_t start, block_t end) 1732 { 1733 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1734 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1735 &(dcc->fstrim_list) : &(dcc->wait_list); 1736 struct discard_cmd *dc = NULL, *iter, *tmp; 1737 unsigned int trimmed = 0; 1738 1739 next: 1740 dc = NULL; 1741 1742 mutex_lock(&dcc->cmd_lock); 1743 list_for_each_entry_safe(iter, tmp, wait_list, list) { 1744 if (iter->di.lstart + iter->di.len <= start || 1745 end <= iter->di.lstart) 1746 continue; 1747 if (iter->di.len < dpolicy->granularity) 1748 continue; 1749 if (iter->state == D_DONE && !iter->ref) { 1750 wait_for_completion_io(&iter->wait); 1751 if (!iter->error) 1752 trimmed += iter->di.len; 1753 __remove_discard_cmd(sbi, iter); 1754 } else { 1755 iter->ref++; 1756 dc = iter; 1757 break; 1758 } 1759 } 1760 mutex_unlock(&dcc->cmd_lock); 1761 1762 if (dc) { 1763 trimmed += __wait_one_discard_bio(sbi, dc); 1764 goto next; 1765 } 1766 1767 return trimmed; 1768 } 1769 1770 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1771 struct discard_policy *dpolicy) 1772 { 1773 struct discard_policy dp; 1774 unsigned int discard_blks; 1775 1776 if (dpolicy) 1777 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX); 1778 1779 /* wait all */ 1780 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, MIN_DISCARD_GRANULARITY); 1781 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1782 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, MIN_DISCARD_GRANULARITY); 1783 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1784 1785 return discard_blks; 1786 } 1787 1788 /* This should be covered by global mutex, &sit_i->sentry_lock */ 1789 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr) 1790 { 1791 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1792 struct discard_cmd *dc; 1793 bool need_wait = false; 1794 1795 mutex_lock(&dcc->cmd_lock); 1796 dc = __lookup_discard_cmd(sbi, blkaddr); 1797 #ifdef CONFIG_BLK_DEV_ZONED 1798 if (dc && f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(dc->bdev)) { 1799 int devi = f2fs_bdev_index(sbi, dc->bdev); 1800 1801 if (devi < 0) { 1802 mutex_unlock(&dcc->cmd_lock); 1803 return; 1804 } 1805 1806 if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) { 1807 /* force submit zone reset */ 1808 if (dc->state == D_PREP) 1809 __submit_zone_reset_cmd(sbi, dc, REQ_SYNC, 1810 &dcc->wait_list, NULL); 1811 dc->ref++; 1812 mutex_unlock(&dcc->cmd_lock); 1813 /* wait zone reset */ 1814 __wait_one_discard_bio(sbi, dc); 1815 return; 1816 } 1817 } 1818 #endif 1819 if (dc) { 1820 if (dc->state == D_PREP) { 1821 __punch_discard_cmd(sbi, dc, blkaddr); 1822 } else { 1823 dc->ref++; 1824 need_wait = true; 1825 } 1826 } 1827 mutex_unlock(&dcc->cmd_lock); 1828 1829 if (need_wait) 1830 __wait_one_discard_bio(sbi, dc); 1831 } 1832 1833 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi) 1834 { 1835 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1836 1837 if (dcc && dcc->f2fs_issue_discard) { 1838 struct task_struct *discard_thread = dcc->f2fs_issue_discard; 1839 1840 dcc->f2fs_issue_discard = NULL; 1841 kthread_stop(discard_thread); 1842 } 1843 } 1844 1845 /** 1846 * f2fs_issue_discard_timeout() - Issue all discard cmd within UMOUNT_DISCARD_TIMEOUT 1847 * @sbi: the f2fs_sb_info data for discard cmd to issue 1848 * 1849 * When UMOUNT_DISCARD_TIMEOUT is exceeded, all remaining discard commands will be dropped 1850 * 1851 * Return true if issued all discard cmd or no discard cmd need issue, otherwise return false. 1852 */ 1853 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi) 1854 { 1855 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1856 struct discard_policy dpolicy; 1857 bool dropped; 1858 1859 if (!atomic_read(&dcc->discard_cmd_cnt)) 1860 return true; 1861 1862 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT, 1863 dcc->discard_granularity); 1864 __issue_discard_cmd(sbi, &dpolicy); 1865 dropped = __drop_discard_cmd(sbi); 1866 1867 /* just to make sure there is no pending discard commands */ 1868 __wait_all_discard_cmd(sbi, NULL); 1869 1870 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt)); 1871 return !dropped; 1872 } 1873 1874 static int issue_discard_thread(void *data) 1875 { 1876 struct f2fs_sb_info *sbi = data; 1877 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1878 wait_queue_head_t *q = &dcc->discard_wait_queue; 1879 struct discard_policy dpolicy; 1880 unsigned int wait_ms = dcc->min_discard_issue_time; 1881 int issued; 1882 1883 set_freezable(); 1884 1885 do { 1886 wait_event_interruptible_timeout(*q, 1887 kthread_should_stop() || freezing(current) || 1888 dcc->discard_wake, 1889 msecs_to_jiffies(wait_ms)); 1890 1891 if (sbi->gc_mode == GC_URGENT_HIGH || 1892 !f2fs_available_free_memory(sbi, DISCARD_CACHE)) 1893 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1894 MIN_DISCARD_GRANULARITY); 1895 else 1896 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG, 1897 dcc->discard_granularity); 1898 1899 if (dcc->discard_wake) 1900 dcc->discard_wake = false; 1901 1902 /* clean up pending candidates before going to sleep */ 1903 if (atomic_read(&dcc->queued_discard)) 1904 __wait_all_discard_cmd(sbi, NULL); 1905 1906 if (try_to_freeze()) 1907 continue; 1908 if (f2fs_readonly(sbi->sb)) 1909 continue; 1910 if (kthread_should_stop()) 1911 return 0; 1912 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) || 1913 !atomic_read(&dcc->discard_cmd_cnt)) { 1914 wait_ms = dpolicy.max_interval; 1915 continue; 1916 } 1917 1918 sb_start_intwrite(sbi->sb); 1919 1920 issued = __issue_discard_cmd(sbi, &dpolicy); 1921 if (issued > 0) { 1922 __wait_all_discard_cmd(sbi, &dpolicy); 1923 wait_ms = dpolicy.min_interval; 1924 } else if (issued == -1) { 1925 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME); 1926 if (!wait_ms) 1927 wait_ms = dpolicy.mid_interval; 1928 } else { 1929 wait_ms = dpolicy.max_interval; 1930 } 1931 if (!atomic_read(&dcc->discard_cmd_cnt)) 1932 wait_ms = dpolicy.max_interval; 1933 1934 sb_end_intwrite(sbi->sb); 1935 1936 } while (!kthread_should_stop()); 1937 return 0; 1938 } 1939 1940 #ifdef CONFIG_BLK_DEV_ZONED 1941 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi, 1942 struct block_device *bdev, block_t blkstart, block_t blklen) 1943 { 1944 sector_t sector, nr_sects; 1945 block_t lblkstart = blkstart; 1946 int devi = 0; 1947 u64 remainder = 0; 1948 1949 if (f2fs_is_multi_device(sbi)) { 1950 devi = f2fs_target_device_index(sbi, blkstart); 1951 if (blkstart < FDEV(devi).start_blk || 1952 blkstart > FDEV(devi).end_blk) { 1953 f2fs_err(sbi, "Invalid block %x", blkstart); 1954 return -EIO; 1955 } 1956 blkstart -= FDEV(devi).start_blk; 1957 } 1958 1959 /* For sequential zones, reset the zone write pointer */ 1960 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) { 1961 sector = SECTOR_FROM_BLOCK(blkstart); 1962 nr_sects = SECTOR_FROM_BLOCK(blklen); 1963 div64_u64_rem(sector, bdev_zone_sectors(bdev), &remainder); 1964 1965 if (remainder || nr_sects != bdev_zone_sectors(bdev)) { 1966 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)", 1967 devi, sbi->s_ndevs ? FDEV(devi).path : "", 1968 blkstart, blklen); 1969 return -EIO; 1970 } 1971 1972 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) { 1973 trace_f2fs_issue_reset_zone(bdev, blkstart); 1974 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, 1975 sector, nr_sects, GFP_NOFS); 1976 } 1977 1978 __queue_zone_reset_cmd(sbi, bdev, blkstart, lblkstart, blklen); 1979 return 0; 1980 } 1981 1982 /* For conventional zones, use regular discard if supported */ 1983 __queue_discard_cmd(sbi, bdev, lblkstart, blklen); 1984 return 0; 1985 } 1986 #endif 1987 1988 static int __issue_discard_async(struct f2fs_sb_info *sbi, 1989 struct block_device *bdev, block_t blkstart, block_t blklen) 1990 { 1991 #ifdef CONFIG_BLK_DEV_ZONED 1992 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) 1993 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen); 1994 #endif 1995 __queue_discard_cmd(sbi, bdev, blkstart, blklen); 1996 return 0; 1997 } 1998 1999 static int f2fs_issue_discard(struct f2fs_sb_info *sbi, 2000 block_t blkstart, block_t blklen) 2001 { 2002 sector_t start = blkstart, len = 0; 2003 struct block_device *bdev; 2004 struct seg_entry *se; 2005 unsigned int offset; 2006 block_t i; 2007 int err = 0; 2008 2009 bdev = f2fs_target_device(sbi, blkstart, NULL); 2010 2011 for (i = blkstart; i < blkstart + blklen; i++, len++) { 2012 if (i != start) { 2013 struct block_device *bdev2 = 2014 f2fs_target_device(sbi, i, NULL); 2015 2016 if (bdev2 != bdev) { 2017 err = __issue_discard_async(sbi, bdev, 2018 start, len); 2019 if (err) 2020 return err; 2021 bdev = bdev2; 2022 start = i; 2023 len = 0; 2024 } 2025 } 2026 2027 se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); 2028 offset = GET_BLKOFF_FROM_SEG0(sbi, i); 2029 2030 if (f2fs_block_unit_discard(sbi) && 2031 !f2fs_test_and_set_bit(offset, se->discard_map)) 2032 sbi->discard_blks--; 2033 } 2034 2035 if (len) 2036 err = __issue_discard_async(sbi, bdev, start, len); 2037 return err; 2038 } 2039 2040 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc, 2041 bool check_only) 2042 { 2043 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 2044 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); 2045 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 2046 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 2047 unsigned long *discard_map = (unsigned long *)se->discard_map; 2048 unsigned long *dmap = SIT_I(sbi)->tmp_map; 2049 unsigned int start = 0, end = -1; 2050 bool force = (cpc->reason & CP_DISCARD); 2051 struct discard_entry *de = NULL; 2052 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list; 2053 int i; 2054 2055 if (se->valid_blocks == BLKS_PER_SEG(sbi) || 2056 !f2fs_hw_support_discard(sbi) || 2057 !f2fs_block_unit_discard(sbi)) 2058 return false; 2059 2060 if (!force) { 2061 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks || 2062 SM_I(sbi)->dcc_info->nr_discards >= 2063 SM_I(sbi)->dcc_info->max_discards) 2064 return false; 2065 } 2066 2067 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ 2068 for (i = 0; i < entries; i++) 2069 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : 2070 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; 2071 2072 while (force || SM_I(sbi)->dcc_info->nr_discards <= 2073 SM_I(sbi)->dcc_info->max_discards) { 2074 start = __find_rev_next_bit(dmap, BLKS_PER_SEG(sbi), end + 1); 2075 if (start >= BLKS_PER_SEG(sbi)) 2076 break; 2077 2078 end = __find_rev_next_zero_bit(dmap, 2079 BLKS_PER_SEG(sbi), start + 1); 2080 if (force && start && end != BLKS_PER_SEG(sbi) && 2081 (end - start) < cpc->trim_minlen) 2082 continue; 2083 2084 if (check_only) 2085 return true; 2086 2087 if (!de) { 2088 de = f2fs_kmem_cache_alloc(discard_entry_slab, 2089 GFP_F2FS_ZERO, true, NULL); 2090 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start); 2091 list_add_tail(&de->list, head); 2092 } 2093 2094 for (i = start; i < end; i++) 2095 __set_bit_le(i, (void *)de->discard_map); 2096 2097 SM_I(sbi)->dcc_info->nr_discards += end - start; 2098 } 2099 return false; 2100 } 2101 2102 static void release_discard_addr(struct discard_entry *entry) 2103 { 2104 list_del(&entry->list); 2105 kmem_cache_free(discard_entry_slab, entry); 2106 } 2107 2108 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi) 2109 { 2110 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); 2111 struct discard_entry *entry, *this; 2112 2113 /* drop caches */ 2114 list_for_each_entry_safe(entry, this, head, list) 2115 release_discard_addr(entry); 2116 } 2117 2118 /* 2119 * Should call f2fs_clear_prefree_segments after checkpoint is done. 2120 */ 2121 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 2122 { 2123 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2124 unsigned int segno; 2125 2126 mutex_lock(&dirty_i->seglist_lock); 2127 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) 2128 __set_test_and_free(sbi, segno, false); 2129 mutex_unlock(&dirty_i->seglist_lock); 2130 } 2131 2132 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi, 2133 struct cp_control *cpc) 2134 { 2135 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2136 struct list_head *head = &dcc->entry_list; 2137 struct discard_entry *entry, *this; 2138 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2139 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; 2140 unsigned int start = 0, end = -1; 2141 unsigned int secno, start_segno; 2142 bool force = (cpc->reason & CP_DISCARD); 2143 bool section_alignment = F2FS_OPTION(sbi).discard_unit == 2144 DISCARD_UNIT_SECTION; 2145 2146 if (f2fs_lfs_mode(sbi) && __is_large_section(sbi)) 2147 section_alignment = true; 2148 2149 mutex_lock(&dirty_i->seglist_lock); 2150 2151 while (1) { 2152 int i; 2153 2154 if (section_alignment && end != -1) 2155 end--; 2156 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); 2157 if (start >= MAIN_SEGS(sbi)) 2158 break; 2159 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), 2160 start + 1); 2161 2162 if (section_alignment) { 2163 start = rounddown(start, SEGS_PER_SEC(sbi)); 2164 end = roundup(end, SEGS_PER_SEC(sbi)); 2165 } 2166 2167 for (i = start; i < end; i++) { 2168 if (test_and_clear_bit(i, prefree_map)) 2169 dirty_i->nr_dirty[PRE]--; 2170 } 2171 2172 if (!f2fs_realtime_discard_enable(sbi)) 2173 continue; 2174 2175 if (force && start >= cpc->trim_start && 2176 (end - 1) <= cpc->trim_end) 2177 continue; 2178 2179 /* Should cover 2MB zoned device for zone-based reset */ 2180 if (!f2fs_sb_has_blkzoned(sbi) && 2181 (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi))) { 2182 f2fs_issue_discard(sbi, START_BLOCK(sbi, start), 2183 (end - start) << sbi->log_blocks_per_seg); 2184 continue; 2185 } 2186 next: 2187 secno = GET_SEC_FROM_SEG(sbi, start); 2188 start_segno = GET_SEG_FROM_SEC(sbi, secno); 2189 if (!IS_CURSEC(sbi, secno) && 2190 !get_valid_blocks(sbi, start, true)) 2191 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno), 2192 BLKS_PER_SEC(sbi)); 2193 2194 start = start_segno + SEGS_PER_SEC(sbi); 2195 if (start < end) 2196 goto next; 2197 else 2198 end = start - 1; 2199 } 2200 mutex_unlock(&dirty_i->seglist_lock); 2201 2202 if (!f2fs_block_unit_discard(sbi)) 2203 goto wakeup; 2204 2205 /* send small discards */ 2206 list_for_each_entry_safe(entry, this, head, list) { 2207 unsigned int cur_pos = 0, next_pos, len, total_len = 0; 2208 bool is_valid = test_bit_le(0, entry->discard_map); 2209 2210 find_next: 2211 if (is_valid) { 2212 next_pos = find_next_zero_bit_le(entry->discard_map, 2213 BLKS_PER_SEG(sbi), cur_pos); 2214 len = next_pos - cur_pos; 2215 2216 if (f2fs_sb_has_blkzoned(sbi) || 2217 (force && len < cpc->trim_minlen)) 2218 goto skip; 2219 2220 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos, 2221 len); 2222 total_len += len; 2223 } else { 2224 next_pos = find_next_bit_le(entry->discard_map, 2225 BLKS_PER_SEG(sbi), cur_pos); 2226 } 2227 skip: 2228 cur_pos = next_pos; 2229 is_valid = !is_valid; 2230 2231 if (cur_pos < BLKS_PER_SEG(sbi)) 2232 goto find_next; 2233 2234 release_discard_addr(entry); 2235 dcc->nr_discards -= total_len; 2236 } 2237 2238 wakeup: 2239 wake_up_discard_thread(sbi, false); 2240 } 2241 2242 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi) 2243 { 2244 dev_t dev = sbi->sb->s_bdev->bd_dev; 2245 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2246 int err = 0; 2247 2248 if (!f2fs_realtime_discard_enable(sbi)) 2249 return 0; 2250 2251 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi, 2252 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev)); 2253 if (IS_ERR(dcc->f2fs_issue_discard)) { 2254 err = PTR_ERR(dcc->f2fs_issue_discard); 2255 dcc->f2fs_issue_discard = NULL; 2256 } 2257 2258 return err; 2259 } 2260 2261 static int create_discard_cmd_control(struct f2fs_sb_info *sbi) 2262 { 2263 struct discard_cmd_control *dcc; 2264 int err = 0, i; 2265 2266 if (SM_I(sbi)->dcc_info) { 2267 dcc = SM_I(sbi)->dcc_info; 2268 goto init_thread; 2269 } 2270 2271 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL); 2272 if (!dcc) 2273 return -ENOMEM; 2274 2275 dcc->discard_io_aware_gran = MAX_PLIST_NUM; 2276 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY; 2277 dcc->max_ordered_discard = DEFAULT_MAX_ORDERED_DISCARD_GRANULARITY; 2278 if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT) 2279 dcc->discard_granularity = BLKS_PER_SEG(sbi); 2280 else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION) 2281 dcc->discard_granularity = BLKS_PER_SEC(sbi); 2282 2283 INIT_LIST_HEAD(&dcc->entry_list); 2284 for (i = 0; i < MAX_PLIST_NUM; i++) 2285 INIT_LIST_HEAD(&dcc->pend_list[i]); 2286 INIT_LIST_HEAD(&dcc->wait_list); 2287 INIT_LIST_HEAD(&dcc->fstrim_list); 2288 mutex_init(&dcc->cmd_lock); 2289 atomic_set(&dcc->issued_discard, 0); 2290 atomic_set(&dcc->queued_discard, 0); 2291 atomic_set(&dcc->discard_cmd_cnt, 0); 2292 dcc->nr_discards = 0; 2293 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg; 2294 dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST; 2295 dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME; 2296 dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME; 2297 dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME; 2298 dcc->discard_urgent_util = DEF_DISCARD_URGENT_UTIL; 2299 dcc->undiscard_blks = 0; 2300 dcc->next_pos = 0; 2301 dcc->root = RB_ROOT_CACHED; 2302 dcc->rbtree_check = false; 2303 2304 init_waitqueue_head(&dcc->discard_wait_queue); 2305 SM_I(sbi)->dcc_info = dcc; 2306 init_thread: 2307 err = f2fs_start_discard_thread(sbi); 2308 if (err) { 2309 kfree(dcc); 2310 SM_I(sbi)->dcc_info = NULL; 2311 } 2312 2313 return err; 2314 } 2315 2316 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi) 2317 { 2318 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2319 2320 if (!dcc) 2321 return; 2322 2323 f2fs_stop_discard_thread(sbi); 2324 2325 /* 2326 * Recovery can cache discard commands, so in error path of 2327 * fill_super(), it needs to give a chance to handle them. 2328 */ 2329 f2fs_issue_discard_timeout(sbi); 2330 2331 kfree(dcc); 2332 SM_I(sbi)->dcc_info = NULL; 2333 } 2334 2335 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 2336 { 2337 struct sit_info *sit_i = SIT_I(sbi); 2338 2339 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { 2340 sit_i->dirty_sentries++; 2341 return false; 2342 } 2343 2344 return true; 2345 } 2346 2347 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 2348 unsigned int segno, int modified) 2349 { 2350 struct seg_entry *se = get_seg_entry(sbi, segno); 2351 2352 se->type = type; 2353 if (modified) 2354 __mark_sit_entry_dirty(sbi, segno); 2355 } 2356 2357 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi, 2358 block_t blkaddr) 2359 { 2360 unsigned int segno = GET_SEGNO(sbi, blkaddr); 2361 2362 if (segno == NULL_SEGNO) 2363 return 0; 2364 return get_seg_entry(sbi, segno)->mtime; 2365 } 2366 2367 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr, 2368 unsigned long long old_mtime) 2369 { 2370 struct seg_entry *se; 2371 unsigned int segno = GET_SEGNO(sbi, blkaddr); 2372 unsigned long long ctime = get_mtime(sbi, false); 2373 unsigned long long mtime = old_mtime ? old_mtime : ctime; 2374 2375 if (segno == NULL_SEGNO) 2376 return; 2377 2378 se = get_seg_entry(sbi, segno); 2379 2380 if (!se->mtime) 2381 se->mtime = mtime; 2382 else 2383 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime, 2384 se->valid_blocks + 1); 2385 2386 if (ctime > SIT_I(sbi)->max_mtime) 2387 SIT_I(sbi)->max_mtime = ctime; 2388 } 2389 2390 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 2391 { 2392 struct seg_entry *se; 2393 unsigned int segno, offset; 2394 long int new_vblocks; 2395 bool exist; 2396 #ifdef CONFIG_F2FS_CHECK_FS 2397 bool mir_exist; 2398 #endif 2399 2400 segno = GET_SEGNO(sbi, blkaddr); 2401 if (segno == NULL_SEGNO) 2402 return; 2403 2404 se = get_seg_entry(sbi, segno); 2405 new_vblocks = se->valid_blocks + del; 2406 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2407 2408 f2fs_bug_on(sbi, (new_vblocks < 0 || 2409 (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno)))); 2410 2411 se->valid_blocks = new_vblocks; 2412 2413 /* Update valid block bitmap */ 2414 if (del > 0) { 2415 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map); 2416 #ifdef CONFIG_F2FS_CHECK_FS 2417 mir_exist = f2fs_test_and_set_bit(offset, 2418 se->cur_valid_map_mir); 2419 if (unlikely(exist != mir_exist)) { 2420 f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d", 2421 blkaddr, exist); 2422 f2fs_bug_on(sbi, 1); 2423 } 2424 #endif 2425 if (unlikely(exist)) { 2426 f2fs_err(sbi, "Bitmap was wrongly set, blk:%u", 2427 blkaddr); 2428 f2fs_bug_on(sbi, 1); 2429 se->valid_blocks--; 2430 del = 0; 2431 } 2432 2433 if (f2fs_block_unit_discard(sbi) && 2434 !f2fs_test_and_set_bit(offset, se->discard_map)) 2435 sbi->discard_blks--; 2436 2437 /* 2438 * SSR should never reuse block which is checkpointed 2439 * or newly invalidated. 2440 */ 2441 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) { 2442 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map)) 2443 se->ckpt_valid_blocks++; 2444 } 2445 } else { 2446 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map); 2447 #ifdef CONFIG_F2FS_CHECK_FS 2448 mir_exist = f2fs_test_and_clear_bit(offset, 2449 se->cur_valid_map_mir); 2450 if (unlikely(exist != mir_exist)) { 2451 f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d", 2452 blkaddr, exist); 2453 f2fs_bug_on(sbi, 1); 2454 } 2455 #endif 2456 if (unlikely(!exist)) { 2457 f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u", 2458 blkaddr); 2459 f2fs_bug_on(sbi, 1); 2460 se->valid_blocks++; 2461 del = 0; 2462 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 2463 /* 2464 * If checkpoints are off, we must not reuse data that 2465 * was used in the previous checkpoint. If it was used 2466 * before, we must track that to know how much space we 2467 * really have. 2468 */ 2469 if (f2fs_test_bit(offset, se->ckpt_valid_map)) { 2470 spin_lock(&sbi->stat_lock); 2471 sbi->unusable_block_count++; 2472 spin_unlock(&sbi->stat_lock); 2473 } 2474 } 2475 2476 if (f2fs_block_unit_discard(sbi) && 2477 f2fs_test_and_clear_bit(offset, se->discard_map)) 2478 sbi->discard_blks++; 2479 } 2480 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) 2481 se->ckpt_valid_blocks += del; 2482 2483 __mark_sit_entry_dirty(sbi, segno); 2484 2485 /* update total number of valid blocks to be written in ckpt area */ 2486 SIT_I(sbi)->written_valid_blocks += del; 2487 2488 if (__is_large_section(sbi)) 2489 get_sec_entry(sbi, segno)->valid_blocks += del; 2490 } 2491 2492 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) 2493 { 2494 unsigned int segno = GET_SEGNO(sbi, addr); 2495 struct sit_info *sit_i = SIT_I(sbi); 2496 2497 f2fs_bug_on(sbi, addr == NULL_ADDR); 2498 if (addr == NEW_ADDR || addr == COMPRESS_ADDR) 2499 return; 2500 2501 f2fs_invalidate_internal_cache(sbi, addr); 2502 2503 /* add it into sit main buffer */ 2504 down_write(&sit_i->sentry_lock); 2505 2506 update_segment_mtime(sbi, addr, 0); 2507 update_sit_entry(sbi, addr, -1); 2508 2509 /* add it into dirty seglist */ 2510 locate_dirty_segment(sbi, segno); 2511 2512 up_write(&sit_i->sentry_lock); 2513 } 2514 2515 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr) 2516 { 2517 struct sit_info *sit_i = SIT_I(sbi); 2518 unsigned int segno, offset; 2519 struct seg_entry *se; 2520 bool is_cp = false; 2521 2522 if (!__is_valid_data_blkaddr(blkaddr)) 2523 return true; 2524 2525 down_read(&sit_i->sentry_lock); 2526 2527 segno = GET_SEGNO(sbi, blkaddr); 2528 se = get_seg_entry(sbi, segno); 2529 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2530 2531 if (f2fs_test_bit(offset, se->ckpt_valid_map)) 2532 is_cp = true; 2533 2534 up_read(&sit_i->sentry_lock); 2535 2536 return is_cp; 2537 } 2538 2539 static unsigned short f2fs_curseg_valid_blocks(struct f2fs_sb_info *sbi, int type) 2540 { 2541 struct curseg_info *curseg = CURSEG_I(sbi, type); 2542 2543 if (sbi->ckpt->alloc_type[type] == SSR) 2544 return BLKS_PER_SEG(sbi); 2545 return curseg->next_blkoff; 2546 } 2547 2548 /* 2549 * Calculate the number of current summary pages for writing 2550 */ 2551 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) 2552 { 2553 int valid_sum_count = 0; 2554 int i, sum_in_page; 2555 2556 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2557 if (sbi->ckpt->alloc_type[i] != SSR && for_ra) 2558 valid_sum_count += 2559 le16_to_cpu(F2FS_CKPT(sbi)->cur_data_blkoff[i]); 2560 else 2561 valid_sum_count += f2fs_curseg_valid_blocks(sbi, i); 2562 } 2563 2564 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE - 2565 SUM_FOOTER_SIZE) / SUMMARY_SIZE; 2566 if (valid_sum_count <= sum_in_page) 2567 return 1; 2568 else if ((valid_sum_count - sum_in_page) <= 2569 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) 2570 return 2; 2571 return 3; 2572 } 2573 2574 /* 2575 * Caller should put this summary page 2576 */ 2577 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) 2578 { 2579 if (unlikely(f2fs_cp_error(sbi))) 2580 return ERR_PTR(-EIO); 2581 return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno)); 2582 } 2583 2584 void f2fs_update_meta_page(struct f2fs_sb_info *sbi, 2585 void *src, block_t blk_addr) 2586 { 2587 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2588 2589 memcpy(page_address(page), src, PAGE_SIZE); 2590 set_page_dirty(page); 2591 f2fs_put_page(page, 1); 2592 } 2593 2594 static void write_sum_page(struct f2fs_sb_info *sbi, 2595 struct f2fs_summary_block *sum_blk, block_t blk_addr) 2596 { 2597 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr); 2598 } 2599 2600 static void write_current_sum_page(struct f2fs_sb_info *sbi, 2601 int type, block_t blk_addr) 2602 { 2603 struct curseg_info *curseg = CURSEG_I(sbi, type); 2604 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2605 struct f2fs_summary_block *src = curseg->sum_blk; 2606 struct f2fs_summary_block *dst; 2607 2608 dst = (struct f2fs_summary_block *)page_address(page); 2609 memset(dst, 0, PAGE_SIZE); 2610 2611 mutex_lock(&curseg->curseg_mutex); 2612 2613 down_read(&curseg->journal_rwsem); 2614 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE); 2615 up_read(&curseg->journal_rwsem); 2616 2617 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE); 2618 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE); 2619 2620 mutex_unlock(&curseg->curseg_mutex); 2621 2622 set_page_dirty(page); 2623 f2fs_put_page(page, 1); 2624 } 2625 2626 static int is_next_segment_free(struct f2fs_sb_info *sbi, 2627 struct curseg_info *curseg, int type) 2628 { 2629 unsigned int segno = curseg->segno + 1; 2630 struct free_segmap_info *free_i = FREE_I(sbi); 2631 2632 if (segno < MAIN_SEGS(sbi) && segno % SEGS_PER_SEC(sbi)) 2633 return !test_bit(segno, free_i->free_segmap); 2634 return 0; 2635 } 2636 2637 /* 2638 * Find a new segment from the free segments bitmap to right order 2639 * This function should be returned with success, otherwise BUG 2640 */ 2641 static void get_new_segment(struct f2fs_sb_info *sbi, 2642 unsigned int *newseg, bool new_sec, bool pinning) 2643 { 2644 struct free_segmap_info *free_i = FREE_I(sbi); 2645 unsigned int segno, secno, zoneno; 2646 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; 2647 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg); 2648 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg); 2649 bool init = true; 2650 int i; 2651 int ret = 0; 2652 2653 spin_lock(&free_i->segmap_lock); 2654 2655 if (!new_sec && ((*newseg + 1) % SEGS_PER_SEC(sbi))) { 2656 segno = find_next_zero_bit(free_i->free_segmap, 2657 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1); 2658 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1)) 2659 goto got_it; 2660 } 2661 2662 /* 2663 * If we format f2fs on zoned storage, let's try to get pinned sections 2664 * from beginning of the storage, which should be a conventional one. 2665 */ 2666 if (f2fs_sb_has_blkzoned(sbi)) { 2667 segno = pinning ? 0 : max(first_zoned_segno(sbi), *newseg); 2668 hint = GET_SEC_FROM_SEG(sbi, segno); 2669 } 2670 2671 find_other_zone: 2672 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 2673 if (secno >= MAIN_SECS(sbi)) { 2674 secno = find_first_zero_bit(free_i->free_secmap, 2675 MAIN_SECS(sbi)); 2676 if (secno >= MAIN_SECS(sbi)) { 2677 ret = -ENOSPC; 2678 goto out_unlock; 2679 } 2680 } 2681 segno = GET_SEG_FROM_SEC(sbi, secno); 2682 zoneno = GET_ZONE_FROM_SEC(sbi, secno); 2683 2684 /* give up on finding another zone */ 2685 if (!init) 2686 goto got_it; 2687 if (sbi->secs_per_zone == 1) 2688 goto got_it; 2689 if (zoneno == old_zoneno) 2690 goto got_it; 2691 for (i = 0; i < NR_CURSEG_TYPE; i++) 2692 if (CURSEG_I(sbi, i)->zone == zoneno) 2693 break; 2694 2695 if (i < NR_CURSEG_TYPE) { 2696 /* zone is in user, try another */ 2697 if (zoneno + 1 >= total_zones) 2698 hint = 0; 2699 else 2700 hint = (zoneno + 1) * sbi->secs_per_zone; 2701 init = false; 2702 goto find_other_zone; 2703 } 2704 got_it: 2705 /* set it as dirty segment in free segmap */ 2706 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); 2707 __set_inuse(sbi, segno); 2708 *newseg = segno; 2709 out_unlock: 2710 spin_unlock(&free_i->segmap_lock); 2711 2712 if (ret) { 2713 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_NO_SEGMENT); 2714 f2fs_bug_on(sbi, 1); 2715 } 2716 } 2717 2718 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 2719 { 2720 struct curseg_info *curseg = CURSEG_I(sbi, type); 2721 struct summary_footer *sum_footer; 2722 unsigned short seg_type = curseg->seg_type; 2723 2724 curseg->inited = true; 2725 curseg->segno = curseg->next_segno; 2726 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno); 2727 curseg->next_blkoff = 0; 2728 curseg->next_segno = NULL_SEGNO; 2729 2730 sum_footer = &(curseg->sum_blk->footer); 2731 memset(sum_footer, 0, sizeof(struct summary_footer)); 2732 2733 sanity_check_seg_type(sbi, seg_type); 2734 2735 if (IS_DATASEG(seg_type)) 2736 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 2737 if (IS_NODESEG(seg_type)) 2738 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 2739 __set_sit_entry_type(sbi, seg_type, curseg->segno, modified); 2740 } 2741 2742 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type) 2743 { 2744 struct curseg_info *curseg = CURSEG_I(sbi, type); 2745 unsigned short seg_type = curseg->seg_type; 2746 2747 sanity_check_seg_type(sbi, seg_type); 2748 if (f2fs_need_rand_seg(sbi)) 2749 return get_random_u32_below(MAIN_SECS(sbi) * SEGS_PER_SEC(sbi)); 2750 2751 if (__is_large_section(sbi)) 2752 return curseg->segno; 2753 2754 /* inmem log may not locate on any segment after mount */ 2755 if (!curseg->inited) 2756 return 0; 2757 2758 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 2759 return 0; 2760 2761 if (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type)) 2762 return 0; 2763 2764 if (SIT_I(sbi)->last_victim[ALLOC_NEXT]) 2765 return SIT_I(sbi)->last_victim[ALLOC_NEXT]; 2766 2767 /* find segments from 0 to reuse freed segments */ 2768 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE) 2769 return 0; 2770 2771 return curseg->segno; 2772 } 2773 2774 /* 2775 * Allocate a current working segment. 2776 * This function always allocates a free segment in LFS manner. 2777 */ 2778 static int new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 2779 { 2780 struct curseg_info *curseg = CURSEG_I(sbi, type); 2781 unsigned int segno = curseg->segno; 2782 bool pinning = type == CURSEG_COLD_DATA_PINNED; 2783 2784 if (curseg->inited) 2785 write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, segno)); 2786 2787 segno = __get_next_segno(sbi, type); 2788 get_new_segment(sbi, &segno, new_sec, pinning); 2789 if (new_sec && pinning && 2790 !f2fs_valid_pinned_area(sbi, START_BLOCK(sbi, segno))) { 2791 __set_free(sbi, segno); 2792 return -EAGAIN; 2793 } 2794 2795 curseg->next_segno = segno; 2796 reset_curseg(sbi, type, 1); 2797 curseg->alloc_type = LFS; 2798 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) 2799 curseg->fragment_remained_chunk = 2800 get_random_u32_inclusive(1, sbi->max_fragment_chunk); 2801 return 0; 2802 } 2803 2804 static int __next_free_blkoff(struct f2fs_sb_info *sbi, 2805 int segno, block_t start) 2806 { 2807 struct seg_entry *se = get_seg_entry(sbi, segno); 2808 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 2809 unsigned long *target_map = SIT_I(sbi)->tmp_map; 2810 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 2811 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 2812 int i; 2813 2814 for (i = 0; i < entries; i++) 2815 target_map[i] = ckpt_map[i] | cur_map[i]; 2816 2817 return __find_rev_next_zero_bit(target_map, BLKS_PER_SEG(sbi), start); 2818 } 2819 2820 static int f2fs_find_next_ssr_block(struct f2fs_sb_info *sbi, 2821 struct curseg_info *seg) 2822 { 2823 return __next_free_blkoff(sbi, seg->segno, seg->next_blkoff + 1); 2824 } 2825 2826 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno) 2827 { 2828 return __next_free_blkoff(sbi, segno, 0) < BLKS_PER_SEG(sbi); 2829 } 2830 2831 /* 2832 * This function always allocates a used segment(from dirty seglist) by SSR 2833 * manner, so it should recover the existing segment information of valid blocks 2834 */ 2835 static void change_curseg(struct f2fs_sb_info *sbi, int type) 2836 { 2837 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2838 struct curseg_info *curseg = CURSEG_I(sbi, type); 2839 unsigned int new_segno = curseg->next_segno; 2840 struct f2fs_summary_block *sum_node; 2841 struct page *sum_page; 2842 2843 write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, curseg->segno)); 2844 2845 __set_test_and_inuse(sbi, new_segno); 2846 2847 mutex_lock(&dirty_i->seglist_lock); 2848 __remove_dirty_segment(sbi, new_segno, PRE); 2849 __remove_dirty_segment(sbi, new_segno, DIRTY); 2850 mutex_unlock(&dirty_i->seglist_lock); 2851 2852 reset_curseg(sbi, type, 1); 2853 curseg->alloc_type = SSR; 2854 curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0); 2855 2856 sum_page = f2fs_get_sum_page(sbi, new_segno); 2857 if (IS_ERR(sum_page)) { 2858 /* GC won't be able to use stale summary pages by cp_error */ 2859 memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE); 2860 return; 2861 } 2862 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 2863 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); 2864 f2fs_put_page(sum_page, 1); 2865 } 2866 2867 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type, 2868 int alloc_mode, unsigned long long age); 2869 2870 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type, 2871 int target_type, int alloc_mode, 2872 unsigned long long age) 2873 { 2874 struct curseg_info *curseg = CURSEG_I(sbi, type); 2875 2876 curseg->seg_type = target_type; 2877 2878 if (get_ssr_segment(sbi, type, alloc_mode, age)) { 2879 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno); 2880 2881 curseg->seg_type = se->type; 2882 change_curseg(sbi, type); 2883 } else { 2884 /* allocate cold segment by default */ 2885 curseg->seg_type = CURSEG_COLD_DATA; 2886 new_curseg(sbi, type, true); 2887 } 2888 stat_inc_seg_type(sbi, curseg); 2889 } 2890 2891 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi) 2892 { 2893 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC); 2894 2895 if (!sbi->am.atgc_enabled) 2896 return; 2897 2898 f2fs_down_read(&SM_I(sbi)->curseg_lock); 2899 2900 mutex_lock(&curseg->curseg_mutex); 2901 down_write(&SIT_I(sbi)->sentry_lock); 2902 2903 get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0); 2904 2905 up_write(&SIT_I(sbi)->sentry_lock); 2906 mutex_unlock(&curseg->curseg_mutex); 2907 2908 f2fs_up_read(&SM_I(sbi)->curseg_lock); 2909 2910 } 2911 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi) 2912 { 2913 __f2fs_init_atgc_curseg(sbi); 2914 } 2915 2916 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type) 2917 { 2918 struct curseg_info *curseg = CURSEG_I(sbi, type); 2919 2920 mutex_lock(&curseg->curseg_mutex); 2921 if (!curseg->inited) 2922 goto out; 2923 2924 if (get_valid_blocks(sbi, curseg->segno, false)) { 2925 write_sum_page(sbi, curseg->sum_blk, 2926 GET_SUM_BLOCK(sbi, curseg->segno)); 2927 } else { 2928 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 2929 __set_test_and_free(sbi, curseg->segno, true); 2930 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 2931 } 2932 out: 2933 mutex_unlock(&curseg->curseg_mutex); 2934 } 2935 2936 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi) 2937 { 2938 __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED); 2939 2940 if (sbi->am.atgc_enabled) 2941 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC); 2942 } 2943 2944 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type) 2945 { 2946 struct curseg_info *curseg = CURSEG_I(sbi, type); 2947 2948 mutex_lock(&curseg->curseg_mutex); 2949 if (!curseg->inited) 2950 goto out; 2951 if (get_valid_blocks(sbi, curseg->segno, false)) 2952 goto out; 2953 2954 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 2955 __set_test_and_inuse(sbi, curseg->segno); 2956 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 2957 out: 2958 mutex_unlock(&curseg->curseg_mutex); 2959 } 2960 2961 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi) 2962 { 2963 __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED); 2964 2965 if (sbi->am.atgc_enabled) 2966 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC); 2967 } 2968 2969 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type, 2970 int alloc_mode, unsigned long long age) 2971 { 2972 struct curseg_info *curseg = CURSEG_I(sbi, type); 2973 unsigned segno = NULL_SEGNO; 2974 unsigned short seg_type = curseg->seg_type; 2975 int i, cnt; 2976 bool reversed = false; 2977 2978 sanity_check_seg_type(sbi, seg_type); 2979 2980 /* f2fs_need_SSR() already forces to do this */ 2981 if (!f2fs_get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) { 2982 curseg->next_segno = segno; 2983 return 1; 2984 } 2985 2986 /* For node segments, let's do SSR more intensively */ 2987 if (IS_NODESEG(seg_type)) { 2988 if (seg_type >= CURSEG_WARM_NODE) { 2989 reversed = true; 2990 i = CURSEG_COLD_NODE; 2991 } else { 2992 i = CURSEG_HOT_NODE; 2993 } 2994 cnt = NR_CURSEG_NODE_TYPE; 2995 } else { 2996 if (seg_type >= CURSEG_WARM_DATA) { 2997 reversed = true; 2998 i = CURSEG_COLD_DATA; 2999 } else { 3000 i = CURSEG_HOT_DATA; 3001 } 3002 cnt = NR_CURSEG_DATA_TYPE; 3003 } 3004 3005 for (; cnt-- > 0; reversed ? i-- : i++) { 3006 if (i == seg_type) 3007 continue; 3008 if (!f2fs_get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) { 3009 curseg->next_segno = segno; 3010 return 1; 3011 } 3012 } 3013 3014 /* find valid_blocks=0 in dirty list */ 3015 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 3016 segno = get_free_segment(sbi); 3017 if (segno != NULL_SEGNO) { 3018 curseg->next_segno = segno; 3019 return 1; 3020 } 3021 } 3022 return 0; 3023 } 3024 3025 static bool need_new_seg(struct f2fs_sb_info *sbi, int type) 3026 { 3027 struct curseg_info *curseg = CURSEG_I(sbi, type); 3028 3029 if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) && 3030 curseg->seg_type == CURSEG_WARM_NODE) 3031 return true; 3032 if (curseg->alloc_type == LFS && 3033 is_next_segment_free(sbi, curseg, type) && 3034 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 3035 return true; 3036 if (!f2fs_need_SSR(sbi) || !get_ssr_segment(sbi, type, SSR, 0)) 3037 return true; 3038 return false; 3039 } 3040 3041 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type, 3042 unsigned int start, unsigned int end) 3043 { 3044 struct curseg_info *curseg = CURSEG_I(sbi, type); 3045 unsigned int segno; 3046 3047 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3048 mutex_lock(&curseg->curseg_mutex); 3049 down_write(&SIT_I(sbi)->sentry_lock); 3050 3051 segno = CURSEG_I(sbi, type)->segno; 3052 if (segno < start || segno > end) 3053 goto unlock; 3054 3055 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0)) 3056 change_curseg(sbi, type); 3057 else 3058 new_curseg(sbi, type, true); 3059 3060 stat_inc_seg_type(sbi, curseg); 3061 3062 locate_dirty_segment(sbi, segno); 3063 unlock: 3064 up_write(&SIT_I(sbi)->sentry_lock); 3065 3066 if (segno != curseg->segno) 3067 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u", 3068 type, segno, curseg->segno); 3069 3070 mutex_unlock(&curseg->curseg_mutex); 3071 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3072 } 3073 3074 static int __allocate_new_segment(struct f2fs_sb_info *sbi, int type, 3075 bool new_sec, bool force) 3076 { 3077 struct curseg_info *curseg = CURSEG_I(sbi, type); 3078 unsigned int old_segno; 3079 3080 if (!force && curseg->inited && 3081 !curseg->next_blkoff && 3082 !get_valid_blocks(sbi, curseg->segno, new_sec) && 3083 !get_ckpt_valid_blocks(sbi, curseg->segno, new_sec)) 3084 return 0; 3085 3086 old_segno = curseg->segno; 3087 if (new_curseg(sbi, type, true)) 3088 return -EAGAIN; 3089 stat_inc_seg_type(sbi, curseg); 3090 locate_dirty_segment(sbi, old_segno); 3091 return 0; 3092 } 3093 3094 int f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force) 3095 { 3096 int ret; 3097 3098 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3099 down_write(&SIT_I(sbi)->sentry_lock); 3100 ret = __allocate_new_segment(sbi, type, true, force); 3101 up_write(&SIT_I(sbi)->sentry_lock); 3102 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3103 3104 return ret; 3105 } 3106 3107 int f2fs_allocate_pinning_section(struct f2fs_sb_info *sbi) 3108 { 3109 int err; 3110 bool gc_required = true; 3111 3112 retry: 3113 f2fs_lock_op(sbi); 3114 err = f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false); 3115 f2fs_unlock_op(sbi); 3116 3117 if (f2fs_sb_has_blkzoned(sbi) && err && gc_required) { 3118 f2fs_down_write(&sbi->gc_lock); 3119 f2fs_gc_range(sbi, 0, GET_SEGNO(sbi, FDEV(0).end_blk), true, 1); 3120 f2fs_up_write(&sbi->gc_lock); 3121 3122 gc_required = false; 3123 goto retry; 3124 } 3125 3126 return err; 3127 } 3128 3129 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi) 3130 { 3131 int i; 3132 3133 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3134 down_write(&SIT_I(sbi)->sentry_lock); 3135 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) 3136 __allocate_new_segment(sbi, i, false, false); 3137 up_write(&SIT_I(sbi)->sentry_lock); 3138 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3139 } 3140 3141 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi, 3142 struct cp_control *cpc) 3143 { 3144 __u64 trim_start = cpc->trim_start; 3145 bool has_candidate = false; 3146 3147 down_write(&SIT_I(sbi)->sentry_lock); 3148 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) { 3149 if (add_discard_addrs(sbi, cpc, true)) { 3150 has_candidate = true; 3151 break; 3152 } 3153 } 3154 up_write(&SIT_I(sbi)->sentry_lock); 3155 3156 cpc->trim_start = trim_start; 3157 return has_candidate; 3158 } 3159 3160 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi, 3161 struct discard_policy *dpolicy, 3162 unsigned int start, unsigned int end) 3163 { 3164 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 3165 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 3166 struct rb_node **insert_p = NULL, *insert_parent = NULL; 3167 struct discard_cmd *dc; 3168 struct blk_plug plug; 3169 int issued; 3170 unsigned int trimmed = 0; 3171 3172 next: 3173 issued = 0; 3174 3175 mutex_lock(&dcc->cmd_lock); 3176 if (unlikely(dcc->rbtree_check)) 3177 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi)); 3178 3179 dc = __lookup_discard_cmd_ret(&dcc->root, start, 3180 &prev_dc, &next_dc, &insert_p, &insert_parent); 3181 if (!dc) 3182 dc = next_dc; 3183 3184 blk_start_plug(&plug); 3185 3186 while (dc && dc->di.lstart <= end) { 3187 struct rb_node *node; 3188 int err = 0; 3189 3190 if (dc->di.len < dpolicy->granularity) 3191 goto skip; 3192 3193 if (dc->state != D_PREP) { 3194 list_move_tail(&dc->list, &dcc->fstrim_list); 3195 goto skip; 3196 } 3197 3198 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued); 3199 3200 if (issued >= dpolicy->max_requests) { 3201 start = dc->di.lstart + dc->di.len; 3202 3203 if (err) 3204 __remove_discard_cmd(sbi, dc); 3205 3206 blk_finish_plug(&plug); 3207 mutex_unlock(&dcc->cmd_lock); 3208 trimmed += __wait_all_discard_cmd(sbi, NULL); 3209 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 3210 goto next; 3211 } 3212 skip: 3213 node = rb_next(&dc->rb_node); 3214 if (err) 3215 __remove_discard_cmd(sbi, dc); 3216 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 3217 3218 if (fatal_signal_pending(current)) 3219 break; 3220 } 3221 3222 blk_finish_plug(&plug); 3223 mutex_unlock(&dcc->cmd_lock); 3224 3225 return trimmed; 3226 } 3227 3228 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) 3229 { 3230 __u64 start = F2FS_BYTES_TO_BLK(range->start); 3231 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; 3232 unsigned int start_segno, end_segno; 3233 block_t start_block, end_block; 3234 struct cp_control cpc; 3235 struct discard_policy dpolicy; 3236 unsigned long long trimmed = 0; 3237 int err = 0; 3238 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi); 3239 3240 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize) 3241 return -EINVAL; 3242 3243 if (end < MAIN_BLKADDR(sbi)) 3244 goto out; 3245 3246 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 3247 f2fs_warn(sbi, "Found FS corruption, run fsck to fix."); 3248 return -EFSCORRUPTED; 3249 } 3250 3251 /* start/end segment number in main_area */ 3252 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); 3253 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : 3254 GET_SEGNO(sbi, end); 3255 if (need_align) { 3256 start_segno = rounddown(start_segno, SEGS_PER_SEC(sbi)); 3257 end_segno = roundup(end_segno + 1, SEGS_PER_SEC(sbi)) - 1; 3258 } 3259 3260 cpc.reason = CP_DISCARD; 3261 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen)); 3262 cpc.trim_start = start_segno; 3263 cpc.trim_end = end_segno; 3264 3265 if (sbi->discard_blks == 0) 3266 goto out; 3267 3268 f2fs_down_write(&sbi->gc_lock); 3269 stat_inc_cp_call_count(sbi, TOTAL_CALL); 3270 err = f2fs_write_checkpoint(sbi, &cpc); 3271 f2fs_up_write(&sbi->gc_lock); 3272 if (err) 3273 goto out; 3274 3275 /* 3276 * We filed discard candidates, but actually we don't need to wait for 3277 * all of them, since they'll be issued in idle time along with runtime 3278 * discard option. User configuration looks like using runtime discard 3279 * or periodic fstrim instead of it. 3280 */ 3281 if (f2fs_realtime_discard_enable(sbi)) 3282 goto out; 3283 3284 start_block = START_BLOCK(sbi, start_segno); 3285 end_block = START_BLOCK(sbi, end_segno + 1); 3286 3287 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen); 3288 trimmed = __issue_discard_cmd_range(sbi, &dpolicy, 3289 start_block, end_block); 3290 3291 trimmed += __wait_discard_cmd_range(sbi, &dpolicy, 3292 start_block, end_block); 3293 out: 3294 if (!err) 3295 range->len = F2FS_BLK_TO_BYTES(trimmed); 3296 return err; 3297 } 3298 3299 int f2fs_rw_hint_to_seg_type(enum rw_hint hint) 3300 { 3301 switch (hint) { 3302 case WRITE_LIFE_SHORT: 3303 return CURSEG_HOT_DATA; 3304 case WRITE_LIFE_EXTREME: 3305 return CURSEG_COLD_DATA; 3306 default: 3307 return CURSEG_WARM_DATA; 3308 } 3309 } 3310 3311 static int __get_segment_type_2(struct f2fs_io_info *fio) 3312 { 3313 if (fio->type == DATA) 3314 return CURSEG_HOT_DATA; 3315 else 3316 return CURSEG_HOT_NODE; 3317 } 3318 3319 static int __get_segment_type_4(struct f2fs_io_info *fio) 3320 { 3321 if (fio->type == DATA) { 3322 struct inode *inode = fio->page->mapping->host; 3323 3324 if (S_ISDIR(inode->i_mode)) 3325 return CURSEG_HOT_DATA; 3326 else 3327 return CURSEG_COLD_DATA; 3328 } else { 3329 if (IS_DNODE(fio->page) && is_cold_node(fio->page)) 3330 return CURSEG_WARM_NODE; 3331 else 3332 return CURSEG_COLD_NODE; 3333 } 3334 } 3335 3336 static int __get_age_segment_type(struct inode *inode, pgoff_t pgofs) 3337 { 3338 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3339 struct extent_info ei = {}; 3340 3341 if (f2fs_lookup_age_extent_cache(inode, pgofs, &ei)) { 3342 if (!ei.age) 3343 return NO_CHECK_TYPE; 3344 if (ei.age <= sbi->hot_data_age_threshold) 3345 return CURSEG_HOT_DATA; 3346 if (ei.age <= sbi->warm_data_age_threshold) 3347 return CURSEG_WARM_DATA; 3348 return CURSEG_COLD_DATA; 3349 } 3350 return NO_CHECK_TYPE; 3351 } 3352 3353 static int __get_segment_type_6(struct f2fs_io_info *fio) 3354 { 3355 if (fio->type == DATA) { 3356 struct inode *inode = fio->page->mapping->host; 3357 int type; 3358 3359 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE)) 3360 return CURSEG_COLD_DATA_PINNED; 3361 3362 if (page_private_gcing(fio->page)) { 3363 if (fio->sbi->am.atgc_enabled && 3364 (fio->io_type == FS_DATA_IO) && 3365 (fio->sbi->gc_mode != GC_URGENT_HIGH) && 3366 __is_valid_data_blkaddr(fio->old_blkaddr) && 3367 !is_inode_flag_set(inode, FI_OPU_WRITE)) 3368 return CURSEG_ALL_DATA_ATGC; 3369 else 3370 return CURSEG_COLD_DATA; 3371 } 3372 if (file_is_cold(inode) || f2fs_need_compress_data(inode)) 3373 return CURSEG_COLD_DATA; 3374 3375 type = __get_age_segment_type(inode, fio->page->index); 3376 if (type != NO_CHECK_TYPE) 3377 return type; 3378 3379 if (file_is_hot(inode) || 3380 is_inode_flag_set(inode, FI_HOT_DATA) || 3381 f2fs_is_cow_file(inode)) 3382 return CURSEG_HOT_DATA; 3383 return f2fs_rw_hint_to_seg_type(inode->i_write_hint); 3384 } else { 3385 if (IS_DNODE(fio->page)) 3386 return is_cold_node(fio->page) ? CURSEG_WARM_NODE : 3387 CURSEG_HOT_NODE; 3388 return CURSEG_COLD_NODE; 3389 } 3390 } 3391 3392 static int __get_segment_type(struct f2fs_io_info *fio) 3393 { 3394 int type = 0; 3395 3396 switch (F2FS_OPTION(fio->sbi).active_logs) { 3397 case 2: 3398 type = __get_segment_type_2(fio); 3399 break; 3400 case 4: 3401 type = __get_segment_type_4(fio); 3402 break; 3403 case 6: 3404 type = __get_segment_type_6(fio); 3405 break; 3406 default: 3407 f2fs_bug_on(fio->sbi, true); 3408 } 3409 3410 if (IS_HOT(type)) 3411 fio->temp = HOT; 3412 else if (IS_WARM(type)) 3413 fio->temp = WARM; 3414 else 3415 fio->temp = COLD; 3416 return type; 3417 } 3418 3419 static void f2fs_randomize_chunk(struct f2fs_sb_info *sbi, 3420 struct curseg_info *seg) 3421 { 3422 /* To allocate block chunks in different sizes, use random number */ 3423 if (--seg->fragment_remained_chunk > 0) 3424 return; 3425 3426 seg->fragment_remained_chunk = 3427 get_random_u32_inclusive(1, sbi->max_fragment_chunk); 3428 seg->next_blkoff += 3429 get_random_u32_inclusive(1, sbi->max_fragment_hole); 3430 } 3431 3432 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, 3433 block_t old_blkaddr, block_t *new_blkaddr, 3434 struct f2fs_summary *sum, int type, 3435 struct f2fs_io_info *fio) 3436 { 3437 struct sit_info *sit_i = SIT_I(sbi); 3438 struct curseg_info *curseg = CURSEG_I(sbi, type); 3439 unsigned long long old_mtime; 3440 bool from_gc = (type == CURSEG_ALL_DATA_ATGC); 3441 struct seg_entry *se = NULL; 3442 bool segment_full = false; 3443 3444 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3445 3446 mutex_lock(&curseg->curseg_mutex); 3447 down_write(&sit_i->sentry_lock); 3448 3449 if (from_gc) { 3450 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO); 3451 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr)); 3452 sanity_check_seg_type(sbi, se->type); 3453 f2fs_bug_on(sbi, IS_NODESEG(se->type)); 3454 } 3455 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 3456 3457 f2fs_bug_on(sbi, curseg->next_blkoff >= BLKS_PER_SEG(sbi)); 3458 3459 f2fs_wait_discard_bio(sbi, *new_blkaddr); 3460 3461 curseg->sum_blk->entries[curseg->next_blkoff] = *sum; 3462 if (curseg->alloc_type == SSR) { 3463 curseg->next_blkoff = f2fs_find_next_ssr_block(sbi, curseg); 3464 } else { 3465 curseg->next_blkoff++; 3466 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) 3467 f2fs_randomize_chunk(sbi, curseg); 3468 } 3469 if (curseg->next_blkoff >= f2fs_usable_blks_in_seg(sbi, curseg->segno)) 3470 segment_full = true; 3471 stat_inc_block_count(sbi, curseg); 3472 3473 if (from_gc) { 3474 old_mtime = get_segment_mtime(sbi, old_blkaddr); 3475 } else { 3476 update_segment_mtime(sbi, old_blkaddr, 0); 3477 old_mtime = 0; 3478 } 3479 update_segment_mtime(sbi, *new_blkaddr, old_mtime); 3480 3481 /* 3482 * SIT information should be updated before segment allocation, 3483 * since SSR needs latest valid block information. 3484 */ 3485 update_sit_entry(sbi, *new_blkaddr, 1); 3486 update_sit_entry(sbi, old_blkaddr, -1); 3487 3488 /* 3489 * If the current segment is full, flush it out and replace it with a 3490 * new segment. 3491 */ 3492 if (segment_full) { 3493 if (type == CURSEG_COLD_DATA_PINNED && 3494 !((curseg->segno + 1) % sbi->segs_per_sec)) { 3495 write_sum_page(sbi, curseg->sum_blk, 3496 GET_SUM_BLOCK(sbi, curseg->segno)); 3497 goto skip_new_segment; 3498 } 3499 3500 if (from_gc) { 3501 get_atssr_segment(sbi, type, se->type, 3502 AT_SSR, se->mtime); 3503 } else { 3504 if (need_new_seg(sbi, type)) 3505 new_curseg(sbi, type, false); 3506 else 3507 change_curseg(sbi, type); 3508 stat_inc_seg_type(sbi, curseg); 3509 } 3510 } 3511 3512 skip_new_segment: 3513 /* 3514 * segment dirty status should be updated after segment allocation, 3515 * so we just need to update status only one time after previous 3516 * segment being closed. 3517 */ 3518 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3519 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr)); 3520 3521 if (IS_DATASEG(curseg->seg_type)) 3522 atomic64_inc(&sbi->allocated_data_blocks); 3523 3524 up_write(&sit_i->sentry_lock); 3525 3526 if (page && IS_NODESEG(curseg->seg_type)) { 3527 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); 3528 3529 f2fs_inode_chksum_set(sbi, page); 3530 } 3531 3532 if (fio) { 3533 struct f2fs_bio_info *io; 3534 3535 INIT_LIST_HEAD(&fio->list); 3536 fio->in_list = 1; 3537 io = sbi->write_io[fio->type] + fio->temp; 3538 spin_lock(&io->io_lock); 3539 list_add_tail(&fio->list, &io->io_list); 3540 spin_unlock(&io->io_lock); 3541 } 3542 3543 mutex_unlock(&curseg->curseg_mutex); 3544 3545 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3546 } 3547 3548 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino, 3549 block_t blkaddr, unsigned int blkcnt) 3550 { 3551 if (!f2fs_is_multi_device(sbi)) 3552 return; 3553 3554 while (1) { 3555 unsigned int devidx = f2fs_target_device_index(sbi, blkaddr); 3556 unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1; 3557 3558 /* update device state for fsync */ 3559 f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO); 3560 3561 /* update device state for checkpoint */ 3562 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) { 3563 spin_lock(&sbi->dev_lock); 3564 f2fs_set_bit(devidx, (char *)&sbi->dirty_device); 3565 spin_unlock(&sbi->dev_lock); 3566 } 3567 3568 if (blkcnt <= blks) 3569 break; 3570 blkcnt -= blks; 3571 blkaddr += blks; 3572 } 3573 } 3574 3575 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) 3576 { 3577 int type = __get_segment_type(fio); 3578 bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA); 3579 3580 if (keep_order) 3581 f2fs_down_read(&fio->sbi->io_order_lock); 3582 3583 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr, 3584 &fio->new_blkaddr, sum, type, fio); 3585 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO) 3586 f2fs_invalidate_internal_cache(fio->sbi, fio->old_blkaddr); 3587 3588 /* writeout dirty page into bdev */ 3589 f2fs_submit_page_write(fio); 3590 3591 f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1); 3592 3593 if (keep_order) 3594 f2fs_up_read(&fio->sbi->io_order_lock); 3595 } 3596 3597 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page, 3598 enum iostat_type io_type) 3599 { 3600 struct f2fs_io_info fio = { 3601 .sbi = sbi, 3602 .type = META, 3603 .temp = HOT, 3604 .op = REQ_OP_WRITE, 3605 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO, 3606 .old_blkaddr = page->index, 3607 .new_blkaddr = page->index, 3608 .page = page, 3609 .encrypted_page = NULL, 3610 .in_list = 0, 3611 }; 3612 3613 if (unlikely(page->index >= MAIN_BLKADDR(sbi))) 3614 fio.op_flags &= ~REQ_META; 3615 3616 set_page_writeback(page); 3617 f2fs_submit_page_write(&fio); 3618 3619 stat_inc_meta_count(sbi, page->index); 3620 f2fs_update_iostat(sbi, NULL, io_type, F2FS_BLKSIZE); 3621 } 3622 3623 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio) 3624 { 3625 struct f2fs_summary sum; 3626 3627 set_summary(&sum, nid, 0, 0); 3628 do_write_page(&sum, fio); 3629 3630 f2fs_update_iostat(fio->sbi, NULL, fio->io_type, F2FS_BLKSIZE); 3631 } 3632 3633 void f2fs_outplace_write_data(struct dnode_of_data *dn, 3634 struct f2fs_io_info *fio) 3635 { 3636 struct f2fs_sb_info *sbi = fio->sbi; 3637 struct f2fs_summary sum; 3638 3639 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); 3640 if (fio->io_type == FS_DATA_IO || fio->io_type == FS_CP_DATA_IO) 3641 f2fs_update_age_extent_cache(dn); 3642 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version); 3643 do_write_page(&sum, fio); 3644 f2fs_update_data_blkaddr(dn, fio->new_blkaddr); 3645 3646 f2fs_update_iostat(sbi, dn->inode, fio->io_type, F2FS_BLKSIZE); 3647 } 3648 3649 int f2fs_inplace_write_data(struct f2fs_io_info *fio) 3650 { 3651 int err; 3652 struct f2fs_sb_info *sbi = fio->sbi; 3653 unsigned int segno; 3654 3655 fio->new_blkaddr = fio->old_blkaddr; 3656 /* i/o temperature is needed for passing down write hints */ 3657 __get_segment_type(fio); 3658 3659 segno = GET_SEGNO(sbi, fio->new_blkaddr); 3660 3661 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) { 3662 set_sbi_flag(sbi, SBI_NEED_FSCK); 3663 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.", 3664 __func__, segno); 3665 err = -EFSCORRUPTED; 3666 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE); 3667 goto drop_bio; 3668 } 3669 3670 if (f2fs_cp_error(sbi)) { 3671 err = -EIO; 3672 goto drop_bio; 3673 } 3674 3675 if (fio->meta_gc) 3676 f2fs_truncate_meta_inode_pages(sbi, fio->new_blkaddr, 1); 3677 3678 stat_inc_inplace_blocks(fio->sbi); 3679 3680 if (fio->bio && !IS_F2FS_IPU_NOCACHE(sbi)) 3681 err = f2fs_merge_page_bio(fio); 3682 else 3683 err = f2fs_submit_page_bio(fio); 3684 if (!err) { 3685 f2fs_update_device_state(fio->sbi, fio->ino, 3686 fio->new_blkaddr, 1); 3687 f2fs_update_iostat(fio->sbi, fio->page->mapping->host, 3688 fio->io_type, F2FS_BLKSIZE); 3689 } 3690 3691 return err; 3692 drop_bio: 3693 if (fio->bio && *(fio->bio)) { 3694 struct bio *bio = *(fio->bio); 3695 3696 bio->bi_status = BLK_STS_IOERR; 3697 bio_endio(bio); 3698 *(fio->bio) = NULL; 3699 } 3700 return err; 3701 } 3702 3703 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi, 3704 unsigned int segno) 3705 { 3706 int i; 3707 3708 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 3709 if (CURSEG_I(sbi, i)->segno == segno) 3710 break; 3711 } 3712 return i; 3713 } 3714 3715 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 3716 block_t old_blkaddr, block_t new_blkaddr, 3717 bool recover_curseg, bool recover_newaddr, 3718 bool from_gc) 3719 { 3720 struct sit_info *sit_i = SIT_I(sbi); 3721 struct curseg_info *curseg; 3722 unsigned int segno, old_cursegno; 3723 struct seg_entry *se; 3724 int type; 3725 unsigned short old_blkoff; 3726 unsigned char old_alloc_type; 3727 3728 segno = GET_SEGNO(sbi, new_blkaddr); 3729 se = get_seg_entry(sbi, segno); 3730 type = se->type; 3731 3732 f2fs_down_write(&SM_I(sbi)->curseg_lock); 3733 3734 if (!recover_curseg) { 3735 /* for recovery flow */ 3736 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { 3737 if (old_blkaddr == NULL_ADDR) 3738 type = CURSEG_COLD_DATA; 3739 else 3740 type = CURSEG_WARM_DATA; 3741 } 3742 } else { 3743 if (IS_CURSEG(sbi, segno)) { 3744 /* se->type is volatile as SSR allocation */ 3745 type = __f2fs_get_curseg(sbi, segno); 3746 f2fs_bug_on(sbi, type == NO_CHECK_TYPE); 3747 } else { 3748 type = CURSEG_WARM_DATA; 3749 } 3750 } 3751 3752 f2fs_bug_on(sbi, !IS_DATASEG(type)); 3753 curseg = CURSEG_I(sbi, type); 3754 3755 mutex_lock(&curseg->curseg_mutex); 3756 down_write(&sit_i->sentry_lock); 3757 3758 old_cursegno = curseg->segno; 3759 old_blkoff = curseg->next_blkoff; 3760 old_alloc_type = curseg->alloc_type; 3761 3762 /* change the current segment */ 3763 if (segno != curseg->segno) { 3764 curseg->next_segno = segno; 3765 change_curseg(sbi, type); 3766 } 3767 3768 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); 3769 curseg->sum_blk->entries[curseg->next_blkoff] = *sum; 3770 3771 if (!recover_curseg || recover_newaddr) { 3772 if (!from_gc) 3773 update_segment_mtime(sbi, new_blkaddr, 0); 3774 update_sit_entry(sbi, new_blkaddr, 1); 3775 } 3776 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) { 3777 f2fs_invalidate_internal_cache(sbi, old_blkaddr); 3778 if (!from_gc) 3779 update_segment_mtime(sbi, old_blkaddr, 0); 3780 update_sit_entry(sbi, old_blkaddr, -1); 3781 } 3782 3783 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3784 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); 3785 3786 locate_dirty_segment(sbi, old_cursegno); 3787 3788 if (recover_curseg) { 3789 if (old_cursegno != curseg->segno) { 3790 curseg->next_segno = old_cursegno; 3791 change_curseg(sbi, type); 3792 } 3793 curseg->next_blkoff = old_blkoff; 3794 curseg->alloc_type = old_alloc_type; 3795 } 3796 3797 up_write(&sit_i->sentry_lock); 3798 mutex_unlock(&curseg->curseg_mutex); 3799 f2fs_up_write(&SM_I(sbi)->curseg_lock); 3800 } 3801 3802 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, 3803 block_t old_addr, block_t new_addr, 3804 unsigned char version, bool recover_curseg, 3805 bool recover_newaddr) 3806 { 3807 struct f2fs_summary sum; 3808 3809 set_summary(&sum, dn->nid, dn->ofs_in_node, version); 3810 3811 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr, 3812 recover_curseg, recover_newaddr, false); 3813 3814 f2fs_update_data_blkaddr(dn, new_addr); 3815 } 3816 3817 void f2fs_wait_on_page_writeback(struct page *page, 3818 enum page_type type, bool ordered, bool locked) 3819 { 3820 if (PageWriteback(page)) { 3821 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 3822 3823 /* submit cached LFS IO */ 3824 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type); 3825 /* submit cached IPU IO */ 3826 f2fs_submit_merged_ipu_write(sbi, NULL, page); 3827 if (ordered) { 3828 wait_on_page_writeback(page); 3829 f2fs_bug_on(sbi, locked && PageWriteback(page)); 3830 } else { 3831 wait_for_stable_page(page); 3832 } 3833 } 3834 } 3835 3836 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr) 3837 { 3838 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3839 struct page *cpage; 3840 3841 if (!f2fs_meta_inode_gc_required(inode)) 3842 return; 3843 3844 if (!__is_valid_data_blkaddr(blkaddr)) 3845 return; 3846 3847 cpage = find_lock_page(META_MAPPING(sbi), blkaddr); 3848 if (cpage) { 3849 f2fs_wait_on_page_writeback(cpage, DATA, true, true); 3850 f2fs_put_page(cpage, 1); 3851 } 3852 } 3853 3854 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr, 3855 block_t len) 3856 { 3857 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3858 block_t i; 3859 3860 if (!f2fs_meta_inode_gc_required(inode)) 3861 return; 3862 3863 for (i = 0; i < len; i++) 3864 f2fs_wait_on_block_writeback(inode, blkaddr + i); 3865 3866 f2fs_truncate_meta_inode_pages(sbi, blkaddr, len); 3867 } 3868 3869 static int read_compacted_summaries(struct f2fs_sb_info *sbi) 3870 { 3871 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3872 struct curseg_info *seg_i; 3873 unsigned char *kaddr; 3874 struct page *page; 3875 block_t start; 3876 int i, j, offset; 3877 3878 start = start_sum_block(sbi); 3879 3880 page = f2fs_get_meta_page(sbi, start++); 3881 if (IS_ERR(page)) 3882 return PTR_ERR(page); 3883 kaddr = (unsigned char *)page_address(page); 3884 3885 /* Step 1: restore nat cache */ 3886 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 3887 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE); 3888 3889 /* Step 2: restore sit cache */ 3890 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 3891 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE); 3892 offset = 2 * SUM_JOURNAL_SIZE; 3893 3894 /* Step 3: restore summary entries */ 3895 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 3896 unsigned short blk_off; 3897 unsigned int segno; 3898 3899 seg_i = CURSEG_I(sbi, i); 3900 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 3901 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 3902 seg_i->next_segno = segno; 3903 reset_curseg(sbi, i, 0); 3904 seg_i->alloc_type = ckpt->alloc_type[i]; 3905 seg_i->next_blkoff = blk_off; 3906 3907 if (seg_i->alloc_type == SSR) 3908 blk_off = BLKS_PER_SEG(sbi); 3909 3910 for (j = 0; j < blk_off; j++) { 3911 struct f2fs_summary *s; 3912 3913 s = (struct f2fs_summary *)(kaddr + offset); 3914 seg_i->sum_blk->entries[j] = *s; 3915 offset += SUMMARY_SIZE; 3916 if (offset + SUMMARY_SIZE <= PAGE_SIZE - 3917 SUM_FOOTER_SIZE) 3918 continue; 3919 3920 f2fs_put_page(page, 1); 3921 page = NULL; 3922 3923 page = f2fs_get_meta_page(sbi, start++); 3924 if (IS_ERR(page)) 3925 return PTR_ERR(page); 3926 kaddr = (unsigned char *)page_address(page); 3927 offset = 0; 3928 } 3929 } 3930 f2fs_put_page(page, 1); 3931 return 0; 3932 } 3933 3934 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 3935 { 3936 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3937 struct f2fs_summary_block *sum; 3938 struct curseg_info *curseg; 3939 struct page *new; 3940 unsigned short blk_off; 3941 unsigned int segno = 0; 3942 block_t blk_addr = 0; 3943 int err = 0; 3944 3945 /* get segment number and block addr */ 3946 if (IS_DATASEG(type)) { 3947 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 3948 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 3949 CURSEG_HOT_DATA]); 3950 if (__exist_node_summaries(sbi)) 3951 blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type); 3952 else 3953 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 3954 } else { 3955 segno = le32_to_cpu(ckpt->cur_node_segno[type - 3956 CURSEG_HOT_NODE]); 3957 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 3958 CURSEG_HOT_NODE]); 3959 if (__exist_node_summaries(sbi)) 3960 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 3961 type - CURSEG_HOT_NODE); 3962 else 3963 blk_addr = GET_SUM_BLOCK(sbi, segno); 3964 } 3965 3966 new = f2fs_get_meta_page(sbi, blk_addr); 3967 if (IS_ERR(new)) 3968 return PTR_ERR(new); 3969 sum = (struct f2fs_summary_block *)page_address(new); 3970 3971 if (IS_NODESEG(type)) { 3972 if (__exist_node_summaries(sbi)) { 3973 struct f2fs_summary *ns = &sum->entries[0]; 3974 int i; 3975 3976 for (i = 0; i < BLKS_PER_SEG(sbi); i++, ns++) { 3977 ns->version = 0; 3978 ns->ofs_in_node = 0; 3979 } 3980 } else { 3981 err = f2fs_restore_node_summary(sbi, segno, sum); 3982 if (err) 3983 goto out; 3984 } 3985 } 3986 3987 /* set uncompleted segment to curseg */ 3988 curseg = CURSEG_I(sbi, type); 3989 mutex_lock(&curseg->curseg_mutex); 3990 3991 /* update journal info */ 3992 down_write(&curseg->journal_rwsem); 3993 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE); 3994 up_write(&curseg->journal_rwsem); 3995 3996 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE); 3997 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE); 3998 curseg->next_segno = segno; 3999 reset_curseg(sbi, type, 0); 4000 curseg->alloc_type = ckpt->alloc_type[type]; 4001 curseg->next_blkoff = blk_off; 4002 mutex_unlock(&curseg->curseg_mutex); 4003 out: 4004 f2fs_put_page(new, 1); 4005 return err; 4006 } 4007 4008 static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 4009 { 4010 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal; 4011 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal; 4012 int type = CURSEG_HOT_DATA; 4013 int err; 4014 4015 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) { 4016 int npages = f2fs_npages_for_summary_flush(sbi, true); 4017 4018 if (npages >= 2) 4019 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages, 4020 META_CP, true); 4021 4022 /* restore for compacted data summary */ 4023 err = read_compacted_summaries(sbi); 4024 if (err) 4025 return err; 4026 type = CURSEG_HOT_NODE; 4027 } 4028 4029 if (__exist_node_summaries(sbi)) 4030 f2fs_ra_meta_pages(sbi, 4031 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type), 4032 NR_CURSEG_PERSIST_TYPE - type, META_CP, true); 4033 4034 for (; type <= CURSEG_COLD_NODE; type++) { 4035 err = read_normal_summaries(sbi, type); 4036 if (err) 4037 return err; 4038 } 4039 4040 /* sanity check for summary blocks */ 4041 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES || 4042 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) { 4043 f2fs_err(sbi, "invalid journal entries nats %u sits %u", 4044 nats_in_cursum(nat_j), sits_in_cursum(sit_j)); 4045 return -EINVAL; 4046 } 4047 4048 return 0; 4049 } 4050 4051 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 4052 { 4053 struct page *page; 4054 unsigned char *kaddr; 4055 struct f2fs_summary *summary; 4056 struct curseg_info *seg_i; 4057 int written_size = 0; 4058 int i, j; 4059 4060 page = f2fs_grab_meta_page(sbi, blkaddr++); 4061 kaddr = (unsigned char *)page_address(page); 4062 memset(kaddr, 0, PAGE_SIZE); 4063 4064 /* Step 1: write nat cache */ 4065 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 4066 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE); 4067 written_size += SUM_JOURNAL_SIZE; 4068 4069 /* Step 2: write sit cache */ 4070 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 4071 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE); 4072 written_size += SUM_JOURNAL_SIZE; 4073 4074 /* Step 3: write summary entries */ 4075 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 4076 seg_i = CURSEG_I(sbi, i); 4077 for (j = 0; j < f2fs_curseg_valid_blocks(sbi, i); j++) { 4078 if (!page) { 4079 page = f2fs_grab_meta_page(sbi, blkaddr++); 4080 kaddr = (unsigned char *)page_address(page); 4081 memset(kaddr, 0, PAGE_SIZE); 4082 written_size = 0; 4083 } 4084 summary = (struct f2fs_summary *)(kaddr + written_size); 4085 *summary = seg_i->sum_blk->entries[j]; 4086 written_size += SUMMARY_SIZE; 4087 4088 if (written_size + SUMMARY_SIZE <= PAGE_SIZE - 4089 SUM_FOOTER_SIZE) 4090 continue; 4091 4092 set_page_dirty(page); 4093 f2fs_put_page(page, 1); 4094 page = NULL; 4095 } 4096 } 4097 if (page) { 4098 set_page_dirty(page); 4099 f2fs_put_page(page, 1); 4100 } 4101 } 4102 4103 static void write_normal_summaries(struct f2fs_sb_info *sbi, 4104 block_t blkaddr, int type) 4105 { 4106 int i, end; 4107 4108 if (IS_DATASEG(type)) 4109 end = type + NR_CURSEG_DATA_TYPE; 4110 else 4111 end = type + NR_CURSEG_NODE_TYPE; 4112 4113 for (i = type; i < end; i++) 4114 write_current_sum_page(sbi, i, blkaddr + (i - type)); 4115 } 4116 4117 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 4118 { 4119 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) 4120 write_compacted_summaries(sbi, start_blk); 4121 else 4122 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 4123 } 4124 4125 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 4126 { 4127 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 4128 } 4129 4130 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type, 4131 unsigned int val, int alloc) 4132 { 4133 int i; 4134 4135 if (type == NAT_JOURNAL) { 4136 for (i = 0; i < nats_in_cursum(journal); i++) { 4137 if (le32_to_cpu(nid_in_journal(journal, i)) == val) 4138 return i; 4139 } 4140 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL)) 4141 return update_nats_in_cursum(journal, 1); 4142 } else if (type == SIT_JOURNAL) { 4143 for (i = 0; i < sits_in_cursum(journal); i++) 4144 if (le32_to_cpu(segno_in_journal(journal, i)) == val) 4145 return i; 4146 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL)) 4147 return update_sits_in_cursum(journal, 1); 4148 } 4149 return -1; 4150 } 4151 4152 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, 4153 unsigned int segno) 4154 { 4155 return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno)); 4156 } 4157 4158 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, 4159 unsigned int start) 4160 { 4161 struct sit_info *sit_i = SIT_I(sbi); 4162 struct page *page; 4163 pgoff_t src_off, dst_off; 4164 4165 src_off = current_sit_addr(sbi, start); 4166 dst_off = next_sit_addr(sbi, src_off); 4167 4168 page = f2fs_grab_meta_page(sbi, dst_off); 4169 seg_info_to_sit_page(sbi, page, start); 4170 4171 set_page_dirty(page); 4172 set_to_next_sit(sit_i, start); 4173 4174 return page; 4175 } 4176 4177 static struct sit_entry_set *grab_sit_entry_set(void) 4178 { 4179 struct sit_entry_set *ses = 4180 f2fs_kmem_cache_alloc(sit_entry_set_slab, 4181 GFP_NOFS, true, NULL); 4182 4183 ses->entry_cnt = 0; 4184 INIT_LIST_HEAD(&ses->set_list); 4185 return ses; 4186 } 4187 4188 static void release_sit_entry_set(struct sit_entry_set *ses) 4189 { 4190 list_del(&ses->set_list); 4191 kmem_cache_free(sit_entry_set_slab, ses); 4192 } 4193 4194 static void adjust_sit_entry_set(struct sit_entry_set *ses, 4195 struct list_head *head) 4196 { 4197 struct sit_entry_set *next = ses; 4198 4199 if (list_is_last(&ses->set_list, head)) 4200 return; 4201 4202 list_for_each_entry_continue(next, head, set_list) 4203 if (ses->entry_cnt <= next->entry_cnt) { 4204 list_move_tail(&ses->set_list, &next->set_list); 4205 return; 4206 } 4207 4208 list_move_tail(&ses->set_list, head); 4209 } 4210 4211 static void add_sit_entry(unsigned int segno, struct list_head *head) 4212 { 4213 struct sit_entry_set *ses; 4214 unsigned int start_segno = START_SEGNO(segno); 4215 4216 list_for_each_entry(ses, head, set_list) { 4217 if (ses->start_segno == start_segno) { 4218 ses->entry_cnt++; 4219 adjust_sit_entry_set(ses, head); 4220 return; 4221 } 4222 } 4223 4224 ses = grab_sit_entry_set(); 4225 4226 ses->start_segno = start_segno; 4227 ses->entry_cnt++; 4228 list_add(&ses->set_list, head); 4229 } 4230 4231 static void add_sits_in_set(struct f2fs_sb_info *sbi) 4232 { 4233 struct f2fs_sm_info *sm_info = SM_I(sbi); 4234 struct list_head *set_list = &sm_info->sit_entry_set; 4235 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; 4236 unsigned int segno; 4237 4238 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) 4239 add_sit_entry(segno, set_list); 4240 } 4241 4242 static void remove_sits_in_journal(struct f2fs_sb_info *sbi) 4243 { 4244 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4245 struct f2fs_journal *journal = curseg->journal; 4246 int i; 4247 4248 down_write(&curseg->journal_rwsem); 4249 for (i = 0; i < sits_in_cursum(journal); i++) { 4250 unsigned int segno; 4251 bool dirtied; 4252 4253 segno = le32_to_cpu(segno_in_journal(journal, i)); 4254 dirtied = __mark_sit_entry_dirty(sbi, segno); 4255 4256 if (!dirtied) 4257 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); 4258 } 4259 update_sits_in_cursum(journal, -i); 4260 up_write(&curseg->journal_rwsem); 4261 } 4262 4263 /* 4264 * CP calls this function, which flushes SIT entries including sit_journal, 4265 * and moves prefree segs to free segs. 4266 */ 4267 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 4268 { 4269 struct sit_info *sit_i = SIT_I(sbi); 4270 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 4271 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4272 struct f2fs_journal *journal = curseg->journal; 4273 struct sit_entry_set *ses, *tmp; 4274 struct list_head *head = &SM_I(sbi)->sit_entry_set; 4275 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS); 4276 struct seg_entry *se; 4277 4278 down_write(&sit_i->sentry_lock); 4279 4280 if (!sit_i->dirty_sentries) 4281 goto out; 4282 4283 /* 4284 * add and account sit entries of dirty bitmap in sit entry 4285 * set temporarily 4286 */ 4287 add_sits_in_set(sbi); 4288 4289 /* 4290 * if there are no enough space in journal to store dirty sit 4291 * entries, remove all entries from journal and add and account 4292 * them in sit entry set. 4293 */ 4294 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) || 4295 !to_journal) 4296 remove_sits_in_journal(sbi); 4297 4298 /* 4299 * there are two steps to flush sit entries: 4300 * #1, flush sit entries to journal in current cold data summary block. 4301 * #2, flush sit entries to sit page. 4302 */ 4303 list_for_each_entry_safe(ses, tmp, head, set_list) { 4304 struct page *page = NULL; 4305 struct f2fs_sit_block *raw_sit = NULL; 4306 unsigned int start_segno = ses->start_segno; 4307 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, 4308 (unsigned long)MAIN_SEGS(sbi)); 4309 unsigned int segno = start_segno; 4310 4311 if (to_journal && 4312 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL)) 4313 to_journal = false; 4314 4315 if (to_journal) { 4316 down_write(&curseg->journal_rwsem); 4317 } else { 4318 page = get_next_sit_page(sbi, start_segno); 4319 raw_sit = page_address(page); 4320 } 4321 4322 /* flush dirty sit entries in region of current sit set */ 4323 for_each_set_bit_from(segno, bitmap, end) { 4324 int offset, sit_offset; 4325 4326 se = get_seg_entry(sbi, segno); 4327 #ifdef CONFIG_F2FS_CHECK_FS 4328 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir, 4329 SIT_VBLOCK_MAP_SIZE)) 4330 f2fs_bug_on(sbi, 1); 4331 #endif 4332 4333 /* add discard candidates */ 4334 if (!(cpc->reason & CP_DISCARD)) { 4335 cpc->trim_start = segno; 4336 add_discard_addrs(sbi, cpc, false); 4337 } 4338 4339 if (to_journal) { 4340 offset = f2fs_lookup_journal_in_cursum(journal, 4341 SIT_JOURNAL, segno, 1); 4342 f2fs_bug_on(sbi, offset < 0); 4343 segno_in_journal(journal, offset) = 4344 cpu_to_le32(segno); 4345 seg_info_to_raw_sit(se, 4346 &sit_in_journal(journal, offset)); 4347 check_block_count(sbi, segno, 4348 &sit_in_journal(journal, offset)); 4349 } else { 4350 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 4351 seg_info_to_raw_sit(se, 4352 &raw_sit->entries[sit_offset]); 4353 check_block_count(sbi, segno, 4354 &raw_sit->entries[sit_offset]); 4355 } 4356 4357 __clear_bit(segno, bitmap); 4358 sit_i->dirty_sentries--; 4359 ses->entry_cnt--; 4360 } 4361 4362 if (to_journal) 4363 up_write(&curseg->journal_rwsem); 4364 else 4365 f2fs_put_page(page, 1); 4366 4367 f2fs_bug_on(sbi, ses->entry_cnt); 4368 release_sit_entry_set(ses); 4369 } 4370 4371 f2fs_bug_on(sbi, !list_empty(head)); 4372 f2fs_bug_on(sbi, sit_i->dirty_sentries); 4373 out: 4374 if (cpc->reason & CP_DISCARD) { 4375 __u64 trim_start = cpc->trim_start; 4376 4377 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) 4378 add_discard_addrs(sbi, cpc, false); 4379 4380 cpc->trim_start = trim_start; 4381 } 4382 up_write(&sit_i->sentry_lock); 4383 4384 set_prefree_as_free_segments(sbi); 4385 } 4386 4387 static int build_sit_info(struct f2fs_sb_info *sbi) 4388 { 4389 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 4390 struct sit_info *sit_i; 4391 unsigned int sit_segs, start; 4392 char *src_bitmap, *bitmap; 4393 unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size; 4394 unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0; 4395 4396 /* allocate memory for SIT information */ 4397 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL); 4398 if (!sit_i) 4399 return -ENOMEM; 4400 4401 SM_I(sbi)->sit_info = sit_i; 4402 4403 sit_i->sentries = 4404 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry), 4405 MAIN_SEGS(sbi)), 4406 GFP_KERNEL); 4407 if (!sit_i->sentries) 4408 return -ENOMEM; 4409 4410 main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4411 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size, 4412 GFP_KERNEL); 4413 if (!sit_i->dirty_sentries_bitmap) 4414 return -ENOMEM; 4415 4416 #ifdef CONFIG_F2FS_CHECK_FS 4417 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map); 4418 #else 4419 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map); 4420 #endif 4421 sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4422 if (!sit_i->bitmap) 4423 return -ENOMEM; 4424 4425 bitmap = sit_i->bitmap; 4426 4427 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4428 sit_i->sentries[start].cur_valid_map = bitmap; 4429 bitmap += SIT_VBLOCK_MAP_SIZE; 4430 4431 sit_i->sentries[start].ckpt_valid_map = bitmap; 4432 bitmap += SIT_VBLOCK_MAP_SIZE; 4433 4434 #ifdef CONFIG_F2FS_CHECK_FS 4435 sit_i->sentries[start].cur_valid_map_mir = bitmap; 4436 bitmap += SIT_VBLOCK_MAP_SIZE; 4437 #endif 4438 4439 if (discard_map) { 4440 sit_i->sentries[start].discard_map = bitmap; 4441 bitmap += SIT_VBLOCK_MAP_SIZE; 4442 } 4443 } 4444 4445 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 4446 if (!sit_i->tmp_map) 4447 return -ENOMEM; 4448 4449 if (__is_large_section(sbi)) { 4450 sit_i->sec_entries = 4451 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry), 4452 MAIN_SECS(sbi)), 4453 GFP_KERNEL); 4454 if (!sit_i->sec_entries) 4455 return -ENOMEM; 4456 } 4457 4458 /* get information related with SIT */ 4459 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 4460 4461 /* setup SIT bitmap from ckeckpoint pack */ 4462 sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 4463 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 4464 4465 sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL); 4466 if (!sit_i->sit_bitmap) 4467 return -ENOMEM; 4468 4469 #ifdef CONFIG_F2FS_CHECK_FS 4470 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, 4471 sit_bitmap_size, GFP_KERNEL); 4472 if (!sit_i->sit_bitmap_mir) 4473 return -ENOMEM; 4474 4475 sit_i->invalid_segmap = f2fs_kvzalloc(sbi, 4476 main_bitmap_size, GFP_KERNEL); 4477 if (!sit_i->invalid_segmap) 4478 return -ENOMEM; 4479 #endif 4480 4481 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 4482 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; 4483 sit_i->written_valid_blocks = 0; 4484 sit_i->bitmap_size = sit_bitmap_size; 4485 sit_i->dirty_sentries = 0; 4486 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 4487 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 4488 sit_i->mounted_time = ktime_get_boottime_seconds(); 4489 init_rwsem(&sit_i->sentry_lock); 4490 return 0; 4491 } 4492 4493 static int build_free_segmap(struct f2fs_sb_info *sbi) 4494 { 4495 struct free_segmap_info *free_i; 4496 unsigned int bitmap_size, sec_bitmap_size; 4497 4498 /* allocate memory for free segmap information */ 4499 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL); 4500 if (!free_i) 4501 return -ENOMEM; 4502 4503 SM_I(sbi)->free_info = free_i; 4504 4505 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4506 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL); 4507 if (!free_i->free_segmap) 4508 return -ENOMEM; 4509 4510 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4511 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL); 4512 if (!free_i->free_secmap) 4513 return -ENOMEM; 4514 4515 /* set all segments as dirty temporarily */ 4516 memset(free_i->free_segmap, 0xff, bitmap_size); 4517 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 4518 4519 /* init free segmap information */ 4520 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); 4521 free_i->free_segments = 0; 4522 free_i->free_sections = 0; 4523 spin_lock_init(&free_i->segmap_lock); 4524 return 0; 4525 } 4526 4527 static int build_curseg(struct f2fs_sb_info *sbi) 4528 { 4529 struct curseg_info *array; 4530 int i; 4531 4532 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, 4533 sizeof(*array)), GFP_KERNEL); 4534 if (!array) 4535 return -ENOMEM; 4536 4537 SM_I(sbi)->curseg_array = array; 4538 4539 for (i = 0; i < NO_CHECK_TYPE; i++) { 4540 mutex_init(&array[i].curseg_mutex); 4541 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL); 4542 if (!array[i].sum_blk) 4543 return -ENOMEM; 4544 init_rwsem(&array[i].journal_rwsem); 4545 array[i].journal = f2fs_kzalloc(sbi, 4546 sizeof(struct f2fs_journal), GFP_KERNEL); 4547 if (!array[i].journal) 4548 return -ENOMEM; 4549 if (i < NR_PERSISTENT_LOG) 4550 array[i].seg_type = CURSEG_HOT_DATA + i; 4551 else if (i == CURSEG_COLD_DATA_PINNED) 4552 array[i].seg_type = CURSEG_COLD_DATA; 4553 else if (i == CURSEG_ALL_DATA_ATGC) 4554 array[i].seg_type = CURSEG_COLD_DATA; 4555 array[i].segno = NULL_SEGNO; 4556 array[i].next_blkoff = 0; 4557 array[i].inited = false; 4558 } 4559 return restore_curseg_summaries(sbi); 4560 } 4561 4562 static int build_sit_entries(struct f2fs_sb_info *sbi) 4563 { 4564 struct sit_info *sit_i = SIT_I(sbi); 4565 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4566 struct f2fs_journal *journal = curseg->journal; 4567 struct seg_entry *se; 4568 struct f2fs_sit_entry sit; 4569 int sit_blk_cnt = SIT_BLK_CNT(sbi); 4570 unsigned int i, start, end; 4571 unsigned int readed, start_blk = 0; 4572 int err = 0; 4573 block_t sit_valid_blocks[2] = {0, 0}; 4574 4575 do { 4576 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS, 4577 META_SIT, true); 4578 4579 start = start_blk * sit_i->sents_per_block; 4580 end = (start_blk + readed) * sit_i->sents_per_block; 4581 4582 for (; start < end && start < MAIN_SEGS(sbi); start++) { 4583 struct f2fs_sit_block *sit_blk; 4584 struct page *page; 4585 4586 se = &sit_i->sentries[start]; 4587 page = get_current_sit_page(sbi, start); 4588 if (IS_ERR(page)) 4589 return PTR_ERR(page); 4590 sit_blk = (struct f2fs_sit_block *)page_address(page); 4591 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 4592 f2fs_put_page(page, 1); 4593 4594 err = check_block_count(sbi, start, &sit); 4595 if (err) 4596 return err; 4597 seg_info_from_raw_sit(se, &sit); 4598 4599 if (se->type >= NR_PERSISTENT_LOG) { 4600 f2fs_err(sbi, "Invalid segment type: %u, segno: %u", 4601 se->type, start); 4602 f2fs_handle_error(sbi, 4603 ERROR_INCONSISTENT_SUM_TYPE); 4604 return -EFSCORRUPTED; 4605 } 4606 4607 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks; 4608 4609 if (!f2fs_block_unit_discard(sbi)) 4610 goto init_discard_map_done; 4611 4612 /* build discard map only one time */ 4613 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4614 memset(se->discard_map, 0xff, 4615 SIT_VBLOCK_MAP_SIZE); 4616 goto init_discard_map_done; 4617 } 4618 memcpy(se->discard_map, se->cur_valid_map, 4619 SIT_VBLOCK_MAP_SIZE); 4620 sbi->discard_blks += BLKS_PER_SEG(sbi) - 4621 se->valid_blocks; 4622 init_discard_map_done: 4623 if (__is_large_section(sbi)) 4624 get_sec_entry(sbi, start)->valid_blocks += 4625 se->valid_blocks; 4626 } 4627 start_blk += readed; 4628 } while (start_blk < sit_blk_cnt); 4629 4630 down_read(&curseg->journal_rwsem); 4631 for (i = 0; i < sits_in_cursum(journal); i++) { 4632 unsigned int old_valid_blocks; 4633 4634 start = le32_to_cpu(segno_in_journal(journal, i)); 4635 if (start >= MAIN_SEGS(sbi)) { 4636 f2fs_err(sbi, "Wrong journal entry on segno %u", 4637 start); 4638 err = -EFSCORRUPTED; 4639 f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL); 4640 break; 4641 } 4642 4643 se = &sit_i->sentries[start]; 4644 sit = sit_in_journal(journal, i); 4645 4646 old_valid_blocks = se->valid_blocks; 4647 4648 sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks; 4649 4650 err = check_block_count(sbi, start, &sit); 4651 if (err) 4652 break; 4653 seg_info_from_raw_sit(se, &sit); 4654 4655 if (se->type >= NR_PERSISTENT_LOG) { 4656 f2fs_err(sbi, "Invalid segment type: %u, segno: %u", 4657 se->type, start); 4658 err = -EFSCORRUPTED; 4659 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE); 4660 break; 4661 } 4662 4663 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks; 4664 4665 if (f2fs_block_unit_discard(sbi)) { 4666 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4667 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE); 4668 } else { 4669 memcpy(se->discard_map, se->cur_valid_map, 4670 SIT_VBLOCK_MAP_SIZE); 4671 sbi->discard_blks += old_valid_blocks; 4672 sbi->discard_blks -= se->valid_blocks; 4673 } 4674 } 4675 4676 if (__is_large_section(sbi)) { 4677 get_sec_entry(sbi, start)->valid_blocks += 4678 se->valid_blocks; 4679 get_sec_entry(sbi, start)->valid_blocks -= 4680 old_valid_blocks; 4681 } 4682 } 4683 up_read(&curseg->journal_rwsem); 4684 4685 if (err) 4686 return err; 4687 4688 if (sit_valid_blocks[NODE] != valid_node_count(sbi)) { 4689 f2fs_err(sbi, "SIT is corrupted node# %u vs %u", 4690 sit_valid_blocks[NODE], valid_node_count(sbi)); 4691 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT); 4692 return -EFSCORRUPTED; 4693 } 4694 4695 if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] > 4696 valid_user_blocks(sbi)) { 4697 f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u", 4698 sit_valid_blocks[DATA], sit_valid_blocks[NODE], 4699 valid_user_blocks(sbi)); 4700 f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT); 4701 return -EFSCORRUPTED; 4702 } 4703 4704 return 0; 4705 } 4706 4707 static void init_free_segmap(struct f2fs_sb_info *sbi) 4708 { 4709 unsigned int start; 4710 int type; 4711 struct seg_entry *sentry; 4712 4713 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4714 if (f2fs_usable_blks_in_seg(sbi, start) == 0) 4715 continue; 4716 sentry = get_seg_entry(sbi, start); 4717 if (!sentry->valid_blocks) 4718 __set_free(sbi, start); 4719 else 4720 SIT_I(sbi)->written_valid_blocks += 4721 sentry->valid_blocks; 4722 } 4723 4724 /* set use the current segments */ 4725 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 4726 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 4727 4728 __set_test_and_inuse(sbi, curseg_t->segno); 4729 } 4730 } 4731 4732 static void init_dirty_segmap(struct f2fs_sb_info *sbi) 4733 { 4734 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4735 struct free_segmap_info *free_i = FREE_I(sbi); 4736 unsigned int segno = 0, offset = 0, secno; 4737 block_t valid_blocks, usable_blks_in_seg; 4738 4739 while (1) { 4740 /* find dirty segment based on free segmap */ 4741 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); 4742 if (segno >= MAIN_SEGS(sbi)) 4743 break; 4744 offset = segno + 1; 4745 valid_blocks = get_valid_blocks(sbi, segno, false); 4746 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); 4747 if (valid_blocks == usable_blks_in_seg || !valid_blocks) 4748 continue; 4749 if (valid_blocks > usable_blks_in_seg) { 4750 f2fs_bug_on(sbi, 1); 4751 continue; 4752 } 4753 mutex_lock(&dirty_i->seglist_lock); 4754 __locate_dirty_segment(sbi, segno, DIRTY); 4755 mutex_unlock(&dirty_i->seglist_lock); 4756 } 4757 4758 if (!__is_large_section(sbi)) 4759 return; 4760 4761 mutex_lock(&dirty_i->seglist_lock); 4762 for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) { 4763 valid_blocks = get_valid_blocks(sbi, segno, true); 4764 secno = GET_SEC_FROM_SEG(sbi, segno); 4765 4766 if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi)) 4767 continue; 4768 if (IS_CURSEC(sbi, secno)) 4769 continue; 4770 set_bit(secno, dirty_i->dirty_secmap); 4771 } 4772 mutex_unlock(&dirty_i->seglist_lock); 4773 } 4774 4775 static int init_victim_secmap(struct f2fs_sb_info *sbi) 4776 { 4777 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4778 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4779 4780 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4781 if (!dirty_i->victim_secmap) 4782 return -ENOMEM; 4783 4784 dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4785 if (!dirty_i->pinned_secmap) 4786 return -ENOMEM; 4787 4788 dirty_i->pinned_secmap_cnt = 0; 4789 dirty_i->enable_pin_section = true; 4790 return 0; 4791 } 4792 4793 static int build_dirty_segmap(struct f2fs_sb_info *sbi) 4794 { 4795 struct dirty_seglist_info *dirty_i; 4796 unsigned int bitmap_size, i; 4797 4798 /* allocate memory for dirty segments list information */ 4799 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info), 4800 GFP_KERNEL); 4801 if (!dirty_i) 4802 return -ENOMEM; 4803 4804 SM_I(sbi)->dirty_info = dirty_i; 4805 mutex_init(&dirty_i->seglist_lock); 4806 4807 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4808 4809 for (i = 0; i < NR_DIRTY_TYPE; i++) { 4810 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size, 4811 GFP_KERNEL); 4812 if (!dirty_i->dirty_segmap[i]) 4813 return -ENOMEM; 4814 } 4815 4816 if (__is_large_section(sbi)) { 4817 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4818 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi, 4819 bitmap_size, GFP_KERNEL); 4820 if (!dirty_i->dirty_secmap) 4821 return -ENOMEM; 4822 } 4823 4824 init_dirty_segmap(sbi); 4825 return init_victim_secmap(sbi); 4826 } 4827 4828 static int sanity_check_curseg(struct f2fs_sb_info *sbi) 4829 { 4830 int i; 4831 4832 /* 4833 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr; 4834 * In LFS curseg, all blkaddr after .next_blkoff should be unused. 4835 */ 4836 for (i = 0; i < NR_PERSISTENT_LOG; i++) { 4837 struct curseg_info *curseg = CURSEG_I(sbi, i); 4838 struct seg_entry *se = get_seg_entry(sbi, curseg->segno); 4839 unsigned int blkofs = curseg->next_blkoff; 4840 4841 if (f2fs_sb_has_readonly(sbi) && 4842 i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE) 4843 continue; 4844 4845 sanity_check_seg_type(sbi, curseg->seg_type); 4846 4847 if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) { 4848 f2fs_err(sbi, 4849 "Current segment has invalid alloc_type:%d", 4850 curseg->alloc_type); 4851 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG); 4852 return -EFSCORRUPTED; 4853 } 4854 4855 if (f2fs_test_bit(blkofs, se->cur_valid_map)) 4856 goto out; 4857 4858 if (curseg->alloc_type == SSR) 4859 continue; 4860 4861 for (blkofs += 1; blkofs < BLKS_PER_SEG(sbi); blkofs++) { 4862 if (!f2fs_test_bit(blkofs, se->cur_valid_map)) 4863 continue; 4864 out: 4865 f2fs_err(sbi, 4866 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u", 4867 i, curseg->segno, curseg->alloc_type, 4868 curseg->next_blkoff, blkofs); 4869 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG); 4870 return -EFSCORRUPTED; 4871 } 4872 } 4873 return 0; 4874 } 4875 4876 #ifdef CONFIG_BLK_DEV_ZONED 4877 4878 static int check_zone_write_pointer(struct f2fs_sb_info *sbi, 4879 struct f2fs_dev_info *fdev, 4880 struct blk_zone *zone) 4881 { 4882 unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno; 4883 block_t zone_block, wp_block, last_valid_block; 4884 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 4885 int i, s, b, ret; 4886 struct seg_entry *se; 4887 4888 if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4889 return 0; 4890 4891 wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block); 4892 wp_segno = GET_SEGNO(sbi, wp_block); 4893 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno); 4894 zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block); 4895 zone_segno = GET_SEGNO(sbi, zone_block); 4896 zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno); 4897 4898 if (zone_segno >= MAIN_SEGS(sbi)) 4899 return 0; 4900 4901 /* 4902 * Skip check of zones cursegs point to, since 4903 * fix_curseg_write_pointer() checks them. 4904 */ 4905 for (i = 0; i < NO_CHECK_TYPE; i++) 4906 if (zone_secno == GET_SEC_FROM_SEG(sbi, 4907 CURSEG_I(sbi, i)->segno)) 4908 return 0; 4909 4910 /* 4911 * Get last valid block of the zone. 4912 */ 4913 last_valid_block = zone_block - 1; 4914 for (s = sbi->segs_per_sec - 1; s >= 0; s--) { 4915 segno = zone_segno + s; 4916 se = get_seg_entry(sbi, segno); 4917 for (b = sbi->blocks_per_seg - 1; b >= 0; b--) 4918 if (f2fs_test_bit(b, se->cur_valid_map)) { 4919 last_valid_block = START_BLOCK(sbi, segno) + b; 4920 break; 4921 } 4922 if (last_valid_block >= zone_block) 4923 break; 4924 } 4925 4926 /* 4927 * The write pointer matches with the valid blocks or 4928 * already points to the end of the zone. 4929 */ 4930 if ((last_valid_block + 1 == wp_block) || 4931 (zone->wp == zone->start + zone->len)) 4932 return 0; 4933 4934 if (last_valid_block + 1 == zone_block) { 4935 /* 4936 * If there is no valid block in the zone and if write pointer 4937 * is not at zone start, reset the write pointer. 4938 */ 4939 f2fs_notice(sbi, 4940 "Zone without valid block has non-zero write " 4941 "pointer. Reset the write pointer: wp[0x%x,0x%x]", 4942 wp_segno, wp_blkoff); 4943 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block, 4944 zone->len >> log_sectors_per_block); 4945 if (ret) 4946 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 4947 fdev->path, ret); 4948 4949 return ret; 4950 } 4951 4952 /* 4953 * If there are valid blocks and the write pointer doesn't 4954 * match with them, we need to report the inconsistency and 4955 * fill the zone till the end to close the zone. This inconsistency 4956 * does not cause write error because the zone will not be selected 4957 * for write operation until it get discarded. 4958 */ 4959 f2fs_notice(sbi, "Valid blocks are not aligned with write pointer: " 4960 "valid block[0x%x,0x%x] wp[0x%x,0x%x]", 4961 GET_SEGNO(sbi, last_valid_block), 4962 GET_BLKOFF_FROM_SEG0(sbi, last_valid_block), 4963 wp_segno, wp_blkoff); 4964 4965 ret = blkdev_zone_mgmt(fdev->bdev, REQ_OP_ZONE_FINISH, 4966 zone->start, zone->len, GFP_NOFS); 4967 if (ret == -EOPNOTSUPP) { 4968 ret = blkdev_issue_zeroout(fdev->bdev, zone->wp, 4969 zone->len - (zone->wp - zone->start), 4970 GFP_NOFS, 0); 4971 if (ret) 4972 f2fs_err(sbi, "Fill up zone failed: %s (errno=%d)", 4973 fdev->path, ret); 4974 } else if (ret) { 4975 f2fs_err(sbi, "Finishing zone failed: %s (errno=%d)", 4976 fdev->path, ret); 4977 } 4978 4979 return ret; 4980 } 4981 4982 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi, 4983 block_t zone_blkaddr) 4984 { 4985 int i; 4986 4987 for (i = 0; i < sbi->s_ndevs; i++) { 4988 if (!bdev_is_zoned(FDEV(i).bdev)) 4989 continue; 4990 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr && 4991 zone_blkaddr <= FDEV(i).end_blk)) 4992 return &FDEV(i); 4993 } 4994 4995 return NULL; 4996 } 4997 4998 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx, 4999 void *data) 5000 { 5001 memcpy(data, zone, sizeof(struct blk_zone)); 5002 return 0; 5003 } 5004 5005 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type) 5006 { 5007 struct curseg_info *cs = CURSEG_I(sbi, type); 5008 struct f2fs_dev_info *zbd; 5009 struct blk_zone zone; 5010 unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off; 5011 block_t cs_zone_block, wp_block; 5012 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 5013 sector_t zone_sector; 5014 int err; 5015 5016 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 5017 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 5018 5019 zbd = get_target_zoned_dev(sbi, cs_zone_block); 5020 if (!zbd) 5021 return 0; 5022 5023 /* report zone for the sector the curseg points to */ 5024 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 5025 << log_sectors_per_block; 5026 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 5027 report_one_zone_cb, &zone); 5028 if (err != 1) { 5029 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 5030 zbd->path, err); 5031 return err; 5032 } 5033 5034 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 5035 return 0; 5036 5037 wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block); 5038 wp_segno = GET_SEGNO(sbi, wp_block); 5039 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno); 5040 wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0); 5041 5042 if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff && 5043 wp_sector_off == 0) 5044 return 0; 5045 5046 f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: " 5047 "curseg[0x%x,0x%x] wp[0x%x,0x%x]", 5048 type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff); 5049 5050 f2fs_notice(sbi, "Assign new section to curseg[%d]: " 5051 "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff); 5052 5053 f2fs_allocate_new_section(sbi, type, true); 5054 5055 /* check consistency of the zone curseg pointed to */ 5056 if (check_zone_write_pointer(sbi, zbd, &zone)) 5057 return -EIO; 5058 5059 /* check newly assigned zone */ 5060 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 5061 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 5062 5063 zbd = get_target_zoned_dev(sbi, cs_zone_block); 5064 if (!zbd) 5065 return 0; 5066 5067 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 5068 << log_sectors_per_block; 5069 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 5070 report_one_zone_cb, &zone); 5071 if (err != 1) { 5072 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 5073 zbd->path, err); 5074 return err; 5075 } 5076 5077 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 5078 return 0; 5079 5080 if (zone.wp != zone.start) { 5081 f2fs_notice(sbi, 5082 "New zone for curseg[%d] is not yet discarded. " 5083 "Reset the zone: curseg[0x%x,0x%x]", 5084 type, cs->segno, cs->next_blkoff); 5085 err = __f2fs_issue_discard_zone(sbi, zbd->bdev, cs_zone_block, 5086 zone.len >> log_sectors_per_block); 5087 if (err) { 5088 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 5089 zbd->path, err); 5090 return err; 5091 } 5092 } 5093 5094 return 0; 5095 } 5096 5097 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 5098 { 5099 int i, ret; 5100 5101 for (i = 0; i < NR_PERSISTENT_LOG; i++) { 5102 ret = fix_curseg_write_pointer(sbi, i); 5103 if (ret) 5104 return ret; 5105 } 5106 5107 return 0; 5108 } 5109 5110 struct check_zone_write_pointer_args { 5111 struct f2fs_sb_info *sbi; 5112 struct f2fs_dev_info *fdev; 5113 }; 5114 5115 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx, 5116 void *data) 5117 { 5118 struct check_zone_write_pointer_args *args; 5119 5120 args = (struct check_zone_write_pointer_args *)data; 5121 5122 return check_zone_write_pointer(args->sbi, args->fdev, zone); 5123 } 5124 5125 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi) 5126 { 5127 int i, ret; 5128 struct check_zone_write_pointer_args args; 5129 5130 for (i = 0; i < sbi->s_ndevs; i++) { 5131 if (!bdev_is_zoned(FDEV(i).bdev)) 5132 continue; 5133 5134 args.sbi = sbi; 5135 args.fdev = &FDEV(i); 5136 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES, 5137 check_zone_write_pointer_cb, &args); 5138 if (ret < 0) 5139 return ret; 5140 } 5141 5142 return 0; 5143 } 5144 5145 /* 5146 * Return the number of usable blocks in a segment. The number of blocks 5147 * returned is always equal to the number of blocks in a segment for 5148 * segments fully contained within a sequential zone capacity or a 5149 * conventional zone. For segments partially contained in a sequential 5150 * zone capacity, the number of usable blocks up to the zone capacity 5151 * is returned. 0 is returned in all other cases. 5152 */ 5153 static inline unsigned int f2fs_usable_zone_blks_in_seg( 5154 struct f2fs_sb_info *sbi, unsigned int segno) 5155 { 5156 block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr; 5157 unsigned int secno; 5158 5159 if (!sbi->unusable_blocks_per_sec) 5160 return BLKS_PER_SEG(sbi); 5161 5162 secno = GET_SEC_FROM_SEG(sbi, segno); 5163 seg_start = START_BLOCK(sbi, segno); 5164 sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno)); 5165 sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi); 5166 5167 /* 5168 * If segment starts before zone capacity and spans beyond 5169 * zone capacity, then usable blocks are from seg start to 5170 * zone capacity. If the segment starts after the zone capacity, 5171 * then there are no usable blocks. 5172 */ 5173 if (seg_start >= sec_cap_blkaddr) 5174 return 0; 5175 if (seg_start + BLKS_PER_SEG(sbi) > sec_cap_blkaddr) 5176 return sec_cap_blkaddr - seg_start; 5177 5178 return BLKS_PER_SEG(sbi); 5179 } 5180 #else 5181 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 5182 { 5183 return 0; 5184 } 5185 5186 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi) 5187 { 5188 return 0; 5189 } 5190 5191 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi, 5192 unsigned int segno) 5193 { 5194 return 0; 5195 } 5196 5197 #endif 5198 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi, 5199 unsigned int segno) 5200 { 5201 if (f2fs_sb_has_blkzoned(sbi)) 5202 return f2fs_usable_zone_blks_in_seg(sbi, segno); 5203 5204 return BLKS_PER_SEG(sbi); 5205 } 5206 5207 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi, 5208 unsigned int segno) 5209 { 5210 if (f2fs_sb_has_blkzoned(sbi)) 5211 return CAP_SEGS_PER_SEC(sbi); 5212 5213 return SEGS_PER_SEC(sbi); 5214 } 5215 5216 /* 5217 * Update min, max modified time for cost-benefit GC algorithm 5218 */ 5219 static void init_min_max_mtime(struct f2fs_sb_info *sbi) 5220 { 5221 struct sit_info *sit_i = SIT_I(sbi); 5222 unsigned int segno; 5223 5224 down_write(&sit_i->sentry_lock); 5225 5226 sit_i->min_mtime = ULLONG_MAX; 5227 5228 for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) { 5229 unsigned int i; 5230 unsigned long long mtime = 0; 5231 5232 for (i = 0; i < SEGS_PER_SEC(sbi); i++) 5233 mtime += get_seg_entry(sbi, segno + i)->mtime; 5234 5235 mtime = div_u64(mtime, SEGS_PER_SEC(sbi)); 5236 5237 if (sit_i->min_mtime > mtime) 5238 sit_i->min_mtime = mtime; 5239 } 5240 sit_i->max_mtime = get_mtime(sbi, false); 5241 sit_i->dirty_max_mtime = 0; 5242 up_write(&sit_i->sentry_lock); 5243 } 5244 5245 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi) 5246 { 5247 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 5248 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 5249 struct f2fs_sm_info *sm_info; 5250 int err; 5251 5252 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL); 5253 if (!sm_info) 5254 return -ENOMEM; 5255 5256 /* init sm info */ 5257 sbi->sm_info = sm_info; 5258 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 5259 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 5260 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 5261 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 5262 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 5263 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 5264 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 5265 sm_info->rec_prefree_segments = sm_info->main_segments * 5266 DEF_RECLAIM_PREFREE_SEGMENTS / 100; 5267 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS) 5268 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS; 5269 5270 if (!f2fs_lfs_mode(sbi)) 5271 sm_info->ipu_policy = BIT(F2FS_IPU_FSYNC); 5272 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; 5273 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; 5274 sm_info->min_seq_blocks = BLKS_PER_SEG(sbi); 5275 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS; 5276 sm_info->min_ssr_sections = reserved_sections(sbi); 5277 5278 INIT_LIST_HEAD(&sm_info->sit_entry_set); 5279 5280 init_f2fs_rwsem(&sm_info->curseg_lock); 5281 5282 err = f2fs_create_flush_cmd_control(sbi); 5283 if (err) 5284 return err; 5285 5286 err = create_discard_cmd_control(sbi); 5287 if (err) 5288 return err; 5289 5290 err = build_sit_info(sbi); 5291 if (err) 5292 return err; 5293 err = build_free_segmap(sbi); 5294 if (err) 5295 return err; 5296 err = build_curseg(sbi); 5297 if (err) 5298 return err; 5299 5300 /* reinit free segmap based on SIT */ 5301 err = build_sit_entries(sbi); 5302 if (err) 5303 return err; 5304 5305 init_free_segmap(sbi); 5306 err = build_dirty_segmap(sbi); 5307 if (err) 5308 return err; 5309 5310 err = sanity_check_curseg(sbi); 5311 if (err) 5312 return err; 5313 5314 init_min_max_mtime(sbi); 5315 return 0; 5316 } 5317 5318 static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 5319 enum dirty_type dirty_type) 5320 { 5321 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5322 5323 mutex_lock(&dirty_i->seglist_lock); 5324 kvfree(dirty_i->dirty_segmap[dirty_type]); 5325 dirty_i->nr_dirty[dirty_type] = 0; 5326 mutex_unlock(&dirty_i->seglist_lock); 5327 } 5328 5329 static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 5330 { 5331 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5332 5333 kvfree(dirty_i->pinned_secmap); 5334 kvfree(dirty_i->victim_secmap); 5335 } 5336 5337 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 5338 { 5339 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5340 int i; 5341 5342 if (!dirty_i) 5343 return; 5344 5345 /* discard pre-free/dirty segments list */ 5346 for (i = 0; i < NR_DIRTY_TYPE; i++) 5347 discard_dirty_segmap(sbi, i); 5348 5349 if (__is_large_section(sbi)) { 5350 mutex_lock(&dirty_i->seglist_lock); 5351 kvfree(dirty_i->dirty_secmap); 5352 mutex_unlock(&dirty_i->seglist_lock); 5353 } 5354 5355 destroy_victim_secmap(sbi); 5356 SM_I(sbi)->dirty_info = NULL; 5357 kfree(dirty_i); 5358 } 5359 5360 static void destroy_curseg(struct f2fs_sb_info *sbi) 5361 { 5362 struct curseg_info *array = SM_I(sbi)->curseg_array; 5363 int i; 5364 5365 if (!array) 5366 return; 5367 SM_I(sbi)->curseg_array = NULL; 5368 for (i = 0; i < NR_CURSEG_TYPE; i++) { 5369 kfree(array[i].sum_blk); 5370 kfree(array[i].journal); 5371 } 5372 kfree(array); 5373 } 5374 5375 static void destroy_free_segmap(struct f2fs_sb_info *sbi) 5376 { 5377 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 5378 5379 if (!free_i) 5380 return; 5381 SM_I(sbi)->free_info = NULL; 5382 kvfree(free_i->free_segmap); 5383 kvfree(free_i->free_secmap); 5384 kfree(free_i); 5385 } 5386 5387 static void destroy_sit_info(struct f2fs_sb_info *sbi) 5388 { 5389 struct sit_info *sit_i = SIT_I(sbi); 5390 5391 if (!sit_i) 5392 return; 5393 5394 if (sit_i->sentries) 5395 kvfree(sit_i->bitmap); 5396 kfree(sit_i->tmp_map); 5397 5398 kvfree(sit_i->sentries); 5399 kvfree(sit_i->sec_entries); 5400 kvfree(sit_i->dirty_sentries_bitmap); 5401 5402 SM_I(sbi)->sit_info = NULL; 5403 kvfree(sit_i->sit_bitmap); 5404 #ifdef CONFIG_F2FS_CHECK_FS 5405 kvfree(sit_i->sit_bitmap_mir); 5406 kvfree(sit_i->invalid_segmap); 5407 #endif 5408 kfree(sit_i); 5409 } 5410 5411 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi) 5412 { 5413 struct f2fs_sm_info *sm_info = SM_I(sbi); 5414 5415 if (!sm_info) 5416 return; 5417 f2fs_destroy_flush_cmd_control(sbi, true); 5418 destroy_discard_cmd_control(sbi); 5419 destroy_dirty_segmap(sbi); 5420 destroy_curseg(sbi); 5421 destroy_free_segmap(sbi); 5422 destroy_sit_info(sbi); 5423 sbi->sm_info = NULL; 5424 kfree(sm_info); 5425 } 5426 5427 int __init f2fs_create_segment_manager_caches(void) 5428 { 5429 discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry", 5430 sizeof(struct discard_entry)); 5431 if (!discard_entry_slab) 5432 goto fail; 5433 5434 discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd", 5435 sizeof(struct discard_cmd)); 5436 if (!discard_cmd_slab) 5437 goto destroy_discard_entry; 5438 5439 sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set", 5440 sizeof(struct sit_entry_set)); 5441 if (!sit_entry_set_slab) 5442 goto destroy_discard_cmd; 5443 5444 revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry", 5445 sizeof(struct revoke_entry)); 5446 if (!revoke_entry_slab) 5447 goto destroy_sit_entry_set; 5448 return 0; 5449 5450 destroy_sit_entry_set: 5451 kmem_cache_destroy(sit_entry_set_slab); 5452 destroy_discard_cmd: 5453 kmem_cache_destroy(discard_cmd_slab); 5454 destroy_discard_entry: 5455 kmem_cache_destroy(discard_entry_slab); 5456 fail: 5457 return -ENOMEM; 5458 } 5459 5460 void f2fs_destroy_segment_manager_caches(void) 5461 { 5462 kmem_cache_destroy(sit_entry_set_slab); 5463 kmem_cache_destroy(discard_cmd_slab); 5464 kmem_cache_destroy(discard_entry_slab); 5465 kmem_cache_destroy(revoke_entry_slab); 5466 } 5467