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 clear_inode_flag(fi->cow_inode, FI_COW_FILE); 196 iput(fi->cow_inode); 197 fi->cow_inode = NULL; 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 if (clean) { 205 truncate_inode_pages_final(inode->i_mapping); 206 f2fs_i_size_write(inode, fi->original_i_size); 207 } 208 } 209 210 static int __replace_atomic_write_block(struct inode *inode, pgoff_t index, 211 block_t new_addr, block_t *old_addr, bool recover) 212 { 213 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 214 struct dnode_of_data dn; 215 struct node_info ni; 216 int err; 217 218 retry: 219 set_new_dnode(&dn, inode, NULL, NULL, 0); 220 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE_RA); 221 if (err) { 222 if (err == -ENOMEM) { 223 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 224 goto retry; 225 } 226 return err; 227 } 228 229 err = f2fs_get_node_info(sbi, dn.nid, &ni, false); 230 if (err) { 231 f2fs_put_dnode(&dn); 232 return err; 233 } 234 235 if (recover) { 236 /* dn.data_blkaddr is always valid */ 237 if (!__is_valid_data_blkaddr(new_addr)) { 238 if (new_addr == NULL_ADDR) 239 dec_valid_block_count(sbi, inode, 1); 240 f2fs_invalidate_blocks(sbi, dn.data_blkaddr); 241 f2fs_update_data_blkaddr(&dn, new_addr); 242 } else { 243 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, 244 new_addr, ni.version, true, true); 245 } 246 } else { 247 blkcnt_t count = 1; 248 249 *old_addr = dn.data_blkaddr; 250 f2fs_truncate_data_blocks_range(&dn, 1); 251 dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count); 252 inc_valid_block_count(sbi, inode, &count); 253 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr, 254 ni.version, true, false); 255 } 256 257 f2fs_put_dnode(&dn); 258 return 0; 259 } 260 261 static void __complete_revoke_list(struct inode *inode, struct list_head *head, 262 bool revoke) 263 { 264 struct revoke_entry *cur, *tmp; 265 pgoff_t start_index = 0; 266 bool truncate = is_inode_flag_set(inode, FI_ATOMIC_REPLACE); 267 268 list_for_each_entry_safe(cur, tmp, head, list) { 269 if (revoke) { 270 __replace_atomic_write_block(inode, cur->index, 271 cur->old_addr, NULL, true); 272 } else if (truncate) { 273 f2fs_truncate_hole(inode, start_index, cur->index); 274 start_index = cur->index + 1; 275 } 276 277 list_del(&cur->list); 278 kmem_cache_free(revoke_entry_slab, cur); 279 } 280 281 if (!revoke && truncate) 282 f2fs_do_truncate_blocks(inode, start_index * PAGE_SIZE, false); 283 } 284 285 static int __f2fs_commit_atomic_write(struct inode *inode) 286 { 287 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 288 struct f2fs_inode_info *fi = F2FS_I(inode); 289 struct inode *cow_inode = fi->cow_inode; 290 struct revoke_entry *new; 291 struct list_head revoke_list; 292 block_t blkaddr; 293 struct dnode_of_data dn; 294 pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); 295 pgoff_t off = 0, blen, index; 296 int ret = 0, i; 297 298 INIT_LIST_HEAD(&revoke_list); 299 300 while (len) { 301 blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len); 302 303 set_new_dnode(&dn, cow_inode, NULL, NULL, 0); 304 ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA); 305 if (ret && ret != -ENOENT) { 306 goto out; 307 } else if (ret == -ENOENT) { 308 ret = 0; 309 if (dn.max_level == 0) 310 goto out; 311 goto next; 312 } 313 314 blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_page, cow_inode), 315 len); 316 index = off; 317 for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) { 318 blkaddr = f2fs_data_blkaddr(&dn); 319 320 if (!__is_valid_data_blkaddr(blkaddr)) { 321 continue; 322 } else if (!f2fs_is_valid_blkaddr(sbi, blkaddr, 323 DATA_GENERIC_ENHANCE)) { 324 f2fs_put_dnode(&dn); 325 ret = -EFSCORRUPTED; 326 f2fs_handle_error(sbi, 327 ERROR_INVALID_BLKADDR); 328 goto out; 329 } 330 331 new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS, 332 true, NULL); 333 334 ret = __replace_atomic_write_block(inode, index, blkaddr, 335 &new->old_addr, false); 336 if (ret) { 337 f2fs_put_dnode(&dn); 338 kmem_cache_free(revoke_entry_slab, new); 339 goto out; 340 } 341 342 f2fs_update_data_blkaddr(&dn, NULL_ADDR); 343 new->index = index; 344 list_add_tail(&new->list, &revoke_list); 345 } 346 f2fs_put_dnode(&dn); 347 next: 348 off += blen; 349 len -= blen; 350 } 351 352 out: 353 if (ret) { 354 sbi->revoked_atomic_block += fi->atomic_write_cnt; 355 } else { 356 sbi->committed_atomic_block += fi->atomic_write_cnt; 357 set_inode_flag(inode, FI_ATOMIC_COMMITTED); 358 } 359 360 __complete_revoke_list(inode, &revoke_list, ret ? true : false); 361 362 return ret; 363 } 364 365 int f2fs_commit_atomic_write(struct inode *inode) 366 { 367 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 368 struct f2fs_inode_info *fi = F2FS_I(inode); 369 int err; 370 371 err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX); 372 if (err) 373 return err; 374 375 f2fs_down_write(&fi->i_gc_rwsem[WRITE]); 376 f2fs_lock_op(sbi); 377 378 err = __f2fs_commit_atomic_write(inode); 379 380 f2fs_unlock_op(sbi); 381 f2fs_up_write(&fi->i_gc_rwsem[WRITE]); 382 383 return err; 384 } 385 386 /* 387 * This function balances dirty node and dentry pages. 388 * In addition, it controls garbage collection. 389 */ 390 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need) 391 { 392 if (time_to_inject(sbi, FAULT_CHECKPOINT)) { 393 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT); 394 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT); 395 } 396 397 /* balance_fs_bg is able to be pending */ 398 if (need && excess_cached_nats(sbi)) 399 f2fs_balance_fs_bg(sbi, false); 400 401 if (!f2fs_is_checkpoint_ready(sbi)) 402 return; 403 404 /* 405 * We should do GC or end up with checkpoint, if there are so many dirty 406 * dir/node pages without enough free segments. 407 */ 408 if (has_not_enough_free_secs(sbi, 0, 0)) { 409 if (test_opt(sbi, GC_MERGE) && sbi->gc_thread && 410 sbi->gc_thread->f2fs_gc_task) { 411 DEFINE_WAIT(wait); 412 413 prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait, 414 TASK_UNINTERRUPTIBLE); 415 wake_up(&sbi->gc_thread->gc_wait_queue_head); 416 io_schedule(); 417 finish_wait(&sbi->gc_thread->fggc_wq, &wait); 418 } else { 419 struct f2fs_gc_control gc_control = { 420 .victim_segno = NULL_SEGNO, 421 .init_gc_type = BG_GC, 422 .no_bg_gc = true, 423 .should_migrate_blocks = false, 424 .err_gc_skipped = false, 425 .nr_free_secs = 1 }; 426 f2fs_down_write(&sbi->gc_lock); 427 f2fs_gc(sbi, &gc_control); 428 } 429 } 430 } 431 432 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi) 433 { 434 int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2; 435 unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS); 436 unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA); 437 unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES); 438 unsigned int meta = get_pages(sbi, F2FS_DIRTY_META); 439 unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA); 440 unsigned int threshold = sbi->blocks_per_seg * factor * 441 DEFAULT_DIRTY_THRESHOLD; 442 unsigned int global_threshold = threshold * 3 / 2; 443 444 if (dents >= threshold || qdata >= threshold || 445 nodes >= threshold || meta >= threshold || 446 imeta >= threshold) 447 return true; 448 return dents + qdata + nodes + meta + imeta > global_threshold; 449 } 450 451 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg) 452 { 453 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 454 return; 455 456 /* try to shrink extent cache when there is no enough memory */ 457 if (!f2fs_available_free_memory(sbi, EXTENT_CACHE)) 458 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER); 459 460 /* check the # of cached NAT entries */ 461 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES)) 462 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK); 463 464 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) 465 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS); 466 else 467 f2fs_build_free_nids(sbi, false, false); 468 469 if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) || 470 excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi)) 471 goto do_sync; 472 473 /* there is background inflight IO or foreground operation recently */ 474 if (is_inflight_io(sbi, REQ_TIME) || 475 (!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem))) 476 return; 477 478 /* exceed periodical checkpoint timeout threshold */ 479 if (f2fs_time_over(sbi, CP_TIME)) 480 goto do_sync; 481 482 /* checkpoint is the only way to shrink partial cached entries */ 483 if (f2fs_available_free_memory(sbi, NAT_ENTRIES) && 484 f2fs_available_free_memory(sbi, INO_ENTRIES)) 485 return; 486 487 do_sync: 488 if (test_opt(sbi, DATA_FLUSH) && from_bg) { 489 struct blk_plug plug; 490 491 mutex_lock(&sbi->flush_lock); 492 493 blk_start_plug(&plug); 494 f2fs_sync_dirty_inodes(sbi, FILE_INODE, false); 495 blk_finish_plug(&plug); 496 497 mutex_unlock(&sbi->flush_lock); 498 } 499 f2fs_sync_fs(sbi->sb, 1); 500 stat_inc_bg_cp_count(sbi->stat_info); 501 } 502 503 static int __submit_flush_wait(struct f2fs_sb_info *sbi, 504 struct block_device *bdev) 505 { 506 int ret = blkdev_issue_flush(bdev); 507 508 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER), 509 test_opt(sbi, FLUSH_MERGE), ret); 510 return ret; 511 } 512 513 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino) 514 { 515 int ret = 0; 516 int i; 517 518 if (!f2fs_is_multi_device(sbi)) 519 return __submit_flush_wait(sbi, sbi->sb->s_bdev); 520 521 for (i = 0; i < sbi->s_ndevs; i++) { 522 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO)) 523 continue; 524 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 525 if (ret) 526 break; 527 } 528 return ret; 529 } 530 531 static int issue_flush_thread(void *data) 532 { 533 struct f2fs_sb_info *sbi = data; 534 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 535 wait_queue_head_t *q = &fcc->flush_wait_queue; 536 repeat: 537 if (kthread_should_stop()) 538 return 0; 539 540 if (!llist_empty(&fcc->issue_list)) { 541 struct flush_cmd *cmd, *next; 542 int ret; 543 544 fcc->dispatch_list = llist_del_all(&fcc->issue_list); 545 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list); 546 547 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode); 548 549 ret = submit_flush_wait(sbi, cmd->ino); 550 atomic_inc(&fcc->issued_flush); 551 552 llist_for_each_entry_safe(cmd, next, 553 fcc->dispatch_list, llnode) { 554 cmd->ret = ret; 555 complete(&cmd->wait); 556 } 557 fcc->dispatch_list = NULL; 558 } 559 560 wait_event_interruptible(*q, 561 kthread_should_stop() || !llist_empty(&fcc->issue_list)); 562 goto repeat; 563 } 564 565 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino) 566 { 567 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 568 struct flush_cmd cmd; 569 int ret; 570 571 if (test_opt(sbi, NOBARRIER)) 572 return 0; 573 574 if (!test_opt(sbi, FLUSH_MERGE)) { 575 atomic_inc(&fcc->queued_flush); 576 ret = submit_flush_wait(sbi, ino); 577 atomic_dec(&fcc->queued_flush); 578 atomic_inc(&fcc->issued_flush); 579 return ret; 580 } 581 582 if (atomic_inc_return(&fcc->queued_flush) == 1 || 583 f2fs_is_multi_device(sbi)) { 584 ret = submit_flush_wait(sbi, ino); 585 atomic_dec(&fcc->queued_flush); 586 587 atomic_inc(&fcc->issued_flush); 588 return ret; 589 } 590 591 cmd.ino = ino; 592 init_completion(&cmd.wait); 593 594 llist_add(&cmd.llnode, &fcc->issue_list); 595 596 /* 597 * update issue_list before we wake up issue_flush thread, this 598 * smp_mb() pairs with another barrier in ___wait_event(), see 599 * more details in comments of waitqueue_active(). 600 */ 601 smp_mb(); 602 603 if (waitqueue_active(&fcc->flush_wait_queue)) 604 wake_up(&fcc->flush_wait_queue); 605 606 if (fcc->f2fs_issue_flush) { 607 wait_for_completion(&cmd.wait); 608 atomic_dec(&fcc->queued_flush); 609 } else { 610 struct llist_node *list; 611 612 list = llist_del_all(&fcc->issue_list); 613 if (!list) { 614 wait_for_completion(&cmd.wait); 615 atomic_dec(&fcc->queued_flush); 616 } else { 617 struct flush_cmd *tmp, *next; 618 619 ret = submit_flush_wait(sbi, ino); 620 621 llist_for_each_entry_safe(tmp, next, list, llnode) { 622 if (tmp == &cmd) { 623 cmd.ret = ret; 624 atomic_dec(&fcc->queued_flush); 625 continue; 626 } 627 tmp->ret = ret; 628 complete(&tmp->wait); 629 } 630 } 631 } 632 633 return cmd.ret; 634 } 635 636 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi) 637 { 638 dev_t dev = sbi->sb->s_bdev->bd_dev; 639 struct flush_cmd_control *fcc; 640 641 if (SM_I(sbi)->fcc_info) { 642 fcc = SM_I(sbi)->fcc_info; 643 if (fcc->f2fs_issue_flush) 644 return 0; 645 goto init_thread; 646 } 647 648 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL); 649 if (!fcc) 650 return -ENOMEM; 651 atomic_set(&fcc->issued_flush, 0); 652 atomic_set(&fcc->queued_flush, 0); 653 init_waitqueue_head(&fcc->flush_wait_queue); 654 init_llist_head(&fcc->issue_list); 655 SM_I(sbi)->fcc_info = fcc; 656 if (!test_opt(sbi, FLUSH_MERGE)) 657 return 0; 658 659 init_thread: 660 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi, 661 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev)); 662 if (IS_ERR(fcc->f2fs_issue_flush)) { 663 int err = PTR_ERR(fcc->f2fs_issue_flush); 664 665 kfree(fcc); 666 SM_I(sbi)->fcc_info = NULL; 667 return err; 668 } 669 670 return 0; 671 } 672 673 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free) 674 { 675 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info; 676 677 if (fcc && fcc->f2fs_issue_flush) { 678 struct task_struct *flush_thread = fcc->f2fs_issue_flush; 679 680 fcc->f2fs_issue_flush = NULL; 681 kthread_stop(flush_thread); 682 } 683 if (free) { 684 kfree(fcc); 685 SM_I(sbi)->fcc_info = NULL; 686 } 687 } 688 689 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi) 690 { 691 int ret = 0, i; 692 693 if (!f2fs_is_multi_device(sbi)) 694 return 0; 695 696 if (test_opt(sbi, NOBARRIER)) 697 return 0; 698 699 for (i = 1; i < sbi->s_ndevs; i++) { 700 int count = DEFAULT_RETRY_IO_COUNT; 701 702 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device)) 703 continue; 704 705 do { 706 ret = __submit_flush_wait(sbi, FDEV(i).bdev); 707 if (ret) 708 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 709 } while (ret && --count); 710 711 if (ret) { 712 f2fs_stop_checkpoint(sbi, false, 713 STOP_CP_REASON_FLUSH_FAIL); 714 break; 715 } 716 717 spin_lock(&sbi->dev_lock); 718 f2fs_clear_bit(i, (char *)&sbi->dirty_device); 719 spin_unlock(&sbi->dev_lock); 720 } 721 722 return ret; 723 } 724 725 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 726 enum dirty_type dirty_type) 727 { 728 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 729 730 /* need not be added */ 731 if (IS_CURSEG(sbi, segno)) 732 return; 733 734 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 735 dirty_i->nr_dirty[dirty_type]++; 736 737 if (dirty_type == DIRTY) { 738 struct seg_entry *sentry = get_seg_entry(sbi, segno); 739 enum dirty_type t = sentry->type; 740 741 if (unlikely(t >= DIRTY)) { 742 f2fs_bug_on(sbi, 1); 743 return; 744 } 745 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t])) 746 dirty_i->nr_dirty[t]++; 747 748 if (__is_large_section(sbi)) { 749 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 750 block_t valid_blocks = 751 get_valid_blocks(sbi, segno, true); 752 753 f2fs_bug_on(sbi, unlikely(!valid_blocks || 754 valid_blocks == CAP_BLKS_PER_SEC(sbi))); 755 756 if (!IS_CURSEC(sbi, secno)) 757 set_bit(secno, dirty_i->dirty_secmap); 758 } 759 } 760 } 761 762 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 763 enum dirty_type dirty_type) 764 { 765 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 766 block_t valid_blocks; 767 768 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) 769 dirty_i->nr_dirty[dirty_type]--; 770 771 if (dirty_type == DIRTY) { 772 struct seg_entry *sentry = get_seg_entry(sbi, segno); 773 enum dirty_type t = sentry->type; 774 775 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) 776 dirty_i->nr_dirty[t]--; 777 778 valid_blocks = get_valid_blocks(sbi, segno, true); 779 if (valid_blocks == 0) { 780 clear_bit(GET_SEC_FROM_SEG(sbi, segno), 781 dirty_i->victim_secmap); 782 #ifdef CONFIG_F2FS_CHECK_FS 783 clear_bit(segno, SIT_I(sbi)->invalid_segmap); 784 #endif 785 } 786 if (__is_large_section(sbi)) { 787 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 788 789 if (!valid_blocks || 790 valid_blocks == CAP_BLKS_PER_SEC(sbi)) { 791 clear_bit(secno, dirty_i->dirty_secmap); 792 return; 793 } 794 795 if (!IS_CURSEC(sbi, secno)) 796 set_bit(secno, dirty_i->dirty_secmap); 797 } 798 } 799 } 800 801 /* 802 * Should not occur error such as -ENOMEM. 803 * Adding dirty entry into seglist is not critical operation. 804 * If a given segment is one of current working segments, it won't be added. 805 */ 806 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) 807 { 808 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 809 unsigned short valid_blocks, ckpt_valid_blocks; 810 unsigned int usable_blocks; 811 812 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) 813 return; 814 815 usable_blocks = f2fs_usable_blks_in_seg(sbi, segno); 816 mutex_lock(&dirty_i->seglist_lock); 817 818 valid_blocks = get_valid_blocks(sbi, segno, false); 819 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false); 820 821 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) || 822 ckpt_valid_blocks == usable_blocks)) { 823 __locate_dirty_segment(sbi, segno, PRE); 824 __remove_dirty_segment(sbi, segno, DIRTY); 825 } else if (valid_blocks < usable_blocks) { 826 __locate_dirty_segment(sbi, segno, DIRTY); 827 } else { 828 /* Recovery routine with SSR needs this */ 829 __remove_dirty_segment(sbi, segno, DIRTY); 830 } 831 832 mutex_unlock(&dirty_i->seglist_lock); 833 } 834 835 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */ 836 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi) 837 { 838 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 839 unsigned int segno; 840 841 mutex_lock(&dirty_i->seglist_lock); 842 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 843 if (get_valid_blocks(sbi, segno, false)) 844 continue; 845 if (IS_CURSEG(sbi, segno)) 846 continue; 847 __locate_dirty_segment(sbi, segno, PRE); 848 __remove_dirty_segment(sbi, segno, DIRTY); 849 } 850 mutex_unlock(&dirty_i->seglist_lock); 851 } 852 853 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi) 854 { 855 int ovp_hole_segs = 856 (overprovision_segments(sbi) - reserved_segments(sbi)); 857 block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg; 858 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 859 block_t holes[2] = {0, 0}; /* DATA and NODE */ 860 block_t unusable; 861 struct seg_entry *se; 862 unsigned int segno; 863 864 mutex_lock(&dirty_i->seglist_lock); 865 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 866 se = get_seg_entry(sbi, segno); 867 if (IS_NODESEG(se->type)) 868 holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) - 869 se->valid_blocks; 870 else 871 holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) - 872 se->valid_blocks; 873 } 874 mutex_unlock(&dirty_i->seglist_lock); 875 876 unusable = max(holes[DATA], holes[NODE]); 877 if (unusable > ovp_holes) 878 return unusable - ovp_holes; 879 return 0; 880 } 881 882 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable) 883 { 884 int ovp_hole_segs = 885 (overprovision_segments(sbi) - reserved_segments(sbi)); 886 if (unusable > F2FS_OPTION(sbi).unusable_cap) 887 return -EAGAIN; 888 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) && 889 dirty_segments(sbi) > ovp_hole_segs) 890 return -EAGAIN; 891 return 0; 892 } 893 894 /* This is only used by SBI_CP_DISABLED */ 895 static unsigned int get_free_segment(struct f2fs_sb_info *sbi) 896 { 897 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 898 unsigned int segno = 0; 899 900 mutex_lock(&dirty_i->seglist_lock); 901 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) { 902 if (get_valid_blocks(sbi, segno, false)) 903 continue; 904 if (get_ckpt_valid_blocks(sbi, segno, false)) 905 continue; 906 mutex_unlock(&dirty_i->seglist_lock); 907 return segno; 908 } 909 mutex_unlock(&dirty_i->seglist_lock); 910 return NULL_SEGNO; 911 } 912 913 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi, 914 struct block_device *bdev, block_t lstart, 915 block_t start, block_t len) 916 { 917 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 918 struct list_head *pend_list; 919 struct discard_cmd *dc; 920 921 f2fs_bug_on(sbi, !len); 922 923 pend_list = &dcc->pend_list[plist_idx(len)]; 924 925 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL); 926 INIT_LIST_HEAD(&dc->list); 927 dc->bdev = bdev; 928 dc->lstart = lstart; 929 dc->start = start; 930 dc->len = len; 931 dc->ref = 0; 932 dc->state = D_PREP; 933 dc->queued = 0; 934 dc->error = 0; 935 init_completion(&dc->wait); 936 list_add_tail(&dc->list, pend_list); 937 spin_lock_init(&dc->lock); 938 dc->bio_ref = 0; 939 atomic_inc(&dcc->discard_cmd_cnt); 940 dcc->undiscard_blks += len; 941 942 return dc; 943 } 944 945 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi, 946 struct block_device *bdev, block_t lstart, 947 block_t start, block_t len, 948 struct rb_node *parent, struct rb_node **p, 949 bool leftmost) 950 { 951 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 952 struct discard_cmd *dc; 953 954 dc = __create_discard_cmd(sbi, bdev, lstart, start, len); 955 956 rb_link_node(&dc->rb_node, parent, p); 957 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost); 958 959 return dc; 960 } 961 962 static void __detach_discard_cmd(struct discard_cmd_control *dcc, 963 struct discard_cmd *dc) 964 { 965 if (dc->state == D_DONE) 966 atomic_sub(dc->queued, &dcc->queued_discard); 967 968 list_del(&dc->list); 969 rb_erase_cached(&dc->rb_node, &dcc->root); 970 dcc->undiscard_blks -= dc->len; 971 972 kmem_cache_free(discard_cmd_slab, dc); 973 974 atomic_dec(&dcc->discard_cmd_cnt); 975 } 976 977 static void __remove_discard_cmd(struct f2fs_sb_info *sbi, 978 struct discard_cmd *dc) 979 { 980 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 981 unsigned long flags; 982 983 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len); 984 985 spin_lock_irqsave(&dc->lock, flags); 986 if (dc->bio_ref) { 987 spin_unlock_irqrestore(&dc->lock, flags); 988 return; 989 } 990 spin_unlock_irqrestore(&dc->lock, flags); 991 992 f2fs_bug_on(sbi, dc->ref); 993 994 if (dc->error == -EOPNOTSUPP) 995 dc->error = 0; 996 997 if (dc->error) 998 printk_ratelimited( 999 "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d", 1000 KERN_INFO, sbi->sb->s_id, 1001 dc->lstart, dc->start, dc->len, dc->error); 1002 __detach_discard_cmd(dcc, dc); 1003 } 1004 1005 static void f2fs_submit_discard_endio(struct bio *bio) 1006 { 1007 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private; 1008 unsigned long flags; 1009 1010 spin_lock_irqsave(&dc->lock, flags); 1011 if (!dc->error) 1012 dc->error = blk_status_to_errno(bio->bi_status); 1013 dc->bio_ref--; 1014 if (!dc->bio_ref && dc->state == D_SUBMIT) { 1015 dc->state = D_DONE; 1016 complete_all(&dc->wait); 1017 } 1018 spin_unlock_irqrestore(&dc->lock, flags); 1019 bio_put(bio); 1020 } 1021 1022 static void __check_sit_bitmap(struct f2fs_sb_info *sbi, 1023 block_t start, block_t end) 1024 { 1025 #ifdef CONFIG_F2FS_CHECK_FS 1026 struct seg_entry *sentry; 1027 unsigned int segno; 1028 block_t blk = start; 1029 unsigned long offset, size, max_blocks = sbi->blocks_per_seg; 1030 unsigned long *map; 1031 1032 while (blk < end) { 1033 segno = GET_SEGNO(sbi, blk); 1034 sentry = get_seg_entry(sbi, segno); 1035 offset = GET_BLKOFF_FROM_SEG0(sbi, blk); 1036 1037 if (end < START_BLOCK(sbi, segno + 1)) 1038 size = GET_BLKOFF_FROM_SEG0(sbi, end); 1039 else 1040 size = max_blocks; 1041 map = (unsigned long *)(sentry->cur_valid_map); 1042 offset = __find_rev_next_bit(map, size, offset); 1043 f2fs_bug_on(sbi, offset != size); 1044 blk = START_BLOCK(sbi, segno + 1); 1045 } 1046 #endif 1047 } 1048 1049 static void __init_discard_policy(struct f2fs_sb_info *sbi, 1050 struct discard_policy *dpolicy, 1051 int discard_type, unsigned int granularity) 1052 { 1053 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1054 1055 /* common policy */ 1056 dpolicy->type = discard_type; 1057 dpolicy->sync = true; 1058 dpolicy->ordered = false; 1059 dpolicy->granularity = granularity; 1060 1061 dpolicy->max_requests = dcc->max_discard_request; 1062 dpolicy->io_aware_gran = MAX_PLIST_NUM; 1063 dpolicy->timeout = false; 1064 1065 if (discard_type == DPOLICY_BG) { 1066 dpolicy->min_interval = dcc->min_discard_issue_time; 1067 dpolicy->mid_interval = dcc->mid_discard_issue_time; 1068 dpolicy->max_interval = dcc->max_discard_issue_time; 1069 dpolicy->io_aware = true; 1070 dpolicy->sync = false; 1071 dpolicy->ordered = true; 1072 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) { 1073 dpolicy->granularity = 1; 1074 if (atomic_read(&dcc->discard_cmd_cnt)) 1075 dpolicy->max_interval = 1076 dcc->min_discard_issue_time; 1077 } 1078 } else if (discard_type == DPOLICY_FORCE) { 1079 dpolicy->min_interval = dcc->min_discard_issue_time; 1080 dpolicy->mid_interval = dcc->mid_discard_issue_time; 1081 dpolicy->max_interval = dcc->max_discard_issue_time; 1082 dpolicy->io_aware = false; 1083 } else if (discard_type == DPOLICY_FSTRIM) { 1084 dpolicy->io_aware = false; 1085 } else if (discard_type == DPOLICY_UMOUNT) { 1086 dpolicy->io_aware = false; 1087 /* we need to issue all to keep CP_TRIMMED_FLAG */ 1088 dpolicy->granularity = 1; 1089 dpolicy->timeout = true; 1090 } 1091 } 1092 1093 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1094 struct block_device *bdev, block_t lstart, 1095 block_t start, block_t len); 1096 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */ 1097 static int __submit_discard_cmd(struct f2fs_sb_info *sbi, 1098 struct discard_policy *dpolicy, 1099 struct discard_cmd *dc, 1100 unsigned int *issued) 1101 { 1102 struct block_device *bdev = dc->bdev; 1103 unsigned int max_discard_blocks = 1104 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev)); 1105 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1106 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1107 &(dcc->fstrim_list) : &(dcc->wait_list); 1108 blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0; 1109 block_t lstart, start, len, total_len; 1110 int err = 0; 1111 1112 if (dc->state != D_PREP) 1113 return 0; 1114 1115 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 1116 return 0; 1117 1118 trace_f2fs_issue_discard(bdev, dc->start, dc->len); 1119 1120 lstart = dc->lstart; 1121 start = dc->start; 1122 len = dc->len; 1123 total_len = len; 1124 1125 dc->len = 0; 1126 1127 while (total_len && *issued < dpolicy->max_requests && !err) { 1128 struct bio *bio = NULL; 1129 unsigned long flags; 1130 bool last = true; 1131 1132 if (len > max_discard_blocks) { 1133 len = max_discard_blocks; 1134 last = false; 1135 } 1136 1137 (*issued)++; 1138 if (*issued == dpolicy->max_requests) 1139 last = true; 1140 1141 dc->len += len; 1142 1143 if (time_to_inject(sbi, FAULT_DISCARD)) { 1144 f2fs_show_injection_info(sbi, FAULT_DISCARD); 1145 err = -EIO; 1146 goto submit; 1147 } 1148 err = __blkdev_issue_discard(bdev, 1149 SECTOR_FROM_BLOCK(start), 1150 SECTOR_FROM_BLOCK(len), 1151 GFP_NOFS, &bio); 1152 submit: 1153 if (err) { 1154 spin_lock_irqsave(&dc->lock, flags); 1155 if (dc->state == D_PARTIAL) 1156 dc->state = D_SUBMIT; 1157 spin_unlock_irqrestore(&dc->lock, flags); 1158 1159 break; 1160 } 1161 1162 f2fs_bug_on(sbi, !bio); 1163 1164 /* 1165 * should keep before submission to avoid D_DONE 1166 * right away 1167 */ 1168 spin_lock_irqsave(&dc->lock, flags); 1169 if (last) 1170 dc->state = D_SUBMIT; 1171 else 1172 dc->state = D_PARTIAL; 1173 dc->bio_ref++; 1174 spin_unlock_irqrestore(&dc->lock, flags); 1175 1176 atomic_inc(&dcc->queued_discard); 1177 dc->queued++; 1178 list_move_tail(&dc->list, wait_list); 1179 1180 /* sanity check on discard range */ 1181 __check_sit_bitmap(sbi, lstart, lstart + len); 1182 1183 bio->bi_private = dc; 1184 bio->bi_end_io = f2fs_submit_discard_endio; 1185 bio->bi_opf |= flag; 1186 submit_bio(bio); 1187 1188 atomic_inc(&dcc->issued_discard); 1189 1190 f2fs_update_iostat(sbi, NULL, FS_DISCARD, 1); 1191 1192 lstart += len; 1193 start += len; 1194 total_len -= len; 1195 len = total_len; 1196 } 1197 1198 if (!err && len) { 1199 dcc->undiscard_blks -= len; 1200 __update_discard_tree_range(sbi, bdev, lstart, start, len); 1201 } 1202 return err; 1203 } 1204 1205 static void __insert_discard_tree(struct f2fs_sb_info *sbi, 1206 struct block_device *bdev, block_t lstart, 1207 block_t start, block_t len, 1208 struct rb_node **insert_p, 1209 struct rb_node *insert_parent) 1210 { 1211 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1212 struct rb_node **p; 1213 struct rb_node *parent = NULL; 1214 bool leftmost = true; 1215 1216 if (insert_p && insert_parent) { 1217 parent = insert_parent; 1218 p = insert_p; 1219 goto do_insert; 1220 } 1221 1222 p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, 1223 lstart, &leftmost); 1224 do_insert: 1225 __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, 1226 p, leftmost); 1227 } 1228 1229 static void __relocate_discard_cmd(struct discard_cmd_control *dcc, 1230 struct discard_cmd *dc) 1231 { 1232 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]); 1233 } 1234 1235 static void __punch_discard_cmd(struct f2fs_sb_info *sbi, 1236 struct discard_cmd *dc, block_t blkaddr) 1237 { 1238 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1239 struct discard_info di = dc->di; 1240 bool modified = false; 1241 1242 if (dc->state == D_DONE || dc->len == 1) { 1243 __remove_discard_cmd(sbi, dc); 1244 return; 1245 } 1246 1247 dcc->undiscard_blks -= di.len; 1248 1249 if (blkaddr > di.lstart) { 1250 dc->len = blkaddr - dc->lstart; 1251 dcc->undiscard_blks += dc->len; 1252 __relocate_discard_cmd(dcc, dc); 1253 modified = true; 1254 } 1255 1256 if (blkaddr < di.lstart + di.len - 1) { 1257 if (modified) { 1258 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1, 1259 di.start + blkaddr + 1 - di.lstart, 1260 di.lstart + di.len - 1 - blkaddr, 1261 NULL, NULL); 1262 } else { 1263 dc->lstart++; 1264 dc->len--; 1265 dc->start++; 1266 dcc->undiscard_blks += dc->len; 1267 __relocate_discard_cmd(dcc, dc); 1268 } 1269 } 1270 } 1271 1272 static void __update_discard_tree_range(struct f2fs_sb_info *sbi, 1273 struct block_device *bdev, block_t lstart, 1274 block_t start, block_t len) 1275 { 1276 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1277 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1278 struct discard_cmd *dc; 1279 struct discard_info di = {0}; 1280 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1281 unsigned int max_discard_blocks = 1282 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev)); 1283 block_t end = lstart + len; 1284 1285 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root, 1286 NULL, lstart, 1287 (struct rb_entry **)&prev_dc, 1288 (struct rb_entry **)&next_dc, 1289 &insert_p, &insert_parent, true, NULL); 1290 if (dc) 1291 prev_dc = dc; 1292 1293 if (!prev_dc) { 1294 di.lstart = lstart; 1295 di.len = next_dc ? next_dc->lstart - lstart : len; 1296 di.len = min(di.len, len); 1297 di.start = start; 1298 } 1299 1300 while (1) { 1301 struct rb_node *node; 1302 bool merged = false; 1303 struct discard_cmd *tdc = NULL; 1304 1305 if (prev_dc) { 1306 di.lstart = prev_dc->lstart + prev_dc->len; 1307 if (di.lstart < lstart) 1308 di.lstart = lstart; 1309 if (di.lstart >= end) 1310 break; 1311 1312 if (!next_dc || next_dc->lstart > end) 1313 di.len = end - di.lstart; 1314 else 1315 di.len = next_dc->lstart - di.lstart; 1316 di.start = start + di.lstart - lstart; 1317 } 1318 1319 if (!di.len) 1320 goto next; 1321 1322 if (prev_dc && prev_dc->state == D_PREP && 1323 prev_dc->bdev == bdev && 1324 __is_discard_back_mergeable(&di, &prev_dc->di, 1325 max_discard_blocks)) { 1326 prev_dc->di.len += di.len; 1327 dcc->undiscard_blks += di.len; 1328 __relocate_discard_cmd(dcc, prev_dc); 1329 di = prev_dc->di; 1330 tdc = prev_dc; 1331 merged = true; 1332 } 1333 1334 if (next_dc && next_dc->state == D_PREP && 1335 next_dc->bdev == bdev && 1336 __is_discard_front_mergeable(&di, &next_dc->di, 1337 max_discard_blocks)) { 1338 next_dc->di.lstart = di.lstart; 1339 next_dc->di.len += di.len; 1340 next_dc->di.start = di.start; 1341 dcc->undiscard_blks += di.len; 1342 __relocate_discard_cmd(dcc, next_dc); 1343 if (tdc) 1344 __remove_discard_cmd(sbi, tdc); 1345 merged = true; 1346 } 1347 1348 if (!merged) { 1349 __insert_discard_tree(sbi, bdev, di.lstart, di.start, 1350 di.len, NULL, NULL); 1351 } 1352 next: 1353 prev_dc = next_dc; 1354 if (!prev_dc) 1355 break; 1356 1357 node = rb_next(&prev_dc->rb_node); 1358 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1359 } 1360 } 1361 1362 static int __queue_discard_cmd(struct f2fs_sb_info *sbi, 1363 struct block_device *bdev, block_t blkstart, block_t blklen) 1364 { 1365 block_t lblkstart = blkstart; 1366 1367 if (!f2fs_bdev_support_discard(bdev)) 1368 return 0; 1369 1370 trace_f2fs_queue_discard(bdev, blkstart, blklen); 1371 1372 if (f2fs_is_multi_device(sbi)) { 1373 int devi = f2fs_target_device_index(sbi, blkstart); 1374 1375 blkstart -= FDEV(devi).start_blk; 1376 } 1377 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock); 1378 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen); 1379 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock); 1380 return 0; 1381 } 1382 1383 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi, 1384 struct discard_policy *dpolicy) 1385 { 1386 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1387 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 1388 struct rb_node **insert_p = NULL, *insert_parent = NULL; 1389 struct discard_cmd *dc; 1390 struct blk_plug plug; 1391 unsigned int pos = dcc->next_pos; 1392 unsigned int issued = 0; 1393 bool io_interrupted = false; 1394 1395 mutex_lock(&dcc->cmd_lock); 1396 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root, 1397 NULL, pos, 1398 (struct rb_entry **)&prev_dc, 1399 (struct rb_entry **)&next_dc, 1400 &insert_p, &insert_parent, true, NULL); 1401 if (!dc) 1402 dc = next_dc; 1403 1404 blk_start_plug(&plug); 1405 1406 while (dc) { 1407 struct rb_node *node; 1408 int err = 0; 1409 1410 if (dc->state != D_PREP) 1411 goto next; 1412 1413 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) { 1414 io_interrupted = true; 1415 break; 1416 } 1417 1418 dcc->next_pos = dc->lstart + dc->len; 1419 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued); 1420 1421 if (issued >= dpolicy->max_requests) 1422 break; 1423 next: 1424 node = rb_next(&dc->rb_node); 1425 if (err) 1426 __remove_discard_cmd(sbi, dc); 1427 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 1428 } 1429 1430 blk_finish_plug(&plug); 1431 1432 if (!dc) 1433 dcc->next_pos = 0; 1434 1435 mutex_unlock(&dcc->cmd_lock); 1436 1437 if (!issued && io_interrupted) 1438 issued = -1; 1439 1440 return issued; 1441 } 1442 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1443 struct discard_policy *dpolicy); 1444 1445 static int __issue_discard_cmd(struct f2fs_sb_info *sbi, 1446 struct discard_policy *dpolicy) 1447 { 1448 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1449 struct list_head *pend_list; 1450 struct discard_cmd *dc, *tmp; 1451 struct blk_plug plug; 1452 int i, issued; 1453 bool io_interrupted = false; 1454 1455 if (dpolicy->timeout) 1456 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT); 1457 1458 retry: 1459 issued = 0; 1460 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1461 if (dpolicy->timeout && 1462 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1463 break; 1464 1465 if (i + 1 < dpolicy->granularity) 1466 break; 1467 1468 if (i + 1 < dcc->max_ordered_discard && dpolicy->ordered) 1469 return __issue_discard_cmd_orderly(sbi, dpolicy); 1470 1471 pend_list = &dcc->pend_list[i]; 1472 1473 mutex_lock(&dcc->cmd_lock); 1474 if (list_empty(pend_list)) 1475 goto next; 1476 if (unlikely(dcc->rbtree_check)) 1477 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi, 1478 &dcc->root, false)); 1479 blk_start_plug(&plug); 1480 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1481 f2fs_bug_on(sbi, dc->state != D_PREP); 1482 1483 if (dpolicy->timeout && 1484 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT)) 1485 break; 1486 1487 if (dpolicy->io_aware && i < dpolicy->io_aware_gran && 1488 !is_idle(sbi, DISCARD_TIME)) { 1489 io_interrupted = true; 1490 break; 1491 } 1492 1493 __submit_discard_cmd(sbi, dpolicy, dc, &issued); 1494 1495 if (issued >= dpolicy->max_requests) 1496 break; 1497 } 1498 blk_finish_plug(&plug); 1499 next: 1500 mutex_unlock(&dcc->cmd_lock); 1501 1502 if (issued >= dpolicy->max_requests || io_interrupted) 1503 break; 1504 } 1505 1506 if (dpolicy->type == DPOLICY_UMOUNT && issued) { 1507 __wait_all_discard_cmd(sbi, dpolicy); 1508 goto retry; 1509 } 1510 1511 if (!issued && io_interrupted) 1512 issued = -1; 1513 1514 return issued; 1515 } 1516 1517 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi) 1518 { 1519 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1520 struct list_head *pend_list; 1521 struct discard_cmd *dc, *tmp; 1522 int i; 1523 bool dropped = false; 1524 1525 mutex_lock(&dcc->cmd_lock); 1526 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1527 pend_list = &dcc->pend_list[i]; 1528 list_for_each_entry_safe(dc, tmp, pend_list, list) { 1529 f2fs_bug_on(sbi, dc->state != D_PREP); 1530 __remove_discard_cmd(sbi, dc); 1531 dropped = true; 1532 } 1533 } 1534 mutex_unlock(&dcc->cmd_lock); 1535 1536 return dropped; 1537 } 1538 1539 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi) 1540 { 1541 __drop_discard_cmd(sbi); 1542 } 1543 1544 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi, 1545 struct discard_cmd *dc) 1546 { 1547 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1548 unsigned int len = 0; 1549 1550 wait_for_completion_io(&dc->wait); 1551 mutex_lock(&dcc->cmd_lock); 1552 f2fs_bug_on(sbi, dc->state != D_DONE); 1553 dc->ref--; 1554 if (!dc->ref) { 1555 if (!dc->error) 1556 len = dc->len; 1557 __remove_discard_cmd(sbi, dc); 1558 } 1559 mutex_unlock(&dcc->cmd_lock); 1560 1561 return len; 1562 } 1563 1564 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi, 1565 struct discard_policy *dpolicy, 1566 block_t start, block_t end) 1567 { 1568 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1569 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ? 1570 &(dcc->fstrim_list) : &(dcc->wait_list); 1571 struct discard_cmd *dc = NULL, *iter, *tmp; 1572 unsigned int trimmed = 0; 1573 1574 next: 1575 dc = NULL; 1576 1577 mutex_lock(&dcc->cmd_lock); 1578 list_for_each_entry_safe(iter, tmp, wait_list, list) { 1579 if (iter->lstart + iter->len <= start || end <= iter->lstart) 1580 continue; 1581 if (iter->len < dpolicy->granularity) 1582 continue; 1583 if (iter->state == D_DONE && !iter->ref) { 1584 wait_for_completion_io(&iter->wait); 1585 if (!iter->error) 1586 trimmed += iter->len; 1587 __remove_discard_cmd(sbi, iter); 1588 } else { 1589 iter->ref++; 1590 dc = iter; 1591 break; 1592 } 1593 } 1594 mutex_unlock(&dcc->cmd_lock); 1595 1596 if (dc) { 1597 trimmed += __wait_one_discard_bio(sbi, dc); 1598 goto next; 1599 } 1600 1601 return trimmed; 1602 } 1603 1604 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi, 1605 struct discard_policy *dpolicy) 1606 { 1607 struct discard_policy dp; 1608 unsigned int discard_blks; 1609 1610 if (dpolicy) 1611 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX); 1612 1613 /* wait all */ 1614 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1); 1615 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1616 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1); 1617 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX); 1618 1619 return discard_blks; 1620 } 1621 1622 /* This should be covered by global mutex, &sit_i->sentry_lock */ 1623 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr) 1624 { 1625 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1626 struct discard_cmd *dc; 1627 bool need_wait = false; 1628 1629 mutex_lock(&dcc->cmd_lock); 1630 dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root, 1631 NULL, blkaddr); 1632 if (dc) { 1633 if (dc->state == D_PREP) { 1634 __punch_discard_cmd(sbi, dc, blkaddr); 1635 } else { 1636 dc->ref++; 1637 need_wait = true; 1638 } 1639 } 1640 mutex_unlock(&dcc->cmd_lock); 1641 1642 if (need_wait) 1643 __wait_one_discard_bio(sbi, dc); 1644 } 1645 1646 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi) 1647 { 1648 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1649 1650 if (dcc && dcc->f2fs_issue_discard) { 1651 struct task_struct *discard_thread = dcc->f2fs_issue_discard; 1652 1653 dcc->f2fs_issue_discard = NULL; 1654 kthread_stop(discard_thread); 1655 } 1656 } 1657 1658 /* This comes from f2fs_put_super */ 1659 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi) 1660 { 1661 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1662 struct discard_policy dpolicy; 1663 bool dropped; 1664 1665 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT, 1666 dcc->discard_granularity); 1667 __issue_discard_cmd(sbi, &dpolicy); 1668 dropped = __drop_discard_cmd(sbi); 1669 1670 /* just to make sure there is no pending discard commands */ 1671 __wait_all_discard_cmd(sbi, NULL); 1672 1673 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt)); 1674 return dropped; 1675 } 1676 1677 static int issue_discard_thread(void *data) 1678 { 1679 struct f2fs_sb_info *sbi = data; 1680 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1681 wait_queue_head_t *q = &dcc->discard_wait_queue; 1682 struct discard_policy dpolicy; 1683 unsigned int wait_ms = dcc->min_discard_issue_time; 1684 int issued; 1685 1686 set_freezable(); 1687 1688 do { 1689 if (sbi->gc_mode == GC_URGENT_HIGH || 1690 !f2fs_available_free_memory(sbi, DISCARD_CACHE)) 1691 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1); 1692 else 1693 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG, 1694 dcc->discard_granularity); 1695 1696 if (!atomic_read(&dcc->discard_cmd_cnt)) 1697 wait_ms = dpolicy.max_interval; 1698 1699 wait_event_interruptible_timeout(*q, 1700 kthread_should_stop() || freezing(current) || 1701 dcc->discard_wake, 1702 msecs_to_jiffies(wait_ms)); 1703 1704 if (dcc->discard_wake) 1705 dcc->discard_wake = 0; 1706 1707 /* clean up pending candidates before going to sleep */ 1708 if (atomic_read(&dcc->queued_discard)) 1709 __wait_all_discard_cmd(sbi, NULL); 1710 1711 if (try_to_freeze()) 1712 continue; 1713 if (f2fs_readonly(sbi->sb)) 1714 continue; 1715 if (kthread_should_stop()) 1716 return 0; 1717 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 1718 wait_ms = dpolicy.max_interval; 1719 continue; 1720 } 1721 if (!atomic_read(&dcc->discard_cmd_cnt)) 1722 continue; 1723 1724 sb_start_intwrite(sbi->sb); 1725 1726 issued = __issue_discard_cmd(sbi, &dpolicy); 1727 if (issued > 0) { 1728 __wait_all_discard_cmd(sbi, &dpolicy); 1729 wait_ms = dpolicy.min_interval; 1730 } else if (issued == -1) { 1731 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME); 1732 if (!wait_ms) 1733 wait_ms = dpolicy.mid_interval; 1734 } else { 1735 wait_ms = dpolicy.max_interval; 1736 } 1737 1738 sb_end_intwrite(sbi->sb); 1739 1740 } while (!kthread_should_stop()); 1741 return 0; 1742 } 1743 1744 #ifdef CONFIG_BLK_DEV_ZONED 1745 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi, 1746 struct block_device *bdev, block_t blkstart, block_t blklen) 1747 { 1748 sector_t sector, nr_sects; 1749 block_t lblkstart = blkstart; 1750 int devi = 0; 1751 1752 if (f2fs_is_multi_device(sbi)) { 1753 devi = f2fs_target_device_index(sbi, blkstart); 1754 if (blkstart < FDEV(devi).start_blk || 1755 blkstart > FDEV(devi).end_blk) { 1756 f2fs_err(sbi, "Invalid block %x", blkstart); 1757 return -EIO; 1758 } 1759 blkstart -= FDEV(devi).start_blk; 1760 } 1761 1762 /* For sequential zones, reset the zone write pointer */ 1763 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) { 1764 sector = SECTOR_FROM_BLOCK(blkstart); 1765 nr_sects = SECTOR_FROM_BLOCK(blklen); 1766 1767 if (sector & (bdev_zone_sectors(bdev) - 1) || 1768 nr_sects != bdev_zone_sectors(bdev)) { 1769 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)", 1770 devi, sbi->s_ndevs ? FDEV(devi).path : "", 1771 blkstart, blklen); 1772 return -EIO; 1773 } 1774 trace_f2fs_issue_reset_zone(bdev, blkstart); 1775 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, 1776 sector, nr_sects, GFP_NOFS); 1777 } 1778 1779 /* For conventional zones, use regular discard if supported */ 1780 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen); 1781 } 1782 #endif 1783 1784 static int __issue_discard_async(struct f2fs_sb_info *sbi, 1785 struct block_device *bdev, block_t blkstart, block_t blklen) 1786 { 1787 #ifdef CONFIG_BLK_DEV_ZONED 1788 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) 1789 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen); 1790 #endif 1791 return __queue_discard_cmd(sbi, bdev, blkstart, blklen); 1792 } 1793 1794 static int f2fs_issue_discard(struct f2fs_sb_info *sbi, 1795 block_t blkstart, block_t blklen) 1796 { 1797 sector_t start = blkstart, len = 0; 1798 struct block_device *bdev; 1799 struct seg_entry *se; 1800 unsigned int offset; 1801 block_t i; 1802 int err = 0; 1803 1804 bdev = f2fs_target_device(sbi, blkstart, NULL); 1805 1806 for (i = blkstart; i < blkstart + blklen; i++, len++) { 1807 if (i != start) { 1808 struct block_device *bdev2 = 1809 f2fs_target_device(sbi, i, NULL); 1810 1811 if (bdev2 != bdev) { 1812 err = __issue_discard_async(sbi, bdev, 1813 start, len); 1814 if (err) 1815 return err; 1816 bdev = bdev2; 1817 start = i; 1818 len = 0; 1819 } 1820 } 1821 1822 se = get_seg_entry(sbi, GET_SEGNO(sbi, i)); 1823 offset = GET_BLKOFF_FROM_SEG0(sbi, i); 1824 1825 if (f2fs_block_unit_discard(sbi) && 1826 !f2fs_test_and_set_bit(offset, se->discard_map)) 1827 sbi->discard_blks--; 1828 } 1829 1830 if (len) 1831 err = __issue_discard_async(sbi, bdev, start, len); 1832 return err; 1833 } 1834 1835 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc, 1836 bool check_only) 1837 { 1838 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 1839 int max_blocks = sbi->blocks_per_seg; 1840 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start); 1841 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 1842 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 1843 unsigned long *discard_map = (unsigned long *)se->discard_map; 1844 unsigned long *dmap = SIT_I(sbi)->tmp_map; 1845 unsigned int start = 0, end = -1; 1846 bool force = (cpc->reason & CP_DISCARD); 1847 struct discard_entry *de = NULL; 1848 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list; 1849 int i; 1850 1851 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi) || 1852 !f2fs_block_unit_discard(sbi)) 1853 return false; 1854 1855 if (!force) { 1856 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks || 1857 SM_I(sbi)->dcc_info->nr_discards >= 1858 SM_I(sbi)->dcc_info->max_discards) 1859 return false; 1860 } 1861 1862 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */ 1863 for (i = 0; i < entries; i++) 1864 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] : 1865 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i]; 1866 1867 while (force || SM_I(sbi)->dcc_info->nr_discards <= 1868 SM_I(sbi)->dcc_info->max_discards) { 1869 start = __find_rev_next_bit(dmap, max_blocks, end + 1); 1870 if (start >= max_blocks) 1871 break; 1872 1873 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1); 1874 if (force && start && end != max_blocks 1875 && (end - start) < cpc->trim_minlen) 1876 continue; 1877 1878 if (check_only) 1879 return true; 1880 1881 if (!de) { 1882 de = f2fs_kmem_cache_alloc(discard_entry_slab, 1883 GFP_F2FS_ZERO, true, NULL); 1884 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start); 1885 list_add_tail(&de->list, head); 1886 } 1887 1888 for (i = start; i < end; i++) 1889 __set_bit_le(i, (void *)de->discard_map); 1890 1891 SM_I(sbi)->dcc_info->nr_discards += end - start; 1892 } 1893 return false; 1894 } 1895 1896 static void release_discard_addr(struct discard_entry *entry) 1897 { 1898 list_del(&entry->list); 1899 kmem_cache_free(discard_entry_slab, entry); 1900 } 1901 1902 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi) 1903 { 1904 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list); 1905 struct discard_entry *entry, *this; 1906 1907 /* drop caches */ 1908 list_for_each_entry_safe(entry, this, head, list) 1909 release_discard_addr(entry); 1910 } 1911 1912 /* 1913 * Should call f2fs_clear_prefree_segments after checkpoint is done. 1914 */ 1915 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 1916 { 1917 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1918 unsigned int segno; 1919 1920 mutex_lock(&dirty_i->seglist_lock); 1921 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi)) 1922 __set_test_and_free(sbi, segno, false); 1923 mutex_unlock(&dirty_i->seglist_lock); 1924 } 1925 1926 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi, 1927 struct cp_control *cpc) 1928 { 1929 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1930 struct list_head *head = &dcc->entry_list; 1931 struct discard_entry *entry, *this; 1932 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1933 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE]; 1934 unsigned int start = 0, end = -1; 1935 unsigned int secno, start_segno; 1936 bool force = (cpc->reason & CP_DISCARD); 1937 bool section_alignment = F2FS_OPTION(sbi).discard_unit == 1938 DISCARD_UNIT_SECTION; 1939 1940 if (f2fs_lfs_mode(sbi) && __is_large_section(sbi)) 1941 section_alignment = true; 1942 1943 mutex_lock(&dirty_i->seglist_lock); 1944 1945 while (1) { 1946 int i; 1947 1948 if (section_alignment && end != -1) 1949 end--; 1950 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1); 1951 if (start >= MAIN_SEGS(sbi)) 1952 break; 1953 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi), 1954 start + 1); 1955 1956 if (section_alignment) { 1957 start = rounddown(start, sbi->segs_per_sec); 1958 end = roundup(end, sbi->segs_per_sec); 1959 } 1960 1961 for (i = start; i < end; i++) { 1962 if (test_and_clear_bit(i, prefree_map)) 1963 dirty_i->nr_dirty[PRE]--; 1964 } 1965 1966 if (!f2fs_realtime_discard_enable(sbi)) 1967 continue; 1968 1969 if (force && start >= cpc->trim_start && 1970 (end - 1) <= cpc->trim_end) 1971 continue; 1972 1973 if (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi)) { 1974 f2fs_issue_discard(sbi, START_BLOCK(sbi, start), 1975 (end - start) << sbi->log_blocks_per_seg); 1976 continue; 1977 } 1978 next: 1979 secno = GET_SEC_FROM_SEG(sbi, start); 1980 start_segno = GET_SEG_FROM_SEC(sbi, secno); 1981 if (!IS_CURSEC(sbi, secno) && 1982 !get_valid_blocks(sbi, start, true)) 1983 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno), 1984 sbi->segs_per_sec << sbi->log_blocks_per_seg); 1985 1986 start = start_segno + sbi->segs_per_sec; 1987 if (start < end) 1988 goto next; 1989 else 1990 end = start - 1; 1991 } 1992 mutex_unlock(&dirty_i->seglist_lock); 1993 1994 if (!f2fs_block_unit_discard(sbi)) 1995 goto wakeup; 1996 1997 /* send small discards */ 1998 list_for_each_entry_safe(entry, this, head, list) { 1999 unsigned int cur_pos = 0, next_pos, len, total_len = 0; 2000 bool is_valid = test_bit_le(0, entry->discard_map); 2001 2002 find_next: 2003 if (is_valid) { 2004 next_pos = find_next_zero_bit_le(entry->discard_map, 2005 sbi->blocks_per_seg, cur_pos); 2006 len = next_pos - cur_pos; 2007 2008 if (f2fs_sb_has_blkzoned(sbi) || 2009 (force && len < cpc->trim_minlen)) 2010 goto skip; 2011 2012 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos, 2013 len); 2014 total_len += len; 2015 } else { 2016 next_pos = find_next_bit_le(entry->discard_map, 2017 sbi->blocks_per_seg, cur_pos); 2018 } 2019 skip: 2020 cur_pos = next_pos; 2021 is_valid = !is_valid; 2022 2023 if (cur_pos < sbi->blocks_per_seg) 2024 goto find_next; 2025 2026 release_discard_addr(entry); 2027 dcc->nr_discards -= total_len; 2028 } 2029 2030 wakeup: 2031 wake_up_discard_thread(sbi, false); 2032 } 2033 2034 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi) 2035 { 2036 dev_t dev = sbi->sb->s_bdev->bd_dev; 2037 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2038 int err = 0; 2039 2040 if (!f2fs_realtime_discard_enable(sbi)) 2041 return 0; 2042 2043 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi, 2044 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev)); 2045 if (IS_ERR(dcc->f2fs_issue_discard)) { 2046 err = PTR_ERR(dcc->f2fs_issue_discard); 2047 dcc->f2fs_issue_discard = NULL; 2048 } 2049 2050 return err; 2051 } 2052 2053 static int create_discard_cmd_control(struct f2fs_sb_info *sbi) 2054 { 2055 struct discard_cmd_control *dcc; 2056 int err = 0, i; 2057 2058 if (SM_I(sbi)->dcc_info) { 2059 dcc = SM_I(sbi)->dcc_info; 2060 goto init_thread; 2061 } 2062 2063 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL); 2064 if (!dcc) 2065 return -ENOMEM; 2066 2067 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY; 2068 dcc->max_ordered_discard = DEFAULT_MAX_ORDERED_DISCARD_GRANULARITY; 2069 if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT) 2070 dcc->discard_granularity = sbi->blocks_per_seg; 2071 else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION) 2072 dcc->discard_granularity = BLKS_PER_SEC(sbi); 2073 2074 INIT_LIST_HEAD(&dcc->entry_list); 2075 for (i = 0; i < MAX_PLIST_NUM; i++) 2076 INIT_LIST_HEAD(&dcc->pend_list[i]); 2077 INIT_LIST_HEAD(&dcc->wait_list); 2078 INIT_LIST_HEAD(&dcc->fstrim_list); 2079 mutex_init(&dcc->cmd_lock); 2080 atomic_set(&dcc->issued_discard, 0); 2081 atomic_set(&dcc->queued_discard, 0); 2082 atomic_set(&dcc->discard_cmd_cnt, 0); 2083 dcc->nr_discards = 0; 2084 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg; 2085 dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST; 2086 dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME; 2087 dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME; 2088 dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME; 2089 dcc->undiscard_blks = 0; 2090 dcc->next_pos = 0; 2091 dcc->root = RB_ROOT_CACHED; 2092 dcc->rbtree_check = false; 2093 2094 init_waitqueue_head(&dcc->discard_wait_queue); 2095 SM_I(sbi)->dcc_info = dcc; 2096 init_thread: 2097 err = f2fs_start_discard_thread(sbi); 2098 if (err) { 2099 kfree(dcc); 2100 SM_I(sbi)->dcc_info = NULL; 2101 } 2102 2103 return err; 2104 } 2105 2106 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi) 2107 { 2108 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2109 2110 if (!dcc) 2111 return; 2112 2113 f2fs_stop_discard_thread(sbi); 2114 2115 /* 2116 * Recovery can cache discard commands, so in error path of 2117 * fill_super(), it needs to give a chance to handle them. 2118 */ 2119 if (unlikely(atomic_read(&dcc->discard_cmd_cnt))) 2120 f2fs_issue_discard_timeout(sbi); 2121 2122 kfree(dcc); 2123 SM_I(sbi)->dcc_info = NULL; 2124 } 2125 2126 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 2127 { 2128 struct sit_info *sit_i = SIT_I(sbi); 2129 2130 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) { 2131 sit_i->dirty_sentries++; 2132 return false; 2133 } 2134 2135 return true; 2136 } 2137 2138 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 2139 unsigned int segno, int modified) 2140 { 2141 struct seg_entry *se = get_seg_entry(sbi, segno); 2142 2143 se->type = type; 2144 if (modified) 2145 __mark_sit_entry_dirty(sbi, segno); 2146 } 2147 2148 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi, 2149 block_t blkaddr) 2150 { 2151 unsigned int segno = GET_SEGNO(sbi, blkaddr); 2152 2153 if (segno == NULL_SEGNO) 2154 return 0; 2155 return get_seg_entry(sbi, segno)->mtime; 2156 } 2157 2158 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr, 2159 unsigned long long old_mtime) 2160 { 2161 struct seg_entry *se; 2162 unsigned int segno = GET_SEGNO(sbi, blkaddr); 2163 unsigned long long ctime = get_mtime(sbi, false); 2164 unsigned long long mtime = old_mtime ? old_mtime : ctime; 2165 2166 if (segno == NULL_SEGNO) 2167 return; 2168 2169 se = get_seg_entry(sbi, segno); 2170 2171 if (!se->mtime) 2172 se->mtime = mtime; 2173 else 2174 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime, 2175 se->valid_blocks + 1); 2176 2177 if (ctime > SIT_I(sbi)->max_mtime) 2178 SIT_I(sbi)->max_mtime = ctime; 2179 } 2180 2181 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 2182 { 2183 struct seg_entry *se; 2184 unsigned int segno, offset; 2185 long int new_vblocks; 2186 bool exist; 2187 #ifdef CONFIG_F2FS_CHECK_FS 2188 bool mir_exist; 2189 #endif 2190 2191 segno = GET_SEGNO(sbi, blkaddr); 2192 2193 se = get_seg_entry(sbi, segno); 2194 new_vblocks = se->valid_blocks + del; 2195 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2196 2197 f2fs_bug_on(sbi, (new_vblocks < 0 || 2198 (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno)))); 2199 2200 se->valid_blocks = new_vblocks; 2201 2202 /* Update valid block bitmap */ 2203 if (del > 0) { 2204 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map); 2205 #ifdef CONFIG_F2FS_CHECK_FS 2206 mir_exist = f2fs_test_and_set_bit(offset, 2207 se->cur_valid_map_mir); 2208 if (unlikely(exist != mir_exist)) { 2209 f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d", 2210 blkaddr, exist); 2211 f2fs_bug_on(sbi, 1); 2212 } 2213 #endif 2214 if (unlikely(exist)) { 2215 f2fs_err(sbi, "Bitmap was wrongly set, blk:%u", 2216 blkaddr); 2217 f2fs_bug_on(sbi, 1); 2218 se->valid_blocks--; 2219 del = 0; 2220 } 2221 2222 if (f2fs_block_unit_discard(sbi) && 2223 !f2fs_test_and_set_bit(offset, se->discard_map)) 2224 sbi->discard_blks--; 2225 2226 /* 2227 * SSR should never reuse block which is checkpointed 2228 * or newly invalidated. 2229 */ 2230 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) { 2231 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map)) 2232 se->ckpt_valid_blocks++; 2233 } 2234 } else { 2235 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map); 2236 #ifdef CONFIG_F2FS_CHECK_FS 2237 mir_exist = f2fs_test_and_clear_bit(offset, 2238 se->cur_valid_map_mir); 2239 if (unlikely(exist != mir_exist)) { 2240 f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d", 2241 blkaddr, exist); 2242 f2fs_bug_on(sbi, 1); 2243 } 2244 #endif 2245 if (unlikely(!exist)) { 2246 f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u", 2247 blkaddr); 2248 f2fs_bug_on(sbi, 1); 2249 se->valid_blocks++; 2250 del = 0; 2251 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 2252 /* 2253 * If checkpoints are off, we must not reuse data that 2254 * was used in the previous checkpoint. If it was used 2255 * before, we must track that to know how much space we 2256 * really have. 2257 */ 2258 if (f2fs_test_bit(offset, se->ckpt_valid_map)) { 2259 spin_lock(&sbi->stat_lock); 2260 sbi->unusable_block_count++; 2261 spin_unlock(&sbi->stat_lock); 2262 } 2263 } 2264 2265 if (f2fs_block_unit_discard(sbi) && 2266 f2fs_test_and_clear_bit(offset, se->discard_map)) 2267 sbi->discard_blks++; 2268 } 2269 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) 2270 se->ckpt_valid_blocks += del; 2271 2272 __mark_sit_entry_dirty(sbi, segno); 2273 2274 /* update total number of valid blocks to be written in ckpt area */ 2275 SIT_I(sbi)->written_valid_blocks += del; 2276 2277 if (__is_large_section(sbi)) 2278 get_sec_entry(sbi, segno)->valid_blocks += del; 2279 } 2280 2281 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) 2282 { 2283 unsigned int segno = GET_SEGNO(sbi, addr); 2284 struct sit_info *sit_i = SIT_I(sbi); 2285 2286 f2fs_bug_on(sbi, addr == NULL_ADDR); 2287 if (addr == NEW_ADDR || addr == COMPRESS_ADDR) 2288 return; 2289 2290 invalidate_mapping_pages(META_MAPPING(sbi), addr, addr); 2291 f2fs_invalidate_compress_page(sbi, addr); 2292 2293 /* add it into sit main buffer */ 2294 down_write(&sit_i->sentry_lock); 2295 2296 update_segment_mtime(sbi, addr, 0); 2297 update_sit_entry(sbi, addr, -1); 2298 2299 /* add it into dirty seglist */ 2300 locate_dirty_segment(sbi, segno); 2301 2302 up_write(&sit_i->sentry_lock); 2303 } 2304 2305 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr) 2306 { 2307 struct sit_info *sit_i = SIT_I(sbi); 2308 unsigned int segno, offset; 2309 struct seg_entry *se; 2310 bool is_cp = false; 2311 2312 if (!__is_valid_data_blkaddr(blkaddr)) 2313 return true; 2314 2315 down_read(&sit_i->sentry_lock); 2316 2317 segno = GET_SEGNO(sbi, blkaddr); 2318 se = get_seg_entry(sbi, segno); 2319 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 2320 2321 if (f2fs_test_bit(offset, se->ckpt_valid_map)) 2322 is_cp = true; 2323 2324 up_read(&sit_i->sentry_lock); 2325 2326 return is_cp; 2327 } 2328 2329 /* 2330 * This function should be resided under the curseg_mutex lock 2331 */ 2332 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, 2333 struct f2fs_summary *sum) 2334 { 2335 struct curseg_info *curseg = CURSEG_I(sbi, type); 2336 void *addr = curseg->sum_blk; 2337 2338 addr += curseg->next_blkoff * sizeof(struct f2fs_summary); 2339 memcpy(addr, sum, sizeof(struct f2fs_summary)); 2340 } 2341 2342 /* 2343 * Calculate the number of current summary pages for writing 2344 */ 2345 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra) 2346 { 2347 int valid_sum_count = 0; 2348 int i, sum_in_page; 2349 2350 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 2351 if (sbi->ckpt->alloc_type[i] == SSR) 2352 valid_sum_count += sbi->blocks_per_seg; 2353 else { 2354 if (for_ra) 2355 valid_sum_count += le16_to_cpu( 2356 F2FS_CKPT(sbi)->cur_data_blkoff[i]); 2357 else 2358 valid_sum_count += curseg_blkoff(sbi, i); 2359 } 2360 } 2361 2362 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE - 2363 SUM_FOOTER_SIZE) / SUMMARY_SIZE; 2364 if (valid_sum_count <= sum_in_page) 2365 return 1; 2366 else if ((valid_sum_count - sum_in_page) <= 2367 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE) 2368 return 2; 2369 return 3; 2370 } 2371 2372 /* 2373 * Caller should put this summary page 2374 */ 2375 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) 2376 { 2377 if (unlikely(f2fs_cp_error(sbi))) 2378 return ERR_PTR(-EIO); 2379 return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno)); 2380 } 2381 2382 void f2fs_update_meta_page(struct f2fs_sb_info *sbi, 2383 void *src, block_t blk_addr) 2384 { 2385 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2386 2387 memcpy(page_address(page), src, PAGE_SIZE); 2388 set_page_dirty(page); 2389 f2fs_put_page(page, 1); 2390 } 2391 2392 static void write_sum_page(struct f2fs_sb_info *sbi, 2393 struct f2fs_summary_block *sum_blk, block_t blk_addr) 2394 { 2395 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr); 2396 } 2397 2398 static void write_current_sum_page(struct f2fs_sb_info *sbi, 2399 int type, block_t blk_addr) 2400 { 2401 struct curseg_info *curseg = CURSEG_I(sbi, type); 2402 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 2403 struct f2fs_summary_block *src = curseg->sum_blk; 2404 struct f2fs_summary_block *dst; 2405 2406 dst = (struct f2fs_summary_block *)page_address(page); 2407 memset(dst, 0, PAGE_SIZE); 2408 2409 mutex_lock(&curseg->curseg_mutex); 2410 2411 down_read(&curseg->journal_rwsem); 2412 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE); 2413 up_read(&curseg->journal_rwsem); 2414 2415 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE); 2416 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE); 2417 2418 mutex_unlock(&curseg->curseg_mutex); 2419 2420 set_page_dirty(page); 2421 f2fs_put_page(page, 1); 2422 } 2423 2424 static int is_next_segment_free(struct f2fs_sb_info *sbi, 2425 struct curseg_info *curseg, int type) 2426 { 2427 unsigned int segno = curseg->segno + 1; 2428 struct free_segmap_info *free_i = FREE_I(sbi); 2429 2430 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec) 2431 return !test_bit(segno, free_i->free_segmap); 2432 return 0; 2433 } 2434 2435 /* 2436 * Find a new segment from the free segments bitmap to right order 2437 * This function should be returned with success, otherwise BUG 2438 */ 2439 static void get_new_segment(struct f2fs_sb_info *sbi, 2440 unsigned int *newseg, bool new_sec, int dir) 2441 { 2442 struct free_segmap_info *free_i = FREE_I(sbi); 2443 unsigned int segno, secno, zoneno; 2444 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone; 2445 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg); 2446 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg); 2447 unsigned int left_start = hint; 2448 bool init = true; 2449 int go_left = 0; 2450 int i; 2451 2452 spin_lock(&free_i->segmap_lock); 2453 2454 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { 2455 segno = find_next_zero_bit(free_i->free_segmap, 2456 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1); 2457 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1)) 2458 goto got_it; 2459 } 2460 find_other_zone: 2461 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint); 2462 if (secno >= MAIN_SECS(sbi)) { 2463 if (dir == ALLOC_RIGHT) { 2464 secno = find_first_zero_bit(free_i->free_secmap, 2465 MAIN_SECS(sbi)); 2466 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi)); 2467 } else { 2468 go_left = 1; 2469 left_start = hint - 1; 2470 } 2471 } 2472 if (go_left == 0) 2473 goto skip_left; 2474 2475 while (test_bit(left_start, free_i->free_secmap)) { 2476 if (left_start > 0) { 2477 left_start--; 2478 continue; 2479 } 2480 left_start = find_first_zero_bit(free_i->free_secmap, 2481 MAIN_SECS(sbi)); 2482 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi)); 2483 break; 2484 } 2485 secno = left_start; 2486 skip_left: 2487 segno = GET_SEG_FROM_SEC(sbi, secno); 2488 zoneno = GET_ZONE_FROM_SEC(sbi, secno); 2489 2490 /* give up on finding another zone */ 2491 if (!init) 2492 goto got_it; 2493 if (sbi->secs_per_zone == 1) 2494 goto got_it; 2495 if (zoneno == old_zoneno) 2496 goto got_it; 2497 if (dir == ALLOC_LEFT) { 2498 if (!go_left && zoneno + 1 >= total_zones) 2499 goto got_it; 2500 if (go_left && zoneno == 0) 2501 goto got_it; 2502 } 2503 for (i = 0; i < NR_CURSEG_TYPE; i++) 2504 if (CURSEG_I(sbi, i)->zone == zoneno) 2505 break; 2506 2507 if (i < NR_CURSEG_TYPE) { 2508 /* zone is in user, try another */ 2509 if (go_left) 2510 hint = zoneno * sbi->secs_per_zone - 1; 2511 else if (zoneno + 1 >= total_zones) 2512 hint = 0; 2513 else 2514 hint = (zoneno + 1) * sbi->secs_per_zone; 2515 init = false; 2516 goto find_other_zone; 2517 } 2518 got_it: 2519 /* set it as dirty segment in free segmap */ 2520 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap)); 2521 __set_inuse(sbi, segno); 2522 *newseg = segno; 2523 spin_unlock(&free_i->segmap_lock); 2524 } 2525 2526 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 2527 { 2528 struct curseg_info *curseg = CURSEG_I(sbi, type); 2529 struct summary_footer *sum_footer; 2530 unsigned short seg_type = curseg->seg_type; 2531 2532 curseg->inited = true; 2533 curseg->segno = curseg->next_segno; 2534 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno); 2535 curseg->next_blkoff = 0; 2536 curseg->next_segno = NULL_SEGNO; 2537 2538 sum_footer = &(curseg->sum_blk->footer); 2539 memset(sum_footer, 0, sizeof(struct summary_footer)); 2540 2541 sanity_check_seg_type(sbi, seg_type); 2542 2543 if (IS_DATASEG(seg_type)) 2544 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 2545 if (IS_NODESEG(seg_type)) 2546 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 2547 __set_sit_entry_type(sbi, seg_type, curseg->segno, modified); 2548 } 2549 2550 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type) 2551 { 2552 struct curseg_info *curseg = CURSEG_I(sbi, type); 2553 unsigned short seg_type = curseg->seg_type; 2554 2555 sanity_check_seg_type(sbi, seg_type); 2556 if (f2fs_need_rand_seg(sbi)) 2557 return prandom_u32_max(MAIN_SECS(sbi) * sbi->segs_per_sec); 2558 2559 /* if segs_per_sec is large than 1, we need to keep original policy. */ 2560 if (__is_large_section(sbi)) 2561 return curseg->segno; 2562 2563 /* inmem log may not locate on any segment after mount */ 2564 if (!curseg->inited) 2565 return 0; 2566 2567 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 2568 return 0; 2569 2570 if (test_opt(sbi, NOHEAP) && 2571 (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type))) 2572 return 0; 2573 2574 if (SIT_I(sbi)->last_victim[ALLOC_NEXT]) 2575 return SIT_I(sbi)->last_victim[ALLOC_NEXT]; 2576 2577 /* find segments from 0 to reuse freed segments */ 2578 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE) 2579 return 0; 2580 2581 return curseg->segno; 2582 } 2583 2584 /* 2585 * Allocate a current working segment. 2586 * This function always allocates a free segment in LFS manner. 2587 */ 2588 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 2589 { 2590 struct curseg_info *curseg = CURSEG_I(sbi, type); 2591 unsigned short seg_type = curseg->seg_type; 2592 unsigned int segno = curseg->segno; 2593 int dir = ALLOC_LEFT; 2594 2595 if (curseg->inited) 2596 write_sum_page(sbi, curseg->sum_blk, 2597 GET_SUM_BLOCK(sbi, segno)); 2598 if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA) 2599 dir = ALLOC_RIGHT; 2600 2601 if (test_opt(sbi, NOHEAP)) 2602 dir = ALLOC_RIGHT; 2603 2604 segno = __get_next_segno(sbi, type); 2605 get_new_segment(sbi, &segno, new_sec, dir); 2606 curseg->next_segno = segno; 2607 reset_curseg(sbi, type, 1); 2608 curseg->alloc_type = LFS; 2609 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) 2610 curseg->fragment_remained_chunk = 2611 prandom_u32_max(sbi->max_fragment_chunk) + 1; 2612 } 2613 2614 static int __next_free_blkoff(struct f2fs_sb_info *sbi, 2615 int segno, block_t start) 2616 { 2617 struct seg_entry *se = get_seg_entry(sbi, segno); 2618 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long); 2619 unsigned long *target_map = SIT_I(sbi)->tmp_map; 2620 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map; 2621 unsigned long *cur_map = (unsigned long *)se->cur_valid_map; 2622 int i; 2623 2624 for (i = 0; i < entries; i++) 2625 target_map[i] = ckpt_map[i] | cur_map[i]; 2626 2627 return __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start); 2628 } 2629 2630 /* 2631 * If a segment is written by LFS manner, next block offset is just obtained 2632 * by increasing the current block offset. However, if a segment is written by 2633 * SSR manner, next block offset obtained by calling __next_free_blkoff 2634 */ 2635 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, 2636 struct curseg_info *seg) 2637 { 2638 if (seg->alloc_type == SSR) { 2639 seg->next_blkoff = 2640 __next_free_blkoff(sbi, seg->segno, 2641 seg->next_blkoff + 1); 2642 } else { 2643 seg->next_blkoff++; 2644 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) { 2645 /* To allocate block chunks in different sizes, use random number */ 2646 if (--seg->fragment_remained_chunk <= 0) { 2647 seg->fragment_remained_chunk = 2648 prandom_u32_max(sbi->max_fragment_chunk) + 1; 2649 seg->next_blkoff += 2650 prandom_u32_max(sbi->max_fragment_hole) + 1; 2651 } 2652 } 2653 } 2654 } 2655 2656 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno) 2657 { 2658 return __next_free_blkoff(sbi, segno, 0) < sbi->blocks_per_seg; 2659 } 2660 2661 /* 2662 * This function always allocates a used segment(from dirty seglist) by SSR 2663 * manner, so it should recover the existing segment information of valid blocks 2664 */ 2665 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool flush) 2666 { 2667 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 2668 struct curseg_info *curseg = CURSEG_I(sbi, type); 2669 unsigned int new_segno = curseg->next_segno; 2670 struct f2fs_summary_block *sum_node; 2671 struct page *sum_page; 2672 2673 if (flush) 2674 write_sum_page(sbi, curseg->sum_blk, 2675 GET_SUM_BLOCK(sbi, curseg->segno)); 2676 2677 __set_test_and_inuse(sbi, new_segno); 2678 2679 mutex_lock(&dirty_i->seglist_lock); 2680 __remove_dirty_segment(sbi, new_segno, PRE); 2681 __remove_dirty_segment(sbi, new_segno, DIRTY); 2682 mutex_unlock(&dirty_i->seglist_lock); 2683 2684 reset_curseg(sbi, type, 1); 2685 curseg->alloc_type = SSR; 2686 curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0); 2687 2688 sum_page = f2fs_get_sum_page(sbi, new_segno); 2689 if (IS_ERR(sum_page)) { 2690 /* GC won't be able to use stale summary pages by cp_error */ 2691 memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE); 2692 return; 2693 } 2694 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 2695 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); 2696 f2fs_put_page(sum_page, 1); 2697 } 2698 2699 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type, 2700 int alloc_mode, unsigned long long age); 2701 2702 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type, 2703 int target_type, int alloc_mode, 2704 unsigned long long age) 2705 { 2706 struct curseg_info *curseg = CURSEG_I(sbi, type); 2707 2708 curseg->seg_type = target_type; 2709 2710 if (get_ssr_segment(sbi, type, alloc_mode, age)) { 2711 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno); 2712 2713 curseg->seg_type = se->type; 2714 change_curseg(sbi, type, true); 2715 } else { 2716 /* allocate cold segment by default */ 2717 curseg->seg_type = CURSEG_COLD_DATA; 2718 new_curseg(sbi, type, true); 2719 } 2720 stat_inc_seg_type(sbi, curseg); 2721 } 2722 2723 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi) 2724 { 2725 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC); 2726 2727 if (!sbi->am.atgc_enabled) 2728 return; 2729 2730 f2fs_down_read(&SM_I(sbi)->curseg_lock); 2731 2732 mutex_lock(&curseg->curseg_mutex); 2733 down_write(&SIT_I(sbi)->sentry_lock); 2734 2735 get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0); 2736 2737 up_write(&SIT_I(sbi)->sentry_lock); 2738 mutex_unlock(&curseg->curseg_mutex); 2739 2740 f2fs_up_read(&SM_I(sbi)->curseg_lock); 2741 2742 } 2743 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi) 2744 { 2745 __f2fs_init_atgc_curseg(sbi); 2746 } 2747 2748 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type) 2749 { 2750 struct curseg_info *curseg = CURSEG_I(sbi, type); 2751 2752 mutex_lock(&curseg->curseg_mutex); 2753 if (!curseg->inited) 2754 goto out; 2755 2756 if (get_valid_blocks(sbi, curseg->segno, false)) { 2757 write_sum_page(sbi, curseg->sum_blk, 2758 GET_SUM_BLOCK(sbi, curseg->segno)); 2759 } else { 2760 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 2761 __set_test_and_free(sbi, curseg->segno, true); 2762 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 2763 } 2764 out: 2765 mutex_unlock(&curseg->curseg_mutex); 2766 } 2767 2768 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi) 2769 { 2770 __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED); 2771 2772 if (sbi->am.atgc_enabled) 2773 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC); 2774 } 2775 2776 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type) 2777 { 2778 struct curseg_info *curseg = CURSEG_I(sbi, type); 2779 2780 mutex_lock(&curseg->curseg_mutex); 2781 if (!curseg->inited) 2782 goto out; 2783 if (get_valid_blocks(sbi, curseg->segno, false)) 2784 goto out; 2785 2786 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 2787 __set_test_and_inuse(sbi, curseg->segno); 2788 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 2789 out: 2790 mutex_unlock(&curseg->curseg_mutex); 2791 } 2792 2793 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi) 2794 { 2795 __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED); 2796 2797 if (sbi->am.atgc_enabled) 2798 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC); 2799 } 2800 2801 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type, 2802 int alloc_mode, unsigned long long age) 2803 { 2804 struct curseg_info *curseg = CURSEG_I(sbi, type); 2805 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; 2806 unsigned segno = NULL_SEGNO; 2807 unsigned short seg_type = curseg->seg_type; 2808 int i, cnt; 2809 bool reversed = false; 2810 2811 sanity_check_seg_type(sbi, seg_type); 2812 2813 /* f2fs_need_SSR() already forces to do this */ 2814 if (!v_ops->get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) { 2815 curseg->next_segno = segno; 2816 return 1; 2817 } 2818 2819 /* For node segments, let's do SSR more intensively */ 2820 if (IS_NODESEG(seg_type)) { 2821 if (seg_type >= CURSEG_WARM_NODE) { 2822 reversed = true; 2823 i = CURSEG_COLD_NODE; 2824 } else { 2825 i = CURSEG_HOT_NODE; 2826 } 2827 cnt = NR_CURSEG_NODE_TYPE; 2828 } else { 2829 if (seg_type >= CURSEG_WARM_DATA) { 2830 reversed = true; 2831 i = CURSEG_COLD_DATA; 2832 } else { 2833 i = CURSEG_HOT_DATA; 2834 } 2835 cnt = NR_CURSEG_DATA_TYPE; 2836 } 2837 2838 for (; cnt-- > 0; reversed ? i-- : i++) { 2839 if (i == seg_type) 2840 continue; 2841 if (!v_ops->get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) { 2842 curseg->next_segno = segno; 2843 return 1; 2844 } 2845 } 2846 2847 /* find valid_blocks=0 in dirty list */ 2848 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 2849 segno = get_free_segment(sbi); 2850 if (segno != NULL_SEGNO) { 2851 curseg->next_segno = segno; 2852 return 1; 2853 } 2854 } 2855 return 0; 2856 } 2857 2858 /* 2859 * flush out current segment and replace it with new segment 2860 * This function should be returned with success, otherwise BUG 2861 */ 2862 static void allocate_segment_by_default(struct f2fs_sb_info *sbi, 2863 int type, bool force) 2864 { 2865 struct curseg_info *curseg = CURSEG_I(sbi, type); 2866 2867 if (force) 2868 new_curseg(sbi, type, true); 2869 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) && 2870 curseg->seg_type == CURSEG_WARM_NODE) 2871 new_curseg(sbi, type, false); 2872 else if (curseg->alloc_type == LFS && 2873 is_next_segment_free(sbi, curseg, type) && 2874 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 2875 new_curseg(sbi, type, false); 2876 else if (f2fs_need_SSR(sbi) && 2877 get_ssr_segment(sbi, type, SSR, 0)) 2878 change_curseg(sbi, type, true); 2879 else 2880 new_curseg(sbi, type, false); 2881 2882 stat_inc_seg_type(sbi, curseg); 2883 } 2884 2885 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type, 2886 unsigned int start, unsigned int end) 2887 { 2888 struct curseg_info *curseg = CURSEG_I(sbi, type); 2889 unsigned int segno; 2890 2891 f2fs_down_read(&SM_I(sbi)->curseg_lock); 2892 mutex_lock(&curseg->curseg_mutex); 2893 down_write(&SIT_I(sbi)->sentry_lock); 2894 2895 segno = CURSEG_I(sbi, type)->segno; 2896 if (segno < start || segno > end) 2897 goto unlock; 2898 2899 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0)) 2900 change_curseg(sbi, type, true); 2901 else 2902 new_curseg(sbi, type, true); 2903 2904 stat_inc_seg_type(sbi, curseg); 2905 2906 locate_dirty_segment(sbi, segno); 2907 unlock: 2908 up_write(&SIT_I(sbi)->sentry_lock); 2909 2910 if (segno != curseg->segno) 2911 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u", 2912 type, segno, curseg->segno); 2913 2914 mutex_unlock(&curseg->curseg_mutex); 2915 f2fs_up_read(&SM_I(sbi)->curseg_lock); 2916 } 2917 2918 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type, 2919 bool new_sec, bool force) 2920 { 2921 struct curseg_info *curseg = CURSEG_I(sbi, type); 2922 unsigned int old_segno; 2923 2924 if (!curseg->inited) 2925 goto alloc; 2926 2927 if (force || curseg->next_blkoff || 2928 get_valid_blocks(sbi, curseg->segno, new_sec)) 2929 goto alloc; 2930 2931 if (!get_ckpt_valid_blocks(sbi, curseg->segno, new_sec)) 2932 return; 2933 alloc: 2934 old_segno = curseg->segno; 2935 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true); 2936 locate_dirty_segment(sbi, old_segno); 2937 } 2938 2939 static void __allocate_new_section(struct f2fs_sb_info *sbi, 2940 int type, bool force) 2941 { 2942 __allocate_new_segment(sbi, type, true, force); 2943 } 2944 2945 void f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force) 2946 { 2947 f2fs_down_read(&SM_I(sbi)->curseg_lock); 2948 down_write(&SIT_I(sbi)->sentry_lock); 2949 __allocate_new_section(sbi, type, force); 2950 up_write(&SIT_I(sbi)->sentry_lock); 2951 f2fs_up_read(&SM_I(sbi)->curseg_lock); 2952 } 2953 2954 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi) 2955 { 2956 int i; 2957 2958 f2fs_down_read(&SM_I(sbi)->curseg_lock); 2959 down_write(&SIT_I(sbi)->sentry_lock); 2960 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) 2961 __allocate_new_segment(sbi, i, false, false); 2962 up_write(&SIT_I(sbi)->sentry_lock); 2963 f2fs_up_read(&SM_I(sbi)->curseg_lock); 2964 } 2965 2966 static const struct segment_allocation default_salloc_ops = { 2967 .allocate_segment = allocate_segment_by_default, 2968 }; 2969 2970 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi, 2971 struct cp_control *cpc) 2972 { 2973 __u64 trim_start = cpc->trim_start; 2974 bool has_candidate = false; 2975 2976 down_write(&SIT_I(sbi)->sentry_lock); 2977 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) { 2978 if (add_discard_addrs(sbi, cpc, true)) { 2979 has_candidate = true; 2980 break; 2981 } 2982 } 2983 up_write(&SIT_I(sbi)->sentry_lock); 2984 2985 cpc->trim_start = trim_start; 2986 return has_candidate; 2987 } 2988 2989 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi, 2990 struct discard_policy *dpolicy, 2991 unsigned int start, unsigned int end) 2992 { 2993 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 2994 struct discard_cmd *prev_dc = NULL, *next_dc = NULL; 2995 struct rb_node **insert_p = NULL, *insert_parent = NULL; 2996 struct discard_cmd *dc; 2997 struct blk_plug plug; 2998 int issued; 2999 unsigned int trimmed = 0; 3000 3001 next: 3002 issued = 0; 3003 3004 mutex_lock(&dcc->cmd_lock); 3005 if (unlikely(dcc->rbtree_check)) 3006 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi, 3007 &dcc->root, false)); 3008 3009 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root, 3010 NULL, start, 3011 (struct rb_entry **)&prev_dc, 3012 (struct rb_entry **)&next_dc, 3013 &insert_p, &insert_parent, true, NULL); 3014 if (!dc) 3015 dc = next_dc; 3016 3017 blk_start_plug(&plug); 3018 3019 while (dc && dc->lstart <= end) { 3020 struct rb_node *node; 3021 int err = 0; 3022 3023 if (dc->len < dpolicy->granularity) 3024 goto skip; 3025 3026 if (dc->state != D_PREP) { 3027 list_move_tail(&dc->list, &dcc->fstrim_list); 3028 goto skip; 3029 } 3030 3031 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued); 3032 3033 if (issued >= dpolicy->max_requests) { 3034 start = dc->lstart + dc->len; 3035 3036 if (err) 3037 __remove_discard_cmd(sbi, dc); 3038 3039 blk_finish_plug(&plug); 3040 mutex_unlock(&dcc->cmd_lock); 3041 trimmed += __wait_all_discard_cmd(sbi, NULL); 3042 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 3043 goto next; 3044 } 3045 skip: 3046 node = rb_next(&dc->rb_node); 3047 if (err) 3048 __remove_discard_cmd(sbi, dc); 3049 dc = rb_entry_safe(node, struct discard_cmd, rb_node); 3050 3051 if (fatal_signal_pending(current)) 3052 break; 3053 } 3054 3055 blk_finish_plug(&plug); 3056 mutex_unlock(&dcc->cmd_lock); 3057 3058 return trimmed; 3059 } 3060 3061 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range) 3062 { 3063 __u64 start = F2FS_BYTES_TO_BLK(range->start); 3064 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1; 3065 unsigned int start_segno, end_segno; 3066 block_t start_block, end_block; 3067 struct cp_control cpc; 3068 struct discard_policy dpolicy; 3069 unsigned long long trimmed = 0; 3070 int err = 0; 3071 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi); 3072 3073 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize) 3074 return -EINVAL; 3075 3076 if (end < MAIN_BLKADDR(sbi)) 3077 goto out; 3078 3079 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 3080 f2fs_warn(sbi, "Found FS corruption, run fsck to fix."); 3081 return -EFSCORRUPTED; 3082 } 3083 3084 /* start/end segment number in main_area */ 3085 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start); 3086 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 : 3087 GET_SEGNO(sbi, end); 3088 if (need_align) { 3089 start_segno = rounddown(start_segno, sbi->segs_per_sec); 3090 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1; 3091 } 3092 3093 cpc.reason = CP_DISCARD; 3094 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen)); 3095 cpc.trim_start = start_segno; 3096 cpc.trim_end = end_segno; 3097 3098 if (sbi->discard_blks == 0) 3099 goto out; 3100 3101 f2fs_down_write(&sbi->gc_lock); 3102 err = f2fs_write_checkpoint(sbi, &cpc); 3103 f2fs_up_write(&sbi->gc_lock); 3104 if (err) 3105 goto out; 3106 3107 /* 3108 * We filed discard candidates, but actually we don't need to wait for 3109 * all of them, since they'll be issued in idle time along with runtime 3110 * discard option. User configuration looks like using runtime discard 3111 * or periodic fstrim instead of it. 3112 */ 3113 if (f2fs_realtime_discard_enable(sbi)) 3114 goto out; 3115 3116 start_block = START_BLOCK(sbi, start_segno); 3117 end_block = START_BLOCK(sbi, end_segno + 1); 3118 3119 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen); 3120 trimmed = __issue_discard_cmd_range(sbi, &dpolicy, 3121 start_block, end_block); 3122 3123 trimmed += __wait_discard_cmd_range(sbi, &dpolicy, 3124 start_block, end_block); 3125 out: 3126 if (!err) 3127 range->len = F2FS_BLK_TO_BYTES(trimmed); 3128 return err; 3129 } 3130 3131 static bool __has_curseg_space(struct f2fs_sb_info *sbi, 3132 struct curseg_info *curseg) 3133 { 3134 return curseg->next_blkoff < f2fs_usable_blks_in_seg(sbi, 3135 curseg->segno); 3136 } 3137 3138 int f2fs_rw_hint_to_seg_type(enum rw_hint hint) 3139 { 3140 switch (hint) { 3141 case WRITE_LIFE_SHORT: 3142 return CURSEG_HOT_DATA; 3143 case WRITE_LIFE_EXTREME: 3144 return CURSEG_COLD_DATA; 3145 default: 3146 return CURSEG_WARM_DATA; 3147 } 3148 } 3149 3150 static int __get_segment_type_2(struct f2fs_io_info *fio) 3151 { 3152 if (fio->type == DATA) 3153 return CURSEG_HOT_DATA; 3154 else 3155 return CURSEG_HOT_NODE; 3156 } 3157 3158 static int __get_segment_type_4(struct f2fs_io_info *fio) 3159 { 3160 if (fio->type == DATA) { 3161 struct inode *inode = fio->page->mapping->host; 3162 3163 if (S_ISDIR(inode->i_mode)) 3164 return CURSEG_HOT_DATA; 3165 else 3166 return CURSEG_COLD_DATA; 3167 } else { 3168 if (IS_DNODE(fio->page) && is_cold_node(fio->page)) 3169 return CURSEG_WARM_NODE; 3170 else 3171 return CURSEG_COLD_NODE; 3172 } 3173 } 3174 3175 static int __get_segment_type_6(struct f2fs_io_info *fio) 3176 { 3177 if (fio->type == DATA) { 3178 struct inode *inode = fio->page->mapping->host; 3179 3180 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE)) 3181 return CURSEG_COLD_DATA_PINNED; 3182 3183 if (page_private_gcing(fio->page)) { 3184 if (fio->sbi->am.atgc_enabled && 3185 (fio->io_type == FS_DATA_IO) && 3186 (fio->sbi->gc_mode != GC_URGENT_HIGH)) 3187 return CURSEG_ALL_DATA_ATGC; 3188 else 3189 return CURSEG_COLD_DATA; 3190 } 3191 if (file_is_cold(inode) || f2fs_need_compress_data(inode)) 3192 return CURSEG_COLD_DATA; 3193 if (file_is_hot(inode) || 3194 is_inode_flag_set(inode, FI_HOT_DATA) || 3195 f2fs_is_cow_file(inode)) 3196 return CURSEG_HOT_DATA; 3197 return f2fs_rw_hint_to_seg_type(inode->i_write_hint); 3198 } else { 3199 if (IS_DNODE(fio->page)) 3200 return is_cold_node(fio->page) ? CURSEG_WARM_NODE : 3201 CURSEG_HOT_NODE; 3202 return CURSEG_COLD_NODE; 3203 } 3204 } 3205 3206 static int __get_segment_type(struct f2fs_io_info *fio) 3207 { 3208 int type = 0; 3209 3210 switch (F2FS_OPTION(fio->sbi).active_logs) { 3211 case 2: 3212 type = __get_segment_type_2(fio); 3213 break; 3214 case 4: 3215 type = __get_segment_type_4(fio); 3216 break; 3217 case 6: 3218 type = __get_segment_type_6(fio); 3219 break; 3220 default: 3221 f2fs_bug_on(fio->sbi, true); 3222 } 3223 3224 if (IS_HOT(type)) 3225 fio->temp = HOT; 3226 else if (IS_WARM(type)) 3227 fio->temp = WARM; 3228 else 3229 fio->temp = COLD; 3230 return type; 3231 } 3232 3233 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, 3234 block_t old_blkaddr, block_t *new_blkaddr, 3235 struct f2fs_summary *sum, int type, 3236 struct f2fs_io_info *fio) 3237 { 3238 struct sit_info *sit_i = SIT_I(sbi); 3239 struct curseg_info *curseg = CURSEG_I(sbi, type); 3240 unsigned long long old_mtime; 3241 bool from_gc = (type == CURSEG_ALL_DATA_ATGC); 3242 struct seg_entry *se = NULL; 3243 3244 f2fs_down_read(&SM_I(sbi)->curseg_lock); 3245 3246 mutex_lock(&curseg->curseg_mutex); 3247 down_write(&sit_i->sentry_lock); 3248 3249 if (from_gc) { 3250 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO); 3251 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr)); 3252 sanity_check_seg_type(sbi, se->type); 3253 f2fs_bug_on(sbi, IS_NODESEG(se->type)); 3254 } 3255 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 3256 3257 f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg); 3258 3259 f2fs_wait_discard_bio(sbi, *new_blkaddr); 3260 3261 /* 3262 * __add_sum_entry should be resided under the curseg_mutex 3263 * because, this function updates a summary entry in the 3264 * current summary block. 3265 */ 3266 __add_sum_entry(sbi, type, sum); 3267 3268 __refresh_next_blkoff(sbi, curseg); 3269 3270 stat_inc_block_count(sbi, curseg); 3271 3272 if (from_gc) { 3273 old_mtime = get_segment_mtime(sbi, old_blkaddr); 3274 } else { 3275 update_segment_mtime(sbi, old_blkaddr, 0); 3276 old_mtime = 0; 3277 } 3278 update_segment_mtime(sbi, *new_blkaddr, old_mtime); 3279 3280 /* 3281 * SIT information should be updated before segment allocation, 3282 * since SSR needs latest valid block information. 3283 */ 3284 update_sit_entry(sbi, *new_blkaddr, 1); 3285 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 3286 update_sit_entry(sbi, old_blkaddr, -1); 3287 3288 if (!__has_curseg_space(sbi, curseg)) { 3289 if (from_gc) 3290 get_atssr_segment(sbi, type, se->type, 3291 AT_SSR, se->mtime); 3292 else 3293 sit_i->s_ops->allocate_segment(sbi, type, false); 3294 } 3295 /* 3296 * segment dirty status should be updated after segment allocation, 3297 * so we just need to update status only one time after previous 3298 * segment being closed. 3299 */ 3300 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3301 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr)); 3302 3303 up_write(&sit_i->sentry_lock); 3304 3305 if (page && IS_NODESEG(type)) { 3306 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); 3307 3308 f2fs_inode_chksum_set(sbi, page); 3309 } 3310 3311 if (fio) { 3312 struct f2fs_bio_info *io; 3313 3314 if (F2FS_IO_ALIGNED(sbi)) 3315 fio->retry = false; 3316 3317 INIT_LIST_HEAD(&fio->list); 3318 fio->in_list = true; 3319 io = sbi->write_io[fio->type] + fio->temp; 3320 spin_lock(&io->io_lock); 3321 list_add_tail(&fio->list, &io->io_list); 3322 spin_unlock(&io->io_lock); 3323 } 3324 3325 mutex_unlock(&curseg->curseg_mutex); 3326 3327 f2fs_up_read(&SM_I(sbi)->curseg_lock); 3328 } 3329 3330 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino, 3331 block_t blkaddr, unsigned int blkcnt) 3332 { 3333 if (!f2fs_is_multi_device(sbi)) 3334 return; 3335 3336 while (1) { 3337 unsigned int devidx = f2fs_target_device_index(sbi, blkaddr); 3338 unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1; 3339 3340 /* update device state for fsync */ 3341 f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO); 3342 3343 /* update device state for checkpoint */ 3344 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) { 3345 spin_lock(&sbi->dev_lock); 3346 f2fs_set_bit(devidx, (char *)&sbi->dirty_device); 3347 spin_unlock(&sbi->dev_lock); 3348 } 3349 3350 if (blkcnt <= blks) 3351 break; 3352 blkcnt -= blks; 3353 blkaddr += blks; 3354 } 3355 } 3356 3357 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio) 3358 { 3359 int type = __get_segment_type(fio); 3360 bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA); 3361 3362 if (keep_order) 3363 f2fs_down_read(&fio->sbi->io_order_lock); 3364 reallocate: 3365 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr, 3366 &fio->new_blkaddr, sum, type, fio); 3367 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO) { 3368 invalidate_mapping_pages(META_MAPPING(fio->sbi), 3369 fio->old_blkaddr, fio->old_blkaddr); 3370 f2fs_invalidate_compress_page(fio->sbi, fio->old_blkaddr); 3371 } 3372 3373 /* writeout dirty page into bdev */ 3374 f2fs_submit_page_write(fio); 3375 if (fio->retry) { 3376 fio->old_blkaddr = fio->new_blkaddr; 3377 goto reallocate; 3378 } 3379 3380 f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1); 3381 3382 if (keep_order) 3383 f2fs_up_read(&fio->sbi->io_order_lock); 3384 } 3385 3386 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page, 3387 enum iostat_type io_type) 3388 { 3389 struct f2fs_io_info fio = { 3390 .sbi = sbi, 3391 .type = META, 3392 .temp = HOT, 3393 .op = REQ_OP_WRITE, 3394 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO, 3395 .old_blkaddr = page->index, 3396 .new_blkaddr = page->index, 3397 .page = page, 3398 .encrypted_page = NULL, 3399 .in_list = false, 3400 }; 3401 3402 if (unlikely(page->index >= MAIN_BLKADDR(sbi))) 3403 fio.op_flags &= ~REQ_META; 3404 3405 set_page_writeback(page); 3406 ClearPageError(page); 3407 f2fs_submit_page_write(&fio); 3408 3409 stat_inc_meta_count(sbi, page->index); 3410 f2fs_update_iostat(sbi, NULL, io_type, F2FS_BLKSIZE); 3411 } 3412 3413 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio) 3414 { 3415 struct f2fs_summary sum; 3416 3417 set_summary(&sum, nid, 0, 0); 3418 do_write_page(&sum, fio); 3419 3420 f2fs_update_iostat(fio->sbi, NULL, fio->io_type, F2FS_BLKSIZE); 3421 } 3422 3423 void f2fs_outplace_write_data(struct dnode_of_data *dn, 3424 struct f2fs_io_info *fio) 3425 { 3426 struct f2fs_sb_info *sbi = fio->sbi; 3427 struct f2fs_summary sum; 3428 3429 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR); 3430 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version); 3431 do_write_page(&sum, fio); 3432 f2fs_update_data_blkaddr(dn, fio->new_blkaddr); 3433 3434 f2fs_update_iostat(sbi, dn->inode, fio->io_type, F2FS_BLKSIZE); 3435 } 3436 3437 int f2fs_inplace_write_data(struct f2fs_io_info *fio) 3438 { 3439 int err; 3440 struct f2fs_sb_info *sbi = fio->sbi; 3441 unsigned int segno; 3442 3443 fio->new_blkaddr = fio->old_blkaddr; 3444 /* i/o temperature is needed for passing down write hints */ 3445 __get_segment_type(fio); 3446 3447 segno = GET_SEGNO(sbi, fio->new_blkaddr); 3448 3449 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) { 3450 set_sbi_flag(sbi, SBI_NEED_FSCK); 3451 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.", 3452 __func__, segno); 3453 err = -EFSCORRUPTED; 3454 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE); 3455 goto drop_bio; 3456 } 3457 3458 if (f2fs_cp_error(sbi)) { 3459 err = -EIO; 3460 goto drop_bio; 3461 } 3462 3463 if (fio->post_read) 3464 invalidate_mapping_pages(META_MAPPING(sbi), 3465 fio->new_blkaddr, fio->new_blkaddr); 3466 3467 stat_inc_inplace_blocks(fio->sbi); 3468 3469 if (fio->bio && !(SM_I(sbi)->ipu_policy & (1 << F2FS_IPU_NOCACHE))) 3470 err = f2fs_merge_page_bio(fio); 3471 else 3472 err = f2fs_submit_page_bio(fio); 3473 if (!err) { 3474 f2fs_update_device_state(fio->sbi, fio->ino, 3475 fio->new_blkaddr, 1); 3476 f2fs_update_iostat(fio->sbi, fio->page->mapping->host, 3477 fio->io_type, F2FS_BLKSIZE); 3478 } 3479 3480 return err; 3481 drop_bio: 3482 if (fio->bio && *(fio->bio)) { 3483 struct bio *bio = *(fio->bio); 3484 3485 bio->bi_status = BLK_STS_IOERR; 3486 bio_endio(bio); 3487 *(fio->bio) = NULL; 3488 } 3489 return err; 3490 } 3491 3492 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi, 3493 unsigned int segno) 3494 { 3495 int i; 3496 3497 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 3498 if (CURSEG_I(sbi, i)->segno == segno) 3499 break; 3500 } 3501 return i; 3502 } 3503 3504 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 3505 block_t old_blkaddr, block_t new_blkaddr, 3506 bool recover_curseg, bool recover_newaddr, 3507 bool from_gc) 3508 { 3509 struct sit_info *sit_i = SIT_I(sbi); 3510 struct curseg_info *curseg; 3511 unsigned int segno, old_cursegno; 3512 struct seg_entry *se; 3513 int type; 3514 unsigned short old_blkoff; 3515 unsigned char old_alloc_type; 3516 3517 segno = GET_SEGNO(sbi, new_blkaddr); 3518 se = get_seg_entry(sbi, segno); 3519 type = se->type; 3520 3521 f2fs_down_write(&SM_I(sbi)->curseg_lock); 3522 3523 if (!recover_curseg) { 3524 /* for recovery flow */ 3525 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { 3526 if (old_blkaddr == NULL_ADDR) 3527 type = CURSEG_COLD_DATA; 3528 else 3529 type = CURSEG_WARM_DATA; 3530 } 3531 } else { 3532 if (IS_CURSEG(sbi, segno)) { 3533 /* se->type is volatile as SSR allocation */ 3534 type = __f2fs_get_curseg(sbi, segno); 3535 f2fs_bug_on(sbi, type == NO_CHECK_TYPE); 3536 } else { 3537 type = CURSEG_WARM_DATA; 3538 } 3539 } 3540 3541 f2fs_bug_on(sbi, !IS_DATASEG(type)); 3542 curseg = CURSEG_I(sbi, type); 3543 3544 mutex_lock(&curseg->curseg_mutex); 3545 down_write(&sit_i->sentry_lock); 3546 3547 old_cursegno = curseg->segno; 3548 old_blkoff = curseg->next_blkoff; 3549 old_alloc_type = curseg->alloc_type; 3550 3551 /* change the current segment */ 3552 if (segno != curseg->segno) { 3553 curseg->next_segno = segno; 3554 change_curseg(sbi, type, true); 3555 } 3556 3557 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr); 3558 __add_sum_entry(sbi, type, sum); 3559 3560 if (!recover_curseg || recover_newaddr) { 3561 if (!from_gc) 3562 update_segment_mtime(sbi, new_blkaddr, 0); 3563 update_sit_entry(sbi, new_blkaddr, 1); 3564 } 3565 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) { 3566 invalidate_mapping_pages(META_MAPPING(sbi), 3567 old_blkaddr, old_blkaddr); 3568 f2fs_invalidate_compress_page(sbi, old_blkaddr); 3569 if (!from_gc) 3570 update_segment_mtime(sbi, old_blkaddr, 0); 3571 update_sit_entry(sbi, old_blkaddr, -1); 3572 } 3573 3574 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 3575 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr)); 3576 3577 locate_dirty_segment(sbi, old_cursegno); 3578 3579 if (recover_curseg) { 3580 if (old_cursegno != curseg->segno) { 3581 curseg->next_segno = old_cursegno; 3582 change_curseg(sbi, type, true); 3583 } 3584 curseg->next_blkoff = old_blkoff; 3585 curseg->alloc_type = old_alloc_type; 3586 } 3587 3588 up_write(&sit_i->sentry_lock); 3589 mutex_unlock(&curseg->curseg_mutex); 3590 f2fs_up_write(&SM_I(sbi)->curseg_lock); 3591 } 3592 3593 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn, 3594 block_t old_addr, block_t new_addr, 3595 unsigned char version, bool recover_curseg, 3596 bool recover_newaddr) 3597 { 3598 struct f2fs_summary sum; 3599 3600 set_summary(&sum, dn->nid, dn->ofs_in_node, version); 3601 3602 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr, 3603 recover_curseg, recover_newaddr, false); 3604 3605 f2fs_update_data_blkaddr(dn, new_addr); 3606 } 3607 3608 void f2fs_wait_on_page_writeback(struct page *page, 3609 enum page_type type, bool ordered, bool locked) 3610 { 3611 if (PageWriteback(page)) { 3612 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 3613 3614 /* submit cached LFS IO */ 3615 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type); 3616 /* sbumit cached IPU IO */ 3617 f2fs_submit_merged_ipu_write(sbi, NULL, page); 3618 if (ordered) { 3619 wait_on_page_writeback(page); 3620 f2fs_bug_on(sbi, locked && PageWriteback(page)); 3621 } else { 3622 wait_for_stable_page(page); 3623 } 3624 } 3625 } 3626 3627 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr) 3628 { 3629 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3630 struct page *cpage; 3631 3632 if (!f2fs_post_read_required(inode)) 3633 return; 3634 3635 if (!__is_valid_data_blkaddr(blkaddr)) 3636 return; 3637 3638 cpage = find_lock_page(META_MAPPING(sbi), blkaddr); 3639 if (cpage) { 3640 f2fs_wait_on_page_writeback(cpage, DATA, true, true); 3641 f2fs_put_page(cpage, 1); 3642 } 3643 } 3644 3645 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr, 3646 block_t len) 3647 { 3648 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3649 block_t i; 3650 3651 if (!f2fs_post_read_required(inode)) 3652 return; 3653 3654 for (i = 0; i < len; i++) 3655 f2fs_wait_on_block_writeback(inode, blkaddr + i); 3656 3657 invalidate_mapping_pages(META_MAPPING(sbi), blkaddr, blkaddr + len - 1); 3658 } 3659 3660 static int read_compacted_summaries(struct f2fs_sb_info *sbi) 3661 { 3662 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3663 struct curseg_info *seg_i; 3664 unsigned char *kaddr; 3665 struct page *page; 3666 block_t start; 3667 int i, j, offset; 3668 3669 start = start_sum_block(sbi); 3670 3671 page = f2fs_get_meta_page(sbi, start++); 3672 if (IS_ERR(page)) 3673 return PTR_ERR(page); 3674 kaddr = (unsigned char *)page_address(page); 3675 3676 /* Step 1: restore nat cache */ 3677 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 3678 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE); 3679 3680 /* Step 2: restore sit cache */ 3681 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 3682 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE); 3683 offset = 2 * SUM_JOURNAL_SIZE; 3684 3685 /* Step 3: restore summary entries */ 3686 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 3687 unsigned short blk_off; 3688 unsigned int segno; 3689 3690 seg_i = CURSEG_I(sbi, i); 3691 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 3692 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 3693 seg_i->next_segno = segno; 3694 reset_curseg(sbi, i, 0); 3695 seg_i->alloc_type = ckpt->alloc_type[i]; 3696 seg_i->next_blkoff = blk_off; 3697 3698 if (seg_i->alloc_type == SSR) 3699 blk_off = sbi->blocks_per_seg; 3700 3701 for (j = 0; j < blk_off; j++) { 3702 struct f2fs_summary *s; 3703 3704 s = (struct f2fs_summary *)(kaddr + offset); 3705 seg_i->sum_blk->entries[j] = *s; 3706 offset += SUMMARY_SIZE; 3707 if (offset + SUMMARY_SIZE <= PAGE_SIZE - 3708 SUM_FOOTER_SIZE) 3709 continue; 3710 3711 f2fs_put_page(page, 1); 3712 page = NULL; 3713 3714 page = f2fs_get_meta_page(sbi, start++); 3715 if (IS_ERR(page)) 3716 return PTR_ERR(page); 3717 kaddr = (unsigned char *)page_address(page); 3718 offset = 0; 3719 } 3720 } 3721 f2fs_put_page(page, 1); 3722 return 0; 3723 } 3724 3725 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 3726 { 3727 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 3728 struct f2fs_summary_block *sum; 3729 struct curseg_info *curseg; 3730 struct page *new; 3731 unsigned short blk_off; 3732 unsigned int segno = 0; 3733 block_t blk_addr = 0; 3734 int err = 0; 3735 3736 /* get segment number and block addr */ 3737 if (IS_DATASEG(type)) { 3738 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 3739 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 3740 CURSEG_HOT_DATA]); 3741 if (__exist_node_summaries(sbi)) 3742 blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type); 3743 else 3744 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 3745 } else { 3746 segno = le32_to_cpu(ckpt->cur_node_segno[type - 3747 CURSEG_HOT_NODE]); 3748 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 3749 CURSEG_HOT_NODE]); 3750 if (__exist_node_summaries(sbi)) 3751 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 3752 type - CURSEG_HOT_NODE); 3753 else 3754 blk_addr = GET_SUM_BLOCK(sbi, segno); 3755 } 3756 3757 new = f2fs_get_meta_page(sbi, blk_addr); 3758 if (IS_ERR(new)) 3759 return PTR_ERR(new); 3760 sum = (struct f2fs_summary_block *)page_address(new); 3761 3762 if (IS_NODESEG(type)) { 3763 if (__exist_node_summaries(sbi)) { 3764 struct f2fs_summary *ns = &sum->entries[0]; 3765 int i; 3766 3767 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { 3768 ns->version = 0; 3769 ns->ofs_in_node = 0; 3770 } 3771 } else { 3772 err = f2fs_restore_node_summary(sbi, segno, sum); 3773 if (err) 3774 goto out; 3775 } 3776 } 3777 3778 /* set uncompleted segment to curseg */ 3779 curseg = CURSEG_I(sbi, type); 3780 mutex_lock(&curseg->curseg_mutex); 3781 3782 /* update journal info */ 3783 down_write(&curseg->journal_rwsem); 3784 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE); 3785 up_write(&curseg->journal_rwsem); 3786 3787 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE); 3788 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE); 3789 curseg->next_segno = segno; 3790 reset_curseg(sbi, type, 0); 3791 curseg->alloc_type = ckpt->alloc_type[type]; 3792 curseg->next_blkoff = blk_off; 3793 mutex_unlock(&curseg->curseg_mutex); 3794 out: 3795 f2fs_put_page(new, 1); 3796 return err; 3797 } 3798 3799 static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 3800 { 3801 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal; 3802 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal; 3803 int type = CURSEG_HOT_DATA; 3804 int err; 3805 3806 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) { 3807 int npages = f2fs_npages_for_summary_flush(sbi, true); 3808 3809 if (npages >= 2) 3810 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages, 3811 META_CP, true); 3812 3813 /* restore for compacted data summary */ 3814 err = read_compacted_summaries(sbi); 3815 if (err) 3816 return err; 3817 type = CURSEG_HOT_NODE; 3818 } 3819 3820 if (__exist_node_summaries(sbi)) 3821 f2fs_ra_meta_pages(sbi, 3822 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type), 3823 NR_CURSEG_PERSIST_TYPE - type, META_CP, true); 3824 3825 for (; type <= CURSEG_COLD_NODE; type++) { 3826 err = read_normal_summaries(sbi, type); 3827 if (err) 3828 return err; 3829 } 3830 3831 /* sanity check for summary blocks */ 3832 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES || 3833 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) { 3834 f2fs_err(sbi, "invalid journal entries nats %u sits %u", 3835 nats_in_cursum(nat_j), sits_in_cursum(sit_j)); 3836 return -EINVAL; 3837 } 3838 3839 return 0; 3840 } 3841 3842 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 3843 { 3844 struct page *page; 3845 unsigned char *kaddr; 3846 struct f2fs_summary *summary; 3847 struct curseg_info *seg_i; 3848 int written_size = 0; 3849 int i, j; 3850 3851 page = f2fs_grab_meta_page(sbi, blkaddr++); 3852 kaddr = (unsigned char *)page_address(page); 3853 memset(kaddr, 0, PAGE_SIZE); 3854 3855 /* Step 1: write nat cache */ 3856 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 3857 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE); 3858 written_size += SUM_JOURNAL_SIZE; 3859 3860 /* Step 2: write sit cache */ 3861 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 3862 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE); 3863 written_size += SUM_JOURNAL_SIZE; 3864 3865 /* Step 3: write summary entries */ 3866 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 3867 unsigned short blkoff; 3868 3869 seg_i = CURSEG_I(sbi, i); 3870 if (sbi->ckpt->alloc_type[i] == SSR) 3871 blkoff = sbi->blocks_per_seg; 3872 else 3873 blkoff = curseg_blkoff(sbi, i); 3874 3875 for (j = 0; j < blkoff; j++) { 3876 if (!page) { 3877 page = f2fs_grab_meta_page(sbi, blkaddr++); 3878 kaddr = (unsigned char *)page_address(page); 3879 memset(kaddr, 0, PAGE_SIZE); 3880 written_size = 0; 3881 } 3882 summary = (struct f2fs_summary *)(kaddr + written_size); 3883 *summary = seg_i->sum_blk->entries[j]; 3884 written_size += SUMMARY_SIZE; 3885 3886 if (written_size + SUMMARY_SIZE <= PAGE_SIZE - 3887 SUM_FOOTER_SIZE) 3888 continue; 3889 3890 set_page_dirty(page); 3891 f2fs_put_page(page, 1); 3892 page = NULL; 3893 } 3894 } 3895 if (page) { 3896 set_page_dirty(page); 3897 f2fs_put_page(page, 1); 3898 } 3899 } 3900 3901 static void write_normal_summaries(struct f2fs_sb_info *sbi, 3902 block_t blkaddr, int type) 3903 { 3904 int i, end; 3905 3906 if (IS_DATASEG(type)) 3907 end = type + NR_CURSEG_DATA_TYPE; 3908 else 3909 end = type + NR_CURSEG_NODE_TYPE; 3910 3911 for (i = type; i < end; i++) 3912 write_current_sum_page(sbi, i, blkaddr + (i - type)); 3913 } 3914 3915 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 3916 { 3917 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) 3918 write_compacted_summaries(sbi, start_blk); 3919 else 3920 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 3921 } 3922 3923 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 3924 { 3925 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 3926 } 3927 3928 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type, 3929 unsigned int val, int alloc) 3930 { 3931 int i; 3932 3933 if (type == NAT_JOURNAL) { 3934 for (i = 0; i < nats_in_cursum(journal); i++) { 3935 if (le32_to_cpu(nid_in_journal(journal, i)) == val) 3936 return i; 3937 } 3938 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL)) 3939 return update_nats_in_cursum(journal, 1); 3940 } else if (type == SIT_JOURNAL) { 3941 for (i = 0; i < sits_in_cursum(journal); i++) 3942 if (le32_to_cpu(segno_in_journal(journal, i)) == val) 3943 return i; 3944 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL)) 3945 return update_sits_in_cursum(journal, 1); 3946 } 3947 return -1; 3948 } 3949 3950 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, 3951 unsigned int segno) 3952 { 3953 return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno)); 3954 } 3955 3956 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, 3957 unsigned int start) 3958 { 3959 struct sit_info *sit_i = SIT_I(sbi); 3960 struct page *page; 3961 pgoff_t src_off, dst_off; 3962 3963 src_off = current_sit_addr(sbi, start); 3964 dst_off = next_sit_addr(sbi, src_off); 3965 3966 page = f2fs_grab_meta_page(sbi, dst_off); 3967 seg_info_to_sit_page(sbi, page, start); 3968 3969 set_page_dirty(page); 3970 set_to_next_sit(sit_i, start); 3971 3972 return page; 3973 } 3974 3975 static struct sit_entry_set *grab_sit_entry_set(void) 3976 { 3977 struct sit_entry_set *ses = 3978 f2fs_kmem_cache_alloc(sit_entry_set_slab, 3979 GFP_NOFS, true, NULL); 3980 3981 ses->entry_cnt = 0; 3982 INIT_LIST_HEAD(&ses->set_list); 3983 return ses; 3984 } 3985 3986 static void release_sit_entry_set(struct sit_entry_set *ses) 3987 { 3988 list_del(&ses->set_list); 3989 kmem_cache_free(sit_entry_set_slab, ses); 3990 } 3991 3992 static void adjust_sit_entry_set(struct sit_entry_set *ses, 3993 struct list_head *head) 3994 { 3995 struct sit_entry_set *next = ses; 3996 3997 if (list_is_last(&ses->set_list, head)) 3998 return; 3999 4000 list_for_each_entry_continue(next, head, set_list) 4001 if (ses->entry_cnt <= next->entry_cnt) { 4002 list_move_tail(&ses->set_list, &next->set_list); 4003 return; 4004 } 4005 4006 list_move_tail(&ses->set_list, head); 4007 } 4008 4009 static void add_sit_entry(unsigned int segno, struct list_head *head) 4010 { 4011 struct sit_entry_set *ses; 4012 unsigned int start_segno = START_SEGNO(segno); 4013 4014 list_for_each_entry(ses, head, set_list) { 4015 if (ses->start_segno == start_segno) { 4016 ses->entry_cnt++; 4017 adjust_sit_entry_set(ses, head); 4018 return; 4019 } 4020 } 4021 4022 ses = grab_sit_entry_set(); 4023 4024 ses->start_segno = start_segno; 4025 ses->entry_cnt++; 4026 list_add(&ses->set_list, head); 4027 } 4028 4029 static void add_sits_in_set(struct f2fs_sb_info *sbi) 4030 { 4031 struct f2fs_sm_info *sm_info = SM_I(sbi); 4032 struct list_head *set_list = &sm_info->sit_entry_set; 4033 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap; 4034 unsigned int segno; 4035 4036 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi)) 4037 add_sit_entry(segno, set_list); 4038 } 4039 4040 static void remove_sits_in_journal(struct f2fs_sb_info *sbi) 4041 { 4042 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4043 struct f2fs_journal *journal = curseg->journal; 4044 int i; 4045 4046 down_write(&curseg->journal_rwsem); 4047 for (i = 0; i < sits_in_cursum(journal); i++) { 4048 unsigned int segno; 4049 bool dirtied; 4050 4051 segno = le32_to_cpu(segno_in_journal(journal, i)); 4052 dirtied = __mark_sit_entry_dirty(sbi, segno); 4053 4054 if (!dirtied) 4055 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set); 4056 } 4057 update_sits_in_cursum(journal, -i); 4058 up_write(&curseg->journal_rwsem); 4059 } 4060 4061 /* 4062 * CP calls this function, which flushes SIT entries including sit_journal, 4063 * and moves prefree segs to free segs. 4064 */ 4065 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc) 4066 { 4067 struct sit_info *sit_i = SIT_I(sbi); 4068 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 4069 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4070 struct f2fs_journal *journal = curseg->journal; 4071 struct sit_entry_set *ses, *tmp; 4072 struct list_head *head = &SM_I(sbi)->sit_entry_set; 4073 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS); 4074 struct seg_entry *se; 4075 4076 down_write(&sit_i->sentry_lock); 4077 4078 if (!sit_i->dirty_sentries) 4079 goto out; 4080 4081 /* 4082 * add and account sit entries of dirty bitmap in sit entry 4083 * set temporarily 4084 */ 4085 add_sits_in_set(sbi); 4086 4087 /* 4088 * if there are no enough space in journal to store dirty sit 4089 * entries, remove all entries from journal and add and account 4090 * them in sit entry set. 4091 */ 4092 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) || 4093 !to_journal) 4094 remove_sits_in_journal(sbi); 4095 4096 /* 4097 * there are two steps to flush sit entries: 4098 * #1, flush sit entries to journal in current cold data summary block. 4099 * #2, flush sit entries to sit page. 4100 */ 4101 list_for_each_entry_safe(ses, tmp, head, set_list) { 4102 struct page *page = NULL; 4103 struct f2fs_sit_block *raw_sit = NULL; 4104 unsigned int start_segno = ses->start_segno; 4105 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK, 4106 (unsigned long)MAIN_SEGS(sbi)); 4107 unsigned int segno = start_segno; 4108 4109 if (to_journal && 4110 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL)) 4111 to_journal = false; 4112 4113 if (to_journal) { 4114 down_write(&curseg->journal_rwsem); 4115 } else { 4116 page = get_next_sit_page(sbi, start_segno); 4117 raw_sit = page_address(page); 4118 } 4119 4120 /* flush dirty sit entries in region of current sit set */ 4121 for_each_set_bit_from(segno, bitmap, end) { 4122 int offset, sit_offset; 4123 4124 se = get_seg_entry(sbi, segno); 4125 #ifdef CONFIG_F2FS_CHECK_FS 4126 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir, 4127 SIT_VBLOCK_MAP_SIZE)) 4128 f2fs_bug_on(sbi, 1); 4129 #endif 4130 4131 /* add discard candidates */ 4132 if (!(cpc->reason & CP_DISCARD)) { 4133 cpc->trim_start = segno; 4134 add_discard_addrs(sbi, cpc, false); 4135 } 4136 4137 if (to_journal) { 4138 offset = f2fs_lookup_journal_in_cursum(journal, 4139 SIT_JOURNAL, segno, 1); 4140 f2fs_bug_on(sbi, offset < 0); 4141 segno_in_journal(journal, offset) = 4142 cpu_to_le32(segno); 4143 seg_info_to_raw_sit(se, 4144 &sit_in_journal(journal, offset)); 4145 check_block_count(sbi, segno, 4146 &sit_in_journal(journal, offset)); 4147 } else { 4148 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 4149 seg_info_to_raw_sit(se, 4150 &raw_sit->entries[sit_offset]); 4151 check_block_count(sbi, segno, 4152 &raw_sit->entries[sit_offset]); 4153 } 4154 4155 __clear_bit(segno, bitmap); 4156 sit_i->dirty_sentries--; 4157 ses->entry_cnt--; 4158 } 4159 4160 if (to_journal) 4161 up_write(&curseg->journal_rwsem); 4162 else 4163 f2fs_put_page(page, 1); 4164 4165 f2fs_bug_on(sbi, ses->entry_cnt); 4166 release_sit_entry_set(ses); 4167 } 4168 4169 f2fs_bug_on(sbi, !list_empty(head)); 4170 f2fs_bug_on(sbi, sit_i->dirty_sentries); 4171 out: 4172 if (cpc->reason & CP_DISCARD) { 4173 __u64 trim_start = cpc->trim_start; 4174 4175 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) 4176 add_discard_addrs(sbi, cpc, false); 4177 4178 cpc->trim_start = trim_start; 4179 } 4180 up_write(&sit_i->sentry_lock); 4181 4182 set_prefree_as_free_segments(sbi); 4183 } 4184 4185 static int build_sit_info(struct f2fs_sb_info *sbi) 4186 { 4187 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 4188 struct sit_info *sit_i; 4189 unsigned int sit_segs, start; 4190 char *src_bitmap, *bitmap; 4191 unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size; 4192 unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0; 4193 4194 /* allocate memory for SIT information */ 4195 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL); 4196 if (!sit_i) 4197 return -ENOMEM; 4198 4199 SM_I(sbi)->sit_info = sit_i; 4200 4201 sit_i->sentries = 4202 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry), 4203 MAIN_SEGS(sbi)), 4204 GFP_KERNEL); 4205 if (!sit_i->sentries) 4206 return -ENOMEM; 4207 4208 main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4209 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size, 4210 GFP_KERNEL); 4211 if (!sit_i->dirty_sentries_bitmap) 4212 return -ENOMEM; 4213 4214 #ifdef CONFIG_F2FS_CHECK_FS 4215 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map); 4216 #else 4217 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map); 4218 #endif 4219 sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4220 if (!sit_i->bitmap) 4221 return -ENOMEM; 4222 4223 bitmap = sit_i->bitmap; 4224 4225 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4226 sit_i->sentries[start].cur_valid_map = bitmap; 4227 bitmap += SIT_VBLOCK_MAP_SIZE; 4228 4229 sit_i->sentries[start].ckpt_valid_map = bitmap; 4230 bitmap += SIT_VBLOCK_MAP_SIZE; 4231 4232 #ifdef CONFIG_F2FS_CHECK_FS 4233 sit_i->sentries[start].cur_valid_map_mir = bitmap; 4234 bitmap += SIT_VBLOCK_MAP_SIZE; 4235 #endif 4236 4237 if (discard_map) { 4238 sit_i->sentries[start].discard_map = bitmap; 4239 bitmap += SIT_VBLOCK_MAP_SIZE; 4240 } 4241 } 4242 4243 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 4244 if (!sit_i->tmp_map) 4245 return -ENOMEM; 4246 4247 if (__is_large_section(sbi)) { 4248 sit_i->sec_entries = 4249 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry), 4250 MAIN_SECS(sbi)), 4251 GFP_KERNEL); 4252 if (!sit_i->sec_entries) 4253 return -ENOMEM; 4254 } 4255 4256 /* get information related with SIT */ 4257 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 4258 4259 /* setup SIT bitmap from ckeckpoint pack */ 4260 sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 4261 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 4262 4263 sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL); 4264 if (!sit_i->sit_bitmap) 4265 return -ENOMEM; 4266 4267 #ifdef CONFIG_F2FS_CHECK_FS 4268 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, 4269 sit_bitmap_size, GFP_KERNEL); 4270 if (!sit_i->sit_bitmap_mir) 4271 return -ENOMEM; 4272 4273 sit_i->invalid_segmap = f2fs_kvzalloc(sbi, 4274 main_bitmap_size, GFP_KERNEL); 4275 if (!sit_i->invalid_segmap) 4276 return -ENOMEM; 4277 #endif 4278 4279 /* init SIT information */ 4280 sit_i->s_ops = &default_salloc_ops; 4281 4282 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 4283 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; 4284 sit_i->written_valid_blocks = 0; 4285 sit_i->bitmap_size = sit_bitmap_size; 4286 sit_i->dirty_sentries = 0; 4287 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 4288 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 4289 sit_i->mounted_time = ktime_get_boottime_seconds(); 4290 init_rwsem(&sit_i->sentry_lock); 4291 return 0; 4292 } 4293 4294 static int build_free_segmap(struct f2fs_sb_info *sbi) 4295 { 4296 struct free_segmap_info *free_i; 4297 unsigned int bitmap_size, sec_bitmap_size; 4298 4299 /* allocate memory for free segmap information */ 4300 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL); 4301 if (!free_i) 4302 return -ENOMEM; 4303 4304 SM_I(sbi)->free_info = free_i; 4305 4306 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4307 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL); 4308 if (!free_i->free_segmap) 4309 return -ENOMEM; 4310 4311 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4312 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL); 4313 if (!free_i->free_secmap) 4314 return -ENOMEM; 4315 4316 /* set all segments as dirty temporarily */ 4317 memset(free_i->free_segmap, 0xff, bitmap_size); 4318 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 4319 4320 /* init free segmap information */ 4321 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi)); 4322 free_i->free_segments = 0; 4323 free_i->free_sections = 0; 4324 spin_lock_init(&free_i->segmap_lock); 4325 return 0; 4326 } 4327 4328 static int build_curseg(struct f2fs_sb_info *sbi) 4329 { 4330 struct curseg_info *array; 4331 int i; 4332 4333 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, 4334 sizeof(*array)), GFP_KERNEL); 4335 if (!array) 4336 return -ENOMEM; 4337 4338 SM_I(sbi)->curseg_array = array; 4339 4340 for (i = 0; i < NO_CHECK_TYPE; i++) { 4341 mutex_init(&array[i].curseg_mutex); 4342 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL); 4343 if (!array[i].sum_blk) 4344 return -ENOMEM; 4345 init_rwsem(&array[i].journal_rwsem); 4346 array[i].journal = f2fs_kzalloc(sbi, 4347 sizeof(struct f2fs_journal), GFP_KERNEL); 4348 if (!array[i].journal) 4349 return -ENOMEM; 4350 if (i < NR_PERSISTENT_LOG) 4351 array[i].seg_type = CURSEG_HOT_DATA + i; 4352 else if (i == CURSEG_COLD_DATA_PINNED) 4353 array[i].seg_type = CURSEG_COLD_DATA; 4354 else if (i == CURSEG_ALL_DATA_ATGC) 4355 array[i].seg_type = CURSEG_COLD_DATA; 4356 array[i].segno = NULL_SEGNO; 4357 array[i].next_blkoff = 0; 4358 array[i].inited = false; 4359 } 4360 return restore_curseg_summaries(sbi); 4361 } 4362 4363 static int build_sit_entries(struct f2fs_sb_info *sbi) 4364 { 4365 struct sit_info *sit_i = SIT_I(sbi); 4366 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 4367 struct f2fs_journal *journal = curseg->journal; 4368 struct seg_entry *se; 4369 struct f2fs_sit_entry sit; 4370 int sit_blk_cnt = SIT_BLK_CNT(sbi); 4371 unsigned int i, start, end; 4372 unsigned int readed, start_blk = 0; 4373 int err = 0; 4374 block_t sit_valid_blocks[2] = {0, 0}; 4375 4376 do { 4377 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS, 4378 META_SIT, true); 4379 4380 start = start_blk * sit_i->sents_per_block; 4381 end = (start_blk + readed) * sit_i->sents_per_block; 4382 4383 for (; start < end && start < MAIN_SEGS(sbi); start++) { 4384 struct f2fs_sit_block *sit_blk; 4385 struct page *page; 4386 4387 se = &sit_i->sentries[start]; 4388 page = get_current_sit_page(sbi, start); 4389 if (IS_ERR(page)) 4390 return PTR_ERR(page); 4391 sit_blk = (struct f2fs_sit_block *)page_address(page); 4392 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 4393 f2fs_put_page(page, 1); 4394 4395 err = check_block_count(sbi, start, &sit); 4396 if (err) 4397 return err; 4398 seg_info_from_raw_sit(se, &sit); 4399 4400 if (se->type >= NR_PERSISTENT_LOG) { 4401 f2fs_err(sbi, "Invalid segment type: %u, segno: %u", 4402 se->type, start); 4403 f2fs_handle_error(sbi, 4404 ERROR_INCONSISTENT_SUM_TYPE); 4405 return -EFSCORRUPTED; 4406 } 4407 4408 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks; 4409 4410 if (f2fs_block_unit_discard(sbi)) { 4411 /* build discard map only one time */ 4412 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4413 memset(se->discard_map, 0xff, 4414 SIT_VBLOCK_MAP_SIZE); 4415 } else { 4416 memcpy(se->discard_map, 4417 se->cur_valid_map, 4418 SIT_VBLOCK_MAP_SIZE); 4419 sbi->discard_blks += 4420 sbi->blocks_per_seg - 4421 se->valid_blocks; 4422 } 4423 } 4424 4425 if (__is_large_section(sbi)) 4426 get_sec_entry(sbi, start)->valid_blocks += 4427 se->valid_blocks; 4428 } 4429 start_blk += readed; 4430 } while (start_blk < sit_blk_cnt); 4431 4432 down_read(&curseg->journal_rwsem); 4433 for (i = 0; i < sits_in_cursum(journal); i++) { 4434 unsigned int old_valid_blocks; 4435 4436 start = le32_to_cpu(segno_in_journal(journal, i)); 4437 if (start >= MAIN_SEGS(sbi)) { 4438 f2fs_err(sbi, "Wrong journal entry on segno %u", 4439 start); 4440 err = -EFSCORRUPTED; 4441 f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL); 4442 break; 4443 } 4444 4445 se = &sit_i->sentries[start]; 4446 sit = sit_in_journal(journal, i); 4447 4448 old_valid_blocks = se->valid_blocks; 4449 4450 sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks; 4451 4452 err = check_block_count(sbi, start, &sit); 4453 if (err) 4454 break; 4455 seg_info_from_raw_sit(se, &sit); 4456 4457 if (se->type >= NR_PERSISTENT_LOG) { 4458 f2fs_err(sbi, "Invalid segment type: %u, segno: %u", 4459 se->type, start); 4460 err = -EFSCORRUPTED; 4461 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE); 4462 break; 4463 } 4464 4465 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks; 4466 4467 if (f2fs_block_unit_discard(sbi)) { 4468 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) { 4469 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE); 4470 } else { 4471 memcpy(se->discard_map, se->cur_valid_map, 4472 SIT_VBLOCK_MAP_SIZE); 4473 sbi->discard_blks += old_valid_blocks; 4474 sbi->discard_blks -= se->valid_blocks; 4475 } 4476 } 4477 4478 if (__is_large_section(sbi)) { 4479 get_sec_entry(sbi, start)->valid_blocks += 4480 se->valid_blocks; 4481 get_sec_entry(sbi, start)->valid_blocks -= 4482 old_valid_blocks; 4483 } 4484 } 4485 up_read(&curseg->journal_rwsem); 4486 4487 if (err) 4488 return err; 4489 4490 if (sit_valid_blocks[NODE] != valid_node_count(sbi)) { 4491 f2fs_err(sbi, "SIT is corrupted node# %u vs %u", 4492 sit_valid_blocks[NODE], valid_node_count(sbi)); 4493 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT); 4494 return -EFSCORRUPTED; 4495 } 4496 4497 if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] > 4498 valid_user_blocks(sbi)) { 4499 f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u", 4500 sit_valid_blocks[DATA], sit_valid_blocks[NODE], 4501 valid_user_blocks(sbi)); 4502 f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT); 4503 return -EFSCORRUPTED; 4504 } 4505 4506 return 0; 4507 } 4508 4509 static void init_free_segmap(struct f2fs_sb_info *sbi) 4510 { 4511 unsigned int start; 4512 int type; 4513 struct seg_entry *sentry; 4514 4515 for (start = 0; start < MAIN_SEGS(sbi); start++) { 4516 if (f2fs_usable_blks_in_seg(sbi, start) == 0) 4517 continue; 4518 sentry = get_seg_entry(sbi, start); 4519 if (!sentry->valid_blocks) 4520 __set_free(sbi, start); 4521 else 4522 SIT_I(sbi)->written_valid_blocks += 4523 sentry->valid_blocks; 4524 } 4525 4526 /* set use the current segments */ 4527 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 4528 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 4529 4530 __set_test_and_inuse(sbi, curseg_t->segno); 4531 } 4532 } 4533 4534 static void init_dirty_segmap(struct f2fs_sb_info *sbi) 4535 { 4536 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4537 struct free_segmap_info *free_i = FREE_I(sbi); 4538 unsigned int segno = 0, offset = 0, secno; 4539 block_t valid_blocks, usable_blks_in_seg; 4540 4541 while (1) { 4542 /* find dirty segment based on free segmap */ 4543 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset); 4544 if (segno >= MAIN_SEGS(sbi)) 4545 break; 4546 offset = segno + 1; 4547 valid_blocks = get_valid_blocks(sbi, segno, false); 4548 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); 4549 if (valid_blocks == usable_blks_in_seg || !valid_blocks) 4550 continue; 4551 if (valid_blocks > usable_blks_in_seg) { 4552 f2fs_bug_on(sbi, 1); 4553 continue; 4554 } 4555 mutex_lock(&dirty_i->seglist_lock); 4556 __locate_dirty_segment(sbi, segno, DIRTY); 4557 mutex_unlock(&dirty_i->seglist_lock); 4558 } 4559 4560 if (!__is_large_section(sbi)) 4561 return; 4562 4563 mutex_lock(&dirty_i->seglist_lock); 4564 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 4565 valid_blocks = get_valid_blocks(sbi, segno, true); 4566 secno = GET_SEC_FROM_SEG(sbi, segno); 4567 4568 if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi)) 4569 continue; 4570 if (IS_CURSEC(sbi, secno)) 4571 continue; 4572 set_bit(secno, dirty_i->dirty_secmap); 4573 } 4574 mutex_unlock(&dirty_i->seglist_lock); 4575 } 4576 4577 static int init_victim_secmap(struct f2fs_sb_info *sbi) 4578 { 4579 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 4580 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4581 4582 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4583 if (!dirty_i->victim_secmap) 4584 return -ENOMEM; 4585 4586 dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL); 4587 if (!dirty_i->pinned_secmap) 4588 return -ENOMEM; 4589 4590 dirty_i->pinned_secmap_cnt = 0; 4591 dirty_i->enable_pin_section = true; 4592 return 0; 4593 } 4594 4595 static int build_dirty_segmap(struct f2fs_sb_info *sbi) 4596 { 4597 struct dirty_seglist_info *dirty_i; 4598 unsigned int bitmap_size, i; 4599 4600 /* allocate memory for dirty segments list information */ 4601 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info), 4602 GFP_KERNEL); 4603 if (!dirty_i) 4604 return -ENOMEM; 4605 4606 SM_I(sbi)->dirty_info = dirty_i; 4607 mutex_init(&dirty_i->seglist_lock); 4608 4609 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi)); 4610 4611 for (i = 0; i < NR_DIRTY_TYPE; i++) { 4612 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size, 4613 GFP_KERNEL); 4614 if (!dirty_i->dirty_segmap[i]) 4615 return -ENOMEM; 4616 } 4617 4618 if (__is_large_section(sbi)) { 4619 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 4620 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi, 4621 bitmap_size, GFP_KERNEL); 4622 if (!dirty_i->dirty_secmap) 4623 return -ENOMEM; 4624 } 4625 4626 init_dirty_segmap(sbi); 4627 return init_victim_secmap(sbi); 4628 } 4629 4630 static int sanity_check_curseg(struct f2fs_sb_info *sbi) 4631 { 4632 int i; 4633 4634 /* 4635 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr; 4636 * In LFS curseg, all blkaddr after .next_blkoff should be unused. 4637 */ 4638 for (i = 0; i < NR_PERSISTENT_LOG; i++) { 4639 struct curseg_info *curseg = CURSEG_I(sbi, i); 4640 struct seg_entry *se = get_seg_entry(sbi, curseg->segno); 4641 unsigned int blkofs = curseg->next_blkoff; 4642 4643 if (f2fs_sb_has_readonly(sbi) && 4644 i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE) 4645 continue; 4646 4647 sanity_check_seg_type(sbi, curseg->seg_type); 4648 4649 if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) { 4650 f2fs_err(sbi, 4651 "Current segment has invalid alloc_type:%d", 4652 curseg->alloc_type); 4653 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG); 4654 return -EFSCORRUPTED; 4655 } 4656 4657 if (f2fs_test_bit(blkofs, se->cur_valid_map)) 4658 goto out; 4659 4660 if (curseg->alloc_type == SSR) 4661 continue; 4662 4663 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) { 4664 if (!f2fs_test_bit(blkofs, se->cur_valid_map)) 4665 continue; 4666 out: 4667 f2fs_err(sbi, 4668 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u", 4669 i, curseg->segno, curseg->alloc_type, 4670 curseg->next_blkoff, blkofs); 4671 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG); 4672 return -EFSCORRUPTED; 4673 } 4674 } 4675 return 0; 4676 } 4677 4678 #ifdef CONFIG_BLK_DEV_ZONED 4679 4680 static int check_zone_write_pointer(struct f2fs_sb_info *sbi, 4681 struct f2fs_dev_info *fdev, 4682 struct blk_zone *zone) 4683 { 4684 unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno; 4685 block_t zone_block, wp_block, last_valid_block; 4686 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 4687 int i, s, b, ret; 4688 struct seg_entry *se; 4689 4690 if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4691 return 0; 4692 4693 wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block); 4694 wp_segno = GET_SEGNO(sbi, wp_block); 4695 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno); 4696 zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block); 4697 zone_segno = GET_SEGNO(sbi, zone_block); 4698 zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno); 4699 4700 if (zone_segno >= MAIN_SEGS(sbi)) 4701 return 0; 4702 4703 /* 4704 * Skip check of zones cursegs point to, since 4705 * fix_curseg_write_pointer() checks them. 4706 */ 4707 for (i = 0; i < NO_CHECK_TYPE; i++) 4708 if (zone_secno == GET_SEC_FROM_SEG(sbi, 4709 CURSEG_I(sbi, i)->segno)) 4710 return 0; 4711 4712 /* 4713 * Get last valid block of the zone. 4714 */ 4715 last_valid_block = zone_block - 1; 4716 for (s = sbi->segs_per_sec - 1; s >= 0; s--) { 4717 segno = zone_segno + s; 4718 se = get_seg_entry(sbi, segno); 4719 for (b = sbi->blocks_per_seg - 1; b >= 0; b--) 4720 if (f2fs_test_bit(b, se->cur_valid_map)) { 4721 last_valid_block = START_BLOCK(sbi, segno) + b; 4722 break; 4723 } 4724 if (last_valid_block >= zone_block) 4725 break; 4726 } 4727 4728 /* 4729 * If last valid block is beyond the write pointer, report the 4730 * inconsistency. This inconsistency does not cause write error 4731 * because the zone will not be selected for write operation until 4732 * it get discarded. Just report it. 4733 */ 4734 if (last_valid_block >= wp_block) { 4735 f2fs_notice(sbi, "Valid block beyond write pointer: " 4736 "valid block[0x%x,0x%x] wp[0x%x,0x%x]", 4737 GET_SEGNO(sbi, last_valid_block), 4738 GET_BLKOFF_FROM_SEG0(sbi, last_valid_block), 4739 wp_segno, wp_blkoff); 4740 return 0; 4741 } 4742 4743 /* 4744 * If there is no valid block in the zone and if write pointer is 4745 * not at zone start, reset the write pointer. 4746 */ 4747 if (last_valid_block + 1 == zone_block && zone->wp != zone->start) { 4748 f2fs_notice(sbi, 4749 "Zone without valid block has non-zero write " 4750 "pointer. Reset the write pointer: wp[0x%x,0x%x]", 4751 wp_segno, wp_blkoff); 4752 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block, 4753 zone->len >> log_sectors_per_block); 4754 if (ret) { 4755 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 4756 fdev->path, ret); 4757 return ret; 4758 } 4759 } 4760 4761 return 0; 4762 } 4763 4764 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi, 4765 block_t zone_blkaddr) 4766 { 4767 int i; 4768 4769 for (i = 0; i < sbi->s_ndevs; i++) { 4770 if (!bdev_is_zoned(FDEV(i).bdev)) 4771 continue; 4772 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr && 4773 zone_blkaddr <= FDEV(i).end_blk)) 4774 return &FDEV(i); 4775 } 4776 4777 return NULL; 4778 } 4779 4780 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx, 4781 void *data) 4782 { 4783 memcpy(data, zone, sizeof(struct blk_zone)); 4784 return 0; 4785 } 4786 4787 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type) 4788 { 4789 struct curseg_info *cs = CURSEG_I(sbi, type); 4790 struct f2fs_dev_info *zbd; 4791 struct blk_zone zone; 4792 unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off; 4793 block_t cs_zone_block, wp_block; 4794 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT; 4795 sector_t zone_sector; 4796 int err; 4797 4798 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 4799 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 4800 4801 zbd = get_target_zoned_dev(sbi, cs_zone_block); 4802 if (!zbd) 4803 return 0; 4804 4805 /* report zone for the sector the curseg points to */ 4806 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 4807 << log_sectors_per_block; 4808 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 4809 report_one_zone_cb, &zone); 4810 if (err != 1) { 4811 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 4812 zbd->path, err); 4813 return err; 4814 } 4815 4816 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4817 return 0; 4818 4819 wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block); 4820 wp_segno = GET_SEGNO(sbi, wp_block); 4821 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno); 4822 wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0); 4823 4824 if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff && 4825 wp_sector_off == 0) 4826 return 0; 4827 4828 f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: " 4829 "curseg[0x%x,0x%x] wp[0x%x,0x%x]", 4830 type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff); 4831 4832 f2fs_notice(sbi, "Assign new section to curseg[%d]: " 4833 "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff); 4834 4835 f2fs_allocate_new_section(sbi, type, true); 4836 4837 /* check consistency of the zone curseg pointed to */ 4838 if (check_zone_write_pointer(sbi, zbd, &zone)) 4839 return -EIO; 4840 4841 /* check newly assigned zone */ 4842 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno); 4843 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section)); 4844 4845 zbd = get_target_zoned_dev(sbi, cs_zone_block); 4846 if (!zbd) 4847 return 0; 4848 4849 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk) 4850 << log_sectors_per_block; 4851 err = blkdev_report_zones(zbd->bdev, zone_sector, 1, 4852 report_one_zone_cb, &zone); 4853 if (err != 1) { 4854 f2fs_err(sbi, "Report zone failed: %s errno=(%d)", 4855 zbd->path, err); 4856 return err; 4857 } 4858 4859 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ) 4860 return 0; 4861 4862 if (zone.wp != zone.start) { 4863 f2fs_notice(sbi, 4864 "New zone for curseg[%d] is not yet discarded. " 4865 "Reset the zone: curseg[0x%x,0x%x]", 4866 type, cs->segno, cs->next_blkoff); 4867 err = __f2fs_issue_discard_zone(sbi, zbd->bdev, 4868 zone_sector >> log_sectors_per_block, 4869 zone.len >> log_sectors_per_block); 4870 if (err) { 4871 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)", 4872 zbd->path, err); 4873 return err; 4874 } 4875 } 4876 4877 return 0; 4878 } 4879 4880 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 4881 { 4882 int i, ret; 4883 4884 for (i = 0; i < NR_PERSISTENT_LOG; i++) { 4885 ret = fix_curseg_write_pointer(sbi, i); 4886 if (ret) 4887 return ret; 4888 } 4889 4890 return 0; 4891 } 4892 4893 struct check_zone_write_pointer_args { 4894 struct f2fs_sb_info *sbi; 4895 struct f2fs_dev_info *fdev; 4896 }; 4897 4898 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx, 4899 void *data) 4900 { 4901 struct check_zone_write_pointer_args *args; 4902 4903 args = (struct check_zone_write_pointer_args *)data; 4904 4905 return check_zone_write_pointer(args->sbi, args->fdev, zone); 4906 } 4907 4908 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi) 4909 { 4910 int i, ret; 4911 struct check_zone_write_pointer_args args; 4912 4913 for (i = 0; i < sbi->s_ndevs; i++) { 4914 if (!bdev_is_zoned(FDEV(i).bdev)) 4915 continue; 4916 4917 args.sbi = sbi; 4918 args.fdev = &FDEV(i); 4919 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES, 4920 check_zone_write_pointer_cb, &args); 4921 if (ret < 0) 4922 return ret; 4923 } 4924 4925 return 0; 4926 } 4927 4928 static bool is_conv_zone(struct f2fs_sb_info *sbi, unsigned int zone_idx, 4929 unsigned int dev_idx) 4930 { 4931 if (!bdev_is_zoned(FDEV(dev_idx).bdev)) 4932 return true; 4933 return !test_bit(zone_idx, FDEV(dev_idx).blkz_seq); 4934 } 4935 4936 /* Return the zone index in the given device */ 4937 static unsigned int get_zone_idx(struct f2fs_sb_info *sbi, unsigned int secno, 4938 int dev_idx) 4939 { 4940 block_t sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno)); 4941 4942 return (sec_start_blkaddr - FDEV(dev_idx).start_blk) >> 4943 sbi->log_blocks_per_blkz; 4944 } 4945 4946 /* 4947 * Return the usable segments in a section based on the zone's 4948 * corresponding zone capacity. Zone is equal to a section. 4949 */ 4950 static inline unsigned int f2fs_usable_zone_segs_in_sec( 4951 struct f2fs_sb_info *sbi, unsigned int segno) 4952 { 4953 unsigned int dev_idx, zone_idx; 4954 4955 dev_idx = f2fs_target_device_index(sbi, START_BLOCK(sbi, segno)); 4956 zone_idx = get_zone_idx(sbi, GET_SEC_FROM_SEG(sbi, segno), dev_idx); 4957 4958 /* Conventional zone's capacity is always equal to zone size */ 4959 if (is_conv_zone(sbi, zone_idx, dev_idx)) 4960 return sbi->segs_per_sec; 4961 4962 if (!sbi->unusable_blocks_per_sec) 4963 return sbi->segs_per_sec; 4964 4965 /* Get the segment count beyond zone capacity block */ 4966 return sbi->segs_per_sec - (sbi->unusable_blocks_per_sec >> 4967 sbi->log_blocks_per_seg); 4968 } 4969 4970 /* 4971 * Return the number of usable blocks in a segment. The number of blocks 4972 * returned is always equal to the number of blocks in a segment for 4973 * segments fully contained within a sequential zone capacity or a 4974 * conventional zone. For segments partially contained in a sequential 4975 * zone capacity, the number of usable blocks up to the zone capacity 4976 * is returned. 0 is returned in all other cases. 4977 */ 4978 static inline unsigned int f2fs_usable_zone_blks_in_seg( 4979 struct f2fs_sb_info *sbi, unsigned int segno) 4980 { 4981 block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr; 4982 unsigned int zone_idx, dev_idx, secno; 4983 4984 secno = GET_SEC_FROM_SEG(sbi, segno); 4985 seg_start = START_BLOCK(sbi, segno); 4986 dev_idx = f2fs_target_device_index(sbi, seg_start); 4987 zone_idx = get_zone_idx(sbi, secno, dev_idx); 4988 4989 /* 4990 * Conventional zone's capacity is always equal to zone size, 4991 * so, blocks per segment is unchanged. 4992 */ 4993 if (is_conv_zone(sbi, zone_idx, dev_idx)) 4994 return sbi->blocks_per_seg; 4995 4996 if (!sbi->unusable_blocks_per_sec) 4997 return sbi->blocks_per_seg; 4998 4999 sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno)); 5000 sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi); 5001 5002 /* 5003 * If segment starts before zone capacity and spans beyond 5004 * zone capacity, then usable blocks are from seg start to 5005 * zone capacity. If the segment starts after the zone capacity, 5006 * then there are no usable blocks. 5007 */ 5008 if (seg_start >= sec_cap_blkaddr) 5009 return 0; 5010 if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr) 5011 return sec_cap_blkaddr - seg_start; 5012 5013 return sbi->blocks_per_seg; 5014 } 5015 #else 5016 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi) 5017 { 5018 return 0; 5019 } 5020 5021 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi) 5022 { 5023 return 0; 5024 } 5025 5026 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi, 5027 unsigned int segno) 5028 { 5029 return 0; 5030 } 5031 5032 static inline unsigned int f2fs_usable_zone_segs_in_sec(struct f2fs_sb_info *sbi, 5033 unsigned int segno) 5034 { 5035 return 0; 5036 } 5037 #endif 5038 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi, 5039 unsigned int segno) 5040 { 5041 if (f2fs_sb_has_blkzoned(sbi)) 5042 return f2fs_usable_zone_blks_in_seg(sbi, segno); 5043 5044 return sbi->blocks_per_seg; 5045 } 5046 5047 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi, 5048 unsigned int segno) 5049 { 5050 if (f2fs_sb_has_blkzoned(sbi)) 5051 return f2fs_usable_zone_segs_in_sec(sbi, segno); 5052 5053 return sbi->segs_per_sec; 5054 } 5055 5056 /* 5057 * Update min, max modified time for cost-benefit GC algorithm 5058 */ 5059 static void init_min_max_mtime(struct f2fs_sb_info *sbi) 5060 { 5061 struct sit_info *sit_i = SIT_I(sbi); 5062 unsigned int segno; 5063 5064 down_write(&sit_i->sentry_lock); 5065 5066 sit_i->min_mtime = ULLONG_MAX; 5067 5068 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) { 5069 unsigned int i; 5070 unsigned long long mtime = 0; 5071 5072 for (i = 0; i < sbi->segs_per_sec; i++) 5073 mtime += get_seg_entry(sbi, segno + i)->mtime; 5074 5075 mtime = div_u64(mtime, sbi->segs_per_sec); 5076 5077 if (sit_i->min_mtime > mtime) 5078 sit_i->min_mtime = mtime; 5079 } 5080 sit_i->max_mtime = get_mtime(sbi, false); 5081 sit_i->dirty_max_mtime = 0; 5082 up_write(&sit_i->sentry_lock); 5083 } 5084 5085 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi) 5086 { 5087 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 5088 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 5089 struct f2fs_sm_info *sm_info; 5090 int err; 5091 5092 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL); 5093 if (!sm_info) 5094 return -ENOMEM; 5095 5096 /* init sm info */ 5097 sbi->sm_info = sm_info; 5098 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 5099 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 5100 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 5101 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 5102 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 5103 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 5104 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 5105 sm_info->rec_prefree_segments = sm_info->main_segments * 5106 DEF_RECLAIM_PREFREE_SEGMENTS / 100; 5107 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS) 5108 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS; 5109 5110 if (!f2fs_lfs_mode(sbi)) 5111 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC; 5112 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL; 5113 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS; 5114 sm_info->min_seq_blocks = sbi->blocks_per_seg; 5115 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS; 5116 sm_info->min_ssr_sections = reserved_sections(sbi); 5117 5118 INIT_LIST_HEAD(&sm_info->sit_entry_set); 5119 5120 init_f2fs_rwsem(&sm_info->curseg_lock); 5121 5122 if (!f2fs_readonly(sbi->sb)) { 5123 err = f2fs_create_flush_cmd_control(sbi); 5124 if (err) 5125 return err; 5126 } 5127 5128 err = create_discard_cmd_control(sbi); 5129 if (err) 5130 return err; 5131 5132 err = build_sit_info(sbi); 5133 if (err) 5134 return err; 5135 err = build_free_segmap(sbi); 5136 if (err) 5137 return err; 5138 err = build_curseg(sbi); 5139 if (err) 5140 return err; 5141 5142 /* reinit free segmap based on SIT */ 5143 err = build_sit_entries(sbi); 5144 if (err) 5145 return err; 5146 5147 init_free_segmap(sbi); 5148 err = build_dirty_segmap(sbi); 5149 if (err) 5150 return err; 5151 5152 err = sanity_check_curseg(sbi); 5153 if (err) 5154 return err; 5155 5156 init_min_max_mtime(sbi); 5157 return 0; 5158 } 5159 5160 static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 5161 enum dirty_type dirty_type) 5162 { 5163 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5164 5165 mutex_lock(&dirty_i->seglist_lock); 5166 kvfree(dirty_i->dirty_segmap[dirty_type]); 5167 dirty_i->nr_dirty[dirty_type] = 0; 5168 mutex_unlock(&dirty_i->seglist_lock); 5169 } 5170 5171 static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 5172 { 5173 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5174 5175 kvfree(dirty_i->pinned_secmap); 5176 kvfree(dirty_i->victim_secmap); 5177 } 5178 5179 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 5180 { 5181 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 5182 int i; 5183 5184 if (!dirty_i) 5185 return; 5186 5187 /* discard pre-free/dirty segments list */ 5188 for (i = 0; i < NR_DIRTY_TYPE; i++) 5189 discard_dirty_segmap(sbi, i); 5190 5191 if (__is_large_section(sbi)) { 5192 mutex_lock(&dirty_i->seglist_lock); 5193 kvfree(dirty_i->dirty_secmap); 5194 mutex_unlock(&dirty_i->seglist_lock); 5195 } 5196 5197 destroy_victim_secmap(sbi); 5198 SM_I(sbi)->dirty_info = NULL; 5199 kfree(dirty_i); 5200 } 5201 5202 static void destroy_curseg(struct f2fs_sb_info *sbi) 5203 { 5204 struct curseg_info *array = SM_I(sbi)->curseg_array; 5205 int i; 5206 5207 if (!array) 5208 return; 5209 SM_I(sbi)->curseg_array = NULL; 5210 for (i = 0; i < NR_CURSEG_TYPE; i++) { 5211 kfree(array[i].sum_blk); 5212 kfree(array[i].journal); 5213 } 5214 kfree(array); 5215 } 5216 5217 static void destroy_free_segmap(struct f2fs_sb_info *sbi) 5218 { 5219 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 5220 5221 if (!free_i) 5222 return; 5223 SM_I(sbi)->free_info = NULL; 5224 kvfree(free_i->free_segmap); 5225 kvfree(free_i->free_secmap); 5226 kfree(free_i); 5227 } 5228 5229 static void destroy_sit_info(struct f2fs_sb_info *sbi) 5230 { 5231 struct sit_info *sit_i = SIT_I(sbi); 5232 5233 if (!sit_i) 5234 return; 5235 5236 if (sit_i->sentries) 5237 kvfree(sit_i->bitmap); 5238 kfree(sit_i->tmp_map); 5239 5240 kvfree(sit_i->sentries); 5241 kvfree(sit_i->sec_entries); 5242 kvfree(sit_i->dirty_sentries_bitmap); 5243 5244 SM_I(sbi)->sit_info = NULL; 5245 kvfree(sit_i->sit_bitmap); 5246 #ifdef CONFIG_F2FS_CHECK_FS 5247 kvfree(sit_i->sit_bitmap_mir); 5248 kvfree(sit_i->invalid_segmap); 5249 #endif 5250 kfree(sit_i); 5251 } 5252 5253 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi) 5254 { 5255 struct f2fs_sm_info *sm_info = SM_I(sbi); 5256 5257 if (!sm_info) 5258 return; 5259 f2fs_destroy_flush_cmd_control(sbi, true); 5260 destroy_discard_cmd_control(sbi); 5261 destroy_dirty_segmap(sbi); 5262 destroy_curseg(sbi); 5263 destroy_free_segmap(sbi); 5264 destroy_sit_info(sbi); 5265 sbi->sm_info = NULL; 5266 kfree(sm_info); 5267 } 5268 5269 int __init f2fs_create_segment_manager_caches(void) 5270 { 5271 discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry", 5272 sizeof(struct discard_entry)); 5273 if (!discard_entry_slab) 5274 goto fail; 5275 5276 discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd", 5277 sizeof(struct discard_cmd)); 5278 if (!discard_cmd_slab) 5279 goto destroy_discard_entry; 5280 5281 sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set", 5282 sizeof(struct sit_entry_set)); 5283 if (!sit_entry_set_slab) 5284 goto destroy_discard_cmd; 5285 5286 revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry", 5287 sizeof(struct revoke_entry)); 5288 if (!revoke_entry_slab) 5289 goto destroy_sit_entry_set; 5290 return 0; 5291 5292 destroy_sit_entry_set: 5293 kmem_cache_destroy(sit_entry_set_slab); 5294 destroy_discard_cmd: 5295 kmem_cache_destroy(discard_cmd_slab); 5296 destroy_discard_entry: 5297 kmem_cache_destroy(discard_entry_slab); 5298 fail: 5299 return -ENOMEM; 5300 } 5301 5302 void f2fs_destroy_segment_manager_caches(void) 5303 { 5304 kmem_cache_destroy(sit_entry_set_slab); 5305 kmem_cache_destroy(discard_cmd_slab); 5306 kmem_cache_destroy(discard_entry_slab); 5307 kmem_cache_destroy(revoke_entry_slab); 5308 } 5309