1 /* 2 * fs/f2fs/segment.c 3 * 4 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 5 * http://www.samsung.com/ 6 * 7 * This program is free software; you can redistribute it and/or modify 8 * it under the terms of the GNU General Public License version 2 as 9 * published by the Free Software Foundation. 10 */ 11 #include <linux/fs.h> 12 #include <linux/f2fs_fs.h> 13 #include <linux/bio.h> 14 #include <linux/blkdev.h> 15 #include <linux/prefetch.h> 16 #include <linux/vmalloc.h> 17 18 #include "f2fs.h" 19 #include "segment.h" 20 #include "node.h" 21 #include <trace/events/f2fs.h> 22 23 /* 24 * This function balances dirty node and dentry pages. 25 * In addition, it controls garbage collection. 26 */ 27 void f2fs_balance_fs(struct f2fs_sb_info *sbi) 28 { 29 /* 30 * We should do GC or end up with checkpoint, if there are so many dirty 31 * dir/node pages without enough free segments. 32 */ 33 if (has_not_enough_free_secs(sbi, 0)) { 34 mutex_lock(&sbi->gc_mutex); 35 f2fs_gc(sbi); 36 } 37 } 38 39 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 40 enum dirty_type dirty_type) 41 { 42 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 43 44 /* need not be added */ 45 if (IS_CURSEG(sbi, segno)) 46 return; 47 48 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 49 dirty_i->nr_dirty[dirty_type]++; 50 51 if (dirty_type == DIRTY) { 52 struct seg_entry *sentry = get_seg_entry(sbi, segno); 53 enum dirty_type t = DIRTY_HOT_DATA; 54 55 dirty_type = sentry->type; 56 57 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 58 dirty_i->nr_dirty[dirty_type]++; 59 60 /* Only one bitmap should be set */ 61 for (; t <= DIRTY_COLD_NODE; t++) { 62 if (t == dirty_type) 63 continue; 64 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) 65 dirty_i->nr_dirty[t]--; 66 } 67 } 68 } 69 70 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 71 enum dirty_type dirty_type) 72 { 73 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 74 75 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) 76 dirty_i->nr_dirty[dirty_type]--; 77 78 if (dirty_type == DIRTY) { 79 enum dirty_type t = DIRTY_HOT_DATA; 80 81 /* clear all the bitmaps */ 82 for (; t <= DIRTY_COLD_NODE; t++) 83 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t])) 84 dirty_i->nr_dirty[t]--; 85 86 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0) 87 clear_bit(GET_SECNO(sbi, segno), 88 dirty_i->victim_secmap); 89 } 90 } 91 92 /* 93 * Should not occur error such as -ENOMEM. 94 * Adding dirty entry into seglist is not critical operation. 95 * If a given segment is one of current working segments, it won't be added. 96 */ 97 void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) 98 { 99 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 100 unsigned short valid_blocks; 101 102 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) 103 return; 104 105 mutex_lock(&dirty_i->seglist_lock); 106 107 valid_blocks = get_valid_blocks(sbi, segno, 0); 108 109 if (valid_blocks == 0) { 110 __locate_dirty_segment(sbi, segno, PRE); 111 __remove_dirty_segment(sbi, segno, DIRTY); 112 } else if (valid_blocks < sbi->blocks_per_seg) { 113 __locate_dirty_segment(sbi, segno, DIRTY); 114 } else { 115 /* Recovery routine with SSR needs this */ 116 __remove_dirty_segment(sbi, segno, DIRTY); 117 } 118 119 mutex_unlock(&dirty_i->seglist_lock); 120 return; 121 } 122 123 /* 124 * Should call clear_prefree_segments after checkpoint is done. 125 */ 126 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 127 { 128 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 129 unsigned int segno, offset = 0; 130 unsigned int total_segs = TOTAL_SEGS(sbi); 131 132 mutex_lock(&dirty_i->seglist_lock); 133 while (1) { 134 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs, 135 offset); 136 if (segno >= total_segs) 137 break; 138 __set_test_and_free(sbi, segno); 139 offset = segno + 1; 140 } 141 mutex_unlock(&dirty_i->seglist_lock); 142 } 143 144 void clear_prefree_segments(struct f2fs_sb_info *sbi) 145 { 146 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 147 unsigned int segno, offset = 0; 148 unsigned int total_segs = TOTAL_SEGS(sbi); 149 150 mutex_lock(&dirty_i->seglist_lock); 151 while (1) { 152 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs, 153 offset); 154 if (segno >= total_segs) 155 break; 156 157 offset = segno + 1; 158 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[PRE])) 159 dirty_i->nr_dirty[PRE]--; 160 161 /* Let's use trim */ 162 if (test_opt(sbi, DISCARD)) 163 blkdev_issue_discard(sbi->sb->s_bdev, 164 START_BLOCK(sbi, segno) << 165 sbi->log_sectors_per_block, 166 1 << (sbi->log_sectors_per_block + 167 sbi->log_blocks_per_seg), 168 GFP_NOFS, 0); 169 } 170 mutex_unlock(&dirty_i->seglist_lock); 171 } 172 173 static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 174 { 175 struct sit_info *sit_i = SIT_I(sbi); 176 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) 177 sit_i->dirty_sentries++; 178 } 179 180 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 181 unsigned int segno, int modified) 182 { 183 struct seg_entry *se = get_seg_entry(sbi, segno); 184 se->type = type; 185 if (modified) 186 __mark_sit_entry_dirty(sbi, segno); 187 } 188 189 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 190 { 191 struct seg_entry *se; 192 unsigned int segno, offset; 193 long int new_vblocks; 194 195 segno = GET_SEGNO(sbi, blkaddr); 196 197 se = get_seg_entry(sbi, segno); 198 new_vblocks = se->valid_blocks + del; 199 offset = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) & (sbi->blocks_per_seg - 1); 200 201 BUG_ON((new_vblocks >> (sizeof(unsigned short) << 3) || 202 (new_vblocks > sbi->blocks_per_seg))); 203 204 se->valid_blocks = new_vblocks; 205 se->mtime = get_mtime(sbi); 206 SIT_I(sbi)->max_mtime = se->mtime; 207 208 /* Update valid block bitmap */ 209 if (del > 0) { 210 if (f2fs_set_bit(offset, se->cur_valid_map)) 211 BUG(); 212 } else { 213 if (!f2fs_clear_bit(offset, se->cur_valid_map)) 214 BUG(); 215 } 216 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) 217 se->ckpt_valid_blocks += del; 218 219 __mark_sit_entry_dirty(sbi, segno); 220 221 /* update total number of valid blocks to be written in ckpt area */ 222 SIT_I(sbi)->written_valid_blocks += del; 223 224 if (sbi->segs_per_sec > 1) 225 get_sec_entry(sbi, segno)->valid_blocks += del; 226 } 227 228 static void refresh_sit_entry(struct f2fs_sb_info *sbi, 229 block_t old_blkaddr, block_t new_blkaddr) 230 { 231 update_sit_entry(sbi, new_blkaddr, 1); 232 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 233 update_sit_entry(sbi, old_blkaddr, -1); 234 } 235 236 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) 237 { 238 unsigned int segno = GET_SEGNO(sbi, addr); 239 struct sit_info *sit_i = SIT_I(sbi); 240 241 BUG_ON(addr == NULL_ADDR); 242 if (addr == NEW_ADDR) 243 return; 244 245 /* add it into sit main buffer */ 246 mutex_lock(&sit_i->sentry_lock); 247 248 update_sit_entry(sbi, addr, -1); 249 250 /* add it into dirty seglist */ 251 locate_dirty_segment(sbi, segno); 252 253 mutex_unlock(&sit_i->sentry_lock); 254 } 255 256 /* 257 * This function should be resided under the curseg_mutex lock 258 */ 259 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, 260 struct f2fs_summary *sum, unsigned short offset) 261 { 262 struct curseg_info *curseg = CURSEG_I(sbi, type); 263 void *addr = curseg->sum_blk; 264 addr += offset * sizeof(struct f2fs_summary); 265 memcpy(addr, sum, sizeof(struct f2fs_summary)); 266 return; 267 } 268 269 /* 270 * Calculate the number of current summary pages for writing 271 */ 272 int npages_for_summary_flush(struct f2fs_sb_info *sbi) 273 { 274 int total_size_bytes = 0; 275 int valid_sum_count = 0; 276 int i, sum_space; 277 278 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 279 if (sbi->ckpt->alloc_type[i] == SSR) 280 valid_sum_count += sbi->blocks_per_seg; 281 else 282 valid_sum_count += curseg_blkoff(sbi, i); 283 } 284 285 total_size_bytes = valid_sum_count * (SUMMARY_SIZE + 1) 286 + sizeof(struct nat_journal) + 2 287 + sizeof(struct sit_journal) + 2; 288 sum_space = PAGE_CACHE_SIZE - SUM_FOOTER_SIZE; 289 if (total_size_bytes < sum_space) 290 return 1; 291 else if (total_size_bytes < 2 * sum_space) 292 return 2; 293 return 3; 294 } 295 296 /* 297 * Caller should put this summary page 298 */ 299 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) 300 { 301 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno)); 302 } 303 304 static void write_sum_page(struct f2fs_sb_info *sbi, 305 struct f2fs_summary_block *sum_blk, block_t blk_addr) 306 { 307 struct page *page = grab_meta_page(sbi, blk_addr); 308 void *kaddr = page_address(page); 309 memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE); 310 set_page_dirty(page); 311 f2fs_put_page(page, 1); 312 } 313 314 static unsigned int check_prefree_segments(struct f2fs_sb_info *sbi, int type) 315 { 316 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 317 unsigned long *prefree_segmap = dirty_i->dirty_segmap[PRE]; 318 unsigned int segno; 319 unsigned int ofs = 0; 320 321 /* 322 * If there is not enough reserved sections, 323 * we should not reuse prefree segments. 324 */ 325 if (has_not_enough_free_secs(sbi, 0)) 326 return NULL_SEGNO; 327 328 /* 329 * NODE page should not reuse prefree segment, 330 * since those information is used for SPOR. 331 */ 332 if (IS_NODESEG(type)) 333 return NULL_SEGNO; 334 next: 335 segno = find_next_bit(prefree_segmap, TOTAL_SEGS(sbi), ofs); 336 ofs += sbi->segs_per_sec; 337 338 if (segno < TOTAL_SEGS(sbi)) { 339 int i; 340 341 /* skip intermediate segments in a section */ 342 if (segno % sbi->segs_per_sec) 343 goto next; 344 345 /* skip if the section is currently used */ 346 if (sec_usage_check(sbi, GET_SECNO(sbi, segno))) 347 goto next; 348 349 /* skip if whole section is not prefree */ 350 for (i = 1; i < sbi->segs_per_sec; i++) 351 if (!test_bit(segno + i, prefree_segmap)) 352 goto next; 353 354 /* skip if whole section was not free at the last checkpoint */ 355 for (i = 0; i < sbi->segs_per_sec; i++) 356 if (get_seg_entry(sbi, segno + i)->ckpt_valid_blocks) 357 goto next; 358 359 return segno; 360 } 361 return NULL_SEGNO; 362 } 363 364 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type) 365 { 366 struct curseg_info *curseg = CURSEG_I(sbi, type); 367 unsigned int segno = curseg->segno; 368 struct free_segmap_info *free_i = FREE_I(sbi); 369 370 if (segno + 1 < TOTAL_SEGS(sbi) && (segno + 1) % sbi->segs_per_sec) 371 return !test_bit(segno + 1, free_i->free_segmap); 372 return 0; 373 } 374 375 /* 376 * Find a new segment from the free segments bitmap to right order 377 * This function should be returned with success, otherwise BUG 378 */ 379 static void get_new_segment(struct f2fs_sb_info *sbi, 380 unsigned int *newseg, bool new_sec, int dir) 381 { 382 struct free_segmap_info *free_i = FREE_I(sbi); 383 unsigned int segno, secno, zoneno; 384 unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone; 385 unsigned int hint = *newseg / sbi->segs_per_sec; 386 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg); 387 unsigned int left_start = hint; 388 bool init = true; 389 int go_left = 0; 390 int i; 391 392 write_lock(&free_i->segmap_lock); 393 394 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { 395 segno = find_next_zero_bit(free_i->free_segmap, 396 TOTAL_SEGS(sbi), *newseg + 1); 397 if (segno - *newseg < sbi->segs_per_sec - 398 (*newseg % sbi->segs_per_sec)) 399 goto got_it; 400 } 401 find_other_zone: 402 secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint); 403 if (secno >= TOTAL_SECS(sbi)) { 404 if (dir == ALLOC_RIGHT) { 405 secno = find_next_zero_bit(free_i->free_secmap, 406 TOTAL_SECS(sbi), 0); 407 BUG_ON(secno >= TOTAL_SECS(sbi)); 408 } else { 409 go_left = 1; 410 left_start = hint - 1; 411 } 412 } 413 if (go_left == 0) 414 goto skip_left; 415 416 while (test_bit(left_start, free_i->free_secmap)) { 417 if (left_start > 0) { 418 left_start--; 419 continue; 420 } 421 left_start = find_next_zero_bit(free_i->free_secmap, 422 TOTAL_SECS(sbi), 0); 423 BUG_ON(left_start >= TOTAL_SECS(sbi)); 424 break; 425 } 426 secno = left_start; 427 skip_left: 428 hint = secno; 429 segno = secno * sbi->segs_per_sec; 430 zoneno = secno / sbi->secs_per_zone; 431 432 /* give up on finding another zone */ 433 if (!init) 434 goto got_it; 435 if (sbi->secs_per_zone == 1) 436 goto got_it; 437 if (zoneno == old_zoneno) 438 goto got_it; 439 if (dir == ALLOC_LEFT) { 440 if (!go_left && zoneno + 1 >= total_zones) 441 goto got_it; 442 if (go_left && zoneno == 0) 443 goto got_it; 444 } 445 for (i = 0; i < NR_CURSEG_TYPE; i++) 446 if (CURSEG_I(sbi, i)->zone == zoneno) 447 break; 448 449 if (i < NR_CURSEG_TYPE) { 450 /* zone is in user, try another */ 451 if (go_left) 452 hint = zoneno * sbi->secs_per_zone - 1; 453 else if (zoneno + 1 >= total_zones) 454 hint = 0; 455 else 456 hint = (zoneno + 1) * sbi->secs_per_zone; 457 init = false; 458 goto find_other_zone; 459 } 460 got_it: 461 /* set it as dirty segment in free segmap */ 462 BUG_ON(test_bit(segno, free_i->free_segmap)); 463 __set_inuse(sbi, segno); 464 *newseg = segno; 465 write_unlock(&free_i->segmap_lock); 466 } 467 468 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 469 { 470 struct curseg_info *curseg = CURSEG_I(sbi, type); 471 struct summary_footer *sum_footer; 472 473 curseg->segno = curseg->next_segno; 474 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno); 475 curseg->next_blkoff = 0; 476 curseg->next_segno = NULL_SEGNO; 477 478 sum_footer = &(curseg->sum_blk->footer); 479 memset(sum_footer, 0, sizeof(struct summary_footer)); 480 if (IS_DATASEG(type)) 481 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 482 if (IS_NODESEG(type)) 483 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 484 __set_sit_entry_type(sbi, type, curseg->segno, modified); 485 } 486 487 /* 488 * Allocate a current working segment. 489 * This function always allocates a free segment in LFS manner. 490 */ 491 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 492 { 493 struct curseg_info *curseg = CURSEG_I(sbi, type); 494 unsigned int segno = curseg->segno; 495 int dir = ALLOC_LEFT; 496 497 write_sum_page(sbi, curseg->sum_blk, 498 GET_SUM_BLOCK(sbi, curseg->segno)); 499 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA) 500 dir = ALLOC_RIGHT; 501 502 if (test_opt(sbi, NOHEAP)) 503 dir = ALLOC_RIGHT; 504 505 get_new_segment(sbi, &segno, new_sec, dir); 506 curseg->next_segno = segno; 507 reset_curseg(sbi, type, 1); 508 curseg->alloc_type = LFS; 509 } 510 511 static void __next_free_blkoff(struct f2fs_sb_info *sbi, 512 struct curseg_info *seg, block_t start) 513 { 514 struct seg_entry *se = get_seg_entry(sbi, seg->segno); 515 block_t ofs; 516 for (ofs = start; ofs < sbi->blocks_per_seg; ofs++) { 517 if (!f2fs_test_bit(ofs, se->ckpt_valid_map) 518 && !f2fs_test_bit(ofs, se->cur_valid_map)) 519 break; 520 } 521 seg->next_blkoff = ofs; 522 } 523 524 /* 525 * If a segment is written by LFS manner, next block offset is just obtained 526 * by increasing the current block offset. However, if a segment is written by 527 * SSR manner, next block offset obtained by calling __next_free_blkoff 528 */ 529 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, 530 struct curseg_info *seg) 531 { 532 if (seg->alloc_type == SSR) 533 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1); 534 else 535 seg->next_blkoff++; 536 } 537 538 /* 539 * This function always allocates a used segment (from dirty seglist) by SSR 540 * manner, so it should recover the existing segment information of valid blocks 541 */ 542 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse) 543 { 544 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 545 struct curseg_info *curseg = CURSEG_I(sbi, type); 546 unsigned int new_segno = curseg->next_segno; 547 struct f2fs_summary_block *sum_node; 548 struct page *sum_page; 549 550 write_sum_page(sbi, curseg->sum_blk, 551 GET_SUM_BLOCK(sbi, curseg->segno)); 552 __set_test_and_inuse(sbi, new_segno); 553 554 mutex_lock(&dirty_i->seglist_lock); 555 __remove_dirty_segment(sbi, new_segno, PRE); 556 __remove_dirty_segment(sbi, new_segno, DIRTY); 557 mutex_unlock(&dirty_i->seglist_lock); 558 559 reset_curseg(sbi, type, 1); 560 curseg->alloc_type = SSR; 561 __next_free_blkoff(sbi, curseg, 0); 562 563 if (reuse) { 564 sum_page = get_sum_page(sbi, new_segno); 565 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 566 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); 567 f2fs_put_page(sum_page, 1); 568 } 569 } 570 571 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type) 572 { 573 struct curseg_info *curseg = CURSEG_I(sbi, type); 574 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; 575 576 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0)) 577 return v_ops->get_victim(sbi, 578 &(curseg)->next_segno, BG_GC, type, SSR); 579 580 /* For data segments, let's do SSR more intensively */ 581 for (; type >= CURSEG_HOT_DATA; type--) 582 if (v_ops->get_victim(sbi, &(curseg)->next_segno, 583 BG_GC, type, SSR)) 584 return 1; 585 return 0; 586 } 587 588 /* 589 * flush out current segment and replace it with new segment 590 * This function should be returned with success, otherwise BUG 591 */ 592 static void allocate_segment_by_default(struct f2fs_sb_info *sbi, 593 int type, bool force) 594 { 595 struct curseg_info *curseg = CURSEG_I(sbi, type); 596 597 if (force) { 598 new_curseg(sbi, type, true); 599 goto out; 600 } 601 602 curseg->next_segno = check_prefree_segments(sbi, type); 603 604 if (curseg->next_segno != NULL_SEGNO) 605 change_curseg(sbi, type, false); 606 else if (type == CURSEG_WARM_NODE) 607 new_curseg(sbi, type, false); 608 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type)) 609 new_curseg(sbi, type, false); 610 else if (need_SSR(sbi) && get_ssr_segment(sbi, type)) 611 change_curseg(sbi, type, true); 612 else 613 new_curseg(sbi, type, false); 614 out: 615 sbi->segment_count[curseg->alloc_type]++; 616 } 617 618 void allocate_new_segments(struct f2fs_sb_info *sbi) 619 { 620 struct curseg_info *curseg; 621 unsigned int old_curseg; 622 int i; 623 624 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 625 curseg = CURSEG_I(sbi, i); 626 old_curseg = curseg->segno; 627 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true); 628 locate_dirty_segment(sbi, old_curseg); 629 } 630 } 631 632 static const struct segment_allocation default_salloc_ops = { 633 .allocate_segment = allocate_segment_by_default, 634 }; 635 636 static void f2fs_end_io_write(struct bio *bio, int err) 637 { 638 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); 639 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; 640 struct bio_private *p = bio->bi_private; 641 642 do { 643 struct page *page = bvec->bv_page; 644 645 if (--bvec >= bio->bi_io_vec) 646 prefetchw(&bvec->bv_page->flags); 647 if (!uptodate) { 648 SetPageError(page); 649 if (page->mapping) 650 set_bit(AS_EIO, &page->mapping->flags); 651 set_ckpt_flags(p->sbi->ckpt, CP_ERROR_FLAG); 652 p->sbi->sb->s_flags |= MS_RDONLY; 653 } 654 end_page_writeback(page); 655 dec_page_count(p->sbi, F2FS_WRITEBACK); 656 } while (bvec >= bio->bi_io_vec); 657 658 if (p->is_sync) 659 complete(p->wait); 660 kfree(p); 661 bio_put(bio); 662 } 663 664 struct bio *f2fs_bio_alloc(struct block_device *bdev, int npages) 665 { 666 struct bio *bio; 667 struct bio_private *priv; 668 retry: 669 priv = kmalloc(sizeof(struct bio_private), GFP_NOFS); 670 if (!priv) { 671 cond_resched(); 672 goto retry; 673 } 674 675 /* No failure on bio allocation */ 676 bio = bio_alloc(GFP_NOIO, npages); 677 bio->bi_bdev = bdev; 678 bio->bi_private = priv; 679 return bio; 680 } 681 682 static void do_submit_bio(struct f2fs_sb_info *sbi, 683 enum page_type type, bool sync) 684 { 685 int rw = sync ? WRITE_SYNC : WRITE; 686 enum page_type btype = type > META ? META : type; 687 688 if (type >= META_FLUSH) 689 rw = WRITE_FLUSH_FUA; 690 691 if (btype == META) 692 rw |= REQ_META; 693 694 if (sbi->bio[btype]) { 695 struct bio_private *p = sbi->bio[btype]->bi_private; 696 p->sbi = sbi; 697 sbi->bio[btype]->bi_end_io = f2fs_end_io_write; 698 699 trace_f2fs_do_submit_bio(sbi->sb, btype, sync, sbi->bio[btype]); 700 701 if (type == META_FLUSH) { 702 DECLARE_COMPLETION_ONSTACK(wait); 703 p->is_sync = true; 704 p->wait = &wait; 705 submit_bio(rw, sbi->bio[btype]); 706 wait_for_completion(&wait); 707 } else { 708 p->is_sync = false; 709 submit_bio(rw, sbi->bio[btype]); 710 } 711 sbi->bio[btype] = NULL; 712 } 713 } 714 715 void f2fs_submit_bio(struct f2fs_sb_info *sbi, enum page_type type, bool sync) 716 { 717 down_write(&sbi->bio_sem); 718 do_submit_bio(sbi, type, sync); 719 up_write(&sbi->bio_sem); 720 } 721 722 static void submit_write_page(struct f2fs_sb_info *sbi, struct page *page, 723 block_t blk_addr, enum page_type type) 724 { 725 struct block_device *bdev = sbi->sb->s_bdev; 726 727 verify_block_addr(sbi, blk_addr); 728 729 down_write(&sbi->bio_sem); 730 731 inc_page_count(sbi, F2FS_WRITEBACK); 732 733 if (sbi->bio[type] && sbi->last_block_in_bio[type] != blk_addr - 1) 734 do_submit_bio(sbi, type, false); 735 alloc_new: 736 if (sbi->bio[type] == NULL) { 737 sbi->bio[type] = f2fs_bio_alloc(bdev, max_hw_blocks(sbi)); 738 sbi->bio[type]->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr); 739 /* 740 * The end_io will be assigned at the sumbission phase. 741 * Until then, let bio_add_page() merge consecutive IOs as much 742 * as possible. 743 */ 744 } 745 746 if (bio_add_page(sbi->bio[type], page, PAGE_CACHE_SIZE, 0) < 747 PAGE_CACHE_SIZE) { 748 do_submit_bio(sbi, type, false); 749 goto alloc_new; 750 } 751 752 sbi->last_block_in_bio[type] = blk_addr; 753 754 up_write(&sbi->bio_sem); 755 trace_f2fs_submit_write_page(page, blk_addr, type); 756 } 757 758 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type) 759 { 760 struct curseg_info *curseg = CURSEG_I(sbi, type); 761 if (curseg->next_blkoff < sbi->blocks_per_seg) 762 return true; 763 return false; 764 } 765 766 static int __get_segment_type_2(struct page *page, enum page_type p_type) 767 { 768 if (p_type == DATA) 769 return CURSEG_HOT_DATA; 770 else 771 return CURSEG_HOT_NODE; 772 } 773 774 static int __get_segment_type_4(struct page *page, enum page_type p_type) 775 { 776 if (p_type == DATA) { 777 struct inode *inode = page->mapping->host; 778 779 if (S_ISDIR(inode->i_mode)) 780 return CURSEG_HOT_DATA; 781 else 782 return CURSEG_COLD_DATA; 783 } else { 784 if (IS_DNODE(page) && !is_cold_node(page)) 785 return CURSEG_HOT_NODE; 786 else 787 return CURSEG_COLD_NODE; 788 } 789 } 790 791 static int __get_segment_type_6(struct page *page, enum page_type p_type) 792 { 793 if (p_type == DATA) { 794 struct inode *inode = page->mapping->host; 795 796 if (S_ISDIR(inode->i_mode)) 797 return CURSEG_HOT_DATA; 798 else if (is_cold_data(page) || is_cold_file(inode)) 799 return CURSEG_COLD_DATA; 800 else 801 return CURSEG_WARM_DATA; 802 } else { 803 if (IS_DNODE(page)) 804 return is_cold_node(page) ? CURSEG_WARM_NODE : 805 CURSEG_HOT_NODE; 806 else 807 return CURSEG_COLD_NODE; 808 } 809 } 810 811 static int __get_segment_type(struct page *page, enum page_type p_type) 812 { 813 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb); 814 switch (sbi->active_logs) { 815 case 2: 816 return __get_segment_type_2(page, p_type); 817 case 4: 818 return __get_segment_type_4(page, p_type); 819 } 820 /* NR_CURSEG_TYPE(6) logs by default */ 821 BUG_ON(sbi->active_logs != NR_CURSEG_TYPE); 822 return __get_segment_type_6(page, p_type); 823 } 824 825 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page, 826 block_t old_blkaddr, block_t *new_blkaddr, 827 struct f2fs_summary *sum, enum page_type p_type) 828 { 829 struct sit_info *sit_i = SIT_I(sbi); 830 struct curseg_info *curseg; 831 unsigned int old_cursegno; 832 int type; 833 834 type = __get_segment_type(page, p_type); 835 curseg = CURSEG_I(sbi, type); 836 837 mutex_lock(&curseg->curseg_mutex); 838 839 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 840 old_cursegno = curseg->segno; 841 842 /* 843 * __add_sum_entry should be resided under the curseg_mutex 844 * because, this function updates a summary entry in the 845 * current summary block. 846 */ 847 __add_sum_entry(sbi, type, sum, curseg->next_blkoff); 848 849 mutex_lock(&sit_i->sentry_lock); 850 __refresh_next_blkoff(sbi, curseg); 851 sbi->block_count[curseg->alloc_type]++; 852 853 /* 854 * SIT information should be updated before segment allocation, 855 * since SSR needs latest valid block information. 856 */ 857 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr); 858 859 if (!__has_curseg_space(sbi, type)) 860 sit_i->s_ops->allocate_segment(sbi, type, false); 861 862 locate_dirty_segment(sbi, old_cursegno); 863 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 864 mutex_unlock(&sit_i->sentry_lock); 865 866 if (p_type == NODE) 867 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); 868 869 /* writeout dirty page into bdev */ 870 submit_write_page(sbi, page, *new_blkaddr, p_type); 871 872 mutex_unlock(&curseg->curseg_mutex); 873 } 874 875 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page) 876 { 877 set_page_writeback(page); 878 submit_write_page(sbi, page, page->index, META); 879 } 880 881 void write_node_page(struct f2fs_sb_info *sbi, struct page *page, 882 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr) 883 { 884 struct f2fs_summary sum; 885 set_summary(&sum, nid, 0, 0); 886 do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, NODE); 887 } 888 889 void write_data_page(struct inode *inode, struct page *page, 890 struct dnode_of_data *dn, block_t old_blkaddr, 891 block_t *new_blkaddr) 892 { 893 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 894 struct f2fs_summary sum; 895 struct node_info ni; 896 897 BUG_ON(old_blkaddr == NULL_ADDR); 898 get_node_info(sbi, dn->nid, &ni); 899 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); 900 901 do_write_page(sbi, page, old_blkaddr, 902 new_blkaddr, &sum, DATA); 903 } 904 905 void rewrite_data_page(struct f2fs_sb_info *sbi, struct page *page, 906 block_t old_blk_addr) 907 { 908 submit_write_page(sbi, page, old_blk_addr, DATA); 909 } 910 911 void recover_data_page(struct f2fs_sb_info *sbi, 912 struct page *page, struct f2fs_summary *sum, 913 block_t old_blkaddr, block_t new_blkaddr) 914 { 915 struct sit_info *sit_i = SIT_I(sbi); 916 struct curseg_info *curseg; 917 unsigned int segno, old_cursegno; 918 struct seg_entry *se; 919 int type; 920 921 segno = GET_SEGNO(sbi, new_blkaddr); 922 se = get_seg_entry(sbi, segno); 923 type = se->type; 924 925 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { 926 if (old_blkaddr == NULL_ADDR) 927 type = CURSEG_COLD_DATA; 928 else 929 type = CURSEG_WARM_DATA; 930 } 931 curseg = CURSEG_I(sbi, type); 932 933 mutex_lock(&curseg->curseg_mutex); 934 mutex_lock(&sit_i->sentry_lock); 935 936 old_cursegno = curseg->segno; 937 938 /* change the current segment */ 939 if (segno != curseg->segno) { 940 curseg->next_segno = segno; 941 change_curseg(sbi, type, true); 942 } 943 944 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) & 945 (sbi->blocks_per_seg - 1); 946 __add_sum_entry(sbi, type, sum, curseg->next_blkoff); 947 948 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr); 949 950 locate_dirty_segment(sbi, old_cursegno); 951 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 952 953 mutex_unlock(&sit_i->sentry_lock); 954 mutex_unlock(&curseg->curseg_mutex); 955 } 956 957 void rewrite_node_page(struct f2fs_sb_info *sbi, 958 struct page *page, struct f2fs_summary *sum, 959 block_t old_blkaddr, block_t new_blkaddr) 960 { 961 struct sit_info *sit_i = SIT_I(sbi); 962 int type = CURSEG_WARM_NODE; 963 struct curseg_info *curseg; 964 unsigned int segno, old_cursegno; 965 block_t next_blkaddr = next_blkaddr_of_node(page); 966 unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr); 967 968 curseg = CURSEG_I(sbi, type); 969 970 mutex_lock(&curseg->curseg_mutex); 971 mutex_lock(&sit_i->sentry_lock); 972 973 segno = GET_SEGNO(sbi, new_blkaddr); 974 old_cursegno = curseg->segno; 975 976 /* change the current segment */ 977 if (segno != curseg->segno) { 978 curseg->next_segno = segno; 979 change_curseg(sbi, type, true); 980 } 981 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) & 982 (sbi->blocks_per_seg - 1); 983 __add_sum_entry(sbi, type, sum, curseg->next_blkoff); 984 985 /* change the current log to the next block addr in advance */ 986 if (next_segno != segno) { 987 curseg->next_segno = next_segno; 988 change_curseg(sbi, type, true); 989 } 990 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, next_blkaddr) & 991 (sbi->blocks_per_seg - 1); 992 993 /* rewrite node page */ 994 set_page_writeback(page); 995 submit_write_page(sbi, page, new_blkaddr, NODE); 996 f2fs_submit_bio(sbi, NODE, true); 997 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr); 998 999 locate_dirty_segment(sbi, old_cursegno); 1000 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 1001 1002 mutex_unlock(&sit_i->sentry_lock); 1003 mutex_unlock(&curseg->curseg_mutex); 1004 } 1005 1006 static int read_compacted_summaries(struct f2fs_sb_info *sbi) 1007 { 1008 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1009 struct curseg_info *seg_i; 1010 unsigned char *kaddr; 1011 struct page *page; 1012 block_t start; 1013 int i, j, offset; 1014 1015 start = start_sum_block(sbi); 1016 1017 page = get_meta_page(sbi, start++); 1018 kaddr = (unsigned char *)page_address(page); 1019 1020 /* Step 1: restore nat cache */ 1021 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 1022 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE); 1023 1024 /* Step 2: restore sit cache */ 1025 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 1026 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE, 1027 SUM_JOURNAL_SIZE); 1028 offset = 2 * SUM_JOURNAL_SIZE; 1029 1030 /* Step 3: restore summary entries */ 1031 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 1032 unsigned short blk_off; 1033 unsigned int segno; 1034 1035 seg_i = CURSEG_I(sbi, i); 1036 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 1037 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 1038 seg_i->next_segno = segno; 1039 reset_curseg(sbi, i, 0); 1040 seg_i->alloc_type = ckpt->alloc_type[i]; 1041 seg_i->next_blkoff = blk_off; 1042 1043 if (seg_i->alloc_type == SSR) 1044 blk_off = sbi->blocks_per_seg; 1045 1046 for (j = 0; j < blk_off; j++) { 1047 struct f2fs_summary *s; 1048 s = (struct f2fs_summary *)(kaddr + offset); 1049 seg_i->sum_blk->entries[j] = *s; 1050 offset += SUMMARY_SIZE; 1051 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE - 1052 SUM_FOOTER_SIZE) 1053 continue; 1054 1055 f2fs_put_page(page, 1); 1056 page = NULL; 1057 1058 page = get_meta_page(sbi, start++); 1059 kaddr = (unsigned char *)page_address(page); 1060 offset = 0; 1061 } 1062 } 1063 f2fs_put_page(page, 1); 1064 return 0; 1065 } 1066 1067 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 1068 { 1069 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1070 struct f2fs_summary_block *sum; 1071 struct curseg_info *curseg; 1072 struct page *new; 1073 unsigned short blk_off; 1074 unsigned int segno = 0; 1075 block_t blk_addr = 0; 1076 1077 /* get segment number and block addr */ 1078 if (IS_DATASEG(type)) { 1079 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 1080 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 1081 CURSEG_HOT_DATA]); 1082 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) 1083 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type); 1084 else 1085 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 1086 } else { 1087 segno = le32_to_cpu(ckpt->cur_node_segno[type - 1088 CURSEG_HOT_NODE]); 1089 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 1090 CURSEG_HOT_NODE]); 1091 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) 1092 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 1093 type - CURSEG_HOT_NODE); 1094 else 1095 blk_addr = GET_SUM_BLOCK(sbi, segno); 1096 } 1097 1098 new = get_meta_page(sbi, blk_addr); 1099 sum = (struct f2fs_summary_block *)page_address(new); 1100 1101 if (IS_NODESEG(type)) { 1102 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) { 1103 struct f2fs_summary *ns = &sum->entries[0]; 1104 int i; 1105 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { 1106 ns->version = 0; 1107 ns->ofs_in_node = 0; 1108 } 1109 } else { 1110 if (restore_node_summary(sbi, segno, sum)) { 1111 f2fs_put_page(new, 1); 1112 return -EINVAL; 1113 } 1114 } 1115 } 1116 1117 /* set uncompleted segment to curseg */ 1118 curseg = CURSEG_I(sbi, type); 1119 mutex_lock(&curseg->curseg_mutex); 1120 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE); 1121 curseg->next_segno = segno; 1122 reset_curseg(sbi, type, 0); 1123 curseg->alloc_type = ckpt->alloc_type[type]; 1124 curseg->next_blkoff = blk_off; 1125 mutex_unlock(&curseg->curseg_mutex); 1126 f2fs_put_page(new, 1); 1127 return 0; 1128 } 1129 1130 static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 1131 { 1132 int type = CURSEG_HOT_DATA; 1133 1134 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) { 1135 /* restore for compacted data summary */ 1136 if (read_compacted_summaries(sbi)) 1137 return -EINVAL; 1138 type = CURSEG_HOT_NODE; 1139 } 1140 1141 for (; type <= CURSEG_COLD_NODE; type++) 1142 if (read_normal_summaries(sbi, type)) 1143 return -EINVAL; 1144 return 0; 1145 } 1146 1147 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 1148 { 1149 struct page *page; 1150 unsigned char *kaddr; 1151 struct f2fs_summary *summary; 1152 struct curseg_info *seg_i; 1153 int written_size = 0; 1154 int i, j; 1155 1156 page = grab_meta_page(sbi, blkaddr++); 1157 kaddr = (unsigned char *)page_address(page); 1158 1159 /* Step 1: write nat cache */ 1160 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 1161 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE); 1162 written_size += SUM_JOURNAL_SIZE; 1163 1164 /* Step 2: write sit cache */ 1165 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 1166 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits, 1167 SUM_JOURNAL_SIZE); 1168 written_size += SUM_JOURNAL_SIZE; 1169 1170 set_page_dirty(page); 1171 1172 /* Step 3: write summary entries */ 1173 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 1174 unsigned short blkoff; 1175 seg_i = CURSEG_I(sbi, i); 1176 if (sbi->ckpt->alloc_type[i] == SSR) 1177 blkoff = sbi->blocks_per_seg; 1178 else 1179 blkoff = curseg_blkoff(sbi, i); 1180 1181 for (j = 0; j < blkoff; j++) { 1182 if (!page) { 1183 page = grab_meta_page(sbi, blkaddr++); 1184 kaddr = (unsigned char *)page_address(page); 1185 written_size = 0; 1186 } 1187 summary = (struct f2fs_summary *)(kaddr + written_size); 1188 *summary = seg_i->sum_blk->entries[j]; 1189 written_size += SUMMARY_SIZE; 1190 set_page_dirty(page); 1191 1192 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE - 1193 SUM_FOOTER_SIZE) 1194 continue; 1195 1196 f2fs_put_page(page, 1); 1197 page = NULL; 1198 } 1199 } 1200 if (page) 1201 f2fs_put_page(page, 1); 1202 } 1203 1204 static void write_normal_summaries(struct f2fs_sb_info *sbi, 1205 block_t blkaddr, int type) 1206 { 1207 int i, end; 1208 if (IS_DATASEG(type)) 1209 end = type + NR_CURSEG_DATA_TYPE; 1210 else 1211 end = type + NR_CURSEG_NODE_TYPE; 1212 1213 for (i = type; i < end; i++) { 1214 struct curseg_info *sum = CURSEG_I(sbi, i); 1215 mutex_lock(&sum->curseg_mutex); 1216 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type)); 1217 mutex_unlock(&sum->curseg_mutex); 1218 } 1219 } 1220 1221 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 1222 { 1223 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) 1224 write_compacted_summaries(sbi, start_blk); 1225 else 1226 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 1227 } 1228 1229 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 1230 { 1231 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG)) 1232 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 1233 return; 1234 } 1235 1236 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type, 1237 unsigned int val, int alloc) 1238 { 1239 int i; 1240 1241 if (type == NAT_JOURNAL) { 1242 for (i = 0; i < nats_in_cursum(sum); i++) { 1243 if (le32_to_cpu(nid_in_journal(sum, i)) == val) 1244 return i; 1245 } 1246 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) 1247 return update_nats_in_cursum(sum, 1); 1248 } else if (type == SIT_JOURNAL) { 1249 for (i = 0; i < sits_in_cursum(sum); i++) 1250 if (le32_to_cpu(segno_in_journal(sum, i)) == val) 1251 return i; 1252 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES) 1253 return update_sits_in_cursum(sum, 1); 1254 } 1255 return -1; 1256 } 1257 1258 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, 1259 unsigned int segno) 1260 { 1261 struct sit_info *sit_i = SIT_I(sbi); 1262 unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno); 1263 block_t blk_addr = sit_i->sit_base_addr + offset; 1264 1265 check_seg_range(sbi, segno); 1266 1267 /* calculate sit block address */ 1268 if (f2fs_test_bit(offset, sit_i->sit_bitmap)) 1269 blk_addr += sit_i->sit_blocks; 1270 1271 return get_meta_page(sbi, blk_addr); 1272 } 1273 1274 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, 1275 unsigned int start) 1276 { 1277 struct sit_info *sit_i = SIT_I(sbi); 1278 struct page *src_page, *dst_page; 1279 pgoff_t src_off, dst_off; 1280 void *src_addr, *dst_addr; 1281 1282 src_off = current_sit_addr(sbi, start); 1283 dst_off = next_sit_addr(sbi, src_off); 1284 1285 /* get current sit block page without lock */ 1286 src_page = get_meta_page(sbi, src_off); 1287 dst_page = grab_meta_page(sbi, dst_off); 1288 BUG_ON(PageDirty(src_page)); 1289 1290 src_addr = page_address(src_page); 1291 dst_addr = page_address(dst_page); 1292 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE); 1293 1294 set_page_dirty(dst_page); 1295 f2fs_put_page(src_page, 1); 1296 1297 set_to_next_sit(sit_i, start); 1298 1299 return dst_page; 1300 } 1301 1302 static bool flush_sits_in_journal(struct f2fs_sb_info *sbi) 1303 { 1304 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 1305 struct f2fs_summary_block *sum = curseg->sum_blk; 1306 int i; 1307 1308 /* 1309 * If the journal area in the current summary is full of sit entries, 1310 * all the sit entries will be flushed. Otherwise the sit entries 1311 * are not able to replace with newly hot sit entries. 1312 */ 1313 if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) { 1314 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) { 1315 unsigned int segno; 1316 segno = le32_to_cpu(segno_in_journal(sum, i)); 1317 __mark_sit_entry_dirty(sbi, segno); 1318 } 1319 update_sits_in_cursum(sum, -sits_in_cursum(sum)); 1320 return 1; 1321 } 1322 return 0; 1323 } 1324 1325 /* 1326 * CP calls this function, which flushes SIT entries including sit_journal, 1327 * and moves prefree segs to free segs. 1328 */ 1329 void flush_sit_entries(struct f2fs_sb_info *sbi) 1330 { 1331 struct sit_info *sit_i = SIT_I(sbi); 1332 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 1333 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 1334 struct f2fs_summary_block *sum = curseg->sum_blk; 1335 unsigned long nsegs = TOTAL_SEGS(sbi); 1336 struct page *page = NULL; 1337 struct f2fs_sit_block *raw_sit = NULL; 1338 unsigned int start = 0, end = 0; 1339 unsigned int segno = -1; 1340 bool flushed; 1341 1342 mutex_lock(&curseg->curseg_mutex); 1343 mutex_lock(&sit_i->sentry_lock); 1344 1345 /* 1346 * "flushed" indicates whether sit entries in journal are flushed 1347 * to the SIT area or not. 1348 */ 1349 flushed = flush_sits_in_journal(sbi); 1350 1351 while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) { 1352 struct seg_entry *se = get_seg_entry(sbi, segno); 1353 int sit_offset, offset; 1354 1355 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 1356 1357 if (flushed) 1358 goto to_sit_page; 1359 1360 offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1); 1361 if (offset >= 0) { 1362 segno_in_journal(sum, offset) = cpu_to_le32(segno); 1363 seg_info_to_raw_sit(se, &sit_in_journal(sum, offset)); 1364 goto flush_done; 1365 } 1366 to_sit_page: 1367 if (!page || (start > segno) || (segno > end)) { 1368 if (page) { 1369 f2fs_put_page(page, 1); 1370 page = NULL; 1371 } 1372 1373 start = START_SEGNO(sit_i, segno); 1374 end = start + SIT_ENTRY_PER_BLOCK - 1; 1375 1376 /* read sit block that will be updated */ 1377 page = get_next_sit_page(sbi, start); 1378 raw_sit = page_address(page); 1379 } 1380 1381 /* udpate entry in SIT block */ 1382 seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]); 1383 flush_done: 1384 __clear_bit(segno, bitmap); 1385 sit_i->dirty_sentries--; 1386 } 1387 mutex_unlock(&sit_i->sentry_lock); 1388 mutex_unlock(&curseg->curseg_mutex); 1389 1390 /* writeout last modified SIT block */ 1391 f2fs_put_page(page, 1); 1392 1393 set_prefree_as_free_segments(sbi); 1394 } 1395 1396 static int build_sit_info(struct f2fs_sb_info *sbi) 1397 { 1398 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 1399 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1400 struct sit_info *sit_i; 1401 unsigned int sit_segs, start; 1402 char *src_bitmap, *dst_bitmap; 1403 unsigned int bitmap_size; 1404 1405 /* allocate memory for SIT information */ 1406 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL); 1407 if (!sit_i) 1408 return -ENOMEM; 1409 1410 SM_I(sbi)->sit_info = sit_i; 1411 1412 sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry)); 1413 if (!sit_i->sentries) 1414 return -ENOMEM; 1415 1416 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi)); 1417 sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL); 1418 if (!sit_i->dirty_sentries_bitmap) 1419 return -ENOMEM; 1420 1421 for (start = 0; start < TOTAL_SEGS(sbi); start++) { 1422 sit_i->sentries[start].cur_valid_map 1423 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 1424 sit_i->sentries[start].ckpt_valid_map 1425 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 1426 if (!sit_i->sentries[start].cur_valid_map 1427 || !sit_i->sentries[start].ckpt_valid_map) 1428 return -ENOMEM; 1429 } 1430 1431 if (sbi->segs_per_sec > 1) { 1432 sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) * 1433 sizeof(struct sec_entry)); 1434 if (!sit_i->sec_entries) 1435 return -ENOMEM; 1436 } 1437 1438 /* get information related with SIT */ 1439 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 1440 1441 /* setup SIT bitmap from ckeckpoint pack */ 1442 bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 1443 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 1444 1445 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); 1446 if (!dst_bitmap) 1447 return -ENOMEM; 1448 1449 /* init SIT information */ 1450 sit_i->s_ops = &default_salloc_ops; 1451 1452 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 1453 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; 1454 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count); 1455 sit_i->sit_bitmap = dst_bitmap; 1456 sit_i->bitmap_size = bitmap_size; 1457 sit_i->dirty_sentries = 0; 1458 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 1459 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 1460 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec; 1461 mutex_init(&sit_i->sentry_lock); 1462 return 0; 1463 } 1464 1465 static int build_free_segmap(struct f2fs_sb_info *sbi) 1466 { 1467 struct f2fs_sm_info *sm_info = SM_I(sbi); 1468 struct free_segmap_info *free_i; 1469 unsigned int bitmap_size, sec_bitmap_size; 1470 1471 /* allocate memory for free segmap information */ 1472 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL); 1473 if (!free_i) 1474 return -ENOMEM; 1475 1476 SM_I(sbi)->free_info = free_i; 1477 1478 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi)); 1479 free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL); 1480 if (!free_i->free_segmap) 1481 return -ENOMEM; 1482 1483 sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi)); 1484 free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL); 1485 if (!free_i->free_secmap) 1486 return -ENOMEM; 1487 1488 /* set all segments as dirty temporarily */ 1489 memset(free_i->free_segmap, 0xff, bitmap_size); 1490 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 1491 1492 /* init free segmap information */ 1493 free_i->start_segno = 1494 (unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr); 1495 free_i->free_segments = 0; 1496 free_i->free_sections = 0; 1497 rwlock_init(&free_i->segmap_lock); 1498 return 0; 1499 } 1500 1501 static int build_curseg(struct f2fs_sb_info *sbi) 1502 { 1503 struct curseg_info *array; 1504 int i; 1505 1506 array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL); 1507 if (!array) 1508 return -ENOMEM; 1509 1510 SM_I(sbi)->curseg_array = array; 1511 1512 for (i = 0; i < NR_CURSEG_TYPE; i++) { 1513 mutex_init(&array[i].curseg_mutex); 1514 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL); 1515 if (!array[i].sum_blk) 1516 return -ENOMEM; 1517 array[i].segno = NULL_SEGNO; 1518 array[i].next_blkoff = 0; 1519 } 1520 return restore_curseg_summaries(sbi); 1521 } 1522 1523 static void build_sit_entries(struct f2fs_sb_info *sbi) 1524 { 1525 struct sit_info *sit_i = SIT_I(sbi); 1526 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 1527 struct f2fs_summary_block *sum = curseg->sum_blk; 1528 unsigned int start; 1529 1530 for (start = 0; start < TOTAL_SEGS(sbi); start++) { 1531 struct seg_entry *se = &sit_i->sentries[start]; 1532 struct f2fs_sit_block *sit_blk; 1533 struct f2fs_sit_entry sit; 1534 struct page *page; 1535 int i; 1536 1537 mutex_lock(&curseg->curseg_mutex); 1538 for (i = 0; i < sits_in_cursum(sum); i++) { 1539 if (le32_to_cpu(segno_in_journal(sum, i)) == start) { 1540 sit = sit_in_journal(sum, i); 1541 mutex_unlock(&curseg->curseg_mutex); 1542 goto got_it; 1543 } 1544 } 1545 mutex_unlock(&curseg->curseg_mutex); 1546 page = get_current_sit_page(sbi, start); 1547 sit_blk = (struct f2fs_sit_block *)page_address(page); 1548 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 1549 f2fs_put_page(page, 1); 1550 got_it: 1551 check_block_count(sbi, start, &sit); 1552 seg_info_from_raw_sit(se, &sit); 1553 if (sbi->segs_per_sec > 1) { 1554 struct sec_entry *e = get_sec_entry(sbi, start); 1555 e->valid_blocks += se->valid_blocks; 1556 } 1557 } 1558 } 1559 1560 static void init_free_segmap(struct f2fs_sb_info *sbi) 1561 { 1562 unsigned int start; 1563 int type; 1564 1565 for (start = 0; start < TOTAL_SEGS(sbi); start++) { 1566 struct seg_entry *sentry = get_seg_entry(sbi, start); 1567 if (!sentry->valid_blocks) 1568 __set_free(sbi, start); 1569 } 1570 1571 /* set use the current segments */ 1572 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 1573 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 1574 __set_test_and_inuse(sbi, curseg_t->segno); 1575 } 1576 } 1577 1578 static void init_dirty_segmap(struct f2fs_sb_info *sbi) 1579 { 1580 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1581 struct free_segmap_info *free_i = FREE_I(sbi); 1582 unsigned int segno = 0, offset = 0; 1583 unsigned short valid_blocks; 1584 1585 while (segno < TOTAL_SEGS(sbi)) { 1586 /* find dirty segment based on free segmap */ 1587 segno = find_next_inuse(free_i, TOTAL_SEGS(sbi), offset); 1588 if (segno >= TOTAL_SEGS(sbi)) 1589 break; 1590 offset = segno + 1; 1591 valid_blocks = get_valid_blocks(sbi, segno, 0); 1592 if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks) 1593 continue; 1594 mutex_lock(&dirty_i->seglist_lock); 1595 __locate_dirty_segment(sbi, segno, DIRTY); 1596 mutex_unlock(&dirty_i->seglist_lock); 1597 } 1598 } 1599 1600 static int init_victim_secmap(struct f2fs_sb_info *sbi) 1601 { 1602 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1603 unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi)); 1604 1605 dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL); 1606 if (!dirty_i->victim_secmap) 1607 return -ENOMEM; 1608 return 0; 1609 } 1610 1611 static int build_dirty_segmap(struct f2fs_sb_info *sbi) 1612 { 1613 struct dirty_seglist_info *dirty_i; 1614 unsigned int bitmap_size, i; 1615 1616 /* allocate memory for dirty segments list information */ 1617 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL); 1618 if (!dirty_i) 1619 return -ENOMEM; 1620 1621 SM_I(sbi)->dirty_info = dirty_i; 1622 mutex_init(&dirty_i->seglist_lock); 1623 1624 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi)); 1625 1626 for (i = 0; i < NR_DIRTY_TYPE; i++) { 1627 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL); 1628 if (!dirty_i->dirty_segmap[i]) 1629 return -ENOMEM; 1630 } 1631 1632 init_dirty_segmap(sbi); 1633 return init_victim_secmap(sbi); 1634 } 1635 1636 /* 1637 * Update min, max modified time for cost-benefit GC algorithm 1638 */ 1639 static void init_min_max_mtime(struct f2fs_sb_info *sbi) 1640 { 1641 struct sit_info *sit_i = SIT_I(sbi); 1642 unsigned int segno; 1643 1644 mutex_lock(&sit_i->sentry_lock); 1645 1646 sit_i->min_mtime = LLONG_MAX; 1647 1648 for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) { 1649 unsigned int i; 1650 unsigned long long mtime = 0; 1651 1652 for (i = 0; i < sbi->segs_per_sec; i++) 1653 mtime += get_seg_entry(sbi, segno + i)->mtime; 1654 1655 mtime = div_u64(mtime, sbi->segs_per_sec); 1656 1657 if (sit_i->min_mtime > mtime) 1658 sit_i->min_mtime = mtime; 1659 } 1660 sit_i->max_mtime = get_mtime(sbi); 1661 mutex_unlock(&sit_i->sentry_lock); 1662 } 1663 1664 int build_segment_manager(struct f2fs_sb_info *sbi) 1665 { 1666 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 1667 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1668 struct f2fs_sm_info *sm_info; 1669 int err; 1670 1671 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL); 1672 if (!sm_info) 1673 return -ENOMEM; 1674 1675 /* init sm info */ 1676 sbi->sm_info = sm_info; 1677 INIT_LIST_HEAD(&sm_info->wblist_head); 1678 spin_lock_init(&sm_info->wblist_lock); 1679 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 1680 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 1681 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 1682 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 1683 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 1684 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 1685 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 1686 1687 err = build_sit_info(sbi); 1688 if (err) 1689 return err; 1690 err = build_free_segmap(sbi); 1691 if (err) 1692 return err; 1693 err = build_curseg(sbi); 1694 if (err) 1695 return err; 1696 1697 /* reinit free segmap based on SIT */ 1698 build_sit_entries(sbi); 1699 1700 init_free_segmap(sbi); 1701 err = build_dirty_segmap(sbi); 1702 if (err) 1703 return err; 1704 1705 init_min_max_mtime(sbi); 1706 return 0; 1707 } 1708 1709 static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 1710 enum dirty_type dirty_type) 1711 { 1712 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1713 1714 mutex_lock(&dirty_i->seglist_lock); 1715 kfree(dirty_i->dirty_segmap[dirty_type]); 1716 dirty_i->nr_dirty[dirty_type] = 0; 1717 mutex_unlock(&dirty_i->seglist_lock); 1718 } 1719 1720 static void destroy_victim_secmap(struct f2fs_sb_info *sbi) 1721 { 1722 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1723 kfree(dirty_i->victim_secmap); 1724 } 1725 1726 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 1727 { 1728 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1729 int i; 1730 1731 if (!dirty_i) 1732 return; 1733 1734 /* discard pre-free/dirty segments list */ 1735 for (i = 0; i < NR_DIRTY_TYPE; i++) 1736 discard_dirty_segmap(sbi, i); 1737 1738 destroy_victim_secmap(sbi); 1739 SM_I(sbi)->dirty_info = NULL; 1740 kfree(dirty_i); 1741 } 1742 1743 static void destroy_curseg(struct f2fs_sb_info *sbi) 1744 { 1745 struct curseg_info *array = SM_I(sbi)->curseg_array; 1746 int i; 1747 1748 if (!array) 1749 return; 1750 SM_I(sbi)->curseg_array = NULL; 1751 for (i = 0; i < NR_CURSEG_TYPE; i++) 1752 kfree(array[i].sum_blk); 1753 kfree(array); 1754 } 1755 1756 static void destroy_free_segmap(struct f2fs_sb_info *sbi) 1757 { 1758 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 1759 if (!free_i) 1760 return; 1761 SM_I(sbi)->free_info = NULL; 1762 kfree(free_i->free_segmap); 1763 kfree(free_i->free_secmap); 1764 kfree(free_i); 1765 } 1766 1767 static void destroy_sit_info(struct f2fs_sb_info *sbi) 1768 { 1769 struct sit_info *sit_i = SIT_I(sbi); 1770 unsigned int start; 1771 1772 if (!sit_i) 1773 return; 1774 1775 if (sit_i->sentries) { 1776 for (start = 0; start < TOTAL_SEGS(sbi); start++) { 1777 kfree(sit_i->sentries[start].cur_valid_map); 1778 kfree(sit_i->sentries[start].ckpt_valid_map); 1779 } 1780 } 1781 vfree(sit_i->sentries); 1782 vfree(sit_i->sec_entries); 1783 kfree(sit_i->dirty_sentries_bitmap); 1784 1785 SM_I(sbi)->sit_info = NULL; 1786 kfree(sit_i->sit_bitmap); 1787 kfree(sit_i); 1788 } 1789 1790 void destroy_segment_manager(struct f2fs_sb_info *sbi) 1791 { 1792 struct f2fs_sm_info *sm_info = SM_I(sbi); 1793 destroy_dirty_segmap(sbi); 1794 destroy_curseg(sbi); 1795 destroy_free_segmap(sbi); 1796 destroy_sit_info(sbi); 1797 sbi->sm_info = NULL; 1798 kfree(sm_info); 1799 } 1800