segment.c revision 53cf95222fad7a962cc03fb61a33e37bcf4f5c9d
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 22/* 23 * This function balances dirty node and dentry pages. 24 * In addition, it controls garbage collection. 25 */ 26void f2fs_balance_fs(struct f2fs_sb_info *sbi) 27{ 28 /* 29 * We should do GC or end up with checkpoint, if there are so many dirty 30 * dir/node pages without enough free segments. 31 */ 32 if (has_not_enough_free_secs(sbi, 0)) { 33 mutex_lock(&sbi->gc_mutex); 34 f2fs_gc(sbi); 35 } 36} 37 38static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 39 enum dirty_type dirty_type) 40{ 41 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 42 43 /* need not be added */ 44 if (IS_CURSEG(sbi, segno)) 45 return; 46 47 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 48 dirty_i->nr_dirty[dirty_type]++; 49 50 if (dirty_type == DIRTY) { 51 struct seg_entry *sentry = get_seg_entry(sbi, segno); 52 dirty_type = sentry->type; 53 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type])) 54 dirty_i->nr_dirty[dirty_type]++; 55 } 56} 57 58static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno, 59 enum dirty_type dirty_type) 60{ 61 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 62 63 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type])) 64 dirty_i->nr_dirty[dirty_type]--; 65 66 if (dirty_type == DIRTY) { 67 struct seg_entry *sentry = get_seg_entry(sbi, segno); 68 dirty_type = sentry->type; 69 if (test_and_clear_bit(segno, 70 dirty_i->dirty_segmap[dirty_type])) 71 dirty_i->nr_dirty[dirty_type]--; 72 clear_bit(segno, dirty_i->victim_segmap[FG_GC]); 73 clear_bit(segno, dirty_i->victim_segmap[BG_GC]); 74 } 75} 76 77/* 78 * Should not occur error such as -ENOMEM. 79 * Adding dirty entry into seglist is not critical operation. 80 * If a given segment is one of current working segments, it won't be added. 81 */ 82void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno) 83{ 84 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 85 unsigned short valid_blocks; 86 87 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno)) 88 return; 89 90 mutex_lock(&dirty_i->seglist_lock); 91 92 valid_blocks = get_valid_blocks(sbi, segno, 0); 93 94 if (valid_blocks == 0) { 95 __locate_dirty_segment(sbi, segno, PRE); 96 __remove_dirty_segment(sbi, segno, DIRTY); 97 } else if (valid_blocks < sbi->blocks_per_seg) { 98 __locate_dirty_segment(sbi, segno, DIRTY); 99 } else { 100 /* Recovery routine with SSR needs this */ 101 __remove_dirty_segment(sbi, segno, DIRTY); 102 } 103 104 mutex_unlock(&dirty_i->seglist_lock); 105 return; 106} 107 108/* 109 * Should call clear_prefree_segments after checkpoint is done. 110 */ 111static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi) 112{ 113 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 114 unsigned int segno, offset = 0; 115 unsigned int total_segs = TOTAL_SEGS(sbi); 116 117 mutex_lock(&dirty_i->seglist_lock); 118 while (1) { 119 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs, 120 offset); 121 if (segno >= total_segs) 122 break; 123 __set_test_and_free(sbi, segno); 124 offset = segno + 1; 125 } 126 mutex_unlock(&dirty_i->seglist_lock); 127} 128 129void clear_prefree_segments(struct f2fs_sb_info *sbi) 130{ 131 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 132 unsigned int segno, offset = 0; 133 unsigned int total_segs = TOTAL_SEGS(sbi); 134 135 mutex_lock(&dirty_i->seglist_lock); 136 while (1) { 137 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs, 138 offset); 139 if (segno >= total_segs) 140 break; 141 142 offset = segno + 1; 143 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[PRE])) 144 dirty_i->nr_dirty[PRE]--; 145 146 /* Let's use trim */ 147 if (test_opt(sbi, DISCARD)) 148 blkdev_issue_discard(sbi->sb->s_bdev, 149 START_BLOCK(sbi, segno) << 150 sbi->log_sectors_per_block, 151 1 << (sbi->log_sectors_per_block + 152 sbi->log_blocks_per_seg), 153 GFP_NOFS, 0); 154 } 155 mutex_unlock(&dirty_i->seglist_lock); 156} 157 158static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno) 159{ 160 struct sit_info *sit_i = SIT_I(sbi); 161 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) 162 sit_i->dirty_sentries++; 163} 164 165static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type, 166 unsigned int segno, int modified) 167{ 168 struct seg_entry *se = get_seg_entry(sbi, segno); 169 se->type = type; 170 if (modified) 171 __mark_sit_entry_dirty(sbi, segno); 172} 173 174static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del) 175{ 176 struct seg_entry *se; 177 unsigned int segno, offset; 178 long int new_vblocks; 179 180 segno = GET_SEGNO(sbi, blkaddr); 181 182 se = get_seg_entry(sbi, segno); 183 new_vblocks = se->valid_blocks + del; 184 offset = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) & (sbi->blocks_per_seg - 1); 185 186 BUG_ON((new_vblocks >> (sizeof(unsigned short) << 3) || 187 (new_vblocks > sbi->blocks_per_seg))); 188 189 se->valid_blocks = new_vblocks; 190 se->mtime = get_mtime(sbi); 191 SIT_I(sbi)->max_mtime = se->mtime; 192 193 /* Update valid block bitmap */ 194 if (del > 0) { 195 if (f2fs_set_bit(offset, se->cur_valid_map)) 196 BUG(); 197 } else { 198 if (!f2fs_clear_bit(offset, se->cur_valid_map)) 199 BUG(); 200 } 201 if (!f2fs_test_bit(offset, se->ckpt_valid_map)) 202 se->ckpt_valid_blocks += del; 203 204 __mark_sit_entry_dirty(sbi, segno); 205 206 /* update total number of valid blocks to be written in ckpt area */ 207 SIT_I(sbi)->written_valid_blocks += del; 208 209 if (sbi->segs_per_sec > 1) 210 get_sec_entry(sbi, segno)->valid_blocks += del; 211} 212 213static void refresh_sit_entry(struct f2fs_sb_info *sbi, 214 block_t old_blkaddr, block_t new_blkaddr) 215{ 216 update_sit_entry(sbi, new_blkaddr, 1); 217 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) 218 update_sit_entry(sbi, old_blkaddr, -1); 219} 220 221void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr) 222{ 223 unsigned int segno = GET_SEGNO(sbi, addr); 224 struct sit_info *sit_i = SIT_I(sbi); 225 226 BUG_ON(addr == NULL_ADDR); 227 if (addr == NEW_ADDR) 228 return; 229 230 /* add it into sit main buffer */ 231 mutex_lock(&sit_i->sentry_lock); 232 233 update_sit_entry(sbi, addr, -1); 234 235 /* add it into dirty seglist */ 236 locate_dirty_segment(sbi, segno); 237 238 mutex_unlock(&sit_i->sentry_lock); 239} 240 241/* 242 * This function should be resided under the curseg_mutex lock 243 */ 244static void __add_sum_entry(struct f2fs_sb_info *sbi, int type, 245 struct f2fs_summary *sum, unsigned short offset) 246{ 247 struct curseg_info *curseg = CURSEG_I(sbi, type); 248 void *addr = curseg->sum_blk; 249 addr += offset * sizeof(struct f2fs_summary); 250 memcpy(addr, sum, sizeof(struct f2fs_summary)); 251 return; 252} 253 254/* 255 * Calculate the number of current summary pages for writing 256 */ 257int npages_for_summary_flush(struct f2fs_sb_info *sbi) 258{ 259 int total_size_bytes = 0; 260 int valid_sum_count = 0; 261 int i, sum_space; 262 263 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 264 if (sbi->ckpt->alloc_type[i] == SSR) 265 valid_sum_count += sbi->blocks_per_seg; 266 else 267 valid_sum_count += curseg_blkoff(sbi, i); 268 } 269 270 total_size_bytes = valid_sum_count * (SUMMARY_SIZE + 1) 271 + sizeof(struct nat_journal) + 2 272 + sizeof(struct sit_journal) + 2; 273 sum_space = PAGE_CACHE_SIZE - SUM_FOOTER_SIZE; 274 if (total_size_bytes < sum_space) 275 return 1; 276 else if (total_size_bytes < 2 * sum_space) 277 return 2; 278 return 3; 279} 280 281/* 282 * Caller should put this summary page 283 */ 284struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno) 285{ 286 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno)); 287} 288 289static void write_sum_page(struct f2fs_sb_info *sbi, 290 struct f2fs_summary_block *sum_blk, block_t blk_addr) 291{ 292 struct page *page = grab_meta_page(sbi, blk_addr); 293 void *kaddr = page_address(page); 294 memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE); 295 set_page_dirty(page); 296 f2fs_put_page(page, 1); 297} 298 299static unsigned int check_prefree_segments(struct f2fs_sb_info *sbi, 300 int ofs_unit, int type) 301{ 302 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 303 unsigned long *prefree_segmap = dirty_i->dirty_segmap[PRE]; 304 unsigned int segno, next_segno, i; 305 int ofs = 0; 306 307 /* 308 * If there is not enough reserved sections, 309 * we should not reuse prefree segments. 310 */ 311 if (has_not_enough_free_secs(sbi, 0)) 312 return NULL_SEGNO; 313 314 /* 315 * NODE page should not reuse prefree segment, 316 * since those information is used for SPOR. 317 */ 318 if (IS_NODESEG(type)) 319 return NULL_SEGNO; 320next: 321 segno = find_next_bit(prefree_segmap, TOTAL_SEGS(sbi), ofs++); 322 ofs = ((segno / ofs_unit) * ofs_unit) + ofs_unit; 323 if (segno < TOTAL_SEGS(sbi)) { 324 /* skip intermediate segments in a section */ 325 if (segno % ofs_unit) 326 goto next; 327 328 /* skip if whole section is not prefree */ 329 next_segno = find_next_zero_bit(prefree_segmap, 330 TOTAL_SEGS(sbi), segno + 1); 331 if (next_segno - segno < ofs_unit) 332 goto next; 333 334 /* skip if whole section was not free at the last checkpoint */ 335 for (i = 0; i < ofs_unit; i++) 336 if (get_seg_entry(sbi, segno)->ckpt_valid_blocks) 337 goto next; 338 return segno; 339 } 340 return NULL_SEGNO; 341} 342 343/* 344 * Find a new segment from the free segments bitmap to right order 345 * This function should be returned with success, otherwise BUG 346 */ 347static void get_new_segment(struct f2fs_sb_info *sbi, 348 unsigned int *newseg, bool new_sec, int dir) 349{ 350 struct free_segmap_info *free_i = FREE_I(sbi); 351 unsigned int segno, secno, zoneno; 352 unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone; 353 unsigned int hint = *newseg / sbi->segs_per_sec; 354 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg); 355 unsigned int left_start = hint; 356 bool init = true; 357 int go_left = 0; 358 int i; 359 360 write_lock(&free_i->segmap_lock); 361 362 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) { 363 segno = find_next_zero_bit(free_i->free_segmap, 364 TOTAL_SEGS(sbi), *newseg + 1); 365 if (segno < TOTAL_SEGS(sbi)) 366 goto got_it; 367 } 368find_other_zone: 369 secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint); 370 if (secno >= TOTAL_SECS(sbi)) { 371 if (dir == ALLOC_RIGHT) { 372 secno = find_next_zero_bit(free_i->free_secmap, 373 TOTAL_SECS(sbi), 0); 374 BUG_ON(secno >= TOTAL_SECS(sbi)); 375 } else { 376 go_left = 1; 377 left_start = hint - 1; 378 } 379 } 380 if (go_left == 0) 381 goto skip_left; 382 383 while (test_bit(left_start, free_i->free_secmap)) { 384 if (left_start > 0) { 385 left_start--; 386 continue; 387 } 388 left_start = find_next_zero_bit(free_i->free_secmap, 389 TOTAL_SECS(sbi), 0); 390 BUG_ON(left_start >= TOTAL_SECS(sbi)); 391 break; 392 } 393 secno = left_start; 394skip_left: 395 hint = secno; 396 segno = secno * sbi->segs_per_sec; 397 zoneno = secno / sbi->secs_per_zone; 398 399 /* give up on finding another zone */ 400 if (!init) 401 goto got_it; 402 if (sbi->secs_per_zone == 1) 403 goto got_it; 404 if (zoneno == old_zoneno) 405 goto got_it; 406 if (dir == ALLOC_LEFT) { 407 if (!go_left && zoneno + 1 >= total_zones) 408 goto got_it; 409 if (go_left && zoneno == 0) 410 goto got_it; 411 } 412 for (i = 0; i < NR_CURSEG_TYPE; i++) 413 if (CURSEG_I(sbi, i)->zone == zoneno) 414 break; 415 416 if (i < NR_CURSEG_TYPE) { 417 /* zone is in user, try another */ 418 if (go_left) 419 hint = zoneno * sbi->secs_per_zone - 1; 420 else if (zoneno + 1 >= total_zones) 421 hint = 0; 422 else 423 hint = (zoneno + 1) * sbi->secs_per_zone; 424 init = false; 425 goto find_other_zone; 426 } 427got_it: 428 /* set it as dirty segment in free segmap */ 429 BUG_ON(test_bit(segno, free_i->free_segmap)); 430 __set_inuse(sbi, segno); 431 *newseg = segno; 432 write_unlock(&free_i->segmap_lock); 433} 434 435static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified) 436{ 437 struct curseg_info *curseg = CURSEG_I(sbi, type); 438 struct summary_footer *sum_footer; 439 440 curseg->segno = curseg->next_segno; 441 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno); 442 curseg->next_blkoff = 0; 443 curseg->next_segno = NULL_SEGNO; 444 445 sum_footer = &(curseg->sum_blk->footer); 446 memset(sum_footer, 0, sizeof(struct summary_footer)); 447 if (IS_DATASEG(type)) 448 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA); 449 if (IS_NODESEG(type)) 450 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE); 451 __set_sit_entry_type(sbi, type, curseg->segno, modified); 452} 453 454/* 455 * Allocate a current working segment. 456 * This function always allocates a free segment in LFS manner. 457 */ 458static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec) 459{ 460 struct curseg_info *curseg = CURSEG_I(sbi, type); 461 unsigned int segno = curseg->segno; 462 int dir = ALLOC_LEFT; 463 464 write_sum_page(sbi, curseg->sum_blk, 465 GET_SUM_BLOCK(sbi, curseg->segno)); 466 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA) 467 dir = ALLOC_RIGHT; 468 469 if (test_opt(sbi, NOHEAP)) 470 dir = ALLOC_RIGHT; 471 472 get_new_segment(sbi, &segno, new_sec, dir); 473 curseg->next_segno = segno; 474 reset_curseg(sbi, type, 1); 475 curseg->alloc_type = LFS; 476} 477 478static void __next_free_blkoff(struct f2fs_sb_info *sbi, 479 struct curseg_info *seg, block_t start) 480{ 481 struct seg_entry *se = get_seg_entry(sbi, seg->segno); 482 block_t ofs; 483 for (ofs = start; ofs < sbi->blocks_per_seg; ofs++) { 484 if (!f2fs_test_bit(ofs, se->ckpt_valid_map) 485 && !f2fs_test_bit(ofs, se->cur_valid_map)) 486 break; 487 } 488 seg->next_blkoff = ofs; 489} 490 491/* 492 * If a segment is written by LFS manner, next block offset is just obtained 493 * by increasing the current block offset. However, if a segment is written by 494 * SSR manner, next block offset obtained by calling __next_free_blkoff 495 */ 496static void __refresh_next_blkoff(struct f2fs_sb_info *sbi, 497 struct curseg_info *seg) 498{ 499 if (seg->alloc_type == SSR) 500 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1); 501 else 502 seg->next_blkoff++; 503} 504 505/* 506 * This function always allocates a used segment (from dirty seglist) by SSR 507 * manner, so it should recover the existing segment information of valid blocks 508 */ 509static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse) 510{ 511 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 512 struct curseg_info *curseg = CURSEG_I(sbi, type); 513 unsigned int new_segno = curseg->next_segno; 514 struct f2fs_summary_block *sum_node; 515 struct page *sum_page; 516 517 write_sum_page(sbi, curseg->sum_blk, 518 GET_SUM_BLOCK(sbi, curseg->segno)); 519 __set_test_and_inuse(sbi, new_segno); 520 521 mutex_lock(&dirty_i->seglist_lock); 522 __remove_dirty_segment(sbi, new_segno, PRE); 523 __remove_dirty_segment(sbi, new_segno, DIRTY); 524 mutex_unlock(&dirty_i->seglist_lock); 525 526 reset_curseg(sbi, type, 1); 527 curseg->alloc_type = SSR; 528 __next_free_blkoff(sbi, curseg, 0); 529 530 if (reuse) { 531 sum_page = get_sum_page(sbi, new_segno); 532 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 533 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE); 534 f2fs_put_page(sum_page, 1); 535 } 536} 537 538static int get_ssr_segment(struct f2fs_sb_info *sbi, int type) 539{ 540 struct curseg_info *curseg = CURSEG_I(sbi, type); 541 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops; 542 543 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0)) 544 return v_ops->get_victim(sbi, 545 &(curseg)->next_segno, BG_GC, type, SSR); 546 547 /* For data segments, let's do SSR more intensively */ 548 for (; type >= CURSEG_HOT_DATA; type--) 549 if (v_ops->get_victim(sbi, &(curseg)->next_segno, 550 BG_GC, type, SSR)) 551 return 1; 552 return 0; 553} 554 555/* 556 * flush out current segment and replace it with new segment 557 * This function should be returned with success, otherwise BUG 558 */ 559static void allocate_segment_by_default(struct f2fs_sb_info *sbi, 560 int type, bool force) 561{ 562 struct curseg_info *curseg = CURSEG_I(sbi, type); 563 unsigned int ofs_unit; 564 565 if (force) { 566 new_curseg(sbi, type, true); 567 goto out; 568 } 569 570 ofs_unit = need_SSR(sbi) ? 1 : sbi->segs_per_sec; 571 curseg->next_segno = check_prefree_segments(sbi, ofs_unit, type); 572 573 if (curseg->next_segno != NULL_SEGNO) 574 change_curseg(sbi, type, false); 575 else if (type == CURSEG_WARM_NODE) 576 new_curseg(sbi, type, false); 577 else if (need_SSR(sbi) && get_ssr_segment(sbi, type)) 578 change_curseg(sbi, type, true); 579 else 580 new_curseg(sbi, type, false); 581out: 582 sbi->segment_count[curseg->alloc_type]++; 583} 584 585void allocate_new_segments(struct f2fs_sb_info *sbi) 586{ 587 struct curseg_info *curseg; 588 unsigned int old_curseg; 589 int i; 590 591 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 592 curseg = CURSEG_I(sbi, i); 593 old_curseg = curseg->segno; 594 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true); 595 locate_dirty_segment(sbi, old_curseg); 596 } 597} 598 599static const struct segment_allocation default_salloc_ops = { 600 .allocate_segment = allocate_segment_by_default, 601}; 602 603static void f2fs_end_io_write(struct bio *bio, int err) 604{ 605 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); 606 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; 607 struct bio_private *p = bio->bi_private; 608 609 do { 610 struct page *page = bvec->bv_page; 611 612 if (--bvec >= bio->bi_io_vec) 613 prefetchw(&bvec->bv_page->flags); 614 if (!uptodate) { 615 SetPageError(page); 616 if (page->mapping) 617 set_bit(AS_EIO, &page->mapping->flags); 618 set_ckpt_flags(p->sbi->ckpt, CP_ERROR_FLAG); 619 p->sbi->sb->s_flags |= MS_RDONLY; 620 } 621 end_page_writeback(page); 622 dec_page_count(p->sbi, F2FS_WRITEBACK); 623 } while (bvec >= bio->bi_io_vec); 624 625 if (p->is_sync) 626 complete(p->wait); 627 kfree(p); 628 bio_put(bio); 629} 630 631struct bio *f2fs_bio_alloc(struct block_device *bdev, int npages) 632{ 633 struct bio *bio; 634 struct bio_private *priv; 635retry: 636 priv = kmalloc(sizeof(struct bio_private), GFP_NOFS); 637 if (!priv) { 638 cond_resched(); 639 goto retry; 640 } 641 642 /* No failure on bio allocation */ 643 bio = bio_alloc(GFP_NOIO, npages); 644 bio->bi_bdev = bdev; 645 bio->bi_private = priv; 646 return bio; 647} 648 649static void do_submit_bio(struct f2fs_sb_info *sbi, 650 enum page_type type, bool sync) 651{ 652 int rw = sync ? WRITE_SYNC : WRITE; 653 enum page_type btype = type > META ? META : type; 654 655 if (type >= META_FLUSH) 656 rw = WRITE_FLUSH_FUA; 657 658 if (sbi->bio[btype]) { 659 struct bio_private *p = sbi->bio[btype]->bi_private; 660 p->sbi = sbi; 661 sbi->bio[btype]->bi_end_io = f2fs_end_io_write; 662 if (type == META_FLUSH) { 663 DECLARE_COMPLETION_ONSTACK(wait); 664 p->is_sync = true; 665 p->wait = &wait; 666 submit_bio(rw, sbi->bio[btype]); 667 wait_for_completion(&wait); 668 } else { 669 p->is_sync = false; 670 submit_bio(rw, sbi->bio[btype]); 671 } 672 sbi->bio[btype] = NULL; 673 } 674} 675 676void f2fs_submit_bio(struct f2fs_sb_info *sbi, enum page_type type, bool sync) 677{ 678 down_write(&sbi->bio_sem); 679 do_submit_bio(sbi, type, sync); 680 up_write(&sbi->bio_sem); 681} 682 683static void submit_write_page(struct f2fs_sb_info *sbi, struct page *page, 684 block_t blk_addr, enum page_type type) 685{ 686 struct block_device *bdev = sbi->sb->s_bdev; 687 688 verify_block_addr(sbi, blk_addr); 689 690 down_write(&sbi->bio_sem); 691 692 inc_page_count(sbi, F2FS_WRITEBACK); 693 694 if (sbi->bio[type] && sbi->last_block_in_bio[type] != blk_addr - 1) 695 do_submit_bio(sbi, type, false); 696alloc_new: 697 if (sbi->bio[type] == NULL) { 698 sbi->bio[type] = f2fs_bio_alloc(bdev, bio_get_nr_vecs(bdev)); 699 sbi->bio[type]->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr); 700 /* 701 * The end_io will be assigned at the sumbission phase. 702 * Until then, let bio_add_page() merge consecutive IOs as much 703 * as possible. 704 */ 705 } 706 707 if (bio_add_page(sbi->bio[type], page, PAGE_CACHE_SIZE, 0) < 708 PAGE_CACHE_SIZE) { 709 do_submit_bio(sbi, type, false); 710 goto alloc_new; 711 } 712 713 sbi->last_block_in_bio[type] = blk_addr; 714 715 up_write(&sbi->bio_sem); 716} 717 718static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type) 719{ 720 struct curseg_info *curseg = CURSEG_I(sbi, type); 721 if (curseg->next_blkoff < sbi->blocks_per_seg) 722 return true; 723 return false; 724} 725 726static int __get_segment_type_2(struct page *page, enum page_type p_type) 727{ 728 if (p_type == DATA) 729 return CURSEG_HOT_DATA; 730 else 731 return CURSEG_HOT_NODE; 732} 733 734static int __get_segment_type_4(struct page *page, enum page_type p_type) 735{ 736 if (p_type == DATA) { 737 struct inode *inode = page->mapping->host; 738 739 if (S_ISDIR(inode->i_mode)) 740 return CURSEG_HOT_DATA; 741 else 742 return CURSEG_COLD_DATA; 743 } else { 744 if (IS_DNODE(page) && !is_cold_node(page)) 745 return CURSEG_HOT_NODE; 746 else 747 return CURSEG_COLD_NODE; 748 } 749} 750 751static int __get_segment_type_6(struct page *page, enum page_type p_type) 752{ 753 if (p_type == DATA) { 754 struct inode *inode = page->mapping->host; 755 756 if (S_ISDIR(inode->i_mode)) 757 return CURSEG_HOT_DATA; 758 else if (is_cold_data(page) || is_cold_file(inode)) 759 return CURSEG_COLD_DATA; 760 else 761 return CURSEG_WARM_DATA; 762 } else { 763 if (IS_DNODE(page)) 764 return is_cold_node(page) ? CURSEG_WARM_NODE : 765 CURSEG_HOT_NODE; 766 else 767 return CURSEG_COLD_NODE; 768 } 769} 770 771static int __get_segment_type(struct page *page, enum page_type p_type) 772{ 773 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb); 774 switch (sbi->active_logs) { 775 case 2: 776 return __get_segment_type_2(page, p_type); 777 case 4: 778 return __get_segment_type_4(page, p_type); 779 } 780 /* NR_CURSEG_TYPE(6) logs by default */ 781 BUG_ON(sbi->active_logs != NR_CURSEG_TYPE); 782 return __get_segment_type_6(page, p_type); 783} 784 785static void do_write_page(struct f2fs_sb_info *sbi, struct page *page, 786 block_t old_blkaddr, block_t *new_blkaddr, 787 struct f2fs_summary *sum, enum page_type p_type) 788{ 789 struct sit_info *sit_i = SIT_I(sbi); 790 struct curseg_info *curseg; 791 unsigned int old_cursegno; 792 int type; 793 794 type = __get_segment_type(page, p_type); 795 curseg = CURSEG_I(sbi, type); 796 797 mutex_lock(&curseg->curseg_mutex); 798 799 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 800 old_cursegno = curseg->segno; 801 802 /* 803 * __add_sum_entry should be resided under the curseg_mutex 804 * because, this function updates a summary entry in the 805 * current summary block. 806 */ 807 __add_sum_entry(sbi, type, sum, curseg->next_blkoff); 808 809 mutex_lock(&sit_i->sentry_lock); 810 __refresh_next_blkoff(sbi, curseg); 811 sbi->block_count[curseg->alloc_type]++; 812 813 /* 814 * SIT information should be updated before segment allocation, 815 * since SSR needs latest valid block information. 816 */ 817 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr); 818 819 if (!__has_curseg_space(sbi, type)) 820 sit_i->s_ops->allocate_segment(sbi, type, false); 821 822 locate_dirty_segment(sbi, old_cursegno); 823 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 824 mutex_unlock(&sit_i->sentry_lock); 825 826 if (p_type == NODE) 827 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); 828 829 /* writeout dirty page into bdev */ 830 submit_write_page(sbi, page, *new_blkaddr, p_type); 831 832 mutex_unlock(&curseg->curseg_mutex); 833} 834 835void write_meta_page(struct f2fs_sb_info *sbi, struct page *page) 836{ 837 set_page_writeback(page); 838 submit_write_page(sbi, page, page->index, META); 839} 840 841void write_node_page(struct f2fs_sb_info *sbi, struct page *page, 842 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr) 843{ 844 struct f2fs_summary sum; 845 set_summary(&sum, nid, 0, 0); 846 do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, NODE); 847} 848 849void write_data_page(struct inode *inode, struct page *page, 850 struct dnode_of_data *dn, block_t old_blkaddr, 851 block_t *new_blkaddr) 852{ 853 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 854 struct f2fs_summary sum; 855 struct node_info ni; 856 857 BUG_ON(old_blkaddr == NULL_ADDR); 858 get_node_info(sbi, dn->nid, &ni); 859 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); 860 861 do_write_page(sbi, page, old_blkaddr, 862 new_blkaddr, &sum, DATA); 863} 864 865void rewrite_data_page(struct f2fs_sb_info *sbi, struct page *page, 866 block_t old_blk_addr) 867{ 868 submit_write_page(sbi, page, old_blk_addr, DATA); 869} 870 871void recover_data_page(struct f2fs_sb_info *sbi, 872 struct page *page, struct f2fs_summary *sum, 873 block_t old_blkaddr, block_t new_blkaddr) 874{ 875 struct sit_info *sit_i = SIT_I(sbi); 876 struct curseg_info *curseg; 877 unsigned int segno, old_cursegno; 878 struct seg_entry *se; 879 int type; 880 881 segno = GET_SEGNO(sbi, new_blkaddr); 882 se = get_seg_entry(sbi, segno); 883 type = se->type; 884 885 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) { 886 if (old_blkaddr == NULL_ADDR) 887 type = CURSEG_COLD_DATA; 888 else 889 type = CURSEG_WARM_DATA; 890 } 891 curseg = CURSEG_I(sbi, type); 892 893 mutex_lock(&curseg->curseg_mutex); 894 mutex_lock(&sit_i->sentry_lock); 895 896 old_cursegno = curseg->segno; 897 898 /* change the current segment */ 899 if (segno != curseg->segno) { 900 curseg->next_segno = segno; 901 change_curseg(sbi, type, true); 902 } 903 904 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) & 905 (sbi->blocks_per_seg - 1); 906 __add_sum_entry(sbi, type, sum, curseg->next_blkoff); 907 908 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr); 909 910 locate_dirty_segment(sbi, old_cursegno); 911 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 912 913 mutex_unlock(&sit_i->sentry_lock); 914 mutex_unlock(&curseg->curseg_mutex); 915} 916 917void rewrite_node_page(struct f2fs_sb_info *sbi, 918 struct page *page, struct f2fs_summary *sum, 919 block_t old_blkaddr, block_t new_blkaddr) 920{ 921 struct sit_info *sit_i = SIT_I(sbi); 922 int type = CURSEG_WARM_NODE; 923 struct curseg_info *curseg; 924 unsigned int segno, old_cursegno; 925 block_t next_blkaddr = next_blkaddr_of_node(page); 926 unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr); 927 928 curseg = CURSEG_I(sbi, type); 929 930 mutex_lock(&curseg->curseg_mutex); 931 mutex_lock(&sit_i->sentry_lock); 932 933 segno = GET_SEGNO(sbi, new_blkaddr); 934 old_cursegno = curseg->segno; 935 936 /* change the current segment */ 937 if (segno != curseg->segno) { 938 curseg->next_segno = segno; 939 change_curseg(sbi, type, true); 940 } 941 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) & 942 (sbi->blocks_per_seg - 1); 943 __add_sum_entry(sbi, type, sum, curseg->next_blkoff); 944 945 /* change the current log to the next block addr in advance */ 946 if (next_segno != segno) { 947 curseg->next_segno = next_segno; 948 change_curseg(sbi, type, true); 949 } 950 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, next_blkaddr) & 951 (sbi->blocks_per_seg - 1); 952 953 /* rewrite node page */ 954 set_page_writeback(page); 955 submit_write_page(sbi, page, new_blkaddr, NODE); 956 f2fs_submit_bio(sbi, NODE, true); 957 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr); 958 959 locate_dirty_segment(sbi, old_cursegno); 960 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr)); 961 962 mutex_unlock(&sit_i->sentry_lock); 963 mutex_unlock(&curseg->curseg_mutex); 964} 965 966static int read_compacted_summaries(struct f2fs_sb_info *sbi) 967{ 968 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 969 struct curseg_info *seg_i; 970 unsigned char *kaddr; 971 struct page *page; 972 block_t start; 973 int i, j, offset; 974 975 start = start_sum_block(sbi); 976 977 page = get_meta_page(sbi, start++); 978 kaddr = (unsigned char *)page_address(page); 979 980 /* Step 1: restore nat cache */ 981 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 982 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE); 983 984 /* Step 2: restore sit cache */ 985 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 986 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE, 987 SUM_JOURNAL_SIZE); 988 offset = 2 * SUM_JOURNAL_SIZE; 989 990 /* Step 3: restore summary entries */ 991 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 992 unsigned short blk_off; 993 unsigned int segno; 994 995 seg_i = CURSEG_I(sbi, i); 996 segno = le32_to_cpu(ckpt->cur_data_segno[i]); 997 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]); 998 seg_i->next_segno = segno; 999 reset_curseg(sbi, i, 0); 1000 seg_i->alloc_type = ckpt->alloc_type[i]; 1001 seg_i->next_blkoff = blk_off; 1002 1003 if (seg_i->alloc_type == SSR) 1004 blk_off = sbi->blocks_per_seg; 1005 1006 for (j = 0; j < blk_off; j++) { 1007 struct f2fs_summary *s; 1008 s = (struct f2fs_summary *)(kaddr + offset); 1009 seg_i->sum_blk->entries[j] = *s; 1010 offset += SUMMARY_SIZE; 1011 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE - 1012 SUM_FOOTER_SIZE) 1013 continue; 1014 1015 f2fs_put_page(page, 1); 1016 page = NULL; 1017 1018 page = get_meta_page(sbi, start++); 1019 kaddr = (unsigned char *)page_address(page); 1020 offset = 0; 1021 } 1022 } 1023 f2fs_put_page(page, 1); 1024 return 0; 1025} 1026 1027static int read_normal_summaries(struct f2fs_sb_info *sbi, int type) 1028{ 1029 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1030 struct f2fs_summary_block *sum; 1031 struct curseg_info *curseg; 1032 struct page *new; 1033 unsigned short blk_off; 1034 unsigned int segno = 0; 1035 block_t blk_addr = 0; 1036 1037 /* get segment number and block addr */ 1038 if (IS_DATASEG(type)) { 1039 segno = le32_to_cpu(ckpt->cur_data_segno[type]); 1040 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type - 1041 CURSEG_HOT_DATA]); 1042 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) 1043 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type); 1044 else 1045 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type); 1046 } else { 1047 segno = le32_to_cpu(ckpt->cur_node_segno[type - 1048 CURSEG_HOT_NODE]); 1049 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type - 1050 CURSEG_HOT_NODE]); 1051 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) 1052 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE, 1053 type - CURSEG_HOT_NODE); 1054 else 1055 blk_addr = GET_SUM_BLOCK(sbi, segno); 1056 } 1057 1058 new = get_meta_page(sbi, blk_addr); 1059 sum = (struct f2fs_summary_block *)page_address(new); 1060 1061 if (IS_NODESEG(type)) { 1062 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) { 1063 struct f2fs_summary *ns = &sum->entries[0]; 1064 int i; 1065 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) { 1066 ns->version = 0; 1067 ns->ofs_in_node = 0; 1068 } 1069 } else { 1070 if (restore_node_summary(sbi, segno, sum)) { 1071 f2fs_put_page(new, 1); 1072 return -EINVAL; 1073 } 1074 } 1075 } 1076 1077 /* set uncompleted segment to curseg */ 1078 curseg = CURSEG_I(sbi, type); 1079 mutex_lock(&curseg->curseg_mutex); 1080 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE); 1081 curseg->next_segno = segno; 1082 reset_curseg(sbi, type, 0); 1083 curseg->alloc_type = ckpt->alloc_type[type]; 1084 curseg->next_blkoff = blk_off; 1085 mutex_unlock(&curseg->curseg_mutex); 1086 f2fs_put_page(new, 1); 1087 return 0; 1088} 1089 1090static int restore_curseg_summaries(struct f2fs_sb_info *sbi) 1091{ 1092 int type = CURSEG_HOT_DATA; 1093 1094 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) { 1095 /* restore for compacted data summary */ 1096 if (read_compacted_summaries(sbi)) 1097 return -EINVAL; 1098 type = CURSEG_HOT_NODE; 1099 } 1100 1101 for (; type <= CURSEG_COLD_NODE; type++) 1102 if (read_normal_summaries(sbi, type)) 1103 return -EINVAL; 1104 return 0; 1105} 1106 1107static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr) 1108{ 1109 struct page *page; 1110 unsigned char *kaddr; 1111 struct f2fs_summary *summary; 1112 struct curseg_info *seg_i; 1113 int written_size = 0; 1114 int i, j; 1115 1116 page = grab_meta_page(sbi, blkaddr++); 1117 kaddr = (unsigned char *)page_address(page); 1118 1119 /* Step 1: write nat cache */ 1120 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA); 1121 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE); 1122 written_size += SUM_JOURNAL_SIZE; 1123 1124 /* Step 2: write sit cache */ 1125 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA); 1126 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits, 1127 SUM_JOURNAL_SIZE); 1128 written_size += SUM_JOURNAL_SIZE; 1129 1130 set_page_dirty(page); 1131 1132 /* Step 3: write summary entries */ 1133 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 1134 unsigned short blkoff; 1135 seg_i = CURSEG_I(sbi, i); 1136 if (sbi->ckpt->alloc_type[i] == SSR) 1137 blkoff = sbi->blocks_per_seg; 1138 else 1139 blkoff = curseg_blkoff(sbi, i); 1140 1141 for (j = 0; j < blkoff; j++) { 1142 if (!page) { 1143 page = grab_meta_page(sbi, blkaddr++); 1144 kaddr = (unsigned char *)page_address(page); 1145 written_size = 0; 1146 } 1147 summary = (struct f2fs_summary *)(kaddr + written_size); 1148 *summary = seg_i->sum_blk->entries[j]; 1149 written_size += SUMMARY_SIZE; 1150 set_page_dirty(page); 1151 1152 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE - 1153 SUM_FOOTER_SIZE) 1154 continue; 1155 1156 f2fs_put_page(page, 1); 1157 page = NULL; 1158 } 1159 } 1160 if (page) 1161 f2fs_put_page(page, 1); 1162} 1163 1164static void write_normal_summaries(struct f2fs_sb_info *sbi, 1165 block_t blkaddr, int type) 1166{ 1167 int i, end; 1168 if (IS_DATASEG(type)) 1169 end = type + NR_CURSEG_DATA_TYPE; 1170 else 1171 end = type + NR_CURSEG_NODE_TYPE; 1172 1173 for (i = type; i < end; i++) { 1174 struct curseg_info *sum = CURSEG_I(sbi, i); 1175 mutex_lock(&sum->curseg_mutex); 1176 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type)); 1177 mutex_unlock(&sum->curseg_mutex); 1178 } 1179} 1180 1181void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 1182{ 1183 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) 1184 write_compacted_summaries(sbi, start_blk); 1185 else 1186 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA); 1187} 1188 1189void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk) 1190{ 1191 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG)) 1192 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE); 1193 return; 1194} 1195 1196int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type, 1197 unsigned int val, int alloc) 1198{ 1199 int i; 1200 1201 if (type == NAT_JOURNAL) { 1202 for (i = 0; i < nats_in_cursum(sum); i++) { 1203 if (le32_to_cpu(nid_in_journal(sum, i)) == val) 1204 return i; 1205 } 1206 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) 1207 return update_nats_in_cursum(sum, 1); 1208 } else if (type == SIT_JOURNAL) { 1209 for (i = 0; i < sits_in_cursum(sum); i++) 1210 if (le32_to_cpu(segno_in_journal(sum, i)) == val) 1211 return i; 1212 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES) 1213 return update_sits_in_cursum(sum, 1); 1214 } 1215 return -1; 1216} 1217 1218static struct page *get_current_sit_page(struct f2fs_sb_info *sbi, 1219 unsigned int segno) 1220{ 1221 struct sit_info *sit_i = SIT_I(sbi); 1222 unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno); 1223 block_t blk_addr = sit_i->sit_base_addr + offset; 1224 1225 check_seg_range(sbi, segno); 1226 1227 /* calculate sit block address */ 1228 if (f2fs_test_bit(offset, sit_i->sit_bitmap)) 1229 blk_addr += sit_i->sit_blocks; 1230 1231 return get_meta_page(sbi, blk_addr); 1232} 1233 1234static struct page *get_next_sit_page(struct f2fs_sb_info *sbi, 1235 unsigned int start) 1236{ 1237 struct sit_info *sit_i = SIT_I(sbi); 1238 struct page *src_page, *dst_page; 1239 pgoff_t src_off, dst_off; 1240 void *src_addr, *dst_addr; 1241 1242 src_off = current_sit_addr(sbi, start); 1243 dst_off = next_sit_addr(sbi, src_off); 1244 1245 /* get current sit block page without lock */ 1246 src_page = get_meta_page(sbi, src_off); 1247 dst_page = grab_meta_page(sbi, dst_off); 1248 BUG_ON(PageDirty(src_page)); 1249 1250 src_addr = page_address(src_page); 1251 dst_addr = page_address(dst_page); 1252 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE); 1253 1254 set_page_dirty(dst_page); 1255 f2fs_put_page(src_page, 1); 1256 1257 set_to_next_sit(sit_i, start); 1258 1259 return dst_page; 1260} 1261 1262static bool flush_sits_in_journal(struct f2fs_sb_info *sbi) 1263{ 1264 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 1265 struct f2fs_summary_block *sum = curseg->sum_blk; 1266 int i; 1267 1268 /* 1269 * If the journal area in the current summary is full of sit entries, 1270 * all the sit entries will be flushed. Otherwise the sit entries 1271 * are not able to replace with newly hot sit entries. 1272 */ 1273 if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) { 1274 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) { 1275 unsigned int segno; 1276 segno = le32_to_cpu(segno_in_journal(sum, i)); 1277 __mark_sit_entry_dirty(sbi, segno); 1278 } 1279 update_sits_in_cursum(sum, -sits_in_cursum(sum)); 1280 return 1; 1281 } 1282 return 0; 1283} 1284 1285/* 1286 * CP calls this function, which flushes SIT entries including sit_journal, 1287 * and moves prefree segs to free segs. 1288 */ 1289void flush_sit_entries(struct f2fs_sb_info *sbi) 1290{ 1291 struct sit_info *sit_i = SIT_I(sbi); 1292 unsigned long *bitmap = sit_i->dirty_sentries_bitmap; 1293 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 1294 struct f2fs_summary_block *sum = curseg->sum_blk; 1295 unsigned long nsegs = TOTAL_SEGS(sbi); 1296 struct page *page = NULL; 1297 struct f2fs_sit_block *raw_sit = NULL; 1298 unsigned int start = 0, end = 0; 1299 unsigned int segno = -1; 1300 bool flushed; 1301 1302 mutex_lock(&curseg->curseg_mutex); 1303 mutex_lock(&sit_i->sentry_lock); 1304 1305 /* 1306 * "flushed" indicates whether sit entries in journal are flushed 1307 * to the SIT area or not. 1308 */ 1309 flushed = flush_sits_in_journal(sbi); 1310 1311 while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) { 1312 struct seg_entry *se = get_seg_entry(sbi, segno); 1313 int sit_offset, offset; 1314 1315 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno); 1316 1317 if (flushed) 1318 goto to_sit_page; 1319 1320 offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1); 1321 if (offset >= 0) { 1322 segno_in_journal(sum, offset) = cpu_to_le32(segno); 1323 seg_info_to_raw_sit(se, &sit_in_journal(sum, offset)); 1324 goto flush_done; 1325 } 1326to_sit_page: 1327 if (!page || (start > segno) || (segno > end)) { 1328 if (page) { 1329 f2fs_put_page(page, 1); 1330 page = NULL; 1331 } 1332 1333 start = START_SEGNO(sit_i, segno); 1334 end = start + SIT_ENTRY_PER_BLOCK - 1; 1335 1336 /* read sit block that will be updated */ 1337 page = get_next_sit_page(sbi, start); 1338 raw_sit = page_address(page); 1339 } 1340 1341 /* udpate entry in SIT block */ 1342 seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]); 1343flush_done: 1344 __clear_bit(segno, bitmap); 1345 sit_i->dirty_sentries--; 1346 } 1347 mutex_unlock(&sit_i->sentry_lock); 1348 mutex_unlock(&curseg->curseg_mutex); 1349 1350 /* writeout last modified SIT block */ 1351 f2fs_put_page(page, 1); 1352 1353 set_prefree_as_free_segments(sbi); 1354} 1355 1356static int build_sit_info(struct f2fs_sb_info *sbi) 1357{ 1358 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 1359 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1360 struct sit_info *sit_i; 1361 unsigned int sit_segs, start; 1362 char *src_bitmap, *dst_bitmap; 1363 unsigned int bitmap_size; 1364 1365 /* allocate memory for SIT information */ 1366 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL); 1367 if (!sit_i) 1368 return -ENOMEM; 1369 1370 SM_I(sbi)->sit_info = sit_i; 1371 1372 sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry)); 1373 if (!sit_i->sentries) 1374 return -ENOMEM; 1375 1376 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi)); 1377 sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL); 1378 if (!sit_i->dirty_sentries_bitmap) 1379 return -ENOMEM; 1380 1381 for (start = 0; start < TOTAL_SEGS(sbi); start++) { 1382 sit_i->sentries[start].cur_valid_map 1383 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 1384 sit_i->sentries[start].ckpt_valid_map 1385 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL); 1386 if (!sit_i->sentries[start].cur_valid_map 1387 || !sit_i->sentries[start].ckpt_valid_map) 1388 return -ENOMEM; 1389 } 1390 1391 if (sbi->segs_per_sec > 1) { 1392 sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) * 1393 sizeof(struct sec_entry)); 1394 if (!sit_i->sec_entries) 1395 return -ENOMEM; 1396 } 1397 1398 /* get information related with SIT */ 1399 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1; 1400 1401 /* setup SIT bitmap from ckeckpoint pack */ 1402 bitmap_size = __bitmap_size(sbi, SIT_BITMAP); 1403 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP); 1404 1405 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL); 1406 if (!dst_bitmap) 1407 return -ENOMEM; 1408 1409 /* init SIT information */ 1410 sit_i->s_ops = &default_salloc_ops; 1411 1412 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr); 1413 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg; 1414 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count); 1415 sit_i->sit_bitmap = dst_bitmap; 1416 sit_i->bitmap_size = bitmap_size; 1417 sit_i->dirty_sentries = 0; 1418 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK; 1419 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time); 1420 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec; 1421 mutex_init(&sit_i->sentry_lock); 1422 return 0; 1423} 1424 1425static int build_free_segmap(struct f2fs_sb_info *sbi) 1426{ 1427 struct f2fs_sm_info *sm_info = SM_I(sbi); 1428 struct free_segmap_info *free_i; 1429 unsigned int bitmap_size, sec_bitmap_size; 1430 1431 /* allocate memory for free segmap information */ 1432 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL); 1433 if (!free_i) 1434 return -ENOMEM; 1435 1436 SM_I(sbi)->free_info = free_i; 1437 1438 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi)); 1439 free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL); 1440 if (!free_i->free_segmap) 1441 return -ENOMEM; 1442 1443 sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi)); 1444 free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL); 1445 if (!free_i->free_secmap) 1446 return -ENOMEM; 1447 1448 /* set all segments as dirty temporarily */ 1449 memset(free_i->free_segmap, 0xff, bitmap_size); 1450 memset(free_i->free_secmap, 0xff, sec_bitmap_size); 1451 1452 /* init free segmap information */ 1453 free_i->start_segno = 1454 (unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr); 1455 free_i->free_segments = 0; 1456 free_i->free_sections = 0; 1457 rwlock_init(&free_i->segmap_lock); 1458 return 0; 1459} 1460 1461static int build_curseg(struct f2fs_sb_info *sbi) 1462{ 1463 struct curseg_info *array; 1464 int i; 1465 1466 array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL); 1467 if (!array) 1468 return -ENOMEM; 1469 1470 SM_I(sbi)->curseg_array = array; 1471 1472 for (i = 0; i < NR_CURSEG_TYPE; i++) { 1473 mutex_init(&array[i].curseg_mutex); 1474 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL); 1475 if (!array[i].sum_blk) 1476 return -ENOMEM; 1477 array[i].segno = NULL_SEGNO; 1478 array[i].next_blkoff = 0; 1479 } 1480 return restore_curseg_summaries(sbi); 1481} 1482 1483static void build_sit_entries(struct f2fs_sb_info *sbi) 1484{ 1485 struct sit_info *sit_i = SIT_I(sbi); 1486 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA); 1487 struct f2fs_summary_block *sum = curseg->sum_blk; 1488 unsigned int start; 1489 1490 for (start = 0; start < TOTAL_SEGS(sbi); start++) { 1491 struct seg_entry *se = &sit_i->sentries[start]; 1492 struct f2fs_sit_block *sit_blk; 1493 struct f2fs_sit_entry sit; 1494 struct page *page; 1495 int i; 1496 1497 mutex_lock(&curseg->curseg_mutex); 1498 for (i = 0; i < sits_in_cursum(sum); i++) { 1499 if (le32_to_cpu(segno_in_journal(sum, i)) == start) { 1500 sit = sit_in_journal(sum, i); 1501 mutex_unlock(&curseg->curseg_mutex); 1502 goto got_it; 1503 } 1504 } 1505 mutex_unlock(&curseg->curseg_mutex); 1506 page = get_current_sit_page(sbi, start); 1507 sit_blk = (struct f2fs_sit_block *)page_address(page); 1508 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)]; 1509 f2fs_put_page(page, 1); 1510got_it: 1511 check_block_count(sbi, start, &sit); 1512 seg_info_from_raw_sit(se, &sit); 1513 if (sbi->segs_per_sec > 1) { 1514 struct sec_entry *e = get_sec_entry(sbi, start); 1515 e->valid_blocks += se->valid_blocks; 1516 } 1517 } 1518} 1519 1520static void init_free_segmap(struct f2fs_sb_info *sbi) 1521{ 1522 unsigned int start; 1523 int type; 1524 1525 for (start = 0; start < TOTAL_SEGS(sbi); start++) { 1526 struct seg_entry *sentry = get_seg_entry(sbi, start); 1527 if (!sentry->valid_blocks) 1528 __set_free(sbi, start); 1529 } 1530 1531 /* set use the current segments */ 1532 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) { 1533 struct curseg_info *curseg_t = CURSEG_I(sbi, type); 1534 __set_test_and_inuse(sbi, curseg_t->segno); 1535 } 1536} 1537 1538static void init_dirty_segmap(struct f2fs_sb_info *sbi) 1539{ 1540 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1541 struct free_segmap_info *free_i = FREE_I(sbi); 1542 unsigned int segno = 0, offset = 0; 1543 unsigned short valid_blocks; 1544 1545 while (segno < TOTAL_SEGS(sbi)) { 1546 /* find dirty segment based on free segmap */ 1547 segno = find_next_inuse(free_i, TOTAL_SEGS(sbi), offset); 1548 if (segno >= TOTAL_SEGS(sbi)) 1549 break; 1550 offset = segno + 1; 1551 valid_blocks = get_valid_blocks(sbi, segno, 0); 1552 if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks) 1553 continue; 1554 mutex_lock(&dirty_i->seglist_lock); 1555 __locate_dirty_segment(sbi, segno, DIRTY); 1556 mutex_unlock(&dirty_i->seglist_lock); 1557 } 1558} 1559 1560static int init_victim_segmap(struct f2fs_sb_info *sbi) 1561{ 1562 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1563 unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi)); 1564 1565 dirty_i->victim_segmap[FG_GC] = kzalloc(bitmap_size, GFP_KERNEL); 1566 dirty_i->victim_segmap[BG_GC] = kzalloc(bitmap_size, GFP_KERNEL); 1567 if (!dirty_i->victim_segmap[FG_GC] || !dirty_i->victim_segmap[BG_GC]) 1568 return -ENOMEM; 1569 return 0; 1570} 1571 1572static int build_dirty_segmap(struct f2fs_sb_info *sbi) 1573{ 1574 struct dirty_seglist_info *dirty_i; 1575 unsigned int bitmap_size, i; 1576 1577 /* allocate memory for dirty segments list information */ 1578 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL); 1579 if (!dirty_i) 1580 return -ENOMEM; 1581 1582 SM_I(sbi)->dirty_info = dirty_i; 1583 mutex_init(&dirty_i->seglist_lock); 1584 1585 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi)); 1586 1587 for (i = 0; i < NR_DIRTY_TYPE; i++) { 1588 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL); 1589 if (!dirty_i->dirty_segmap[i]) 1590 return -ENOMEM; 1591 } 1592 1593 init_dirty_segmap(sbi); 1594 return init_victim_segmap(sbi); 1595} 1596 1597/* 1598 * Update min, max modified time for cost-benefit GC algorithm 1599 */ 1600static void init_min_max_mtime(struct f2fs_sb_info *sbi) 1601{ 1602 struct sit_info *sit_i = SIT_I(sbi); 1603 unsigned int segno; 1604 1605 mutex_lock(&sit_i->sentry_lock); 1606 1607 sit_i->min_mtime = LLONG_MAX; 1608 1609 for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) { 1610 unsigned int i; 1611 unsigned long long mtime = 0; 1612 1613 for (i = 0; i < sbi->segs_per_sec; i++) 1614 mtime += get_seg_entry(sbi, segno + i)->mtime; 1615 1616 mtime = div_u64(mtime, sbi->segs_per_sec); 1617 1618 if (sit_i->min_mtime > mtime) 1619 sit_i->min_mtime = mtime; 1620 } 1621 sit_i->max_mtime = get_mtime(sbi); 1622 mutex_unlock(&sit_i->sentry_lock); 1623} 1624 1625int build_segment_manager(struct f2fs_sb_info *sbi) 1626{ 1627 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi); 1628 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1629 struct f2fs_sm_info *sm_info; 1630 int err; 1631 1632 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL); 1633 if (!sm_info) 1634 return -ENOMEM; 1635 1636 /* init sm info */ 1637 sbi->sm_info = sm_info; 1638 INIT_LIST_HEAD(&sm_info->wblist_head); 1639 spin_lock_init(&sm_info->wblist_lock); 1640 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr); 1641 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr); 1642 sm_info->segment_count = le32_to_cpu(raw_super->segment_count); 1643 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count); 1644 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count); 1645 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main); 1646 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr); 1647 1648 err = build_sit_info(sbi); 1649 if (err) 1650 return err; 1651 err = build_free_segmap(sbi); 1652 if (err) 1653 return err; 1654 err = build_curseg(sbi); 1655 if (err) 1656 return err; 1657 1658 /* reinit free segmap based on SIT */ 1659 build_sit_entries(sbi); 1660 1661 init_free_segmap(sbi); 1662 err = build_dirty_segmap(sbi); 1663 if (err) 1664 return err; 1665 1666 init_min_max_mtime(sbi); 1667 return 0; 1668} 1669 1670static void discard_dirty_segmap(struct f2fs_sb_info *sbi, 1671 enum dirty_type dirty_type) 1672{ 1673 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1674 1675 mutex_lock(&dirty_i->seglist_lock); 1676 kfree(dirty_i->dirty_segmap[dirty_type]); 1677 dirty_i->nr_dirty[dirty_type] = 0; 1678 mutex_unlock(&dirty_i->seglist_lock); 1679} 1680 1681void reset_victim_segmap(struct f2fs_sb_info *sbi) 1682{ 1683 unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi)); 1684 memset(DIRTY_I(sbi)->victim_segmap[FG_GC], 0, bitmap_size); 1685} 1686 1687static void destroy_victim_segmap(struct f2fs_sb_info *sbi) 1688{ 1689 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1690 1691 kfree(dirty_i->victim_segmap[FG_GC]); 1692 kfree(dirty_i->victim_segmap[BG_GC]); 1693} 1694 1695static void destroy_dirty_segmap(struct f2fs_sb_info *sbi) 1696{ 1697 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 1698 int i; 1699 1700 if (!dirty_i) 1701 return; 1702 1703 /* discard pre-free/dirty segments list */ 1704 for (i = 0; i < NR_DIRTY_TYPE; i++) 1705 discard_dirty_segmap(sbi, i); 1706 1707 destroy_victim_segmap(sbi); 1708 SM_I(sbi)->dirty_info = NULL; 1709 kfree(dirty_i); 1710} 1711 1712static void destroy_curseg(struct f2fs_sb_info *sbi) 1713{ 1714 struct curseg_info *array = SM_I(sbi)->curseg_array; 1715 int i; 1716 1717 if (!array) 1718 return; 1719 SM_I(sbi)->curseg_array = NULL; 1720 for (i = 0; i < NR_CURSEG_TYPE; i++) 1721 kfree(array[i].sum_blk); 1722 kfree(array); 1723} 1724 1725static void destroy_free_segmap(struct f2fs_sb_info *sbi) 1726{ 1727 struct free_segmap_info *free_i = SM_I(sbi)->free_info; 1728 if (!free_i) 1729 return; 1730 SM_I(sbi)->free_info = NULL; 1731 kfree(free_i->free_segmap); 1732 kfree(free_i->free_secmap); 1733 kfree(free_i); 1734} 1735 1736static void destroy_sit_info(struct f2fs_sb_info *sbi) 1737{ 1738 struct sit_info *sit_i = SIT_I(sbi); 1739 unsigned int start; 1740 1741 if (!sit_i) 1742 return; 1743 1744 if (sit_i->sentries) { 1745 for (start = 0; start < TOTAL_SEGS(sbi); start++) { 1746 kfree(sit_i->sentries[start].cur_valid_map); 1747 kfree(sit_i->sentries[start].ckpt_valid_map); 1748 } 1749 } 1750 vfree(sit_i->sentries); 1751 vfree(sit_i->sec_entries); 1752 kfree(sit_i->dirty_sentries_bitmap); 1753 1754 SM_I(sbi)->sit_info = NULL; 1755 kfree(sit_i->sit_bitmap); 1756 kfree(sit_i); 1757} 1758 1759void destroy_segment_manager(struct f2fs_sb_info *sbi) 1760{ 1761 struct f2fs_sm_info *sm_info = SM_I(sbi); 1762 destroy_dirty_segmap(sbi); 1763 destroy_curseg(sbi); 1764 destroy_free_segmap(sbi); 1765 destroy_sit_info(sbi); 1766 sbi->sm_info = NULL; 1767 kfree(sm_info); 1768} 1769