fs/f2fs/segment.c

   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/kthread.h>
  17 #include <linux/vmalloc.h>
  18 #include <linux/swap.h>
  19
  20 #include "f2fs.h"
  21 #include "segment.h"
  22 #include "node.h"
  23 #include <trace/events/f2fs.h>
  24
  25 #define __reverse_ffz(x) __reverse_ffs(~(x))
  26
  27 static struct kmem_cache *discard_entry_slab;
  28
  29 /*
  30  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  31  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  32  */
  33 static inline unsigned long __reverse_ffs(unsigned long word)
  34 {
  35         int num = 0;
  36
  37 #if BITS_PER_LONG == 64
  38         if ((word & 0xffffffff) == 0) {
  39                 num += 32;
  40                 word >>= 32;
  41         }
  42 #endif
  43         if ((word & 0xffff) == 0) {
  44                 num += 16;
  45                 word >>= 16;
  46         }
  47         if ((word & 0xff) == 0) {
  48                 num += 8;
  49                 word >>= 8;
  50         }
  51         if ((word & 0xf0) == 0)
  52                 num += 4;
  53         else
  54                 word >>= 4;
  55         if ((word & 0xc) == 0)
  56                 num += 2;
  57         else
  58                 word >>= 2;
  59         if ((word & 0x2) == 0)
  60                 num += 1;
  61         return num;
  62 }
  63
  64 /*
  65  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c becasue
  66  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  67  * Example:
  68  *                             LSB <--> MSB
  69  *   f2fs_set_bit(0, bitmap) => 0000 0001
  70  *   f2fs_set_bit(7, bitmap) => 1000 0000
  71  */
  72 static unsigned long __find_rev_next_bit(const unsigned long *addr,
  73                         unsigned long size, unsigned long offset)
  74 {
  75         const unsigned long *p = addr + BIT_WORD(offset);
  76         unsigned long result = offset & ~(BITS_PER_LONG - 1);
  77         unsigned long tmp;
  78         unsigned long mask, submask;
  79         unsigned long quot, rest;
  80
  81         if (offset >= size)
  82                 return size;
  83
  84         size -= result;
  85         offset %= BITS_PER_LONG;
  86         if (!offset)
  87                 goto aligned;
  88
  89         tmp = *(p++);
  90         quot = (offset >> 3) << 3;
  91         rest = offset & 0x7;
  92         mask = ~0UL << quot;
  93         submask = (unsigned char)(0xff << rest) >> rest;
  94         submask <<= quot;
  95         mask &= submask;
  96         tmp &= mask;
  97         if (size < BITS_PER_LONG)
  98                 goto found_first;
  99         if (tmp)
 100                 goto found_middle;
 101
 102         size -= BITS_PER_LONG;
 103         result += BITS_PER_LONG;
 104 aligned:
 105         while (size & ~(BITS_PER_LONG-1)) {
 106                 tmp = *(p++);
 107                 if (tmp)
 108                         goto found_middle;
 109                 result += BITS_PER_LONG;
 110                 size -= BITS_PER_LONG;
 111         }
 112         if (!size)
 113                 return result;
 114         tmp = *p;
 115 found_first:
 116         tmp &= (~0UL >> (BITS_PER_LONG - size));
 117         if (tmp == 0UL)         /* Are any bits set? */
 118                 return result + size;   /* Nope. */
 119 found_middle:
 120         return result + __reverse_ffs(tmp);
 121 }
 122
 123 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 124                         unsigned long size, unsigned long offset)
 125 {
 126         const unsigned long *p = addr + BIT_WORD(offset);
 127         unsigned long result = offset & ~(BITS_PER_LONG - 1);
 128         unsigned long tmp;
 129         unsigned long mask, submask;
 130         unsigned long quot, rest;
 131
 132         if (offset >= size)
 133                 return size;
 134
 135         size -= result;
 136         offset %= BITS_PER_LONG;
 137         if (!offset)
 138                 goto aligned;
 139
 140         tmp = *(p++);
 141         quot = (offset >> 3) << 3;
 142         rest = offset & 0x7;
 143         mask = ~(~0UL << quot);
 144         submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest);
 145         submask <<= quot;
 146         mask += submask;
 147         tmp |= mask;
 148         if (size < BITS_PER_LONG)
 149                 goto found_first;
 150         if (~tmp)
 151                 goto found_middle;
 152
 153         size -= BITS_PER_LONG;
 154         result += BITS_PER_LONG;
 155 aligned:
 156         while (size & ~(BITS_PER_LONG - 1)) {
 157                 tmp = *(p++);
 158                 if (~tmp)
 159                         goto found_middle;
 160                 result += BITS_PER_LONG;
 161                 size -= BITS_PER_LONG;
 162         }
 163         if (!size)
 164                 return result;
 165         tmp = *p;
 166
 167 found_first:
 168         tmp |= ~0UL << size;
 169         if (tmp == ~0UL)        /* Are any bits zero? */
 170                 return result + size;   /* Nope. */
 171 found_middle:
 172         return result + __reverse_ffz(tmp);
 173 }
 174
 175 /*
 176  * This function balances dirty node and dentry pages.
 177  * In addition, it controls garbage collection.
 178  */
 179 void f2fs_balance_fs(struct f2fs_sb_info *sbi)
 180 {
 181         /*
 182          * We should do GC or end up with checkpoint, if there are so many dirty
 183          * dir/node pages without enough free segments.
 184          */
 185         if (has_not_enough_free_secs(sbi, 0)) {
 186                 mutex_lock(&sbi->gc_mutex);
 187                 f2fs_gc(sbi);
 188         }
 189 }
 190
 191 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 192 {
 193         /* check the # of cached NAT entries and prefree segments */
 194         if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) ||
 195                                 excess_prefree_segs(sbi))
 196                 f2fs_sync_fs(sbi->sb, true);
 197 }
 198
 199 static int issue_flush_thread(void *data)
 200 {
 201         struct f2fs_sb_info *sbi = data;
 202         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 203         wait_queue_head_t *q = &fcc->flush_wait_queue;
 204 repeat:
 205         if (kthread_should_stop())
 206                 return 0;
 207
 208         spin_lock(&fcc->issue_lock);
 209         if (fcc->issue_list) {
 210                 fcc->dispatch_list = fcc->issue_list;
 211                 fcc->issue_list = fcc->issue_tail = NULL;
 212         }
 213         spin_unlock(&fcc->issue_lock);
 214
 215         if (fcc->dispatch_list) {
 216                 struct bio *bio = bio_alloc(GFP_NOIO, 0);
 217                 struct flush_cmd *cmd, *next;
 218                 int ret;
 219
 220                 bio->bi_bdev = sbi->sb->s_bdev;
 221                 ret = submit_bio_wait(WRITE_FLUSH, bio);
 222
 223                 for (cmd = fcc->dispatch_list; cmd; cmd = next) {
 224                         cmd->ret = ret;
 225                         next = cmd->next;
 226                         complete(&cmd->wait);
 227                 }
 228                 bio_put(bio);
 229                 fcc->dispatch_list = NULL;
 230         }
 231
 232         wait_event_interruptible(*q,
 233                         kthread_should_stop() || fcc->issue_list);
 234         goto repeat;
 235 }
 236
 237 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
 238 {
 239         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 240         struct flush_cmd cmd;
 241
 242         if (!test_opt(sbi, FLUSH_MERGE))
 243                 return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL);
 244
 245         init_completion(&cmd.wait);
 246         cmd.next = NULL;
 247
 248         spin_lock(&fcc->issue_lock);
 249         if (fcc->issue_list)
 250                 fcc->issue_tail->next = &cmd;
 251         else
 252                 fcc->issue_list = &cmd;
 253         fcc->issue_tail = &cmd;
 254         spin_unlock(&fcc->issue_lock);
 255
 256         if (!fcc->dispatch_list)
 257                 wake_up(&fcc->flush_wait_queue);
 258
 259         wait_for_completion(&cmd.wait);
 260
 261         return cmd.ret;
 262 }
 263
 264 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
 265 {
 266         dev_t dev = sbi->sb->s_bdev->bd_dev;
 267         struct flush_cmd_control *fcc;
 268         int err = 0;
 269
 270         fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
 271         if (!fcc)
 272                 return -ENOMEM;
 273         spin_lock_init(&fcc->issue_lock);
 274         init_waitqueue_head(&fcc->flush_wait_queue);
 275         sbi->sm_info->cmd_control_info = fcc;
 276         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 277                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 278         if (IS_ERR(fcc->f2fs_issue_flush)) {
 279                 err = PTR_ERR(fcc->f2fs_issue_flush);
 280                 kfree(fcc);
 281                 sbi->sm_info->cmd_control_info = NULL;
 282                 return err;
 283         }
 284
 285         return err;
 286 }
 287
 288 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
 289 {
 290         struct flush_cmd_control *fcc =
 291                                 sbi->sm_info->cmd_control_info;
 292
 293         if (fcc && fcc->f2fs_issue_flush)
 294                 kthread_stop(fcc->f2fs_issue_flush);
 295         kfree(fcc);
 296         sbi->sm_info->cmd_control_info = NULL;
 297 }
 298
 299 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 300                 enum dirty_type dirty_type)
 301 {
 302         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 303
 304         /* need not be added */
 305         if (IS_CURSEG(sbi, segno))
 306                 return;
 307
 308         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 309                 dirty_i->nr_dirty[dirty_type]++;
 310
 311         if (dirty_type == DIRTY) {
 312                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 313                 enum dirty_type t = sentry->type;
 314
 315                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 316                         dirty_i->nr_dirty[t]++;
 317         }
 318 }
 319
 320 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 321                 enum dirty_type dirty_type)
 322 {
 323         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 324
 325         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 326                 dirty_i->nr_dirty[dirty_type]--;
 327
 328         if (dirty_type == DIRTY) {
 329                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 330                 enum dirty_type t = sentry->type;
 331
 332                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 333                         dirty_i->nr_dirty[t]--;
 334
 335                 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
 336                         clear_bit(GET_SECNO(sbi, segno),
 337                                                 dirty_i->victim_secmap);
 338         }
 339 }
 340
 341 /*
 342  * Should not occur error such as -ENOMEM.
 343  * Adding dirty entry into seglist is not critical operation.
 344  * If a given segment is one of current working segments, it won't be added.
 345  */
 346 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 347 {
 348         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 349         unsigned short valid_blocks;
 350
 351         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 352                 return;
 353
 354         mutex_lock(&dirty_i->seglist_lock);
 355
 356         valid_blocks = get_valid_blocks(sbi, segno, 0);
 357
 358         if (valid_blocks == 0) {
 359                 __locate_dirty_segment(sbi, segno, PRE);
 360                 __remove_dirty_segment(sbi, segno, DIRTY);
 361         } else if (valid_blocks < sbi->blocks_per_seg) {
 362                 __locate_dirty_segment(sbi, segno, DIRTY);
 363         } else {
 364                 /* Recovery routine with SSR needs this */
 365                 __remove_dirty_segment(sbi, segno, DIRTY);
 366         }
 367
 368         mutex_unlock(&dirty_i->seglist_lock);
 369 }
 370
 371 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
 372                                 block_t blkstart, block_t blklen)
 373 {
 374         sector_t start = SECTOR_FROM_BLOCK(sbi, blkstart);
 375         sector_t len = SECTOR_FROM_BLOCK(sbi, blklen);
 376         trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
 377         return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
 378 }
 379
 380 void discard_next_dnode(struct f2fs_sb_info *sbi)
 381 {
 382         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 383         block_t blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 384
 385         if (f2fs_issue_discard(sbi, blkaddr, 1)) {
 386                 struct page *page = grab_meta_page(sbi, blkaddr);
 387                 /* zero-filled page */
 388                 set_page_dirty(page);
 389                 f2fs_put_page(page, 1);
 390         }
 391 }
 392
 393 static void add_discard_addrs(struct f2fs_sb_info *sbi,
 394                         unsigned int segno, struct seg_entry *se)
 395 {
 396         struct list_head *head = &SM_I(sbi)->discard_list;
 397         struct discard_entry *new;
 398         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
 399         int max_blocks = sbi->blocks_per_seg;
 400         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
 401         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
 402         unsigned long dmap[entries];
 403         unsigned int start = 0, end = -1;
 404         int i;
 405
 406         if (!test_opt(sbi, DISCARD))
 407                 return;
 408
 409         /* zero block will be discarded through the prefree list */
 410         if (!se->valid_blocks || se->valid_blocks == max_blocks)
 411                 return;
 412
 413         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
 414         for (i = 0; i < entries; i++)
 415                 dmap[i] = (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
 416
 417         while (SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
 418                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
 419                 if (start >= max_blocks)
 420                         break;
 421
 422                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
 423
 424                 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
 425                 INIT_LIST_HEAD(&new->list);
 426                 new->blkaddr = START_BLOCK(sbi, segno) + start;
 427                 new->len = end - start;
 428
 429                 list_add_tail(&new->list, head);
 430                 SM_I(sbi)->nr_discards += end - start;
 431         }
 432 }
 433
 434 /*
 435  * Should call clear_prefree_segments after checkpoint is done.
 436  */
 437 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
 438 {
 439         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 440         unsigned int segno = -1;
 441         unsigned int total_segs = TOTAL_SEGS(sbi);
 442
 443         mutex_lock(&dirty_i->seglist_lock);
 444         while (1) {
 445                 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
 446                                 segno + 1);
 447                 if (segno >= total_segs)
 448                         break;
 449                 __set_test_and_free(sbi, segno);
 450         }
 451         mutex_unlock(&dirty_i->seglist_lock);
 452 }
 453
 454 void clear_prefree_segments(struct f2fs_sb_info *sbi)
 455 {
 456         struct list_head *head = &(SM_I(sbi)->discard_list);
 457         struct discard_entry *entry, *this;
 458         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 459         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
 460         unsigned int total_segs = TOTAL_SEGS(sbi);
 461         unsigned int start = 0, end = -1;
 462
 463         mutex_lock(&dirty_i->seglist_lock);
 464
 465         while (1) {
 466                 int i;
 467                 start = find_next_bit(prefree_map, total_segs, end + 1);
 468                 if (start >= total_segs)
 469                         break;
 470                 end = find_next_zero_bit(prefree_map, total_segs, start + 1);
 471
 472                 for (i = start; i < end; i++)
 473                         clear_bit(i, prefree_map);
 474
 475                 dirty_i->nr_dirty[PRE] -= end - start;
 476
 477                 if (!test_opt(sbi, DISCARD))
 478                         continue;
 479
 480                 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
 481                                 (end - start) << sbi->log_blocks_per_seg);
 482         }
 483         mutex_unlock(&dirty_i->seglist_lock);
 484
 485         /* send small discards */
 486         list_for_each_entry_safe(entry, this, head, list) {
 487                 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
 488                 list_del(&entry->list);
 489                 SM_I(sbi)->nr_discards -= entry->len;
 490                 kmem_cache_free(discard_entry_slab, entry);
 491         }
 492 }
 493
 494 static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
 495 {
 496         struct sit_info *sit_i = SIT_I(sbi);
 497         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap))
 498                 sit_i->dirty_sentries++;
 499 }
 500
 501 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
 502                                         unsigned int segno, int modified)
 503 {
 504         struct seg_entry *se = get_seg_entry(sbi, segno);
 505         se->type = type;
 506         if (modified)
 507                 __mark_sit_entry_dirty(sbi, segno);
 508 }
 509
 510 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
 511 {
 512         struct seg_entry *se;
 513         unsigned int segno, offset;
 514         long int new_vblocks;
 515
 516         segno = GET_SEGNO(sbi, blkaddr);
 517
 518         se = get_seg_entry(sbi, segno);
 519         new_vblocks = se->valid_blocks + del;
 520         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 521
 522         f2fs_bug_on((new_vblocks >> (sizeof(unsigned short) << 3) ||
 523                                 (new_vblocks > sbi->blocks_per_seg)));
 524
 525         se->valid_blocks = new_vblocks;
 526         se->mtime = get_mtime(sbi);
 527         SIT_I(sbi)->max_mtime = se->mtime;
 528
 529         /* Update valid block bitmap */
 530         if (del > 0) {
 531                 if (f2fs_set_bit(offset, se->cur_valid_map))
 532                         BUG();
 533         } else {
 534                 if (!f2fs_clear_bit(offset, se->cur_valid_map))
 535                         BUG();
 536         }
 537         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
 538                 se->ckpt_valid_blocks += del;
 539
 540         __mark_sit_entry_dirty(sbi, segno);
 541
 542         /* update total number of valid blocks to be written in ckpt area */
 543         SIT_I(sbi)->written_valid_blocks += del;
 544
 545         if (sbi->segs_per_sec > 1)
 546                 get_sec_entry(sbi, segno)->valid_blocks += del;
 547 }
 548
 549 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
 550 {
 551         update_sit_entry(sbi, new, 1);
 552         if (GET_SEGNO(sbi, old) != NULL_SEGNO)
 553                 update_sit_entry(sbi, old, -1);
 554
 555         locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
 556         locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
 557 }
 558
 559 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
 560 {
 561         unsigned int segno = GET_SEGNO(sbi, addr);
 562         struct sit_info *sit_i = SIT_I(sbi);
 563
 564         f2fs_bug_on(addr == NULL_ADDR);
 565         if (addr == NEW_ADDR)
 566                 return;
 567
 568         /* add it into sit main buffer */
 569         mutex_lock(&sit_i->sentry_lock);
 570
 571         update_sit_entry(sbi, addr, -1);
 572
 573         /* add it into dirty seglist */
 574         locate_dirty_segment(sbi, segno);
 575
 576         mutex_unlock(&sit_i->sentry_lock);
 577 }
 578
 579 /*
 580  * This function should be resided under the curseg_mutex lock
 581  */
 582 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
 583                                         struct f2fs_summary *sum)
 584 {
 585         struct curseg_info *curseg = CURSEG_I(sbi, type);
 586         void *addr = curseg->sum_blk;
 587         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
 588         memcpy(addr, sum, sizeof(struct f2fs_summary));
 589 }
 590
 591 /*
 592  * Calculate the number of current summary pages for writing
 593  */
 594 int npages_for_summary_flush(struct f2fs_sb_info *sbi)
 595 {
 596         int valid_sum_count = 0;
 597         int i, sum_in_page;
 598
 599         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 600                 if (sbi->ckpt->alloc_type[i] == SSR)
 601                         valid_sum_count += sbi->blocks_per_seg;
 602                 else
 603                         valid_sum_count += curseg_blkoff(sbi, i);
 604         }
 605
 606         sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
 607                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
 608         if (valid_sum_count <= sum_in_page)
 609                 return 1;
 610         else if ((valid_sum_count - sum_in_page) <=
 611                 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
 612                 return 2;
 613         return 3;
 614 }
 615
 616 /*
 617  * Caller should put this summary page
 618  */
 619 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
 620 {
 621         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
 622 }
 623
 624 static void write_sum_page(struct f2fs_sb_info *sbi,
 625                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
 626 {
 627         struct page *page = grab_meta_page(sbi, blk_addr);
 628         void *kaddr = page_address(page);
 629         memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
 630         set_page_dirty(page);
 631         f2fs_put_page(page, 1);
 632 }
 633
 634 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
 635 {
 636         struct curseg_info *curseg = CURSEG_I(sbi, type);
 637         unsigned int segno = curseg->segno + 1;
 638         struct free_segmap_info *free_i = FREE_I(sbi);
 639
 640         if (segno < TOTAL_SEGS(sbi) && segno % sbi->segs_per_sec)
 641                 return !test_bit(segno, free_i->free_segmap);
 642         return 0;
 643 }
 644
 645 /*
 646  * Find a new segment from the free segments bitmap to right order
 647  * This function should be returned with success, otherwise BUG
 648  */
 649 static void get_new_segment(struct f2fs_sb_info *sbi,
 650                         unsigned int *newseg, bool new_sec, int dir)
 651 {
 652         struct free_segmap_info *free_i = FREE_I(sbi);
 653         unsigned int segno, secno, zoneno;
 654         unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone;
 655         unsigned int hint = *newseg / sbi->segs_per_sec;
 656         unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
 657         unsigned int left_start = hint;
 658         bool init = true;
 659         int go_left = 0;
 660         int i;
 661
 662         write_lock(&free_i->segmap_lock);
 663
 664         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
 665                 segno = find_next_zero_bit(free_i->free_segmap,
 666                                         TOTAL_SEGS(sbi), *newseg + 1);
 667                 if (segno - *newseg < sbi->segs_per_sec -
 668                                         (*newseg % sbi->segs_per_sec))
 669                         goto got_it;
 670         }
 671 find_other_zone:
 672         secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint);
 673         if (secno >= TOTAL_SECS(sbi)) {
 674                 if (dir == ALLOC_RIGHT) {
 675                         secno = find_next_zero_bit(free_i->free_secmap,
 676                                                         TOTAL_SECS(sbi), 0);
 677                         f2fs_bug_on(secno >= TOTAL_SECS(sbi));
 678                 } else {
 679                         go_left = 1;
 680                         left_start = hint - 1;
 681                 }
 682         }
 683         if (go_left == 0)
 684                 goto skip_left;
 685
 686         while (test_bit(left_start, free_i->free_secmap)) {
 687                 if (left_start > 0) {
 688                         left_start--;
 689                         continue;
 690                 }
 691                 left_start = find_next_zero_bit(free_i->free_secmap,
 692                                                         TOTAL_SECS(sbi), 0);
 693                 f2fs_bug_on(left_start >= TOTAL_SECS(sbi));
 694                 break;
 695         }
 696         secno = left_start;
 697 skip_left:
 698         hint = secno;
 699         segno = secno * sbi->segs_per_sec;
 700         zoneno = secno / sbi->secs_per_zone;
 701
 702         /* give up on finding another zone */
 703         if (!init)
 704                 goto got_it;
 705         if (sbi->secs_per_zone == 1)
 706                 goto got_it;
 707         if (zoneno == old_zoneno)
 708                 goto got_it;
 709         if (dir == ALLOC_LEFT) {
 710                 if (!go_left && zoneno + 1 >= total_zones)
 711                         goto got_it;
 712                 if (go_left && zoneno == 0)
 713                         goto got_it;
 714         }
 715         for (i = 0; i < NR_CURSEG_TYPE; i++)
 716                 if (CURSEG_I(sbi, i)->zone == zoneno)
 717                         break;
 718
 719         if (i < NR_CURSEG_TYPE) {
 720                 /* zone is in user, try another */
 721                 if (go_left)
 722                         hint = zoneno * sbi->secs_per_zone - 1;
 723                 else if (zoneno + 1 >= total_zones)
 724                         hint = 0;
 725                 else
 726                         hint = (zoneno + 1) * sbi->secs_per_zone;
 727                 init = false;
 728                 goto find_other_zone;
 729         }
 730 got_it:
 731         /* set it as dirty segment in free segmap */
 732         f2fs_bug_on(test_bit(segno, free_i->free_segmap));
 733         __set_inuse(sbi, segno);
 734         *newseg = segno;
 735         write_unlock(&free_i->segmap_lock);
 736 }
 737
 738 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
 739 {
 740         struct curseg_info *curseg = CURSEG_I(sbi, type);
 741         struct summary_footer *sum_footer;
 742
 743         curseg->segno = curseg->next_segno;
 744         curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
 745         curseg->next_blkoff = 0;
 746         curseg->next_segno = NULL_SEGNO;
 747
 748         sum_footer = &(curseg->sum_blk->footer);
 749         memset(sum_footer, 0, sizeof(struct summary_footer));
 750         if (IS_DATASEG(type))
 751                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
 752         if (IS_NODESEG(type))
 753                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
 754         __set_sit_entry_type(sbi, type, curseg->segno, modified);
 755 }
 756
 757 /*
 758  * Allocate a current working segment.
 759  * This function always allocates a free segment in LFS manner.
 760  */
 761 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
 762 {
 763         struct curseg_info *curseg = CURSEG_I(sbi, type);
 764         unsigned int segno = curseg->segno;
 765         int dir = ALLOC_LEFT;
 766
 767         write_sum_page(sbi, curseg->sum_blk,
 768                                 GET_SUM_BLOCK(sbi, segno));
 769         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
 770                 dir = ALLOC_RIGHT;
 771
 772         if (test_opt(sbi, NOHEAP))
 773                 dir = ALLOC_RIGHT;
 774
 775         get_new_segment(sbi, &segno, new_sec, dir);
 776         curseg->next_segno = segno;
 777         reset_curseg(sbi, type, 1);
 778         curseg->alloc_type = LFS;
 779 }
 780
 781 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
 782                         struct curseg_info *seg, block_t start)
 783 {
 784         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
 785         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
 786         unsigned long target_map[entries];
 787         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
 788         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
 789         int i, pos;
 790
 791         for (i = 0; i < entries; i++)
 792                 target_map[i] = ckpt_map[i] | cur_map[i];
 793
 794         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
 795
 796         seg->next_blkoff = pos;
 797 }
 798
 799 /*
 800  * If a segment is written by LFS manner, next block offset is just obtained
 801  * by increasing the current block offset. However, if a segment is written by
 802  * SSR manner, next block offset obtained by calling __next_free_blkoff
 803  */
 804 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
 805                                 struct curseg_info *seg)
 806 {
 807         if (seg->alloc_type == SSR)
 808                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
 809         else
 810                 seg->next_blkoff++;
 811 }
 812
 813 /*
 814  * This function always allocates a used segment (from dirty seglist) by SSR
 815  * manner, so it should recover the existing segment information of valid blocks
 816  */
 817 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
 818 {
 819         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 820         struct curseg_info *curseg = CURSEG_I(sbi, type);
 821         unsigned int new_segno = curseg->next_segno;
 822         struct f2fs_summary_block *sum_node;
 823         struct page *sum_page;
 824
 825         write_sum_page(sbi, curseg->sum_blk,
 826                                 GET_SUM_BLOCK(sbi, curseg->segno));
 827         __set_test_and_inuse(sbi, new_segno);
 828
 829         mutex_lock(&dirty_i->seglist_lock);
 830         __remove_dirty_segment(sbi, new_segno, PRE);
 831         __remove_dirty_segment(sbi, new_segno, DIRTY);
 832         mutex_unlock(&dirty_i->seglist_lock);
 833
 834         reset_curseg(sbi, type, 1);
 835         curseg->alloc_type = SSR;
 836         __next_free_blkoff(sbi, curseg, 0);
 837
 838         if (reuse) {
 839                 sum_page = get_sum_page(sbi, new_segno);
 840                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
 841                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
 842                 f2fs_put_page(sum_page, 1);
 843         }
 844 }
 845
 846 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
 847 {
 848         struct curseg_info *curseg = CURSEG_I(sbi, type);
 849         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
 850
 851         if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
 852                 return v_ops->get_victim(sbi,
 853                                 &(curseg)->next_segno, BG_GC, type, SSR);
 854
 855         /* For data segments, let's do SSR more intensively */
 856         for (; type >= CURSEG_HOT_DATA; type--)
 857                 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
 858                                                 BG_GC, type, SSR))
 859                         return 1;
 860         return 0;
 861 }
 862
 863 /*
 864  * flush out current segment and replace it with new segment
 865  * This function should be returned with success, otherwise BUG
 866  */
 867 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
 868                                                 int type, bool force)
 869 {
 870         struct curseg_info *curseg = CURSEG_I(sbi, type);
 871
 872         if (force)
 873                 new_curseg(sbi, type, true);
 874         else if (type == CURSEG_WARM_NODE)
 875                 new_curseg(sbi, type, false);
 876         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
 877                 new_curseg(sbi, type, false);
 878         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
 879                 change_curseg(sbi, type, true);
 880         else
 881                 new_curseg(sbi, type, false);
 882
 883         stat_inc_seg_type(sbi, curseg);
 884 }
 885
 886 void allocate_new_segments(struct f2fs_sb_info *sbi)
 887 {
 888         struct curseg_info *curseg;
 889         unsigned int old_curseg;
 890         int i;
 891
 892         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 893                 curseg = CURSEG_I(sbi, i);
 894                 old_curseg = curseg->segno;
 895                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
 896                 locate_dirty_segment(sbi, old_curseg);
 897         }
 898 }
 899
 900 static const struct segment_allocation default_salloc_ops = {
 901         .allocate_segment = allocate_segment_by_default,
 902 };
 903
 904 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
 905 {
 906         struct curseg_info *curseg = CURSEG_I(sbi, type);
 907         if (curseg->next_blkoff < sbi->blocks_per_seg)
 908                 return true;
 909         return false;
 910 }
 911
 912 static int __get_segment_type_2(struct page *page, enum page_type p_type)
 913 {
 914         if (p_type == DATA)
 915                 return CURSEG_HOT_DATA;
 916         else
 917                 return CURSEG_HOT_NODE;
 918 }
 919
 920 static int __get_segment_type_4(struct page *page, enum page_type p_type)
 921 {
 922         if (p_type == DATA) {
 923                 struct inode *inode = page->mapping->host;
 924
 925                 if (S_ISDIR(inode->i_mode))
 926                         return CURSEG_HOT_DATA;
 927                 else
 928                         return CURSEG_COLD_DATA;
 929         } else {
 930                 if (IS_DNODE(page) && !is_cold_node(page))
 931                         return CURSEG_HOT_NODE;
 932                 else
 933                         return CURSEG_COLD_NODE;
 934         }
 935 }
 936
 937 static int __get_segment_type_6(struct page *page, enum page_type p_type)
 938 {
 939         if (p_type == DATA) {
 940                 struct inode *inode = page->mapping->host;
 941
 942                 if (S_ISDIR(inode->i_mode))
 943                         return CURSEG_HOT_DATA;
 944                 else if (is_cold_data(page) || file_is_cold(inode))
 945                         return CURSEG_COLD_DATA;
 946                 else
 947                         return CURSEG_WARM_DATA;
 948         } else {
 949                 if (IS_DNODE(page))
 950                         return is_cold_node(page) ? CURSEG_WARM_NODE :
 951                                                 CURSEG_HOT_NODE;
 952                 else
 953                         return CURSEG_COLD_NODE;
 954         }
 955 }
 956
 957 static int __get_segment_type(struct page *page, enum page_type p_type)
 958 {
 959         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 960         switch (sbi->active_logs) {
 961         case 2:
 962                 return __get_segment_type_2(page, p_type);
 963         case 4:
 964                 return __get_segment_type_4(page, p_type);
 965         }
 966         /* NR_CURSEG_TYPE(6) logs by default */
 967         f2fs_bug_on(sbi->active_logs != NR_CURSEG_TYPE);
 968         return __get_segment_type_6(page, p_type);
 969 }
 970
 971 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
 972                 block_t old_blkaddr, block_t *new_blkaddr,
 973                 struct f2fs_summary *sum, int type)
 974 {
 975         struct sit_info *sit_i = SIT_I(sbi);
 976         struct curseg_info *curseg;
 977         unsigned int old_cursegno;
 978
 979         curseg = CURSEG_I(sbi, type);
 980
 981         mutex_lock(&curseg->curseg_mutex);
 982
 983         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 984         old_cursegno = curseg->segno;
 985
 986         /*
 987          * __add_sum_entry should be resided under the curseg_mutex
 988          * because, this function updates a summary entry in the
 989          * current summary block.
 990          */
 991         __add_sum_entry(sbi, type, sum);
 992
 993         mutex_lock(&sit_i->sentry_lock);
 994         __refresh_next_blkoff(sbi, curseg);
 995
 996         stat_inc_block_count(sbi, curseg);
 997
 998         if (!__has_curseg_space(sbi, type))
 999                 sit_i->s_ops->allocate_segment(sbi, type, false);
1000         /*
1001          * SIT information should be updated before segment allocation,
1002          * since SSR needs latest valid block information.
1003          */
1004         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1005         locate_dirty_segment(sbi, old_cursegno);
1006
1007         mutex_unlock(&sit_i->sentry_lock);
1008
1009         if (page && IS_NODESEG(type))
1010                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1011
1012         mutex_unlock(&curseg->curseg_mutex);
1013 }
1014
1015 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
1016                         block_t old_blkaddr, block_t *new_blkaddr,
1017                         struct f2fs_summary *sum, struct f2fs_io_info *fio)
1018 {
1019         int type = __get_segment_type(page, fio->type);
1020
1021         allocate_data_block(sbi, page, old_blkaddr, new_blkaddr, sum, type);
1022
1023         /* writeout dirty page into bdev */
1024         f2fs_submit_page_mbio(sbi, page, *new_blkaddr, fio);
1025 }
1026
1027 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1028 {
1029         struct f2fs_io_info fio = {
1030                 .type = META,
1031                 .rw = WRITE_SYNC | REQ_META | REQ_PRIO
1032         };
1033
1034         set_page_writeback(page);
1035         f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1036 }
1037
1038 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
1039                 struct f2fs_io_info *fio,
1040                 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
1041 {
1042         struct f2fs_summary sum;
1043         set_summary(&sum, nid, 0, 0);
1044         do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, fio);
1045 }
1046
1047 void write_data_page(struct page *page, struct dnode_of_data *dn,
1048                 block_t *new_blkaddr, struct f2fs_io_info *fio)
1049 {
1050         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
1051         struct f2fs_summary sum;
1052         struct node_info ni;
1053
1054         f2fs_bug_on(dn->data_blkaddr == NULL_ADDR);
1055         get_node_info(sbi, dn->nid, &ni);
1056         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1057
1058         do_write_page(sbi, page, dn->data_blkaddr, new_blkaddr, &sum, fio);
1059 }
1060
1061 void rewrite_data_page(struct page *page, block_t old_blkaddr,
1062                                         struct f2fs_io_info *fio)
1063 {
1064         struct inode *inode = page->mapping->host;
1065         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1066         f2fs_submit_page_mbio(sbi, page, old_blkaddr, fio);
1067 }
1068
1069 void recover_data_page(struct f2fs_sb_info *sbi,
1070                         struct page *page, struct f2fs_summary *sum,
1071                         block_t old_blkaddr, block_t new_blkaddr)
1072 {
1073         struct sit_info *sit_i = SIT_I(sbi);
1074         struct curseg_info *curseg;
1075         unsigned int segno, old_cursegno;
1076         struct seg_entry *se;
1077         int type;
1078
1079         segno = GET_SEGNO(sbi, new_blkaddr);
1080         se = get_seg_entry(sbi, segno);
1081         type = se->type;
1082
1083         if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1084                 if (old_blkaddr == NULL_ADDR)
1085                         type = CURSEG_COLD_DATA;
1086                 else
1087                         type = CURSEG_WARM_DATA;
1088         }
1089         curseg = CURSEG_I(sbi, type);
1090
1091         mutex_lock(&curseg->curseg_mutex);
1092         mutex_lock(&sit_i->sentry_lock);
1093
1094         old_cursegno = curseg->segno;
1095
1096         /* change the current segment */
1097         if (segno != curseg->segno) {
1098                 curseg->next_segno = segno;
1099                 change_curseg(sbi, type, true);
1100         }
1101
1102         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1103         __add_sum_entry(sbi, type, sum);
1104
1105         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1106         locate_dirty_segment(sbi, old_cursegno);
1107
1108         mutex_unlock(&sit_i->sentry_lock);
1109         mutex_unlock(&curseg->curseg_mutex);
1110 }
1111
1112 void rewrite_node_page(struct f2fs_sb_info *sbi,
1113                         struct page *page, struct f2fs_summary *sum,
1114                         block_t old_blkaddr, block_t new_blkaddr)
1115 {
1116         struct sit_info *sit_i = SIT_I(sbi);
1117         int type = CURSEG_WARM_NODE;
1118         struct curseg_info *curseg;
1119         unsigned int segno, old_cursegno;
1120         block_t next_blkaddr = next_blkaddr_of_node(page);
1121         unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr);
1122         struct f2fs_io_info fio = {
1123                 .type = NODE,
1124                 .rw = WRITE_SYNC,
1125         };
1126
1127         curseg = CURSEG_I(sbi, type);
1128
1129         mutex_lock(&curseg->curseg_mutex);
1130         mutex_lock(&sit_i->sentry_lock);
1131
1132         segno = GET_SEGNO(sbi, new_blkaddr);
1133         old_cursegno = curseg->segno;
1134
1135         /* change the current segment */
1136         if (segno != curseg->segno) {
1137                 curseg->next_segno = segno;
1138                 change_curseg(sbi, type, true);
1139         }
1140         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1141         __add_sum_entry(sbi, type, sum);
1142
1143         /* change the current log to the next block addr in advance */
1144         if (next_segno != segno) {
1145                 curseg->next_segno = next_segno;
1146                 change_curseg(sbi, type, true);
1147         }
1148         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, next_blkaddr);
1149
1150         /* rewrite node page */
1151         set_page_writeback(page);
1152         f2fs_submit_page_mbio(sbi, page, new_blkaddr, &fio);
1153         f2fs_submit_merged_bio(sbi, NODE, WRITE);
1154         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1155         locate_dirty_segment(sbi, old_cursegno);
1156
1157         mutex_unlock(&sit_i->sentry_lock);
1158         mutex_unlock(&curseg->curseg_mutex);
1159 }
1160
1161 static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1162                                         struct page *page, enum page_type type)
1163 {
1164         enum page_type btype = PAGE_TYPE_OF_BIO(type);
1165         struct f2fs_bio_info *io = &sbi->write_io[btype];
1166         struct bio_vec *bvec;
1167         int i;
1168
1169         down_read(&io->io_rwsem);
1170         if (!io->bio)
1171                 goto out;
1172
1173         bio_for_each_segment_all(bvec, io->bio, i) {
1174                 if (page == bvec->bv_page) {
1175                         up_read(&io->io_rwsem);
1176                         return true;
1177                 }
1178         }
1179
1180 out:
1181         up_read(&io->io_rwsem);
1182         return false;
1183 }
1184
1185 void f2fs_wait_on_page_writeback(struct page *page,
1186                                 enum page_type type)
1187 {
1188         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1189         if (PageWriteback(page)) {
1190                 if (is_merged_page(sbi, page, type))
1191                         f2fs_submit_merged_bio(sbi, type, WRITE);
1192                 wait_on_page_writeback(page);
1193         }
1194 }
1195
1196 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1197 {
1198         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1199         struct curseg_info *seg_i;
1200         unsigned char *kaddr;
1201         struct page *page;
1202         block_t start;
1203         int i, j, offset;
1204
1205         start = start_sum_block(sbi);
1206
1207         page = get_meta_page(sbi, start++);
1208         kaddr = (unsigned char *)page_address(page);
1209
1210         /* Step 1: restore nat cache */
1211         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1212         memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1213
1214         /* Step 2: restore sit cache */
1215         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1216         memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1217                                                 SUM_JOURNAL_SIZE);
1218         offset = 2 * SUM_JOURNAL_SIZE;
1219
1220         /* Step 3: restore summary entries */
1221         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1222                 unsigned short blk_off;
1223                 unsigned int segno;
1224
1225                 seg_i = CURSEG_I(sbi, i);
1226                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1227                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1228                 seg_i->next_segno = segno;
1229                 reset_curseg(sbi, i, 0);
1230                 seg_i->alloc_type = ckpt->alloc_type[i];
1231                 seg_i->next_blkoff = blk_off;
1232
1233                 if (seg_i->alloc_type == SSR)
1234                         blk_off = sbi->blocks_per_seg;
1235
1236                 for (j = 0; j < blk_off; j++) {
1237                         struct f2fs_summary *s;
1238                         s = (struct f2fs_summary *)(kaddr + offset);
1239                         seg_i->sum_blk->entries[j] = *s;
1240                         offset += SUMMARY_SIZE;
1241                         if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1242                                                 SUM_FOOTER_SIZE)
1243                                 continue;
1244
1245                         f2fs_put_page(page, 1);
1246                         page = NULL;
1247
1248                         page = get_meta_page(sbi, start++);
1249                         kaddr = (unsigned char *)page_address(page);
1250                         offset = 0;
1251                 }
1252         }
1253         f2fs_put_page(page, 1);
1254         return 0;
1255 }
1256
1257 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1258 {
1259         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1260         struct f2fs_summary_block *sum;
1261         struct curseg_info *curseg;
1262         struct page *new;
1263         unsigned short blk_off;
1264         unsigned int segno = 0;
1265         block_t blk_addr = 0;
1266
1267         /* get segment number and block addr */
1268         if (IS_DATASEG(type)) {
1269                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1270                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1271                                                         CURSEG_HOT_DATA]);
1272                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1273                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1274                 else
1275                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1276         } else {
1277                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1278                                                         CURSEG_HOT_NODE]);
1279                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1280                                                         CURSEG_HOT_NODE]);
1281                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1282                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1283                                                         type - CURSEG_HOT_NODE);
1284                 else
1285                         blk_addr = GET_SUM_BLOCK(sbi, segno);
1286         }
1287
1288         new = get_meta_page(sbi, blk_addr);
1289         sum = (struct f2fs_summary_block *)page_address(new);
1290
1291         if (IS_NODESEG(type)) {
1292                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1293                         struct f2fs_summary *ns = &sum->entries[0];
1294                         int i;
1295                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1296                                 ns->version = 0;
1297                                 ns->ofs_in_node = 0;
1298                         }
1299                 } else {
1300                         int err;
1301
1302                         err = restore_node_summary(sbi, segno, sum);
1303                         if (err) {
1304                                 f2fs_put_page(new, 1);
1305                                 return err;
1306                         }
1307                 }
1308         }
1309
1310         /* set uncompleted segment to curseg */
1311         curseg = CURSEG_I(sbi, type);
1312         mutex_lock(&curseg->curseg_mutex);
1313         memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1314         curseg->next_segno = segno;
1315         reset_curseg(sbi, type, 0);
1316         curseg->alloc_type = ckpt->alloc_type[type];
1317         curseg->next_blkoff = blk_off;
1318         mutex_unlock(&curseg->curseg_mutex);
1319         f2fs_put_page(new, 1);
1320         return 0;
1321 }
1322
1323 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1324 {
1325         int type = CURSEG_HOT_DATA;
1326         int err;
1327
1328         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1329                 /* restore for compacted data summary */
1330                 if (read_compacted_summaries(sbi))
1331                         return -EINVAL;
1332                 type = CURSEG_HOT_NODE;
1333         }
1334
1335         for (; type <= CURSEG_COLD_NODE; type++) {
1336                 err = read_normal_summaries(sbi, type);
1337                 if (err)
1338                         return err;
1339         }
1340
1341         return 0;
1342 }
1343
1344 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1345 {
1346         struct page *page;
1347         unsigned char *kaddr;
1348         struct f2fs_summary *summary;
1349         struct curseg_info *seg_i;
1350         int written_size = 0;
1351         int i, j;
1352
1353         page = grab_meta_page(sbi, blkaddr++);
1354         kaddr = (unsigned char *)page_address(page);
1355
1356         /* Step 1: write nat cache */
1357         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1358         memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1359         written_size += SUM_JOURNAL_SIZE;
1360
1361         /* Step 2: write sit cache */
1362         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1363         memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1364                                                 SUM_JOURNAL_SIZE);
1365         written_size += SUM_JOURNAL_SIZE;
1366
1367         /* Step 3: write summary entries */
1368         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1369                 unsigned short blkoff;
1370                 seg_i = CURSEG_I(sbi, i);
1371                 if (sbi->ckpt->alloc_type[i] == SSR)
1372                         blkoff = sbi->blocks_per_seg;
1373                 else
1374                         blkoff = curseg_blkoff(sbi, i);
1375
1376                 for (j = 0; j < blkoff; j++) {
1377                         if (!page) {
1378                                 page = grab_meta_page(sbi, blkaddr++);
1379                                 kaddr = (unsigned char *)page_address(page);
1380                                 written_size = 0;
1381                         }
1382                         summary = (struct f2fs_summary *)(kaddr + written_size);
1383                         *summary = seg_i->sum_blk->entries[j];
1384                         written_size += SUMMARY_SIZE;
1385
1386                         if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1387                                                         SUM_FOOTER_SIZE)
1388                                 continue;
1389
1390                         set_page_dirty(page);
1391                         f2fs_put_page(page, 1);
1392                         page = NULL;
1393                 }
1394         }
1395         if (page) {
1396                 set_page_dirty(page);
1397                 f2fs_put_page(page, 1);
1398         }
1399 }
1400
1401 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1402                                         block_t blkaddr, int type)
1403 {
1404         int i, end;
1405         if (IS_DATASEG(type))
1406                 end = type + NR_CURSEG_DATA_TYPE;
1407         else
1408                 end = type + NR_CURSEG_NODE_TYPE;
1409
1410         for (i = type; i < end; i++) {
1411                 struct curseg_info *sum = CURSEG_I(sbi, i);
1412                 mutex_lock(&sum->curseg_mutex);
1413                 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1414                 mutex_unlock(&sum->curseg_mutex);
1415         }
1416 }
1417
1418 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1419 {
1420         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1421                 write_compacted_summaries(sbi, start_blk);
1422         else
1423                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1424 }
1425
1426 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1427 {
1428         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1429                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1430 }
1431
1432 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1433                                         unsigned int val, int alloc)
1434 {
1435         int i;
1436
1437         if (type == NAT_JOURNAL) {
1438                 for (i = 0; i < nats_in_cursum(sum); i++) {
1439                         if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1440                                 return i;
1441                 }
1442                 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1443                         return update_nats_in_cursum(sum, 1);
1444         } else if (type == SIT_JOURNAL) {
1445                 for (i = 0; i < sits_in_cursum(sum); i++)
1446                         if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1447                                 return i;
1448                 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1449                         return update_sits_in_cursum(sum, 1);
1450         }
1451         return -1;
1452 }
1453
1454 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1455                                         unsigned int segno)
1456 {
1457         struct sit_info *sit_i = SIT_I(sbi);
1458         unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno);
1459         block_t blk_addr = sit_i->sit_base_addr + offset;
1460
1461         check_seg_range(sbi, segno);
1462
1463         /* calculate sit block address */
1464         if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1465                 blk_addr += sit_i->sit_blocks;
1466
1467         return get_meta_page(sbi, blk_addr);
1468 }
1469
1470 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1471                                         unsigned int start)
1472 {
1473         struct sit_info *sit_i = SIT_I(sbi);
1474         struct page *src_page, *dst_page;
1475         pgoff_t src_off, dst_off;
1476         void *src_addr, *dst_addr;
1477
1478         src_off = current_sit_addr(sbi, start);
1479         dst_off = next_sit_addr(sbi, src_off);
1480
1481         /* get current sit block page without lock */
1482         src_page = get_meta_page(sbi, src_off);
1483         dst_page = grab_meta_page(sbi, dst_off);
1484         f2fs_bug_on(PageDirty(src_page));
1485
1486         src_addr = page_address(src_page);
1487         dst_addr = page_address(dst_page);
1488         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1489
1490         set_page_dirty(dst_page);
1491         f2fs_put_page(src_page, 1);
1492
1493         set_to_next_sit(sit_i, start);
1494
1495         return dst_page;
1496 }
1497
1498 static bool flush_sits_in_journal(struct f2fs_sb_info *sbi)
1499 {
1500         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1501         struct f2fs_summary_block *sum = curseg->sum_blk;
1502         int i;
1503
1504         /*
1505          * If the journal area in the current summary is full of sit entries,
1506          * all the sit entries will be flushed. Otherwise the sit entries
1507          * are not able to replace with newly hot sit entries.
1508          */
1509         if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) {
1510                 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1511                         unsigned int segno;
1512                         segno = le32_to_cpu(segno_in_journal(sum, i));
1513                         __mark_sit_entry_dirty(sbi, segno);
1514                 }
1515                 update_sits_in_cursum(sum, -sits_in_cursum(sum));
1516                 return true;
1517         }
1518         return false;
1519 }
1520
1521 /*
1522  * CP calls this function, which flushes SIT entries including sit_journal,
1523  * and moves prefree segs to free segs.
1524  */
1525 void flush_sit_entries(struct f2fs_sb_info *sbi)
1526 {
1527         struct sit_info *sit_i = SIT_I(sbi);
1528         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1529         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1530         struct f2fs_summary_block *sum = curseg->sum_blk;
1531         unsigned long nsegs = TOTAL_SEGS(sbi);
1532         struct page *page = NULL;
1533         struct f2fs_sit_block *raw_sit = NULL;
1534         unsigned int start = 0, end = 0;
1535         unsigned int segno = -1;
1536         bool flushed;
1537
1538         mutex_lock(&curseg->curseg_mutex);
1539         mutex_lock(&sit_i->sentry_lock);
1540
1541         /*
1542          * "flushed" indicates whether sit entries in journal are flushed
1543          * to the SIT area or not.
1544          */
1545         flushed = flush_sits_in_journal(sbi);
1546
1547         while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) {
1548                 struct seg_entry *se = get_seg_entry(sbi, segno);
1549                 int sit_offset, offset;
1550
1551                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1552
1553                 /* add discard candidates */
1554                 if (SM_I(sbi)->nr_discards < SM_I(sbi)->max_discards)
1555                         add_discard_addrs(sbi, segno, se);
1556
1557                 if (flushed)
1558                         goto to_sit_page;
1559
1560                 offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1);
1561                 if (offset >= 0) {
1562                         segno_in_journal(sum, offset) = cpu_to_le32(segno);
1563                         seg_info_to_raw_sit(se, &sit_in_journal(sum, offset));
1564                         goto flush_done;
1565                 }
1566 to_sit_page:
1567                 if (!page || (start > segno) || (segno > end)) {
1568                         if (page) {
1569                                 f2fs_put_page(page, 1);
1570                                 page = NULL;
1571                         }
1572
1573                         start = START_SEGNO(sit_i, segno);
1574                         end = start + SIT_ENTRY_PER_BLOCK - 1;
1575
1576                         /* read sit block that will be updated */
1577                         page = get_next_sit_page(sbi, start);
1578                         raw_sit = page_address(page);
1579                 }
1580
1581                 /* udpate entry in SIT block */
1582                 seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]);
1583 flush_done:
1584                 __clear_bit(segno, bitmap);
1585                 sit_i->dirty_sentries--;
1586         }
1587         mutex_unlock(&sit_i->sentry_lock);
1588         mutex_unlock(&curseg->curseg_mutex);
1589
1590         /* writeout last modified SIT block */
1591         f2fs_put_page(page, 1);
1592
1593         set_prefree_as_free_segments(sbi);
1594 }
1595
1596 static int build_sit_info(struct f2fs_sb_info *sbi)
1597 {
1598         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1599         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1600         struct sit_info *sit_i;
1601         unsigned int sit_segs, start;
1602         char *src_bitmap, *dst_bitmap;
1603         unsigned int bitmap_size;
1604
1605         /* allocate memory for SIT information */
1606         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1607         if (!sit_i)
1608                 return -ENOMEM;
1609
1610         SM_I(sbi)->sit_info = sit_i;
1611
1612         sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1613         if (!sit_i->sentries)
1614                 return -ENOMEM;
1615
1616         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1617         sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1618         if (!sit_i->dirty_sentries_bitmap)
1619                 return -ENOMEM;
1620
1621         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1622                 sit_i->sentries[start].cur_valid_map
1623                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1624                 sit_i->sentries[start].ckpt_valid_map
1625                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1626                 if (!sit_i->sentries[start].cur_valid_map
1627                                 || !sit_i->sentries[start].ckpt_valid_map)
1628                         return -ENOMEM;
1629         }
1630
1631         if (sbi->segs_per_sec > 1) {
1632                 sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) *
1633                                         sizeof(struct sec_entry));
1634                 if (!sit_i->sec_entries)
1635                         return -ENOMEM;
1636         }
1637
1638         /* get information related with SIT */
1639         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1640
1641         /* setup SIT bitmap from ckeckpoint pack */
1642         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1643         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1644
1645         dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1646         if (!dst_bitmap)
1647                 return -ENOMEM;
1648
1649         /* init SIT information */
1650         sit_i->s_ops = &default_salloc_ops;
1651
1652         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1653         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1654         sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1655         sit_i->sit_bitmap = dst_bitmap;
1656         sit_i->bitmap_size = bitmap_size;
1657         sit_i->dirty_sentries = 0;
1658         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1659         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1660         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1661         mutex_init(&sit_i->sentry_lock);
1662         return 0;
1663 }
1664
1665 static int build_free_segmap(struct f2fs_sb_info *sbi)
1666 {
1667         struct f2fs_sm_info *sm_info = SM_I(sbi);
1668         struct free_segmap_info *free_i;
1669         unsigned int bitmap_size, sec_bitmap_size;
1670
1671         /* allocate memory for free segmap information */
1672         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1673         if (!free_i)
1674                 return -ENOMEM;
1675
1676         SM_I(sbi)->free_info = free_i;
1677
1678         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1679         free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1680         if (!free_i->free_segmap)
1681                 return -ENOMEM;
1682
1683         sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1684         free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1685         if (!free_i->free_secmap)
1686                 return -ENOMEM;
1687
1688         /* set all segments as dirty temporarily */
1689         memset(free_i->free_segmap, 0xff, bitmap_size);
1690         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1691
1692         /* init free segmap information */
1693         free_i->start_segno =
1694                 (unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr);
1695         free_i->free_segments = 0;
1696         free_i->free_sections = 0;
1697         rwlock_init(&free_i->segmap_lock);
1698         return 0;
1699 }
1700
1701 static int build_curseg(struct f2fs_sb_info *sbi)
1702 {
1703         struct curseg_info *array;
1704         int i;
1705
1706         array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
1707         if (!array)
1708                 return -ENOMEM;
1709
1710         SM_I(sbi)->curseg_array = array;
1711
1712         for (i = 0; i < NR_CURSEG_TYPE; i++) {
1713                 mutex_init(&array[i].curseg_mutex);
1714                 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1715                 if (!array[i].sum_blk)
1716                         return -ENOMEM;
1717                 array[i].segno = NULL_SEGNO;
1718                 array[i].next_blkoff = 0;
1719         }
1720         return restore_curseg_summaries(sbi);
1721 }
1722
1723 static void build_sit_entries(struct f2fs_sb_info *sbi)
1724 {
1725         struct sit_info *sit_i = SIT_I(sbi);
1726         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1727         struct f2fs_summary_block *sum = curseg->sum_blk;
1728         int sit_blk_cnt = SIT_BLK_CNT(sbi);
1729         unsigned int i, start, end;
1730         unsigned int readed, start_blk = 0;
1731         int nrpages = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1732
1733         do {
1734                 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT);
1735
1736                 start = start_blk * sit_i->sents_per_block;
1737                 end = (start_blk + readed) * sit_i->sents_per_block;
1738
1739                 for (; start < end && start < TOTAL_SEGS(sbi); start++) {
1740                         struct seg_entry *se = &sit_i->sentries[start];
1741                         struct f2fs_sit_block *sit_blk;
1742                         struct f2fs_sit_entry sit;
1743                         struct page *page;
1744
1745                         mutex_lock(&curseg->curseg_mutex);
1746                         for (i = 0; i < sits_in_cursum(sum); i++) {
1747                                 if (le32_to_cpu(segno_in_journal(sum, i))
1748                                                                 == start) {
1749                                         sit = sit_in_journal(sum, i);
1750                                         mutex_unlock(&curseg->curseg_mutex);
1751                                         goto got_it;
1752                                 }
1753                         }
1754                         mutex_unlock(&curseg->curseg_mutex);
1755
1756                         page = get_current_sit_page(sbi, start);
1757                         sit_blk = (struct f2fs_sit_block *)page_address(page);
1758                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1759                         f2fs_put_page(page, 1);
1760 got_it:
1761                         check_block_count(sbi, start, &sit);
1762                         seg_info_from_raw_sit(se, &sit);
1763                         if (sbi->segs_per_sec > 1) {
1764                                 struct sec_entry *e = get_sec_entry(sbi, start);
1765                                 e->valid_blocks += se->valid_blocks;
1766                         }
1767                 }
1768                 start_blk += readed;
1769         } while (start_blk < sit_blk_cnt);
1770 }
1771
1772 static void init_free_segmap(struct f2fs_sb_info *sbi)
1773 {
1774         unsigned int start;
1775         int type;
1776
1777         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1778                 struct seg_entry *sentry = get_seg_entry(sbi, start);
1779                 if (!sentry->valid_blocks)
1780                         __set_free(sbi, start);
1781         }
1782
1783         /* set use the current segments */
1784         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1785                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1786                 __set_test_and_inuse(sbi, curseg_t->segno);
1787         }
1788 }
1789
1790 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1791 {
1792         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1793         struct free_segmap_info *free_i = FREE_I(sbi);
1794         unsigned int segno = 0, offset = 0, total_segs = TOTAL_SEGS(sbi);
1795         unsigned short valid_blocks;
1796
1797         while (1) {
1798                 /* find dirty segment based on free segmap */
1799                 segno = find_next_inuse(free_i, total_segs, offset);
1800                 if (segno >= total_segs)
1801                         break;
1802                 offset = segno + 1;
1803                 valid_blocks = get_valid_blocks(sbi, segno, 0);
1804                 if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks)
1805                         continue;
1806                 mutex_lock(&dirty_i->seglist_lock);
1807                 __locate_dirty_segment(sbi, segno, DIRTY);
1808                 mutex_unlock(&dirty_i->seglist_lock);
1809         }
1810 }
1811
1812 static int init_victim_secmap(struct f2fs_sb_info *sbi)
1813 {
1814         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1815         unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1816
1817         dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
1818         if (!dirty_i->victim_secmap)
1819                 return -ENOMEM;
1820         return 0;
1821 }
1822
1823 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1824 {
1825         struct dirty_seglist_info *dirty_i;
1826         unsigned int bitmap_size, i;
1827
1828         /* allocate memory for dirty segments list information */
1829         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1830         if (!dirty_i)
1831                 return -ENOMEM;
1832
1833         SM_I(sbi)->dirty_info = dirty_i;
1834         mutex_init(&dirty_i->seglist_lock);
1835
1836         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1837
1838         for (i = 0; i < NR_DIRTY_TYPE; i++) {
1839                 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
1840                 if (!dirty_i->dirty_segmap[i])
1841                         return -ENOMEM;
1842         }
1843
1844         init_dirty_segmap(sbi);
1845         return init_victim_secmap(sbi);
1846 }
1847
1848 /*
1849  * Update min, max modified time for cost-benefit GC algorithm
1850  */
1851 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1852 {
1853         struct sit_info *sit_i = SIT_I(sbi);
1854         unsigned int segno;
1855
1856         mutex_lock(&sit_i->sentry_lock);
1857
1858         sit_i->min_mtime = LLONG_MAX;
1859
1860         for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1861                 unsigned int i;
1862                 unsigned long long mtime = 0;
1863
1864                 for (i = 0; i < sbi->segs_per_sec; i++)
1865                         mtime += get_seg_entry(sbi, segno + i)->mtime;
1866
1867                 mtime = div_u64(mtime, sbi->segs_per_sec);
1868
1869                 if (sit_i->min_mtime > mtime)
1870                         sit_i->min_mtime = mtime;
1871         }
1872         sit_i->max_mtime = get_mtime(sbi);
1873         mutex_unlock(&sit_i->sentry_lock);
1874 }
1875
1876 int build_segment_manager(struct f2fs_sb_info *sbi)
1877 {
1878         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1879         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1880         struct f2fs_sm_info *sm_info;
1881         int err;
1882
1883         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1884         if (!sm_info)
1885                 return -ENOMEM;
1886
1887         /* init sm info */
1888         sbi->sm_info = sm_info;
1889         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1890         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1891         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1892         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1893         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1894         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1895         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
1896         sm_info->rec_prefree_segments = sm_info->main_segments *
1897                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
1898         sm_info->ipu_policy = F2FS_IPU_DISABLE;
1899         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
1900
1901         INIT_LIST_HEAD(&sm_info->discard_list);
1902         sm_info->nr_discards = 0;
1903         sm_info->max_discards = 0;
1904
1905         if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
1906                 err = create_flush_cmd_control(sbi);
1907                 if (err)
1908                         return err;
1909         }
1910
1911         err = build_sit_info(sbi);
1912         if (err)
1913                 return err;
1914         err = build_free_segmap(sbi);
1915         if (err)
1916                 return err;
1917         err = build_curseg(sbi);
1918         if (err)
1919                 return err;
1920
1921         /* reinit free segmap based on SIT */
1922         build_sit_entries(sbi);
1923
1924         init_free_segmap(sbi);
1925         err = build_dirty_segmap(sbi);
1926         if (err)
1927                 return err;
1928
1929         init_min_max_mtime(sbi);
1930         return 0;
1931 }
1932
1933 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
1934                 enum dirty_type dirty_type)
1935 {
1936         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1937
1938         mutex_lock(&dirty_i->seglist_lock);
1939         kfree(dirty_i->dirty_segmap[dirty_type]);
1940         dirty_i->nr_dirty[dirty_type] = 0;
1941         mutex_unlock(&dirty_i->seglist_lock);
1942 }
1943
1944 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
1945 {
1946         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1947         kfree(dirty_i->victim_secmap);
1948 }
1949
1950 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
1951 {
1952         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1953         int i;
1954
1955         if (!dirty_i)
1956                 return;
1957
1958         /* discard pre-free/dirty segments list */
1959         for (i = 0; i < NR_DIRTY_TYPE; i++)
1960                 discard_dirty_segmap(sbi, i);
1961
1962         destroy_victim_secmap(sbi);
1963         SM_I(sbi)->dirty_info = NULL;
1964         kfree(dirty_i);
1965 }
1966
1967 static void destroy_curseg(struct f2fs_sb_info *sbi)
1968 {
1969         struct curseg_info *array = SM_I(sbi)->curseg_array;
1970         int i;
1971
1972         if (!array)
1973                 return;
1974         SM_I(sbi)->curseg_array = NULL;
1975         for (i = 0; i < NR_CURSEG_TYPE; i++)
1976                 kfree(array[i].sum_blk);
1977         kfree(array);
1978 }
1979
1980 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
1981 {
1982         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
1983         if (!free_i)
1984                 return;
1985         SM_I(sbi)->free_info = NULL;
1986         kfree(free_i->free_segmap);
1987         kfree(free_i->free_secmap);
1988         kfree(free_i);
1989 }
1990
1991 static void destroy_sit_info(struct f2fs_sb_info *sbi)
1992 {
1993         struct sit_info *sit_i = SIT_I(sbi);
1994         unsigned int start;
1995
1996         if (!sit_i)
1997                 return;
1998
1999         if (sit_i->sentries) {
2000                 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
2001                         kfree(sit_i->sentries[start].cur_valid_map);
2002                         kfree(sit_i->sentries[start].ckpt_valid_map);
2003                 }
2004         }
2005         vfree(sit_i->sentries);
2006         vfree(sit_i->sec_entries);
2007         kfree(sit_i->dirty_sentries_bitmap);
2008
2009         SM_I(sbi)->sit_info = NULL;
2010         kfree(sit_i->sit_bitmap);
2011         kfree(sit_i);
2012 }
2013
2014 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2015 {
2016         struct f2fs_sm_info *sm_info = SM_I(sbi);
2017
2018         if (!sm_info)
2019                 return;
2020         destroy_flush_cmd_control(sbi);
2021         destroy_dirty_segmap(sbi);
2022         destroy_curseg(sbi);
2023         destroy_free_segmap(sbi);
2024         destroy_sit_info(sbi);
2025         sbi->sm_info = NULL;
2026         kfree(sm_info);
2027 }
2028
2029 int __init create_segment_manager_caches(void)
2030 {
2031         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2032                         sizeof(struct discard_entry));
2033         if (!discard_entry_slab)
2034                 return -ENOMEM;
2035         return 0;
2036 }
2037
2038 void destroy_segment_manager_caches(void)
2039 {
2040         kmem_cache_destroy(discard_entry_slab);
2041 }