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 static struct kmem_cache *sit_entry_set_slab;
  29
  30 /*
  31  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  32  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  33  */
  34 static inline unsigned long __reverse_ffs(unsigned long word)
  35 {
  36         int num = 0;
  37
  38 #if BITS_PER_LONG == 64
  39         if ((word & 0xffffffff) == 0) {
  40                 num += 32;
  41                 word >>= 32;
  42         }
  43 #endif
  44         if ((word & 0xffff) == 0) {
  45                 num += 16;
  46                 word >>= 16;
  47         }
  48         if ((word & 0xff) == 0) {
  49                 num += 8;
  50                 word >>= 8;
  51         }
  52         if ((word & 0xf0) == 0)
  53                 num += 4;
  54         else
  55                 word >>= 4;
  56         if ((word & 0xc) == 0)
  57                 num += 2;
  58         else
  59                 word >>= 2;
  60         if ((word & 0x2) == 0)
  61                 num += 1;
  62         return num;
  63 }
  64
  65 /*
  66  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  67  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  68  * Example:
  69  *                             LSB <--> MSB
  70  *   f2fs_set_bit(0, bitmap) => 0000 0001
  71  *   f2fs_set_bit(7, bitmap) => 1000 0000
  72  */
  73 static unsigned long __find_rev_next_bit(const unsigned long *addr,
  74                         unsigned long size, unsigned long offset)
  75 {
  76         const unsigned long *p = addr + BIT_WORD(offset);
  77         unsigned long result = offset & ~(BITS_PER_LONG - 1);
  78         unsigned long tmp;
  79         unsigned long mask, submask;
  80         unsigned long quot, rest;
  81
  82         if (offset >= size)
  83                 return size;
  84
  85         size -= result;
  86         offset %= BITS_PER_LONG;
  87         if (!offset)
  88                 goto aligned;
  89
  90         tmp = *(p++);
  91         quot = (offset >> 3) << 3;
  92         rest = offset & 0x7;
  93         mask = ~0UL << quot;
  94         submask = (unsigned char)(0xff << rest) >> rest;
  95         submask <<= quot;
  96         mask &= submask;
  97         tmp &= mask;
  98         if (size < BITS_PER_LONG)
  99                 goto found_first;
 100         if (tmp)
 101                 goto found_middle;
 102
 103         size -= BITS_PER_LONG;
 104         result += BITS_PER_LONG;
 105 aligned:
 106         while (size & ~(BITS_PER_LONG-1)) {
 107                 tmp = *(p++);
 108                 if (tmp)
 109                         goto found_middle;
 110                 result += BITS_PER_LONG;
 111                 size -= BITS_PER_LONG;
 112         }
 113         if (!size)
 114                 return result;
 115         tmp = *p;
 116 found_first:
 117         tmp &= (~0UL >> (BITS_PER_LONG - size));
 118         if (tmp == 0UL)         /* Are any bits set? */
 119                 return result + size;   /* Nope. */
 120 found_middle:
 121         return result + __reverse_ffs(tmp);
 122 }
 123
 124 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 125                         unsigned long size, unsigned long offset)
 126 {
 127         const unsigned long *p = addr + BIT_WORD(offset);
 128         unsigned long result = offset & ~(BITS_PER_LONG - 1);
 129         unsigned long tmp;
 130         unsigned long mask, submask;
 131         unsigned long quot, rest;
 132
 133         if (offset >= size)
 134                 return size;
 135
 136         size -= result;
 137         offset %= BITS_PER_LONG;
 138         if (!offset)
 139                 goto aligned;
 140
 141         tmp = *(p++);
 142         quot = (offset >> 3) << 3;
 143         rest = offset & 0x7;
 144         mask = ~(~0UL << quot);
 145         submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest);
 146         submask <<= quot;
 147         mask += submask;
 148         tmp |= mask;
 149         if (size < BITS_PER_LONG)
 150                 goto found_first;
 151         if (~tmp)
 152                 goto found_middle;
 153
 154         size -= BITS_PER_LONG;
 155         result += BITS_PER_LONG;
 156 aligned:
 157         while (size & ~(BITS_PER_LONG - 1)) {
 158                 tmp = *(p++);
 159                 if (~tmp)
 160                         goto found_middle;
 161                 result += BITS_PER_LONG;
 162                 size -= BITS_PER_LONG;
 163         }
 164         if (!size)
 165                 return result;
 166         tmp = *p;
 167
 168 found_first:
 169         tmp |= ~0UL << size;
 170         if (tmp == ~0UL)        /* Are any bits zero? */
 171                 return result + size;   /* Nope. */
 172 found_middle:
 173         return result + __reverse_ffz(tmp);
 174 }
 175
 176 /*
 177  * This function balances dirty node and dentry pages.
 178  * In addition, it controls garbage collection.
 179  */
 180 void f2fs_balance_fs(struct f2fs_sb_info *sbi)
 181 {
 182         /*
 183          * We should do GC or end up with checkpoint, if there are so many dirty
 184          * dir/node pages without enough free segments.
 185          */
 186         if (has_not_enough_free_secs(sbi, 0)) {
 187                 mutex_lock(&sbi->gc_mutex);
 188                 f2fs_gc(sbi);
 189         }
 190 }
 191
 192 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 193 {
 194         /* check the # of cached NAT entries and prefree segments */
 195         if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) ||
 196                                 excess_prefree_segs(sbi))
 197                 f2fs_sync_fs(sbi->sb, true);
 198 }
 199
 200 static int issue_flush_thread(void *data)
 201 {
 202         struct f2fs_sb_info *sbi = data;
 203         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 204         wait_queue_head_t *q = &fcc->flush_wait_queue;
 205 repeat:
 206         if (kthread_should_stop())
 207                 return 0;
 208
 209         spin_lock(&fcc->issue_lock);
 210         if (fcc->issue_list) {
 211                 fcc->dispatch_list = fcc->issue_list;
 212                 fcc->issue_list = fcc->issue_tail = NULL;
 213         }
 214         spin_unlock(&fcc->issue_lock);
 215
 216         if (fcc->dispatch_list) {
 217                 struct bio *bio = bio_alloc(GFP_NOIO, 0);
 218                 struct flush_cmd *cmd, *next;
 219                 int ret;
 220
 221                 bio->bi_bdev = sbi->sb->s_bdev;
 222                 ret = submit_bio_wait(WRITE_FLUSH, bio);
 223
 224                 for (cmd = fcc->dispatch_list; cmd; cmd = next) {
 225                         cmd->ret = ret;
 226                         next = cmd->next;
 227                         complete(&cmd->wait);
 228                 }
 229                 bio_put(bio);
 230                 fcc->dispatch_list = NULL;
 231         }
 232
 233         wait_event_interruptible(*q,
 234                         kthread_should_stop() || fcc->issue_list);
 235         goto repeat;
 236 }
 237
 238 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
 239 {
 240         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 241         struct flush_cmd cmd;
 242
 243         trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
 244                                         test_opt(sbi, FLUSH_MERGE));
 245
 246         if (test_opt(sbi, NOBARRIER))
 247                 return 0;
 248
 249         if (!test_opt(sbi, FLUSH_MERGE))
 250                 return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL);
 251
 252         init_completion(&cmd.wait);
 253         cmd.next = NULL;
 254
 255         spin_lock(&fcc->issue_lock);
 256         if (fcc->issue_list)
 257                 fcc->issue_tail->next = &cmd;
 258         else
 259                 fcc->issue_list = &cmd;
 260         fcc->issue_tail = &cmd;
 261         spin_unlock(&fcc->issue_lock);
 262
 263         if (!fcc->dispatch_list)
 264                 wake_up(&fcc->flush_wait_queue);
 265
 266         wait_for_completion(&cmd.wait);
 267
 268         return cmd.ret;
 269 }
 270
 271 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
 272 {
 273         dev_t dev = sbi->sb->s_bdev->bd_dev;
 274         struct flush_cmd_control *fcc;
 275         int err = 0;
 276
 277         fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
 278         if (!fcc)
 279                 return -ENOMEM;
 280         spin_lock_init(&fcc->issue_lock);
 281         init_waitqueue_head(&fcc->flush_wait_queue);
 282         SM_I(sbi)->cmd_control_info = fcc;
 283         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 284                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 285         if (IS_ERR(fcc->f2fs_issue_flush)) {
 286                 err = PTR_ERR(fcc->f2fs_issue_flush);
 287                 kfree(fcc);
 288                 SM_I(sbi)->cmd_control_info = NULL;
 289                 return err;
 290         }
 291
 292         return err;
 293 }
 294
 295 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
 296 {
 297         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 298
 299         if (fcc && fcc->f2fs_issue_flush)
 300                 kthread_stop(fcc->f2fs_issue_flush);
 301         kfree(fcc);
 302         SM_I(sbi)->cmd_control_info = NULL;
 303 }
 304
 305 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 306                 enum dirty_type dirty_type)
 307 {
 308         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 309
 310         /* need not be added */
 311         if (IS_CURSEG(sbi, segno))
 312                 return;
 313
 314         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 315                 dirty_i->nr_dirty[dirty_type]++;
 316
 317         if (dirty_type == DIRTY) {
 318                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 319                 enum dirty_type t = sentry->type;
 320
 321                 if (unlikely(t >= DIRTY)) {
 322                         f2fs_bug_on(sbi, 1);
 323                         return;
 324                 }
 325                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 326                         dirty_i->nr_dirty[t]++;
 327         }
 328 }
 329
 330 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 331                 enum dirty_type dirty_type)
 332 {
 333         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 334
 335         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 336                 dirty_i->nr_dirty[dirty_type]--;
 337
 338         if (dirty_type == DIRTY) {
 339                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 340                 enum dirty_type t = sentry->type;
 341
 342                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 343                         dirty_i->nr_dirty[t]--;
 344
 345                 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
 346                         clear_bit(GET_SECNO(sbi, segno),
 347                                                 dirty_i->victim_secmap);
 348         }
 349 }
 350
 351 /*
 352  * Should not occur error such as -ENOMEM.
 353  * Adding dirty entry into seglist is not critical operation.
 354  * If a given segment is one of current working segments, it won't be added.
 355  */
 356 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 357 {
 358         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 359         unsigned short valid_blocks;
 360
 361         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 362                 return;
 363
 364         mutex_lock(&dirty_i->seglist_lock);
 365
 366         valid_blocks = get_valid_blocks(sbi, segno, 0);
 367
 368         if (valid_blocks == 0) {
 369                 __locate_dirty_segment(sbi, segno, PRE);
 370                 __remove_dirty_segment(sbi, segno, DIRTY);
 371         } else if (valid_blocks < sbi->blocks_per_seg) {
 372                 __locate_dirty_segment(sbi, segno, DIRTY);
 373         } else {
 374                 /* Recovery routine with SSR needs this */
 375                 __remove_dirty_segment(sbi, segno, DIRTY);
 376         }
 377
 378         mutex_unlock(&dirty_i->seglist_lock);
 379 }
 380
 381 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
 382                                 block_t blkstart, block_t blklen)
 383 {
 384         sector_t start = SECTOR_FROM_BLOCK(sbi, blkstart);
 385         sector_t len = SECTOR_FROM_BLOCK(sbi, blklen);
 386         trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
 387         return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
 388 }
 389
 390 void discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
 391 {
 392         if (f2fs_issue_discard(sbi, blkaddr, 1)) {
 393                 struct page *page = grab_meta_page(sbi, blkaddr);
 394                 /* zero-filled page */
 395                 set_page_dirty(page);
 396                 f2fs_put_page(page, 1);
 397         }
 398 }
 399
 400 static void add_discard_addrs(struct f2fs_sb_info *sbi,
 401                         unsigned int segno, struct seg_entry *se)
 402 {
 403         struct list_head *head = &SM_I(sbi)->discard_list;
 404         struct discard_entry *new;
 405         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
 406         int max_blocks = sbi->blocks_per_seg;
 407         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
 408         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
 409         unsigned long dmap[entries];
 410         unsigned int start = 0, end = -1;
 411         int i;
 412
 413         if (!test_opt(sbi, DISCARD))
 414                 return;
 415
 416         /* zero block will be discarded through the prefree list */
 417         if (!se->valid_blocks || se->valid_blocks == max_blocks)
 418                 return;
 419
 420         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
 421         for (i = 0; i < entries; i++)
 422                 dmap[i] = (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
 423
 424         while (SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
 425                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
 426                 if (start >= max_blocks)
 427                         break;
 428
 429                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
 430
 431                 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
 432                 INIT_LIST_HEAD(&new->list);
 433                 new->blkaddr = START_BLOCK(sbi, segno) + start;
 434                 new->len = end - start;
 435
 436                 list_add_tail(&new->list, head);
 437                 SM_I(sbi)->nr_discards += end - start;
 438         }
 439 }
 440
 441 /*
 442  * Should call clear_prefree_segments after checkpoint is done.
 443  */
 444 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
 445 {
 446         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 447         unsigned int segno;
 448         unsigned int total_segs = TOTAL_SEGS(sbi);
 449
 450         mutex_lock(&dirty_i->seglist_lock);
 451         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], total_segs)
 452                 __set_test_and_free(sbi, segno);
 453         mutex_unlock(&dirty_i->seglist_lock);
 454 }
 455
 456 void clear_prefree_segments(struct f2fs_sb_info *sbi)
 457 {
 458         struct list_head *head = &(SM_I(sbi)->discard_list);
 459         struct discard_entry *entry, *this;
 460         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 461         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
 462         unsigned int total_segs = TOTAL_SEGS(sbi);
 463         unsigned int start = 0, end = -1;
 464
 465         mutex_lock(&dirty_i->seglist_lock);
 466
 467         while (1) {
 468                 int i;
 469                 start = find_next_bit(prefree_map, total_segs, end + 1);
 470                 if (start >= total_segs)
 471                         break;
 472                 end = find_next_zero_bit(prefree_map, total_segs, start + 1);
 473
 474                 for (i = start; i < end; i++)
 475                         clear_bit(i, prefree_map);
 476
 477                 dirty_i->nr_dirty[PRE] -= end - start;
 478
 479                 if (!test_opt(sbi, DISCARD))
 480                         continue;
 481
 482                 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
 483                                 (end - start) << sbi->log_blocks_per_seg);
 484         }
 485         mutex_unlock(&dirty_i->seglist_lock);
 486
 487         /* send small discards */
 488         list_for_each_entry_safe(entry, this, head, list) {
 489                 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
 490                 list_del(&entry->list);
 491                 SM_I(sbi)->nr_discards -= entry->len;
 492                 kmem_cache_free(discard_entry_slab, entry);
 493         }
 494 }
 495
 496 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
 497 {
 498         struct sit_info *sit_i = SIT_I(sbi);
 499
 500         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
 501                 sit_i->dirty_sentries++;
 502                 return false;
 503         }
 504
 505         return true;
 506 }
 507
 508 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
 509                                         unsigned int segno, int modified)
 510 {
 511         struct seg_entry *se = get_seg_entry(sbi, segno);
 512         se->type = type;
 513         if (modified)
 514                 __mark_sit_entry_dirty(sbi, segno);
 515 }
 516
 517 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
 518 {
 519         struct seg_entry *se;
 520         unsigned int segno, offset;
 521         long int new_vblocks;
 522
 523         segno = GET_SEGNO(sbi, blkaddr);
 524
 525         se = get_seg_entry(sbi, segno);
 526         new_vblocks = se->valid_blocks + del;
 527         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 528
 529         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
 530                                 (new_vblocks > sbi->blocks_per_seg)));
 531
 532         se->valid_blocks = new_vblocks;
 533         se->mtime = get_mtime(sbi);
 534         SIT_I(sbi)->max_mtime = se->mtime;
 535
 536         /* Update valid block bitmap */
 537         if (del > 0) {
 538                 if (f2fs_set_bit(offset, se->cur_valid_map))
 539                         f2fs_bug_on(sbi, 1);
 540         } else {
 541                 if (!f2fs_clear_bit(offset, se->cur_valid_map))
 542                         f2fs_bug_on(sbi, 1);
 543         }
 544         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
 545                 se->ckpt_valid_blocks += del;
 546
 547         __mark_sit_entry_dirty(sbi, segno);
 548
 549         /* update total number of valid blocks to be written in ckpt area */
 550         SIT_I(sbi)->written_valid_blocks += del;
 551
 552         if (sbi->segs_per_sec > 1)
 553                 get_sec_entry(sbi, segno)->valid_blocks += del;
 554 }
 555
 556 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
 557 {
 558         update_sit_entry(sbi, new, 1);
 559         if (GET_SEGNO(sbi, old) != NULL_SEGNO)
 560                 update_sit_entry(sbi, old, -1);
 561
 562         locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
 563         locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
 564 }
 565
 566 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
 567 {
 568         unsigned int segno = GET_SEGNO(sbi, addr);
 569         struct sit_info *sit_i = SIT_I(sbi);
 570
 571         f2fs_bug_on(sbi, addr == NULL_ADDR);
 572         if (addr == NEW_ADDR)
 573                 return;
 574
 575         /* add it into sit main buffer */
 576         mutex_lock(&sit_i->sentry_lock);
 577
 578         update_sit_entry(sbi, addr, -1);
 579
 580         /* add it into dirty seglist */
 581         locate_dirty_segment(sbi, segno);
 582
 583         mutex_unlock(&sit_i->sentry_lock);
 584 }
 585
 586 /*
 587  * This function should be resided under the curseg_mutex lock
 588  */
 589 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
 590                                         struct f2fs_summary *sum)
 591 {
 592         struct curseg_info *curseg = CURSEG_I(sbi, type);
 593         void *addr = curseg->sum_blk;
 594         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
 595         memcpy(addr, sum, sizeof(struct f2fs_summary));
 596 }
 597
 598 /*
 599  * Calculate the number of current summary pages for writing
 600  */
 601 int npages_for_summary_flush(struct f2fs_sb_info *sbi)
 602 {
 603         int valid_sum_count = 0;
 604         int i, sum_in_page;
 605
 606         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 607                 if (sbi->ckpt->alloc_type[i] == SSR)
 608                         valid_sum_count += sbi->blocks_per_seg;
 609                 else
 610                         valid_sum_count += curseg_blkoff(sbi, i);
 611         }
 612
 613         sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
 614                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
 615         if (valid_sum_count <= sum_in_page)
 616                 return 1;
 617         else if ((valid_sum_count - sum_in_page) <=
 618                 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
 619                 return 2;
 620         return 3;
 621 }
 622
 623 /*
 624  * Caller should put this summary page
 625  */
 626 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
 627 {
 628         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
 629 }
 630
 631 static void write_sum_page(struct f2fs_sb_info *sbi,
 632                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
 633 {
 634         struct page *page = grab_meta_page(sbi, blk_addr);
 635         void *kaddr = page_address(page);
 636         memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
 637         set_page_dirty(page);
 638         f2fs_put_page(page, 1);
 639 }
 640
 641 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
 642 {
 643         struct curseg_info *curseg = CURSEG_I(sbi, type);
 644         unsigned int segno = curseg->segno + 1;
 645         struct free_segmap_info *free_i = FREE_I(sbi);
 646
 647         if (segno < TOTAL_SEGS(sbi) && segno % sbi->segs_per_sec)
 648                 return !test_bit(segno, free_i->free_segmap);
 649         return 0;
 650 }
 651
 652 /*
 653  * Find a new segment from the free segments bitmap to right order
 654  * This function should be returned with success, otherwise BUG
 655  */
 656 static void get_new_segment(struct f2fs_sb_info *sbi,
 657                         unsigned int *newseg, bool new_sec, int dir)
 658 {
 659         struct free_segmap_info *free_i = FREE_I(sbi);
 660         unsigned int segno, secno, zoneno;
 661         unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone;
 662         unsigned int hint = *newseg / sbi->segs_per_sec;
 663         unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
 664         unsigned int left_start = hint;
 665         bool init = true;
 666         int go_left = 0;
 667         int i;
 668
 669         write_lock(&free_i->segmap_lock);
 670
 671         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
 672                 segno = find_next_zero_bit(free_i->free_segmap,
 673                                         TOTAL_SEGS(sbi), *newseg + 1);
 674                 if (segno - *newseg < sbi->segs_per_sec -
 675                                         (*newseg % sbi->segs_per_sec))
 676                         goto got_it;
 677         }
 678 find_other_zone:
 679         secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint);
 680         if (secno >= TOTAL_SECS(sbi)) {
 681                 if (dir == ALLOC_RIGHT) {
 682                         secno = find_next_zero_bit(free_i->free_secmap,
 683                                                         TOTAL_SECS(sbi), 0);
 684                         f2fs_bug_on(sbi, secno >= TOTAL_SECS(sbi));
 685                 } else {
 686                         go_left = 1;
 687                         left_start = hint - 1;
 688                 }
 689         }
 690         if (go_left == 0)
 691                 goto skip_left;
 692
 693         while (test_bit(left_start, free_i->free_secmap)) {
 694                 if (left_start > 0) {
 695                         left_start--;
 696                         continue;
 697                 }
 698                 left_start = find_next_zero_bit(free_i->free_secmap,
 699                                                         TOTAL_SECS(sbi), 0);
 700                 f2fs_bug_on(sbi, left_start >= TOTAL_SECS(sbi));
 701                 break;
 702         }
 703         secno = left_start;
 704 skip_left:
 705         hint = secno;
 706         segno = secno * sbi->segs_per_sec;
 707         zoneno = secno / sbi->secs_per_zone;
 708
 709         /* give up on finding another zone */
 710         if (!init)
 711                 goto got_it;
 712         if (sbi->secs_per_zone == 1)
 713                 goto got_it;
 714         if (zoneno == old_zoneno)
 715                 goto got_it;
 716         if (dir == ALLOC_LEFT) {
 717                 if (!go_left && zoneno + 1 >= total_zones)
 718                         goto got_it;
 719                 if (go_left && zoneno == 0)
 720                         goto got_it;
 721         }
 722         for (i = 0; i < NR_CURSEG_TYPE; i++)
 723                 if (CURSEG_I(sbi, i)->zone == zoneno)
 724                         break;
 725
 726         if (i < NR_CURSEG_TYPE) {
 727                 /* zone is in user, try another */
 728                 if (go_left)
 729                         hint = zoneno * sbi->secs_per_zone - 1;
 730                 else if (zoneno + 1 >= total_zones)
 731                         hint = 0;
 732                 else
 733                         hint = (zoneno + 1) * sbi->secs_per_zone;
 734                 init = false;
 735                 goto find_other_zone;
 736         }
 737 got_it:
 738         /* set it as dirty segment in free segmap */
 739         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
 740         __set_inuse(sbi, segno);
 741         *newseg = segno;
 742         write_unlock(&free_i->segmap_lock);
 743 }
 744
 745 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
 746 {
 747         struct curseg_info *curseg = CURSEG_I(sbi, type);
 748         struct summary_footer *sum_footer;
 749
 750         curseg->segno = curseg->next_segno;
 751         curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
 752         curseg->next_blkoff = 0;
 753         curseg->next_segno = NULL_SEGNO;
 754
 755         sum_footer = &(curseg->sum_blk->footer);
 756         memset(sum_footer, 0, sizeof(struct summary_footer));
 757         if (IS_DATASEG(type))
 758                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
 759         if (IS_NODESEG(type))
 760                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
 761         __set_sit_entry_type(sbi, type, curseg->segno, modified);
 762 }
 763
 764 /*
 765  * Allocate a current working segment.
 766  * This function always allocates a free segment in LFS manner.
 767  */
 768 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
 769 {
 770         struct curseg_info *curseg = CURSEG_I(sbi, type);
 771         unsigned int segno = curseg->segno;
 772         int dir = ALLOC_LEFT;
 773
 774         write_sum_page(sbi, curseg->sum_blk,
 775                                 GET_SUM_BLOCK(sbi, segno));
 776         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
 777                 dir = ALLOC_RIGHT;
 778
 779         if (test_opt(sbi, NOHEAP))
 780                 dir = ALLOC_RIGHT;
 781
 782         get_new_segment(sbi, &segno, new_sec, dir);
 783         curseg->next_segno = segno;
 784         reset_curseg(sbi, type, 1);
 785         curseg->alloc_type = LFS;
 786 }
 787
 788 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
 789                         struct curseg_info *seg, block_t start)
 790 {
 791         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
 792         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
 793         unsigned long target_map[entries];
 794         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
 795         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
 796         int i, pos;
 797
 798         for (i = 0; i < entries; i++)
 799                 target_map[i] = ckpt_map[i] | cur_map[i];
 800
 801         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
 802
 803         seg->next_blkoff = pos;
 804 }
 805
 806 /*
 807  * If a segment is written by LFS manner, next block offset is just obtained
 808  * by increasing the current block offset. However, if a segment is written by
 809  * SSR manner, next block offset obtained by calling __next_free_blkoff
 810  */
 811 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
 812                                 struct curseg_info *seg)
 813 {
 814         if (seg->alloc_type == SSR)
 815                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
 816         else
 817                 seg->next_blkoff++;
 818 }
 819
 820 /*
 821  * This function always allocates a used segment(from dirty seglist) by SSR
 822  * manner, so it should recover the existing segment information of valid blocks
 823  */
 824 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
 825 {
 826         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 827         struct curseg_info *curseg = CURSEG_I(sbi, type);
 828         unsigned int new_segno = curseg->next_segno;
 829         struct f2fs_summary_block *sum_node;
 830         struct page *sum_page;
 831
 832         write_sum_page(sbi, curseg->sum_blk,
 833                                 GET_SUM_BLOCK(sbi, curseg->segno));
 834         __set_test_and_inuse(sbi, new_segno);
 835
 836         mutex_lock(&dirty_i->seglist_lock);
 837         __remove_dirty_segment(sbi, new_segno, PRE);
 838         __remove_dirty_segment(sbi, new_segno, DIRTY);
 839         mutex_unlock(&dirty_i->seglist_lock);
 840
 841         reset_curseg(sbi, type, 1);
 842         curseg->alloc_type = SSR;
 843         __next_free_blkoff(sbi, curseg, 0);
 844
 845         if (reuse) {
 846                 sum_page = get_sum_page(sbi, new_segno);
 847                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
 848                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
 849                 f2fs_put_page(sum_page, 1);
 850         }
 851 }
 852
 853 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
 854 {
 855         struct curseg_info *curseg = CURSEG_I(sbi, type);
 856         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
 857
 858         if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
 859                 return v_ops->get_victim(sbi,
 860                                 &(curseg)->next_segno, BG_GC, type, SSR);
 861
 862         /* For data segments, let's do SSR more intensively */
 863         for (; type >= CURSEG_HOT_DATA; type--)
 864                 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
 865                                                 BG_GC, type, SSR))
 866                         return 1;
 867         return 0;
 868 }
 869
 870 /*
 871  * flush out current segment and replace it with new segment
 872  * This function should be returned with success, otherwise BUG
 873  */
 874 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
 875                                                 int type, bool force)
 876 {
 877         struct curseg_info *curseg = CURSEG_I(sbi, type);
 878
 879         if (force)
 880                 new_curseg(sbi, type, true);
 881         else if (type == CURSEG_WARM_NODE)
 882                 new_curseg(sbi, type, false);
 883         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
 884                 new_curseg(sbi, type, false);
 885         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
 886                 change_curseg(sbi, type, true);
 887         else
 888                 new_curseg(sbi, type, false);
 889
 890         stat_inc_seg_type(sbi, curseg);
 891 }
 892
 893 void allocate_new_segments(struct f2fs_sb_info *sbi)
 894 {
 895         struct curseg_info *curseg;
 896         unsigned int old_curseg;
 897         int i;
 898
 899         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 900                 curseg = CURSEG_I(sbi, i);
 901                 old_curseg = curseg->segno;
 902                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
 903                 locate_dirty_segment(sbi, old_curseg);
 904         }
 905 }
 906
 907 static const struct segment_allocation default_salloc_ops = {
 908         .allocate_segment = allocate_segment_by_default,
 909 };
 910
 911 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
 912 {
 913         struct curseg_info *curseg = CURSEG_I(sbi, type);
 914         if (curseg->next_blkoff < sbi->blocks_per_seg)
 915                 return true;
 916         return false;
 917 }
 918
 919 static int __get_segment_type_2(struct page *page, enum page_type p_type)
 920 {
 921         if (p_type == DATA)
 922                 return CURSEG_HOT_DATA;
 923         else
 924                 return CURSEG_HOT_NODE;
 925 }
 926
 927 static int __get_segment_type_4(struct page *page, enum page_type p_type)
 928 {
 929         if (p_type == DATA) {
 930                 struct inode *inode = page->mapping->host;
 931
 932                 if (S_ISDIR(inode->i_mode))
 933                         return CURSEG_HOT_DATA;
 934                 else
 935                         return CURSEG_COLD_DATA;
 936         } else {
 937                 if (IS_DNODE(page) && !is_cold_node(page))
 938                         return CURSEG_HOT_NODE;
 939                 else
 940                         return CURSEG_COLD_NODE;
 941         }
 942 }
 943
 944 static int __get_segment_type_6(struct page *page, enum page_type p_type)
 945 {
 946         if (p_type == DATA) {
 947                 struct inode *inode = page->mapping->host;
 948
 949                 if (S_ISDIR(inode->i_mode))
 950                         return CURSEG_HOT_DATA;
 951                 else if (is_cold_data(page) || file_is_cold(inode))
 952                         return CURSEG_COLD_DATA;
 953                 else
 954                         return CURSEG_WARM_DATA;
 955         } else {
 956                 if (IS_DNODE(page))
 957                         return is_cold_node(page) ? CURSEG_WARM_NODE :
 958                                                 CURSEG_HOT_NODE;
 959                 else
 960                         return CURSEG_COLD_NODE;
 961         }
 962 }
 963
 964 static int __get_segment_type(struct page *page, enum page_type p_type)
 965 {
 966         switch (F2FS_P_SB(page)->active_logs) {
 967         case 2:
 968                 return __get_segment_type_2(page, p_type);
 969         case 4:
 970                 return __get_segment_type_4(page, p_type);
 971         }
 972         /* NR_CURSEG_TYPE(6) logs by default */
 973         f2fs_bug_on(F2FS_P_SB(page),
 974                 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
 975         return __get_segment_type_6(page, p_type);
 976 }
 977
 978 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
 979                 block_t old_blkaddr, block_t *new_blkaddr,
 980                 struct f2fs_summary *sum, int type)
 981 {
 982         struct sit_info *sit_i = SIT_I(sbi);
 983         struct curseg_info *curseg;
 984
 985         curseg = CURSEG_I(sbi, type);
 986
 987         mutex_lock(&curseg->curseg_mutex);
 988
 989         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 990
 991         /*
 992          * __add_sum_entry should be resided under the curseg_mutex
 993          * because, this function updates a summary entry in the
 994          * current summary block.
 995          */
 996         __add_sum_entry(sbi, type, sum);
 997
 998         mutex_lock(&sit_i->sentry_lock);
 999         __refresh_next_blkoff(sbi, curseg);
1000
1001         stat_inc_block_count(sbi, curseg);
1002
1003         if (!__has_curseg_space(sbi, type))
1004                 sit_i->s_ops->allocate_segment(sbi, type, false);
1005         /*
1006          * SIT information should be updated before segment allocation,
1007          * since SSR needs latest valid block information.
1008          */
1009         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1010
1011         mutex_unlock(&sit_i->sentry_lock);
1012
1013         if (page && IS_NODESEG(type))
1014                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1015
1016         mutex_unlock(&curseg->curseg_mutex);
1017 }
1018
1019 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
1020                         block_t old_blkaddr, block_t *new_blkaddr,
1021                         struct f2fs_summary *sum, struct f2fs_io_info *fio)
1022 {
1023         int type = __get_segment_type(page, fio->type);
1024
1025         allocate_data_block(sbi, page, old_blkaddr, new_blkaddr, sum, type);
1026
1027         /* writeout dirty page into bdev */
1028         f2fs_submit_page_mbio(sbi, page, *new_blkaddr, fio);
1029 }
1030
1031 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1032 {
1033         struct f2fs_io_info fio = {
1034                 .type = META,
1035                 .rw = WRITE_SYNC | REQ_META | REQ_PRIO
1036         };
1037
1038         set_page_writeback(page);
1039         f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1040 }
1041
1042 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
1043                 struct f2fs_io_info *fio,
1044                 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
1045 {
1046         struct f2fs_summary sum;
1047         set_summary(&sum, nid, 0, 0);
1048         do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, fio);
1049 }
1050
1051 void write_data_page(struct page *page, struct dnode_of_data *dn,
1052                 block_t *new_blkaddr, struct f2fs_io_info *fio)
1053 {
1054         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1055         struct f2fs_summary sum;
1056         struct node_info ni;
1057
1058         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1059         get_node_info(sbi, dn->nid, &ni);
1060         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1061
1062         do_write_page(sbi, page, dn->data_blkaddr, new_blkaddr, &sum, fio);
1063 }
1064
1065 void rewrite_data_page(struct page *page, block_t old_blkaddr,
1066                                         struct f2fs_io_info *fio)
1067 {
1068         f2fs_submit_page_mbio(F2FS_P_SB(page), page, old_blkaddr, fio);
1069 }
1070
1071 void recover_data_page(struct f2fs_sb_info *sbi,
1072                         struct page *page, struct f2fs_summary *sum,
1073                         block_t old_blkaddr, block_t new_blkaddr)
1074 {
1075         struct sit_info *sit_i = SIT_I(sbi);
1076         struct curseg_info *curseg;
1077         unsigned int segno, old_cursegno;
1078         struct seg_entry *se;
1079         int type;
1080
1081         segno = GET_SEGNO(sbi, new_blkaddr);
1082         se = get_seg_entry(sbi, segno);
1083         type = se->type;
1084
1085         if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1086                 if (old_blkaddr == NULL_ADDR)
1087                         type = CURSEG_COLD_DATA;
1088                 else
1089                         type = CURSEG_WARM_DATA;
1090         }
1091         curseg = CURSEG_I(sbi, type);
1092
1093         mutex_lock(&curseg->curseg_mutex);
1094         mutex_lock(&sit_i->sentry_lock);
1095
1096         old_cursegno = curseg->segno;
1097
1098         /* change the current segment */
1099         if (segno != curseg->segno) {
1100                 curseg->next_segno = segno;
1101                 change_curseg(sbi, type, true);
1102         }
1103
1104         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1105         __add_sum_entry(sbi, type, sum);
1106
1107         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1108         locate_dirty_segment(sbi, old_cursegno);
1109
1110         mutex_unlock(&sit_i->sentry_lock);
1111         mutex_unlock(&curseg->curseg_mutex);
1112 }
1113
1114 static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1115                                         struct page *page, enum page_type type)
1116 {
1117         enum page_type btype = PAGE_TYPE_OF_BIO(type);
1118         struct f2fs_bio_info *io = &sbi->write_io[btype];
1119         struct bio_vec *bvec;
1120         int i;
1121
1122         down_read(&io->io_rwsem);
1123         if (!io->bio)
1124                 goto out;
1125
1126         bio_for_each_segment_all(bvec, io->bio, i) {
1127                 if (page == bvec->bv_page) {
1128                         up_read(&io->io_rwsem);
1129                         return true;
1130                 }
1131         }
1132
1133 out:
1134         up_read(&io->io_rwsem);
1135         return false;
1136 }
1137
1138 void f2fs_wait_on_page_writeback(struct page *page,
1139                                 enum page_type type)
1140 {
1141         if (PageWriteback(page)) {
1142                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1143
1144                 if (is_merged_page(sbi, page, type))
1145                         f2fs_submit_merged_bio(sbi, type, WRITE);
1146                 wait_on_page_writeback(page);
1147         }
1148 }
1149
1150 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1151 {
1152         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1153         struct curseg_info *seg_i;
1154         unsigned char *kaddr;
1155         struct page *page;
1156         block_t start;
1157         int i, j, offset;
1158
1159         start = start_sum_block(sbi);
1160
1161         page = get_meta_page(sbi, start++);
1162         kaddr = (unsigned char *)page_address(page);
1163
1164         /* Step 1: restore nat cache */
1165         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1166         memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1167
1168         /* Step 2: restore sit cache */
1169         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1170         memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1171                                                 SUM_JOURNAL_SIZE);
1172         offset = 2 * SUM_JOURNAL_SIZE;
1173
1174         /* Step 3: restore summary entries */
1175         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1176                 unsigned short blk_off;
1177                 unsigned int segno;
1178
1179                 seg_i = CURSEG_I(sbi, i);
1180                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1181                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1182                 seg_i->next_segno = segno;
1183                 reset_curseg(sbi, i, 0);
1184                 seg_i->alloc_type = ckpt->alloc_type[i];
1185                 seg_i->next_blkoff = blk_off;
1186
1187                 if (seg_i->alloc_type == SSR)
1188                         blk_off = sbi->blocks_per_seg;
1189
1190                 for (j = 0; j < blk_off; j++) {
1191                         struct f2fs_summary *s;
1192                         s = (struct f2fs_summary *)(kaddr + offset);
1193                         seg_i->sum_blk->entries[j] = *s;
1194                         offset += SUMMARY_SIZE;
1195                         if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1196                                                 SUM_FOOTER_SIZE)
1197                                 continue;
1198
1199                         f2fs_put_page(page, 1);
1200                         page = NULL;
1201
1202                         page = get_meta_page(sbi, start++);
1203                         kaddr = (unsigned char *)page_address(page);
1204                         offset = 0;
1205                 }
1206         }
1207         f2fs_put_page(page, 1);
1208         return 0;
1209 }
1210
1211 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1212 {
1213         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1214         struct f2fs_summary_block *sum;
1215         struct curseg_info *curseg;
1216         struct page *new;
1217         unsigned short blk_off;
1218         unsigned int segno = 0;
1219         block_t blk_addr = 0;
1220
1221         /* get segment number and block addr */
1222         if (IS_DATASEG(type)) {
1223                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1224                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1225                                                         CURSEG_HOT_DATA]);
1226                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1227                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1228                 else
1229                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1230         } else {
1231                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1232                                                         CURSEG_HOT_NODE]);
1233                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1234                                                         CURSEG_HOT_NODE]);
1235                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1236                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1237                                                         type - CURSEG_HOT_NODE);
1238                 else
1239                         blk_addr = GET_SUM_BLOCK(sbi, segno);
1240         }
1241
1242         new = get_meta_page(sbi, blk_addr);
1243         sum = (struct f2fs_summary_block *)page_address(new);
1244
1245         if (IS_NODESEG(type)) {
1246                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1247                         struct f2fs_summary *ns = &sum->entries[0];
1248                         int i;
1249                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1250                                 ns->version = 0;
1251                                 ns->ofs_in_node = 0;
1252                         }
1253                 } else {
1254                         int err;
1255
1256                         err = restore_node_summary(sbi, segno, sum);
1257                         if (err) {
1258                                 f2fs_put_page(new, 1);
1259                                 return err;
1260                         }
1261                 }
1262         }
1263
1264         /* set uncompleted segment to curseg */
1265         curseg = CURSEG_I(sbi, type);
1266         mutex_lock(&curseg->curseg_mutex);
1267         memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1268         curseg->next_segno = segno;
1269         reset_curseg(sbi, type, 0);
1270         curseg->alloc_type = ckpt->alloc_type[type];
1271         curseg->next_blkoff = blk_off;
1272         mutex_unlock(&curseg->curseg_mutex);
1273         f2fs_put_page(new, 1);
1274         return 0;
1275 }
1276
1277 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1278 {
1279         int type = CURSEG_HOT_DATA;
1280         int err;
1281
1282         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1283                 /* restore for compacted data summary */
1284                 if (read_compacted_summaries(sbi))
1285                         return -EINVAL;
1286                 type = CURSEG_HOT_NODE;
1287         }
1288
1289         for (; type <= CURSEG_COLD_NODE; type++) {
1290                 err = read_normal_summaries(sbi, type);
1291                 if (err)
1292                         return err;
1293         }
1294
1295         return 0;
1296 }
1297
1298 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1299 {
1300         struct page *page;
1301         unsigned char *kaddr;
1302         struct f2fs_summary *summary;
1303         struct curseg_info *seg_i;
1304         int written_size = 0;
1305         int i, j;
1306
1307         page = grab_meta_page(sbi, blkaddr++);
1308         kaddr = (unsigned char *)page_address(page);
1309
1310         /* Step 1: write nat cache */
1311         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1312         memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1313         written_size += SUM_JOURNAL_SIZE;
1314
1315         /* Step 2: write sit cache */
1316         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1317         memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1318                                                 SUM_JOURNAL_SIZE);
1319         written_size += SUM_JOURNAL_SIZE;
1320
1321         /* Step 3: write summary entries */
1322         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1323                 unsigned short blkoff;
1324                 seg_i = CURSEG_I(sbi, i);
1325                 if (sbi->ckpt->alloc_type[i] == SSR)
1326                         blkoff = sbi->blocks_per_seg;
1327                 else
1328                         blkoff = curseg_blkoff(sbi, i);
1329
1330                 for (j = 0; j < blkoff; j++) {
1331                         if (!page) {
1332                                 page = grab_meta_page(sbi, blkaddr++);
1333                                 kaddr = (unsigned char *)page_address(page);
1334                                 written_size = 0;
1335                         }
1336                         summary = (struct f2fs_summary *)(kaddr + written_size);
1337                         *summary = seg_i->sum_blk->entries[j];
1338                         written_size += SUMMARY_SIZE;
1339
1340                         if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1341                                                         SUM_FOOTER_SIZE)
1342                                 continue;
1343
1344                         set_page_dirty(page);
1345                         f2fs_put_page(page, 1);
1346                         page = NULL;
1347                 }
1348         }
1349         if (page) {
1350                 set_page_dirty(page);
1351                 f2fs_put_page(page, 1);
1352         }
1353 }
1354
1355 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1356                                         block_t blkaddr, int type)
1357 {
1358         int i, end;
1359         if (IS_DATASEG(type))
1360                 end = type + NR_CURSEG_DATA_TYPE;
1361         else
1362                 end = type + NR_CURSEG_NODE_TYPE;
1363
1364         for (i = type; i < end; i++) {
1365                 struct curseg_info *sum = CURSEG_I(sbi, i);
1366                 mutex_lock(&sum->curseg_mutex);
1367                 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1368                 mutex_unlock(&sum->curseg_mutex);
1369         }
1370 }
1371
1372 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1373 {
1374         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1375                 write_compacted_summaries(sbi, start_blk);
1376         else
1377                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1378 }
1379
1380 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1381 {
1382         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1383                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1384 }
1385
1386 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1387                                         unsigned int val, int alloc)
1388 {
1389         int i;
1390
1391         if (type == NAT_JOURNAL) {
1392                 for (i = 0; i < nats_in_cursum(sum); i++) {
1393                         if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1394                                 return i;
1395                 }
1396                 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1397                         return update_nats_in_cursum(sum, 1);
1398         } else if (type == SIT_JOURNAL) {
1399                 for (i = 0; i < sits_in_cursum(sum); i++)
1400                         if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1401                                 return i;
1402                 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1403                         return update_sits_in_cursum(sum, 1);
1404         }
1405         return -1;
1406 }
1407
1408 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1409                                         unsigned int segno)
1410 {
1411         struct sit_info *sit_i = SIT_I(sbi);
1412         unsigned int offset = SIT_BLOCK_OFFSET(segno);
1413         block_t blk_addr = sit_i->sit_base_addr + offset;
1414
1415         check_seg_range(sbi, segno);
1416
1417         /* calculate sit block address */
1418         if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1419                 blk_addr += sit_i->sit_blocks;
1420
1421         return get_meta_page(sbi, blk_addr);
1422 }
1423
1424 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1425                                         unsigned int start)
1426 {
1427         struct sit_info *sit_i = SIT_I(sbi);
1428         struct page *src_page, *dst_page;
1429         pgoff_t src_off, dst_off;
1430         void *src_addr, *dst_addr;
1431
1432         src_off = current_sit_addr(sbi, start);
1433         dst_off = next_sit_addr(sbi, src_off);
1434
1435         /* get current sit block page without lock */
1436         src_page = get_meta_page(sbi, src_off);
1437         dst_page = grab_meta_page(sbi, dst_off);
1438         f2fs_bug_on(sbi, PageDirty(src_page));
1439
1440         src_addr = page_address(src_page);
1441         dst_addr = page_address(dst_page);
1442         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1443
1444         set_page_dirty(dst_page);
1445         f2fs_put_page(src_page, 1);
1446
1447         set_to_next_sit(sit_i, start);
1448
1449         return dst_page;
1450 }
1451
1452 static struct sit_entry_set *grab_sit_entry_set(void)
1453 {
1454         struct sit_entry_set *ses =
1455                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_ATOMIC);
1456
1457         ses->entry_cnt = 0;
1458         INIT_LIST_HEAD(&ses->set_list);
1459         return ses;
1460 }
1461
1462 static void release_sit_entry_set(struct sit_entry_set *ses)
1463 {
1464         list_del(&ses->set_list);
1465         kmem_cache_free(sit_entry_set_slab, ses);
1466 }
1467
1468 static void adjust_sit_entry_set(struct sit_entry_set *ses,
1469                                                 struct list_head *head)
1470 {
1471         struct sit_entry_set *next = ses;
1472
1473         if (list_is_last(&ses->set_list, head))
1474                 return;
1475
1476         list_for_each_entry_continue(next, head, set_list)
1477                 if (ses->entry_cnt <= next->entry_cnt)
1478                         break;
1479
1480         list_move_tail(&ses->set_list, &next->set_list);
1481 }
1482
1483 static void add_sit_entry(unsigned int segno, struct list_head *head)
1484 {
1485         struct sit_entry_set *ses;
1486         unsigned int start_segno = START_SEGNO(segno);
1487
1488         list_for_each_entry(ses, head, set_list) {
1489                 if (ses->start_segno == start_segno) {
1490                         ses->entry_cnt++;
1491                         adjust_sit_entry_set(ses, head);
1492                         return;
1493                 }
1494         }
1495
1496         ses = grab_sit_entry_set();
1497
1498         ses->start_segno = start_segno;
1499         ses->entry_cnt++;
1500         list_add(&ses->set_list, head);
1501 }
1502
1503 static void add_sits_in_set(struct f2fs_sb_info *sbi)
1504 {
1505         struct f2fs_sm_info *sm_info = SM_I(sbi);
1506         struct list_head *set_list = &sm_info->sit_entry_set;
1507         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
1508         unsigned long nsegs = TOTAL_SEGS(sbi);
1509         unsigned int segno;
1510
1511         for_each_set_bit(segno, bitmap, nsegs)
1512                 add_sit_entry(segno, set_list);
1513 }
1514
1515 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
1516 {
1517         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1518         struct f2fs_summary_block *sum = curseg->sum_blk;
1519         int i;
1520
1521         for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1522                 unsigned int segno;
1523                 bool dirtied;
1524
1525                 segno = le32_to_cpu(segno_in_journal(sum, i));
1526                 dirtied = __mark_sit_entry_dirty(sbi, segno);
1527
1528                 if (!dirtied)
1529                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
1530         }
1531         update_sits_in_cursum(sum, -sits_in_cursum(sum));
1532 }
1533
1534 /*
1535  * CP calls this function, which flushes SIT entries including sit_journal,
1536  * and moves prefree segs to free segs.
1537  */
1538 void flush_sit_entries(struct f2fs_sb_info *sbi)
1539 {
1540         struct sit_info *sit_i = SIT_I(sbi);
1541         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1542         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1543         struct f2fs_summary_block *sum = curseg->sum_blk;
1544         struct sit_entry_set *ses, *tmp;
1545         struct list_head *head = &SM_I(sbi)->sit_entry_set;
1546         unsigned long nsegs = TOTAL_SEGS(sbi);
1547         bool to_journal = true;
1548
1549         mutex_lock(&curseg->curseg_mutex);
1550         mutex_lock(&sit_i->sentry_lock);
1551
1552         /*
1553          * add and account sit entries of dirty bitmap in sit entry
1554          * set temporarily
1555          */
1556         add_sits_in_set(sbi);
1557
1558         /*
1559          * if there are no enough space in journal to store dirty sit
1560          * entries, remove all entries from journal and add and account
1561          * them in sit entry set.
1562          */
1563         if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL))
1564                 remove_sits_in_journal(sbi);
1565
1566         if (!sit_i->dirty_sentries)
1567                 goto out;
1568
1569         /*
1570          * there are two steps to flush sit entries:
1571          * #1, flush sit entries to journal in current cold data summary block.
1572          * #2, flush sit entries to sit page.
1573          */
1574         list_for_each_entry_safe(ses, tmp, head, set_list) {
1575                 struct page *page;
1576                 struct f2fs_sit_block *raw_sit = NULL;
1577                 unsigned int start_segno = ses->start_segno;
1578                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
1579                                                                 nsegs);
1580                 unsigned int segno = start_segno;
1581
1582                 if (to_journal &&
1583                         !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL))
1584                         to_journal = false;
1585
1586                 if (!to_journal) {
1587                         page = get_next_sit_page(sbi, start_segno);
1588                         raw_sit = page_address(page);
1589                 }
1590
1591                 /* flush dirty sit entries in region of current sit set */
1592                 for_each_set_bit_from(segno, bitmap, end) {
1593                         int offset, sit_offset;
1594                         struct seg_entry *se = get_seg_entry(sbi, segno);
1595
1596                         /* add discard candidates */
1597                         if (SM_I(sbi)->nr_discards < SM_I(sbi)->max_discards)
1598                                 add_discard_addrs(sbi, segno, se);
1599
1600                         if (to_journal) {
1601                                 offset = lookup_journal_in_cursum(sum,
1602                                                         SIT_JOURNAL, segno, 1);
1603                                 f2fs_bug_on(sbi, offset < 0);
1604                                 segno_in_journal(sum, offset) =
1605                                                         cpu_to_le32(segno);
1606                                 seg_info_to_raw_sit(se,
1607                                                 &sit_in_journal(sum, offset));
1608                         } else {
1609                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1610                                 seg_info_to_raw_sit(se,
1611                                                 &raw_sit->entries[sit_offset]);
1612                         }
1613
1614                         __clear_bit(segno, bitmap);
1615                         sit_i->dirty_sentries--;
1616                         ses->entry_cnt--;
1617                 }
1618
1619                 if (!to_journal)
1620                         f2fs_put_page(page, 1);
1621
1622                 f2fs_bug_on(sbi, ses->entry_cnt);
1623                 release_sit_entry_set(ses);
1624         }
1625
1626         f2fs_bug_on(sbi, !list_empty(head));
1627         f2fs_bug_on(sbi, sit_i->dirty_sentries);
1628
1629 out:
1630         mutex_unlock(&sit_i->sentry_lock);
1631         mutex_unlock(&curseg->curseg_mutex);
1632
1633         set_prefree_as_free_segments(sbi);
1634 }
1635
1636 static int build_sit_info(struct f2fs_sb_info *sbi)
1637 {
1638         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1639         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1640         struct sit_info *sit_i;
1641         unsigned int sit_segs, start;
1642         char *src_bitmap, *dst_bitmap;
1643         unsigned int bitmap_size;
1644
1645         /* allocate memory for SIT information */
1646         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1647         if (!sit_i)
1648                 return -ENOMEM;
1649
1650         SM_I(sbi)->sit_info = sit_i;
1651
1652         sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1653         if (!sit_i->sentries)
1654                 return -ENOMEM;
1655
1656         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1657         sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1658         if (!sit_i->dirty_sentries_bitmap)
1659                 return -ENOMEM;
1660
1661         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1662                 sit_i->sentries[start].cur_valid_map
1663                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1664                 sit_i->sentries[start].ckpt_valid_map
1665                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1666                 if (!sit_i->sentries[start].cur_valid_map
1667                                 || !sit_i->sentries[start].ckpt_valid_map)
1668                         return -ENOMEM;
1669         }
1670
1671         if (sbi->segs_per_sec > 1) {
1672                 sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) *
1673                                         sizeof(struct sec_entry));
1674                 if (!sit_i->sec_entries)
1675                         return -ENOMEM;
1676         }
1677
1678         /* get information related with SIT */
1679         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1680
1681         /* setup SIT bitmap from ckeckpoint pack */
1682         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1683         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1684
1685         dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1686         if (!dst_bitmap)
1687                 return -ENOMEM;
1688
1689         /* init SIT information */
1690         sit_i->s_ops = &default_salloc_ops;
1691
1692         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1693         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1694         sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1695         sit_i->sit_bitmap = dst_bitmap;
1696         sit_i->bitmap_size = bitmap_size;
1697         sit_i->dirty_sentries = 0;
1698         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1699         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1700         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1701         mutex_init(&sit_i->sentry_lock);
1702         return 0;
1703 }
1704
1705 static int build_free_segmap(struct f2fs_sb_info *sbi)
1706 {
1707         struct f2fs_sm_info *sm_info = SM_I(sbi);
1708         struct free_segmap_info *free_i;
1709         unsigned int bitmap_size, sec_bitmap_size;
1710
1711         /* allocate memory for free segmap information */
1712         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1713         if (!free_i)
1714                 return -ENOMEM;
1715
1716         SM_I(sbi)->free_info = free_i;
1717
1718         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1719         free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1720         if (!free_i->free_segmap)
1721                 return -ENOMEM;
1722
1723         sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1724         free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1725         if (!free_i->free_secmap)
1726                 return -ENOMEM;
1727
1728         /* set all segments as dirty temporarily */
1729         memset(free_i->free_segmap, 0xff, bitmap_size);
1730         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1731
1732         /* init free segmap information */
1733         free_i->start_segno =
1734                 (unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr);
1735         free_i->free_segments = 0;
1736         free_i->free_sections = 0;
1737         rwlock_init(&free_i->segmap_lock);
1738         return 0;
1739 }
1740
1741 static int build_curseg(struct f2fs_sb_info *sbi)
1742 {
1743         struct curseg_info *array;
1744         int i;
1745
1746         array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
1747         if (!array)
1748                 return -ENOMEM;
1749
1750         SM_I(sbi)->curseg_array = array;
1751
1752         for (i = 0; i < NR_CURSEG_TYPE; i++) {
1753                 mutex_init(&array[i].curseg_mutex);
1754                 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1755                 if (!array[i].sum_blk)
1756                         return -ENOMEM;
1757                 array[i].segno = NULL_SEGNO;
1758                 array[i].next_blkoff = 0;
1759         }
1760         return restore_curseg_summaries(sbi);
1761 }
1762
1763 static void build_sit_entries(struct f2fs_sb_info *sbi)
1764 {
1765         struct sit_info *sit_i = SIT_I(sbi);
1766         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1767         struct f2fs_summary_block *sum = curseg->sum_blk;
1768         int sit_blk_cnt = SIT_BLK_CNT(sbi);
1769         unsigned int i, start, end;
1770         unsigned int readed, start_blk = 0;
1771         int nrpages = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1772
1773         do {
1774                 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT);
1775
1776                 start = start_blk * sit_i->sents_per_block;
1777                 end = (start_blk + readed) * sit_i->sents_per_block;
1778
1779                 for (; start < end && start < TOTAL_SEGS(sbi); start++) {
1780                         struct seg_entry *se = &sit_i->sentries[start];
1781                         struct f2fs_sit_block *sit_blk;
1782                         struct f2fs_sit_entry sit;
1783                         struct page *page;
1784
1785                         mutex_lock(&curseg->curseg_mutex);
1786                         for (i = 0; i < sits_in_cursum(sum); i++) {
1787                                 if (le32_to_cpu(segno_in_journal(sum, i))
1788                                                                 == start) {
1789                                         sit = sit_in_journal(sum, i);
1790                                         mutex_unlock(&curseg->curseg_mutex);
1791                                         goto got_it;
1792                                 }
1793                         }
1794                         mutex_unlock(&curseg->curseg_mutex);
1795
1796                         page = get_current_sit_page(sbi, start);
1797                         sit_blk = (struct f2fs_sit_block *)page_address(page);
1798                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1799                         f2fs_put_page(page, 1);
1800 got_it:
1801                         check_block_count(sbi, start, &sit);
1802                         seg_info_from_raw_sit(se, &sit);
1803                         if (sbi->segs_per_sec > 1) {
1804                                 struct sec_entry *e = get_sec_entry(sbi, start);
1805                                 e->valid_blocks += se->valid_blocks;
1806                         }
1807                 }
1808                 start_blk += readed;
1809         } while (start_blk < sit_blk_cnt);
1810 }
1811
1812 static void init_free_segmap(struct f2fs_sb_info *sbi)
1813 {
1814         unsigned int start;
1815         int type;
1816
1817         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1818                 struct seg_entry *sentry = get_seg_entry(sbi, start);
1819                 if (!sentry->valid_blocks)
1820                         __set_free(sbi, start);
1821         }
1822
1823         /* set use the current segments */
1824         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1825                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1826                 __set_test_and_inuse(sbi, curseg_t->segno);
1827         }
1828 }
1829
1830 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1831 {
1832         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1833         struct free_segmap_info *free_i = FREE_I(sbi);
1834         unsigned int segno = 0, offset = 0, total_segs = TOTAL_SEGS(sbi);
1835         unsigned short valid_blocks;
1836
1837         while (1) {
1838                 /* find dirty segment based on free segmap */
1839                 segno = find_next_inuse(free_i, total_segs, offset);
1840                 if (segno >= total_segs)
1841                         break;
1842                 offset = segno + 1;
1843                 valid_blocks = get_valid_blocks(sbi, segno, 0);
1844                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
1845                         continue;
1846                 if (valid_blocks > sbi->blocks_per_seg) {
1847                         f2fs_bug_on(sbi, 1);
1848                         continue;
1849                 }
1850                 mutex_lock(&dirty_i->seglist_lock);
1851                 __locate_dirty_segment(sbi, segno, DIRTY);
1852                 mutex_unlock(&dirty_i->seglist_lock);
1853         }
1854 }
1855
1856 static int init_victim_secmap(struct f2fs_sb_info *sbi)
1857 {
1858         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1859         unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1860
1861         dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
1862         if (!dirty_i->victim_secmap)
1863                 return -ENOMEM;
1864         return 0;
1865 }
1866
1867 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1868 {
1869         struct dirty_seglist_info *dirty_i;
1870         unsigned int bitmap_size, i;
1871
1872         /* allocate memory for dirty segments list information */
1873         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1874         if (!dirty_i)
1875                 return -ENOMEM;
1876
1877         SM_I(sbi)->dirty_info = dirty_i;
1878         mutex_init(&dirty_i->seglist_lock);
1879
1880         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1881
1882         for (i = 0; i < NR_DIRTY_TYPE; i++) {
1883                 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
1884                 if (!dirty_i->dirty_segmap[i])
1885                         return -ENOMEM;
1886         }
1887
1888         init_dirty_segmap(sbi);
1889         return init_victim_secmap(sbi);
1890 }
1891
1892 /*
1893  * Update min, max modified time for cost-benefit GC algorithm
1894  */
1895 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1896 {
1897         struct sit_info *sit_i = SIT_I(sbi);
1898         unsigned int segno;
1899
1900         mutex_lock(&sit_i->sentry_lock);
1901
1902         sit_i->min_mtime = LLONG_MAX;
1903
1904         for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1905                 unsigned int i;
1906                 unsigned long long mtime = 0;
1907
1908                 for (i = 0; i < sbi->segs_per_sec; i++)
1909                         mtime += get_seg_entry(sbi, segno + i)->mtime;
1910
1911                 mtime = div_u64(mtime, sbi->segs_per_sec);
1912
1913                 if (sit_i->min_mtime > mtime)
1914                         sit_i->min_mtime = mtime;
1915         }
1916         sit_i->max_mtime = get_mtime(sbi);
1917         mutex_unlock(&sit_i->sentry_lock);
1918 }
1919
1920 int build_segment_manager(struct f2fs_sb_info *sbi)
1921 {
1922         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1923         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1924         struct f2fs_sm_info *sm_info;
1925         int err;
1926
1927         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1928         if (!sm_info)
1929                 return -ENOMEM;
1930
1931         /* init sm info */
1932         sbi->sm_info = sm_info;
1933         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1934         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1935         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1936         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1937         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1938         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1939         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
1940         sm_info->rec_prefree_segments = sm_info->main_segments *
1941                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
1942         sm_info->ipu_policy = F2FS_IPU_DISABLE;
1943         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
1944
1945         INIT_LIST_HEAD(&sm_info->discard_list);
1946         sm_info->nr_discards = 0;
1947         sm_info->max_discards = 0;
1948
1949         INIT_LIST_HEAD(&sm_info->sit_entry_set);
1950
1951         if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
1952                 err = create_flush_cmd_control(sbi);
1953                 if (err)
1954                         return err;
1955         }
1956
1957         err = build_sit_info(sbi);
1958         if (err)
1959                 return err;
1960         err = build_free_segmap(sbi);
1961         if (err)
1962                 return err;
1963         err = build_curseg(sbi);
1964         if (err)
1965                 return err;
1966
1967         /* reinit free segmap based on SIT */
1968         build_sit_entries(sbi);
1969
1970         init_free_segmap(sbi);
1971         err = build_dirty_segmap(sbi);
1972         if (err)
1973                 return err;
1974
1975         init_min_max_mtime(sbi);
1976         return 0;
1977 }
1978
1979 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
1980                 enum dirty_type dirty_type)
1981 {
1982         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1983
1984         mutex_lock(&dirty_i->seglist_lock);
1985         kfree(dirty_i->dirty_segmap[dirty_type]);
1986         dirty_i->nr_dirty[dirty_type] = 0;
1987         mutex_unlock(&dirty_i->seglist_lock);
1988 }
1989
1990 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
1991 {
1992         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1993         kfree(dirty_i->victim_secmap);
1994 }
1995
1996 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
1997 {
1998         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1999         int i;
2000
2001         if (!dirty_i)
2002                 return;
2003
2004         /* discard pre-free/dirty segments list */
2005         for (i = 0; i < NR_DIRTY_TYPE; i++)
2006                 discard_dirty_segmap(sbi, i);
2007
2008         destroy_victim_secmap(sbi);
2009         SM_I(sbi)->dirty_info = NULL;
2010         kfree(dirty_i);
2011 }
2012
2013 static void destroy_curseg(struct f2fs_sb_info *sbi)
2014 {
2015         struct curseg_info *array = SM_I(sbi)->curseg_array;
2016         int i;
2017
2018         if (!array)
2019                 return;
2020         SM_I(sbi)->curseg_array = NULL;
2021         for (i = 0; i < NR_CURSEG_TYPE; i++)
2022                 kfree(array[i].sum_blk);
2023         kfree(array);
2024 }
2025
2026 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2027 {
2028         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2029         if (!free_i)
2030                 return;
2031         SM_I(sbi)->free_info = NULL;
2032         kfree(free_i->free_segmap);
2033         kfree(free_i->free_secmap);
2034         kfree(free_i);
2035 }
2036
2037 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2038 {
2039         struct sit_info *sit_i = SIT_I(sbi);
2040         unsigned int start;
2041
2042         if (!sit_i)
2043                 return;
2044
2045         if (sit_i->sentries) {
2046                 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
2047                         kfree(sit_i->sentries[start].cur_valid_map);
2048                         kfree(sit_i->sentries[start].ckpt_valid_map);
2049                 }
2050         }
2051         vfree(sit_i->sentries);
2052         vfree(sit_i->sec_entries);
2053         kfree(sit_i->dirty_sentries_bitmap);
2054
2055         SM_I(sbi)->sit_info = NULL;
2056         kfree(sit_i->sit_bitmap);
2057         kfree(sit_i);
2058 }
2059
2060 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2061 {
2062         struct f2fs_sm_info *sm_info = SM_I(sbi);
2063
2064         if (!sm_info)
2065                 return;
2066         destroy_flush_cmd_control(sbi);
2067         destroy_dirty_segmap(sbi);
2068         destroy_curseg(sbi);
2069         destroy_free_segmap(sbi);
2070         destroy_sit_info(sbi);
2071         sbi->sm_info = NULL;
2072         kfree(sm_info);
2073 }
2074
2075 int __init create_segment_manager_caches(void)
2076 {
2077         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2078                         sizeof(struct discard_entry));
2079         if (!discard_entry_slab)
2080                 goto fail;
2081
2082         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2083                         sizeof(struct nat_entry_set));
2084         if (!sit_entry_set_slab)
2085                 goto destory_discard_entry;
2086         return 0;
2087
2088 destory_discard_entry:
2089         kmem_cache_destroy(discard_entry_slab);
2090 fail:
2091         return -ENOMEM;
2092 }
2093
2094 void destroy_segment_manager_caches(void)
2095 {
2096         kmem_cache_destroy(sit_entry_set_slab);
2097         kmem_cache_destroy(discard_entry_slab);
2098 }