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 static struct kmem_cache *inmem_entry_slab;
  30
  31 /*
  32  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  33  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  34  */
  35 static inline unsigned long __reverse_ffs(unsigned long word)
  36 {
  37         int num = 0;
  38
  39 #if BITS_PER_LONG == 64
  40         if ((word & 0xffffffff) == 0) {
  41                 num += 32;
  42                 word >>= 32;
  43         }
  44 #endif
  45         if ((word & 0xffff) == 0) {
  46                 num += 16;
  47                 word >>= 16;
  48         }
  49         if ((word & 0xff) == 0) {
  50                 num += 8;
  51                 word >>= 8;
  52         }
  53         if ((word & 0xf0) == 0)
  54                 num += 4;
  55         else
  56                 word >>= 4;
  57         if ((word & 0xc) == 0)
  58                 num += 2;
  59         else
  60                 word >>= 2;
  61         if ((word & 0x2) == 0)
  62                 num += 1;
  63         return num;
  64 }
  65
  66 /*
  67  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  68  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  69  * Example:
  70  *                             LSB <--> MSB
  71  *   f2fs_set_bit(0, bitmap) => 0000 0001
  72  *   f2fs_set_bit(7, bitmap) => 1000 0000
  73  */
  74 static unsigned long __find_rev_next_bit(const unsigned long *addr,
  75                         unsigned long size, unsigned long offset)
  76 {
  77         const unsigned long *p = addr + BIT_WORD(offset);
  78         unsigned long result = offset & ~(BITS_PER_LONG - 1);
  79         unsigned long tmp;
  80         unsigned long mask, submask;
  81         unsigned long quot, rest;
  82
  83         if (offset >= size)
  84                 return size;
  85
  86         size -= result;
  87         offset %= BITS_PER_LONG;
  88         if (!offset)
  89                 goto aligned;
  90
  91         tmp = *(p++);
  92         quot = (offset >> 3) << 3;
  93         rest = offset & 0x7;
  94         mask = ~0UL << quot;
  95         submask = (unsigned char)(0xff << rest) >> rest;
  96         submask <<= quot;
  97         mask &= submask;
  98         tmp &= mask;
  99         if (size < BITS_PER_LONG)
 100                 goto found_first;
 101         if (tmp)
 102                 goto found_middle;
 103
 104         size -= BITS_PER_LONG;
 105         result += BITS_PER_LONG;
 106 aligned:
 107         while (size & ~(BITS_PER_LONG-1)) {
 108                 tmp = *(p++);
 109                 if (tmp)
 110                         goto found_middle;
 111                 result += BITS_PER_LONG;
 112                 size -= BITS_PER_LONG;
 113         }
 114         if (!size)
 115                 return result;
 116         tmp = *p;
 117 found_first:
 118         tmp &= (~0UL >> (BITS_PER_LONG - size));
 119         if (tmp == 0UL)         /* Are any bits set? */
 120                 return result + size;   /* Nope. */
 121 found_middle:
 122         return result + __reverse_ffs(tmp);
 123 }
 124
 125 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 126                         unsigned long size, unsigned long offset)
 127 {
 128         const unsigned long *p = addr + BIT_WORD(offset);
 129         unsigned long result = offset & ~(BITS_PER_LONG - 1);
 130         unsigned long tmp;
 131         unsigned long mask, submask;
 132         unsigned long quot, rest;
 133
 134         if (offset >= size)
 135                 return size;
 136
 137         size -= result;
 138         offset %= BITS_PER_LONG;
 139         if (!offset)
 140                 goto aligned;
 141
 142         tmp = *(p++);
 143         quot = (offset >> 3) << 3;
 144         rest = offset & 0x7;
 145         mask = ~(~0UL << quot);
 146         submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest);
 147         submask <<= quot;
 148         mask += submask;
 149         tmp |= mask;
 150         if (size < BITS_PER_LONG)
 151                 goto found_first;
 152         if (~tmp)
 153                 goto found_middle;
 154
 155         size -= BITS_PER_LONG;
 156         result += BITS_PER_LONG;
 157 aligned:
 158         while (size & ~(BITS_PER_LONG - 1)) {
 159                 tmp = *(p++);
 160                 if (~tmp)
 161                         goto found_middle;
 162                 result += BITS_PER_LONG;
 163                 size -= BITS_PER_LONG;
 164         }
 165         if (!size)
 166                 return result;
 167         tmp = *p;
 168
 169 found_first:
 170         tmp |= ~0UL << size;
 171         if (tmp == ~0UL)        /* Are any bits zero? */
 172                 return result + size;   /* Nope. */
 173 found_middle:
 174         return result + __reverse_ffz(tmp);
 175 }
 176
 177 void register_inmem_page(struct inode *inode, struct page *page)
 178 {
 179         struct f2fs_inode_info *fi = F2FS_I(inode);
 180         struct inmem_pages *new;
 181         int err;
 182 retry:
 183         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 184
 185         /* add atomic page indices to the list */
 186         new->page = page;
 187         INIT_LIST_HEAD(&new->list);
 188
 189         /* increase reference count with clean state */
 190         mutex_lock(&fi->inmem_lock);
 191         err = radix_tree_insert(&fi->inmem_root, page->index, new);
 192         if (err == -EEXIST) {
 193                 mutex_unlock(&fi->inmem_lock);
 194                 kmem_cache_free(inmem_entry_slab, new);
 195                 return;
 196         } else if (err) {
 197                 mutex_unlock(&fi->inmem_lock);
 198                 kmem_cache_free(inmem_entry_slab, new);
 199                 goto retry;
 200         }
 201         get_page(page);
 202         list_add_tail(&new->list, &fi->inmem_pages);
 203         mutex_unlock(&fi->inmem_lock);
 204 }
 205
 206 void invalidate_inmem_page(struct inode *inode, struct page *page)
 207 {
 208         struct f2fs_inode_info *fi = F2FS_I(inode);
 209         struct inmem_pages *cur;
 210
 211         mutex_lock(&fi->inmem_lock);
 212         cur = radix_tree_lookup(&fi->inmem_root, page->index);
 213         if (cur) {
 214                 radix_tree_delete(&fi->inmem_root, cur->page->index);
 215                 f2fs_put_page(cur->page, 0);
 216                 list_del(&cur->list);
 217                 kmem_cache_free(inmem_entry_slab, cur);
 218         }
 219         mutex_unlock(&fi->inmem_lock);
 220 }
 221
 222 void commit_inmem_pages(struct inode *inode, bool abort)
 223 {
 224         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 225         struct f2fs_inode_info *fi = F2FS_I(inode);
 226         struct inmem_pages *cur, *tmp;
 227         bool submit_bio = false;
 228         struct f2fs_io_info fio = {
 229                 .type = DATA,
 230                 .rw = WRITE_SYNC,
 231         };
 232
 233         f2fs_balance_fs(sbi);
 234         f2fs_lock_op(sbi);
 235
 236         mutex_lock(&fi->inmem_lock);
 237         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 238                 lock_page(cur->page);
 239                 if (!abort && cur->page->mapping == inode->i_mapping) {
 240                         f2fs_wait_on_page_writeback(cur->page, DATA);
 241                         if (clear_page_dirty_for_io(cur->page))
 242                                 inode_dec_dirty_pages(inode);
 243                         do_write_data_page(cur->page, &fio);
 244                         submit_bio = true;
 245                 }
 246                 radix_tree_delete(&fi->inmem_root, cur->page->index);
 247                 f2fs_put_page(cur->page, 1);
 248                 list_del(&cur->list);
 249                 kmem_cache_free(inmem_entry_slab, cur);
 250         }
 251         if (submit_bio)
 252                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 253         mutex_unlock(&fi->inmem_lock);
 254
 255         filemap_fdatawait_range(inode->i_mapping, 0, LLONG_MAX);
 256         f2fs_unlock_op(sbi);
 257 }
 258
 259 /*
 260  * This function balances dirty node and dentry pages.
 261  * In addition, it controls garbage collection.
 262  */
 263 void f2fs_balance_fs(struct f2fs_sb_info *sbi)
 264 {
 265         /*
 266          * We should do GC or end up with checkpoint, if there are so many dirty
 267          * dir/node pages without enough free segments.
 268          */
 269         if (has_not_enough_free_secs(sbi, 0)) {
 270                 mutex_lock(&sbi->gc_mutex);
 271                 f2fs_gc(sbi);
 272         }
 273 }
 274
 275 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 276 {
 277         /* check the # of cached NAT entries and prefree segments */
 278         if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) ||
 279                                 excess_prefree_segs(sbi))
 280                 f2fs_sync_fs(sbi->sb, true);
 281 }
 282
 283 static int issue_flush_thread(void *data)
 284 {
 285         struct f2fs_sb_info *sbi = data;
 286         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 287         wait_queue_head_t *q = &fcc->flush_wait_queue;
 288 repeat:
 289         if (kthread_should_stop())
 290                 return 0;
 291
 292         if (!llist_empty(&fcc->issue_list)) {
 293                 struct bio *bio = bio_alloc(GFP_NOIO, 0);
 294                 struct flush_cmd *cmd, *next;
 295                 int ret;
 296
 297                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 298                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 299
 300                 bio->bi_bdev = sbi->sb->s_bdev;
 301                 ret = submit_bio_wait(WRITE_FLUSH, bio);
 302
 303                 llist_for_each_entry_safe(cmd, next,
 304                                           fcc->dispatch_list, llnode) {
 305                         cmd->ret = ret;
 306                         complete(&cmd->wait);
 307                 }
 308                 bio_put(bio);
 309                 fcc->dispatch_list = NULL;
 310         }
 311
 312         wait_event_interruptible(*q,
 313                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 314         goto repeat;
 315 }
 316
 317 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
 318 {
 319         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 320         struct flush_cmd cmd;
 321
 322         trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
 323                                         test_opt(sbi, FLUSH_MERGE));
 324
 325         if (test_opt(sbi, NOBARRIER))
 326                 return 0;
 327
 328         if (!test_opt(sbi, FLUSH_MERGE))
 329                 return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL);
 330
 331         init_completion(&cmd.wait);
 332
 333         llist_add(&cmd.llnode, &fcc->issue_list);
 334
 335         if (!fcc->dispatch_list)
 336                 wake_up(&fcc->flush_wait_queue);
 337
 338         wait_for_completion(&cmd.wait);
 339
 340         return cmd.ret;
 341 }
 342
 343 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
 344 {
 345         dev_t dev = sbi->sb->s_bdev->bd_dev;
 346         struct flush_cmd_control *fcc;
 347         int err = 0;
 348
 349         fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
 350         if (!fcc)
 351                 return -ENOMEM;
 352         init_waitqueue_head(&fcc->flush_wait_queue);
 353         init_llist_head(&fcc->issue_list);
 354         SM_I(sbi)->cmd_control_info = fcc;
 355         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 356                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 357         if (IS_ERR(fcc->f2fs_issue_flush)) {
 358                 err = PTR_ERR(fcc->f2fs_issue_flush);
 359                 kfree(fcc);
 360                 SM_I(sbi)->cmd_control_info = NULL;
 361                 return err;
 362         }
 363
 364         return err;
 365 }
 366
 367 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
 368 {
 369         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 370
 371         if (fcc && fcc->f2fs_issue_flush)
 372                 kthread_stop(fcc->f2fs_issue_flush);
 373         kfree(fcc);
 374         SM_I(sbi)->cmd_control_info = NULL;
 375 }
 376
 377 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 378                 enum dirty_type dirty_type)
 379 {
 380         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 381
 382         /* need not be added */
 383         if (IS_CURSEG(sbi, segno))
 384                 return;
 385
 386         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 387                 dirty_i->nr_dirty[dirty_type]++;
 388
 389         if (dirty_type == DIRTY) {
 390                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 391                 enum dirty_type t = sentry->type;
 392
 393                 if (unlikely(t >= DIRTY)) {
 394                         f2fs_bug_on(sbi, 1);
 395                         return;
 396                 }
 397                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 398                         dirty_i->nr_dirty[t]++;
 399         }
 400 }
 401
 402 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 403                 enum dirty_type dirty_type)
 404 {
 405         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 406
 407         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 408                 dirty_i->nr_dirty[dirty_type]--;
 409
 410         if (dirty_type == DIRTY) {
 411                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 412                 enum dirty_type t = sentry->type;
 413
 414                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 415                         dirty_i->nr_dirty[t]--;
 416
 417                 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
 418                         clear_bit(GET_SECNO(sbi, segno),
 419                                                 dirty_i->victim_secmap);
 420         }
 421 }
 422
 423 /*
 424  * Should not occur error such as -ENOMEM.
 425  * Adding dirty entry into seglist is not critical operation.
 426  * If a given segment is one of current working segments, it won't be added.
 427  */
 428 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 429 {
 430         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 431         unsigned short valid_blocks;
 432
 433         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 434                 return;
 435
 436         mutex_lock(&dirty_i->seglist_lock);
 437
 438         valid_blocks = get_valid_blocks(sbi, segno, 0);
 439
 440         if (valid_blocks == 0) {
 441                 __locate_dirty_segment(sbi, segno, PRE);
 442                 __remove_dirty_segment(sbi, segno, DIRTY);
 443         } else if (valid_blocks < sbi->blocks_per_seg) {
 444                 __locate_dirty_segment(sbi, segno, DIRTY);
 445         } else {
 446                 /* Recovery routine with SSR needs this */
 447                 __remove_dirty_segment(sbi, segno, DIRTY);
 448         }
 449
 450         mutex_unlock(&dirty_i->seglist_lock);
 451 }
 452
 453 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
 454                                 block_t blkstart, block_t blklen)
 455 {
 456         sector_t start = SECTOR_FROM_BLOCK(blkstart);
 457         sector_t len = SECTOR_FROM_BLOCK(blklen);
 458         trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
 459         return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
 460 }
 461
 462 void discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
 463 {
 464         if (f2fs_issue_discard(sbi, blkaddr, 1)) {
 465                 struct page *page = grab_meta_page(sbi, blkaddr);
 466                 /* zero-filled page */
 467                 set_page_dirty(page);
 468                 f2fs_put_page(page, 1);
 469         }
 470 }
 471
 472 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 473 {
 474         struct list_head *head = &SM_I(sbi)->discard_list;
 475         struct discard_entry *new;
 476         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
 477         int max_blocks = sbi->blocks_per_seg;
 478         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
 479         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
 480         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
 481         unsigned long dmap[entries];
 482         unsigned int start = 0, end = -1;
 483         bool force = (cpc->reason == CP_DISCARD);
 484         int i;
 485
 486         if (!force && !test_opt(sbi, DISCARD))
 487                 return;
 488
 489         if (force && !se->valid_blocks) {
 490                 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 491                 /*
 492                  * if this segment is registered in the prefree list, then
 493                  * we should skip adding a discard candidate, and let the
 494                  * checkpoint do that later.
 495                  */
 496                 mutex_lock(&dirty_i->seglist_lock);
 497                 if (test_bit(cpc->trim_start, dirty_i->dirty_segmap[PRE])) {
 498                         mutex_unlock(&dirty_i->seglist_lock);
 499                         cpc->trimmed += sbi->blocks_per_seg;
 500                         return;
 501                 }
 502                 mutex_unlock(&dirty_i->seglist_lock);
 503
 504                 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
 505                 INIT_LIST_HEAD(&new->list);
 506                 new->blkaddr = START_BLOCK(sbi, cpc->trim_start);
 507                 new->len = sbi->blocks_per_seg;
 508                 list_add_tail(&new->list, head);
 509                 SM_I(sbi)->nr_discards += sbi->blocks_per_seg;
 510                 cpc->trimmed += sbi->blocks_per_seg;
 511                 return;
 512         }
 513
 514         /* zero block will be discarded through the prefree list */
 515         if (!se->valid_blocks || se->valid_blocks == max_blocks)
 516                 return;
 517
 518         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
 519         for (i = 0; i < entries; i++)
 520                 dmap[i] = ~(cur_map[i] | ckpt_map[i]);
 521
 522         while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
 523                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
 524                 if (start >= max_blocks)
 525                         break;
 526
 527                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
 528
 529                 if (end - start < cpc->trim_minlen)
 530                         continue;
 531
 532                 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
 533                 INIT_LIST_HEAD(&new->list);
 534                 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
 535                 new->len = end - start;
 536                 cpc->trimmed += end - start;
 537
 538                 list_add_tail(&new->list, head);
 539                 SM_I(sbi)->nr_discards += end - start;
 540         }
 541 }
 542
 543 void release_discard_addrs(struct f2fs_sb_info *sbi)
 544 {
 545         struct list_head *head = &(SM_I(sbi)->discard_list);
 546         struct discard_entry *entry, *this;
 547
 548         /* drop caches */
 549         list_for_each_entry_safe(entry, this, head, list) {
 550                 list_del(&entry->list);
 551                 kmem_cache_free(discard_entry_slab, entry);
 552         }
 553 }
 554
 555 /*
 556  * Should call clear_prefree_segments after checkpoint is done.
 557  */
 558 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
 559 {
 560         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 561         unsigned int segno;
 562
 563         mutex_lock(&dirty_i->seglist_lock);
 564         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
 565                 __set_test_and_free(sbi, segno);
 566         mutex_unlock(&dirty_i->seglist_lock);
 567 }
 568
 569 void clear_prefree_segments(struct f2fs_sb_info *sbi)
 570 {
 571         struct list_head *head = &(SM_I(sbi)->discard_list);
 572         struct discard_entry *entry, *this;
 573         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 574         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
 575         unsigned int start = 0, end = -1;
 576
 577         mutex_lock(&dirty_i->seglist_lock);
 578
 579         while (1) {
 580                 int i;
 581                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
 582                 if (start >= MAIN_SEGS(sbi))
 583                         break;
 584                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
 585                                                                 start + 1);
 586
 587                 for (i = start; i < end; i++)
 588                         clear_bit(i, prefree_map);
 589
 590                 dirty_i->nr_dirty[PRE] -= end - start;
 591
 592                 if (!test_opt(sbi, DISCARD))
 593                         continue;
 594
 595                 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
 596                                 (end - start) << sbi->log_blocks_per_seg);
 597         }
 598         mutex_unlock(&dirty_i->seglist_lock);
 599
 600         /* send small discards */
 601         list_for_each_entry_safe(entry, this, head, list) {
 602                 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
 603                 list_del(&entry->list);
 604                 SM_I(sbi)->nr_discards -= entry->len;
 605                 kmem_cache_free(discard_entry_slab, entry);
 606         }
 607 }
 608
 609 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
 610 {
 611         struct sit_info *sit_i = SIT_I(sbi);
 612
 613         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
 614                 sit_i->dirty_sentries++;
 615                 return false;
 616         }
 617
 618         return true;
 619 }
 620
 621 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
 622                                         unsigned int segno, int modified)
 623 {
 624         struct seg_entry *se = get_seg_entry(sbi, segno);
 625         se->type = type;
 626         if (modified)
 627                 __mark_sit_entry_dirty(sbi, segno);
 628 }
 629
 630 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
 631 {
 632         struct seg_entry *se;
 633         unsigned int segno, offset;
 634         long int new_vblocks;
 635
 636         segno = GET_SEGNO(sbi, blkaddr);
 637
 638         se = get_seg_entry(sbi, segno);
 639         new_vblocks = se->valid_blocks + del;
 640         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 641
 642         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
 643                                 (new_vblocks > sbi->blocks_per_seg)));
 644
 645         se->valid_blocks = new_vblocks;
 646         se->mtime = get_mtime(sbi);
 647         SIT_I(sbi)->max_mtime = se->mtime;
 648
 649         /* Update valid block bitmap */
 650         if (del > 0) {
 651                 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
 652                         f2fs_bug_on(sbi, 1);
 653         } else {
 654                 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
 655                         f2fs_bug_on(sbi, 1);
 656         }
 657         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
 658                 se->ckpt_valid_blocks += del;
 659
 660         __mark_sit_entry_dirty(sbi, segno);
 661
 662         /* update total number of valid blocks to be written in ckpt area */
 663         SIT_I(sbi)->written_valid_blocks += del;
 664
 665         if (sbi->segs_per_sec > 1)
 666                 get_sec_entry(sbi, segno)->valid_blocks += del;
 667 }
 668
 669 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
 670 {
 671         update_sit_entry(sbi, new, 1);
 672         if (GET_SEGNO(sbi, old) != NULL_SEGNO)
 673                 update_sit_entry(sbi, old, -1);
 674
 675         locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
 676         locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
 677 }
 678
 679 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
 680 {
 681         unsigned int segno = GET_SEGNO(sbi, addr);
 682         struct sit_info *sit_i = SIT_I(sbi);
 683
 684         f2fs_bug_on(sbi, addr == NULL_ADDR);
 685         if (addr == NEW_ADDR)
 686                 return;
 687
 688         /* add it into sit main buffer */
 689         mutex_lock(&sit_i->sentry_lock);
 690
 691         update_sit_entry(sbi, addr, -1);
 692
 693         /* add it into dirty seglist */
 694         locate_dirty_segment(sbi, segno);
 695
 696         mutex_unlock(&sit_i->sentry_lock);
 697 }
 698
 699 /*
 700  * This function should be resided under the curseg_mutex lock
 701  */
 702 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
 703                                         struct f2fs_summary *sum)
 704 {
 705         struct curseg_info *curseg = CURSEG_I(sbi, type);
 706         void *addr = curseg->sum_blk;
 707         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
 708         memcpy(addr, sum, sizeof(struct f2fs_summary));
 709 }
 710
 711 /*
 712  * Calculate the number of current summary pages for writing
 713  */
 714 int npages_for_summary_flush(struct f2fs_sb_info *sbi)
 715 {
 716         int valid_sum_count = 0;
 717         int i, sum_in_page;
 718
 719         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 720                 if (sbi->ckpt->alloc_type[i] == SSR)
 721                         valid_sum_count += sbi->blocks_per_seg;
 722                 else
 723                         valid_sum_count += curseg_blkoff(sbi, i);
 724         }
 725
 726         sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
 727                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
 728         if (valid_sum_count <= sum_in_page)
 729                 return 1;
 730         else if ((valid_sum_count - sum_in_page) <=
 731                 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
 732                 return 2;
 733         return 3;
 734 }
 735
 736 /*
 737  * Caller should put this summary page
 738  */
 739 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
 740 {
 741         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
 742 }
 743
 744 static void write_sum_page(struct f2fs_sb_info *sbi,
 745                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
 746 {
 747         struct page *page = grab_meta_page(sbi, blk_addr);
 748         void *kaddr = page_address(page);
 749         memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
 750         set_page_dirty(page);
 751         f2fs_put_page(page, 1);
 752 }
 753
 754 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
 755 {
 756         struct curseg_info *curseg = CURSEG_I(sbi, type);
 757         unsigned int segno = curseg->segno + 1;
 758         struct free_segmap_info *free_i = FREE_I(sbi);
 759
 760         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
 761                 return !test_bit(segno, free_i->free_segmap);
 762         return 0;
 763 }
 764
 765 /*
 766  * Find a new segment from the free segments bitmap to right order
 767  * This function should be returned with success, otherwise BUG
 768  */
 769 static void get_new_segment(struct f2fs_sb_info *sbi,
 770                         unsigned int *newseg, bool new_sec, int dir)
 771 {
 772         struct free_segmap_info *free_i = FREE_I(sbi);
 773         unsigned int segno, secno, zoneno;
 774         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
 775         unsigned int hint = *newseg / sbi->segs_per_sec;
 776         unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
 777         unsigned int left_start = hint;
 778         bool init = true;
 779         int go_left = 0;
 780         int i;
 781
 782         write_lock(&free_i->segmap_lock);
 783
 784         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
 785                 segno = find_next_zero_bit(free_i->free_segmap,
 786                                         MAIN_SEGS(sbi), *newseg + 1);
 787                 if (segno - *newseg < sbi->segs_per_sec -
 788                                         (*newseg % sbi->segs_per_sec))
 789                         goto got_it;
 790         }
 791 find_other_zone:
 792         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
 793         if (secno >= MAIN_SECS(sbi)) {
 794                 if (dir == ALLOC_RIGHT) {
 795                         secno = find_next_zero_bit(free_i->free_secmap,
 796                                                         MAIN_SECS(sbi), 0);
 797                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
 798                 } else {
 799                         go_left = 1;
 800                         left_start = hint - 1;
 801                 }
 802         }
 803         if (go_left == 0)
 804                 goto skip_left;
 805
 806         while (test_bit(left_start, free_i->free_secmap)) {
 807                 if (left_start > 0) {
 808                         left_start--;
 809                         continue;
 810                 }
 811                 left_start = find_next_zero_bit(free_i->free_secmap,
 812                                                         MAIN_SECS(sbi), 0);
 813                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
 814                 break;
 815         }
 816         secno = left_start;
 817 skip_left:
 818         hint = secno;
 819         segno = secno * sbi->segs_per_sec;
 820         zoneno = secno / sbi->secs_per_zone;
 821
 822         /* give up on finding another zone */
 823         if (!init)
 824                 goto got_it;
 825         if (sbi->secs_per_zone == 1)
 826                 goto got_it;
 827         if (zoneno == old_zoneno)
 828                 goto got_it;
 829         if (dir == ALLOC_LEFT) {
 830                 if (!go_left && zoneno + 1 >= total_zones)
 831                         goto got_it;
 832                 if (go_left && zoneno == 0)
 833                         goto got_it;
 834         }
 835         for (i = 0; i < NR_CURSEG_TYPE; i++)
 836                 if (CURSEG_I(sbi, i)->zone == zoneno)
 837                         break;
 838
 839         if (i < NR_CURSEG_TYPE) {
 840                 /* zone is in user, try another */
 841                 if (go_left)
 842                         hint = zoneno * sbi->secs_per_zone - 1;
 843                 else if (zoneno + 1 >= total_zones)
 844                         hint = 0;
 845                 else
 846                         hint = (zoneno + 1) * sbi->secs_per_zone;
 847                 init = false;
 848                 goto find_other_zone;
 849         }
 850 got_it:
 851         /* set it as dirty segment in free segmap */
 852         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
 853         __set_inuse(sbi, segno);
 854         *newseg = segno;
 855         write_unlock(&free_i->segmap_lock);
 856 }
 857
 858 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
 859 {
 860         struct curseg_info *curseg = CURSEG_I(sbi, type);
 861         struct summary_footer *sum_footer;
 862
 863         curseg->segno = curseg->next_segno;
 864         curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
 865         curseg->next_blkoff = 0;
 866         curseg->next_segno = NULL_SEGNO;
 867
 868         sum_footer = &(curseg->sum_blk->footer);
 869         memset(sum_footer, 0, sizeof(struct summary_footer));
 870         if (IS_DATASEG(type))
 871                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
 872         if (IS_NODESEG(type))
 873                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
 874         __set_sit_entry_type(sbi, type, curseg->segno, modified);
 875 }
 876
 877 /*
 878  * Allocate a current working segment.
 879  * This function always allocates a free segment in LFS manner.
 880  */
 881 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
 882 {
 883         struct curseg_info *curseg = CURSEG_I(sbi, type);
 884         unsigned int segno = curseg->segno;
 885         int dir = ALLOC_LEFT;
 886
 887         write_sum_page(sbi, curseg->sum_blk,
 888                                 GET_SUM_BLOCK(sbi, segno));
 889         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
 890                 dir = ALLOC_RIGHT;
 891
 892         if (test_opt(sbi, NOHEAP))
 893                 dir = ALLOC_RIGHT;
 894
 895         get_new_segment(sbi, &segno, new_sec, dir);
 896         curseg->next_segno = segno;
 897         reset_curseg(sbi, type, 1);
 898         curseg->alloc_type = LFS;
 899 }
 900
 901 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
 902                         struct curseg_info *seg, block_t start)
 903 {
 904         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
 905         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
 906         unsigned long target_map[entries];
 907         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
 908         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
 909         int i, pos;
 910
 911         for (i = 0; i < entries; i++)
 912                 target_map[i] = ckpt_map[i] | cur_map[i];
 913
 914         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
 915
 916         seg->next_blkoff = pos;
 917 }
 918
 919 /*
 920  * If a segment is written by LFS manner, next block offset is just obtained
 921  * by increasing the current block offset. However, if a segment is written by
 922  * SSR manner, next block offset obtained by calling __next_free_blkoff
 923  */
 924 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
 925                                 struct curseg_info *seg)
 926 {
 927         if (seg->alloc_type == SSR)
 928                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
 929         else
 930                 seg->next_blkoff++;
 931 }
 932
 933 /*
 934  * This function always allocates a used segment(from dirty seglist) by SSR
 935  * manner, so it should recover the existing segment information of valid blocks
 936  */
 937 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
 938 {
 939         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 940         struct curseg_info *curseg = CURSEG_I(sbi, type);
 941         unsigned int new_segno = curseg->next_segno;
 942         struct f2fs_summary_block *sum_node;
 943         struct page *sum_page;
 944
 945         write_sum_page(sbi, curseg->sum_blk,
 946                                 GET_SUM_BLOCK(sbi, curseg->segno));
 947         __set_test_and_inuse(sbi, new_segno);
 948
 949         mutex_lock(&dirty_i->seglist_lock);
 950         __remove_dirty_segment(sbi, new_segno, PRE);
 951         __remove_dirty_segment(sbi, new_segno, DIRTY);
 952         mutex_unlock(&dirty_i->seglist_lock);
 953
 954         reset_curseg(sbi, type, 1);
 955         curseg->alloc_type = SSR;
 956         __next_free_blkoff(sbi, curseg, 0);
 957
 958         if (reuse) {
 959                 sum_page = get_sum_page(sbi, new_segno);
 960                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
 961                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
 962                 f2fs_put_page(sum_page, 1);
 963         }
 964 }
 965
 966 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
 967 {
 968         struct curseg_info *curseg = CURSEG_I(sbi, type);
 969         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
 970
 971         if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
 972                 return v_ops->get_victim(sbi,
 973                                 &(curseg)->next_segno, BG_GC, type, SSR);
 974
 975         /* For data segments, let's do SSR more intensively */
 976         for (; type >= CURSEG_HOT_DATA; type--)
 977                 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
 978                                                 BG_GC, type, SSR))
 979                         return 1;
 980         return 0;
 981 }
 982
 983 /*
 984  * flush out current segment and replace it with new segment
 985  * This function should be returned with success, otherwise BUG
 986  */
 987 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
 988                                                 int type, bool force)
 989 {
 990         struct curseg_info *curseg = CURSEG_I(sbi, type);
 991
 992         if (force)
 993                 new_curseg(sbi, type, true);
 994         else if (type == CURSEG_WARM_NODE)
 995                 new_curseg(sbi, type, false);
 996         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
 997                 new_curseg(sbi, type, false);
 998         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
 999                 change_curseg(sbi, type, true);
1000         else
1001                 new_curseg(sbi, type, false);
1002
1003         stat_inc_seg_type(sbi, curseg);
1004 }
1005
1006 void allocate_new_segments(struct f2fs_sb_info *sbi)
1007 {
1008         struct curseg_info *curseg;
1009         unsigned int old_curseg;
1010         int i;
1011
1012         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1013                 curseg = CURSEG_I(sbi, i);
1014                 old_curseg = curseg->segno;
1015                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
1016                 locate_dirty_segment(sbi, old_curseg);
1017         }
1018 }
1019
1020 static const struct segment_allocation default_salloc_ops = {
1021         .allocate_segment = allocate_segment_by_default,
1022 };
1023
1024 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1025 {
1026         __u64 start = range->start >> sbi->log_blocksize;
1027         __u64 end = start + (range->len >> sbi->log_blocksize) - 1;
1028         unsigned int start_segno, end_segno;
1029         struct cp_control cpc;
1030
1031         if (range->minlen > SEGMENT_SIZE(sbi) || start >= MAX_BLKADDR(sbi) ||
1032                                                 range->len < sbi->blocksize)
1033                 return -EINVAL;
1034
1035         cpc.trimmed = 0;
1036         if (end <= MAIN_BLKADDR(sbi))
1037                 goto out;
1038
1039         /* start/end segment number in main_area */
1040         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1041         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1042                                                 GET_SEGNO(sbi, end);
1043         cpc.reason = CP_DISCARD;
1044         cpc.trim_start = start_segno;
1045         cpc.trim_end = end_segno;
1046         cpc.trim_minlen = range->minlen >> sbi->log_blocksize;
1047
1048         /* do checkpoint to issue discard commands safely */
1049         mutex_lock(&sbi->gc_mutex);
1050         write_checkpoint(sbi, &cpc);
1051         mutex_unlock(&sbi->gc_mutex);
1052 out:
1053         range->len = cpc.trimmed << sbi->log_blocksize;
1054         return 0;
1055 }
1056
1057 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1058 {
1059         struct curseg_info *curseg = CURSEG_I(sbi, type);
1060         if (curseg->next_blkoff < sbi->blocks_per_seg)
1061                 return true;
1062         return false;
1063 }
1064
1065 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1066 {
1067         if (p_type == DATA)
1068                 return CURSEG_HOT_DATA;
1069         else
1070                 return CURSEG_HOT_NODE;
1071 }
1072
1073 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1074 {
1075         if (p_type == DATA) {
1076                 struct inode *inode = page->mapping->host;
1077
1078                 if (S_ISDIR(inode->i_mode))
1079                         return CURSEG_HOT_DATA;
1080                 else
1081                         return CURSEG_COLD_DATA;
1082         } else {
1083                 if (IS_DNODE(page) && !is_cold_node(page))
1084                         return CURSEG_HOT_NODE;
1085                 else
1086                         return CURSEG_COLD_NODE;
1087         }
1088 }
1089
1090 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1091 {
1092         if (p_type == DATA) {
1093                 struct inode *inode = page->mapping->host;
1094
1095                 if (S_ISDIR(inode->i_mode))
1096                         return CURSEG_HOT_DATA;
1097                 else if (is_cold_data(page) || file_is_cold(inode))
1098                         return CURSEG_COLD_DATA;
1099                 else
1100                         return CURSEG_WARM_DATA;
1101         } else {
1102                 if (IS_DNODE(page))
1103                         return is_cold_node(page) ? CURSEG_WARM_NODE :
1104                                                 CURSEG_HOT_NODE;
1105                 else
1106                         return CURSEG_COLD_NODE;
1107         }
1108 }
1109
1110 static int __get_segment_type(struct page *page, enum page_type p_type)
1111 {
1112         switch (F2FS_P_SB(page)->active_logs) {
1113         case 2:
1114                 return __get_segment_type_2(page, p_type);
1115         case 4:
1116                 return __get_segment_type_4(page, p_type);
1117         }
1118         /* NR_CURSEG_TYPE(6) logs by default */
1119         f2fs_bug_on(F2FS_P_SB(page),
1120                 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1121         return __get_segment_type_6(page, p_type);
1122 }
1123
1124 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1125                 block_t old_blkaddr, block_t *new_blkaddr,
1126                 struct f2fs_summary *sum, int type)
1127 {
1128         struct sit_info *sit_i = SIT_I(sbi);
1129         struct curseg_info *curseg;
1130
1131         curseg = CURSEG_I(sbi, type);
1132
1133         mutex_lock(&curseg->curseg_mutex);
1134
1135         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1136
1137         /*
1138          * __add_sum_entry should be resided under the curseg_mutex
1139          * because, this function updates a summary entry in the
1140          * current summary block.
1141          */
1142         __add_sum_entry(sbi, type, sum);
1143
1144         mutex_lock(&sit_i->sentry_lock);
1145         __refresh_next_blkoff(sbi, curseg);
1146
1147         stat_inc_block_count(sbi, curseg);
1148
1149         if (!__has_curseg_space(sbi, type))
1150                 sit_i->s_ops->allocate_segment(sbi, type, false);
1151         /*
1152          * SIT information should be updated before segment allocation,
1153          * since SSR needs latest valid block information.
1154          */
1155         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1156
1157         mutex_unlock(&sit_i->sentry_lock);
1158
1159         if (page && IS_NODESEG(type))
1160                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1161
1162         mutex_unlock(&curseg->curseg_mutex);
1163 }
1164
1165 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
1166                         block_t old_blkaddr, block_t *new_blkaddr,
1167                         struct f2fs_summary *sum, struct f2fs_io_info *fio)
1168 {
1169         int type = __get_segment_type(page, fio->type);
1170
1171         allocate_data_block(sbi, page, old_blkaddr, new_blkaddr, sum, type);
1172
1173         /* writeout dirty page into bdev */
1174         f2fs_submit_page_mbio(sbi, page, *new_blkaddr, fio);
1175 }
1176
1177 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1178 {
1179         struct f2fs_io_info fio = {
1180                 .type = META,
1181                 .rw = WRITE_SYNC | REQ_META | REQ_PRIO
1182         };
1183
1184         set_page_writeback(page);
1185         f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1186 }
1187
1188 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
1189                 struct f2fs_io_info *fio,
1190                 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
1191 {
1192         struct f2fs_summary sum;
1193         set_summary(&sum, nid, 0, 0);
1194         do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, fio);
1195 }
1196
1197 void write_data_page(struct page *page, struct dnode_of_data *dn,
1198                 block_t *new_blkaddr, struct f2fs_io_info *fio)
1199 {
1200         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1201         struct f2fs_summary sum;
1202         struct node_info ni;
1203
1204         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1205         get_node_info(sbi, dn->nid, &ni);
1206         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1207
1208         do_write_page(sbi, page, dn->data_blkaddr, new_blkaddr, &sum, fio);
1209 }
1210
1211 void rewrite_data_page(struct page *page, block_t old_blkaddr,
1212                                         struct f2fs_io_info *fio)
1213 {
1214         f2fs_submit_page_mbio(F2FS_P_SB(page), page, old_blkaddr, fio);
1215 }
1216
1217 void recover_data_page(struct f2fs_sb_info *sbi,
1218                         struct page *page, struct f2fs_summary *sum,
1219                         block_t old_blkaddr, block_t new_blkaddr)
1220 {
1221         struct sit_info *sit_i = SIT_I(sbi);
1222         struct curseg_info *curseg;
1223         unsigned int segno, old_cursegno;
1224         struct seg_entry *se;
1225         int type;
1226
1227         segno = GET_SEGNO(sbi, new_blkaddr);
1228         se = get_seg_entry(sbi, segno);
1229         type = se->type;
1230
1231         if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1232                 if (old_blkaddr == NULL_ADDR)
1233                         type = CURSEG_COLD_DATA;
1234                 else
1235                         type = CURSEG_WARM_DATA;
1236         }
1237         curseg = CURSEG_I(sbi, type);
1238
1239         mutex_lock(&curseg->curseg_mutex);
1240         mutex_lock(&sit_i->sentry_lock);
1241
1242         old_cursegno = curseg->segno;
1243
1244         /* change the current segment */
1245         if (segno != curseg->segno) {
1246                 curseg->next_segno = segno;
1247                 change_curseg(sbi, type, true);
1248         }
1249
1250         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1251         __add_sum_entry(sbi, type, sum);
1252
1253         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1254         locate_dirty_segment(sbi, old_cursegno);
1255
1256         mutex_unlock(&sit_i->sentry_lock);
1257         mutex_unlock(&curseg->curseg_mutex);
1258 }
1259
1260 static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1261                                         struct page *page, enum page_type type)
1262 {
1263         enum page_type btype = PAGE_TYPE_OF_BIO(type);
1264         struct f2fs_bio_info *io = &sbi->write_io[btype];
1265         struct bio_vec *bvec;
1266         int i;
1267
1268         down_read(&io->io_rwsem);
1269         if (!io->bio)
1270                 goto out;
1271
1272         bio_for_each_segment_all(bvec, io->bio, i) {
1273                 if (page == bvec->bv_page) {
1274                         up_read(&io->io_rwsem);
1275                         return true;
1276                 }
1277         }
1278
1279 out:
1280         up_read(&io->io_rwsem);
1281         return false;
1282 }
1283
1284 void f2fs_wait_on_page_writeback(struct page *page,
1285                                 enum page_type type)
1286 {
1287         if (PageWriteback(page)) {
1288                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1289
1290                 if (is_merged_page(sbi, page, type))
1291                         f2fs_submit_merged_bio(sbi, type, WRITE);
1292                 wait_on_page_writeback(page);
1293         }
1294 }
1295
1296 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1297 {
1298         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1299         struct curseg_info *seg_i;
1300         unsigned char *kaddr;
1301         struct page *page;
1302         block_t start;
1303         int i, j, offset;
1304
1305         start = start_sum_block(sbi);
1306
1307         page = get_meta_page(sbi, start++);
1308         kaddr = (unsigned char *)page_address(page);
1309
1310         /* Step 1: restore nat cache */
1311         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1312         memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1313
1314         /* Step 2: restore sit cache */
1315         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1316         memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1317                                                 SUM_JOURNAL_SIZE);
1318         offset = 2 * SUM_JOURNAL_SIZE;
1319
1320         /* Step 3: restore summary entries */
1321         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1322                 unsigned short blk_off;
1323                 unsigned int segno;
1324
1325                 seg_i = CURSEG_I(sbi, i);
1326                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1327                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1328                 seg_i->next_segno = segno;
1329                 reset_curseg(sbi, i, 0);
1330                 seg_i->alloc_type = ckpt->alloc_type[i];
1331                 seg_i->next_blkoff = blk_off;
1332
1333                 if (seg_i->alloc_type == SSR)
1334                         blk_off = sbi->blocks_per_seg;
1335
1336                 for (j = 0; j < blk_off; j++) {
1337                         struct f2fs_summary *s;
1338                         s = (struct f2fs_summary *)(kaddr + offset);
1339                         seg_i->sum_blk->entries[j] = *s;
1340                         offset += SUMMARY_SIZE;
1341                         if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1342                                                 SUM_FOOTER_SIZE)
1343                                 continue;
1344
1345                         f2fs_put_page(page, 1);
1346                         page = NULL;
1347
1348                         page = get_meta_page(sbi, start++);
1349                         kaddr = (unsigned char *)page_address(page);
1350                         offset = 0;
1351                 }
1352         }
1353         f2fs_put_page(page, 1);
1354         return 0;
1355 }
1356
1357 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1358 {
1359         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1360         struct f2fs_summary_block *sum;
1361         struct curseg_info *curseg;
1362         struct page *new;
1363         unsigned short blk_off;
1364         unsigned int segno = 0;
1365         block_t blk_addr = 0;
1366
1367         /* get segment number and block addr */
1368         if (IS_DATASEG(type)) {
1369                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1370                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1371                                                         CURSEG_HOT_DATA]);
1372                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1373                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1374                 else
1375                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1376         } else {
1377                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1378                                                         CURSEG_HOT_NODE]);
1379                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1380                                                         CURSEG_HOT_NODE]);
1381                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1382                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1383                                                         type - CURSEG_HOT_NODE);
1384                 else
1385                         blk_addr = GET_SUM_BLOCK(sbi, segno);
1386         }
1387
1388         new = get_meta_page(sbi, blk_addr);
1389         sum = (struct f2fs_summary_block *)page_address(new);
1390
1391         if (IS_NODESEG(type)) {
1392                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1393                         struct f2fs_summary *ns = &sum->entries[0];
1394                         int i;
1395                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1396                                 ns->version = 0;
1397                                 ns->ofs_in_node = 0;
1398                         }
1399                 } else {
1400                         int err;
1401
1402                         err = restore_node_summary(sbi, segno, sum);
1403                         if (err) {
1404                                 f2fs_put_page(new, 1);
1405                                 return err;
1406                         }
1407                 }
1408         }
1409
1410         /* set uncompleted segment to curseg */
1411         curseg = CURSEG_I(sbi, type);
1412         mutex_lock(&curseg->curseg_mutex);
1413         memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1414         curseg->next_segno = segno;
1415         reset_curseg(sbi, type, 0);
1416         curseg->alloc_type = ckpt->alloc_type[type];
1417         curseg->next_blkoff = blk_off;
1418         mutex_unlock(&curseg->curseg_mutex);
1419         f2fs_put_page(new, 1);
1420         return 0;
1421 }
1422
1423 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1424 {
1425         int type = CURSEG_HOT_DATA;
1426         int err;
1427
1428         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1429                 /* restore for compacted data summary */
1430                 if (read_compacted_summaries(sbi))
1431                         return -EINVAL;
1432                 type = CURSEG_HOT_NODE;
1433         }
1434
1435         for (; type <= CURSEG_COLD_NODE; type++) {
1436                 err = read_normal_summaries(sbi, type);
1437                 if (err)
1438                         return err;
1439         }
1440
1441         return 0;
1442 }
1443
1444 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1445 {
1446         struct page *page;
1447         unsigned char *kaddr;
1448         struct f2fs_summary *summary;
1449         struct curseg_info *seg_i;
1450         int written_size = 0;
1451         int i, j;
1452
1453         page = grab_meta_page(sbi, blkaddr++);
1454         kaddr = (unsigned char *)page_address(page);
1455
1456         /* Step 1: write nat cache */
1457         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1458         memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1459         written_size += SUM_JOURNAL_SIZE;
1460
1461         /* Step 2: write sit cache */
1462         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1463         memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1464                                                 SUM_JOURNAL_SIZE);
1465         written_size += SUM_JOURNAL_SIZE;
1466
1467         /* Step 3: write summary entries */
1468         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1469                 unsigned short blkoff;
1470                 seg_i = CURSEG_I(sbi, i);
1471                 if (sbi->ckpt->alloc_type[i] == SSR)
1472                         blkoff = sbi->blocks_per_seg;
1473                 else
1474                         blkoff = curseg_blkoff(sbi, i);
1475
1476                 for (j = 0; j < blkoff; j++) {
1477                         if (!page) {
1478                                 page = grab_meta_page(sbi, blkaddr++);
1479                                 kaddr = (unsigned char *)page_address(page);
1480                                 written_size = 0;
1481                         }
1482                         summary = (struct f2fs_summary *)(kaddr + written_size);
1483                         *summary = seg_i->sum_blk->entries[j];
1484                         written_size += SUMMARY_SIZE;
1485
1486                         if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1487                                                         SUM_FOOTER_SIZE)
1488                                 continue;
1489
1490                         set_page_dirty(page);
1491                         f2fs_put_page(page, 1);
1492                         page = NULL;
1493                 }
1494         }
1495         if (page) {
1496                 set_page_dirty(page);
1497                 f2fs_put_page(page, 1);
1498         }
1499 }
1500
1501 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1502                                         block_t blkaddr, int type)
1503 {
1504         int i, end;
1505         if (IS_DATASEG(type))
1506                 end = type + NR_CURSEG_DATA_TYPE;
1507         else
1508                 end = type + NR_CURSEG_NODE_TYPE;
1509
1510         for (i = type; i < end; i++) {
1511                 struct curseg_info *sum = CURSEG_I(sbi, i);
1512                 mutex_lock(&sum->curseg_mutex);
1513                 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1514                 mutex_unlock(&sum->curseg_mutex);
1515         }
1516 }
1517
1518 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1519 {
1520         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1521                 write_compacted_summaries(sbi, start_blk);
1522         else
1523                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1524 }
1525
1526 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1527 {
1528         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1529                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1530 }
1531
1532 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1533                                         unsigned int val, int alloc)
1534 {
1535         int i;
1536
1537         if (type == NAT_JOURNAL) {
1538                 for (i = 0; i < nats_in_cursum(sum); i++) {
1539                         if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1540                                 return i;
1541                 }
1542                 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1543                         return update_nats_in_cursum(sum, 1);
1544         } else if (type == SIT_JOURNAL) {
1545                 for (i = 0; i < sits_in_cursum(sum); i++)
1546                         if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1547                                 return i;
1548                 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1549                         return update_sits_in_cursum(sum, 1);
1550         }
1551         return -1;
1552 }
1553
1554 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1555                                         unsigned int segno)
1556 {
1557         return get_meta_page(sbi, current_sit_addr(sbi, segno));
1558 }
1559
1560 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1561                                         unsigned int start)
1562 {
1563         struct sit_info *sit_i = SIT_I(sbi);
1564         struct page *src_page, *dst_page;
1565         pgoff_t src_off, dst_off;
1566         void *src_addr, *dst_addr;
1567
1568         src_off = current_sit_addr(sbi, start);
1569         dst_off = next_sit_addr(sbi, src_off);
1570
1571         /* get current sit block page without lock */
1572         src_page = get_meta_page(sbi, src_off);
1573         dst_page = grab_meta_page(sbi, dst_off);
1574         f2fs_bug_on(sbi, PageDirty(src_page));
1575
1576         src_addr = page_address(src_page);
1577         dst_addr = page_address(dst_page);
1578         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1579
1580         set_page_dirty(dst_page);
1581         f2fs_put_page(src_page, 1);
1582
1583         set_to_next_sit(sit_i, start);
1584
1585         return dst_page;
1586 }
1587
1588 static struct sit_entry_set *grab_sit_entry_set(void)
1589 {
1590         struct sit_entry_set *ses =
1591                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_ATOMIC);
1592
1593         ses->entry_cnt = 0;
1594         INIT_LIST_HEAD(&ses->set_list);
1595         return ses;
1596 }
1597
1598 static void release_sit_entry_set(struct sit_entry_set *ses)
1599 {
1600         list_del(&ses->set_list);
1601         kmem_cache_free(sit_entry_set_slab, ses);
1602 }
1603
1604 static void adjust_sit_entry_set(struct sit_entry_set *ses,
1605                                                 struct list_head *head)
1606 {
1607         struct sit_entry_set *next = ses;
1608
1609         if (list_is_last(&ses->set_list, head))
1610                 return;
1611
1612         list_for_each_entry_continue(next, head, set_list)
1613                 if (ses->entry_cnt <= next->entry_cnt)
1614                         break;
1615
1616         list_move_tail(&ses->set_list, &next->set_list);
1617 }
1618
1619 static void add_sit_entry(unsigned int segno, struct list_head *head)
1620 {
1621         struct sit_entry_set *ses;
1622         unsigned int start_segno = START_SEGNO(segno);
1623
1624         list_for_each_entry(ses, head, set_list) {
1625                 if (ses->start_segno == start_segno) {
1626                         ses->entry_cnt++;
1627                         adjust_sit_entry_set(ses, head);
1628                         return;
1629                 }
1630         }
1631
1632         ses = grab_sit_entry_set();
1633
1634         ses->start_segno = start_segno;
1635         ses->entry_cnt++;
1636         list_add(&ses->set_list, head);
1637 }
1638
1639 static void add_sits_in_set(struct f2fs_sb_info *sbi)
1640 {
1641         struct f2fs_sm_info *sm_info = SM_I(sbi);
1642         struct list_head *set_list = &sm_info->sit_entry_set;
1643         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
1644         unsigned int segno;
1645
1646         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
1647                 add_sit_entry(segno, set_list);
1648 }
1649
1650 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
1651 {
1652         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1653         struct f2fs_summary_block *sum = curseg->sum_blk;
1654         int i;
1655
1656         for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1657                 unsigned int segno;
1658                 bool dirtied;
1659
1660                 segno = le32_to_cpu(segno_in_journal(sum, i));
1661                 dirtied = __mark_sit_entry_dirty(sbi, segno);
1662
1663                 if (!dirtied)
1664                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
1665         }
1666         update_sits_in_cursum(sum, -sits_in_cursum(sum));
1667 }
1668
1669 /*
1670  * CP calls this function, which flushes SIT entries including sit_journal,
1671  * and moves prefree segs to free segs.
1672  */
1673 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1674 {
1675         struct sit_info *sit_i = SIT_I(sbi);
1676         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1677         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1678         struct f2fs_summary_block *sum = curseg->sum_blk;
1679         struct sit_entry_set *ses, *tmp;
1680         struct list_head *head = &SM_I(sbi)->sit_entry_set;
1681         bool to_journal = true;
1682         struct seg_entry *se;
1683
1684         mutex_lock(&curseg->curseg_mutex);
1685         mutex_lock(&sit_i->sentry_lock);
1686
1687         /*
1688          * add and account sit entries of dirty bitmap in sit entry
1689          * set temporarily
1690          */
1691         add_sits_in_set(sbi);
1692
1693         /*
1694          * if there are no enough space in journal to store dirty sit
1695          * entries, remove all entries from journal and add and account
1696          * them in sit entry set.
1697          */
1698         if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL))
1699                 remove_sits_in_journal(sbi);
1700
1701         if (!sit_i->dirty_sentries)
1702                 goto out;
1703
1704         /*
1705          * there are two steps to flush sit entries:
1706          * #1, flush sit entries to journal in current cold data summary block.
1707          * #2, flush sit entries to sit page.
1708          */
1709         list_for_each_entry_safe(ses, tmp, head, set_list) {
1710                 struct page *page = NULL;
1711                 struct f2fs_sit_block *raw_sit = NULL;
1712                 unsigned int start_segno = ses->start_segno;
1713                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
1714                                                 (unsigned long)MAIN_SEGS(sbi));
1715                 unsigned int segno = start_segno;
1716
1717                 if (to_journal &&
1718                         !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL))
1719                         to_journal = false;
1720
1721                 if (!to_journal) {
1722                         page = get_next_sit_page(sbi, start_segno);
1723                         raw_sit = page_address(page);
1724                 }
1725
1726                 /* flush dirty sit entries in region of current sit set */
1727                 for_each_set_bit_from(segno, bitmap, end) {
1728                         int offset, sit_offset;
1729
1730                         se = get_seg_entry(sbi, segno);
1731
1732                         /* add discard candidates */
1733                         if (SM_I(sbi)->nr_discards < SM_I(sbi)->max_discards) {
1734                                 cpc->trim_start = segno;
1735                                 add_discard_addrs(sbi, cpc);
1736                         }
1737
1738                         if (to_journal) {
1739                                 offset = lookup_journal_in_cursum(sum,
1740                                                         SIT_JOURNAL, segno, 1);
1741                                 f2fs_bug_on(sbi, offset < 0);
1742                                 segno_in_journal(sum, offset) =
1743                                                         cpu_to_le32(segno);
1744                                 seg_info_to_raw_sit(se,
1745                                                 &sit_in_journal(sum, offset));
1746                         } else {
1747                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1748                                 seg_info_to_raw_sit(se,
1749                                                 &raw_sit->entries[sit_offset]);
1750                         }
1751
1752                         __clear_bit(segno, bitmap);
1753                         sit_i->dirty_sentries--;
1754                         ses->entry_cnt--;
1755                 }
1756
1757                 if (!to_journal)
1758                         f2fs_put_page(page, 1);
1759
1760                 f2fs_bug_on(sbi, ses->entry_cnt);
1761                 release_sit_entry_set(ses);
1762         }
1763
1764         f2fs_bug_on(sbi, !list_empty(head));
1765         f2fs_bug_on(sbi, sit_i->dirty_sentries);
1766 out:
1767         if (cpc->reason == CP_DISCARD) {
1768                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
1769                         add_discard_addrs(sbi, cpc);
1770         }
1771         mutex_unlock(&sit_i->sentry_lock);
1772         mutex_unlock(&curseg->curseg_mutex);
1773
1774         set_prefree_as_free_segments(sbi);
1775 }
1776
1777 static int build_sit_info(struct f2fs_sb_info *sbi)
1778 {
1779         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1780         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1781         struct sit_info *sit_i;
1782         unsigned int sit_segs, start;
1783         char *src_bitmap, *dst_bitmap;
1784         unsigned int bitmap_size;
1785
1786         /* allocate memory for SIT information */
1787         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1788         if (!sit_i)
1789                 return -ENOMEM;
1790
1791         SM_I(sbi)->sit_info = sit_i;
1792
1793         sit_i->sentries = vzalloc(MAIN_SEGS(sbi) * sizeof(struct seg_entry));
1794         if (!sit_i->sentries)
1795                 return -ENOMEM;
1796
1797         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
1798         sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1799         if (!sit_i->dirty_sentries_bitmap)
1800                 return -ENOMEM;
1801
1802         for (start = 0; start < MAIN_SEGS(sbi); start++) {
1803                 sit_i->sentries[start].cur_valid_map
1804                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1805                 sit_i->sentries[start].ckpt_valid_map
1806                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1807                 if (!sit_i->sentries[start].cur_valid_map
1808                                 || !sit_i->sentries[start].ckpt_valid_map)
1809                         return -ENOMEM;
1810         }
1811
1812         if (sbi->segs_per_sec > 1) {
1813                 sit_i->sec_entries = vzalloc(MAIN_SECS(sbi) *
1814                                         sizeof(struct sec_entry));
1815                 if (!sit_i->sec_entries)
1816                         return -ENOMEM;
1817         }
1818
1819         /* get information related with SIT */
1820         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1821
1822         /* setup SIT bitmap from ckeckpoint pack */
1823         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1824         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1825
1826         dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1827         if (!dst_bitmap)
1828                 return -ENOMEM;
1829
1830         /* init SIT information */
1831         sit_i->s_ops = &default_salloc_ops;
1832
1833         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1834         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1835         sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1836         sit_i->sit_bitmap = dst_bitmap;
1837         sit_i->bitmap_size = bitmap_size;
1838         sit_i->dirty_sentries = 0;
1839         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1840         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1841         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1842         mutex_init(&sit_i->sentry_lock);
1843         return 0;
1844 }
1845
1846 static int build_free_segmap(struct f2fs_sb_info *sbi)
1847 {
1848         struct free_segmap_info *free_i;
1849         unsigned int bitmap_size, sec_bitmap_size;
1850
1851         /* allocate memory for free segmap information */
1852         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1853         if (!free_i)
1854                 return -ENOMEM;
1855
1856         SM_I(sbi)->free_info = free_i;
1857
1858         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
1859         free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1860         if (!free_i->free_segmap)
1861                 return -ENOMEM;
1862
1863         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
1864         free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1865         if (!free_i->free_secmap)
1866                 return -ENOMEM;
1867
1868         /* set all segments as dirty temporarily */
1869         memset(free_i->free_segmap, 0xff, bitmap_size);
1870         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1871
1872         /* init free segmap information */
1873         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
1874         free_i->free_segments = 0;
1875         free_i->free_sections = 0;
1876         rwlock_init(&free_i->segmap_lock);
1877         return 0;
1878 }
1879
1880 static int build_curseg(struct f2fs_sb_info *sbi)
1881 {
1882         struct curseg_info *array;
1883         int i;
1884
1885         array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
1886         if (!array)
1887                 return -ENOMEM;
1888
1889         SM_I(sbi)->curseg_array = array;
1890
1891         for (i = 0; i < NR_CURSEG_TYPE; i++) {
1892                 mutex_init(&array[i].curseg_mutex);
1893                 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1894                 if (!array[i].sum_blk)
1895                         return -ENOMEM;
1896                 array[i].segno = NULL_SEGNO;
1897                 array[i].next_blkoff = 0;
1898         }
1899         return restore_curseg_summaries(sbi);
1900 }
1901
1902 static void build_sit_entries(struct f2fs_sb_info *sbi)
1903 {
1904         struct sit_info *sit_i = SIT_I(sbi);
1905         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1906         struct f2fs_summary_block *sum = curseg->sum_blk;
1907         int sit_blk_cnt = SIT_BLK_CNT(sbi);
1908         unsigned int i, start, end;
1909         unsigned int readed, start_blk = 0;
1910         int nrpages = MAX_BIO_BLOCKS(sbi);
1911
1912         do {
1913                 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT);
1914
1915                 start = start_blk * sit_i->sents_per_block;
1916                 end = (start_blk + readed) * sit_i->sents_per_block;
1917
1918                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
1919                         struct seg_entry *se = &sit_i->sentries[start];
1920                         struct f2fs_sit_block *sit_blk;
1921                         struct f2fs_sit_entry sit;
1922                         struct page *page;
1923
1924                         mutex_lock(&curseg->curseg_mutex);
1925                         for (i = 0; i < sits_in_cursum(sum); i++) {
1926                                 if (le32_to_cpu(segno_in_journal(sum, i))
1927                                                                 == start) {
1928                                         sit = sit_in_journal(sum, i);
1929                                         mutex_unlock(&curseg->curseg_mutex);
1930                                         goto got_it;
1931                                 }
1932                         }
1933                         mutex_unlock(&curseg->curseg_mutex);
1934
1935                         page = get_current_sit_page(sbi, start);
1936                         sit_blk = (struct f2fs_sit_block *)page_address(page);
1937                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1938                         f2fs_put_page(page, 1);
1939 got_it:
1940                         check_block_count(sbi, start, &sit);
1941                         seg_info_from_raw_sit(se, &sit);
1942                         if (sbi->segs_per_sec > 1) {
1943                                 struct sec_entry *e = get_sec_entry(sbi, start);
1944                                 e->valid_blocks += se->valid_blocks;
1945                         }
1946                 }
1947                 start_blk += readed;
1948         } while (start_blk < sit_blk_cnt);
1949 }
1950
1951 static void init_free_segmap(struct f2fs_sb_info *sbi)
1952 {
1953         unsigned int start;
1954         int type;
1955
1956         for (start = 0; start < MAIN_SEGS(sbi); start++) {
1957                 struct seg_entry *sentry = get_seg_entry(sbi, start);
1958                 if (!sentry->valid_blocks)
1959                         __set_free(sbi, start);
1960         }
1961
1962         /* set use the current segments */
1963         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1964                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1965                 __set_test_and_inuse(sbi, curseg_t->segno);
1966         }
1967 }
1968
1969 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1970 {
1971         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1972         struct free_segmap_info *free_i = FREE_I(sbi);
1973         unsigned int segno = 0, offset = 0;
1974         unsigned short valid_blocks;
1975
1976         while (1) {
1977                 /* find dirty segment based on free segmap */
1978                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
1979                 if (segno >= MAIN_SEGS(sbi))
1980                         break;
1981                 offset = segno + 1;
1982                 valid_blocks = get_valid_blocks(sbi, segno, 0);
1983                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
1984                         continue;
1985                 if (valid_blocks > sbi->blocks_per_seg) {
1986                         f2fs_bug_on(sbi, 1);
1987                         continue;
1988                 }
1989                 mutex_lock(&dirty_i->seglist_lock);
1990                 __locate_dirty_segment(sbi, segno, DIRTY);
1991                 mutex_unlock(&dirty_i->seglist_lock);
1992         }
1993 }
1994
1995 static int init_victim_secmap(struct f2fs_sb_info *sbi)
1996 {
1997         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1998         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
1999
2000         dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
2001         if (!dirty_i->victim_secmap)
2002                 return -ENOMEM;
2003         return 0;
2004 }
2005
2006 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2007 {
2008         struct dirty_seglist_info *dirty_i;
2009         unsigned int bitmap_size, i;
2010
2011         /* allocate memory for dirty segments list information */
2012         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2013         if (!dirty_i)
2014                 return -ENOMEM;
2015
2016         SM_I(sbi)->dirty_info = dirty_i;
2017         mutex_init(&dirty_i->seglist_lock);
2018
2019         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2020
2021         for (i = 0; i < NR_DIRTY_TYPE; i++) {
2022                 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
2023                 if (!dirty_i->dirty_segmap[i])
2024                         return -ENOMEM;
2025         }
2026
2027         init_dirty_segmap(sbi);
2028         return init_victim_secmap(sbi);
2029 }
2030
2031 /*
2032  * Update min, max modified time for cost-benefit GC algorithm
2033  */
2034 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2035 {
2036         struct sit_info *sit_i = SIT_I(sbi);
2037         unsigned int segno;
2038
2039         mutex_lock(&sit_i->sentry_lock);
2040
2041         sit_i->min_mtime = LLONG_MAX;
2042
2043         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2044                 unsigned int i;
2045                 unsigned long long mtime = 0;
2046
2047                 for (i = 0; i < sbi->segs_per_sec; i++)
2048                         mtime += get_seg_entry(sbi, segno + i)->mtime;
2049
2050                 mtime = div_u64(mtime, sbi->segs_per_sec);
2051
2052                 if (sit_i->min_mtime > mtime)
2053                         sit_i->min_mtime = mtime;
2054         }
2055         sit_i->max_mtime = get_mtime(sbi);
2056         mutex_unlock(&sit_i->sentry_lock);
2057 }
2058
2059 int build_segment_manager(struct f2fs_sb_info *sbi)
2060 {
2061         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2062         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2063         struct f2fs_sm_info *sm_info;
2064         int err;
2065
2066         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2067         if (!sm_info)
2068                 return -ENOMEM;
2069
2070         /* init sm info */
2071         sbi->sm_info = sm_info;
2072         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2073         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2074         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2075         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2076         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2077         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2078         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2079         sm_info->rec_prefree_segments = sm_info->main_segments *
2080                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2081         sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2082         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2083         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2084
2085         INIT_LIST_HEAD(&sm_info->discard_list);
2086         sm_info->nr_discards = 0;
2087         sm_info->max_discards = 0;
2088
2089         INIT_LIST_HEAD(&sm_info->sit_entry_set);
2090
2091         if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2092                 err = create_flush_cmd_control(sbi);
2093                 if (err)
2094                         return err;
2095         }
2096
2097         err = build_sit_info(sbi);
2098         if (err)
2099                 return err;
2100         err = build_free_segmap(sbi);
2101         if (err)
2102                 return err;
2103         err = build_curseg(sbi);
2104         if (err)
2105                 return err;
2106
2107         /* reinit free segmap based on SIT */
2108         build_sit_entries(sbi);
2109
2110         init_free_segmap(sbi);
2111         err = build_dirty_segmap(sbi);
2112         if (err)
2113                 return err;
2114
2115         init_min_max_mtime(sbi);
2116         return 0;
2117 }
2118
2119 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2120                 enum dirty_type dirty_type)
2121 {
2122         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2123
2124         mutex_lock(&dirty_i->seglist_lock);
2125         kfree(dirty_i->dirty_segmap[dirty_type]);
2126         dirty_i->nr_dirty[dirty_type] = 0;
2127         mutex_unlock(&dirty_i->seglist_lock);
2128 }
2129
2130 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2131 {
2132         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2133         kfree(dirty_i->victim_secmap);
2134 }
2135
2136 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2137 {
2138         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2139         int i;
2140
2141         if (!dirty_i)
2142                 return;
2143
2144         /* discard pre-free/dirty segments list */
2145         for (i = 0; i < NR_DIRTY_TYPE; i++)
2146                 discard_dirty_segmap(sbi, i);
2147
2148         destroy_victim_secmap(sbi);
2149         SM_I(sbi)->dirty_info = NULL;
2150         kfree(dirty_i);
2151 }
2152
2153 static void destroy_curseg(struct f2fs_sb_info *sbi)
2154 {
2155         struct curseg_info *array = SM_I(sbi)->curseg_array;
2156         int i;
2157
2158         if (!array)
2159                 return;
2160         SM_I(sbi)->curseg_array = NULL;
2161         for (i = 0; i < NR_CURSEG_TYPE; i++)
2162                 kfree(array[i].sum_blk);
2163         kfree(array);
2164 }
2165
2166 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2167 {
2168         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2169         if (!free_i)
2170                 return;
2171         SM_I(sbi)->free_info = NULL;
2172         kfree(free_i->free_segmap);
2173         kfree(free_i->free_secmap);
2174         kfree(free_i);
2175 }
2176
2177 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2178 {
2179         struct sit_info *sit_i = SIT_I(sbi);
2180         unsigned int start;
2181
2182         if (!sit_i)
2183                 return;
2184
2185         if (sit_i->sentries) {
2186                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2187                         kfree(sit_i->sentries[start].cur_valid_map);
2188                         kfree(sit_i->sentries[start].ckpt_valid_map);
2189                 }
2190         }
2191         vfree(sit_i->sentries);
2192         vfree(sit_i->sec_entries);
2193         kfree(sit_i->dirty_sentries_bitmap);
2194
2195         SM_I(sbi)->sit_info = NULL;
2196         kfree(sit_i->sit_bitmap);
2197         kfree(sit_i);
2198 }
2199
2200 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2201 {
2202         struct f2fs_sm_info *sm_info = SM_I(sbi);
2203
2204         if (!sm_info)
2205                 return;
2206         destroy_flush_cmd_control(sbi);
2207         destroy_dirty_segmap(sbi);
2208         destroy_curseg(sbi);
2209         destroy_free_segmap(sbi);
2210         destroy_sit_info(sbi);
2211         sbi->sm_info = NULL;
2212         kfree(sm_info);
2213 }
2214
2215 int __init create_segment_manager_caches(void)
2216 {
2217         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2218                         sizeof(struct discard_entry));
2219         if (!discard_entry_slab)
2220                 goto fail;
2221
2222         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2223                         sizeof(struct nat_entry_set));
2224         if (!sit_entry_set_slab)
2225                 goto destory_discard_entry;
2226
2227         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2228                         sizeof(struct inmem_pages));
2229         if (!inmem_entry_slab)
2230                 goto destroy_sit_entry_set;
2231         return 0;
2232
2233 destroy_sit_entry_set:
2234         kmem_cache_destroy(sit_entry_set_slab);
2235 destory_discard_entry:
2236         kmem_cache_destroy(discard_entry_slab);
2237 fail:
2238         return -ENOMEM;
2239 }
2240
2241 void destroy_segment_manager_caches(void)
2242 {
2243         kmem_cache_destroy(sit_entry_set_slab);
2244         kmem_cache_destroy(discard_entry_slab);
2245         kmem_cache_destroy(inmem_entry_slab);
2246 }