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
 183         SetPagePrivate(page);
 184
 185         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 186
 187         /* add atomic page indices to the list */
 188         new->page = page;
 189         INIT_LIST_HEAD(&new->list);
 190 retry:
 191         /* increase reference count with clean state */
 192         mutex_lock(&fi->inmem_lock);
 193         err = radix_tree_insert(&fi->inmem_root, page->index, new);
 194         if (err == -EEXIST) {
 195                 mutex_unlock(&fi->inmem_lock);
 196                 kmem_cache_free(inmem_entry_slab, new);
 197                 return;
 198         } else if (err) {
 199                 mutex_unlock(&fi->inmem_lock);
 200                 goto retry;
 201         }
 202         get_page(page);
 203         list_add_tail(&new->list, &fi->inmem_pages);
 204         mutex_unlock(&fi->inmem_lock);
 205 }
 206
 207 void invalidate_inmem_page(struct inode *inode, struct page *page)
 208 {
 209         struct f2fs_inode_info *fi = F2FS_I(inode);
 210         struct inmem_pages *cur;
 211
 212         mutex_lock(&fi->inmem_lock);
 213         cur = radix_tree_lookup(&fi->inmem_root, page->index);
 214         if (cur) {
 215                 radix_tree_delete(&fi->inmem_root, cur->page->index);
 216                 f2fs_put_page(cur->page, 0);
 217                 list_del(&cur->list);
 218                 kmem_cache_free(inmem_entry_slab, cur);
 219         }
 220         mutex_unlock(&fi->inmem_lock);
 221 }
 222
 223 void commit_inmem_pages(struct inode *inode, bool abort)
 224 {
 225         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 226         struct f2fs_inode_info *fi = F2FS_I(inode);
 227         struct inmem_pages *cur, *tmp;
 228         bool submit_bio = false;
 229         struct f2fs_io_info fio = {
 230                 .type = DATA,
 231                 .rw = WRITE_SYNC,
 232         };
 233
 234         /*
 235          * The abort is true only when f2fs_evict_inode is called.
 236          * Basically, the f2fs_evict_inode doesn't produce any data writes, so
 237          * that we don't need to call f2fs_balance_fs.
 238          * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this
 239          * inode becomes free by iget_locked in f2fs_iget.
 240          */
 241         if (!abort)
 242                 f2fs_balance_fs(sbi);
 243
 244         f2fs_lock_op(sbi);
 245
 246         mutex_lock(&fi->inmem_lock);
 247         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 248                 lock_page(cur->page);
 249                 if (!abort && cur->page->mapping == inode->i_mapping) {
 250                         f2fs_wait_on_page_writeback(cur->page, DATA);
 251                         if (clear_page_dirty_for_io(cur->page))
 252                                 inode_dec_dirty_pages(inode);
 253                         do_write_data_page(cur->page, &fio);
 254                         submit_bio = true;
 255                 }
 256                 radix_tree_delete(&fi->inmem_root, cur->page->index);
 257                 f2fs_put_page(cur->page, 1);
 258                 list_del(&cur->list);
 259                 kmem_cache_free(inmem_entry_slab, cur);
 260         }
 261         if (submit_bio)
 262                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 263         mutex_unlock(&fi->inmem_lock);
 264
 265         filemap_fdatawait_range(inode->i_mapping, 0, LLONG_MAX);
 266         f2fs_unlock_op(sbi);
 267 }
 268
 269 /*
 270  * This function balances dirty node and dentry pages.
 271  * In addition, it controls garbage collection.
 272  */
 273 void f2fs_balance_fs(struct f2fs_sb_info *sbi)
 274 {
 275         /*
 276          * We should do GC or end up with checkpoint, if there are so many dirty
 277          * dir/node pages without enough free segments.
 278          */
 279         if (has_not_enough_free_secs(sbi, 0)) {
 280                 mutex_lock(&sbi->gc_mutex);
 281                 f2fs_gc(sbi);
 282         }
 283 }
 284
 285 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 286 {
 287         /* check the # of cached NAT entries and prefree segments */
 288         if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) ||
 289                         excess_prefree_segs(sbi) ||
 290                         available_free_memory(sbi, INO_ENTRIES))
 291                 f2fs_sync_fs(sbi->sb, true);
 292 }
 293
 294 static int issue_flush_thread(void *data)
 295 {
 296         struct f2fs_sb_info *sbi = data;
 297         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 298         wait_queue_head_t *q = &fcc->flush_wait_queue;
 299 repeat:
 300         if (kthread_should_stop())
 301                 return 0;
 302
 303         if (!llist_empty(&fcc->issue_list)) {
 304                 struct bio *bio = bio_alloc(GFP_NOIO, 0);
 305                 struct flush_cmd *cmd, *next;
 306                 int ret;
 307
 308                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 309                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 310
 311                 bio->bi_bdev = sbi->sb->s_bdev;
 312                 ret = submit_bio_wait(WRITE_FLUSH, bio);
 313
 314                 llist_for_each_entry_safe(cmd, next,
 315                                           fcc->dispatch_list, llnode) {
 316                         cmd->ret = ret;
 317                         complete(&cmd->wait);
 318                 }
 319                 bio_put(bio);
 320                 fcc->dispatch_list = NULL;
 321         }
 322
 323         wait_event_interruptible(*q,
 324                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 325         goto repeat;
 326 }
 327
 328 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
 329 {
 330         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 331         struct flush_cmd cmd;
 332
 333         trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
 334                                         test_opt(sbi, FLUSH_MERGE));
 335
 336         if (test_opt(sbi, NOBARRIER))
 337                 return 0;
 338
 339         if (!test_opt(sbi, FLUSH_MERGE))
 340                 return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL);
 341
 342         init_completion(&cmd.wait);
 343
 344         llist_add(&cmd.llnode, &fcc->issue_list);
 345
 346         if (!fcc->dispatch_list)
 347                 wake_up(&fcc->flush_wait_queue);
 348
 349         wait_for_completion(&cmd.wait);
 350
 351         return cmd.ret;
 352 }
 353
 354 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
 355 {
 356         dev_t dev = sbi->sb->s_bdev->bd_dev;
 357         struct flush_cmd_control *fcc;
 358         int err = 0;
 359
 360         fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
 361         if (!fcc)
 362                 return -ENOMEM;
 363         init_waitqueue_head(&fcc->flush_wait_queue);
 364         init_llist_head(&fcc->issue_list);
 365         SM_I(sbi)->cmd_control_info = fcc;
 366         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 367                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 368         if (IS_ERR(fcc->f2fs_issue_flush)) {
 369                 err = PTR_ERR(fcc->f2fs_issue_flush);
 370                 kfree(fcc);
 371                 SM_I(sbi)->cmd_control_info = NULL;
 372                 return err;
 373         }
 374
 375         return err;
 376 }
 377
 378 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
 379 {
 380         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 381
 382         if (fcc && fcc->f2fs_issue_flush)
 383                 kthread_stop(fcc->f2fs_issue_flush);
 384         kfree(fcc);
 385         SM_I(sbi)->cmd_control_info = NULL;
 386 }
 387
 388 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 389                 enum dirty_type dirty_type)
 390 {
 391         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 392
 393         /* need not be added */
 394         if (IS_CURSEG(sbi, segno))
 395                 return;
 396
 397         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 398                 dirty_i->nr_dirty[dirty_type]++;
 399
 400         if (dirty_type == DIRTY) {
 401                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 402                 enum dirty_type t = sentry->type;
 403
 404                 if (unlikely(t >= DIRTY)) {
 405                         f2fs_bug_on(sbi, 1);
 406                         return;
 407                 }
 408                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 409                         dirty_i->nr_dirty[t]++;
 410         }
 411 }
 412
 413 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 414                 enum dirty_type dirty_type)
 415 {
 416         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 417
 418         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 419                 dirty_i->nr_dirty[dirty_type]--;
 420
 421         if (dirty_type == DIRTY) {
 422                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 423                 enum dirty_type t = sentry->type;
 424
 425                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 426                         dirty_i->nr_dirty[t]--;
 427
 428                 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
 429                         clear_bit(GET_SECNO(sbi, segno),
 430                                                 dirty_i->victim_secmap);
 431         }
 432 }
 433
 434 /*
 435  * Should not occur error such as -ENOMEM.
 436  * Adding dirty entry into seglist is not critical operation.
 437  * If a given segment is one of current working segments, it won't be added.
 438  */
 439 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 440 {
 441         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 442         unsigned short valid_blocks;
 443
 444         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 445                 return;
 446
 447         mutex_lock(&dirty_i->seglist_lock);
 448
 449         valid_blocks = get_valid_blocks(sbi, segno, 0);
 450
 451         if (valid_blocks == 0) {
 452                 __locate_dirty_segment(sbi, segno, PRE);
 453                 __remove_dirty_segment(sbi, segno, DIRTY);
 454         } else if (valid_blocks < sbi->blocks_per_seg) {
 455                 __locate_dirty_segment(sbi, segno, DIRTY);
 456         } else {
 457                 /* Recovery routine with SSR needs this */
 458                 __remove_dirty_segment(sbi, segno, DIRTY);
 459         }
 460
 461         mutex_unlock(&dirty_i->seglist_lock);
 462 }
 463
 464 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
 465                                 block_t blkstart, block_t blklen)
 466 {
 467         sector_t start = SECTOR_FROM_BLOCK(blkstart);
 468         sector_t len = SECTOR_FROM_BLOCK(blklen);
 469         trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
 470         return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
 471 }
 472
 473 void discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
 474 {
 475         if (f2fs_issue_discard(sbi, blkaddr, 1)) {
 476                 struct page *page = grab_meta_page(sbi, blkaddr);
 477                 /* zero-filled page */
 478                 set_page_dirty(page);
 479                 f2fs_put_page(page, 1);
 480         }
 481 }
 482
 483 static void __add_discard_entry(struct f2fs_sb_info *sbi,
 484                 struct cp_control *cpc, unsigned int start, unsigned int end)
 485 {
 486         struct list_head *head = &SM_I(sbi)->discard_list;
 487         struct discard_entry *new, *last;
 488
 489         if (!list_empty(head)) {
 490                 last = list_last_entry(head, struct discard_entry, list);
 491                 if (START_BLOCK(sbi, cpc->trim_start) + start ==
 492                                                 last->blkaddr + last->len) {
 493                         last->len += end - start;
 494                         goto done;
 495                 }
 496         }
 497
 498         new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
 499         INIT_LIST_HEAD(&new->list);
 500         new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
 501         new->len = end - start;
 502         list_add_tail(&new->list, head);
 503 done:
 504         SM_I(sbi)->nr_discards += end - start;
 505         cpc->trimmed += end - start;
 506 }
 507
 508 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 509 {
 510         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
 511         int max_blocks = sbi->blocks_per_seg;
 512         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
 513         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
 514         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
 515         unsigned long dmap[entries];
 516         unsigned int start = 0, end = -1;
 517         bool force = (cpc->reason == CP_DISCARD);
 518         int i;
 519
 520         if (!force && !test_opt(sbi, DISCARD))
 521                 return;
 522
 523         if (force && !se->valid_blocks) {
 524                 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 525                 /*
 526                  * if this segment is registered in the prefree list, then
 527                  * we should skip adding a discard candidate, and let the
 528                  * checkpoint do that later.
 529                  */
 530                 mutex_lock(&dirty_i->seglist_lock);
 531                 if (test_bit(cpc->trim_start, dirty_i->dirty_segmap[PRE])) {
 532                         mutex_unlock(&dirty_i->seglist_lock);
 533                         cpc->trimmed += sbi->blocks_per_seg;
 534                         return;
 535                 }
 536                 mutex_unlock(&dirty_i->seglist_lock);
 537
 538                 __add_discard_entry(sbi, cpc, 0, sbi->blocks_per_seg);
 539                 return;
 540         }
 541
 542         /* zero block will be discarded through the prefree list */
 543         if (!se->valid_blocks || se->valid_blocks == max_blocks)
 544                 return;
 545
 546         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
 547         for (i = 0; i < entries; i++)
 548                 dmap[i] = ~(cur_map[i] | ckpt_map[i]);
 549
 550         while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
 551                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
 552                 if (start >= max_blocks)
 553                         break;
 554
 555                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
 556
 557                 if (end - start < cpc->trim_minlen)
 558                         continue;
 559
 560                 __add_discard_entry(sbi, cpc, start, end);
 561         }
 562 }
 563
 564 void release_discard_addrs(struct f2fs_sb_info *sbi)
 565 {
 566         struct list_head *head = &(SM_I(sbi)->discard_list);
 567         struct discard_entry *entry, *this;
 568
 569         /* drop caches */
 570         list_for_each_entry_safe(entry, this, head, list) {
 571                 list_del(&entry->list);
 572                 kmem_cache_free(discard_entry_slab, entry);
 573         }
 574 }
 575
 576 /*
 577  * Should call clear_prefree_segments after checkpoint is done.
 578  */
 579 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
 580 {
 581         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 582         unsigned int segno;
 583
 584         mutex_lock(&dirty_i->seglist_lock);
 585         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
 586                 __set_test_and_free(sbi, segno);
 587         mutex_unlock(&dirty_i->seglist_lock);
 588 }
 589
 590 void clear_prefree_segments(struct f2fs_sb_info *sbi)
 591 {
 592         struct list_head *head = &(SM_I(sbi)->discard_list);
 593         struct discard_entry *entry, *this;
 594         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 595         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
 596         unsigned int start = 0, end = -1;
 597
 598         mutex_lock(&dirty_i->seglist_lock);
 599
 600         while (1) {
 601                 int i;
 602                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
 603                 if (start >= MAIN_SEGS(sbi))
 604                         break;
 605                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
 606                                                                 start + 1);
 607
 608                 for (i = start; i < end; i++)
 609                         clear_bit(i, prefree_map);
 610
 611                 dirty_i->nr_dirty[PRE] -= end - start;
 612
 613                 if (!test_opt(sbi, DISCARD))
 614                         continue;
 615
 616                 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
 617                                 (end - start) << sbi->log_blocks_per_seg);
 618         }
 619         mutex_unlock(&dirty_i->seglist_lock);
 620
 621         /* send small discards */
 622         list_for_each_entry_safe(entry, this, head, list) {
 623                 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
 624                 list_del(&entry->list);
 625                 SM_I(sbi)->nr_discards -= entry->len;
 626                 kmem_cache_free(discard_entry_slab, entry);
 627         }
 628 }
 629
 630 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
 631 {
 632         struct sit_info *sit_i = SIT_I(sbi);
 633
 634         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
 635                 sit_i->dirty_sentries++;
 636                 return false;
 637         }
 638
 639         return true;
 640 }
 641
 642 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
 643                                         unsigned int segno, int modified)
 644 {
 645         struct seg_entry *se = get_seg_entry(sbi, segno);
 646         se->type = type;
 647         if (modified)
 648                 __mark_sit_entry_dirty(sbi, segno);
 649 }
 650
 651 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
 652 {
 653         struct seg_entry *se;
 654         unsigned int segno, offset;
 655         long int new_vblocks;
 656
 657         segno = GET_SEGNO(sbi, blkaddr);
 658
 659         se = get_seg_entry(sbi, segno);
 660         new_vblocks = se->valid_blocks + del;
 661         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 662
 663         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
 664                                 (new_vblocks > sbi->blocks_per_seg)));
 665
 666         se->valid_blocks = new_vblocks;
 667         se->mtime = get_mtime(sbi);
 668         SIT_I(sbi)->max_mtime = se->mtime;
 669
 670         /* Update valid block bitmap */
 671         if (del > 0) {
 672                 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
 673                         f2fs_bug_on(sbi, 1);
 674         } else {
 675                 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
 676                         f2fs_bug_on(sbi, 1);
 677         }
 678         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
 679                 se->ckpt_valid_blocks += del;
 680
 681         __mark_sit_entry_dirty(sbi, segno);
 682
 683         /* update total number of valid blocks to be written in ckpt area */
 684         SIT_I(sbi)->written_valid_blocks += del;
 685
 686         if (sbi->segs_per_sec > 1)
 687                 get_sec_entry(sbi, segno)->valid_blocks += del;
 688 }
 689
 690 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
 691 {
 692         update_sit_entry(sbi, new, 1);
 693         if (GET_SEGNO(sbi, old) != NULL_SEGNO)
 694                 update_sit_entry(sbi, old, -1);
 695
 696         locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
 697         locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
 698 }
 699
 700 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
 701 {
 702         unsigned int segno = GET_SEGNO(sbi, addr);
 703         struct sit_info *sit_i = SIT_I(sbi);
 704
 705         f2fs_bug_on(sbi, addr == NULL_ADDR);
 706         if (addr == NEW_ADDR)
 707                 return;
 708
 709         /* add it into sit main buffer */
 710         mutex_lock(&sit_i->sentry_lock);
 711
 712         update_sit_entry(sbi, addr, -1);
 713
 714         /* add it into dirty seglist */
 715         locate_dirty_segment(sbi, segno);
 716
 717         mutex_unlock(&sit_i->sentry_lock);
 718 }
 719
 720 /*
 721  * This function should be resided under the curseg_mutex lock
 722  */
 723 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
 724                                         struct f2fs_summary *sum)
 725 {
 726         struct curseg_info *curseg = CURSEG_I(sbi, type);
 727         void *addr = curseg->sum_blk;
 728         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
 729         memcpy(addr, sum, sizeof(struct f2fs_summary));
 730 }
 731
 732 /*
 733  * Calculate the number of current summary pages for writing
 734  */
 735 int npages_for_summary_flush(struct f2fs_sb_info *sbi)
 736 {
 737         int valid_sum_count = 0;
 738         int i, sum_in_page;
 739
 740         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 741                 if (sbi->ckpt->alloc_type[i] == SSR)
 742                         valid_sum_count += sbi->blocks_per_seg;
 743                 else
 744                         valid_sum_count += curseg_blkoff(sbi, i);
 745         }
 746
 747         sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
 748                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
 749         if (valid_sum_count <= sum_in_page)
 750                 return 1;
 751         else if ((valid_sum_count - sum_in_page) <=
 752                 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
 753                 return 2;
 754         return 3;
 755 }
 756
 757 /*
 758  * Caller should put this summary page
 759  */
 760 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
 761 {
 762         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
 763 }
 764
 765 static void write_sum_page(struct f2fs_sb_info *sbi,
 766                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
 767 {
 768         struct page *page = grab_meta_page(sbi, blk_addr);
 769         void *kaddr = page_address(page);
 770         memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
 771         set_page_dirty(page);
 772         f2fs_put_page(page, 1);
 773 }
 774
 775 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
 776 {
 777         struct curseg_info *curseg = CURSEG_I(sbi, type);
 778         unsigned int segno = curseg->segno + 1;
 779         struct free_segmap_info *free_i = FREE_I(sbi);
 780
 781         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
 782                 return !test_bit(segno, free_i->free_segmap);
 783         return 0;
 784 }
 785
 786 /*
 787  * Find a new segment from the free segments bitmap to right order
 788  * This function should be returned with success, otherwise BUG
 789  */
 790 static void get_new_segment(struct f2fs_sb_info *sbi,
 791                         unsigned int *newseg, bool new_sec, int dir)
 792 {
 793         struct free_segmap_info *free_i = FREE_I(sbi);
 794         unsigned int segno, secno, zoneno;
 795         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
 796         unsigned int hint = *newseg / sbi->segs_per_sec;
 797         unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
 798         unsigned int left_start = hint;
 799         bool init = true;
 800         int go_left = 0;
 801         int i;
 802
 803         write_lock(&free_i->segmap_lock);
 804
 805         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
 806                 segno = find_next_zero_bit(free_i->free_segmap,
 807                                         MAIN_SEGS(sbi), *newseg + 1);
 808                 if (segno - *newseg < sbi->segs_per_sec -
 809                                         (*newseg % sbi->segs_per_sec))
 810                         goto got_it;
 811         }
 812 find_other_zone:
 813         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
 814         if (secno >= MAIN_SECS(sbi)) {
 815                 if (dir == ALLOC_RIGHT) {
 816                         secno = find_next_zero_bit(free_i->free_secmap,
 817                                                         MAIN_SECS(sbi), 0);
 818                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
 819                 } else {
 820                         go_left = 1;
 821                         left_start = hint - 1;
 822                 }
 823         }
 824         if (go_left == 0)
 825                 goto skip_left;
 826
 827         while (test_bit(left_start, free_i->free_secmap)) {
 828                 if (left_start > 0) {
 829                         left_start--;
 830                         continue;
 831                 }
 832                 left_start = find_next_zero_bit(free_i->free_secmap,
 833                                                         MAIN_SECS(sbi), 0);
 834                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
 835                 break;
 836         }
 837         secno = left_start;
 838 skip_left:
 839         hint = secno;
 840         segno = secno * sbi->segs_per_sec;
 841         zoneno = secno / sbi->secs_per_zone;
 842
 843         /* give up on finding another zone */
 844         if (!init)
 845                 goto got_it;
 846         if (sbi->secs_per_zone == 1)
 847                 goto got_it;
 848         if (zoneno == old_zoneno)
 849                 goto got_it;
 850         if (dir == ALLOC_LEFT) {
 851                 if (!go_left && zoneno + 1 >= total_zones)
 852                         goto got_it;
 853                 if (go_left && zoneno == 0)
 854                         goto got_it;
 855         }
 856         for (i = 0; i < NR_CURSEG_TYPE; i++)
 857                 if (CURSEG_I(sbi, i)->zone == zoneno)
 858                         break;
 859
 860         if (i < NR_CURSEG_TYPE) {
 861                 /* zone is in user, try another */
 862                 if (go_left)
 863                         hint = zoneno * sbi->secs_per_zone - 1;
 864                 else if (zoneno + 1 >= total_zones)
 865                         hint = 0;
 866                 else
 867                         hint = (zoneno + 1) * sbi->secs_per_zone;
 868                 init = false;
 869                 goto find_other_zone;
 870         }
 871 got_it:
 872         /* set it as dirty segment in free segmap */
 873         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
 874         __set_inuse(sbi, segno);
 875         *newseg = segno;
 876         write_unlock(&free_i->segmap_lock);
 877 }
 878
 879 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
 880 {
 881         struct curseg_info *curseg = CURSEG_I(sbi, type);
 882         struct summary_footer *sum_footer;
 883
 884         curseg->segno = curseg->next_segno;
 885         curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
 886         curseg->next_blkoff = 0;
 887         curseg->next_segno = NULL_SEGNO;
 888
 889         sum_footer = &(curseg->sum_blk->footer);
 890         memset(sum_footer, 0, sizeof(struct summary_footer));
 891         if (IS_DATASEG(type))
 892                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
 893         if (IS_NODESEG(type))
 894                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
 895         __set_sit_entry_type(sbi, type, curseg->segno, modified);
 896 }
 897
 898 /*
 899  * Allocate a current working segment.
 900  * This function always allocates a free segment in LFS manner.
 901  */
 902 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
 903 {
 904         struct curseg_info *curseg = CURSEG_I(sbi, type);
 905         unsigned int segno = curseg->segno;
 906         int dir = ALLOC_LEFT;
 907
 908         write_sum_page(sbi, curseg->sum_blk,
 909                                 GET_SUM_BLOCK(sbi, segno));
 910         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
 911                 dir = ALLOC_RIGHT;
 912
 913         if (test_opt(sbi, NOHEAP))
 914                 dir = ALLOC_RIGHT;
 915
 916         get_new_segment(sbi, &segno, new_sec, dir);
 917         curseg->next_segno = segno;
 918         reset_curseg(sbi, type, 1);
 919         curseg->alloc_type = LFS;
 920 }
 921
 922 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
 923                         struct curseg_info *seg, block_t start)
 924 {
 925         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
 926         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
 927         unsigned long target_map[entries];
 928         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
 929         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
 930         int i, pos;
 931
 932         for (i = 0; i < entries; i++)
 933                 target_map[i] = ckpt_map[i] | cur_map[i];
 934
 935         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
 936
 937         seg->next_blkoff = pos;
 938 }
 939
 940 /*
 941  * If a segment is written by LFS manner, next block offset is just obtained
 942  * by increasing the current block offset. However, if a segment is written by
 943  * SSR manner, next block offset obtained by calling __next_free_blkoff
 944  */
 945 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
 946                                 struct curseg_info *seg)
 947 {
 948         if (seg->alloc_type == SSR)
 949                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
 950         else
 951                 seg->next_blkoff++;
 952 }
 953
 954 /*
 955  * This function always allocates a used segment(from dirty seglist) by SSR
 956  * manner, so it should recover the existing segment information of valid blocks
 957  */
 958 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
 959 {
 960         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 961         struct curseg_info *curseg = CURSEG_I(sbi, type);
 962         unsigned int new_segno = curseg->next_segno;
 963         struct f2fs_summary_block *sum_node;
 964         struct page *sum_page;
 965
 966         write_sum_page(sbi, curseg->sum_blk,
 967                                 GET_SUM_BLOCK(sbi, curseg->segno));
 968         __set_test_and_inuse(sbi, new_segno);
 969
 970         mutex_lock(&dirty_i->seglist_lock);
 971         __remove_dirty_segment(sbi, new_segno, PRE);
 972         __remove_dirty_segment(sbi, new_segno, DIRTY);
 973         mutex_unlock(&dirty_i->seglist_lock);
 974
 975         reset_curseg(sbi, type, 1);
 976         curseg->alloc_type = SSR;
 977         __next_free_blkoff(sbi, curseg, 0);
 978
 979         if (reuse) {
 980                 sum_page = get_sum_page(sbi, new_segno);
 981                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
 982                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
 983                 f2fs_put_page(sum_page, 1);
 984         }
 985 }
 986
 987 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
 988 {
 989         struct curseg_info *curseg = CURSEG_I(sbi, type);
 990         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
 991
 992         if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
 993                 return v_ops->get_victim(sbi,
 994                                 &(curseg)->next_segno, BG_GC, type, SSR);
 995
 996         /* For data segments, let's do SSR more intensively */
 997         for (; type >= CURSEG_HOT_DATA; type--)
 998                 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
 999                                                 BG_GC, type, SSR))
1000                         return 1;
1001         return 0;
1002 }
1003
1004 /*
1005  * flush out current segment and replace it with new segment
1006  * This function should be returned with success, otherwise BUG
1007  */
1008 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1009                                                 int type, bool force)
1010 {
1011         struct curseg_info *curseg = CURSEG_I(sbi, type);
1012
1013         if (force)
1014                 new_curseg(sbi, type, true);
1015         else if (type == CURSEG_WARM_NODE)
1016                 new_curseg(sbi, type, false);
1017         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1018                 new_curseg(sbi, type, false);
1019         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1020                 change_curseg(sbi, type, true);
1021         else
1022                 new_curseg(sbi, type, false);
1023
1024         stat_inc_seg_type(sbi, curseg);
1025 }
1026
1027 void allocate_new_segments(struct f2fs_sb_info *sbi)
1028 {
1029         struct curseg_info *curseg;
1030         unsigned int old_curseg;
1031         int i;
1032
1033         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1034                 curseg = CURSEG_I(sbi, i);
1035                 old_curseg = curseg->segno;
1036                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
1037                 locate_dirty_segment(sbi, old_curseg);
1038         }
1039 }
1040
1041 static const struct segment_allocation default_salloc_ops = {
1042         .allocate_segment = allocate_segment_by_default,
1043 };
1044
1045 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1046 {
1047         __u64 start = range->start >> sbi->log_blocksize;
1048         __u64 end = start + (range->len >> sbi->log_blocksize) - 1;
1049         unsigned int start_segno, end_segno;
1050         struct cp_control cpc;
1051
1052         if (range->minlen > SEGMENT_SIZE(sbi) || start >= MAX_BLKADDR(sbi) ||
1053                                                 range->len < sbi->blocksize)
1054                 return -EINVAL;
1055
1056         cpc.trimmed = 0;
1057         if (end <= MAIN_BLKADDR(sbi))
1058                 goto out;
1059
1060         /* start/end segment number in main_area */
1061         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1062         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1063                                                 GET_SEGNO(sbi, end);
1064         cpc.reason = CP_DISCARD;
1065         cpc.trim_start = start_segno;
1066         cpc.trim_end = end_segno;
1067         cpc.trim_minlen = range->minlen >> sbi->log_blocksize;
1068
1069         /* do checkpoint to issue discard commands safely */
1070         mutex_lock(&sbi->gc_mutex);
1071         write_checkpoint(sbi, &cpc);
1072         mutex_unlock(&sbi->gc_mutex);
1073 out:
1074         range->len = cpc.trimmed << sbi->log_blocksize;
1075         return 0;
1076 }
1077
1078 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1079 {
1080         struct curseg_info *curseg = CURSEG_I(sbi, type);
1081         if (curseg->next_blkoff < sbi->blocks_per_seg)
1082                 return true;
1083         return false;
1084 }
1085
1086 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1087 {
1088         if (p_type == DATA)
1089                 return CURSEG_HOT_DATA;
1090         else
1091                 return CURSEG_HOT_NODE;
1092 }
1093
1094 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1095 {
1096         if (p_type == DATA) {
1097                 struct inode *inode = page->mapping->host;
1098
1099                 if (S_ISDIR(inode->i_mode))
1100                         return CURSEG_HOT_DATA;
1101                 else
1102                         return CURSEG_COLD_DATA;
1103         } else {
1104                 if (IS_DNODE(page) && is_cold_node(page))
1105                         return CURSEG_WARM_NODE;
1106                 else
1107                         return CURSEG_COLD_NODE;
1108         }
1109 }
1110
1111 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1112 {
1113         if (p_type == DATA) {
1114                 struct inode *inode = page->mapping->host;
1115
1116                 if (S_ISDIR(inode->i_mode))
1117                         return CURSEG_HOT_DATA;
1118                 else if (is_cold_data(page) || file_is_cold(inode))
1119                         return CURSEG_COLD_DATA;
1120                 else
1121                         return CURSEG_WARM_DATA;
1122         } else {
1123                 if (IS_DNODE(page))
1124                         return is_cold_node(page) ? CURSEG_WARM_NODE :
1125                                                 CURSEG_HOT_NODE;
1126                 else
1127                         return CURSEG_COLD_NODE;
1128         }
1129 }
1130
1131 static int __get_segment_type(struct page *page, enum page_type p_type)
1132 {
1133         switch (F2FS_P_SB(page)->active_logs) {
1134         case 2:
1135                 return __get_segment_type_2(page, p_type);
1136         case 4:
1137                 return __get_segment_type_4(page, p_type);
1138         }
1139         /* NR_CURSEG_TYPE(6) logs by default */
1140         f2fs_bug_on(F2FS_P_SB(page),
1141                 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1142         return __get_segment_type_6(page, p_type);
1143 }
1144
1145 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1146                 block_t old_blkaddr, block_t *new_blkaddr,
1147                 struct f2fs_summary *sum, int type)
1148 {
1149         struct sit_info *sit_i = SIT_I(sbi);
1150         struct curseg_info *curseg;
1151
1152         curseg = CURSEG_I(sbi, type);
1153
1154         mutex_lock(&curseg->curseg_mutex);
1155
1156         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1157
1158         /*
1159          * __add_sum_entry should be resided under the curseg_mutex
1160          * because, this function updates a summary entry in the
1161          * current summary block.
1162          */
1163         __add_sum_entry(sbi, type, sum);
1164
1165         mutex_lock(&sit_i->sentry_lock);
1166         __refresh_next_blkoff(sbi, curseg);
1167
1168         stat_inc_block_count(sbi, curseg);
1169
1170         if (!__has_curseg_space(sbi, type))
1171                 sit_i->s_ops->allocate_segment(sbi, type, false);
1172         /*
1173          * SIT information should be updated before segment allocation,
1174          * since SSR needs latest valid block information.
1175          */
1176         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1177
1178         mutex_unlock(&sit_i->sentry_lock);
1179
1180         if (page && IS_NODESEG(type))
1181                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1182
1183         mutex_unlock(&curseg->curseg_mutex);
1184 }
1185
1186 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
1187                         block_t old_blkaddr, block_t *new_blkaddr,
1188                         struct f2fs_summary *sum, struct f2fs_io_info *fio)
1189 {
1190         int type = __get_segment_type(page, fio->type);
1191
1192         allocate_data_block(sbi, page, old_blkaddr, new_blkaddr, sum, type);
1193
1194         /* writeout dirty page into bdev */
1195         f2fs_submit_page_mbio(sbi, page, *new_blkaddr, fio);
1196 }
1197
1198 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1199 {
1200         struct f2fs_io_info fio = {
1201                 .type = META,
1202                 .rw = WRITE_SYNC | REQ_META | REQ_PRIO
1203         };
1204
1205         set_page_writeback(page);
1206         f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1207 }
1208
1209 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
1210                 struct f2fs_io_info *fio,
1211                 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
1212 {
1213         struct f2fs_summary sum;
1214         set_summary(&sum, nid, 0, 0);
1215         do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, fio);
1216 }
1217
1218 void write_data_page(struct page *page, struct dnode_of_data *dn,
1219                 block_t *new_blkaddr, struct f2fs_io_info *fio)
1220 {
1221         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1222         struct f2fs_summary sum;
1223         struct node_info ni;
1224
1225         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1226         get_node_info(sbi, dn->nid, &ni);
1227         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1228
1229         do_write_page(sbi, page, dn->data_blkaddr, new_blkaddr, &sum, fio);
1230 }
1231
1232 void rewrite_data_page(struct page *page, block_t old_blkaddr,
1233                                         struct f2fs_io_info *fio)
1234 {
1235         f2fs_submit_page_mbio(F2FS_P_SB(page), page, old_blkaddr, fio);
1236 }
1237
1238 void recover_data_page(struct f2fs_sb_info *sbi,
1239                         struct page *page, struct f2fs_summary *sum,
1240                         block_t old_blkaddr, block_t new_blkaddr)
1241 {
1242         struct sit_info *sit_i = SIT_I(sbi);
1243         struct curseg_info *curseg;
1244         unsigned int segno, old_cursegno;
1245         struct seg_entry *se;
1246         int type;
1247
1248         segno = GET_SEGNO(sbi, new_blkaddr);
1249         se = get_seg_entry(sbi, segno);
1250         type = se->type;
1251
1252         if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1253                 if (old_blkaddr == NULL_ADDR)
1254                         type = CURSEG_COLD_DATA;
1255                 else
1256                         type = CURSEG_WARM_DATA;
1257         }
1258         curseg = CURSEG_I(sbi, type);
1259
1260         mutex_lock(&curseg->curseg_mutex);
1261         mutex_lock(&sit_i->sentry_lock);
1262
1263         old_cursegno = curseg->segno;
1264
1265         /* change the current segment */
1266         if (segno != curseg->segno) {
1267                 curseg->next_segno = segno;
1268                 change_curseg(sbi, type, true);
1269         }
1270
1271         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1272         __add_sum_entry(sbi, type, sum);
1273
1274         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1275         locate_dirty_segment(sbi, old_cursegno);
1276
1277         mutex_unlock(&sit_i->sentry_lock);
1278         mutex_unlock(&curseg->curseg_mutex);
1279 }
1280
1281 static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1282                                         struct page *page, enum page_type type)
1283 {
1284         enum page_type btype = PAGE_TYPE_OF_BIO(type);
1285         struct f2fs_bio_info *io = &sbi->write_io[btype];
1286         struct bio_vec *bvec;
1287         int i;
1288
1289         down_read(&io->io_rwsem);
1290         if (!io->bio)
1291                 goto out;
1292
1293         bio_for_each_segment_all(bvec, io->bio, i) {
1294                 if (page == bvec->bv_page) {
1295                         up_read(&io->io_rwsem);
1296                         return true;
1297                 }
1298         }
1299
1300 out:
1301         up_read(&io->io_rwsem);
1302         return false;
1303 }
1304
1305 void f2fs_wait_on_page_writeback(struct page *page,
1306                                 enum page_type type)
1307 {
1308         if (PageWriteback(page)) {
1309                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1310
1311                 if (is_merged_page(sbi, page, type))
1312                         f2fs_submit_merged_bio(sbi, type, WRITE);
1313                 wait_on_page_writeback(page);
1314         }
1315 }
1316
1317 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1318 {
1319         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1320         struct curseg_info *seg_i;
1321         unsigned char *kaddr;
1322         struct page *page;
1323         block_t start;
1324         int i, j, offset;
1325
1326         start = start_sum_block(sbi);
1327
1328         page = get_meta_page(sbi, start++);
1329         kaddr = (unsigned char *)page_address(page);
1330
1331         /* Step 1: restore nat cache */
1332         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1333         memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1334
1335         /* Step 2: restore sit cache */
1336         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1337         memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1338                                                 SUM_JOURNAL_SIZE);
1339         offset = 2 * SUM_JOURNAL_SIZE;
1340
1341         /* Step 3: restore summary entries */
1342         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1343                 unsigned short blk_off;
1344                 unsigned int segno;
1345
1346                 seg_i = CURSEG_I(sbi, i);
1347                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1348                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1349                 seg_i->next_segno = segno;
1350                 reset_curseg(sbi, i, 0);
1351                 seg_i->alloc_type = ckpt->alloc_type[i];
1352                 seg_i->next_blkoff = blk_off;
1353
1354                 if (seg_i->alloc_type == SSR)
1355                         blk_off = sbi->blocks_per_seg;
1356
1357                 for (j = 0; j < blk_off; j++) {
1358                         struct f2fs_summary *s;
1359                         s = (struct f2fs_summary *)(kaddr + offset);
1360                         seg_i->sum_blk->entries[j] = *s;
1361                         offset += SUMMARY_SIZE;
1362                         if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1363                                                 SUM_FOOTER_SIZE)
1364                                 continue;
1365
1366                         f2fs_put_page(page, 1);
1367                         page = NULL;
1368
1369                         page = get_meta_page(sbi, start++);
1370                         kaddr = (unsigned char *)page_address(page);
1371                         offset = 0;
1372                 }
1373         }
1374         f2fs_put_page(page, 1);
1375         return 0;
1376 }
1377
1378 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1379 {
1380         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1381         struct f2fs_summary_block *sum;
1382         struct curseg_info *curseg;
1383         struct page *new;
1384         unsigned short blk_off;
1385         unsigned int segno = 0;
1386         block_t blk_addr = 0;
1387
1388         /* get segment number and block addr */
1389         if (IS_DATASEG(type)) {
1390                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1391                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1392                                                         CURSEG_HOT_DATA]);
1393                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1394                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1395                 else
1396                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1397         } else {
1398                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1399                                                         CURSEG_HOT_NODE]);
1400                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1401                                                         CURSEG_HOT_NODE]);
1402                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1403                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1404                                                         type - CURSEG_HOT_NODE);
1405                 else
1406                         blk_addr = GET_SUM_BLOCK(sbi, segno);
1407         }
1408
1409         new = get_meta_page(sbi, blk_addr);
1410         sum = (struct f2fs_summary_block *)page_address(new);
1411
1412         if (IS_NODESEG(type)) {
1413                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1414                         struct f2fs_summary *ns = &sum->entries[0];
1415                         int i;
1416                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1417                                 ns->version = 0;
1418                                 ns->ofs_in_node = 0;
1419                         }
1420                 } else {
1421                         int err;
1422
1423                         err = restore_node_summary(sbi, segno, sum);
1424                         if (err) {
1425                                 f2fs_put_page(new, 1);
1426                                 return err;
1427                         }
1428                 }
1429         }
1430
1431         /* set uncompleted segment to curseg */
1432         curseg = CURSEG_I(sbi, type);
1433         mutex_lock(&curseg->curseg_mutex);
1434         memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1435         curseg->next_segno = segno;
1436         reset_curseg(sbi, type, 0);
1437         curseg->alloc_type = ckpt->alloc_type[type];
1438         curseg->next_blkoff = blk_off;
1439         mutex_unlock(&curseg->curseg_mutex);
1440         f2fs_put_page(new, 1);
1441         return 0;
1442 }
1443
1444 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1445 {
1446         int type = CURSEG_HOT_DATA;
1447         int err;
1448
1449         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1450                 /* restore for compacted data summary */
1451                 if (read_compacted_summaries(sbi))
1452                         return -EINVAL;
1453                 type = CURSEG_HOT_NODE;
1454         }
1455
1456         for (; type <= CURSEG_COLD_NODE; type++) {
1457                 err = read_normal_summaries(sbi, type);
1458                 if (err)
1459                         return err;
1460         }
1461
1462         return 0;
1463 }
1464
1465 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1466 {
1467         struct page *page;
1468         unsigned char *kaddr;
1469         struct f2fs_summary *summary;
1470         struct curseg_info *seg_i;
1471         int written_size = 0;
1472         int i, j;
1473
1474         page = grab_meta_page(sbi, blkaddr++);
1475         kaddr = (unsigned char *)page_address(page);
1476
1477         /* Step 1: write nat cache */
1478         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1479         memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1480         written_size += SUM_JOURNAL_SIZE;
1481
1482         /* Step 2: write sit cache */
1483         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1484         memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1485                                                 SUM_JOURNAL_SIZE);
1486         written_size += SUM_JOURNAL_SIZE;
1487
1488         /* Step 3: write summary entries */
1489         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1490                 unsigned short blkoff;
1491                 seg_i = CURSEG_I(sbi, i);
1492                 if (sbi->ckpt->alloc_type[i] == SSR)
1493                         blkoff = sbi->blocks_per_seg;
1494                 else
1495                         blkoff = curseg_blkoff(sbi, i);
1496
1497                 for (j = 0; j < blkoff; j++) {
1498                         if (!page) {
1499                                 page = grab_meta_page(sbi, blkaddr++);
1500                                 kaddr = (unsigned char *)page_address(page);
1501                                 written_size = 0;
1502                         }
1503                         summary = (struct f2fs_summary *)(kaddr + written_size);
1504                         *summary = seg_i->sum_blk->entries[j];
1505                         written_size += SUMMARY_SIZE;
1506
1507                         if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1508                                                         SUM_FOOTER_SIZE)
1509                                 continue;
1510
1511                         set_page_dirty(page);
1512                         f2fs_put_page(page, 1);
1513                         page = NULL;
1514                 }
1515         }
1516         if (page) {
1517                 set_page_dirty(page);
1518                 f2fs_put_page(page, 1);
1519         }
1520 }
1521
1522 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1523                                         block_t blkaddr, int type)
1524 {
1525         int i, end;
1526         if (IS_DATASEG(type))
1527                 end = type + NR_CURSEG_DATA_TYPE;
1528         else
1529                 end = type + NR_CURSEG_NODE_TYPE;
1530
1531         for (i = type; i < end; i++) {
1532                 struct curseg_info *sum = CURSEG_I(sbi, i);
1533                 mutex_lock(&sum->curseg_mutex);
1534                 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1535                 mutex_unlock(&sum->curseg_mutex);
1536         }
1537 }
1538
1539 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1540 {
1541         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1542                 write_compacted_summaries(sbi, start_blk);
1543         else
1544                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1545 }
1546
1547 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1548 {
1549         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1550                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1551 }
1552
1553 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1554                                         unsigned int val, int alloc)
1555 {
1556         int i;
1557
1558         if (type == NAT_JOURNAL) {
1559                 for (i = 0; i < nats_in_cursum(sum); i++) {
1560                         if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1561                                 return i;
1562                 }
1563                 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1564                         return update_nats_in_cursum(sum, 1);
1565         } else if (type == SIT_JOURNAL) {
1566                 for (i = 0; i < sits_in_cursum(sum); i++)
1567                         if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1568                                 return i;
1569                 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1570                         return update_sits_in_cursum(sum, 1);
1571         }
1572         return -1;
1573 }
1574
1575 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1576                                         unsigned int segno)
1577 {
1578         return get_meta_page(sbi, current_sit_addr(sbi, segno));
1579 }
1580
1581 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1582                                         unsigned int start)
1583 {
1584         struct sit_info *sit_i = SIT_I(sbi);
1585         struct page *src_page, *dst_page;
1586         pgoff_t src_off, dst_off;
1587         void *src_addr, *dst_addr;
1588
1589         src_off = current_sit_addr(sbi, start);
1590         dst_off = next_sit_addr(sbi, src_off);
1591
1592         /* get current sit block page without lock */
1593         src_page = get_meta_page(sbi, src_off);
1594         dst_page = grab_meta_page(sbi, dst_off);
1595         f2fs_bug_on(sbi, PageDirty(src_page));
1596
1597         src_addr = page_address(src_page);
1598         dst_addr = page_address(dst_page);
1599         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1600
1601         set_page_dirty(dst_page);
1602         f2fs_put_page(src_page, 1);
1603
1604         set_to_next_sit(sit_i, start);
1605
1606         return dst_page;
1607 }
1608
1609 static struct sit_entry_set *grab_sit_entry_set(void)
1610 {
1611         struct sit_entry_set *ses =
1612                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_ATOMIC);
1613
1614         ses->entry_cnt = 0;
1615         INIT_LIST_HEAD(&ses->set_list);
1616         return ses;
1617 }
1618
1619 static void release_sit_entry_set(struct sit_entry_set *ses)
1620 {
1621         list_del(&ses->set_list);
1622         kmem_cache_free(sit_entry_set_slab, ses);
1623 }
1624
1625 static void adjust_sit_entry_set(struct sit_entry_set *ses,
1626                                                 struct list_head *head)
1627 {
1628         struct sit_entry_set *next = ses;
1629
1630         if (list_is_last(&ses->set_list, head))
1631                 return;
1632
1633         list_for_each_entry_continue(next, head, set_list)
1634                 if (ses->entry_cnt <= next->entry_cnt)
1635                         break;
1636
1637         list_move_tail(&ses->set_list, &next->set_list);
1638 }
1639
1640 static void add_sit_entry(unsigned int segno, struct list_head *head)
1641 {
1642         struct sit_entry_set *ses;
1643         unsigned int start_segno = START_SEGNO(segno);
1644
1645         list_for_each_entry(ses, head, set_list) {
1646                 if (ses->start_segno == start_segno) {
1647                         ses->entry_cnt++;
1648                         adjust_sit_entry_set(ses, head);
1649                         return;
1650                 }
1651         }
1652
1653         ses = grab_sit_entry_set();
1654
1655         ses->start_segno = start_segno;
1656         ses->entry_cnt++;
1657         list_add(&ses->set_list, head);
1658 }
1659
1660 static void add_sits_in_set(struct f2fs_sb_info *sbi)
1661 {
1662         struct f2fs_sm_info *sm_info = SM_I(sbi);
1663         struct list_head *set_list = &sm_info->sit_entry_set;
1664         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
1665         unsigned int segno;
1666
1667         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
1668                 add_sit_entry(segno, set_list);
1669 }
1670
1671 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
1672 {
1673         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1674         struct f2fs_summary_block *sum = curseg->sum_blk;
1675         int i;
1676
1677         for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1678                 unsigned int segno;
1679                 bool dirtied;
1680
1681                 segno = le32_to_cpu(segno_in_journal(sum, i));
1682                 dirtied = __mark_sit_entry_dirty(sbi, segno);
1683
1684                 if (!dirtied)
1685                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
1686         }
1687         update_sits_in_cursum(sum, -sits_in_cursum(sum));
1688 }
1689
1690 /*
1691  * CP calls this function, which flushes SIT entries including sit_journal,
1692  * and moves prefree segs to free segs.
1693  */
1694 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1695 {
1696         struct sit_info *sit_i = SIT_I(sbi);
1697         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1698         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1699         struct f2fs_summary_block *sum = curseg->sum_blk;
1700         struct sit_entry_set *ses, *tmp;
1701         struct list_head *head = &SM_I(sbi)->sit_entry_set;
1702         bool to_journal = true;
1703         struct seg_entry *se;
1704
1705         mutex_lock(&curseg->curseg_mutex);
1706         mutex_lock(&sit_i->sentry_lock);
1707
1708         /*
1709          * add and account sit entries of dirty bitmap in sit entry
1710          * set temporarily
1711          */
1712         add_sits_in_set(sbi);
1713
1714         /*
1715          * if there are no enough space in journal to store dirty sit
1716          * entries, remove all entries from journal and add and account
1717          * them in sit entry set.
1718          */
1719         if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL))
1720                 remove_sits_in_journal(sbi);
1721
1722         if (!sit_i->dirty_sentries)
1723                 goto out;
1724
1725         /*
1726          * there are two steps to flush sit entries:
1727          * #1, flush sit entries to journal in current cold data summary block.
1728          * #2, flush sit entries to sit page.
1729          */
1730         list_for_each_entry_safe(ses, tmp, head, set_list) {
1731                 struct page *page = NULL;
1732                 struct f2fs_sit_block *raw_sit = NULL;
1733                 unsigned int start_segno = ses->start_segno;
1734                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
1735                                                 (unsigned long)MAIN_SEGS(sbi));
1736                 unsigned int segno = start_segno;
1737
1738                 if (to_journal &&
1739                         !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL))
1740                         to_journal = false;
1741
1742                 if (!to_journal) {
1743                         page = get_next_sit_page(sbi, start_segno);
1744                         raw_sit = page_address(page);
1745                 }
1746
1747                 /* flush dirty sit entries in region of current sit set */
1748                 for_each_set_bit_from(segno, bitmap, end) {
1749                         int offset, sit_offset;
1750
1751                         se = get_seg_entry(sbi, segno);
1752
1753                         /* add discard candidates */
1754                         if (SM_I(sbi)->nr_discards < SM_I(sbi)->max_discards) {
1755                                 cpc->trim_start = segno;
1756                                 add_discard_addrs(sbi, cpc);
1757                         }
1758
1759                         if (to_journal) {
1760                                 offset = lookup_journal_in_cursum(sum,
1761                                                         SIT_JOURNAL, segno, 1);
1762                                 f2fs_bug_on(sbi, offset < 0);
1763                                 segno_in_journal(sum, offset) =
1764                                                         cpu_to_le32(segno);
1765                                 seg_info_to_raw_sit(se,
1766                                                 &sit_in_journal(sum, offset));
1767                         } else {
1768                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1769                                 seg_info_to_raw_sit(se,
1770                                                 &raw_sit->entries[sit_offset]);
1771                         }
1772
1773                         __clear_bit(segno, bitmap);
1774                         sit_i->dirty_sentries--;
1775                         ses->entry_cnt--;
1776                 }
1777
1778                 if (!to_journal)
1779                         f2fs_put_page(page, 1);
1780
1781                 f2fs_bug_on(sbi, ses->entry_cnt);
1782                 release_sit_entry_set(ses);
1783         }
1784
1785         f2fs_bug_on(sbi, !list_empty(head));
1786         f2fs_bug_on(sbi, sit_i->dirty_sentries);
1787 out:
1788         if (cpc->reason == CP_DISCARD) {
1789                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
1790                         add_discard_addrs(sbi, cpc);
1791         }
1792         mutex_unlock(&sit_i->sentry_lock);
1793         mutex_unlock(&curseg->curseg_mutex);
1794
1795         set_prefree_as_free_segments(sbi);
1796 }
1797
1798 static int build_sit_info(struct f2fs_sb_info *sbi)
1799 {
1800         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1801         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1802         struct sit_info *sit_i;
1803         unsigned int sit_segs, start;
1804         char *src_bitmap, *dst_bitmap;
1805         unsigned int bitmap_size;
1806
1807         /* allocate memory for SIT information */
1808         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1809         if (!sit_i)
1810                 return -ENOMEM;
1811
1812         SM_I(sbi)->sit_info = sit_i;
1813
1814         sit_i->sentries = vzalloc(MAIN_SEGS(sbi) * sizeof(struct seg_entry));
1815         if (!sit_i->sentries)
1816                 return -ENOMEM;
1817
1818         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
1819         sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1820         if (!sit_i->dirty_sentries_bitmap)
1821                 return -ENOMEM;
1822
1823         for (start = 0; start < MAIN_SEGS(sbi); start++) {
1824                 sit_i->sentries[start].cur_valid_map
1825                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1826                 sit_i->sentries[start].ckpt_valid_map
1827                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1828                 if (!sit_i->sentries[start].cur_valid_map
1829                                 || !sit_i->sentries[start].ckpt_valid_map)
1830                         return -ENOMEM;
1831         }
1832
1833         if (sbi->segs_per_sec > 1) {
1834                 sit_i->sec_entries = vzalloc(MAIN_SECS(sbi) *
1835                                         sizeof(struct sec_entry));
1836                 if (!sit_i->sec_entries)
1837                         return -ENOMEM;
1838         }
1839
1840         /* get information related with SIT */
1841         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1842
1843         /* setup SIT bitmap from ckeckpoint pack */
1844         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1845         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1846
1847         dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1848         if (!dst_bitmap)
1849                 return -ENOMEM;
1850
1851         /* init SIT information */
1852         sit_i->s_ops = &default_salloc_ops;
1853
1854         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1855         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1856         sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1857         sit_i->sit_bitmap = dst_bitmap;
1858         sit_i->bitmap_size = bitmap_size;
1859         sit_i->dirty_sentries = 0;
1860         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1861         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1862         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1863         mutex_init(&sit_i->sentry_lock);
1864         return 0;
1865 }
1866
1867 static int build_free_segmap(struct f2fs_sb_info *sbi)
1868 {
1869         struct free_segmap_info *free_i;
1870         unsigned int bitmap_size, sec_bitmap_size;
1871
1872         /* allocate memory for free segmap information */
1873         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1874         if (!free_i)
1875                 return -ENOMEM;
1876
1877         SM_I(sbi)->free_info = free_i;
1878
1879         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
1880         free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1881         if (!free_i->free_segmap)
1882                 return -ENOMEM;
1883
1884         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
1885         free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1886         if (!free_i->free_secmap)
1887                 return -ENOMEM;
1888
1889         /* set all segments as dirty temporarily */
1890         memset(free_i->free_segmap, 0xff, bitmap_size);
1891         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1892
1893         /* init free segmap information */
1894         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
1895         free_i->free_segments = 0;
1896         free_i->free_sections = 0;
1897         rwlock_init(&free_i->segmap_lock);
1898         return 0;
1899 }
1900
1901 static int build_curseg(struct f2fs_sb_info *sbi)
1902 {
1903         struct curseg_info *array;
1904         int i;
1905
1906         array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
1907         if (!array)
1908                 return -ENOMEM;
1909
1910         SM_I(sbi)->curseg_array = array;
1911
1912         for (i = 0; i < NR_CURSEG_TYPE; i++) {
1913                 mutex_init(&array[i].curseg_mutex);
1914                 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1915                 if (!array[i].sum_blk)
1916                         return -ENOMEM;
1917                 array[i].segno = NULL_SEGNO;
1918                 array[i].next_blkoff = 0;
1919         }
1920         return restore_curseg_summaries(sbi);
1921 }
1922
1923 static void build_sit_entries(struct f2fs_sb_info *sbi)
1924 {
1925         struct sit_info *sit_i = SIT_I(sbi);
1926         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1927         struct f2fs_summary_block *sum = curseg->sum_blk;
1928         int sit_blk_cnt = SIT_BLK_CNT(sbi);
1929         unsigned int i, start, end;
1930         unsigned int readed, start_blk = 0;
1931         int nrpages = MAX_BIO_BLOCKS(sbi);
1932
1933         do {
1934                 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT);
1935
1936                 start = start_blk * sit_i->sents_per_block;
1937                 end = (start_blk + readed) * sit_i->sents_per_block;
1938
1939                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
1940                         struct seg_entry *se = &sit_i->sentries[start];
1941                         struct f2fs_sit_block *sit_blk;
1942                         struct f2fs_sit_entry sit;
1943                         struct page *page;
1944
1945                         mutex_lock(&curseg->curseg_mutex);
1946                         for (i = 0; i < sits_in_cursum(sum); i++) {
1947                                 if (le32_to_cpu(segno_in_journal(sum, i))
1948                                                                 == start) {
1949                                         sit = sit_in_journal(sum, i);
1950                                         mutex_unlock(&curseg->curseg_mutex);
1951                                         goto got_it;
1952                                 }
1953                         }
1954                         mutex_unlock(&curseg->curseg_mutex);
1955
1956                         page = get_current_sit_page(sbi, start);
1957                         sit_blk = (struct f2fs_sit_block *)page_address(page);
1958                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1959                         f2fs_put_page(page, 1);
1960 got_it:
1961                         check_block_count(sbi, start, &sit);
1962                         seg_info_from_raw_sit(se, &sit);
1963                         if (sbi->segs_per_sec > 1) {
1964                                 struct sec_entry *e = get_sec_entry(sbi, start);
1965                                 e->valid_blocks += se->valid_blocks;
1966                         }
1967                 }
1968                 start_blk += readed;
1969         } while (start_blk < sit_blk_cnt);
1970 }
1971
1972 static void init_free_segmap(struct f2fs_sb_info *sbi)
1973 {
1974         unsigned int start;
1975         int type;
1976
1977         for (start = 0; start < MAIN_SEGS(sbi); start++) {
1978                 struct seg_entry *sentry = get_seg_entry(sbi, start);
1979                 if (!sentry->valid_blocks)
1980                         __set_free(sbi, start);
1981         }
1982
1983         /* set use the current segments */
1984         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1985                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1986                 __set_test_and_inuse(sbi, curseg_t->segno);
1987         }
1988 }
1989
1990 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1991 {
1992         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1993         struct free_segmap_info *free_i = FREE_I(sbi);
1994         unsigned int segno = 0, offset = 0;
1995         unsigned short valid_blocks;
1996
1997         while (1) {
1998                 /* find dirty segment based on free segmap */
1999                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2000                 if (segno >= MAIN_SEGS(sbi))
2001                         break;
2002                 offset = segno + 1;
2003                 valid_blocks = get_valid_blocks(sbi, segno, 0);
2004                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2005                         continue;
2006                 if (valid_blocks > sbi->blocks_per_seg) {
2007                         f2fs_bug_on(sbi, 1);
2008                         continue;
2009                 }
2010                 mutex_lock(&dirty_i->seglist_lock);
2011                 __locate_dirty_segment(sbi, segno, DIRTY);
2012                 mutex_unlock(&dirty_i->seglist_lock);
2013         }
2014 }
2015
2016 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2017 {
2018         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2019         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2020
2021         dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
2022         if (!dirty_i->victim_secmap)
2023                 return -ENOMEM;
2024         return 0;
2025 }
2026
2027 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2028 {
2029         struct dirty_seglist_info *dirty_i;
2030         unsigned int bitmap_size, i;
2031
2032         /* allocate memory for dirty segments list information */
2033         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2034         if (!dirty_i)
2035                 return -ENOMEM;
2036
2037         SM_I(sbi)->dirty_info = dirty_i;
2038         mutex_init(&dirty_i->seglist_lock);
2039
2040         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2041
2042         for (i = 0; i < NR_DIRTY_TYPE; i++) {
2043                 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
2044                 if (!dirty_i->dirty_segmap[i])
2045                         return -ENOMEM;
2046         }
2047
2048         init_dirty_segmap(sbi);
2049         return init_victim_secmap(sbi);
2050 }
2051
2052 /*
2053  * Update min, max modified time for cost-benefit GC algorithm
2054  */
2055 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2056 {
2057         struct sit_info *sit_i = SIT_I(sbi);
2058         unsigned int segno;
2059
2060         mutex_lock(&sit_i->sentry_lock);
2061
2062         sit_i->min_mtime = LLONG_MAX;
2063
2064         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2065                 unsigned int i;
2066                 unsigned long long mtime = 0;
2067
2068                 for (i = 0; i < sbi->segs_per_sec; i++)
2069                         mtime += get_seg_entry(sbi, segno + i)->mtime;
2070
2071                 mtime = div_u64(mtime, sbi->segs_per_sec);
2072
2073                 if (sit_i->min_mtime > mtime)
2074                         sit_i->min_mtime = mtime;
2075         }
2076         sit_i->max_mtime = get_mtime(sbi);
2077         mutex_unlock(&sit_i->sentry_lock);
2078 }
2079
2080 int build_segment_manager(struct f2fs_sb_info *sbi)
2081 {
2082         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2083         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2084         struct f2fs_sm_info *sm_info;
2085         int err;
2086
2087         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2088         if (!sm_info)
2089                 return -ENOMEM;
2090
2091         /* init sm info */
2092         sbi->sm_info = sm_info;
2093         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2094         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2095         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2096         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2097         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2098         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2099         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2100         sm_info->rec_prefree_segments = sm_info->main_segments *
2101                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2102         sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2103         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2104         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2105
2106         INIT_LIST_HEAD(&sm_info->discard_list);
2107         sm_info->nr_discards = 0;
2108         sm_info->max_discards = 0;
2109
2110         INIT_LIST_HEAD(&sm_info->sit_entry_set);
2111
2112         if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2113                 err = create_flush_cmd_control(sbi);
2114                 if (err)
2115                         return err;
2116         }
2117
2118         err = build_sit_info(sbi);
2119         if (err)
2120                 return err;
2121         err = build_free_segmap(sbi);
2122         if (err)
2123                 return err;
2124         err = build_curseg(sbi);
2125         if (err)
2126                 return err;
2127
2128         /* reinit free segmap based on SIT */
2129         build_sit_entries(sbi);
2130
2131         init_free_segmap(sbi);
2132         err = build_dirty_segmap(sbi);
2133         if (err)
2134                 return err;
2135
2136         init_min_max_mtime(sbi);
2137         return 0;
2138 }
2139
2140 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2141                 enum dirty_type dirty_type)
2142 {
2143         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2144
2145         mutex_lock(&dirty_i->seglist_lock);
2146         kfree(dirty_i->dirty_segmap[dirty_type]);
2147         dirty_i->nr_dirty[dirty_type] = 0;
2148         mutex_unlock(&dirty_i->seglist_lock);
2149 }
2150
2151 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2152 {
2153         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2154         kfree(dirty_i->victim_secmap);
2155 }
2156
2157 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2158 {
2159         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2160         int i;
2161
2162         if (!dirty_i)
2163                 return;
2164
2165         /* discard pre-free/dirty segments list */
2166         for (i = 0; i < NR_DIRTY_TYPE; i++)
2167                 discard_dirty_segmap(sbi, i);
2168
2169         destroy_victim_secmap(sbi);
2170         SM_I(sbi)->dirty_info = NULL;
2171         kfree(dirty_i);
2172 }
2173
2174 static void destroy_curseg(struct f2fs_sb_info *sbi)
2175 {
2176         struct curseg_info *array = SM_I(sbi)->curseg_array;
2177         int i;
2178
2179         if (!array)
2180                 return;
2181         SM_I(sbi)->curseg_array = NULL;
2182         for (i = 0; i < NR_CURSEG_TYPE; i++)
2183                 kfree(array[i].sum_blk);
2184         kfree(array);
2185 }
2186
2187 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2188 {
2189         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2190         if (!free_i)
2191                 return;
2192         SM_I(sbi)->free_info = NULL;
2193         kfree(free_i->free_segmap);
2194         kfree(free_i->free_secmap);
2195         kfree(free_i);
2196 }
2197
2198 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2199 {
2200         struct sit_info *sit_i = SIT_I(sbi);
2201         unsigned int start;
2202
2203         if (!sit_i)
2204                 return;
2205
2206         if (sit_i->sentries) {
2207                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2208                         kfree(sit_i->sentries[start].cur_valid_map);
2209                         kfree(sit_i->sentries[start].ckpt_valid_map);
2210                 }
2211         }
2212         vfree(sit_i->sentries);
2213         vfree(sit_i->sec_entries);
2214         kfree(sit_i->dirty_sentries_bitmap);
2215
2216         SM_I(sbi)->sit_info = NULL;
2217         kfree(sit_i->sit_bitmap);
2218         kfree(sit_i);
2219 }
2220
2221 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2222 {
2223         struct f2fs_sm_info *sm_info = SM_I(sbi);
2224
2225         if (!sm_info)
2226                 return;
2227         destroy_flush_cmd_control(sbi);
2228         destroy_dirty_segmap(sbi);
2229         destroy_curseg(sbi);
2230         destroy_free_segmap(sbi);
2231         destroy_sit_info(sbi);
2232         sbi->sm_info = NULL;
2233         kfree(sm_info);
2234 }
2235
2236 int __init create_segment_manager_caches(void)
2237 {
2238         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2239                         sizeof(struct discard_entry));
2240         if (!discard_entry_slab)
2241                 goto fail;
2242
2243         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2244                         sizeof(struct sit_entry_set));
2245         if (!sit_entry_set_slab)
2246                 goto destory_discard_entry;
2247
2248         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2249                         sizeof(struct inmem_pages));
2250         if (!inmem_entry_slab)
2251                 goto destroy_sit_entry_set;
2252         return 0;
2253
2254 destroy_sit_entry_set:
2255         kmem_cache_destroy(sit_entry_set_slab);
2256 destory_discard_entry:
2257         kmem_cache_destroy(discard_entry_slab);
2258 fail:
2259         return -ENOMEM;
2260 }
2261
2262 void destroy_segment_manager_caches(void)
2263 {
2264         kmem_cache_destroy(sit_entry_set_slab);
2265         kmem_cache_destroy(discard_entry_slab);
2266         kmem_cache_destroy(inmem_entry_slab);
2267 }