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.h"
  24 #include <trace/events/f2fs.h>
  25
  26 #define __reverse_ffz(x) __reverse_ffs(~(x))
  27
  28 static struct kmem_cache *discard_entry_slab;
  29 static struct kmem_cache *sit_entry_set_slab;
  30 static struct kmem_cache *inmem_entry_slab;
  31
  32 /*
  33  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  34  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  35  */
  36 static inline unsigned long __reverse_ffs(unsigned long word)
  37 {
  38         int num = 0;
  39
  40 #if BITS_PER_LONG == 64
  41         if ((word & 0xffffffff) == 0) {
  42                 num += 32;
  43                 word >>= 32;
  44         }
  45 #endif
  46         if ((word & 0xffff) == 0) {
  47                 num += 16;
  48                 word >>= 16;
  49         }
  50         if ((word & 0xff) == 0) {
  51                 num += 8;
  52                 word >>= 8;
  53         }
  54         if ((word & 0xf0) == 0)
  55                 num += 4;
  56         else
  57                 word >>= 4;
  58         if ((word & 0xc) == 0)
  59                 num += 2;
  60         else
  61                 word >>= 2;
  62         if ((word & 0x2) == 0)
  63                 num += 1;
  64         return num;
  65 }
  66
  67 /*
  68  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  69  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  70  * Example:
  71  *                             LSB <--> MSB
  72  *   f2fs_set_bit(0, bitmap) => 0000 0001
  73  *   f2fs_set_bit(7, bitmap) => 1000 0000
  74  */
  75 static unsigned long __find_rev_next_bit(const unsigned long *addr,
  76                         unsigned long size, unsigned long offset)
  77 {
  78         const unsigned long *p = addr + BIT_WORD(offset);
  79         unsigned long result = offset & ~(BITS_PER_LONG - 1);
  80         unsigned long tmp;
  81         unsigned long mask, submask;
  82         unsigned long quot, rest;
  83
  84         if (offset >= size)
  85                 return size;
  86
  87         size -= result;
  88         offset %= BITS_PER_LONG;
  89         if (!offset)
  90                 goto aligned;
  91
  92         tmp = *(p++);
  93         quot = (offset >> 3) << 3;
  94         rest = offset & 0x7;
  95         mask = ~0UL << quot;
  96         submask = (unsigned char)(0xff << rest) >> rest;
  97         submask <<= quot;
  98         mask &= submask;
  99         tmp &= mask;
 100         if (size < BITS_PER_LONG)
 101                 goto found_first;
 102         if (tmp)
 103                 goto found_middle;
 104
 105         size -= BITS_PER_LONG;
 106         result += BITS_PER_LONG;
 107 aligned:
 108         while (size & ~(BITS_PER_LONG-1)) {
 109                 tmp = *(p++);
 110                 if (tmp)
 111                         goto found_middle;
 112                 result += BITS_PER_LONG;
 113                 size -= BITS_PER_LONG;
 114         }
 115         if (!size)
 116                 return result;
 117         tmp = *p;
 118 found_first:
 119         tmp &= (~0UL >> (BITS_PER_LONG - size));
 120         if (tmp == 0UL)         /* Are any bits set? */
 121                 return result + size;   /* Nope. */
 122 found_middle:
 123         return result + __reverse_ffs(tmp);
 124 }
 125
 126 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 127                         unsigned long size, unsigned long offset)
 128 {
 129         const unsigned long *p = addr + BIT_WORD(offset);
 130         unsigned long result = offset & ~(BITS_PER_LONG - 1);
 131         unsigned long tmp;
 132         unsigned long mask, submask;
 133         unsigned long quot, rest;
 134
 135         if (offset >= size)
 136                 return size;
 137
 138         size -= result;
 139         offset %= BITS_PER_LONG;
 140         if (!offset)
 141                 goto aligned;
 142
 143         tmp = *(p++);
 144         quot = (offset >> 3) << 3;
 145         rest = offset & 0x7;
 146         mask = ~(~0UL << quot);
 147         submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest);
 148         submask <<= quot;
 149         mask += submask;
 150         tmp |= mask;
 151         if (size < BITS_PER_LONG)
 152                 goto found_first;
 153         if (~tmp)
 154                 goto found_middle;
 155
 156         size -= BITS_PER_LONG;
 157         result += BITS_PER_LONG;
 158 aligned:
 159         while (size & ~(BITS_PER_LONG - 1)) {
 160                 tmp = *(p++);
 161                 if (~tmp)
 162                         goto found_middle;
 163                 result += BITS_PER_LONG;
 164                 size -= BITS_PER_LONG;
 165         }
 166         if (!size)
 167                 return result;
 168         tmp = *p;
 169
 170 found_first:
 171         tmp |= ~0UL << size;
 172         if (tmp == ~0UL)        /* Are any bits zero? */
 173                 return result + size;   /* Nope. */
 174 found_middle:
 175         return result + __reverse_ffz(tmp);
 176 }
 177
 178 void register_inmem_page(struct inode *inode, struct page *page)
 179 {
 180         struct f2fs_inode_info *fi = F2FS_I(inode);
 181         struct inmem_pages *new;
 182         int err;
 183
 184         SetPagePrivate(page);
 185         f2fs_trace_pid(page);
 186
 187         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 188
 189         /* add atomic page indices to the list */
 190         new->page = page;
 191         INIT_LIST_HEAD(&new->list);
 192 retry:
 193         /* increase reference count with clean state */
 194         mutex_lock(&fi->inmem_lock);
 195         err = radix_tree_insert(&fi->inmem_root, page->index, new);
 196         if (err == -EEXIST) {
 197                 mutex_unlock(&fi->inmem_lock);
 198                 kmem_cache_free(inmem_entry_slab, new);
 199                 return;
 200         } else if (err) {
 201                 mutex_unlock(&fi->inmem_lock);
 202                 goto retry;
 203         }
 204         get_page(page);
 205         list_add_tail(&new->list, &fi->inmem_pages);
 206         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 207         mutex_unlock(&fi->inmem_lock);
 208 }
 209
 210 void commit_inmem_pages(struct inode *inode, bool abort)
 211 {
 212         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 213         struct f2fs_inode_info *fi = F2FS_I(inode);
 214         struct inmem_pages *cur, *tmp;
 215         bool submit_bio = false;
 216         struct f2fs_io_info fio = {
 217                 .type = DATA,
 218                 .rw = WRITE_SYNC | REQ_PRIO,
 219         };
 220
 221         /*
 222          * The abort is true only when f2fs_evict_inode is called.
 223          * Basically, the f2fs_evict_inode doesn't produce any data writes, so
 224          * that we don't need to call f2fs_balance_fs.
 225          * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this
 226          * inode becomes free by iget_locked in f2fs_iget.
 227          */
 228         if (!abort) {
 229                 f2fs_balance_fs(sbi);
 230                 f2fs_lock_op(sbi);
 231         }
 232
 233         mutex_lock(&fi->inmem_lock);
 234         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 235                 if (!abort) {
 236                         lock_page(cur->page);
 237                         if (cur->page->mapping == inode->i_mapping) {
 238                                 f2fs_wait_on_page_writeback(cur->page, DATA);
 239                                 if (clear_page_dirty_for_io(cur->page))
 240                                         inode_dec_dirty_pages(inode);
 241                                 do_write_data_page(cur->page, &fio);
 242                                 submit_bio = true;
 243                         }
 244                         f2fs_put_page(cur->page, 1);
 245                 } else {
 246                         put_page(cur->page);
 247                 }
 248                 radix_tree_delete(&fi->inmem_root, cur->page->index);
 249                 list_del(&cur->list);
 250                 kmem_cache_free(inmem_entry_slab, cur);
 251                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 252         }
 253         mutex_unlock(&fi->inmem_lock);
 254
 255         if (!abort) {
 256                 f2fs_unlock_op(sbi);
 257                 if (submit_bio)
 258                         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 259         }
 260 }
 261
 262 /*
 263  * This function balances dirty node and dentry pages.
 264  * In addition, it controls garbage collection.
 265  */
 266 void f2fs_balance_fs(struct f2fs_sb_info *sbi)
 267 {
 268         /*
 269          * We should do GC or end up with checkpoint, if there are so many dirty
 270          * dir/node pages without enough free segments.
 271          */
 272         if (has_not_enough_free_secs(sbi, 0)) {
 273                 mutex_lock(&sbi->gc_mutex);
 274                 f2fs_gc(sbi);
 275         }
 276 }
 277
 278 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 279 {
 280         /* check the # of cached NAT entries and prefree segments */
 281         if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) ||
 282                         excess_prefree_segs(sbi) ||
 283                         !available_free_memory(sbi, INO_ENTRIES))
 284                 f2fs_sync_fs(sbi->sb, true);
 285 }
 286
 287 static int issue_flush_thread(void *data)
 288 {
 289         struct f2fs_sb_info *sbi = data;
 290         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 291         wait_queue_head_t *q = &fcc->flush_wait_queue;
 292 repeat:
 293         if (kthread_should_stop())
 294                 return 0;
 295
 296         if (!llist_empty(&fcc->issue_list)) {
 297                 struct bio *bio = bio_alloc(GFP_NOIO, 0);
 298                 struct flush_cmd *cmd, *next;
 299                 int ret;
 300
 301                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 302                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 303
 304                 bio->bi_bdev = sbi->sb->s_bdev;
 305                 ret = submit_bio_wait(WRITE_FLUSH, bio);
 306
 307                 llist_for_each_entry_safe(cmd, next,
 308                                           fcc->dispatch_list, llnode) {
 309                         cmd->ret = ret;
 310                         complete(&cmd->wait);
 311                 }
 312                 bio_put(bio);
 313                 fcc->dispatch_list = NULL;
 314         }
 315
 316         wait_event_interruptible(*q,
 317                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 318         goto repeat;
 319 }
 320
 321 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
 322 {
 323         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 324         struct flush_cmd cmd;
 325
 326         trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
 327                                         test_opt(sbi, FLUSH_MERGE));
 328
 329         if (test_opt(sbi, NOBARRIER))
 330                 return 0;
 331
 332         if (!test_opt(sbi, FLUSH_MERGE))
 333                 return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL);
 334
 335         init_completion(&cmd.wait);
 336
 337         llist_add(&cmd.llnode, &fcc->issue_list);
 338
 339         if (!fcc->dispatch_list)
 340                 wake_up(&fcc->flush_wait_queue);
 341
 342         wait_for_completion(&cmd.wait);
 343
 344         return cmd.ret;
 345 }
 346
 347 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
 348 {
 349         dev_t dev = sbi->sb->s_bdev->bd_dev;
 350         struct flush_cmd_control *fcc;
 351         int err = 0;
 352
 353         fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
 354         if (!fcc)
 355                 return -ENOMEM;
 356         init_waitqueue_head(&fcc->flush_wait_queue);
 357         init_llist_head(&fcc->issue_list);
 358         SM_I(sbi)->cmd_control_info = fcc;
 359         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 360                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 361         if (IS_ERR(fcc->f2fs_issue_flush)) {
 362                 err = PTR_ERR(fcc->f2fs_issue_flush);
 363                 kfree(fcc);
 364                 SM_I(sbi)->cmd_control_info = NULL;
 365                 return err;
 366         }
 367
 368         return err;
 369 }
 370
 371 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
 372 {
 373         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 374
 375         if (fcc && fcc->f2fs_issue_flush)
 376                 kthread_stop(fcc->f2fs_issue_flush);
 377         kfree(fcc);
 378         SM_I(sbi)->cmd_control_info = NULL;
 379 }
 380
 381 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 382                 enum dirty_type dirty_type)
 383 {
 384         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 385
 386         /* need not be added */
 387         if (IS_CURSEG(sbi, segno))
 388                 return;
 389
 390         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 391                 dirty_i->nr_dirty[dirty_type]++;
 392
 393         if (dirty_type == DIRTY) {
 394                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 395                 enum dirty_type t = sentry->type;
 396
 397                 if (unlikely(t >= DIRTY)) {
 398                         f2fs_bug_on(sbi, 1);
 399                         return;
 400                 }
 401                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 402                         dirty_i->nr_dirty[t]++;
 403         }
 404 }
 405
 406 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 407                 enum dirty_type dirty_type)
 408 {
 409         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 410
 411         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 412                 dirty_i->nr_dirty[dirty_type]--;
 413
 414         if (dirty_type == DIRTY) {
 415                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 416                 enum dirty_type t = sentry->type;
 417
 418                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 419                         dirty_i->nr_dirty[t]--;
 420
 421                 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
 422                         clear_bit(GET_SECNO(sbi, segno),
 423                                                 dirty_i->victim_secmap);
 424         }
 425 }
 426
 427 /*
 428  * Should not occur error such as -ENOMEM.
 429  * Adding dirty entry into seglist is not critical operation.
 430  * If a given segment is one of current working segments, it won't be added.
 431  */
 432 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 433 {
 434         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 435         unsigned short valid_blocks;
 436
 437         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 438                 return;
 439
 440         mutex_lock(&dirty_i->seglist_lock);
 441
 442         valid_blocks = get_valid_blocks(sbi, segno, 0);
 443
 444         if (valid_blocks == 0) {
 445                 __locate_dirty_segment(sbi, segno, PRE);
 446                 __remove_dirty_segment(sbi, segno, DIRTY);
 447         } else if (valid_blocks < sbi->blocks_per_seg) {
 448                 __locate_dirty_segment(sbi, segno, DIRTY);
 449         } else {
 450                 /* Recovery routine with SSR needs this */
 451                 __remove_dirty_segment(sbi, segno, DIRTY);
 452         }
 453
 454         mutex_unlock(&dirty_i->seglist_lock);
 455 }
 456
 457 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
 458                                 block_t blkstart, block_t blklen)
 459 {
 460         sector_t start = SECTOR_FROM_BLOCK(blkstart);
 461         sector_t len = SECTOR_FROM_BLOCK(blklen);
 462         trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
 463         return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
 464 }
 465
 466 void discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
 467 {
 468         if (f2fs_issue_discard(sbi, blkaddr, 1)) {
 469                 struct page *page = grab_meta_page(sbi, blkaddr);
 470                 /* zero-filled page */
 471                 set_page_dirty(page);
 472                 f2fs_put_page(page, 1);
 473         }
 474 }
 475
 476 static void __add_discard_entry(struct f2fs_sb_info *sbi,
 477                 struct cp_control *cpc, unsigned int start, unsigned int end)
 478 {
 479         struct list_head *head = &SM_I(sbi)->discard_list;
 480         struct discard_entry *new, *last;
 481
 482         if (!list_empty(head)) {
 483                 last = list_last_entry(head, struct discard_entry, list);
 484                 if (START_BLOCK(sbi, cpc->trim_start) + start ==
 485                                                 last->blkaddr + last->len) {
 486                         last->len += end - start;
 487                         goto done;
 488                 }
 489         }
 490
 491         new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
 492         INIT_LIST_HEAD(&new->list);
 493         new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
 494         new->len = end - start;
 495         list_add_tail(&new->list, head);
 496 done:
 497         SM_I(sbi)->nr_discards += end - start;
 498         cpc->trimmed += end - start;
 499 }
 500
 501 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 502 {
 503         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
 504         int max_blocks = sbi->blocks_per_seg;
 505         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
 506         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
 507         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
 508         unsigned long dmap[entries];
 509         unsigned int start = 0, end = -1;
 510         bool force = (cpc->reason == CP_DISCARD);
 511         int i;
 512
 513         if (!force && (!test_opt(sbi, DISCARD) ||
 514                         SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards))
 515                 return;
 516
 517         if (force && !se->valid_blocks) {
 518                 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 519                 /*
 520                  * if this segment is registered in the prefree list, then
 521                  * we should skip adding a discard candidate, and let the
 522                  * checkpoint do that later.
 523                  */
 524                 mutex_lock(&dirty_i->seglist_lock);
 525                 if (test_bit(cpc->trim_start, dirty_i->dirty_segmap[PRE])) {
 526                         mutex_unlock(&dirty_i->seglist_lock);
 527                         cpc->trimmed += sbi->blocks_per_seg;
 528                         return;
 529                 }
 530                 mutex_unlock(&dirty_i->seglist_lock);
 531
 532                 __add_discard_entry(sbi, cpc, 0, sbi->blocks_per_seg);
 533                 return;
 534         }
 535
 536         /* zero block will be discarded through the prefree list */
 537         if (!se->valid_blocks || se->valid_blocks == max_blocks)
 538                 return;
 539
 540         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
 541         for (i = 0; i < entries; i++)
 542                 dmap[i] = force ? ~ckpt_map[i] :
 543                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
 544
 545         while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
 546                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
 547                 if (start >= max_blocks)
 548                         break;
 549
 550                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
 551
 552                 if (end - start < cpc->trim_minlen)
 553                         continue;
 554
 555                 __add_discard_entry(sbi, cpc, start, end);
 556         }
 557 }
 558
 559 void release_discard_addrs(struct f2fs_sb_info *sbi)
 560 {
 561         struct list_head *head = &(SM_I(sbi)->discard_list);
 562         struct discard_entry *entry, *this;
 563
 564         /* drop caches */
 565         list_for_each_entry_safe(entry, this, head, list) {
 566                 list_del(&entry->list);
 567                 kmem_cache_free(discard_entry_slab, entry);
 568         }
 569 }
 570
 571 /*
 572  * Should call clear_prefree_segments after checkpoint is done.
 573  */
 574 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
 575 {
 576         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 577         unsigned int segno;
 578
 579         mutex_lock(&dirty_i->seglist_lock);
 580         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
 581                 __set_test_and_free(sbi, segno);
 582         mutex_unlock(&dirty_i->seglist_lock);
 583 }
 584
 585 void clear_prefree_segments(struct f2fs_sb_info *sbi)
 586 {
 587         struct list_head *head = &(SM_I(sbi)->discard_list);
 588         struct discard_entry *entry, *this;
 589         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 590         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
 591         unsigned int start = 0, end = -1;
 592
 593         mutex_lock(&dirty_i->seglist_lock);
 594
 595         while (1) {
 596                 int i;
 597                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
 598                 if (start >= MAIN_SEGS(sbi))
 599                         break;
 600                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
 601                                                                 start + 1);
 602
 603                 for (i = start; i < end; i++)
 604                         clear_bit(i, prefree_map);
 605
 606                 dirty_i->nr_dirty[PRE] -= end - start;
 607
 608                 if (!test_opt(sbi, DISCARD))
 609                         continue;
 610
 611                 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
 612                                 (end - start) << sbi->log_blocks_per_seg);
 613         }
 614         mutex_unlock(&dirty_i->seglist_lock);
 615
 616         /* send small discards */
 617         list_for_each_entry_safe(entry, this, head, list) {
 618                 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
 619                 list_del(&entry->list);
 620                 SM_I(sbi)->nr_discards -= entry->len;
 621                 kmem_cache_free(discard_entry_slab, entry);
 622         }
 623 }
 624
 625 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
 626 {
 627         struct sit_info *sit_i = SIT_I(sbi);
 628
 629         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
 630                 sit_i->dirty_sentries++;
 631                 return false;
 632         }
 633
 634         return true;
 635 }
 636
 637 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
 638                                         unsigned int segno, int modified)
 639 {
 640         struct seg_entry *se = get_seg_entry(sbi, segno);
 641         se->type = type;
 642         if (modified)
 643                 __mark_sit_entry_dirty(sbi, segno);
 644 }
 645
 646 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
 647 {
 648         struct seg_entry *se;
 649         unsigned int segno, offset;
 650         long int new_vblocks;
 651
 652         segno = GET_SEGNO(sbi, blkaddr);
 653
 654         se = get_seg_entry(sbi, segno);
 655         new_vblocks = se->valid_blocks + del;
 656         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 657
 658         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
 659                                 (new_vblocks > sbi->blocks_per_seg)));
 660
 661         se->valid_blocks = new_vblocks;
 662         se->mtime = get_mtime(sbi);
 663         SIT_I(sbi)->max_mtime = se->mtime;
 664
 665         /* Update valid block bitmap */
 666         if (del > 0) {
 667                 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
 668                         f2fs_bug_on(sbi, 1);
 669         } else {
 670                 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
 671                         f2fs_bug_on(sbi, 1);
 672         }
 673         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
 674                 se->ckpt_valid_blocks += del;
 675
 676         __mark_sit_entry_dirty(sbi, segno);
 677
 678         /* update total number of valid blocks to be written in ckpt area */
 679         SIT_I(sbi)->written_valid_blocks += del;
 680
 681         if (sbi->segs_per_sec > 1)
 682                 get_sec_entry(sbi, segno)->valid_blocks += del;
 683 }
 684
 685 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
 686 {
 687         update_sit_entry(sbi, new, 1);
 688         if (GET_SEGNO(sbi, old) != NULL_SEGNO)
 689                 update_sit_entry(sbi, old, -1);
 690
 691         locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
 692         locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
 693 }
 694
 695 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
 696 {
 697         unsigned int segno = GET_SEGNO(sbi, addr);
 698         struct sit_info *sit_i = SIT_I(sbi);
 699
 700         f2fs_bug_on(sbi, addr == NULL_ADDR);
 701         if (addr == NEW_ADDR)
 702                 return;
 703
 704         /* add it into sit main buffer */
 705         mutex_lock(&sit_i->sentry_lock);
 706
 707         update_sit_entry(sbi, addr, -1);
 708
 709         /* add it into dirty seglist */
 710         locate_dirty_segment(sbi, segno);
 711
 712         mutex_unlock(&sit_i->sentry_lock);
 713 }
 714
 715 /*
 716  * This function should be resided under the curseg_mutex lock
 717  */
 718 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
 719                                         struct f2fs_summary *sum)
 720 {
 721         struct curseg_info *curseg = CURSEG_I(sbi, type);
 722         void *addr = curseg->sum_blk;
 723         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
 724         memcpy(addr, sum, sizeof(struct f2fs_summary));
 725 }
 726
 727 /*
 728  * Calculate the number of current summary pages for writing
 729  */
 730 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
 731 {
 732         int valid_sum_count = 0;
 733         int i, sum_in_page;
 734
 735         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 736                 if (sbi->ckpt->alloc_type[i] == SSR)
 737                         valid_sum_count += sbi->blocks_per_seg;
 738                 else {
 739                         if (for_ra)
 740                                 valid_sum_count += le16_to_cpu(
 741                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
 742                         else
 743                                 valid_sum_count += curseg_blkoff(sbi, i);
 744                 }
 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 static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1028 {
1029         struct curseg_info *curseg = CURSEG_I(sbi, type);
1030         unsigned int old_segno;
1031
1032         old_segno = curseg->segno;
1033         SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1034         locate_dirty_segment(sbi, old_segno);
1035 }
1036
1037 void allocate_new_segments(struct f2fs_sb_info *sbi)
1038 {
1039         int i;
1040
1041         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1042                 __allocate_new_segments(sbi, i);
1043 }
1044
1045 static const struct segment_allocation default_salloc_ops = {
1046         .allocate_segment = allocate_segment_by_default,
1047 };
1048
1049 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1050 {
1051         __u64 start = range->start >> sbi->log_blocksize;
1052         __u64 end = start + (range->len >> sbi->log_blocksize) - 1;
1053         unsigned int start_segno, end_segno;
1054         struct cp_control cpc;
1055
1056         if (range->minlen > SEGMENT_SIZE(sbi) || start >= MAX_BLKADDR(sbi) ||
1057                                                 range->len < sbi->blocksize)
1058                 return -EINVAL;
1059
1060         cpc.trimmed = 0;
1061         if (end <= MAIN_BLKADDR(sbi))
1062                 goto out;
1063
1064         /* start/end segment number in main_area */
1065         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1066         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1067                                                 GET_SEGNO(sbi, end);
1068         cpc.reason = CP_DISCARD;
1069         cpc.trim_start = start_segno;
1070         cpc.trim_end = end_segno;
1071         cpc.trim_minlen = range->minlen >> sbi->log_blocksize;
1072
1073         /* do checkpoint to issue discard commands safely */
1074         mutex_lock(&sbi->gc_mutex);
1075         write_checkpoint(sbi, &cpc);
1076         mutex_unlock(&sbi->gc_mutex);
1077 out:
1078         range->len = cpc.trimmed << sbi->log_blocksize;
1079         return 0;
1080 }
1081
1082 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1083 {
1084         struct curseg_info *curseg = CURSEG_I(sbi, type);
1085         if (curseg->next_blkoff < sbi->blocks_per_seg)
1086                 return true;
1087         return false;
1088 }
1089
1090 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1091 {
1092         if (p_type == DATA)
1093                 return CURSEG_HOT_DATA;
1094         else
1095                 return CURSEG_HOT_NODE;
1096 }
1097
1098 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1099 {
1100         if (p_type == DATA) {
1101                 struct inode *inode = page->mapping->host;
1102
1103                 if (S_ISDIR(inode->i_mode))
1104                         return CURSEG_HOT_DATA;
1105                 else
1106                         return CURSEG_COLD_DATA;
1107         } else {
1108                 if (IS_DNODE(page) && is_cold_node(page))
1109                         return CURSEG_WARM_NODE;
1110                 else
1111                         return CURSEG_COLD_NODE;
1112         }
1113 }
1114
1115 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1116 {
1117         if (p_type == DATA) {
1118                 struct inode *inode = page->mapping->host;
1119
1120                 if (S_ISDIR(inode->i_mode))
1121                         return CURSEG_HOT_DATA;
1122                 else if (is_cold_data(page) || file_is_cold(inode))
1123                         return CURSEG_COLD_DATA;
1124                 else
1125                         return CURSEG_WARM_DATA;
1126         } else {
1127                 if (IS_DNODE(page))
1128                         return is_cold_node(page) ? CURSEG_WARM_NODE :
1129                                                 CURSEG_HOT_NODE;
1130                 else
1131                         return CURSEG_COLD_NODE;
1132         }
1133 }
1134
1135 static int __get_segment_type(struct page *page, enum page_type p_type)
1136 {
1137         switch (F2FS_P_SB(page)->active_logs) {
1138         case 2:
1139                 return __get_segment_type_2(page, p_type);
1140         case 4:
1141                 return __get_segment_type_4(page, p_type);
1142         }
1143         /* NR_CURSEG_TYPE(6) logs by default */
1144         f2fs_bug_on(F2FS_P_SB(page),
1145                 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1146         return __get_segment_type_6(page, p_type);
1147 }
1148
1149 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1150                 block_t old_blkaddr, block_t *new_blkaddr,
1151                 struct f2fs_summary *sum, int type)
1152 {
1153         struct sit_info *sit_i = SIT_I(sbi);
1154         struct curseg_info *curseg;
1155         bool direct_io = (type == CURSEG_DIRECT_IO);
1156
1157         type = direct_io ? CURSEG_WARM_DATA : type;
1158
1159         curseg = CURSEG_I(sbi, type);
1160
1161         mutex_lock(&curseg->curseg_mutex);
1162
1163         /* direct_io'ed data is aligned to the segment for better performance */
1164         if (direct_io && curseg->next_blkoff)
1165                 __allocate_new_segments(sbi, type);
1166
1167         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1168
1169         /*
1170          * __add_sum_entry should be resided under the curseg_mutex
1171          * because, this function updates a summary entry in the
1172          * current summary block.
1173          */
1174         __add_sum_entry(sbi, type, sum);
1175
1176         mutex_lock(&sit_i->sentry_lock);
1177         __refresh_next_blkoff(sbi, curseg);
1178
1179         stat_inc_block_count(sbi, curseg);
1180
1181         if (!__has_curseg_space(sbi, type))
1182                 sit_i->s_ops->allocate_segment(sbi, type, false);
1183         /*
1184          * SIT information should be updated before segment allocation,
1185          * since SSR needs latest valid block information.
1186          */
1187         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1188
1189         mutex_unlock(&sit_i->sentry_lock);
1190
1191         if (page && IS_NODESEG(type))
1192                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1193
1194         mutex_unlock(&curseg->curseg_mutex);
1195 }
1196
1197 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
1198                         struct f2fs_summary *sum,
1199                         struct f2fs_io_info *fio)
1200 {
1201         int type = __get_segment_type(page, fio->type);
1202
1203         allocate_data_block(sbi, page, fio->blk_addr, &fio->blk_addr, sum, type);
1204
1205         /* writeout dirty page into bdev */
1206         f2fs_submit_page_mbio(sbi, page, fio);
1207 }
1208
1209 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1210 {
1211         struct f2fs_io_info fio = {
1212                 .type = META,
1213                 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1214                 .blk_addr = page->index,
1215         };
1216
1217         set_page_writeback(page);
1218         f2fs_submit_page_mbio(sbi, page, &fio);
1219 }
1220
1221 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
1222                         unsigned int nid, struct f2fs_io_info *fio)
1223 {
1224         struct f2fs_summary sum;
1225         set_summary(&sum, nid, 0, 0);
1226         do_write_page(sbi, page, &sum, fio);
1227 }
1228
1229 void write_data_page(struct page *page, struct dnode_of_data *dn,
1230                                 struct f2fs_io_info *fio)
1231 {
1232         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1233         struct f2fs_summary sum;
1234         struct node_info ni;
1235
1236         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1237         get_node_info(sbi, dn->nid, &ni);
1238         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1239         do_write_page(sbi, page, &sum, fio);
1240         dn->data_blkaddr = fio->blk_addr;
1241 }
1242
1243 void rewrite_data_page(struct page *page, struct f2fs_io_info *fio)
1244 {
1245         stat_inc_inplace_blocks(F2FS_P_SB(page));
1246         f2fs_submit_page_mbio(F2FS_P_SB(page), page, fio);
1247 }
1248
1249 void recover_data_page(struct f2fs_sb_info *sbi,
1250                         struct page *page, struct f2fs_summary *sum,
1251                         block_t old_blkaddr, block_t new_blkaddr)
1252 {
1253         struct sit_info *sit_i = SIT_I(sbi);
1254         struct curseg_info *curseg;
1255         unsigned int segno, old_cursegno;
1256         struct seg_entry *se;
1257         int type;
1258
1259         segno = GET_SEGNO(sbi, new_blkaddr);
1260         se = get_seg_entry(sbi, segno);
1261         type = se->type;
1262
1263         if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1264                 if (old_blkaddr == NULL_ADDR)
1265                         type = CURSEG_COLD_DATA;
1266                 else
1267                         type = CURSEG_WARM_DATA;
1268         }
1269         curseg = CURSEG_I(sbi, type);
1270
1271         mutex_lock(&curseg->curseg_mutex);
1272         mutex_lock(&sit_i->sentry_lock);
1273
1274         old_cursegno = curseg->segno;
1275
1276         /* change the current segment */
1277         if (segno != curseg->segno) {
1278                 curseg->next_segno = segno;
1279                 change_curseg(sbi, type, true);
1280         }
1281
1282         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1283         __add_sum_entry(sbi, type, sum);
1284
1285         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1286         locate_dirty_segment(sbi, old_cursegno);
1287
1288         mutex_unlock(&sit_i->sentry_lock);
1289         mutex_unlock(&curseg->curseg_mutex);
1290 }
1291
1292 static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1293                                         struct page *page, enum page_type type)
1294 {
1295         enum page_type btype = PAGE_TYPE_OF_BIO(type);
1296         struct f2fs_bio_info *io = &sbi->write_io[btype];
1297         struct bio_vec *bvec;
1298         int i;
1299
1300         down_read(&io->io_rwsem);
1301         if (!io->bio)
1302                 goto out;
1303
1304         bio_for_each_segment_all(bvec, io->bio, i) {
1305                 if (page == bvec->bv_page) {
1306                         up_read(&io->io_rwsem);
1307                         return true;
1308                 }
1309         }
1310
1311 out:
1312         up_read(&io->io_rwsem);
1313         return false;
1314 }
1315
1316 void f2fs_wait_on_page_writeback(struct page *page,
1317                                 enum page_type type)
1318 {
1319         if (PageWriteback(page)) {
1320                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1321
1322                 if (is_merged_page(sbi, page, type))
1323                         f2fs_submit_merged_bio(sbi, type, WRITE);
1324                 wait_on_page_writeback(page);
1325         }
1326 }
1327
1328 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1329 {
1330         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1331         struct curseg_info *seg_i;
1332         unsigned char *kaddr;
1333         struct page *page;
1334         block_t start;
1335         int i, j, offset;
1336
1337         start = start_sum_block(sbi);
1338
1339         page = get_meta_page(sbi, start++);
1340         kaddr = (unsigned char *)page_address(page);
1341
1342         /* Step 1: restore nat cache */
1343         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1344         memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1345
1346         /* Step 2: restore sit cache */
1347         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1348         memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1349                                                 SUM_JOURNAL_SIZE);
1350         offset = 2 * SUM_JOURNAL_SIZE;
1351
1352         /* Step 3: restore summary entries */
1353         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1354                 unsigned short blk_off;
1355                 unsigned int segno;
1356
1357                 seg_i = CURSEG_I(sbi, i);
1358                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1359                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1360                 seg_i->next_segno = segno;
1361                 reset_curseg(sbi, i, 0);
1362                 seg_i->alloc_type = ckpt->alloc_type[i];
1363                 seg_i->next_blkoff = blk_off;
1364
1365                 if (seg_i->alloc_type == SSR)
1366                         blk_off = sbi->blocks_per_seg;
1367
1368                 for (j = 0; j < blk_off; j++) {
1369                         struct f2fs_summary *s;
1370                         s = (struct f2fs_summary *)(kaddr + offset);
1371                         seg_i->sum_blk->entries[j] = *s;
1372                         offset += SUMMARY_SIZE;
1373                         if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1374                                                 SUM_FOOTER_SIZE)
1375                                 continue;
1376
1377                         f2fs_put_page(page, 1);
1378                         page = NULL;
1379
1380                         page = get_meta_page(sbi, start++);
1381                         kaddr = (unsigned char *)page_address(page);
1382                         offset = 0;
1383                 }
1384         }
1385         f2fs_put_page(page, 1);
1386         return 0;
1387 }
1388
1389 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1390 {
1391         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1392         struct f2fs_summary_block *sum;
1393         struct curseg_info *curseg;
1394         struct page *new;
1395         unsigned short blk_off;
1396         unsigned int segno = 0;
1397         block_t blk_addr = 0;
1398
1399         /* get segment number and block addr */
1400         if (IS_DATASEG(type)) {
1401                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1402                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1403                                                         CURSEG_HOT_DATA]);
1404                 if (__exist_node_summaries(sbi))
1405                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1406                 else
1407                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1408         } else {
1409                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1410                                                         CURSEG_HOT_NODE]);
1411                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1412                                                         CURSEG_HOT_NODE]);
1413                 if (__exist_node_summaries(sbi))
1414                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1415                                                         type - CURSEG_HOT_NODE);
1416                 else
1417                         blk_addr = GET_SUM_BLOCK(sbi, segno);
1418         }
1419
1420         new = get_meta_page(sbi, blk_addr);
1421         sum = (struct f2fs_summary_block *)page_address(new);
1422
1423         if (IS_NODESEG(type)) {
1424                 if (__exist_node_summaries(sbi)) {
1425                         struct f2fs_summary *ns = &sum->entries[0];
1426                         int i;
1427                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1428                                 ns->version = 0;
1429                                 ns->ofs_in_node = 0;
1430                         }
1431                 } else {
1432                         int err;
1433
1434                         err = restore_node_summary(sbi, segno, sum);
1435                         if (err) {
1436                                 f2fs_put_page(new, 1);
1437                                 return err;
1438                         }
1439                 }
1440         }
1441
1442         /* set uncompleted segment to curseg */
1443         curseg = CURSEG_I(sbi, type);
1444         mutex_lock(&curseg->curseg_mutex);
1445         memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1446         curseg->next_segno = segno;
1447         reset_curseg(sbi, type, 0);
1448         curseg->alloc_type = ckpt->alloc_type[type];
1449         curseg->next_blkoff = blk_off;
1450         mutex_unlock(&curseg->curseg_mutex);
1451         f2fs_put_page(new, 1);
1452         return 0;
1453 }
1454
1455 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1456 {
1457         int type = CURSEG_HOT_DATA;
1458         int err;
1459
1460         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1461                 int npages = npages_for_summary_flush(sbi, true);
1462
1463                 if (npages >= 2)
1464                         ra_meta_pages(sbi, start_sum_block(sbi), npages,
1465                                                                 META_CP);
1466
1467                 /* restore for compacted data summary */
1468                 if (read_compacted_summaries(sbi))
1469                         return -EINVAL;
1470                 type = CURSEG_HOT_NODE;
1471         }
1472
1473         if (__exist_node_summaries(sbi))
1474                 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1475                                         NR_CURSEG_TYPE - type, META_CP);
1476
1477         for (; type <= CURSEG_COLD_NODE; type++) {
1478                 err = read_normal_summaries(sbi, type);
1479                 if (err)
1480                         return err;
1481         }
1482
1483         return 0;
1484 }
1485
1486 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1487 {
1488         struct page *page;
1489         unsigned char *kaddr;
1490         struct f2fs_summary *summary;
1491         struct curseg_info *seg_i;
1492         int written_size = 0;
1493         int i, j;
1494
1495         page = grab_meta_page(sbi, blkaddr++);
1496         kaddr = (unsigned char *)page_address(page);
1497
1498         /* Step 1: write nat cache */
1499         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1500         memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1501         written_size += SUM_JOURNAL_SIZE;
1502
1503         /* Step 2: write sit cache */
1504         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1505         memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1506                                                 SUM_JOURNAL_SIZE);
1507         written_size += SUM_JOURNAL_SIZE;
1508
1509         /* Step 3: write summary entries */
1510         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1511                 unsigned short blkoff;
1512                 seg_i = CURSEG_I(sbi, i);
1513                 if (sbi->ckpt->alloc_type[i] == SSR)
1514                         blkoff = sbi->blocks_per_seg;
1515                 else
1516                         blkoff = curseg_blkoff(sbi, i);
1517
1518                 for (j = 0; j < blkoff; j++) {
1519                         if (!page) {
1520                                 page = grab_meta_page(sbi, blkaddr++);
1521                                 kaddr = (unsigned char *)page_address(page);
1522                                 written_size = 0;
1523                         }
1524                         summary = (struct f2fs_summary *)(kaddr + written_size);
1525                         *summary = seg_i->sum_blk->entries[j];
1526                         written_size += SUMMARY_SIZE;
1527
1528                         if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1529                                                         SUM_FOOTER_SIZE)
1530                                 continue;
1531
1532                         set_page_dirty(page);
1533                         f2fs_put_page(page, 1);
1534                         page = NULL;
1535                 }
1536         }
1537         if (page) {
1538                 set_page_dirty(page);
1539                 f2fs_put_page(page, 1);
1540         }
1541 }
1542
1543 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1544                                         block_t blkaddr, int type)
1545 {
1546         int i, end;
1547         if (IS_DATASEG(type))
1548                 end = type + NR_CURSEG_DATA_TYPE;
1549         else
1550                 end = type + NR_CURSEG_NODE_TYPE;
1551
1552         for (i = type; i < end; i++) {
1553                 struct curseg_info *sum = CURSEG_I(sbi, i);
1554                 mutex_lock(&sum->curseg_mutex);
1555                 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1556                 mutex_unlock(&sum->curseg_mutex);
1557         }
1558 }
1559
1560 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1561 {
1562         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1563                 write_compacted_summaries(sbi, start_blk);
1564         else
1565                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1566 }
1567
1568 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1569 {
1570         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1571 }
1572
1573 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1574                                         unsigned int val, int alloc)
1575 {
1576         int i;
1577
1578         if (type == NAT_JOURNAL) {
1579                 for (i = 0; i < nats_in_cursum(sum); i++) {
1580                         if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1581                                 return i;
1582                 }
1583                 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1584                         return update_nats_in_cursum(sum, 1);
1585         } else if (type == SIT_JOURNAL) {
1586                 for (i = 0; i < sits_in_cursum(sum); i++)
1587                         if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1588                                 return i;
1589                 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1590                         return update_sits_in_cursum(sum, 1);
1591         }
1592         return -1;
1593 }
1594
1595 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1596                                         unsigned int segno)
1597 {
1598         return get_meta_page(sbi, current_sit_addr(sbi, segno));
1599 }
1600
1601 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1602                                         unsigned int start)
1603 {
1604         struct sit_info *sit_i = SIT_I(sbi);
1605         struct page *src_page, *dst_page;
1606         pgoff_t src_off, dst_off;
1607         void *src_addr, *dst_addr;
1608
1609         src_off = current_sit_addr(sbi, start);
1610         dst_off = next_sit_addr(sbi, src_off);
1611
1612         /* get current sit block page without lock */
1613         src_page = get_meta_page(sbi, src_off);
1614         dst_page = grab_meta_page(sbi, dst_off);
1615         f2fs_bug_on(sbi, PageDirty(src_page));
1616
1617         src_addr = page_address(src_page);
1618         dst_addr = page_address(dst_page);
1619         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1620
1621         set_page_dirty(dst_page);
1622         f2fs_put_page(src_page, 1);
1623
1624         set_to_next_sit(sit_i, start);
1625
1626         return dst_page;
1627 }
1628
1629 static struct sit_entry_set *grab_sit_entry_set(void)
1630 {
1631         struct sit_entry_set *ses =
1632                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_ATOMIC);
1633
1634         ses->entry_cnt = 0;
1635         INIT_LIST_HEAD(&ses->set_list);
1636         return ses;
1637 }
1638
1639 static void release_sit_entry_set(struct sit_entry_set *ses)
1640 {
1641         list_del(&ses->set_list);
1642         kmem_cache_free(sit_entry_set_slab, ses);
1643 }
1644
1645 static void adjust_sit_entry_set(struct sit_entry_set *ses,
1646                                                 struct list_head *head)
1647 {
1648         struct sit_entry_set *next = ses;
1649
1650         if (list_is_last(&ses->set_list, head))
1651                 return;
1652
1653         list_for_each_entry_continue(next, head, set_list)
1654                 if (ses->entry_cnt <= next->entry_cnt)
1655                         break;
1656
1657         list_move_tail(&ses->set_list, &next->set_list);
1658 }
1659
1660 static void add_sit_entry(unsigned int segno, struct list_head *head)
1661 {
1662         struct sit_entry_set *ses;
1663         unsigned int start_segno = START_SEGNO(segno);
1664
1665         list_for_each_entry(ses, head, set_list) {
1666                 if (ses->start_segno == start_segno) {
1667                         ses->entry_cnt++;
1668                         adjust_sit_entry_set(ses, head);
1669                         return;
1670                 }
1671         }
1672
1673         ses = grab_sit_entry_set();
1674
1675         ses->start_segno = start_segno;
1676         ses->entry_cnt++;
1677         list_add(&ses->set_list, head);
1678 }
1679
1680 static void add_sits_in_set(struct f2fs_sb_info *sbi)
1681 {
1682         struct f2fs_sm_info *sm_info = SM_I(sbi);
1683         struct list_head *set_list = &sm_info->sit_entry_set;
1684         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
1685         unsigned int segno;
1686
1687         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
1688                 add_sit_entry(segno, set_list);
1689 }
1690
1691 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
1692 {
1693         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1694         struct f2fs_summary_block *sum = curseg->sum_blk;
1695         int i;
1696
1697         for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1698                 unsigned int segno;
1699                 bool dirtied;
1700
1701                 segno = le32_to_cpu(segno_in_journal(sum, i));
1702                 dirtied = __mark_sit_entry_dirty(sbi, segno);
1703
1704                 if (!dirtied)
1705                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
1706         }
1707         update_sits_in_cursum(sum, -sits_in_cursum(sum));
1708 }
1709
1710 /*
1711  * CP calls this function, which flushes SIT entries including sit_journal,
1712  * and moves prefree segs to free segs.
1713  */
1714 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1715 {
1716         struct sit_info *sit_i = SIT_I(sbi);
1717         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1718         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1719         struct f2fs_summary_block *sum = curseg->sum_blk;
1720         struct sit_entry_set *ses, *tmp;
1721         struct list_head *head = &SM_I(sbi)->sit_entry_set;
1722         bool to_journal = true;
1723         struct seg_entry *se;
1724
1725         mutex_lock(&curseg->curseg_mutex);
1726         mutex_lock(&sit_i->sentry_lock);
1727
1728         /*
1729          * add and account sit entries of dirty bitmap in sit entry
1730          * set temporarily
1731          */
1732         add_sits_in_set(sbi);
1733
1734         /*
1735          * if there are no enough space in journal to store dirty sit
1736          * entries, remove all entries from journal and add and account
1737          * them in sit entry set.
1738          */
1739         if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL))
1740                 remove_sits_in_journal(sbi);
1741
1742         if (!sit_i->dirty_sentries)
1743                 goto out;
1744
1745         /*
1746          * there are two steps to flush sit entries:
1747          * #1, flush sit entries to journal in current cold data summary block.
1748          * #2, flush sit entries to sit page.
1749          */
1750         list_for_each_entry_safe(ses, tmp, head, set_list) {
1751                 struct page *page = NULL;
1752                 struct f2fs_sit_block *raw_sit = NULL;
1753                 unsigned int start_segno = ses->start_segno;
1754                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
1755                                                 (unsigned long)MAIN_SEGS(sbi));
1756                 unsigned int segno = start_segno;
1757
1758                 if (to_journal &&
1759                         !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL))
1760                         to_journal = false;
1761
1762                 if (!to_journal) {
1763                         page = get_next_sit_page(sbi, start_segno);
1764                         raw_sit = page_address(page);
1765                 }
1766
1767                 /* flush dirty sit entries in region of current sit set */
1768                 for_each_set_bit_from(segno, bitmap, end) {
1769                         int offset, sit_offset;
1770
1771                         se = get_seg_entry(sbi, segno);
1772
1773                         /* add discard candidates */
1774                         if (cpc->reason != CP_DISCARD) {
1775                                 cpc->trim_start = segno;
1776                                 add_discard_addrs(sbi, cpc);
1777                         }
1778
1779                         if (to_journal) {
1780                                 offset = lookup_journal_in_cursum(sum,
1781                                                         SIT_JOURNAL, segno, 1);
1782                                 f2fs_bug_on(sbi, offset < 0);
1783                                 segno_in_journal(sum, offset) =
1784                                                         cpu_to_le32(segno);
1785                                 seg_info_to_raw_sit(se,
1786                                                 &sit_in_journal(sum, offset));
1787                         } else {
1788                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1789                                 seg_info_to_raw_sit(se,
1790                                                 &raw_sit->entries[sit_offset]);
1791                         }
1792
1793                         __clear_bit(segno, bitmap);
1794                         sit_i->dirty_sentries--;
1795                         ses->entry_cnt--;
1796                 }
1797
1798                 if (!to_journal)
1799                         f2fs_put_page(page, 1);
1800
1801                 f2fs_bug_on(sbi, ses->entry_cnt);
1802                 release_sit_entry_set(ses);
1803         }
1804
1805         f2fs_bug_on(sbi, !list_empty(head));
1806         f2fs_bug_on(sbi, sit_i->dirty_sentries);
1807 out:
1808         if (cpc->reason == CP_DISCARD) {
1809                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
1810                         add_discard_addrs(sbi, cpc);
1811         }
1812         mutex_unlock(&sit_i->sentry_lock);
1813         mutex_unlock(&curseg->curseg_mutex);
1814
1815         set_prefree_as_free_segments(sbi);
1816 }
1817
1818 static int build_sit_info(struct f2fs_sb_info *sbi)
1819 {
1820         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1821         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1822         struct sit_info *sit_i;
1823         unsigned int sit_segs, start;
1824         char *src_bitmap, *dst_bitmap;
1825         unsigned int bitmap_size;
1826
1827         /* allocate memory for SIT information */
1828         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1829         if (!sit_i)
1830                 return -ENOMEM;
1831
1832         SM_I(sbi)->sit_info = sit_i;
1833
1834         sit_i->sentries = vzalloc(MAIN_SEGS(sbi) * sizeof(struct seg_entry));
1835         if (!sit_i->sentries)
1836                 return -ENOMEM;
1837
1838         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
1839         sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1840         if (!sit_i->dirty_sentries_bitmap)
1841                 return -ENOMEM;
1842
1843         for (start = 0; start < MAIN_SEGS(sbi); start++) {
1844                 sit_i->sentries[start].cur_valid_map
1845                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1846                 sit_i->sentries[start].ckpt_valid_map
1847                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1848                 if (!sit_i->sentries[start].cur_valid_map
1849                                 || !sit_i->sentries[start].ckpt_valid_map)
1850                         return -ENOMEM;
1851         }
1852
1853         if (sbi->segs_per_sec > 1) {
1854                 sit_i->sec_entries = vzalloc(MAIN_SECS(sbi) *
1855                                         sizeof(struct sec_entry));
1856                 if (!sit_i->sec_entries)
1857                         return -ENOMEM;
1858         }
1859
1860         /* get information related with SIT */
1861         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1862
1863         /* setup SIT bitmap from ckeckpoint pack */
1864         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1865         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1866
1867         dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1868         if (!dst_bitmap)
1869                 return -ENOMEM;
1870
1871         /* init SIT information */
1872         sit_i->s_ops = &default_salloc_ops;
1873
1874         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1875         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1876         sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1877         sit_i->sit_bitmap = dst_bitmap;
1878         sit_i->bitmap_size = bitmap_size;
1879         sit_i->dirty_sentries = 0;
1880         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1881         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1882         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1883         mutex_init(&sit_i->sentry_lock);
1884         return 0;
1885 }
1886
1887 static int build_free_segmap(struct f2fs_sb_info *sbi)
1888 {
1889         struct free_segmap_info *free_i;
1890         unsigned int bitmap_size, sec_bitmap_size;
1891
1892         /* allocate memory for free segmap information */
1893         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1894         if (!free_i)
1895                 return -ENOMEM;
1896
1897         SM_I(sbi)->free_info = free_i;
1898
1899         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
1900         free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1901         if (!free_i->free_segmap)
1902                 return -ENOMEM;
1903
1904         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
1905         free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1906         if (!free_i->free_secmap)
1907                 return -ENOMEM;
1908
1909         /* set all segments as dirty temporarily */
1910         memset(free_i->free_segmap, 0xff, bitmap_size);
1911         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1912
1913         /* init free segmap information */
1914         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
1915         free_i->free_segments = 0;
1916         free_i->free_sections = 0;
1917         rwlock_init(&free_i->segmap_lock);
1918         return 0;
1919 }
1920
1921 static int build_curseg(struct f2fs_sb_info *sbi)
1922 {
1923         struct curseg_info *array;
1924         int i;
1925
1926         array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
1927         if (!array)
1928                 return -ENOMEM;
1929
1930         SM_I(sbi)->curseg_array = array;
1931
1932         for (i = 0; i < NR_CURSEG_TYPE; i++) {
1933                 mutex_init(&array[i].curseg_mutex);
1934                 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1935                 if (!array[i].sum_blk)
1936                         return -ENOMEM;
1937                 array[i].segno = NULL_SEGNO;
1938                 array[i].next_blkoff = 0;
1939         }
1940         return restore_curseg_summaries(sbi);
1941 }
1942
1943 static void build_sit_entries(struct f2fs_sb_info *sbi)
1944 {
1945         struct sit_info *sit_i = SIT_I(sbi);
1946         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1947         struct f2fs_summary_block *sum = curseg->sum_blk;
1948         int sit_blk_cnt = SIT_BLK_CNT(sbi);
1949         unsigned int i, start, end;
1950         unsigned int readed, start_blk = 0;
1951         int nrpages = MAX_BIO_BLOCKS(sbi);
1952
1953         do {
1954                 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT);
1955
1956                 start = start_blk * sit_i->sents_per_block;
1957                 end = (start_blk + readed) * sit_i->sents_per_block;
1958
1959                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
1960                         struct seg_entry *se = &sit_i->sentries[start];
1961                         struct f2fs_sit_block *sit_blk;
1962                         struct f2fs_sit_entry sit;
1963                         struct page *page;
1964
1965                         mutex_lock(&curseg->curseg_mutex);
1966                         for (i = 0; i < sits_in_cursum(sum); i++) {
1967                                 if (le32_to_cpu(segno_in_journal(sum, i))
1968                                                                 == start) {
1969                                         sit = sit_in_journal(sum, i);
1970                                         mutex_unlock(&curseg->curseg_mutex);
1971                                         goto got_it;
1972                                 }
1973                         }
1974                         mutex_unlock(&curseg->curseg_mutex);
1975
1976                         page = get_current_sit_page(sbi, start);
1977                         sit_blk = (struct f2fs_sit_block *)page_address(page);
1978                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1979                         f2fs_put_page(page, 1);
1980 got_it:
1981                         check_block_count(sbi, start, &sit);
1982                         seg_info_from_raw_sit(se, &sit);
1983                         if (sbi->segs_per_sec > 1) {
1984                                 struct sec_entry *e = get_sec_entry(sbi, start);
1985                                 e->valid_blocks += se->valid_blocks;
1986                         }
1987                 }
1988                 start_blk += readed;
1989         } while (start_blk < sit_blk_cnt);
1990 }
1991
1992 static void init_free_segmap(struct f2fs_sb_info *sbi)
1993 {
1994         unsigned int start;
1995         int type;
1996
1997         for (start = 0; start < MAIN_SEGS(sbi); start++) {
1998                 struct seg_entry *sentry = get_seg_entry(sbi, start);
1999                 if (!sentry->valid_blocks)
2000                         __set_free(sbi, start);
2001         }
2002
2003         /* set use the current segments */
2004         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2005                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2006                 __set_test_and_inuse(sbi, curseg_t->segno);
2007         }
2008 }
2009
2010 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2011 {
2012         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2013         struct free_segmap_info *free_i = FREE_I(sbi);
2014         unsigned int segno = 0, offset = 0;
2015         unsigned short valid_blocks;
2016
2017         while (1) {
2018                 /* find dirty segment based on free segmap */
2019                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2020                 if (segno >= MAIN_SEGS(sbi))
2021                         break;
2022                 offset = segno + 1;
2023                 valid_blocks = get_valid_blocks(sbi, segno, 0);
2024                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2025                         continue;
2026                 if (valid_blocks > sbi->blocks_per_seg) {
2027                         f2fs_bug_on(sbi, 1);
2028                         continue;
2029                 }
2030                 mutex_lock(&dirty_i->seglist_lock);
2031                 __locate_dirty_segment(sbi, segno, DIRTY);
2032                 mutex_unlock(&dirty_i->seglist_lock);
2033         }
2034 }
2035
2036 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2037 {
2038         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2039         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2040
2041         dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
2042         if (!dirty_i->victim_secmap)
2043                 return -ENOMEM;
2044         return 0;
2045 }
2046
2047 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2048 {
2049         struct dirty_seglist_info *dirty_i;
2050         unsigned int bitmap_size, i;
2051
2052         /* allocate memory for dirty segments list information */
2053         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2054         if (!dirty_i)
2055                 return -ENOMEM;
2056
2057         SM_I(sbi)->dirty_info = dirty_i;
2058         mutex_init(&dirty_i->seglist_lock);
2059
2060         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2061
2062         for (i = 0; i < NR_DIRTY_TYPE; i++) {
2063                 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
2064                 if (!dirty_i->dirty_segmap[i])
2065                         return -ENOMEM;
2066         }
2067
2068         init_dirty_segmap(sbi);
2069         return init_victim_secmap(sbi);
2070 }
2071
2072 /*
2073  * Update min, max modified time for cost-benefit GC algorithm
2074  */
2075 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2076 {
2077         struct sit_info *sit_i = SIT_I(sbi);
2078         unsigned int segno;
2079
2080         mutex_lock(&sit_i->sentry_lock);
2081
2082         sit_i->min_mtime = LLONG_MAX;
2083
2084         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2085                 unsigned int i;
2086                 unsigned long long mtime = 0;
2087
2088                 for (i = 0; i < sbi->segs_per_sec; i++)
2089                         mtime += get_seg_entry(sbi, segno + i)->mtime;
2090
2091                 mtime = div_u64(mtime, sbi->segs_per_sec);
2092
2093                 if (sit_i->min_mtime > mtime)
2094                         sit_i->min_mtime = mtime;
2095         }
2096         sit_i->max_mtime = get_mtime(sbi);
2097         mutex_unlock(&sit_i->sentry_lock);
2098 }
2099
2100 int build_segment_manager(struct f2fs_sb_info *sbi)
2101 {
2102         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2103         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2104         struct f2fs_sm_info *sm_info;
2105         int err;
2106
2107         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2108         if (!sm_info)
2109                 return -ENOMEM;
2110
2111         /* init sm info */
2112         sbi->sm_info = sm_info;
2113         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2114         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2115         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2116         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2117         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2118         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2119         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2120         sm_info->rec_prefree_segments = sm_info->main_segments *
2121                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2122         sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2123         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2124         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2125
2126         INIT_LIST_HEAD(&sm_info->discard_list);
2127         sm_info->nr_discards = 0;
2128         sm_info->max_discards = 0;
2129
2130         INIT_LIST_HEAD(&sm_info->sit_entry_set);
2131
2132         if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2133                 err = create_flush_cmd_control(sbi);
2134                 if (err)
2135                         return err;
2136         }
2137
2138         err = build_sit_info(sbi);
2139         if (err)
2140                 return err;
2141         err = build_free_segmap(sbi);
2142         if (err)
2143                 return err;
2144         err = build_curseg(sbi);
2145         if (err)
2146                 return err;
2147
2148         /* reinit free segmap based on SIT */
2149         build_sit_entries(sbi);
2150
2151         init_free_segmap(sbi);
2152         err = build_dirty_segmap(sbi);
2153         if (err)
2154                 return err;
2155
2156         init_min_max_mtime(sbi);
2157         return 0;
2158 }
2159
2160 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2161                 enum dirty_type dirty_type)
2162 {
2163         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2164
2165         mutex_lock(&dirty_i->seglist_lock);
2166         kfree(dirty_i->dirty_segmap[dirty_type]);
2167         dirty_i->nr_dirty[dirty_type] = 0;
2168         mutex_unlock(&dirty_i->seglist_lock);
2169 }
2170
2171 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2172 {
2173         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2174         kfree(dirty_i->victim_secmap);
2175 }
2176
2177 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2178 {
2179         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2180         int i;
2181
2182         if (!dirty_i)
2183                 return;
2184
2185         /* discard pre-free/dirty segments list */
2186         for (i = 0; i < NR_DIRTY_TYPE; i++)
2187                 discard_dirty_segmap(sbi, i);
2188
2189         destroy_victim_secmap(sbi);
2190         SM_I(sbi)->dirty_info = NULL;
2191         kfree(dirty_i);
2192 }
2193
2194 static void destroy_curseg(struct f2fs_sb_info *sbi)
2195 {
2196         struct curseg_info *array = SM_I(sbi)->curseg_array;
2197         int i;
2198
2199         if (!array)
2200                 return;
2201         SM_I(sbi)->curseg_array = NULL;
2202         for (i = 0; i < NR_CURSEG_TYPE; i++)
2203                 kfree(array[i].sum_blk);
2204         kfree(array);
2205 }
2206
2207 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2208 {
2209         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2210         if (!free_i)
2211                 return;
2212         SM_I(sbi)->free_info = NULL;
2213         kfree(free_i->free_segmap);
2214         kfree(free_i->free_secmap);
2215         kfree(free_i);
2216 }
2217
2218 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2219 {
2220         struct sit_info *sit_i = SIT_I(sbi);
2221         unsigned int start;
2222
2223         if (!sit_i)
2224                 return;
2225
2226         if (sit_i->sentries) {
2227                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2228                         kfree(sit_i->sentries[start].cur_valid_map);
2229                         kfree(sit_i->sentries[start].ckpt_valid_map);
2230                 }
2231         }
2232         vfree(sit_i->sentries);
2233         vfree(sit_i->sec_entries);
2234         kfree(sit_i->dirty_sentries_bitmap);
2235
2236         SM_I(sbi)->sit_info = NULL;
2237         kfree(sit_i->sit_bitmap);
2238         kfree(sit_i);
2239 }
2240
2241 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2242 {
2243         struct f2fs_sm_info *sm_info = SM_I(sbi);
2244
2245         if (!sm_info)
2246                 return;
2247         destroy_flush_cmd_control(sbi);
2248         destroy_dirty_segmap(sbi);
2249         destroy_curseg(sbi);
2250         destroy_free_segmap(sbi);
2251         destroy_sit_info(sbi);
2252         sbi->sm_info = NULL;
2253         kfree(sm_info);
2254 }
2255
2256 int __init create_segment_manager_caches(void)
2257 {
2258         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2259                         sizeof(struct discard_entry));
2260         if (!discard_entry_slab)
2261                 goto fail;
2262
2263         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2264                         sizeof(struct sit_entry_set));
2265         if (!sit_entry_set_slab)
2266                 goto destory_discard_entry;
2267
2268         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2269                         sizeof(struct inmem_pages));
2270         if (!inmem_entry_slab)
2271                 goto destroy_sit_entry_set;
2272         return 0;
2273
2274 destroy_sit_entry_set:
2275         kmem_cache_destroy(sit_entry_set_slab);
2276 destory_discard_entry:
2277         kmem_cache_destroy(discard_entry_slab);
2278 fail:
2279         return -ENOMEM;
2280 }
2281
2282 void destroy_segment_manager_caches(void)
2283 {
2284         kmem_cache_destroy(sit_entry_set_slab);
2285         kmem_cache_destroy(discard_entry_slab);
2286         kmem_cache_destroy(inmem_entry_slab);
2287 }