kernel/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         trace_f2fs_register_inmem_page(page, INMEM);
 210 }
 211
 212 void commit_inmem_pages(struct inode *inode, bool abort)
 213 {
 214         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 215         struct f2fs_inode_info *fi = F2FS_I(inode);
 216         struct inmem_pages *cur, *tmp;
 217         bool submit_bio = false;
 218         struct f2fs_io_info fio = {
 219                 .type = DATA,
 220                 .rw = WRITE_SYNC | REQ_PRIO,
 221         };
 222
 223         /*
 224          * The abort is true only when f2fs_evict_inode is called.
 225          * Basically, the f2fs_evict_inode doesn't produce any data writes, so
 226          * that we don't need to call f2fs_balance_fs.
 227          * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this
 228          * inode becomes free by iget_locked in f2fs_iget.
 229          */
 230         if (!abort) {
 231                 f2fs_balance_fs(sbi);
 232                 f2fs_lock_op(sbi);
 233         }
 234
 235         mutex_lock(&fi->inmem_lock);
 236         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 237                 if (!abort) {
 238                         lock_page(cur->page);
 239                         if (cur->page->mapping == inode->i_mapping) {
 240                                 f2fs_wait_on_page_writeback(cur->page, DATA);
 241                                 if (clear_page_dirty_for_io(cur->page))
 242                                         inode_dec_dirty_pages(inode);
 243                                 trace_f2fs_commit_inmem_page(cur->page, INMEM);
 244                                 do_write_data_page(cur->page, &fio);
 245                                 submit_bio = true;
 246                         }
 247                         f2fs_put_page(cur->page, 1);
 248                 } else {
 249                         trace_f2fs_commit_inmem_page(cur->page, INMEM_DROP);
 250                         put_page(cur->page);
 251                 }
 252                 radix_tree_delete(&fi->inmem_root, cur->page->index);
 253                 list_del(&cur->list);
 254                 kmem_cache_free(inmem_entry_slab, cur);
 255                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 256         }
 257         mutex_unlock(&fi->inmem_lock);
 258
 259         if (!abort) {
 260                 f2fs_unlock_op(sbi);
 261                 if (submit_bio)
 262                         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 263         }
 264 }
 265
 266 /*
 267  * This function balances dirty node and dentry pages.
 268  * In addition, it controls garbage collection.
 269  */
 270 void f2fs_balance_fs(struct f2fs_sb_info *sbi)
 271 {
 272         /*
 273          * We should do GC or end up with checkpoint, if there are so many dirty
 274          * dir/node pages without enough free segments.
 275          */
 276         if (has_not_enough_free_secs(sbi, 0)) {
 277                 mutex_lock(&sbi->gc_mutex);
 278                 f2fs_gc(sbi);
 279         }
 280 }
 281
 282 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 283 {
 284         /* try to shrink extent cache when there is no enough memory */
 285         f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 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 = SIT_I(sbi)->tmp_map;
 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                         SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards))
 522                 return;
 523
 524         if (force && !se->valid_blocks) {
 525                 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 526                 /*
 527                  * if this segment is registered in the prefree list, then
 528                  * we should skip adding a discard candidate, and let the
 529                  * checkpoint do that later.
 530                  */
 531                 mutex_lock(&dirty_i->seglist_lock);
 532                 if (test_bit(cpc->trim_start, dirty_i->dirty_segmap[PRE])) {
 533                         mutex_unlock(&dirty_i->seglist_lock);
 534                         cpc->trimmed += sbi->blocks_per_seg;
 535                         return;
 536                 }
 537                 mutex_unlock(&dirty_i->seglist_lock);
 538
 539                 __add_discard_entry(sbi, cpc, 0, sbi->blocks_per_seg);
 540                 return;
 541         }
 542
 543         /* zero block will be discarded through the prefree list */
 544         if (!se->valid_blocks || se->valid_blocks == max_blocks)
 545                 return;
 546
 547         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
 548         for (i = 0; i < entries; i++)
 549                 dmap[i] = force ? ~ckpt_map[i] :
 550                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
 551
 552         while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
 553                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
 554                 if (start >= max_blocks)
 555                         break;
 556
 557                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
 558
 559                 if (force && end - start < cpc->trim_minlen)
 560                         continue;
 561
 562                 __add_discard_entry(sbi, cpc, start, end);
 563         }
 564 }
 565
 566 void release_discard_addrs(struct f2fs_sb_info *sbi)
 567 {
 568         struct list_head *head = &(SM_I(sbi)->discard_list);
 569         struct discard_entry *entry, *this;
 570
 571         /* drop caches */
 572         list_for_each_entry_safe(entry, this, head, list) {
 573                 list_del(&entry->list);
 574                 kmem_cache_free(discard_entry_slab, entry);
 575         }
 576 }
 577
 578 /*
 579  * Should call clear_prefree_segments after checkpoint is done.
 580  */
 581 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
 582 {
 583         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 584         unsigned int segno;
 585
 586         mutex_lock(&dirty_i->seglist_lock);
 587         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
 588                 __set_test_and_free(sbi, segno);
 589         mutex_unlock(&dirty_i->seglist_lock);
 590 }
 591
 592 void clear_prefree_segments(struct f2fs_sb_info *sbi)
 593 {
 594         struct list_head *head = &(SM_I(sbi)->discard_list);
 595         struct discard_entry *entry, *this;
 596         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 597         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
 598         unsigned int start = 0, end = -1;
 599
 600         mutex_lock(&dirty_i->seglist_lock);
 601
 602         while (1) {
 603                 int i;
 604                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
 605                 if (start >= MAIN_SEGS(sbi))
 606                         break;
 607                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
 608                                                                 start + 1);
 609
 610                 for (i = start; i < end; i++)
 611                         clear_bit(i, prefree_map);
 612
 613                 dirty_i->nr_dirty[PRE] -= end - start;
 614
 615                 if (!test_opt(sbi, DISCARD))
 616                         continue;
 617
 618                 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
 619                                 (end - start) << sbi->log_blocks_per_seg);
 620         }
 621         mutex_unlock(&dirty_i->seglist_lock);
 622
 623         /* send small discards */
 624         list_for_each_entry_safe(entry, this, head, list) {
 625                 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
 626                 list_del(&entry->list);
 627                 SM_I(sbi)->nr_discards -= entry->len;
 628                 kmem_cache_free(discard_entry_slab, entry);
 629         }
 630 }
 631
 632 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
 633 {
 634         struct sit_info *sit_i = SIT_I(sbi);
 635
 636         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
 637                 sit_i->dirty_sentries++;
 638                 return false;
 639         }
 640
 641         return true;
 642 }
 643
 644 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
 645                                         unsigned int segno, int modified)
 646 {
 647         struct seg_entry *se = get_seg_entry(sbi, segno);
 648         se->type = type;
 649         if (modified)
 650                 __mark_sit_entry_dirty(sbi, segno);
 651 }
 652
 653 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
 654 {
 655         struct seg_entry *se;
 656         unsigned int segno, offset;
 657         long int new_vblocks;
 658
 659         segno = GET_SEGNO(sbi, blkaddr);
 660
 661         se = get_seg_entry(sbi, segno);
 662         new_vblocks = se->valid_blocks + del;
 663         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 664
 665         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
 666                                 (new_vblocks > sbi->blocks_per_seg)));
 667
 668         se->valid_blocks = new_vblocks;
 669         se->mtime = get_mtime(sbi);
 670         SIT_I(sbi)->max_mtime = se->mtime;
 671
 672         /* Update valid block bitmap */
 673         if (del > 0) {
 674                 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
 675                         f2fs_bug_on(sbi, 1);
 676         } else {
 677                 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
 678                         f2fs_bug_on(sbi, 1);
 679         }
 680         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
 681                 se->ckpt_valid_blocks += del;
 682
 683         __mark_sit_entry_dirty(sbi, segno);
 684
 685         /* update total number of valid blocks to be written in ckpt area */
 686         SIT_I(sbi)->written_valid_blocks += del;
 687
 688         if (sbi->segs_per_sec > 1)
 689                 get_sec_entry(sbi, segno)->valid_blocks += del;
 690 }
 691
 692 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
 693 {
 694         update_sit_entry(sbi, new, 1);
 695         if (GET_SEGNO(sbi, old) != NULL_SEGNO)
 696                 update_sit_entry(sbi, old, -1);
 697
 698         locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
 699         locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
 700 }
 701
 702 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
 703 {
 704         unsigned int segno = GET_SEGNO(sbi, addr);
 705         struct sit_info *sit_i = SIT_I(sbi);
 706
 707         f2fs_bug_on(sbi, addr == NULL_ADDR);
 708         if (addr == NEW_ADDR)
 709                 return;
 710
 711         /* add it into sit main buffer */
 712         mutex_lock(&sit_i->sentry_lock);
 713
 714         update_sit_entry(sbi, addr, -1);
 715
 716         /* add it into dirty seglist */
 717         locate_dirty_segment(sbi, segno);
 718
 719         mutex_unlock(&sit_i->sentry_lock);
 720 }
 721
 722 /*
 723  * This function should be resided under the curseg_mutex lock
 724  */
 725 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
 726                                         struct f2fs_summary *sum)
 727 {
 728         struct curseg_info *curseg = CURSEG_I(sbi, type);
 729         void *addr = curseg->sum_blk;
 730         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
 731         memcpy(addr, sum, sizeof(struct f2fs_summary));
 732 }
 733
 734 /*
 735  * Calculate the number of current summary pages for writing
 736  */
 737 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
 738 {
 739         int valid_sum_count = 0;
 740         int i, sum_in_page;
 741
 742         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 743                 if (sbi->ckpt->alloc_type[i] == SSR)
 744                         valid_sum_count += sbi->blocks_per_seg;
 745                 else {
 746                         if (for_ra)
 747                                 valid_sum_count += le16_to_cpu(
 748                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
 749                         else
 750                                 valid_sum_count += curseg_blkoff(sbi, i);
 751                 }
 752         }
 753
 754         sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
 755                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
 756         if (valid_sum_count <= sum_in_page)
 757                 return 1;
 758         else if ((valid_sum_count - sum_in_page) <=
 759                 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
 760                 return 2;
 761         return 3;
 762 }
 763
 764 /*
 765  * Caller should put this summary page
 766  */
 767 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
 768 {
 769         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
 770 }
 771
 772 static void write_sum_page(struct f2fs_sb_info *sbi,
 773                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
 774 {
 775         struct page *page = grab_meta_page(sbi, blk_addr);
 776         void *kaddr = page_address(page);
 777         memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
 778         set_page_dirty(page);
 779         f2fs_put_page(page, 1);
 780 }
 781
 782 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
 783 {
 784         struct curseg_info *curseg = CURSEG_I(sbi, type);
 785         unsigned int segno = curseg->segno + 1;
 786         struct free_segmap_info *free_i = FREE_I(sbi);
 787
 788         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
 789                 return !test_bit(segno, free_i->free_segmap);
 790         return 0;
 791 }
 792
 793 /*
 794  * Find a new segment from the free segments bitmap to right order
 795  * This function should be returned with success, otherwise BUG
 796  */
 797 static void get_new_segment(struct f2fs_sb_info *sbi,
 798                         unsigned int *newseg, bool new_sec, int dir)
 799 {
 800         struct free_segmap_info *free_i = FREE_I(sbi);
 801         unsigned int segno, secno, zoneno;
 802         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
 803         unsigned int hint = *newseg / sbi->segs_per_sec;
 804         unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
 805         unsigned int left_start = hint;
 806         bool init = true;
 807         int go_left = 0;
 808         int i;
 809
 810         spin_lock(&free_i->segmap_lock);
 811
 812         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
 813                 segno = find_next_zero_bit(free_i->free_segmap,
 814                                         MAIN_SEGS(sbi), *newseg + 1);
 815                 if (segno - *newseg < sbi->segs_per_sec -
 816                                         (*newseg % sbi->segs_per_sec))
 817                         goto got_it;
 818         }
 819 find_other_zone:
 820         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
 821         if (secno >= MAIN_SECS(sbi)) {
 822                 if (dir == ALLOC_RIGHT) {
 823                         secno = find_next_zero_bit(free_i->free_secmap,
 824                                                         MAIN_SECS(sbi), 0);
 825                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
 826                 } else {
 827                         go_left = 1;
 828                         left_start = hint - 1;
 829                 }
 830         }
 831         if (go_left == 0)
 832                 goto skip_left;
 833
 834         while (test_bit(left_start, free_i->free_secmap)) {
 835                 if (left_start > 0) {
 836                         left_start--;
 837                         continue;
 838                 }
 839                 left_start = find_next_zero_bit(free_i->free_secmap,
 840                                                         MAIN_SECS(sbi), 0);
 841                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
 842                 break;
 843         }
 844         secno = left_start;
 845 skip_left:
 846         hint = secno;
 847         segno = secno * sbi->segs_per_sec;
 848         zoneno = secno / sbi->secs_per_zone;
 849
 850         /* give up on finding another zone */
 851         if (!init)
 852                 goto got_it;
 853         if (sbi->secs_per_zone == 1)
 854                 goto got_it;
 855         if (zoneno == old_zoneno)
 856                 goto got_it;
 857         if (dir == ALLOC_LEFT) {
 858                 if (!go_left && zoneno + 1 >= total_zones)
 859                         goto got_it;
 860                 if (go_left && zoneno == 0)
 861                         goto got_it;
 862         }
 863         for (i = 0; i < NR_CURSEG_TYPE; i++)
 864                 if (CURSEG_I(sbi, i)->zone == zoneno)
 865                         break;
 866
 867         if (i < NR_CURSEG_TYPE) {
 868                 /* zone is in user, try another */
 869                 if (go_left)
 870                         hint = zoneno * sbi->secs_per_zone - 1;
 871                 else if (zoneno + 1 >= total_zones)
 872                         hint = 0;
 873                 else
 874                         hint = (zoneno + 1) * sbi->secs_per_zone;
 875                 init = false;
 876                 goto find_other_zone;
 877         }
 878 got_it:
 879         /* set it as dirty segment in free segmap */
 880         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
 881         __set_inuse(sbi, segno);
 882         *newseg = segno;
 883         spin_unlock(&free_i->segmap_lock);
 884 }
 885
 886 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
 887 {
 888         struct curseg_info *curseg = CURSEG_I(sbi, type);
 889         struct summary_footer *sum_footer;
 890
 891         curseg->segno = curseg->next_segno;
 892         curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
 893         curseg->next_blkoff = 0;
 894         curseg->next_segno = NULL_SEGNO;
 895
 896         sum_footer = &(curseg->sum_blk->footer);
 897         memset(sum_footer, 0, sizeof(struct summary_footer));
 898         if (IS_DATASEG(type))
 899                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
 900         if (IS_NODESEG(type))
 901                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
 902         __set_sit_entry_type(sbi, type, curseg->segno, modified);
 903 }
 904
 905 /*
 906  * Allocate a current working segment.
 907  * This function always allocates a free segment in LFS manner.
 908  */
 909 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
 910 {
 911         struct curseg_info *curseg = CURSEG_I(sbi, type);
 912         unsigned int segno = curseg->segno;
 913         int dir = ALLOC_LEFT;
 914
 915         write_sum_page(sbi, curseg->sum_blk,
 916                                 GET_SUM_BLOCK(sbi, segno));
 917         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
 918                 dir = ALLOC_RIGHT;
 919
 920         if (test_opt(sbi, NOHEAP))
 921                 dir = ALLOC_RIGHT;
 922
 923         get_new_segment(sbi, &segno, new_sec, dir);
 924         curseg->next_segno = segno;
 925         reset_curseg(sbi, type, 1);
 926         curseg->alloc_type = LFS;
 927 }
 928
 929 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
 930                         struct curseg_info *seg, block_t start)
 931 {
 932         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
 933         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
 934         unsigned long *target_map = SIT_I(sbi)->tmp_map;
 935         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
 936         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
 937         int i, pos;
 938
 939         for (i = 0; i < entries; i++)
 940                 target_map[i] = ckpt_map[i] | cur_map[i];
 941
 942         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
 943
 944         seg->next_blkoff = pos;
 945 }
 946
 947 /*
 948  * If a segment is written by LFS manner, next block offset is just obtained
 949  * by increasing the current block offset. However, if a segment is written by
 950  * SSR manner, next block offset obtained by calling __next_free_blkoff
 951  */
 952 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
 953                                 struct curseg_info *seg)
 954 {
 955         if (seg->alloc_type == SSR)
 956                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
 957         else
 958                 seg->next_blkoff++;
 959 }
 960
 961 /*
 962  * This function always allocates a used segment(from dirty seglist) by SSR
 963  * manner, so it should recover the existing segment information of valid blocks
 964  */
 965 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
 966 {
 967         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 968         struct curseg_info *curseg = CURSEG_I(sbi, type);
 969         unsigned int new_segno = curseg->next_segno;
 970         struct f2fs_summary_block *sum_node;
 971         struct page *sum_page;
 972
 973         write_sum_page(sbi, curseg->sum_blk,
 974                                 GET_SUM_BLOCK(sbi, curseg->segno));
 975         __set_test_and_inuse(sbi, new_segno);
 976
 977         mutex_lock(&dirty_i->seglist_lock);
 978         __remove_dirty_segment(sbi, new_segno, PRE);
 979         __remove_dirty_segment(sbi, new_segno, DIRTY);
 980         mutex_unlock(&dirty_i->seglist_lock);
 981
 982         reset_curseg(sbi, type, 1);
 983         curseg->alloc_type = SSR;
 984         __next_free_blkoff(sbi, curseg, 0);
 985
 986         if (reuse) {
 987                 sum_page = get_sum_page(sbi, new_segno);
 988                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
 989                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
 990                 f2fs_put_page(sum_page, 1);
 991         }
 992 }
 993
 994 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
 995 {
 996         struct curseg_info *curseg = CURSEG_I(sbi, type);
 997         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
 998
 999         if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1000                 return v_ops->get_victim(sbi,
1001                                 &(curseg)->next_segno, BG_GC, type, SSR);
1002
1003         /* For data segments, let's do SSR more intensively */
1004         for (; type >= CURSEG_HOT_DATA; type--)
1005                 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1006                                                 BG_GC, type, SSR))
1007                         return 1;
1008         return 0;
1009 }
1010
1011 /*
1012  * flush out current segment and replace it with new segment
1013  * This function should be returned with success, otherwise BUG
1014  */
1015 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1016                                                 int type, bool force)
1017 {
1018         struct curseg_info *curseg = CURSEG_I(sbi, type);
1019
1020         if (force)
1021                 new_curseg(sbi, type, true);
1022         else if (type == CURSEG_WARM_NODE)
1023                 new_curseg(sbi, type, false);
1024         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1025                 new_curseg(sbi, type, false);
1026         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1027                 change_curseg(sbi, type, true);
1028         else
1029                 new_curseg(sbi, type, false);
1030
1031         stat_inc_seg_type(sbi, curseg);
1032 }
1033
1034 static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1035 {
1036         struct curseg_info *curseg = CURSEG_I(sbi, type);
1037         unsigned int old_segno;
1038
1039         old_segno = curseg->segno;
1040         SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1041         locate_dirty_segment(sbi, old_segno);
1042 }
1043
1044 void allocate_new_segments(struct f2fs_sb_info *sbi)
1045 {
1046         int i;
1047
1048         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1049                 __allocate_new_segments(sbi, i);
1050 }
1051
1052 static const struct segment_allocation default_salloc_ops = {
1053         .allocate_segment = allocate_segment_by_default,
1054 };
1055
1056 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1057 {
1058         __u64 start = F2FS_BYTES_TO_BLK(range->start);
1059         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1060         unsigned int start_segno, end_segno;
1061         struct cp_control cpc;
1062
1063         if (range->minlen > SEGMENT_SIZE(sbi) || start >= MAX_BLKADDR(sbi) ||
1064                                                 range->len < sbi->blocksize)
1065                 return -EINVAL;
1066
1067         cpc.trimmed = 0;
1068         if (end <= MAIN_BLKADDR(sbi))
1069                 goto out;
1070
1071         /* start/end segment number in main_area */
1072         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1073         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1074                                                 GET_SEGNO(sbi, end);
1075         cpc.reason = CP_DISCARD;
1076         cpc.trim_minlen = F2FS_BYTES_TO_BLK(range->minlen);
1077
1078         /* do checkpoint to issue discard commands safely */
1079         for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1080                 cpc.trim_start = start_segno;
1081                 cpc.trim_end = min_t(unsigned int, rounddown(start_segno +
1082                                 BATCHED_TRIM_SEGMENTS(sbi),
1083                                 sbi->segs_per_sec) - 1, end_segno);
1084
1085                 mutex_lock(&sbi->gc_mutex);
1086                 write_checkpoint(sbi, &cpc);
1087                 mutex_unlock(&sbi->gc_mutex);
1088         }
1089 out:
1090         range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1091         return 0;
1092 }
1093
1094 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1095 {
1096         struct curseg_info *curseg = CURSEG_I(sbi, type);
1097         if (curseg->next_blkoff < sbi->blocks_per_seg)
1098                 return true;
1099         return false;
1100 }
1101
1102 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1103 {
1104         if (p_type == DATA)
1105                 return CURSEG_HOT_DATA;
1106         else
1107                 return CURSEG_HOT_NODE;
1108 }
1109
1110 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1111 {
1112         if (p_type == DATA) {
1113                 struct inode *inode = page->mapping->host;
1114
1115                 if (S_ISDIR(inode->i_mode))
1116                         return CURSEG_HOT_DATA;
1117                 else
1118                         return CURSEG_COLD_DATA;
1119         } else {
1120                 if (IS_DNODE(page) && is_cold_node(page))
1121                         return CURSEG_WARM_NODE;
1122                 else
1123                         return CURSEG_COLD_NODE;
1124         }
1125 }
1126
1127 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1128 {
1129         if (p_type == DATA) {
1130                 struct inode *inode = page->mapping->host;
1131
1132                 if (S_ISDIR(inode->i_mode))
1133                         return CURSEG_HOT_DATA;
1134                 else if (is_cold_data(page) || file_is_cold(inode))
1135                         return CURSEG_COLD_DATA;
1136                 else
1137                         return CURSEG_WARM_DATA;
1138         } else {
1139                 if (IS_DNODE(page))
1140                         return is_cold_node(page) ? CURSEG_WARM_NODE :
1141                                                 CURSEG_HOT_NODE;
1142                 else
1143                         return CURSEG_COLD_NODE;
1144         }
1145 }
1146
1147 static int __get_segment_type(struct page *page, enum page_type p_type)
1148 {
1149         switch (F2FS_P_SB(page)->active_logs) {
1150         case 2:
1151                 return __get_segment_type_2(page, p_type);
1152         case 4:
1153                 return __get_segment_type_4(page, p_type);
1154         }
1155         /* NR_CURSEG_TYPE(6) logs by default */
1156         f2fs_bug_on(F2FS_P_SB(page),
1157                 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1158         return __get_segment_type_6(page, p_type);
1159 }
1160
1161 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1162                 block_t old_blkaddr, block_t *new_blkaddr,
1163                 struct f2fs_summary *sum, int type)
1164 {
1165         struct sit_info *sit_i = SIT_I(sbi);
1166         struct curseg_info *curseg;
1167         bool direct_io = (type == CURSEG_DIRECT_IO);
1168
1169         type = direct_io ? CURSEG_WARM_DATA : type;
1170
1171         curseg = CURSEG_I(sbi, type);
1172
1173         mutex_lock(&curseg->curseg_mutex);
1174         mutex_lock(&sit_i->sentry_lock);
1175
1176         /* direct_io'ed data is aligned to the segment for better performance */
1177         if (direct_io && curseg->next_blkoff)
1178                 __allocate_new_segments(sbi, type);
1179
1180         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1181
1182         /*
1183          * __add_sum_entry should be resided under the curseg_mutex
1184          * because, this function updates a summary entry in the
1185          * current summary block.
1186          */
1187         __add_sum_entry(sbi, type, sum);
1188
1189         __refresh_next_blkoff(sbi, curseg);
1190
1191         stat_inc_block_count(sbi, curseg);
1192
1193         if (!__has_curseg_space(sbi, type))
1194                 sit_i->s_ops->allocate_segment(sbi, type, false);
1195         /*
1196          * SIT information should be updated before segment allocation,
1197          * since SSR needs latest valid block information.
1198          */
1199         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1200
1201         mutex_unlock(&sit_i->sentry_lock);
1202
1203         if (page && IS_NODESEG(type))
1204                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1205
1206         mutex_unlock(&curseg->curseg_mutex);
1207 }
1208
1209 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
1210                         struct f2fs_summary *sum,
1211                         struct f2fs_io_info *fio)
1212 {
1213         int type = __get_segment_type(page, fio->type);
1214
1215         allocate_data_block(sbi, page, fio->blk_addr, &fio->blk_addr, sum, type);
1216
1217         /* writeout dirty page into bdev */
1218         f2fs_submit_page_mbio(sbi, page, fio);
1219 }
1220
1221 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1222 {
1223         struct f2fs_io_info fio = {
1224                 .type = META,
1225                 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1226                 .blk_addr = page->index,
1227         };
1228
1229         set_page_writeback(page);
1230         f2fs_submit_page_mbio(sbi, page, &fio);
1231 }
1232
1233 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
1234                         unsigned int nid, struct f2fs_io_info *fio)
1235 {
1236         struct f2fs_summary sum;
1237         set_summary(&sum, nid, 0, 0);
1238         do_write_page(sbi, page, &sum, fio);
1239 }
1240
1241 void write_data_page(struct page *page, struct dnode_of_data *dn,
1242                                 struct f2fs_io_info *fio)
1243 {
1244         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1245         struct f2fs_summary sum;
1246         struct node_info ni;
1247
1248         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1249         get_node_info(sbi, dn->nid, &ni);
1250         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1251         do_write_page(sbi, page, &sum, fio);
1252         dn->data_blkaddr = fio->blk_addr;
1253 }
1254
1255 void rewrite_data_page(struct page *page, struct f2fs_io_info *fio)
1256 {
1257         stat_inc_inplace_blocks(F2FS_P_SB(page));
1258         f2fs_submit_page_mbio(F2FS_P_SB(page), page, fio);
1259 }
1260
1261 void recover_data_page(struct f2fs_sb_info *sbi,
1262                         struct page *page, struct f2fs_summary *sum,
1263                         block_t old_blkaddr, block_t new_blkaddr)
1264 {
1265         struct sit_info *sit_i = SIT_I(sbi);
1266         struct curseg_info *curseg;
1267         unsigned int segno, old_cursegno;
1268         struct seg_entry *se;
1269         int type;
1270
1271         segno = GET_SEGNO(sbi, new_blkaddr);
1272         se = get_seg_entry(sbi, segno);
1273         type = se->type;
1274
1275         if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1276                 if (old_blkaddr == NULL_ADDR)
1277                         type = CURSEG_COLD_DATA;
1278                 else
1279                         type = CURSEG_WARM_DATA;
1280         }
1281         curseg = CURSEG_I(sbi, type);
1282
1283         mutex_lock(&curseg->curseg_mutex);
1284         mutex_lock(&sit_i->sentry_lock);
1285
1286         old_cursegno = curseg->segno;
1287
1288         /* change the current segment */
1289         if (segno != curseg->segno) {
1290                 curseg->next_segno = segno;
1291                 change_curseg(sbi, type, true);
1292         }
1293
1294         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1295         __add_sum_entry(sbi, type, sum);
1296
1297         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1298         locate_dirty_segment(sbi, old_cursegno);
1299
1300         mutex_unlock(&sit_i->sentry_lock);
1301         mutex_unlock(&curseg->curseg_mutex);
1302 }
1303
1304 static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1305                                         struct page *page, enum page_type type)
1306 {
1307         enum page_type btype = PAGE_TYPE_OF_BIO(type);
1308         struct f2fs_bio_info *io = &sbi->write_io[btype];
1309         struct bio_vec *bvec;
1310         int i;
1311
1312         down_read(&io->io_rwsem);
1313         if (!io->bio)
1314                 goto out;
1315
1316         bio_for_each_segment_all(bvec, io->bio, i) {
1317                 if (page == bvec->bv_page) {
1318                         up_read(&io->io_rwsem);
1319                         return true;
1320                 }
1321         }
1322
1323 out:
1324         up_read(&io->io_rwsem);
1325         return false;
1326 }
1327
1328 void f2fs_wait_on_page_writeback(struct page *page,
1329                                 enum page_type type)
1330 {
1331         if (PageWriteback(page)) {
1332                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1333
1334                 if (is_merged_page(sbi, page, type))
1335                         f2fs_submit_merged_bio(sbi, type, WRITE);
1336                 wait_on_page_writeback(page);
1337         }
1338 }
1339
1340 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1341 {
1342         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1343         struct curseg_info *seg_i;
1344         unsigned char *kaddr;
1345         struct page *page;
1346         block_t start;
1347         int i, j, offset;
1348
1349         start = start_sum_block(sbi);
1350
1351         page = get_meta_page(sbi, start++);
1352         kaddr = (unsigned char *)page_address(page);
1353
1354         /* Step 1: restore nat cache */
1355         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1356         memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1357
1358         /* Step 2: restore sit cache */
1359         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1360         memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1361                                                 SUM_JOURNAL_SIZE);
1362         offset = 2 * SUM_JOURNAL_SIZE;
1363
1364         /* Step 3: restore summary entries */
1365         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1366                 unsigned short blk_off;
1367                 unsigned int segno;
1368
1369                 seg_i = CURSEG_I(sbi, i);
1370                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1371                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1372                 seg_i->next_segno = segno;
1373                 reset_curseg(sbi, i, 0);
1374                 seg_i->alloc_type = ckpt->alloc_type[i];
1375                 seg_i->next_blkoff = blk_off;
1376
1377                 if (seg_i->alloc_type == SSR)
1378                         blk_off = sbi->blocks_per_seg;
1379
1380                 for (j = 0; j < blk_off; j++) {
1381                         struct f2fs_summary *s;
1382                         s = (struct f2fs_summary *)(kaddr + offset);
1383                         seg_i->sum_blk->entries[j] = *s;
1384                         offset += SUMMARY_SIZE;
1385                         if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1386                                                 SUM_FOOTER_SIZE)
1387                                 continue;
1388
1389                         f2fs_put_page(page, 1);
1390                         page = NULL;
1391
1392                         page = get_meta_page(sbi, start++);
1393                         kaddr = (unsigned char *)page_address(page);
1394                         offset = 0;
1395                 }
1396         }
1397         f2fs_put_page(page, 1);
1398         return 0;
1399 }
1400
1401 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1402 {
1403         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1404         struct f2fs_summary_block *sum;
1405         struct curseg_info *curseg;
1406         struct page *new;
1407         unsigned short blk_off;
1408         unsigned int segno = 0;
1409         block_t blk_addr = 0;
1410
1411         /* get segment number and block addr */
1412         if (IS_DATASEG(type)) {
1413                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1414                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1415                                                         CURSEG_HOT_DATA]);
1416                 if (__exist_node_summaries(sbi))
1417                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1418                 else
1419                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1420         } else {
1421                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1422                                                         CURSEG_HOT_NODE]);
1423                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1424                                                         CURSEG_HOT_NODE]);
1425                 if (__exist_node_summaries(sbi))
1426                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1427                                                         type - CURSEG_HOT_NODE);
1428                 else
1429                         blk_addr = GET_SUM_BLOCK(sbi, segno);
1430         }
1431
1432         new = get_meta_page(sbi, blk_addr);
1433         sum = (struct f2fs_summary_block *)page_address(new);
1434
1435         if (IS_NODESEG(type)) {
1436                 if (__exist_node_summaries(sbi)) {
1437                         struct f2fs_summary *ns = &sum->entries[0];
1438                         int i;
1439                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1440                                 ns->version = 0;
1441                                 ns->ofs_in_node = 0;
1442                         }
1443                 } else {
1444                         int err;
1445
1446                         err = restore_node_summary(sbi, segno, sum);
1447                         if (err) {
1448                                 f2fs_put_page(new, 1);
1449                                 return err;
1450                         }
1451                 }
1452         }
1453
1454         /* set uncompleted segment to curseg */
1455         curseg = CURSEG_I(sbi, type);
1456         mutex_lock(&curseg->curseg_mutex);
1457         memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1458         curseg->next_segno = segno;
1459         reset_curseg(sbi, type, 0);
1460         curseg->alloc_type = ckpt->alloc_type[type];
1461         curseg->next_blkoff = blk_off;
1462         mutex_unlock(&curseg->curseg_mutex);
1463         f2fs_put_page(new, 1);
1464         return 0;
1465 }
1466
1467 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1468 {
1469         int type = CURSEG_HOT_DATA;
1470         int err;
1471
1472         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1473                 int npages = npages_for_summary_flush(sbi, true);
1474
1475                 if (npages >= 2)
1476                         ra_meta_pages(sbi, start_sum_block(sbi), npages,
1477                                                                 META_CP);
1478
1479                 /* restore for compacted data summary */
1480                 if (read_compacted_summaries(sbi))
1481                         return -EINVAL;
1482                 type = CURSEG_HOT_NODE;
1483         }
1484
1485         if (__exist_node_summaries(sbi))
1486                 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1487                                         NR_CURSEG_TYPE - type, META_CP);
1488
1489         for (; type <= CURSEG_COLD_NODE; type++) {
1490                 err = read_normal_summaries(sbi, type);
1491                 if (err)
1492                         return err;
1493         }
1494
1495         return 0;
1496 }
1497
1498 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1499 {
1500         struct page *page;
1501         unsigned char *kaddr;
1502         struct f2fs_summary *summary;
1503         struct curseg_info *seg_i;
1504         int written_size = 0;
1505         int i, j;
1506
1507         page = grab_meta_page(sbi, blkaddr++);
1508         kaddr = (unsigned char *)page_address(page);
1509
1510         /* Step 1: write nat cache */
1511         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1512         memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1513         written_size += SUM_JOURNAL_SIZE;
1514
1515         /* Step 2: write sit cache */
1516         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1517         memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1518                                                 SUM_JOURNAL_SIZE);
1519         written_size += SUM_JOURNAL_SIZE;
1520
1521         /* Step 3: write summary entries */
1522         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1523                 unsigned short blkoff;
1524                 seg_i = CURSEG_I(sbi, i);
1525                 if (sbi->ckpt->alloc_type[i] == SSR)
1526                         blkoff = sbi->blocks_per_seg;
1527                 else
1528                         blkoff = curseg_blkoff(sbi, i);
1529
1530                 for (j = 0; j < blkoff; j++) {
1531                         if (!page) {
1532                                 page = grab_meta_page(sbi, blkaddr++);
1533                                 kaddr = (unsigned char *)page_address(page);
1534                                 written_size = 0;
1535                         }
1536                         summary = (struct f2fs_summary *)(kaddr + written_size);
1537                         *summary = seg_i->sum_blk->entries[j];
1538                         written_size += SUMMARY_SIZE;
1539
1540                         if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1541                                                         SUM_FOOTER_SIZE)
1542                                 continue;
1543
1544                         set_page_dirty(page);
1545                         f2fs_put_page(page, 1);
1546                         page = NULL;
1547                 }
1548         }
1549         if (page) {
1550                 set_page_dirty(page);
1551                 f2fs_put_page(page, 1);
1552         }
1553 }
1554
1555 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1556                                         block_t blkaddr, int type)
1557 {
1558         int i, end;
1559         if (IS_DATASEG(type))
1560                 end = type + NR_CURSEG_DATA_TYPE;
1561         else
1562                 end = type + NR_CURSEG_NODE_TYPE;
1563
1564         for (i = type; i < end; i++) {
1565                 struct curseg_info *sum = CURSEG_I(sbi, i);
1566                 mutex_lock(&sum->curseg_mutex);
1567                 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1568                 mutex_unlock(&sum->curseg_mutex);
1569         }
1570 }
1571
1572 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1573 {
1574         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1575                 write_compacted_summaries(sbi, start_blk);
1576         else
1577                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1578 }
1579
1580 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1581 {
1582         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1583 }
1584
1585 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1586                                         unsigned int val, int alloc)
1587 {
1588         int i;
1589
1590         if (type == NAT_JOURNAL) {
1591                 for (i = 0; i < nats_in_cursum(sum); i++) {
1592                         if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1593                                 return i;
1594                 }
1595                 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1596                         return update_nats_in_cursum(sum, 1);
1597         } else if (type == SIT_JOURNAL) {
1598                 for (i = 0; i < sits_in_cursum(sum); i++)
1599                         if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1600                                 return i;
1601                 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1602                         return update_sits_in_cursum(sum, 1);
1603         }
1604         return -1;
1605 }
1606
1607 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1608                                         unsigned int segno)
1609 {
1610         return get_meta_page(sbi, current_sit_addr(sbi, segno));
1611 }
1612
1613 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1614                                         unsigned int start)
1615 {
1616         struct sit_info *sit_i = SIT_I(sbi);
1617         struct page *src_page, *dst_page;
1618         pgoff_t src_off, dst_off;
1619         void *src_addr, *dst_addr;
1620
1621         src_off = current_sit_addr(sbi, start);
1622         dst_off = next_sit_addr(sbi, src_off);
1623
1624         /* get current sit block page without lock */
1625         src_page = get_meta_page(sbi, src_off);
1626         dst_page = grab_meta_page(sbi, dst_off);
1627         f2fs_bug_on(sbi, PageDirty(src_page));
1628
1629         src_addr = page_address(src_page);
1630         dst_addr = page_address(dst_page);
1631         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1632
1633         set_page_dirty(dst_page);
1634         f2fs_put_page(src_page, 1);
1635
1636         set_to_next_sit(sit_i, start);
1637
1638         return dst_page;
1639 }
1640
1641 static struct sit_entry_set *grab_sit_entry_set(void)
1642 {
1643         struct sit_entry_set *ses =
1644                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_ATOMIC);
1645
1646         ses->entry_cnt = 0;
1647         INIT_LIST_HEAD(&ses->set_list);
1648         return ses;
1649 }
1650
1651 static void release_sit_entry_set(struct sit_entry_set *ses)
1652 {
1653         list_del(&ses->set_list);
1654         kmem_cache_free(sit_entry_set_slab, ses);
1655 }
1656
1657 static void adjust_sit_entry_set(struct sit_entry_set *ses,
1658                                                 struct list_head *head)
1659 {
1660         struct sit_entry_set *next = ses;
1661
1662         if (list_is_last(&ses->set_list, head))
1663                 return;
1664
1665         list_for_each_entry_continue(next, head, set_list)
1666                 if (ses->entry_cnt <= next->entry_cnt)
1667                         break;
1668
1669         list_move_tail(&ses->set_list, &next->set_list);
1670 }
1671
1672 static void add_sit_entry(unsigned int segno, struct list_head *head)
1673 {
1674         struct sit_entry_set *ses;
1675         unsigned int start_segno = START_SEGNO(segno);
1676
1677         list_for_each_entry(ses, head, set_list) {
1678                 if (ses->start_segno == start_segno) {
1679                         ses->entry_cnt++;
1680                         adjust_sit_entry_set(ses, head);
1681                         return;
1682                 }
1683         }
1684
1685         ses = grab_sit_entry_set();
1686
1687         ses->start_segno = start_segno;
1688         ses->entry_cnt++;
1689         list_add(&ses->set_list, head);
1690 }
1691
1692 static void add_sits_in_set(struct f2fs_sb_info *sbi)
1693 {
1694         struct f2fs_sm_info *sm_info = SM_I(sbi);
1695         struct list_head *set_list = &sm_info->sit_entry_set;
1696         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
1697         unsigned int segno;
1698
1699         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
1700                 add_sit_entry(segno, set_list);
1701 }
1702
1703 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
1704 {
1705         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1706         struct f2fs_summary_block *sum = curseg->sum_blk;
1707         int i;
1708
1709         for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1710                 unsigned int segno;
1711                 bool dirtied;
1712
1713                 segno = le32_to_cpu(segno_in_journal(sum, i));
1714                 dirtied = __mark_sit_entry_dirty(sbi, segno);
1715
1716                 if (!dirtied)
1717                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
1718         }
1719         update_sits_in_cursum(sum, -sits_in_cursum(sum));
1720 }
1721
1722 /*
1723  * CP calls this function, which flushes SIT entries including sit_journal,
1724  * and moves prefree segs to free segs.
1725  */
1726 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1727 {
1728         struct sit_info *sit_i = SIT_I(sbi);
1729         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1730         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1731         struct f2fs_summary_block *sum = curseg->sum_blk;
1732         struct sit_entry_set *ses, *tmp;
1733         struct list_head *head = &SM_I(sbi)->sit_entry_set;
1734         bool to_journal = true;
1735         struct seg_entry *se;
1736
1737         mutex_lock(&curseg->curseg_mutex);
1738         mutex_lock(&sit_i->sentry_lock);
1739
1740         if (!sit_i->dirty_sentries)
1741                 goto out;
1742
1743         /*
1744          * add and account sit entries of dirty bitmap in sit entry
1745          * set temporarily
1746          */
1747         add_sits_in_set(sbi);
1748
1749         /*
1750          * if there are no enough space in journal to store dirty sit
1751          * entries, remove all entries from journal and add and account
1752          * them in sit entry set.
1753          */
1754         if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL))
1755                 remove_sits_in_journal(sbi);
1756
1757         /*
1758          * there are two steps to flush sit entries:
1759          * #1, flush sit entries to journal in current cold data summary block.
1760          * #2, flush sit entries to sit page.
1761          */
1762         list_for_each_entry_safe(ses, tmp, head, set_list) {
1763                 struct page *page = NULL;
1764                 struct f2fs_sit_block *raw_sit = NULL;
1765                 unsigned int start_segno = ses->start_segno;
1766                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
1767                                                 (unsigned long)MAIN_SEGS(sbi));
1768                 unsigned int segno = start_segno;
1769
1770                 if (to_journal &&
1771                         !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL))
1772                         to_journal = false;
1773
1774                 if (!to_journal) {
1775                         page = get_next_sit_page(sbi, start_segno);
1776                         raw_sit = page_address(page);
1777                 }
1778
1779                 /* flush dirty sit entries in region of current sit set */
1780                 for_each_set_bit_from(segno, bitmap, end) {
1781                         int offset, sit_offset;
1782
1783                         se = get_seg_entry(sbi, segno);
1784
1785                         /* add discard candidates */
1786                         if (cpc->reason != CP_DISCARD) {
1787                                 cpc->trim_start = segno;
1788                                 add_discard_addrs(sbi, cpc);
1789                         }
1790
1791                         if (to_journal) {
1792                                 offset = lookup_journal_in_cursum(sum,
1793                                                         SIT_JOURNAL, segno, 1);
1794                                 f2fs_bug_on(sbi, offset < 0);
1795                                 segno_in_journal(sum, offset) =
1796                                                         cpu_to_le32(segno);
1797                                 seg_info_to_raw_sit(se,
1798                                                 &sit_in_journal(sum, offset));
1799                         } else {
1800                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1801                                 seg_info_to_raw_sit(se,
1802                                                 &raw_sit->entries[sit_offset]);
1803                         }
1804
1805                         __clear_bit(segno, bitmap);
1806                         sit_i->dirty_sentries--;
1807                         ses->entry_cnt--;
1808                 }
1809
1810                 if (!to_journal)
1811                         f2fs_put_page(page, 1);
1812
1813                 f2fs_bug_on(sbi, ses->entry_cnt);
1814                 release_sit_entry_set(ses);
1815         }
1816
1817         f2fs_bug_on(sbi, !list_empty(head));
1818         f2fs_bug_on(sbi, sit_i->dirty_sentries);
1819 out:
1820         if (cpc->reason == CP_DISCARD) {
1821                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
1822                         add_discard_addrs(sbi, cpc);
1823         }
1824         mutex_unlock(&sit_i->sentry_lock);
1825         mutex_unlock(&curseg->curseg_mutex);
1826
1827         set_prefree_as_free_segments(sbi);
1828 }
1829
1830 static int build_sit_info(struct f2fs_sb_info *sbi)
1831 {
1832         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1833         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1834         struct sit_info *sit_i;
1835         unsigned int sit_segs, start;
1836         char *src_bitmap, *dst_bitmap;
1837         unsigned int bitmap_size;
1838
1839         /* allocate memory for SIT information */
1840         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1841         if (!sit_i)
1842                 return -ENOMEM;
1843
1844         SM_I(sbi)->sit_info = sit_i;
1845
1846         sit_i->sentries = vzalloc(MAIN_SEGS(sbi) * sizeof(struct seg_entry));
1847         if (!sit_i->sentries)
1848                 return -ENOMEM;
1849
1850         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
1851         sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1852         if (!sit_i->dirty_sentries_bitmap)
1853                 return -ENOMEM;
1854
1855         for (start = 0; start < MAIN_SEGS(sbi); start++) {
1856                 sit_i->sentries[start].cur_valid_map
1857                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1858                 sit_i->sentries[start].ckpt_valid_map
1859                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1860                 if (!sit_i->sentries[start].cur_valid_map
1861                                 || !sit_i->sentries[start].ckpt_valid_map)
1862                         return -ENOMEM;
1863         }
1864
1865         sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1866         if (!sit_i->tmp_map)
1867                 return -ENOMEM;
1868
1869         if (sbi->segs_per_sec > 1) {
1870                 sit_i->sec_entries = vzalloc(MAIN_SECS(sbi) *
1871                                         sizeof(struct sec_entry));
1872                 if (!sit_i->sec_entries)
1873                         return -ENOMEM;
1874         }
1875
1876         /* get information related with SIT */
1877         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1878
1879         /* setup SIT bitmap from ckeckpoint pack */
1880         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1881         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1882
1883         dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1884         if (!dst_bitmap)
1885                 return -ENOMEM;
1886
1887         /* init SIT information */
1888         sit_i->s_ops = &default_salloc_ops;
1889
1890         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1891         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1892         sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1893         sit_i->sit_bitmap = dst_bitmap;
1894         sit_i->bitmap_size = bitmap_size;
1895         sit_i->dirty_sentries = 0;
1896         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1897         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1898         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1899         mutex_init(&sit_i->sentry_lock);
1900         return 0;
1901 }
1902
1903 static int build_free_segmap(struct f2fs_sb_info *sbi)
1904 {
1905         struct free_segmap_info *free_i;
1906         unsigned int bitmap_size, sec_bitmap_size;
1907
1908         /* allocate memory for free segmap information */
1909         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1910         if (!free_i)
1911                 return -ENOMEM;
1912
1913         SM_I(sbi)->free_info = free_i;
1914
1915         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
1916         free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1917         if (!free_i->free_segmap)
1918                 return -ENOMEM;
1919
1920         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
1921         free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1922         if (!free_i->free_secmap)
1923                 return -ENOMEM;
1924
1925         /* set all segments as dirty temporarily */
1926         memset(free_i->free_segmap, 0xff, bitmap_size);
1927         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1928
1929         /* init free segmap information */
1930         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
1931         free_i->free_segments = 0;
1932         free_i->free_sections = 0;
1933         spin_lock_init(&free_i->segmap_lock);
1934         return 0;
1935 }
1936
1937 static int build_curseg(struct f2fs_sb_info *sbi)
1938 {
1939         struct curseg_info *array;
1940         int i;
1941
1942         array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
1943         if (!array)
1944                 return -ENOMEM;
1945
1946         SM_I(sbi)->curseg_array = array;
1947
1948         for (i = 0; i < NR_CURSEG_TYPE; i++) {
1949                 mutex_init(&array[i].curseg_mutex);
1950                 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1951                 if (!array[i].sum_blk)
1952                         return -ENOMEM;
1953                 array[i].segno = NULL_SEGNO;
1954                 array[i].next_blkoff = 0;
1955         }
1956         return restore_curseg_summaries(sbi);
1957 }
1958
1959 static void build_sit_entries(struct f2fs_sb_info *sbi)
1960 {
1961         struct sit_info *sit_i = SIT_I(sbi);
1962         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1963         struct f2fs_summary_block *sum = curseg->sum_blk;
1964         int sit_blk_cnt = SIT_BLK_CNT(sbi);
1965         unsigned int i, start, end;
1966         unsigned int readed, start_blk = 0;
1967         int nrpages = MAX_BIO_BLOCKS(sbi);
1968
1969         do {
1970                 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT);
1971
1972                 start = start_blk * sit_i->sents_per_block;
1973                 end = (start_blk + readed) * sit_i->sents_per_block;
1974
1975                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
1976                         struct seg_entry *se = &sit_i->sentries[start];
1977                         struct f2fs_sit_block *sit_blk;
1978                         struct f2fs_sit_entry sit;
1979                         struct page *page;
1980
1981                         mutex_lock(&curseg->curseg_mutex);
1982                         for (i = 0; i < sits_in_cursum(sum); i++) {
1983                                 if (le32_to_cpu(segno_in_journal(sum, i))
1984                                                                 == start) {
1985                                         sit = sit_in_journal(sum, i);
1986                                         mutex_unlock(&curseg->curseg_mutex);
1987                                         goto got_it;
1988                                 }
1989                         }
1990                         mutex_unlock(&curseg->curseg_mutex);
1991
1992                         page = get_current_sit_page(sbi, start);
1993                         sit_blk = (struct f2fs_sit_block *)page_address(page);
1994                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1995                         f2fs_put_page(page, 1);
1996 got_it:
1997                         check_block_count(sbi, start, &sit);
1998                         seg_info_from_raw_sit(se, &sit);
1999                         if (sbi->segs_per_sec > 1) {
2000                                 struct sec_entry *e = get_sec_entry(sbi, start);
2001                                 e->valid_blocks += se->valid_blocks;
2002                         }
2003                 }
2004                 start_blk += readed;
2005         } while (start_blk < sit_blk_cnt);
2006 }
2007
2008 static void init_free_segmap(struct f2fs_sb_info *sbi)
2009 {
2010         unsigned int start;
2011         int type;
2012
2013         for (start = 0; start < MAIN_SEGS(sbi); start++) {
2014                 struct seg_entry *sentry = get_seg_entry(sbi, start);
2015                 if (!sentry->valid_blocks)
2016                         __set_free(sbi, start);
2017         }
2018
2019         /* set use the current segments */
2020         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2021                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2022                 __set_test_and_inuse(sbi, curseg_t->segno);
2023         }
2024 }
2025
2026 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2027 {
2028         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2029         struct free_segmap_info *free_i = FREE_I(sbi);
2030         unsigned int segno = 0, offset = 0;
2031         unsigned short valid_blocks;
2032
2033         while (1) {
2034                 /* find dirty segment based on free segmap */
2035                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2036                 if (segno >= MAIN_SEGS(sbi))
2037                         break;
2038                 offset = segno + 1;
2039                 valid_blocks = get_valid_blocks(sbi, segno, 0);
2040                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2041                         continue;
2042                 if (valid_blocks > sbi->blocks_per_seg) {
2043                         f2fs_bug_on(sbi, 1);
2044                         continue;
2045                 }
2046                 mutex_lock(&dirty_i->seglist_lock);
2047                 __locate_dirty_segment(sbi, segno, DIRTY);
2048                 mutex_unlock(&dirty_i->seglist_lock);
2049         }
2050 }
2051
2052 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2053 {
2054         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2055         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2056
2057         dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
2058         if (!dirty_i->victim_secmap)
2059                 return -ENOMEM;
2060         return 0;
2061 }
2062
2063 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2064 {
2065         struct dirty_seglist_info *dirty_i;
2066         unsigned int bitmap_size, i;
2067
2068         /* allocate memory for dirty segments list information */
2069         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2070         if (!dirty_i)
2071                 return -ENOMEM;
2072
2073         SM_I(sbi)->dirty_info = dirty_i;
2074         mutex_init(&dirty_i->seglist_lock);
2075
2076         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2077
2078         for (i = 0; i < NR_DIRTY_TYPE; i++) {
2079                 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
2080                 if (!dirty_i->dirty_segmap[i])
2081                         return -ENOMEM;
2082         }
2083
2084         init_dirty_segmap(sbi);
2085         return init_victim_secmap(sbi);
2086 }
2087
2088 /*
2089  * Update min, max modified time for cost-benefit GC algorithm
2090  */
2091 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2092 {
2093         struct sit_info *sit_i = SIT_I(sbi);
2094         unsigned int segno;
2095
2096         mutex_lock(&sit_i->sentry_lock);
2097
2098         sit_i->min_mtime = LLONG_MAX;
2099
2100         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2101                 unsigned int i;
2102                 unsigned long long mtime = 0;
2103
2104                 for (i = 0; i < sbi->segs_per_sec; i++)
2105                         mtime += get_seg_entry(sbi, segno + i)->mtime;
2106
2107                 mtime = div_u64(mtime, sbi->segs_per_sec);
2108
2109                 if (sit_i->min_mtime > mtime)
2110                         sit_i->min_mtime = mtime;
2111         }
2112         sit_i->max_mtime = get_mtime(sbi);
2113         mutex_unlock(&sit_i->sentry_lock);
2114 }
2115
2116 int build_segment_manager(struct f2fs_sb_info *sbi)
2117 {
2118         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2119         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2120         struct f2fs_sm_info *sm_info;
2121         int err;
2122
2123         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2124         if (!sm_info)
2125                 return -ENOMEM;
2126
2127         /* init sm info */
2128         sbi->sm_info = sm_info;
2129         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2130         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2131         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2132         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2133         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2134         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2135         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2136         sm_info->rec_prefree_segments = sm_info->main_segments *
2137                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2138         sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2139         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2140         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2141
2142         INIT_LIST_HEAD(&sm_info->discard_list);
2143         sm_info->nr_discards = 0;
2144         sm_info->max_discards = 0;
2145
2146         sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2147
2148         INIT_LIST_HEAD(&sm_info->sit_entry_set);
2149
2150         if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2151                 err = create_flush_cmd_control(sbi);
2152                 if (err)
2153                         return err;
2154         }
2155
2156         err = build_sit_info(sbi);
2157         if (err)
2158                 return err;
2159         err = build_free_segmap(sbi);
2160         if (err)
2161                 return err;
2162         err = build_curseg(sbi);
2163         if (err)
2164                 return err;
2165
2166         /* reinit free segmap based on SIT */
2167         build_sit_entries(sbi);
2168
2169         init_free_segmap(sbi);
2170         err = build_dirty_segmap(sbi);
2171         if (err)
2172                 return err;
2173
2174         init_min_max_mtime(sbi);
2175         return 0;
2176 }
2177
2178 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2179                 enum dirty_type dirty_type)
2180 {
2181         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2182
2183         mutex_lock(&dirty_i->seglist_lock);
2184         kfree(dirty_i->dirty_segmap[dirty_type]);
2185         dirty_i->nr_dirty[dirty_type] = 0;
2186         mutex_unlock(&dirty_i->seglist_lock);
2187 }
2188
2189 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2190 {
2191         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2192         kfree(dirty_i->victim_secmap);
2193 }
2194
2195 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2196 {
2197         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2198         int i;
2199
2200         if (!dirty_i)
2201                 return;
2202
2203         /* discard pre-free/dirty segments list */
2204         for (i = 0; i < NR_DIRTY_TYPE; i++)
2205                 discard_dirty_segmap(sbi, i);
2206
2207         destroy_victim_secmap(sbi);
2208         SM_I(sbi)->dirty_info = NULL;
2209         kfree(dirty_i);
2210 }
2211
2212 static void destroy_curseg(struct f2fs_sb_info *sbi)
2213 {
2214         struct curseg_info *array = SM_I(sbi)->curseg_array;
2215         int i;
2216
2217         if (!array)
2218                 return;
2219         SM_I(sbi)->curseg_array = NULL;
2220         for (i = 0; i < NR_CURSEG_TYPE; i++)
2221                 kfree(array[i].sum_blk);
2222         kfree(array);
2223 }
2224
2225 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2226 {
2227         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2228         if (!free_i)
2229                 return;
2230         SM_I(sbi)->free_info = NULL;
2231         kfree(free_i->free_segmap);
2232         kfree(free_i->free_secmap);
2233         kfree(free_i);
2234 }
2235
2236 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2237 {
2238         struct sit_info *sit_i = SIT_I(sbi);
2239         unsigned int start;
2240
2241         if (!sit_i)
2242                 return;
2243
2244         if (sit_i->sentries) {
2245                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2246                         kfree(sit_i->sentries[start].cur_valid_map);
2247                         kfree(sit_i->sentries[start].ckpt_valid_map);
2248                 }
2249         }
2250         kfree(sit_i->tmp_map);
2251
2252         vfree(sit_i->sentries);
2253         vfree(sit_i->sec_entries);
2254         kfree(sit_i->dirty_sentries_bitmap);
2255
2256         SM_I(sbi)->sit_info = NULL;
2257         kfree(sit_i->sit_bitmap);
2258         kfree(sit_i);
2259 }
2260
2261 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2262 {
2263         struct f2fs_sm_info *sm_info = SM_I(sbi);
2264
2265         if (!sm_info)
2266                 return;
2267         destroy_flush_cmd_control(sbi);
2268         destroy_dirty_segmap(sbi);
2269         destroy_curseg(sbi);
2270         destroy_free_segmap(sbi);
2271         destroy_sit_info(sbi);
2272         sbi->sm_info = NULL;
2273         kfree(sm_info);
2274 }
2275
2276 int __init create_segment_manager_caches(void)
2277 {
2278         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2279                         sizeof(struct discard_entry));
2280         if (!discard_entry_slab)
2281                 goto fail;
2282
2283         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2284                         sizeof(struct sit_entry_set));
2285         if (!sit_entry_set_slab)
2286                 goto destory_discard_entry;
2287
2288         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2289                         sizeof(struct inmem_pages));
2290         if (!inmem_entry_slab)
2291                 goto destroy_sit_entry_set;
2292         return 0;
2293
2294 destroy_sit_entry_set:
2295         kmem_cache_destroy(sit_entry_set_slab);
2296 destory_discard_entry:
2297         kmem_cache_destroy(discard_entry_slab);
2298 fail:
2299         return -ENOMEM;
2300 }
2301
2302 void destroy_segment_manager_caches(void)
2303 {
2304         kmem_cache_destroy(sit_entry_set_slab);
2305         kmem_cache_destroy(discard_entry_slab);
2306         kmem_cache_destroy(inmem_entry_slab);
2307 }