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         while (!f2fs_test_bit(offset, (unsigned char *)addr))
  79                 offset++;
  80
  81         if (offset > size)
  82                 offset = size;
  83
  84         return offset;
  85 #if 0
  86         const unsigned long *p = addr + BIT_WORD(offset);
  87         unsigned long result = offset & ~(BITS_PER_LONG - 1);
  88         unsigned long tmp;
  89         unsigned long mask, submask;
  90         unsigned long quot, rest;
  91
  92         if (offset >= size)
  93                 return size;
  94
  95         size -= result;
  96         offset %= BITS_PER_LONG;
  97         if (!offset)
  98                 goto aligned;
  99
 100         tmp = *(p++);
 101         quot = (offset >> 3) << 3;
 102         rest = offset & 0x7;
 103         mask = ~0UL << quot;
 104         submask = (unsigned char)(0xff << rest) >> rest;
 105         submask <<= quot;
 106         mask &= submask;
 107         tmp &= mask;
 108         if (size < BITS_PER_LONG)
 109                 goto found_first;
 110         if (tmp)
 111                 goto found_middle;
 112
 113         size -= BITS_PER_LONG;
 114         result += BITS_PER_LONG;
 115 aligned:
 116         while (size & ~(BITS_PER_LONG-1)) {
 117                 tmp = *(p++);
 118                 if (tmp)
 119                         goto found_middle;
 120                 result += BITS_PER_LONG;
 121                 size -= BITS_PER_LONG;
 122         }
 123         if (!size)
 124                 return result;
 125         tmp = *p;
 126 found_first:
 127         tmp &= (~0UL >> (BITS_PER_LONG - size));
 128         if (tmp == 0UL)         /* Are any bits set? */
 129                 return result + size;   /* Nope. */
 130 found_middle:
 131         return result + __reverse_ffs(tmp);
 132 #endif
 133 }
 134
 135 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 136                         unsigned long size, unsigned long offset)
 137 {
 138         while (f2fs_test_bit(offset, (unsigned char *)addr))
 139                 offset++;
 140
 141         if (offset > size)
 142                 offset = size;
 143
 144         return offset;
 145 #if 0
 146         const unsigned long *p = addr + BIT_WORD(offset);
 147         unsigned long result = offset & ~(BITS_PER_LONG - 1);
 148         unsigned long tmp;
 149         unsigned long mask, submask;
 150         unsigned long quot, rest;
 151
 152         if (offset >= size)
 153                 return size;
 154
 155         size -= result;
 156         offset %= BITS_PER_LONG;
 157         if (!offset)
 158                 goto aligned;
 159
 160         tmp = *(p++);
 161         quot = (offset >> 3) << 3;
 162         rest = offset & 0x7;
 163         mask = ~(~0UL << quot);
 164         submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest);
 165         submask <<= quot;
 166         mask += submask;
 167         tmp |= mask;
 168         if (size < BITS_PER_LONG)
 169                 goto found_first;
 170         if (~tmp)
 171                 goto found_middle;
 172
 173         size -= BITS_PER_LONG;
 174         result += BITS_PER_LONG;
 175 aligned:
 176         while (size & ~(BITS_PER_LONG - 1)) {
 177                 tmp = *(p++);
 178                 if (~tmp)
 179                         goto found_middle;
 180                 result += BITS_PER_LONG;
 181                 size -= BITS_PER_LONG;
 182         }
 183         if (!size)
 184                 return result;
 185         tmp = *p;
 186
 187 found_first:
 188         tmp |= ~0UL << size;
 189         if (tmp == ~0UL)        /* Are any bits zero? */
 190                 return result + size;   /* Nope. */
 191 found_middle:
 192         return result + __reverse_ffz(tmp);
 193 #endif
 194 }
 195
 196 void register_inmem_page(struct inode *inode, struct page *page)
 197 {
 198         struct f2fs_inode_info *fi = F2FS_I(inode);
 199         struct inmem_pages *new;
 200         int err;
 201
 202         SetPagePrivate(page);
 203         f2fs_trace_pid(page);
 204
 205         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 206
 207         /* add atomic page indices to the list */
 208         new->page = page;
 209         INIT_LIST_HEAD(&new->list);
 210 retry:
 211         /* increase reference count with clean state */
 212         mutex_lock(&fi->inmem_lock);
 213         err = radix_tree_insert(&fi->inmem_root, page->index, new);
 214         if (err == -EEXIST) {
 215                 mutex_unlock(&fi->inmem_lock);
 216                 kmem_cache_free(inmem_entry_slab, new);
 217                 return;
 218         } else if (err) {
 219                 mutex_unlock(&fi->inmem_lock);
 220                 goto retry;
 221         }
 222         get_page(page);
 223         list_add_tail(&new->list, &fi->inmem_pages);
 224         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 225         mutex_unlock(&fi->inmem_lock);
 226
 227         trace_f2fs_register_inmem_page(page, INMEM);
 228 }
 229
 230 void commit_inmem_pages(struct inode *inode, bool abort)
 231 {
 232         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 233         struct f2fs_inode_info *fi = F2FS_I(inode);
 234         struct inmem_pages *cur, *tmp;
 235         bool submit_bio = false;
 236         struct f2fs_io_info fio = {
 237                 .sbi = sbi,
 238                 .type = DATA,
 239                 .rw = WRITE_SYNC | REQ_PRIO,
 240                 .encrypted_page = NULL,
 241         };
 242
 243         /*
 244          * The abort is true only when f2fs_evict_inode is called.
 245          * Basically, the f2fs_evict_inode doesn't produce any data writes, so
 246          * that we don't need to call f2fs_balance_fs.
 247          * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this
 248          * inode becomes free by iget_locked in f2fs_iget.
 249          */
 250         if (!abort) {
 251                 f2fs_balance_fs(sbi);
 252                 f2fs_lock_op(sbi);
 253         }
 254
 255         mutex_lock(&fi->inmem_lock);
 256         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 257                 if (!abort) {
 258                         lock_page(cur->page);
 259                         if (cur->page->mapping == inode->i_mapping) {
 260                                 set_page_dirty(cur->page);
 261                                 f2fs_wait_on_page_writeback(cur->page, DATA);
 262                                 if (clear_page_dirty_for_io(cur->page))
 263                                         inode_dec_dirty_pages(inode);
 264                                 trace_f2fs_commit_inmem_page(cur->page, INMEM);
 265                                 fio.page = cur->page;
 266                                 do_write_data_page(&fio);
 267                                 submit_bio = true;
 268                         }
 269                         f2fs_put_page(cur->page, 1);
 270                 } else {
 271                         trace_f2fs_commit_inmem_page(cur->page, INMEM_DROP);
 272                         put_page(cur->page);
 273                 }
 274                 radix_tree_delete(&fi->inmem_root, cur->page->index);
 275                 list_del(&cur->list);
 276                 kmem_cache_free(inmem_entry_slab, cur);
 277                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 278         }
 279         mutex_unlock(&fi->inmem_lock);
 280
 281         if (!abort) {
 282                 f2fs_unlock_op(sbi);
 283                 if (submit_bio)
 284                         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 285         }
 286 }
 287
 288 /*
 289  * This function balances dirty node and dentry pages.
 290  * In addition, it controls garbage collection.
 291  */
 292 void f2fs_balance_fs(struct f2fs_sb_info *sbi)
 293 {
 294         /*
 295          * We should do GC or end up with checkpoint, if there are so many dirty
 296          * dir/node pages without enough free segments.
 297          */
 298         if (has_not_enough_free_secs(sbi, 0)) {
 299                 mutex_lock(&sbi->gc_mutex);
 300                 f2fs_gc(sbi);
 301         }
 302 }
 303
 304 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 305 {
 306         /* try to shrink extent cache when there is no enough memory */
 307         f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 308
 309         /* check the # of cached NAT entries and prefree segments */
 310         if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) ||
 311                         excess_prefree_segs(sbi) ||
 312                         !available_free_memory(sbi, INO_ENTRIES))
 313                 f2fs_sync_fs(sbi->sb, true);
 314 }
 315
 316 static int issue_flush_thread(void *data)
 317 {
 318         struct f2fs_sb_info *sbi = data;
 319         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 320         wait_queue_head_t *q = &fcc->flush_wait_queue;
 321 repeat:
 322         if (kthread_should_stop())
 323                 return 0;
 324
 325         if (!llist_empty(&fcc->issue_list)) {
 326                 struct bio *bio = bio_alloc(GFP_NOIO, 0);
 327                 struct flush_cmd *cmd, *next;
 328                 int ret;
 329
 330                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 331                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 332
 333                 bio->bi_bdev = sbi->sb->s_bdev;
 334                 ret = submit_bio_wait(WRITE_FLUSH, bio);
 335
 336                 llist_for_each_entry_safe(cmd, next,
 337                                           fcc->dispatch_list, llnode) {
 338                         cmd->ret = ret;
 339                         complete(&cmd->wait);
 340                 }
 341                 bio_put(bio);
 342                 fcc->dispatch_list = NULL;
 343         }
 344
 345         wait_event_interruptible(*q,
 346                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 347         goto repeat;
 348 }
 349
 350 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
 351 {
 352         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 353         struct flush_cmd cmd;
 354
 355         trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
 356                                         test_opt(sbi, FLUSH_MERGE));
 357
 358         if (test_opt(sbi, NOBARRIER))
 359                 return 0;
 360
 361         if (!test_opt(sbi, FLUSH_MERGE))
 362                 return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL);
 363
 364         init_completion(&cmd.wait);
 365
 366         llist_add(&cmd.llnode, &fcc->issue_list);
 367
 368         if (!fcc->dispatch_list)
 369                 wake_up(&fcc->flush_wait_queue);
 370
 371         wait_for_completion(&cmd.wait);
 372
 373         return cmd.ret;
 374 }
 375
 376 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
 377 {
 378         dev_t dev = sbi->sb->s_bdev->bd_dev;
 379         struct flush_cmd_control *fcc;
 380         int err = 0;
 381
 382         fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
 383         if (!fcc)
 384                 return -ENOMEM;
 385         init_waitqueue_head(&fcc->flush_wait_queue);
 386         init_llist_head(&fcc->issue_list);
 387         SM_I(sbi)->cmd_control_info = fcc;
 388         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 389                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 390         if (IS_ERR(fcc->f2fs_issue_flush)) {
 391                 err = PTR_ERR(fcc->f2fs_issue_flush);
 392                 kfree(fcc);
 393                 SM_I(sbi)->cmd_control_info = NULL;
 394                 return err;
 395         }
 396
 397         return err;
 398 }
 399
 400 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
 401 {
 402         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 403
 404         if (fcc && fcc->f2fs_issue_flush)
 405                 kthread_stop(fcc->f2fs_issue_flush);
 406         kfree(fcc);
 407         SM_I(sbi)->cmd_control_info = NULL;
 408 }
 409
 410 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 411                 enum dirty_type dirty_type)
 412 {
 413         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 414
 415         /* need not be added */
 416         if (IS_CURSEG(sbi, segno))
 417                 return;
 418
 419         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 420                 dirty_i->nr_dirty[dirty_type]++;
 421
 422         if (dirty_type == DIRTY) {
 423                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 424                 enum dirty_type t = sentry->type;
 425
 426                 if (unlikely(t >= DIRTY)) {
 427                         f2fs_bug_on(sbi, 1);
 428                         return;
 429                 }
 430                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 431                         dirty_i->nr_dirty[t]++;
 432         }
 433 }
 434
 435 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 436                 enum dirty_type dirty_type)
 437 {
 438         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 439
 440         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 441                 dirty_i->nr_dirty[dirty_type]--;
 442
 443         if (dirty_type == DIRTY) {
 444                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 445                 enum dirty_type t = sentry->type;
 446
 447                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 448                         dirty_i->nr_dirty[t]--;
 449
 450                 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
 451                         clear_bit(GET_SECNO(sbi, segno),
 452                                                 dirty_i->victim_secmap);
 453         }
 454 }
 455
 456 /*
 457  * Should not occur error such as -ENOMEM.
 458  * Adding dirty entry into seglist is not critical operation.
 459  * If a given segment is one of current working segments, it won't be added.
 460  */
 461 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 462 {
 463         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 464         unsigned short valid_blocks;
 465
 466         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 467                 return;
 468
 469         mutex_lock(&dirty_i->seglist_lock);
 470
 471         valid_blocks = get_valid_blocks(sbi, segno, 0);
 472
 473         if (valid_blocks == 0) {
 474                 __locate_dirty_segment(sbi, segno, PRE);
 475                 __remove_dirty_segment(sbi, segno, DIRTY);
 476         } else if (valid_blocks < sbi->blocks_per_seg) {
 477                 __locate_dirty_segment(sbi, segno, DIRTY);
 478         } else {
 479                 /* Recovery routine with SSR needs this */
 480                 __remove_dirty_segment(sbi, segno, DIRTY);
 481         }
 482
 483         mutex_unlock(&dirty_i->seglist_lock);
 484 }
 485
 486 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
 487                                 block_t blkstart, block_t blklen)
 488 {
 489         sector_t start = SECTOR_FROM_BLOCK(blkstart);
 490         sector_t len = SECTOR_FROM_BLOCK(blklen);
 491         struct seg_entry *se;
 492         unsigned int offset;
 493         block_t i;
 494
 495         for (i = blkstart; i < blkstart + blklen; i++) {
 496                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
 497                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
 498
 499                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
 500                         sbi->discard_blks--;
 501         }
 502         trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
 503         return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
 504 }
 505
 506 void discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
 507 {
 508         int err = -ENOTSUPP;
 509
 510         if (test_opt(sbi, DISCARD)) {
 511                 struct seg_entry *se = get_seg_entry(sbi,
 512                                 GET_SEGNO(sbi, blkaddr));
 513                 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 514
 515                 if (f2fs_test_bit(offset, se->discard_map))
 516                         return;
 517
 518                 err = f2fs_issue_discard(sbi, blkaddr, 1);
 519         }
 520
 521         if (err)
 522                 update_meta_page(sbi, NULL, blkaddr);
 523 }
 524
 525 static void __add_discard_entry(struct f2fs_sb_info *sbi,
 526                 struct cp_control *cpc, struct seg_entry *se,
 527                 unsigned int start, unsigned int end)
 528 {
 529         struct list_head *head = &SM_I(sbi)->discard_list;
 530         struct discard_entry *new, *last;
 531
 532         if (!list_empty(head)) {
 533                 last = list_last_entry(head, struct discard_entry, list);
 534                 if (START_BLOCK(sbi, cpc->trim_start) + start ==
 535                                                 last->blkaddr + last->len) {
 536                         last->len += end - start;
 537                         goto done;
 538                 }
 539         }
 540
 541         new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
 542         INIT_LIST_HEAD(&new->list);
 543         new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
 544         new->len = end - start;
 545         list_add_tail(&new->list, head);
 546 done:
 547         SM_I(sbi)->nr_discards += end - start;
 548 }
 549
 550 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 551 {
 552         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
 553         int max_blocks = sbi->blocks_per_seg;
 554         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
 555         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
 556         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
 557         unsigned long *discard_map = (unsigned long *)se->discard_map;
 558         unsigned long *dmap = SIT_I(sbi)->tmp_map;
 559         unsigned int start = 0, end = -1;
 560         bool force = (cpc->reason == CP_DISCARD);
 561         int i;
 562
 563         if (se->valid_blocks == max_blocks)
 564                 return;
 565
 566         if (!force) {
 567                 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
 568                     SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
 569                         return;
 570         }
 571
 572         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
 573         for (i = 0; i < entries; i++)
 574                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
 575                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
 576
 577         while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
 578                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
 579                 if (start >= max_blocks)
 580                         break;
 581
 582                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
 583                 __add_discard_entry(sbi, cpc, se, start, end);
 584         }
 585 }
 586
 587 void release_discard_addrs(struct f2fs_sb_info *sbi)
 588 {
 589         struct list_head *head = &(SM_I(sbi)->discard_list);
 590         struct discard_entry *entry, *this;
 591
 592         /* drop caches */
 593         list_for_each_entry_safe(entry, this, head, list) {
 594                 list_del(&entry->list);
 595                 kmem_cache_free(discard_entry_slab, entry);
 596         }
 597 }
 598
 599 /*
 600  * Should call clear_prefree_segments after checkpoint is done.
 601  */
 602 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
 603 {
 604         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 605         unsigned int segno;
 606
 607         mutex_lock(&dirty_i->seglist_lock);
 608         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
 609                 __set_test_and_free(sbi, segno);
 610         mutex_unlock(&dirty_i->seglist_lock);
 611 }
 612
 613 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 614 {
 615         struct list_head *head = &(SM_I(sbi)->discard_list);
 616         struct discard_entry *entry, *this;
 617         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 618         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
 619         unsigned int start = 0, end = -1;
 620
 621         mutex_lock(&dirty_i->seglist_lock);
 622
 623         while (1) {
 624                 int i;
 625                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
 626                 if (start >= MAIN_SEGS(sbi))
 627                         break;
 628                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
 629                                                                 start + 1);
 630
 631                 for (i = start; i < end; i++)
 632                         clear_bit(i, prefree_map);
 633
 634                 dirty_i->nr_dirty[PRE] -= end - start;
 635
 636                 if (!test_opt(sbi, DISCARD))
 637                         continue;
 638
 639                 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
 640                                 (end - start) << sbi->log_blocks_per_seg);
 641         }
 642         mutex_unlock(&dirty_i->seglist_lock);
 643
 644         /* send small discards */
 645         list_for_each_entry_safe(entry, this, head, list) {
 646                 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
 647                         goto skip;
 648                 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
 649                 cpc->trimmed += entry->len;
 650 skip:
 651                 list_del(&entry->list);
 652                 SM_I(sbi)->nr_discards -= entry->len;
 653                 kmem_cache_free(discard_entry_slab, entry);
 654         }
 655 }
 656
 657 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
 658 {
 659         struct sit_info *sit_i = SIT_I(sbi);
 660
 661         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
 662                 sit_i->dirty_sentries++;
 663                 return false;
 664         }
 665
 666         return true;
 667 }
 668
 669 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
 670                                         unsigned int segno, int modified)
 671 {
 672         struct seg_entry *se = get_seg_entry(sbi, segno);
 673         se->type = type;
 674         if (modified)
 675                 __mark_sit_entry_dirty(sbi, segno);
 676 }
 677
 678 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
 679 {
 680         struct seg_entry *se;
 681         unsigned int segno, offset;
 682         long int new_vblocks;
 683
 684         segno = GET_SEGNO(sbi, blkaddr);
 685
 686         se = get_seg_entry(sbi, segno);
 687         new_vblocks = se->valid_blocks + del;
 688         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 689
 690         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
 691                                 (new_vblocks > sbi->blocks_per_seg)));
 692
 693         se->valid_blocks = new_vblocks;
 694         se->mtime = get_mtime(sbi);
 695         SIT_I(sbi)->max_mtime = se->mtime;
 696
 697         /* Update valid block bitmap */
 698         if (del > 0) {
 699                 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
 700                         f2fs_bug_on(sbi, 1);
 701                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
 702                         sbi->discard_blks--;
 703         } else {
 704                 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
 705                         f2fs_bug_on(sbi, 1);
 706                 if (f2fs_test_and_clear_bit(offset, se->discard_map))
 707                         sbi->discard_blks++;
 708         }
 709         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
 710                 se->ckpt_valid_blocks += del;
 711
 712         __mark_sit_entry_dirty(sbi, segno);
 713
 714         /* update total number of valid blocks to be written in ckpt area */
 715         SIT_I(sbi)->written_valid_blocks += del;
 716
 717         if (sbi->segs_per_sec > 1)
 718                 get_sec_entry(sbi, segno)->valid_blocks += del;
 719 }
 720
 721 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
 722 {
 723         update_sit_entry(sbi, new, 1);
 724         if (GET_SEGNO(sbi, old) != NULL_SEGNO)
 725                 update_sit_entry(sbi, old, -1);
 726
 727         locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
 728         locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
 729 }
 730
 731 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
 732 {
 733         unsigned int segno = GET_SEGNO(sbi, addr);
 734         struct sit_info *sit_i = SIT_I(sbi);
 735
 736         f2fs_bug_on(sbi, addr == NULL_ADDR);
 737         if (addr == NEW_ADDR)
 738                 return;
 739
 740         /* add it into sit main buffer */
 741         mutex_lock(&sit_i->sentry_lock);
 742
 743         update_sit_entry(sbi, addr, -1);
 744
 745         /* add it into dirty seglist */
 746         locate_dirty_segment(sbi, segno);
 747
 748         mutex_unlock(&sit_i->sentry_lock);
 749 }
 750
 751 /*
 752  * This function should be resided under the curseg_mutex lock
 753  */
 754 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
 755                                         struct f2fs_summary *sum)
 756 {
 757         struct curseg_info *curseg = CURSEG_I(sbi, type);
 758         void *addr = curseg->sum_blk;
 759         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
 760         memcpy(addr, sum, sizeof(struct f2fs_summary));
 761 }
 762
 763 /*
 764  * Calculate the number of current summary pages for writing
 765  */
 766 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
 767 {
 768         int valid_sum_count = 0;
 769         int i, sum_in_page;
 770
 771         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 772                 if (sbi->ckpt->alloc_type[i] == SSR)
 773                         valid_sum_count += sbi->blocks_per_seg;
 774                 else {
 775                         if (for_ra)
 776                                 valid_sum_count += le16_to_cpu(
 777                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
 778                         else
 779                                 valid_sum_count += curseg_blkoff(sbi, i);
 780                 }
 781         }
 782
 783         sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
 784                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
 785         if (valid_sum_count <= sum_in_page)
 786                 return 1;
 787         else if ((valid_sum_count - sum_in_page) <=
 788                 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
 789                 return 2;
 790         return 3;
 791 }
 792
 793 /*
 794  * Caller should put this summary page
 795  */
 796 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
 797 {
 798         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
 799 }
 800
 801 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
 802 {
 803         struct page *page = grab_meta_page(sbi, blk_addr);
 804         void *dst = page_address(page);
 805
 806         if (src)
 807                 memcpy(dst, src, PAGE_CACHE_SIZE);
 808         else
 809                 memset(dst, 0, PAGE_CACHE_SIZE);
 810         set_page_dirty(page);
 811         f2fs_put_page(page, 1);
 812 }
 813
 814 static void write_sum_page(struct f2fs_sb_info *sbi,
 815                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
 816 {
 817         update_meta_page(sbi, (void *)sum_blk, blk_addr);
 818 }
 819
 820 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
 821 {
 822         struct curseg_info *curseg = CURSEG_I(sbi, type);
 823         unsigned int segno = curseg->segno + 1;
 824         struct free_segmap_info *free_i = FREE_I(sbi);
 825
 826         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
 827                 return !test_bit(segno, free_i->free_segmap);
 828         return 0;
 829 }
 830
 831 /*
 832  * Find a new segment from the free segments bitmap to right order
 833  * This function should be returned with success, otherwise BUG
 834  */
 835 static void get_new_segment(struct f2fs_sb_info *sbi,
 836                         unsigned int *newseg, bool new_sec, int dir)
 837 {
 838         struct free_segmap_info *free_i = FREE_I(sbi);
 839         unsigned int segno, secno, zoneno;
 840         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
 841         unsigned int hint = *newseg / sbi->segs_per_sec;
 842         unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
 843         unsigned int left_start = hint;
 844         bool init = true;
 845         int go_left = 0;
 846         int i;
 847
 848         spin_lock(&free_i->segmap_lock);
 849
 850         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
 851                 segno = find_next_zero_bit(free_i->free_segmap,
 852                                         MAIN_SEGS(sbi), *newseg + 1);
 853                 if (segno - *newseg < sbi->segs_per_sec -
 854                                         (*newseg % sbi->segs_per_sec))
 855                         goto got_it;
 856         }
 857 find_other_zone:
 858         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
 859         if (secno >= MAIN_SECS(sbi)) {
 860                 if (dir == ALLOC_RIGHT) {
 861                         secno = find_next_zero_bit(free_i->free_secmap,
 862                                                         MAIN_SECS(sbi), 0);
 863                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
 864                 } else {
 865                         go_left = 1;
 866                         left_start = hint - 1;
 867                 }
 868         }
 869         if (go_left == 0)
 870                 goto skip_left;
 871
 872         while (test_bit(left_start, free_i->free_secmap)) {
 873                 if (left_start > 0) {
 874                         left_start--;
 875                         continue;
 876                 }
 877                 left_start = find_next_zero_bit(free_i->free_secmap,
 878                                                         MAIN_SECS(sbi), 0);
 879                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
 880                 break;
 881         }
 882         secno = left_start;
 883 skip_left:
 884         hint = secno;
 885         segno = secno * sbi->segs_per_sec;
 886         zoneno = secno / sbi->secs_per_zone;
 887
 888         /* give up on finding another zone */
 889         if (!init)
 890                 goto got_it;
 891         if (sbi->secs_per_zone == 1)
 892                 goto got_it;
 893         if (zoneno == old_zoneno)
 894                 goto got_it;
 895         if (dir == ALLOC_LEFT) {
 896                 if (!go_left && zoneno + 1 >= total_zones)
 897                         goto got_it;
 898                 if (go_left && zoneno == 0)
 899                         goto got_it;
 900         }
 901         for (i = 0; i < NR_CURSEG_TYPE; i++)
 902                 if (CURSEG_I(sbi, i)->zone == zoneno)
 903                         break;
 904
 905         if (i < NR_CURSEG_TYPE) {
 906                 /* zone is in user, try another */
 907                 if (go_left)
 908                         hint = zoneno * sbi->secs_per_zone - 1;
 909                 else if (zoneno + 1 >= total_zones)
 910                         hint = 0;
 911                 else
 912                         hint = (zoneno + 1) * sbi->secs_per_zone;
 913                 init = false;
 914                 goto find_other_zone;
 915         }
 916 got_it:
 917         /* set it as dirty segment in free segmap */
 918         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
 919         __set_inuse(sbi, segno);
 920         *newseg = segno;
 921         spin_unlock(&free_i->segmap_lock);
 922 }
 923
 924 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
 925 {
 926         struct curseg_info *curseg = CURSEG_I(sbi, type);
 927         struct summary_footer *sum_footer;
 928
 929         curseg->segno = curseg->next_segno;
 930         curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
 931         curseg->next_blkoff = 0;
 932         curseg->next_segno = NULL_SEGNO;
 933
 934         sum_footer = &(curseg->sum_blk->footer);
 935         memset(sum_footer, 0, sizeof(struct summary_footer));
 936         if (IS_DATASEG(type))
 937                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
 938         if (IS_NODESEG(type))
 939                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
 940         __set_sit_entry_type(sbi, type, curseg->segno, modified);
 941 }
 942
 943 /*
 944  * Allocate a current working segment.
 945  * This function always allocates a free segment in LFS manner.
 946  */
 947 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
 948 {
 949         struct curseg_info *curseg = CURSEG_I(sbi, type);
 950         unsigned int segno = curseg->segno;
 951         int dir = ALLOC_LEFT;
 952
 953         write_sum_page(sbi, curseg->sum_blk,
 954                                 GET_SUM_BLOCK(sbi, segno));
 955         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
 956                 dir = ALLOC_RIGHT;
 957
 958         if (test_opt(sbi, NOHEAP))
 959                 dir = ALLOC_RIGHT;
 960
 961         get_new_segment(sbi, &segno, new_sec, dir);
 962         curseg->next_segno = segno;
 963         reset_curseg(sbi, type, 1);
 964         curseg->alloc_type = LFS;
 965 }
 966
 967 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
 968                         struct curseg_info *seg, block_t start)
 969 {
 970         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
 971         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
 972         unsigned long *target_map = SIT_I(sbi)->tmp_map;
 973         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
 974         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
 975         int i, pos;
 976
 977         for (i = 0; i < entries; i++)
 978                 target_map[i] = ckpt_map[i] | cur_map[i];
 979
 980         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
 981
 982         seg->next_blkoff = pos;
 983 }
 984
 985 /*
 986  * If a segment is written by LFS manner, next block offset is just obtained
 987  * by increasing the current block offset. However, if a segment is written by
 988  * SSR manner, next block offset obtained by calling __next_free_blkoff
 989  */
 990 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
 991                                 struct curseg_info *seg)
 992 {
 993         if (seg->alloc_type == SSR)
 994                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
 995         else
 996                 seg->next_blkoff++;
 997 }
 998
 999 /*
1000  * This function always allocates a used segment(from dirty seglist) by SSR
1001  * manner, so it should recover the existing segment information of valid blocks
1002  */
1003 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1004 {
1005         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1006         struct curseg_info *curseg = CURSEG_I(sbi, type);
1007         unsigned int new_segno = curseg->next_segno;
1008         struct f2fs_summary_block *sum_node;
1009         struct page *sum_page;
1010
1011         write_sum_page(sbi, curseg->sum_blk,
1012                                 GET_SUM_BLOCK(sbi, curseg->segno));
1013         __set_test_and_inuse(sbi, new_segno);
1014
1015         mutex_lock(&dirty_i->seglist_lock);
1016         __remove_dirty_segment(sbi, new_segno, PRE);
1017         __remove_dirty_segment(sbi, new_segno, DIRTY);
1018         mutex_unlock(&dirty_i->seglist_lock);
1019
1020         reset_curseg(sbi, type, 1);
1021         curseg->alloc_type = SSR;
1022         __next_free_blkoff(sbi, curseg, 0);
1023
1024         if (reuse) {
1025                 sum_page = get_sum_page(sbi, new_segno);
1026                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1027                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1028                 f2fs_put_page(sum_page, 1);
1029         }
1030 }
1031
1032 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1033 {
1034         struct curseg_info *curseg = CURSEG_I(sbi, type);
1035         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1036
1037         if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1038                 return v_ops->get_victim(sbi,
1039                                 &(curseg)->next_segno, BG_GC, type, SSR);
1040
1041         /* For data segments, let's do SSR more intensively */
1042         for (; type >= CURSEG_HOT_DATA; type--)
1043                 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1044                                                 BG_GC, type, SSR))
1045                         return 1;
1046         return 0;
1047 }
1048
1049 /*
1050  * flush out current segment and replace it with new segment
1051  * This function should be returned with success, otherwise BUG
1052  */
1053 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1054                                                 int type, bool force)
1055 {
1056         struct curseg_info *curseg = CURSEG_I(sbi, type);
1057
1058         if (force)
1059                 new_curseg(sbi, type, true);
1060         else if (type == CURSEG_WARM_NODE)
1061                 new_curseg(sbi, type, false);
1062         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1063                 new_curseg(sbi, type, false);
1064         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1065                 change_curseg(sbi, type, true);
1066         else
1067                 new_curseg(sbi, type, false);
1068
1069         stat_inc_seg_type(sbi, curseg);
1070 }
1071
1072 static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1073 {
1074         struct curseg_info *curseg = CURSEG_I(sbi, type);
1075         unsigned int old_segno;
1076
1077         old_segno = curseg->segno;
1078         SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1079         locate_dirty_segment(sbi, old_segno);
1080 }
1081
1082 void allocate_new_segments(struct f2fs_sb_info *sbi)
1083 {
1084         int i;
1085
1086         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1087                 __allocate_new_segments(sbi, i);
1088 }
1089
1090 static const struct segment_allocation default_salloc_ops = {
1091         .allocate_segment = allocate_segment_by_default,
1092 };
1093
1094 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1095 {
1096         __u64 start = F2FS_BYTES_TO_BLK(range->start);
1097         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1098         unsigned int start_segno, end_segno;
1099         struct cp_control cpc;
1100
1101         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1102                 return -EINVAL;
1103
1104         cpc.trimmed = 0;
1105         if (end <= MAIN_BLKADDR(sbi))
1106                 goto out;
1107
1108         /* start/end segment number in main_area */
1109         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1110         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1111                                                 GET_SEGNO(sbi, end);
1112         cpc.reason = CP_DISCARD;
1113         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1114
1115         /* do checkpoint to issue discard commands safely */
1116         for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1117                 cpc.trim_start = start_segno;
1118
1119                 if (sbi->discard_blks == 0)
1120                         break;
1121                 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1122                         cpc.trim_end = end_segno;
1123                 else
1124                         cpc.trim_end = min_t(unsigned int,
1125                                 rounddown(start_segno +
1126                                 BATCHED_TRIM_SEGMENTS(sbi),
1127                                 sbi->segs_per_sec) - 1, end_segno);
1128
1129                 mutex_lock(&sbi->gc_mutex);
1130                 write_checkpoint(sbi, &cpc);
1131                 mutex_unlock(&sbi->gc_mutex);
1132         }
1133 out:
1134         range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1135         return 0;
1136 }
1137
1138 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1139 {
1140         struct curseg_info *curseg = CURSEG_I(sbi, type);
1141         if (curseg->next_blkoff < sbi->blocks_per_seg)
1142                 return true;
1143         return false;
1144 }
1145
1146 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1147 {
1148         if (p_type == DATA)
1149                 return CURSEG_HOT_DATA;
1150         else
1151                 return CURSEG_HOT_NODE;
1152 }
1153
1154 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1155 {
1156         if (p_type == DATA) {
1157                 struct inode *inode = page->mapping->host;
1158
1159                 if (S_ISDIR(inode->i_mode))
1160                         return CURSEG_HOT_DATA;
1161                 else
1162                         return CURSEG_COLD_DATA;
1163         } else {
1164                 if (IS_DNODE(page) && is_cold_node(page))
1165                         return CURSEG_WARM_NODE;
1166                 else
1167                         return CURSEG_COLD_NODE;
1168         }
1169 }
1170
1171 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1172 {
1173         if (p_type == DATA) {
1174                 struct inode *inode = page->mapping->host;
1175
1176                 if (S_ISDIR(inode->i_mode))
1177                         return CURSEG_HOT_DATA;
1178                 else if (is_cold_data(page) || file_is_cold(inode))
1179                         return CURSEG_COLD_DATA;
1180                 else
1181                         return CURSEG_WARM_DATA;
1182         } else {
1183                 if (IS_DNODE(page))
1184                         return is_cold_node(page) ? CURSEG_WARM_NODE :
1185                                                 CURSEG_HOT_NODE;
1186                 else
1187                         return CURSEG_COLD_NODE;
1188         }
1189 }
1190
1191 static int __get_segment_type(struct page *page, enum page_type p_type)
1192 {
1193         switch (F2FS_P_SB(page)->active_logs) {
1194         case 2:
1195                 return __get_segment_type_2(page, p_type);
1196         case 4:
1197                 return __get_segment_type_4(page, p_type);
1198         }
1199         /* NR_CURSEG_TYPE(6) logs by default */
1200         f2fs_bug_on(F2FS_P_SB(page),
1201                 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1202         return __get_segment_type_6(page, p_type);
1203 }
1204
1205 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1206                 block_t old_blkaddr, block_t *new_blkaddr,
1207                 struct f2fs_summary *sum, int type)
1208 {
1209         struct sit_info *sit_i = SIT_I(sbi);
1210         struct curseg_info *curseg;
1211         bool direct_io = (type == CURSEG_DIRECT_IO);
1212
1213         type = direct_io ? CURSEG_WARM_DATA : type;
1214
1215         curseg = CURSEG_I(sbi, type);
1216
1217         mutex_lock(&curseg->curseg_mutex);
1218         mutex_lock(&sit_i->sentry_lock);
1219
1220         /* direct_io'ed data is aligned to the segment for better performance */
1221         if (direct_io && curseg->next_blkoff)
1222                 __allocate_new_segments(sbi, type);
1223
1224         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1225
1226         /*
1227          * __add_sum_entry should be resided under the curseg_mutex
1228          * because, this function updates a summary entry in the
1229          * current summary block.
1230          */
1231         __add_sum_entry(sbi, type, sum);
1232
1233         __refresh_next_blkoff(sbi, curseg);
1234
1235         stat_inc_block_count(sbi, curseg);
1236
1237         if (!__has_curseg_space(sbi, type))
1238                 sit_i->s_ops->allocate_segment(sbi, type, false);
1239         /*
1240          * SIT information should be updated before segment allocation,
1241          * since SSR needs latest valid block information.
1242          */
1243         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1244
1245         mutex_unlock(&sit_i->sentry_lock);
1246
1247         if (page && IS_NODESEG(type))
1248                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1249
1250         mutex_unlock(&curseg->curseg_mutex);
1251 }
1252
1253 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1254 {
1255         int type = __get_segment_type(fio->page, fio->type);
1256
1257         allocate_data_block(fio->sbi, fio->page, fio->blk_addr,
1258                                         &fio->blk_addr, sum, type);
1259
1260         /* writeout dirty page into bdev */
1261         f2fs_submit_page_mbio(fio);
1262 }
1263
1264 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1265 {
1266         struct f2fs_io_info fio = {
1267                 .sbi = sbi,
1268                 .type = META,
1269                 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1270                 .blk_addr = page->index,
1271                 .page = page,
1272                 .encrypted_page = NULL,
1273         };
1274
1275         set_page_writeback(page);
1276         f2fs_submit_page_mbio(&fio);
1277 }
1278
1279 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1280 {
1281         struct f2fs_summary sum;
1282
1283         set_summary(&sum, nid, 0, 0);
1284         do_write_page(&sum, fio);
1285 }
1286
1287 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1288 {
1289         struct f2fs_sb_info *sbi = fio->sbi;
1290         struct f2fs_summary sum;
1291         struct node_info ni;
1292
1293         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1294         get_node_info(sbi, dn->nid, &ni);
1295         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1296         do_write_page(&sum, fio);
1297         dn->data_blkaddr = fio->blk_addr;
1298 }
1299
1300 void rewrite_data_page(struct f2fs_io_info *fio)
1301 {
1302         stat_inc_inplace_blocks(fio->sbi);
1303         f2fs_submit_page_mbio(fio);
1304 }
1305
1306 static void __f2fs_replace_block(struct f2fs_sb_info *sbi,
1307                                 struct f2fs_summary *sum,
1308                                 block_t old_blkaddr, block_t new_blkaddr,
1309                                 bool recover_curseg)
1310 {
1311         struct sit_info *sit_i = SIT_I(sbi);
1312         struct curseg_info *curseg;
1313         unsigned int segno, old_cursegno;
1314         struct seg_entry *se;
1315         int type;
1316         unsigned short old_blkoff;
1317
1318         segno = GET_SEGNO(sbi, new_blkaddr);
1319         se = get_seg_entry(sbi, segno);
1320         type = se->type;
1321
1322         if (!recover_curseg) {
1323                 /* for recovery flow */
1324                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1325                         if (old_blkaddr == NULL_ADDR)
1326                                 type = CURSEG_COLD_DATA;
1327                         else
1328                                 type = CURSEG_WARM_DATA;
1329                 }
1330         } else {
1331                 if (!IS_CURSEG(sbi, segno))
1332                         type = CURSEG_WARM_DATA;
1333         }
1334
1335         curseg = CURSEG_I(sbi, type);
1336
1337         mutex_lock(&curseg->curseg_mutex);
1338         mutex_lock(&sit_i->sentry_lock);
1339
1340         old_cursegno = curseg->segno;
1341         old_blkoff = curseg->next_blkoff;
1342
1343         /* change the current segment */
1344         if (segno != curseg->segno) {
1345                 curseg->next_segno = segno;
1346                 change_curseg(sbi, type, true);
1347         }
1348
1349         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1350         __add_sum_entry(sbi, type, sum);
1351
1352         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1353         locate_dirty_segment(sbi, old_cursegno);
1354
1355         if (recover_curseg) {
1356                 if (old_cursegno != curseg->segno) {
1357                         curseg->next_segno = old_cursegno;
1358                         change_curseg(sbi, type, true);
1359                 }
1360                 curseg->next_blkoff = old_blkoff;
1361         }
1362
1363         mutex_unlock(&sit_i->sentry_lock);
1364         mutex_unlock(&curseg->curseg_mutex);
1365 }
1366
1367 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1368                                 block_t old_addr, block_t new_addr,
1369                                 unsigned char version, bool recover_curseg)
1370 {
1371         struct f2fs_summary sum;
1372
1373         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1374
1375         __f2fs_replace_block(sbi, &sum, old_addr, new_addr, recover_curseg);
1376
1377         dn->data_blkaddr = new_addr;
1378         set_data_blkaddr(dn);
1379         f2fs_update_extent_cache(dn);
1380 }
1381
1382 static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1383                                         struct page *page, enum page_type type)
1384 {
1385         enum page_type btype = PAGE_TYPE_OF_BIO(type);
1386         struct f2fs_bio_info *io = &sbi->write_io[btype];
1387         struct bio_vec *bvec;
1388         struct page *target;
1389         int i;
1390
1391         down_read(&io->io_rwsem);
1392         if (!io->bio) {
1393                 up_read(&io->io_rwsem);
1394                 return false;
1395         }
1396
1397         bio_for_each_segment_all(bvec, io->bio, i) {
1398
1399                 if (bvec->bv_page->mapping) {
1400                         target = bvec->bv_page;
1401                 } else {
1402                         struct f2fs_crypto_ctx *ctx;
1403
1404                         /* encrypted page */
1405                         ctx = (struct f2fs_crypto_ctx *)page_private(
1406                                                                 bvec->bv_page);
1407                         target = ctx->w.control_page;
1408                 }
1409
1410                 if (page == target) {
1411                         up_read(&io->io_rwsem);
1412                         return true;
1413                 }
1414         }
1415
1416         up_read(&io->io_rwsem);
1417         return false;
1418 }
1419
1420 void f2fs_wait_on_page_writeback(struct page *page,
1421                                 enum page_type type)
1422 {
1423         if (PageWriteback(page)) {
1424                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1425
1426                 if (is_merged_page(sbi, page, type))
1427                         f2fs_submit_merged_bio(sbi, type, WRITE);
1428                 wait_on_page_writeback(page);
1429         }
1430 }
1431
1432 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1433 {
1434         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1435         struct curseg_info *seg_i;
1436         unsigned char *kaddr;
1437         struct page *page;
1438         block_t start;
1439         int i, j, offset;
1440
1441         start = start_sum_block(sbi);
1442
1443         page = get_meta_page(sbi, start++);
1444         kaddr = (unsigned char *)page_address(page);
1445
1446         /* Step 1: restore nat cache */
1447         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1448         memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1449
1450         /* Step 2: restore sit cache */
1451         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1452         memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1453                                                 SUM_JOURNAL_SIZE);
1454         offset = 2 * SUM_JOURNAL_SIZE;
1455
1456         /* Step 3: restore summary entries */
1457         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1458                 unsigned short blk_off;
1459                 unsigned int segno;
1460
1461                 seg_i = CURSEG_I(sbi, i);
1462                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1463                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1464                 seg_i->next_segno = segno;
1465                 reset_curseg(sbi, i, 0);
1466                 seg_i->alloc_type = ckpt->alloc_type[i];
1467                 seg_i->next_blkoff = blk_off;
1468
1469                 if (seg_i->alloc_type == SSR)
1470                         blk_off = sbi->blocks_per_seg;
1471
1472                 for (j = 0; j < blk_off; j++) {
1473                         struct f2fs_summary *s;
1474                         s = (struct f2fs_summary *)(kaddr + offset);
1475                         seg_i->sum_blk->entries[j] = *s;
1476                         offset += SUMMARY_SIZE;
1477                         if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1478                                                 SUM_FOOTER_SIZE)
1479                                 continue;
1480
1481                         f2fs_put_page(page, 1);
1482                         page = NULL;
1483
1484                         page = get_meta_page(sbi, start++);
1485                         kaddr = (unsigned char *)page_address(page);
1486                         offset = 0;
1487                 }
1488         }
1489         f2fs_put_page(page, 1);
1490         return 0;
1491 }
1492
1493 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1494 {
1495         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1496         struct f2fs_summary_block *sum;
1497         struct curseg_info *curseg;
1498         struct page *new;
1499         unsigned short blk_off;
1500         unsigned int segno = 0;
1501         block_t blk_addr = 0;
1502
1503         /* get segment number and block addr */
1504         if (IS_DATASEG(type)) {
1505                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1506                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1507                                                         CURSEG_HOT_DATA]);
1508                 if (__exist_node_summaries(sbi))
1509                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1510                 else
1511                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1512         } else {
1513                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1514                                                         CURSEG_HOT_NODE]);
1515                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1516                                                         CURSEG_HOT_NODE]);
1517                 if (__exist_node_summaries(sbi))
1518                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1519                                                         type - CURSEG_HOT_NODE);
1520                 else
1521                         blk_addr = GET_SUM_BLOCK(sbi, segno);
1522         }
1523
1524         new = get_meta_page(sbi, blk_addr);
1525         sum = (struct f2fs_summary_block *)page_address(new);
1526
1527         if (IS_NODESEG(type)) {
1528                 if (__exist_node_summaries(sbi)) {
1529                         struct f2fs_summary *ns = &sum->entries[0];
1530                         int i;
1531                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1532                                 ns->version = 0;
1533                                 ns->ofs_in_node = 0;
1534                         }
1535                 } else {
1536                         int err;
1537
1538                         err = restore_node_summary(sbi, segno, sum);
1539                         if (err) {
1540                                 f2fs_put_page(new, 1);
1541                                 return err;
1542                         }
1543                 }
1544         }
1545
1546         /* set uncompleted segment to curseg */
1547         curseg = CURSEG_I(sbi, type);
1548         mutex_lock(&curseg->curseg_mutex);
1549         memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1550         curseg->next_segno = segno;
1551         reset_curseg(sbi, type, 0);
1552         curseg->alloc_type = ckpt->alloc_type[type];
1553         curseg->next_blkoff = blk_off;
1554         mutex_unlock(&curseg->curseg_mutex);
1555         f2fs_put_page(new, 1);
1556         return 0;
1557 }
1558
1559 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1560 {
1561         int type = CURSEG_HOT_DATA;
1562         int err;
1563
1564         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1565                 int npages = npages_for_summary_flush(sbi, true);
1566
1567                 if (npages >= 2)
1568                         ra_meta_pages(sbi, start_sum_block(sbi), npages,
1569                                                                 META_CP);
1570
1571                 /* restore for compacted data summary */
1572                 if (read_compacted_summaries(sbi))
1573                         return -EINVAL;
1574                 type = CURSEG_HOT_NODE;
1575         }
1576
1577         if (__exist_node_summaries(sbi))
1578                 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1579                                         NR_CURSEG_TYPE - type, META_CP);
1580
1581         for (; type <= CURSEG_COLD_NODE; type++) {
1582                 err = read_normal_summaries(sbi, type);
1583                 if (err)
1584                         return err;
1585         }
1586
1587         return 0;
1588 }
1589
1590 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1591 {
1592         struct page *page;
1593         unsigned char *kaddr;
1594         struct f2fs_summary *summary;
1595         struct curseg_info *seg_i;
1596         int written_size = 0;
1597         int i, j;
1598
1599         page = grab_meta_page(sbi, blkaddr++);
1600         kaddr = (unsigned char *)page_address(page);
1601
1602         /* Step 1: write nat cache */
1603         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1604         memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1605         written_size += SUM_JOURNAL_SIZE;
1606
1607         /* Step 2: write sit cache */
1608         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1609         memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1610                                                 SUM_JOURNAL_SIZE);
1611         written_size += SUM_JOURNAL_SIZE;
1612
1613         /* Step 3: write summary entries */
1614         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1615                 unsigned short blkoff;
1616                 seg_i = CURSEG_I(sbi, i);
1617                 if (sbi->ckpt->alloc_type[i] == SSR)
1618                         blkoff = sbi->blocks_per_seg;
1619                 else
1620                         blkoff = curseg_blkoff(sbi, i);
1621
1622                 for (j = 0; j < blkoff; j++) {
1623                         if (!page) {
1624                                 page = grab_meta_page(sbi, blkaddr++);
1625                                 kaddr = (unsigned char *)page_address(page);
1626                                 written_size = 0;
1627                         }
1628                         summary = (struct f2fs_summary *)(kaddr + written_size);
1629                         *summary = seg_i->sum_blk->entries[j];
1630                         written_size += SUMMARY_SIZE;
1631
1632                         if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1633                                                         SUM_FOOTER_SIZE)
1634                                 continue;
1635
1636                         set_page_dirty(page);
1637                         f2fs_put_page(page, 1);
1638                         page = NULL;
1639                 }
1640         }
1641         if (page) {
1642                 set_page_dirty(page);
1643                 f2fs_put_page(page, 1);
1644         }
1645 }
1646
1647 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1648                                         block_t blkaddr, int type)
1649 {
1650         int i, end;
1651         if (IS_DATASEG(type))
1652                 end = type + NR_CURSEG_DATA_TYPE;
1653         else
1654                 end = type + NR_CURSEG_NODE_TYPE;
1655
1656         for (i = type; i < end; i++) {
1657                 struct curseg_info *sum = CURSEG_I(sbi, i);
1658                 mutex_lock(&sum->curseg_mutex);
1659                 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1660                 mutex_unlock(&sum->curseg_mutex);
1661         }
1662 }
1663
1664 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1665 {
1666         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1667                 write_compacted_summaries(sbi, start_blk);
1668         else
1669                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1670 }
1671
1672 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1673 {
1674         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1675 }
1676
1677 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1678                                         unsigned int val, int alloc)
1679 {
1680         int i;
1681
1682         if (type == NAT_JOURNAL) {
1683                 for (i = 0; i < nats_in_cursum(sum); i++) {
1684                         if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1685                                 return i;
1686                 }
1687                 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1688                         return update_nats_in_cursum(sum, 1);
1689         } else if (type == SIT_JOURNAL) {
1690                 for (i = 0; i < sits_in_cursum(sum); i++)
1691                         if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1692                                 return i;
1693                 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1694                         return update_sits_in_cursum(sum, 1);
1695         }
1696         return -1;
1697 }
1698
1699 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1700                                         unsigned int segno)
1701 {
1702         return get_meta_page(sbi, current_sit_addr(sbi, segno));
1703 }
1704
1705 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1706                                         unsigned int start)
1707 {
1708         struct sit_info *sit_i = SIT_I(sbi);
1709         struct page *src_page, *dst_page;
1710         pgoff_t src_off, dst_off;
1711         void *src_addr, *dst_addr;
1712
1713         src_off = current_sit_addr(sbi, start);
1714         dst_off = next_sit_addr(sbi, src_off);
1715
1716         /* get current sit block page without lock */
1717         src_page = get_meta_page(sbi, src_off);
1718         dst_page = grab_meta_page(sbi, dst_off);
1719         f2fs_bug_on(sbi, PageDirty(src_page));
1720
1721         src_addr = page_address(src_page);
1722         dst_addr = page_address(dst_page);
1723         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1724
1725         set_page_dirty(dst_page);
1726         f2fs_put_page(src_page, 1);
1727
1728         set_to_next_sit(sit_i, start);
1729
1730         return dst_page;
1731 }
1732
1733 static struct sit_entry_set *grab_sit_entry_set(void)
1734 {
1735         struct sit_entry_set *ses =
1736                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_ATOMIC);
1737
1738         ses->entry_cnt = 0;
1739         INIT_LIST_HEAD(&ses->set_list);
1740         return ses;
1741 }
1742
1743 static void release_sit_entry_set(struct sit_entry_set *ses)
1744 {
1745         list_del(&ses->set_list);
1746         kmem_cache_free(sit_entry_set_slab, ses);
1747 }
1748
1749 static void adjust_sit_entry_set(struct sit_entry_set *ses,
1750                                                 struct list_head *head)
1751 {
1752         struct sit_entry_set *next = ses;
1753
1754         if (list_is_last(&ses->set_list, head))
1755                 return;
1756
1757         list_for_each_entry_continue(next, head, set_list)
1758                 if (ses->entry_cnt <= next->entry_cnt)
1759                         break;
1760
1761         list_move_tail(&ses->set_list, &next->set_list);
1762 }
1763
1764 static void add_sit_entry(unsigned int segno, struct list_head *head)
1765 {
1766         struct sit_entry_set *ses;
1767         unsigned int start_segno = START_SEGNO(segno);
1768
1769         list_for_each_entry(ses, head, set_list) {
1770                 if (ses->start_segno == start_segno) {
1771                         ses->entry_cnt++;
1772                         adjust_sit_entry_set(ses, head);
1773                         return;
1774                 }
1775         }
1776
1777         ses = grab_sit_entry_set();
1778
1779         ses->start_segno = start_segno;
1780         ses->entry_cnt++;
1781         list_add(&ses->set_list, head);
1782 }
1783
1784 static void add_sits_in_set(struct f2fs_sb_info *sbi)
1785 {
1786         struct f2fs_sm_info *sm_info = SM_I(sbi);
1787         struct list_head *set_list = &sm_info->sit_entry_set;
1788         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
1789         unsigned int segno;
1790
1791         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
1792                 add_sit_entry(segno, set_list);
1793 }
1794
1795 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
1796 {
1797         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1798         struct f2fs_summary_block *sum = curseg->sum_blk;
1799         int i;
1800
1801         for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1802                 unsigned int segno;
1803                 bool dirtied;
1804
1805                 segno = le32_to_cpu(segno_in_journal(sum, i));
1806                 dirtied = __mark_sit_entry_dirty(sbi, segno);
1807
1808                 if (!dirtied)
1809                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
1810         }
1811         update_sits_in_cursum(sum, -sits_in_cursum(sum));
1812 }
1813
1814 /*
1815  * CP calls this function, which flushes SIT entries including sit_journal,
1816  * and moves prefree segs to free segs.
1817  */
1818 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1819 {
1820         struct sit_info *sit_i = SIT_I(sbi);
1821         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1822         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1823         struct f2fs_summary_block *sum = curseg->sum_blk;
1824         struct sit_entry_set *ses, *tmp;
1825         struct list_head *head = &SM_I(sbi)->sit_entry_set;
1826         bool to_journal = true;
1827         struct seg_entry *se;
1828
1829         mutex_lock(&curseg->curseg_mutex);
1830         mutex_lock(&sit_i->sentry_lock);
1831
1832         if (!sit_i->dirty_sentries)
1833                 goto out;
1834
1835         /*
1836          * add and account sit entries of dirty bitmap in sit entry
1837          * set temporarily
1838          */
1839         add_sits_in_set(sbi);
1840
1841         /*
1842          * if there are no enough space in journal to store dirty sit
1843          * entries, remove all entries from journal and add and account
1844          * them in sit entry set.
1845          */
1846         if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL))
1847                 remove_sits_in_journal(sbi);
1848
1849         /*
1850          * there are two steps to flush sit entries:
1851          * #1, flush sit entries to journal in current cold data summary block.
1852          * #2, flush sit entries to sit page.
1853          */
1854         list_for_each_entry_safe(ses, tmp, head, set_list) {
1855                 struct page *page = NULL;
1856                 struct f2fs_sit_block *raw_sit = NULL;
1857                 unsigned int start_segno = ses->start_segno;
1858                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
1859                                                 (unsigned long)MAIN_SEGS(sbi));
1860                 unsigned int segno = start_segno;
1861
1862                 if (to_journal &&
1863                         !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL))
1864                         to_journal = false;
1865
1866                 if (!to_journal) {
1867                         page = get_next_sit_page(sbi, start_segno);
1868                         raw_sit = page_address(page);
1869                 }
1870
1871                 /* flush dirty sit entries in region of current sit set */
1872                 for_each_set_bit_from(segno, bitmap, end) {
1873                         int offset, sit_offset;
1874
1875                         se = get_seg_entry(sbi, segno);
1876
1877                         /* add discard candidates */
1878                         if (cpc->reason != CP_DISCARD) {
1879                                 cpc->trim_start = segno;
1880                                 add_discard_addrs(sbi, cpc);
1881                         }
1882
1883                         if (to_journal) {
1884                                 offset = lookup_journal_in_cursum(sum,
1885                                                         SIT_JOURNAL, segno, 1);
1886                                 f2fs_bug_on(sbi, offset < 0);
1887                                 segno_in_journal(sum, offset) =
1888                                                         cpu_to_le32(segno);
1889                                 seg_info_to_raw_sit(se,
1890                                                 &sit_in_journal(sum, offset));
1891                         } else {
1892                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1893                                 seg_info_to_raw_sit(se,
1894                                                 &raw_sit->entries[sit_offset]);
1895                         }
1896
1897                         __clear_bit(segno, bitmap);
1898                         sit_i->dirty_sentries--;
1899                         ses->entry_cnt--;
1900                 }
1901
1902                 if (!to_journal)
1903                         f2fs_put_page(page, 1);
1904
1905                 f2fs_bug_on(sbi, ses->entry_cnt);
1906                 release_sit_entry_set(ses);
1907         }
1908
1909         f2fs_bug_on(sbi, !list_empty(head));
1910         f2fs_bug_on(sbi, sit_i->dirty_sentries);
1911 out:
1912         if (cpc->reason == CP_DISCARD) {
1913                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
1914                         add_discard_addrs(sbi, cpc);
1915         }
1916         mutex_unlock(&sit_i->sentry_lock);
1917         mutex_unlock(&curseg->curseg_mutex);
1918
1919         set_prefree_as_free_segments(sbi);
1920 }
1921
1922 static int build_sit_info(struct f2fs_sb_info *sbi)
1923 {
1924         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1925         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1926         struct sit_info *sit_i;
1927         unsigned int sit_segs, start;
1928         char *src_bitmap, *dst_bitmap;
1929         unsigned int bitmap_size;
1930
1931         /* allocate memory for SIT information */
1932         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1933         if (!sit_i)
1934                 return -ENOMEM;
1935
1936         SM_I(sbi)->sit_info = sit_i;
1937
1938         sit_i->sentries = vzalloc(MAIN_SEGS(sbi) * sizeof(struct seg_entry));
1939         if (!sit_i->sentries)
1940                 return -ENOMEM;
1941
1942         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
1943         sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1944         if (!sit_i->dirty_sentries_bitmap)
1945                 return -ENOMEM;
1946
1947         for (start = 0; start < MAIN_SEGS(sbi); start++) {
1948                 sit_i->sentries[start].cur_valid_map
1949                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1950                 sit_i->sentries[start].ckpt_valid_map
1951                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1952                 sit_i->sentries[start].discard_map
1953                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1954                 if (!sit_i->sentries[start].cur_valid_map ||
1955                                 !sit_i->sentries[start].ckpt_valid_map ||
1956                                 !sit_i->sentries[start].discard_map)
1957                         return -ENOMEM;
1958         }
1959
1960         sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1961         if (!sit_i->tmp_map)
1962                 return -ENOMEM;
1963
1964         if (sbi->segs_per_sec > 1) {
1965                 sit_i->sec_entries = vzalloc(MAIN_SECS(sbi) *
1966                                         sizeof(struct sec_entry));
1967                 if (!sit_i->sec_entries)
1968                         return -ENOMEM;
1969         }
1970
1971         /* get information related with SIT */
1972         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1973
1974         /* setup SIT bitmap from ckeckpoint pack */
1975         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1976         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1977
1978         dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1979         if (!dst_bitmap)
1980                 return -ENOMEM;
1981
1982         /* init SIT information */
1983         sit_i->s_ops = &default_salloc_ops;
1984
1985         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1986         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1987         sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1988         sit_i->sit_bitmap = dst_bitmap;
1989         sit_i->bitmap_size = bitmap_size;
1990         sit_i->dirty_sentries = 0;
1991         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1992         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1993         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1994         mutex_init(&sit_i->sentry_lock);
1995         return 0;
1996 }
1997
1998 static int build_free_segmap(struct f2fs_sb_info *sbi)
1999 {
2000         struct free_segmap_info *free_i;
2001         unsigned int bitmap_size, sec_bitmap_size;
2002
2003         /* allocate memory for free segmap information */
2004         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2005         if (!free_i)
2006                 return -ENOMEM;
2007
2008         SM_I(sbi)->free_info = free_i;
2009
2010         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2011         free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
2012         if (!free_i->free_segmap)
2013                 return -ENOMEM;
2014
2015         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2016         free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
2017         if (!free_i->free_secmap)
2018                 return -ENOMEM;
2019
2020         /* set all segments as dirty temporarily */
2021         memset(free_i->free_segmap, 0xff, bitmap_size);
2022         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2023
2024         /* init free segmap information */
2025         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2026         free_i->free_segments = 0;
2027         free_i->free_sections = 0;
2028         spin_lock_init(&free_i->segmap_lock);
2029         return 0;
2030 }
2031
2032 static int build_curseg(struct f2fs_sb_info *sbi)
2033 {
2034         struct curseg_info *array;
2035         int i;
2036
2037         array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2038         if (!array)
2039                 return -ENOMEM;
2040
2041         SM_I(sbi)->curseg_array = array;
2042
2043         for (i = 0; i < NR_CURSEG_TYPE; i++) {
2044                 mutex_init(&array[i].curseg_mutex);
2045                 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
2046                 if (!array[i].sum_blk)
2047                         return -ENOMEM;
2048                 array[i].segno = NULL_SEGNO;
2049                 array[i].next_blkoff = 0;
2050         }
2051         return restore_curseg_summaries(sbi);
2052 }
2053
2054 static void build_sit_entries(struct f2fs_sb_info *sbi)
2055 {
2056         struct sit_info *sit_i = SIT_I(sbi);
2057         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2058         struct f2fs_summary_block *sum = curseg->sum_blk;
2059         int sit_blk_cnt = SIT_BLK_CNT(sbi);
2060         unsigned int i, start, end;
2061         unsigned int readed, start_blk = 0;
2062         int nrpages = MAX_BIO_BLOCKS(sbi);
2063
2064         do {
2065                 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT);
2066
2067                 start = start_blk * sit_i->sents_per_block;
2068                 end = (start_blk + readed) * sit_i->sents_per_block;
2069
2070                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2071                         struct seg_entry *se = &sit_i->sentries[start];
2072                         struct f2fs_sit_block *sit_blk;
2073                         struct f2fs_sit_entry sit;
2074                         struct page *page;
2075
2076                         mutex_lock(&curseg->curseg_mutex);
2077                         for (i = 0; i < sits_in_cursum(sum); i++) {
2078                                 if (le32_to_cpu(segno_in_journal(sum, i))
2079                                                                 == start) {
2080                                         sit = sit_in_journal(sum, i);
2081                                         mutex_unlock(&curseg->curseg_mutex);
2082                                         goto got_it;
2083                                 }
2084                         }
2085                         mutex_unlock(&curseg->curseg_mutex);
2086
2087                         page = get_current_sit_page(sbi, start);
2088                         sit_blk = (struct f2fs_sit_block *)page_address(page);
2089                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2090                         f2fs_put_page(page, 1);
2091 got_it:
2092                         check_block_count(sbi, start, &sit);
2093                         seg_info_from_raw_sit(se, &sit);
2094
2095                         /* build discard map only one time */
2096                         memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2097                         sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2098
2099                         if (sbi->segs_per_sec > 1) {
2100                                 struct sec_entry *e = get_sec_entry(sbi, start);
2101                                 e->valid_blocks += se->valid_blocks;
2102                         }
2103                 }
2104                 start_blk += readed;
2105         } while (start_blk < sit_blk_cnt);
2106 }
2107
2108 static void init_free_segmap(struct f2fs_sb_info *sbi)
2109 {
2110         unsigned int start;
2111         int type;
2112
2113         for (start = 0; start < MAIN_SEGS(sbi); start++) {
2114                 struct seg_entry *sentry = get_seg_entry(sbi, start);
2115                 if (!sentry->valid_blocks)
2116                         __set_free(sbi, start);
2117         }
2118
2119         /* set use the current segments */
2120         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2121                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2122                 __set_test_and_inuse(sbi, curseg_t->segno);
2123         }
2124 }
2125
2126 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2127 {
2128         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2129         struct free_segmap_info *free_i = FREE_I(sbi);
2130         unsigned int segno = 0, offset = 0;
2131         unsigned short valid_blocks;
2132
2133         while (1) {
2134                 /* find dirty segment based on free segmap */
2135                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2136                 if (segno >= MAIN_SEGS(sbi))
2137                         break;
2138                 offset = segno + 1;
2139                 valid_blocks = get_valid_blocks(sbi, segno, 0);
2140                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2141                         continue;
2142                 if (valid_blocks > sbi->blocks_per_seg) {
2143                         f2fs_bug_on(sbi, 1);
2144                         continue;
2145                 }
2146                 mutex_lock(&dirty_i->seglist_lock);
2147                 __locate_dirty_segment(sbi, segno, DIRTY);
2148                 mutex_unlock(&dirty_i->seglist_lock);
2149         }
2150 }
2151
2152 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2153 {
2154         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2155         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2156
2157         dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
2158         if (!dirty_i->victim_secmap)
2159                 return -ENOMEM;
2160         return 0;
2161 }
2162
2163 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2164 {
2165         struct dirty_seglist_info *dirty_i;
2166         unsigned int bitmap_size, i;
2167
2168         /* allocate memory for dirty segments list information */
2169         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2170         if (!dirty_i)
2171                 return -ENOMEM;
2172
2173         SM_I(sbi)->dirty_info = dirty_i;
2174         mutex_init(&dirty_i->seglist_lock);
2175
2176         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2177
2178         for (i = 0; i < NR_DIRTY_TYPE; i++) {
2179                 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
2180                 if (!dirty_i->dirty_segmap[i])
2181                         return -ENOMEM;
2182         }
2183
2184         init_dirty_segmap(sbi);
2185         return init_victim_secmap(sbi);
2186 }
2187
2188 /*
2189  * Update min, max modified time for cost-benefit GC algorithm
2190  */
2191 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2192 {
2193         struct sit_info *sit_i = SIT_I(sbi);
2194         unsigned int segno;
2195
2196         mutex_lock(&sit_i->sentry_lock);
2197
2198         sit_i->min_mtime = LLONG_MAX;
2199
2200         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2201                 unsigned int i;
2202                 unsigned long long mtime = 0;
2203
2204                 for (i = 0; i < sbi->segs_per_sec; i++)
2205                         mtime += get_seg_entry(sbi, segno + i)->mtime;
2206
2207                 mtime = div_u64(mtime, sbi->segs_per_sec);
2208
2209                 if (sit_i->min_mtime > mtime)
2210                         sit_i->min_mtime = mtime;
2211         }
2212         sit_i->max_mtime = get_mtime(sbi);
2213         mutex_unlock(&sit_i->sentry_lock);
2214 }
2215
2216 int build_segment_manager(struct f2fs_sb_info *sbi)
2217 {
2218         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2219         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2220         struct f2fs_sm_info *sm_info;
2221         int err;
2222
2223         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2224         if (!sm_info)
2225                 return -ENOMEM;
2226
2227         /* init sm info */
2228         sbi->sm_info = sm_info;
2229         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2230         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2231         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2232         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2233         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2234         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2235         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2236         sm_info->rec_prefree_segments = sm_info->main_segments *
2237                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2238         sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2239         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2240         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2241
2242         INIT_LIST_HEAD(&sm_info->discard_list);
2243         sm_info->nr_discards = 0;
2244         sm_info->max_discards = 0;
2245
2246         sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2247
2248         INIT_LIST_HEAD(&sm_info->sit_entry_set);
2249
2250         if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2251                 err = create_flush_cmd_control(sbi);
2252                 if (err)
2253                         return err;
2254         }
2255
2256         err = build_sit_info(sbi);
2257         if (err)
2258                 return err;
2259         err = build_free_segmap(sbi);
2260         if (err)
2261                 return err;
2262         err = build_curseg(sbi);
2263         if (err)
2264                 return err;
2265
2266         /* reinit free segmap based on SIT */
2267         build_sit_entries(sbi);
2268
2269         init_free_segmap(sbi);
2270         err = build_dirty_segmap(sbi);
2271         if (err)
2272                 return err;
2273
2274         init_min_max_mtime(sbi);
2275         return 0;
2276 }
2277
2278 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2279                 enum dirty_type dirty_type)
2280 {
2281         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2282
2283         mutex_lock(&dirty_i->seglist_lock);
2284         kfree(dirty_i->dirty_segmap[dirty_type]);
2285         dirty_i->nr_dirty[dirty_type] = 0;
2286         mutex_unlock(&dirty_i->seglist_lock);
2287 }
2288
2289 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2290 {
2291         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2292         kfree(dirty_i->victim_secmap);
2293 }
2294
2295 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2296 {
2297         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2298         int i;
2299
2300         if (!dirty_i)
2301                 return;
2302
2303         /* discard pre-free/dirty segments list */
2304         for (i = 0; i < NR_DIRTY_TYPE; i++)
2305                 discard_dirty_segmap(sbi, i);
2306
2307         destroy_victim_secmap(sbi);
2308         SM_I(sbi)->dirty_info = NULL;
2309         kfree(dirty_i);
2310 }
2311
2312 static void destroy_curseg(struct f2fs_sb_info *sbi)
2313 {
2314         struct curseg_info *array = SM_I(sbi)->curseg_array;
2315         int i;
2316
2317         if (!array)
2318                 return;
2319         SM_I(sbi)->curseg_array = NULL;
2320         for (i = 0; i < NR_CURSEG_TYPE; i++)
2321                 kfree(array[i].sum_blk);
2322         kfree(array);
2323 }
2324
2325 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2326 {
2327         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2328         if (!free_i)
2329                 return;
2330         SM_I(sbi)->free_info = NULL;
2331         kfree(free_i->free_segmap);
2332         kfree(free_i->free_secmap);
2333         kfree(free_i);
2334 }
2335
2336 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2337 {
2338         struct sit_info *sit_i = SIT_I(sbi);
2339         unsigned int start;
2340
2341         if (!sit_i)
2342                 return;
2343
2344         if (sit_i->sentries) {
2345                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2346                         kfree(sit_i->sentries[start].cur_valid_map);
2347                         kfree(sit_i->sentries[start].ckpt_valid_map);
2348                         kfree(sit_i->sentries[start].discard_map);
2349                 }
2350         }
2351         kfree(sit_i->tmp_map);
2352
2353         vfree(sit_i->sentries);
2354         vfree(sit_i->sec_entries);
2355         kfree(sit_i->dirty_sentries_bitmap);
2356
2357         SM_I(sbi)->sit_info = NULL;
2358         kfree(sit_i->sit_bitmap);
2359         kfree(sit_i);
2360 }
2361
2362 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2363 {
2364         struct f2fs_sm_info *sm_info = SM_I(sbi);
2365
2366         if (!sm_info)
2367                 return;
2368         destroy_flush_cmd_control(sbi);
2369         destroy_dirty_segmap(sbi);
2370         destroy_curseg(sbi);
2371         destroy_free_segmap(sbi);
2372         destroy_sit_info(sbi);
2373         sbi->sm_info = NULL;
2374         kfree(sm_info);
2375 }
2376
2377 int __init create_segment_manager_caches(void)
2378 {
2379         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2380                         sizeof(struct discard_entry));
2381         if (!discard_entry_slab)
2382                 goto fail;
2383
2384         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2385                         sizeof(struct sit_entry_set));
2386         if (!sit_entry_set_slab)
2387                 goto destory_discard_entry;
2388
2389         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2390                         sizeof(struct inmem_pages));
2391         if (!inmem_entry_slab)
2392                 goto destroy_sit_entry_set;
2393         return 0;
2394
2395 destroy_sit_entry_set:
2396         kmem_cache_destroy(sit_entry_set_slab);
2397 destory_discard_entry:
2398         kmem_cache_destroy(discard_entry_slab);
2399 fail:
2400         return -ENOMEM;
2401 }
2402
2403 void destroy_segment_manager_caches(void)
2404 {
2405         kmem_cache_destroy(sit_entry_set_slab);
2406         kmem_cache_destroy(discard_entry_slab);
2407         kmem_cache_destroy(inmem_entry_slab);
2408 }