fs/f2fs/segment.c

   1 /*
   2  * fs/f2fs/segment.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/bio.h>
  14 #include <linux/blkdev.h>
  15 #include <linux/prefetch.h>
  16 #include <linux/kthread.h>
  17 #include <linux/vmalloc.h>
  18 #include <linux/swap.h>
  19
  20 #include "f2fs.h"
  21 #include "segment.h"
  22 #include "node.h"
  23 #include <trace/events/f2fs.h>
  24
  25 #define __reverse_ffz(x) __reverse_ffs(~(x))
  26
  27 static struct kmem_cache *discard_entry_slab;
  28 static struct kmem_cache *sit_entry_set_slab;
  29 static struct kmem_cache *inmem_entry_slab;
  30
  31 /*
  32  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  33  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  34  */
  35 static inline unsigned long __reverse_ffs(unsigned long word)
  36 {
  37         int num = 0;
  38
  39 #if BITS_PER_LONG == 64
  40         if ((word & 0xffffffff) == 0) {
  41                 num += 32;
  42                 word >>= 32;
  43         }
  44 #endif
  45         if ((word & 0xffff) == 0) {
  46                 num += 16;
  47                 word >>= 16;
  48         }
  49         if ((word & 0xff) == 0) {
  50                 num += 8;
  51                 word >>= 8;
  52         }
  53         if ((word & 0xf0) == 0)
  54                 num += 4;
  55         else
  56                 word >>= 4;
  57         if ((word & 0xc) == 0)
  58                 num += 2;
  59         else
  60                 word >>= 2;
  61         if ((word & 0x2) == 0)
  62                 num += 1;
  63         return num;
  64 }
  65
  66 /*
  67  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  68  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  69  * Example:
  70  *                             LSB <--> MSB
  71  *   f2fs_set_bit(0, bitmap) => 0000 0001
  72  *   f2fs_set_bit(7, bitmap) => 1000 0000
  73  */
  74 static unsigned long __find_rev_next_bit(const unsigned long *addr,
  75                         unsigned long size, unsigned long offset)
  76 {
  77         const unsigned long *p = addr + BIT_WORD(offset);
  78         unsigned long result = offset & ~(BITS_PER_LONG - 1);
  79         unsigned long tmp;
  80         unsigned long mask, submask;
  81         unsigned long quot, rest;
  82
  83         if (offset >= size)
  84                 return size;
  85
  86         size -= result;
  87         offset %= BITS_PER_LONG;
  88         if (!offset)
  89                 goto aligned;
  90
  91         tmp = *(p++);
  92         quot = (offset >> 3) << 3;
  93         rest = offset & 0x7;
  94         mask = ~0UL << quot;
  95         submask = (unsigned char)(0xff << rest) >> rest;
  96         submask <<= quot;
  97         mask &= submask;
  98         tmp &= mask;
  99         if (size < BITS_PER_LONG)
 100                 goto found_first;
 101         if (tmp)
 102                 goto found_middle;
 103
 104         size -= BITS_PER_LONG;
 105         result += BITS_PER_LONG;
 106 aligned:
 107         while (size & ~(BITS_PER_LONG-1)) {
 108                 tmp = *(p++);
 109                 if (tmp)
 110                         goto found_middle;
 111                 result += BITS_PER_LONG;
 112                 size -= BITS_PER_LONG;
 113         }
 114         if (!size)
 115                 return result;
 116         tmp = *p;
 117 found_first:
 118         tmp &= (~0UL >> (BITS_PER_LONG - size));
 119         if (tmp == 0UL)         /* Are any bits set? */
 120                 return result + size;   /* Nope. */
 121 found_middle:
 122         return result + __reverse_ffs(tmp);
 123 }
 124
 125 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 126                         unsigned long size, unsigned long offset)
 127 {
 128         const unsigned long *p = addr + BIT_WORD(offset);
 129         unsigned long result = offset & ~(BITS_PER_LONG - 1);
 130         unsigned long tmp;
 131         unsigned long mask, submask;
 132         unsigned long quot, rest;
 133
 134         if (offset >= size)
 135                 return size;
 136
 137         size -= result;
 138         offset %= BITS_PER_LONG;
 139         if (!offset)
 140                 goto aligned;
 141
 142         tmp = *(p++);
 143         quot = (offset >> 3) << 3;
 144         rest = offset & 0x7;
 145         mask = ~(~0UL << quot);
 146         submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest);
 147         submask <<= quot;
 148         mask += submask;
 149         tmp |= mask;
 150         if (size < BITS_PER_LONG)
 151                 goto found_first;
 152         if (~tmp)
 153                 goto found_middle;
 154
 155         size -= BITS_PER_LONG;
 156         result += BITS_PER_LONG;
 157 aligned:
 158         while (size & ~(BITS_PER_LONG - 1)) {
 159                 tmp = *(p++);
 160                 if (~tmp)
 161                         goto found_middle;
 162                 result += BITS_PER_LONG;
 163                 size -= BITS_PER_LONG;
 164         }
 165         if (!size)
 166                 return result;
 167         tmp = *p;
 168
 169 found_first:
 170         tmp |= ~0UL << size;
 171         if (tmp == ~0UL)        /* Are any bits zero? */
 172                 return result + size;   /* Nope. */
 173 found_middle:
 174         return result + __reverse_ffz(tmp);
 175 }
 176
 177 void register_inmem_page(struct inode *inode, struct page *page)
 178 {
 179         struct f2fs_inode_info *fi = F2FS_I(inode);
 180         struct inmem_pages *new;
 181
 182         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 183
 184         /* add atomic page indices to the list */
 185         new->page = page;
 186         INIT_LIST_HEAD(&new->list);
 187
 188         /* increase reference count with clean state */
 189         mutex_lock(&fi->inmem_lock);
 190         get_page(page);
 191         list_add_tail(&new->list, &fi->inmem_pages);
 192         mutex_unlock(&fi->inmem_lock);
 193 }
 194
 195 void commit_inmem_pages(struct inode *inode, bool abort)
 196 {
 197         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 198         struct f2fs_inode_info *fi = F2FS_I(inode);
 199         struct inmem_pages *cur, *tmp;
 200         bool submit_bio = false;
 201         struct f2fs_io_info fio = {
 202                 .type = DATA,
 203                 .rw = WRITE_SYNC,
 204         };
 205
 206         f2fs_balance_fs(sbi);
 207         f2fs_lock_op(sbi);
 208
 209         mutex_lock(&fi->inmem_lock);
 210         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 211                 lock_page(cur->page);
 212                 if (!abort && cur->page->mapping == inode->i_mapping) {
 213                         f2fs_wait_on_page_writeback(cur->page, DATA);
 214                         if (clear_page_dirty_for_io(cur->page))
 215                                 inode_dec_dirty_pages(inode);
 216                         do_write_data_page(cur->page, &fio);
 217                         submit_bio = true;
 218                 }
 219                 f2fs_put_page(cur->page, 1);
 220                 list_del(&cur->list);
 221                 kmem_cache_free(inmem_entry_slab, cur);
 222         }
 223         if (submit_bio)
 224                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 225         mutex_unlock(&fi->inmem_lock);
 226
 227         filemap_fdatawait_range(inode->i_mapping, 0, LLONG_MAX);
 228         f2fs_unlock_op(sbi);
 229 }
 230
 231 /*
 232  * This function balances dirty node and dentry pages.
 233  * In addition, it controls garbage collection.
 234  */
 235 void f2fs_balance_fs(struct f2fs_sb_info *sbi)
 236 {
 237         /*
 238          * We should do GC or end up with checkpoint, if there are so many dirty
 239          * dir/node pages without enough free segments.
 240          */
 241         if (has_not_enough_free_secs(sbi, 0)) {
 242                 mutex_lock(&sbi->gc_mutex);
 243                 f2fs_gc(sbi);
 244         }
 245 }
 246
 247 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 248 {
 249         /* check the # of cached NAT entries and prefree segments */
 250         if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) ||
 251                                 excess_prefree_segs(sbi))
 252                 f2fs_sync_fs(sbi->sb, true);
 253 }
 254
 255 static int issue_flush_thread(void *data)
 256 {
 257         struct f2fs_sb_info *sbi = data;
 258         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 259         wait_queue_head_t *q = &fcc->flush_wait_queue;
 260 repeat:
 261         if (kthread_should_stop())
 262                 return 0;
 263
 264         if (!llist_empty(&fcc->issue_list)) {
 265                 struct bio *bio = bio_alloc(GFP_NOIO, 0);
 266                 struct flush_cmd *cmd, *next;
 267                 int ret;
 268
 269                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 270                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 271
 272                 bio->bi_bdev = sbi->sb->s_bdev;
 273                 ret = submit_bio_wait(WRITE_FLUSH, bio);
 274
 275                 llist_for_each_entry_safe(cmd, next,
 276                                           fcc->dispatch_list, llnode) {
 277                         cmd->ret = ret;
 278                         complete(&cmd->wait);
 279                 }
 280                 bio_put(bio);
 281                 fcc->dispatch_list = NULL;
 282         }
 283
 284         wait_event_interruptible(*q,
 285                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 286         goto repeat;
 287 }
 288
 289 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
 290 {
 291         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 292         struct flush_cmd cmd;
 293
 294         trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
 295                                         test_opt(sbi, FLUSH_MERGE));
 296
 297         if (test_opt(sbi, NOBARRIER))
 298                 return 0;
 299
 300         if (!test_opt(sbi, FLUSH_MERGE))
 301                 return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL);
 302
 303         init_completion(&cmd.wait);
 304
 305         llist_add(&cmd.llnode, &fcc->issue_list);
 306
 307         if (!fcc->dispatch_list)
 308                 wake_up(&fcc->flush_wait_queue);
 309
 310         wait_for_completion(&cmd.wait);
 311
 312         return cmd.ret;
 313 }
 314
 315 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
 316 {
 317         dev_t dev = sbi->sb->s_bdev->bd_dev;
 318         struct flush_cmd_control *fcc;
 319         int err = 0;
 320
 321         fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
 322         if (!fcc)
 323                 return -ENOMEM;
 324         init_waitqueue_head(&fcc->flush_wait_queue);
 325         init_llist_head(&fcc->issue_list);
 326         SM_I(sbi)->cmd_control_info = fcc;
 327         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 328                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 329         if (IS_ERR(fcc->f2fs_issue_flush)) {
 330                 err = PTR_ERR(fcc->f2fs_issue_flush);
 331                 kfree(fcc);
 332                 SM_I(sbi)->cmd_control_info = NULL;
 333                 return err;
 334         }
 335
 336         return err;
 337 }
 338
 339 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
 340 {
 341         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 342
 343         if (fcc && fcc->f2fs_issue_flush)
 344                 kthread_stop(fcc->f2fs_issue_flush);
 345         kfree(fcc);
 346         SM_I(sbi)->cmd_control_info = NULL;
 347 }
 348
 349 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 350                 enum dirty_type dirty_type)
 351 {
 352         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 353
 354         /* need not be added */
 355         if (IS_CURSEG(sbi, segno))
 356                 return;
 357
 358         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 359                 dirty_i->nr_dirty[dirty_type]++;
 360
 361         if (dirty_type == DIRTY) {
 362                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 363                 enum dirty_type t = sentry->type;
 364
 365                 if (unlikely(t >= DIRTY)) {
 366                         f2fs_bug_on(sbi, 1);
 367                         return;
 368                 }
 369                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 370                         dirty_i->nr_dirty[t]++;
 371         }
 372 }
 373
 374 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 375                 enum dirty_type dirty_type)
 376 {
 377         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 378
 379         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 380                 dirty_i->nr_dirty[dirty_type]--;
 381
 382         if (dirty_type == DIRTY) {
 383                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 384                 enum dirty_type t = sentry->type;
 385
 386                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 387                         dirty_i->nr_dirty[t]--;
 388
 389                 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
 390                         clear_bit(GET_SECNO(sbi, segno),
 391                                                 dirty_i->victim_secmap);
 392         }
 393 }
 394
 395 /*
 396  * Should not occur error such as -ENOMEM.
 397  * Adding dirty entry into seglist is not critical operation.
 398  * If a given segment is one of current working segments, it won't be added.
 399  */
 400 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 401 {
 402         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 403         unsigned short valid_blocks;
 404
 405         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 406                 return;
 407
 408         mutex_lock(&dirty_i->seglist_lock);
 409
 410         valid_blocks = get_valid_blocks(sbi, segno, 0);
 411
 412         if (valid_blocks == 0) {
 413                 __locate_dirty_segment(sbi, segno, PRE);
 414                 __remove_dirty_segment(sbi, segno, DIRTY);
 415         } else if (valid_blocks < sbi->blocks_per_seg) {
 416                 __locate_dirty_segment(sbi, segno, DIRTY);
 417         } else {
 418                 /* Recovery routine with SSR needs this */
 419                 __remove_dirty_segment(sbi, segno, DIRTY);
 420         }
 421
 422         mutex_unlock(&dirty_i->seglist_lock);
 423 }
 424
 425 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
 426                                 block_t blkstart, block_t blklen)
 427 {
 428         sector_t start = SECTOR_FROM_BLOCK(blkstart);
 429         sector_t len = SECTOR_FROM_BLOCK(blklen);
 430         trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
 431         return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
 432 }
 433
 434 void discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
 435 {
 436         if (f2fs_issue_discard(sbi, blkaddr, 1)) {
 437                 struct page *page = grab_meta_page(sbi, blkaddr);
 438                 /* zero-filled page */
 439                 set_page_dirty(page);
 440                 f2fs_put_page(page, 1);
 441         }
 442 }
 443
 444 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 445 {
 446         struct list_head *head = &SM_I(sbi)->discard_list;
 447         struct discard_entry *new;
 448         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
 449         int max_blocks = sbi->blocks_per_seg;
 450         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
 451         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
 452         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
 453         unsigned long dmap[entries];
 454         unsigned int start = 0, end = -1;
 455         bool force = (cpc->reason == CP_DISCARD);
 456         int i;
 457
 458         if (!force && !test_opt(sbi, DISCARD))
 459                 return;
 460
 461         if (force && !se->valid_blocks) {
 462                 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 463                 /*
 464                  * if this segment is registered in the prefree list, then
 465                  * we should skip adding a discard candidate, and let the
 466                  * checkpoint do that later.
 467                  */
 468                 mutex_lock(&dirty_i->seglist_lock);
 469                 if (test_bit(cpc->trim_start, dirty_i->dirty_segmap[PRE])) {
 470                         mutex_unlock(&dirty_i->seglist_lock);
 471                         cpc->trimmed += sbi->blocks_per_seg;
 472                         return;
 473                 }
 474                 mutex_unlock(&dirty_i->seglist_lock);
 475
 476                 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
 477                 INIT_LIST_HEAD(&new->list);
 478                 new->blkaddr = START_BLOCK(sbi, cpc->trim_start);
 479                 new->len = sbi->blocks_per_seg;
 480                 list_add_tail(&new->list, head);
 481                 SM_I(sbi)->nr_discards += sbi->blocks_per_seg;
 482                 cpc->trimmed += sbi->blocks_per_seg;
 483                 return;
 484         }
 485
 486         /* zero block will be discarded through the prefree list */
 487         if (!se->valid_blocks || se->valid_blocks == max_blocks)
 488                 return;
 489
 490         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
 491         for (i = 0; i < entries; i++)
 492                 dmap[i] = (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
 493
 494         while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
 495                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
 496                 if (start >= max_blocks)
 497                         break;
 498
 499                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
 500
 501                 if (end - start < cpc->trim_minlen)
 502                         continue;
 503
 504                 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
 505                 INIT_LIST_HEAD(&new->list);
 506                 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
 507                 new->len = end - start;
 508                 cpc->trimmed += end - start;
 509
 510                 list_add_tail(&new->list, head);
 511                 SM_I(sbi)->nr_discards += end - start;
 512         }
 513 }
 514
 515 void release_discard_addrs(struct f2fs_sb_info *sbi)
 516 {
 517         struct list_head *head = &(SM_I(sbi)->discard_list);
 518         struct discard_entry *entry, *this;
 519
 520         /* drop caches */
 521         list_for_each_entry_safe(entry, this, head, list) {
 522                 list_del(&entry->list);
 523                 kmem_cache_free(discard_entry_slab, entry);
 524         }
 525 }
 526
 527 /*
 528  * Should call clear_prefree_segments after checkpoint is done.
 529  */
 530 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
 531 {
 532         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 533         unsigned int segno;
 534
 535         mutex_lock(&dirty_i->seglist_lock);
 536         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
 537                 __set_test_and_free(sbi, segno);
 538         mutex_unlock(&dirty_i->seglist_lock);
 539 }
 540
 541 void clear_prefree_segments(struct f2fs_sb_info *sbi)
 542 {
 543         struct list_head *head = &(SM_I(sbi)->discard_list);
 544         struct discard_entry *entry, *this;
 545         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 546         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
 547         unsigned int start = 0, end = -1;
 548
 549         mutex_lock(&dirty_i->seglist_lock);
 550
 551         while (1) {
 552                 int i;
 553                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
 554                 if (start >= MAIN_SEGS(sbi))
 555                         break;
 556                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
 557                                                                 start + 1);
 558
 559                 for (i = start; i < end; i++)
 560                         clear_bit(i, prefree_map);
 561
 562                 dirty_i->nr_dirty[PRE] -= end - start;
 563
 564                 if (!test_opt(sbi, DISCARD))
 565                         continue;
 566
 567                 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
 568                                 (end - start) << sbi->log_blocks_per_seg);
 569         }
 570         mutex_unlock(&dirty_i->seglist_lock);
 571
 572         /* send small discards */
 573         list_for_each_entry_safe(entry, this, head, list) {
 574                 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
 575                 list_del(&entry->list);
 576                 SM_I(sbi)->nr_discards -= entry->len;
 577                 kmem_cache_free(discard_entry_slab, entry);
 578         }
 579 }
 580
 581 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
 582 {
 583         struct sit_info *sit_i = SIT_I(sbi);
 584
 585         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
 586                 sit_i->dirty_sentries++;
 587                 return false;
 588         }
 589
 590         return true;
 591 }
 592
 593 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
 594                                         unsigned int segno, int modified)
 595 {
 596         struct seg_entry *se = get_seg_entry(sbi, segno);
 597         se->type = type;
 598         if (modified)
 599                 __mark_sit_entry_dirty(sbi, segno);
 600 }
 601
 602 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
 603 {
 604         struct seg_entry *se;
 605         unsigned int segno, offset;
 606         long int new_vblocks;
 607
 608         segno = GET_SEGNO(sbi, blkaddr);
 609
 610         se = get_seg_entry(sbi, segno);
 611         new_vblocks = se->valid_blocks + del;
 612         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 613
 614         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
 615                                 (new_vblocks > sbi->blocks_per_seg)));
 616
 617         se->valid_blocks = new_vblocks;
 618         se->mtime = get_mtime(sbi);
 619         SIT_I(sbi)->max_mtime = se->mtime;
 620
 621         /* Update valid block bitmap */
 622         if (del > 0) {
 623                 if (f2fs_set_bit(offset, se->cur_valid_map))
 624                         f2fs_bug_on(sbi, 1);
 625         } else {
 626                 if (!f2fs_clear_bit(offset, se->cur_valid_map))
 627                         f2fs_bug_on(sbi, 1);
 628         }
 629         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
 630                 se->ckpt_valid_blocks += del;
 631
 632         __mark_sit_entry_dirty(sbi, segno);
 633
 634         /* update total number of valid blocks to be written in ckpt area */
 635         SIT_I(sbi)->written_valid_blocks += del;
 636
 637         if (sbi->segs_per_sec > 1)
 638                 get_sec_entry(sbi, segno)->valid_blocks += del;
 639 }
 640
 641 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
 642 {
 643         update_sit_entry(sbi, new, 1);
 644         if (GET_SEGNO(sbi, old) != NULL_SEGNO)
 645                 update_sit_entry(sbi, old, -1);
 646
 647         locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
 648         locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
 649 }
 650
 651 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
 652 {
 653         unsigned int segno = GET_SEGNO(sbi, addr);
 654         struct sit_info *sit_i = SIT_I(sbi);
 655
 656         f2fs_bug_on(sbi, addr == NULL_ADDR);
 657         if (addr == NEW_ADDR)
 658                 return;
 659
 660         /* add it into sit main buffer */
 661         mutex_lock(&sit_i->sentry_lock);
 662
 663         update_sit_entry(sbi, addr, -1);
 664
 665         /* add it into dirty seglist */
 666         locate_dirty_segment(sbi, segno);
 667
 668         mutex_unlock(&sit_i->sentry_lock);
 669 }
 670
 671 /*
 672  * This function should be resided under the curseg_mutex lock
 673  */
 674 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
 675                                         struct f2fs_summary *sum)
 676 {
 677         struct curseg_info *curseg = CURSEG_I(sbi, type);
 678         void *addr = curseg->sum_blk;
 679         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
 680         memcpy(addr, sum, sizeof(struct f2fs_summary));
 681 }
 682
 683 /*
 684  * Calculate the number of current summary pages for writing
 685  */
 686 int npages_for_summary_flush(struct f2fs_sb_info *sbi)
 687 {
 688         int valid_sum_count = 0;
 689         int i, sum_in_page;
 690
 691         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 692                 if (sbi->ckpt->alloc_type[i] == SSR)
 693                         valid_sum_count += sbi->blocks_per_seg;
 694                 else
 695                         valid_sum_count += curseg_blkoff(sbi, i);
 696         }
 697
 698         sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
 699                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
 700         if (valid_sum_count <= sum_in_page)
 701                 return 1;
 702         else if ((valid_sum_count - sum_in_page) <=
 703                 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
 704                 return 2;
 705         return 3;
 706 }
 707
 708 /*
 709  * Caller should put this summary page
 710  */
 711 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
 712 {
 713         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
 714 }
 715
 716 static void write_sum_page(struct f2fs_sb_info *sbi,
 717                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
 718 {
 719         struct page *page = grab_meta_page(sbi, blk_addr);
 720         void *kaddr = page_address(page);
 721         memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
 722         set_page_dirty(page);
 723         f2fs_put_page(page, 1);
 724 }
 725
 726 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
 727 {
 728         struct curseg_info *curseg = CURSEG_I(sbi, type);
 729         unsigned int segno = curseg->segno + 1;
 730         struct free_segmap_info *free_i = FREE_I(sbi);
 731
 732         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
 733                 return !test_bit(segno, free_i->free_segmap);
 734         return 0;
 735 }
 736
 737 /*
 738  * Find a new segment from the free segments bitmap to right order
 739  * This function should be returned with success, otherwise BUG
 740  */
 741 static void get_new_segment(struct f2fs_sb_info *sbi,
 742                         unsigned int *newseg, bool new_sec, int dir)
 743 {
 744         struct free_segmap_info *free_i = FREE_I(sbi);
 745         unsigned int segno, secno, zoneno;
 746         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
 747         unsigned int hint = *newseg / sbi->segs_per_sec;
 748         unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
 749         unsigned int left_start = hint;
 750         bool init = true;
 751         int go_left = 0;
 752         int i;
 753
 754         write_lock(&free_i->segmap_lock);
 755
 756         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
 757                 segno = find_next_zero_bit(free_i->free_segmap,
 758                                         MAIN_SEGS(sbi), *newseg + 1);
 759                 if (segno - *newseg < sbi->segs_per_sec -
 760                                         (*newseg % sbi->segs_per_sec))
 761                         goto got_it;
 762         }
 763 find_other_zone:
 764         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
 765         if (secno >= MAIN_SECS(sbi)) {
 766                 if (dir == ALLOC_RIGHT) {
 767                         secno = find_next_zero_bit(free_i->free_secmap,
 768                                                         MAIN_SECS(sbi), 0);
 769                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
 770                 } else {
 771                         go_left = 1;
 772                         left_start = hint - 1;
 773                 }
 774         }
 775         if (go_left == 0)
 776                 goto skip_left;
 777
 778         while (test_bit(left_start, free_i->free_secmap)) {
 779                 if (left_start > 0) {
 780                         left_start--;
 781                         continue;
 782                 }
 783                 left_start = find_next_zero_bit(free_i->free_secmap,
 784                                                         MAIN_SECS(sbi), 0);
 785                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
 786                 break;
 787         }
 788         secno = left_start;
 789 skip_left:
 790         hint = secno;
 791         segno = secno * sbi->segs_per_sec;
 792         zoneno = secno / sbi->secs_per_zone;
 793
 794         /* give up on finding another zone */
 795         if (!init)
 796                 goto got_it;
 797         if (sbi->secs_per_zone == 1)
 798                 goto got_it;
 799         if (zoneno == old_zoneno)
 800                 goto got_it;
 801         if (dir == ALLOC_LEFT) {
 802                 if (!go_left && zoneno + 1 >= total_zones)
 803                         goto got_it;
 804                 if (go_left && zoneno == 0)
 805                         goto got_it;
 806         }
 807         for (i = 0; i < NR_CURSEG_TYPE; i++)
 808                 if (CURSEG_I(sbi, i)->zone == zoneno)
 809                         break;
 810
 811         if (i < NR_CURSEG_TYPE) {
 812                 /* zone is in user, try another */
 813                 if (go_left)
 814                         hint = zoneno * sbi->secs_per_zone - 1;
 815                 else if (zoneno + 1 >= total_zones)
 816                         hint = 0;
 817                 else
 818                         hint = (zoneno + 1) * sbi->secs_per_zone;
 819                 init = false;
 820                 goto find_other_zone;
 821         }
 822 got_it:
 823         /* set it as dirty segment in free segmap */
 824         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
 825         __set_inuse(sbi, segno);
 826         *newseg = segno;
 827         write_unlock(&free_i->segmap_lock);
 828 }
 829
 830 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
 831 {
 832         struct curseg_info *curseg = CURSEG_I(sbi, type);
 833         struct summary_footer *sum_footer;
 834
 835         curseg->segno = curseg->next_segno;
 836         curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
 837         curseg->next_blkoff = 0;
 838         curseg->next_segno = NULL_SEGNO;
 839
 840         sum_footer = &(curseg->sum_blk->footer);
 841         memset(sum_footer, 0, sizeof(struct summary_footer));
 842         if (IS_DATASEG(type))
 843                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
 844         if (IS_NODESEG(type))
 845                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
 846         __set_sit_entry_type(sbi, type, curseg->segno, modified);
 847 }
 848
 849 /*
 850  * Allocate a current working segment.
 851  * This function always allocates a free segment in LFS manner.
 852  */
 853 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
 854 {
 855         struct curseg_info *curseg = CURSEG_I(sbi, type);
 856         unsigned int segno = curseg->segno;
 857         int dir = ALLOC_LEFT;
 858
 859         write_sum_page(sbi, curseg->sum_blk,
 860                                 GET_SUM_BLOCK(sbi, segno));
 861         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
 862                 dir = ALLOC_RIGHT;
 863
 864         if (test_opt(sbi, NOHEAP))
 865                 dir = ALLOC_RIGHT;
 866
 867         get_new_segment(sbi, &segno, new_sec, dir);
 868         curseg->next_segno = segno;
 869         reset_curseg(sbi, type, 1);
 870         curseg->alloc_type = LFS;
 871 }
 872
 873 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
 874                         struct curseg_info *seg, block_t start)
 875 {
 876         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
 877         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
 878         unsigned long target_map[entries];
 879         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
 880         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
 881         int i, pos;
 882
 883         for (i = 0; i < entries; i++)
 884                 target_map[i] = ckpt_map[i] | cur_map[i];
 885
 886         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
 887
 888         seg->next_blkoff = pos;
 889 }
 890
 891 /*
 892  * If a segment is written by LFS manner, next block offset is just obtained
 893  * by increasing the current block offset. However, if a segment is written by
 894  * SSR manner, next block offset obtained by calling __next_free_blkoff
 895  */
 896 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
 897                                 struct curseg_info *seg)
 898 {
 899         if (seg->alloc_type == SSR)
 900                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
 901         else
 902                 seg->next_blkoff++;
 903 }
 904
 905 /*
 906  * This function always allocates a used segment(from dirty seglist) by SSR
 907  * manner, so it should recover the existing segment information of valid blocks
 908  */
 909 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
 910 {
 911         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 912         struct curseg_info *curseg = CURSEG_I(sbi, type);
 913         unsigned int new_segno = curseg->next_segno;
 914         struct f2fs_summary_block *sum_node;
 915         struct page *sum_page;
 916
 917         write_sum_page(sbi, curseg->sum_blk,
 918                                 GET_SUM_BLOCK(sbi, curseg->segno));
 919         __set_test_and_inuse(sbi, new_segno);
 920
 921         mutex_lock(&dirty_i->seglist_lock);
 922         __remove_dirty_segment(sbi, new_segno, PRE);
 923         __remove_dirty_segment(sbi, new_segno, DIRTY);
 924         mutex_unlock(&dirty_i->seglist_lock);
 925
 926         reset_curseg(sbi, type, 1);
 927         curseg->alloc_type = SSR;
 928         __next_free_blkoff(sbi, curseg, 0);
 929
 930         if (reuse) {
 931                 sum_page = get_sum_page(sbi, new_segno);
 932                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
 933                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
 934                 f2fs_put_page(sum_page, 1);
 935         }
 936 }
 937
 938 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
 939 {
 940         struct curseg_info *curseg = CURSEG_I(sbi, type);
 941         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
 942
 943         if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
 944                 return v_ops->get_victim(sbi,
 945                                 &(curseg)->next_segno, BG_GC, type, SSR);
 946
 947         /* For data segments, let's do SSR more intensively */
 948         for (; type >= CURSEG_HOT_DATA; type--)
 949                 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
 950                                                 BG_GC, type, SSR))
 951                         return 1;
 952         return 0;
 953 }
 954
 955 /*
 956  * flush out current segment and replace it with new segment
 957  * This function should be returned with success, otherwise BUG
 958  */
 959 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
 960                                                 int type, bool force)
 961 {
 962         struct curseg_info *curseg = CURSEG_I(sbi, type);
 963
 964         if (force)
 965                 new_curseg(sbi, type, true);
 966         else if (type == CURSEG_WARM_NODE)
 967                 new_curseg(sbi, type, false);
 968         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
 969                 new_curseg(sbi, type, false);
 970         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
 971                 change_curseg(sbi, type, true);
 972         else
 973                 new_curseg(sbi, type, false);
 974
 975         stat_inc_seg_type(sbi, curseg);
 976 }
 977
 978 void allocate_new_segments(struct f2fs_sb_info *sbi)
 979 {
 980         struct curseg_info *curseg;
 981         unsigned int old_curseg;
 982         int i;
 983
 984         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 985                 curseg = CURSEG_I(sbi, i);
 986                 old_curseg = curseg->segno;
 987                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
 988                 locate_dirty_segment(sbi, old_curseg);
 989         }
 990 }
 991
 992 static const struct segment_allocation default_salloc_ops = {
 993         .allocate_segment = allocate_segment_by_default,
 994 };
 995
 996 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
 997 {
 998         __u64 start = range->start >> sbi->log_blocksize;
 999         __u64 end = start + (range->len >> sbi->log_blocksize) - 1;
1000         unsigned int start_segno, end_segno;
1001         struct cp_control cpc;
1002
1003         if (range->minlen > SEGMENT_SIZE(sbi) || start >= MAX_BLKADDR(sbi) ||
1004                                                 range->len < sbi->blocksize)
1005                 return -EINVAL;
1006
1007         if (end <= MAIN_BLKADDR(sbi))
1008                 goto out;
1009
1010         /* start/end segment number in main_area */
1011         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1012         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1013                                                 GET_SEGNO(sbi, end);
1014         cpc.reason = CP_DISCARD;
1015         cpc.trim_start = start_segno;
1016         cpc.trim_end = end_segno;
1017         cpc.trim_minlen = range->minlen >> sbi->log_blocksize;
1018         cpc.trimmed = 0;
1019
1020         /* do checkpoint to issue discard commands safely */
1021         write_checkpoint(sbi, &cpc);
1022 out:
1023         range->len = cpc.trimmed << sbi->log_blocksize;
1024         return 0;
1025 }
1026
1027 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1028 {
1029         struct curseg_info *curseg = CURSEG_I(sbi, type);
1030         if (curseg->next_blkoff < sbi->blocks_per_seg)
1031                 return true;
1032         return false;
1033 }
1034
1035 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1036 {
1037         if (p_type == DATA)
1038                 return CURSEG_HOT_DATA;
1039         else
1040                 return CURSEG_HOT_NODE;
1041 }
1042
1043 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1044 {
1045         if (p_type == DATA) {
1046                 struct inode *inode = page->mapping->host;
1047
1048                 if (S_ISDIR(inode->i_mode))
1049                         return CURSEG_HOT_DATA;
1050                 else
1051                         return CURSEG_COLD_DATA;
1052         } else {
1053                 if (IS_DNODE(page) && !is_cold_node(page))
1054                         return CURSEG_HOT_NODE;
1055                 else
1056                         return CURSEG_COLD_NODE;
1057         }
1058 }
1059
1060 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1061 {
1062         if (p_type == DATA) {
1063                 struct inode *inode = page->mapping->host;
1064
1065                 if (S_ISDIR(inode->i_mode))
1066                         return CURSEG_HOT_DATA;
1067                 else if (is_cold_data(page) || file_is_cold(inode))
1068                         return CURSEG_COLD_DATA;
1069                 else
1070                         return CURSEG_WARM_DATA;
1071         } else {
1072                 if (IS_DNODE(page))
1073                         return is_cold_node(page) ? CURSEG_WARM_NODE :
1074                                                 CURSEG_HOT_NODE;
1075                 else
1076                         return CURSEG_COLD_NODE;
1077         }
1078 }
1079
1080 static int __get_segment_type(struct page *page, enum page_type p_type)
1081 {
1082         switch (F2FS_P_SB(page)->active_logs) {
1083         case 2:
1084                 return __get_segment_type_2(page, p_type);
1085         case 4:
1086                 return __get_segment_type_4(page, p_type);
1087         }
1088         /* NR_CURSEG_TYPE(6) logs by default */
1089         f2fs_bug_on(F2FS_P_SB(page),
1090                 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1091         return __get_segment_type_6(page, p_type);
1092 }
1093
1094 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1095                 block_t old_blkaddr, block_t *new_blkaddr,
1096                 struct f2fs_summary *sum, int type)
1097 {
1098         struct sit_info *sit_i = SIT_I(sbi);
1099         struct curseg_info *curseg;
1100
1101         curseg = CURSEG_I(sbi, type);
1102
1103         mutex_lock(&curseg->curseg_mutex);
1104
1105         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1106
1107         /*
1108          * __add_sum_entry should be resided under the curseg_mutex
1109          * because, this function updates a summary entry in the
1110          * current summary block.
1111          */
1112         __add_sum_entry(sbi, type, sum);
1113
1114         mutex_lock(&sit_i->sentry_lock);
1115         __refresh_next_blkoff(sbi, curseg);
1116
1117         stat_inc_block_count(sbi, curseg);
1118
1119         if (!__has_curseg_space(sbi, type))
1120                 sit_i->s_ops->allocate_segment(sbi, type, false);
1121         /*
1122          * SIT information should be updated before segment allocation,
1123          * since SSR needs latest valid block information.
1124          */
1125         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1126
1127         mutex_unlock(&sit_i->sentry_lock);
1128
1129         if (page && IS_NODESEG(type))
1130                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1131
1132         mutex_unlock(&curseg->curseg_mutex);
1133 }
1134
1135 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
1136                         block_t old_blkaddr, block_t *new_blkaddr,
1137                         struct f2fs_summary *sum, struct f2fs_io_info *fio)
1138 {
1139         int type = __get_segment_type(page, fio->type);
1140
1141         allocate_data_block(sbi, page, old_blkaddr, new_blkaddr, sum, type);
1142
1143         /* writeout dirty page into bdev */
1144         f2fs_submit_page_mbio(sbi, page, *new_blkaddr, fio);
1145 }
1146
1147 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1148 {
1149         struct f2fs_io_info fio = {
1150                 .type = META,
1151                 .rw = WRITE_SYNC | REQ_META | REQ_PRIO
1152         };
1153
1154         set_page_writeback(page);
1155         f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1156 }
1157
1158 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
1159                 struct f2fs_io_info *fio,
1160                 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
1161 {
1162         struct f2fs_summary sum;
1163         set_summary(&sum, nid, 0, 0);
1164         do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, fio);
1165 }
1166
1167 void write_data_page(struct page *page, struct dnode_of_data *dn,
1168                 block_t *new_blkaddr, struct f2fs_io_info *fio)
1169 {
1170         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1171         struct f2fs_summary sum;
1172         struct node_info ni;
1173
1174         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1175         get_node_info(sbi, dn->nid, &ni);
1176         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1177
1178         do_write_page(sbi, page, dn->data_blkaddr, new_blkaddr, &sum, fio);
1179 }
1180
1181 void rewrite_data_page(struct page *page, block_t old_blkaddr,
1182                                         struct f2fs_io_info *fio)
1183 {
1184         f2fs_submit_page_mbio(F2FS_P_SB(page), page, old_blkaddr, fio);
1185 }
1186
1187 void recover_data_page(struct f2fs_sb_info *sbi,
1188                         struct page *page, struct f2fs_summary *sum,
1189                         block_t old_blkaddr, block_t new_blkaddr)
1190 {
1191         struct sit_info *sit_i = SIT_I(sbi);
1192         struct curseg_info *curseg;
1193         unsigned int segno, old_cursegno;
1194         struct seg_entry *se;
1195         int type;
1196
1197         segno = GET_SEGNO(sbi, new_blkaddr);
1198         se = get_seg_entry(sbi, segno);
1199         type = se->type;
1200
1201         if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1202                 if (old_blkaddr == NULL_ADDR)
1203                         type = CURSEG_COLD_DATA;
1204                 else
1205                         type = CURSEG_WARM_DATA;
1206         }
1207         curseg = CURSEG_I(sbi, type);
1208
1209         mutex_lock(&curseg->curseg_mutex);
1210         mutex_lock(&sit_i->sentry_lock);
1211
1212         old_cursegno = curseg->segno;
1213
1214         /* change the current segment */
1215         if (segno != curseg->segno) {
1216                 curseg->next_segno = segno;
1217                 change_curseg(sbi, type, true);
1218         }
1219
1220         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1221         __add_sum_entry(sbi, type, sum);
1222
1223         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1224         locate_dirty_segment(sbi, old_cursegno);
1225
1226         mutex_unlock(&sit_i->sentry_lock);
1227         mutex_unlock(&curseg->curseg_mutex);
1228 }
1229
1230 static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1231                                         struct page *page, enum page_type type)
1232 {
1233         enum page_type btype = PAGE_TYPE_OF_BIO(type);
1234         struct f2fs_bio_info *io = &sbi->write_io[btype];
1235         struct bio_vec *bvec;
1236         int i;
1237
1238         down_read(&io->io_rwsem);
1239         if (!io->bio)
1240                 goto out;
1241
1242         bio_for_each_segment_all(bvec, io->bio, i) {
1243                 if (page == bvec->bv_page) {
1244                         up_read(&io->io_rwsem);
1245                         return true;
1246                 }
1247         }
1248
1249 out:
1250         up_read(&io->io_rwsem);
1251         return false;
1252 }
1253
1254 void f2fs_wait_on_page_writeback(struct page *page,
1255                                 enum page_type type)
1256 {
1257         if (PageWriteback(page)) {
1258                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1259
1260                 if (is_merged_page(sbi, page, type))
1261                         f2fs_submit_merged_bio(sbi, type, WRITE);
1262                 wait_on_page_writeback(page);
1263         }
1264 }
1265
1266 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1267 {
1268         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1269         struct curseg_info *seg_i;
1270         unsigned char *kaddr;
1271         struct page *page;
1272         block_t start;
1273         int i, j, offset;
1274
1275         start = start_sum_block(sbi);
1276
1277         page = get_meta_page(sbi, start++);
1278         kaddr = (unsigned char *)page_address(page);
1279
1280         /* Step 1: restore nat cache */
1281         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1282         memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1283
1284         /* Step 2: restore sit cache */
1285         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1286         memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1287                                                 SUM_JOURNAL_SIZE);
1288         offset = 2 * SUM_JOURNAL_SIZE;
1289
1290         /* Step 3: restore summary entries */
1291         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1292                 unsigned short blk_off;
1293                 unsigned int segno;
1294
1295                 seg_i = CURSEG_I(sbi, i);
1296                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1297                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1298                 seg_i->next_segno = segno;
1299                 reset_curseg(sbi, i, 0);
1300                 seg_i->alloc_type = ckpt->alloc_type[i];
1301                 seg_i->next_blkoff = blk_off;
1302
1303                 if (seg_i->alloc_type == SSR)
1304                         blk_off = sbi->blocks_per_seg;
1305
1306                 for (j = 0; j < blk_off; j++) {
1307                         struct f2fs_summary *s;
1308                         s = (struct f2fs_summary *)(kaddr + offset);
1309                         seg_i->sum_blk->entries[j] = *s;
1310                         offset += SUMMARY_SIZE;
1311                         if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1312                                                 SUM_FOOTER_SIZE)
1313                                 continue;
1314
1315                         f2fs_put_page(page, 1);
1316                         page = NULL;
1317
1318                         page = get_meta_page(sbi, start++);
1319                         kaddr = (unsigned char *)page_address(page);
1320                         offset = 0;
1321                 }
1322         }
1323         f2fs_put_page(page, 1);
1324         return 0;
1325 }
1326
1327 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1328 {
1329         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1330         struct f2fs_summary_block *sum;
1331         struct curseg_info *curseg;
1332         struct page *new;
1333         unsigned short blk_off;
1334         unsigned int segno = 0;
1335         block_t blk_addr = 0;
1336
1337         /* get segment number and block addr */
1338         if (IS_DATASEG(type)) {
1339                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1340                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1341                                                         CURSEG_HOT_DATA]);
1342                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1343                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1344                 else
1345                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1346         } else {
1347                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1348                                                         CURSEG_HOT_NODE]);
1349                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1350                                                         CURSEG_HOT_NODE]);
1351                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1352                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1353                                                         type - CURSEG_HOT_NODE);
1354                 else
1355                         blk_addr = GET_SUM_BLOCK(sbi, segno);
1356         }
1357
1358         new = get_meta_page(sbi, blk_addr);
1359         sum = (struct f2fs_summary_block *)page_address(new);
1360
1361         if (IS_NODESEG(type)) {
1362                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1363                         struct f2fs_summary *ns = &sum->entries[0];
1364                         int i;
1365                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1366                                 ns->version = 0;
1367                                 ns->ofs_in_node = 0;
1368                         }
1369                 } else {
1370                         int err;
1371
1372                         err = restore_node_summary(sbi, segno, sum);
1373                         if (err) {
1374                                 f2fs_put_page(new, 1);
1375                                 return err;
1376                         }
1377                 }
1378         }
1379
1380         /* set uncompleted segment to curseg */
1381         curseg = CURSEG_I(sbi, type);
1382         mutex_lock(&curseg->curseg_mutex);
1383         memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1384         curseg->next_segno = segno;
1385         reset_curseg(sbi, type, 0);
1386         curseg->alloc_type = ckpt->alloc_type[type];
1387         curseg->next_blkoff = blk_off;
1388         mutex_unlock(&curseg->curseg_mutex);
1389         f2fs_put_page(new, 1);
1390         return 0;
1391 }
1392
1393 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1394 {
1395         int type = CURSEG_HOT_DATA;
1396         int err;
1397
1398         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1399                 /* restore for compacted data summary */
1400                 if (read_compacted_summaries(sbi))
1401                         return -EINVAL;
1402                 type = CURSEG_HOT_NODE;
1403         }
1404
1405         for (; type <= CURSEG_COLD_NODE; type++) {
1406                 err = read_normal_summaries(sbi, type);
1407                 if (err)
1408                         return err;
1409         }
1410
1411         return 0;
1412 }
1413
1414 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1415 {
1416         struct page *page;
1417         unsigned char *kaddr;
1418         struct f2fs_summary *summary;
1419         struct curseg_info *seg_i;
1420         int written_size = 0;
1421         int i, j;
1422
1423         page = grab_meta_page(sbi, blkaddr++);
1424         kaddr = (unsigned char *)page_address(page);
1425
1426         /* Step 1: write nat cache */
1427         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1428         memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1429         written_size += SUM_JOURNAL_SIZE;
1430
1431         /* Step 2: write sit cache */
1432         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1433         memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1434                                                 SUM_JOURNAL_SIZE);
1435         written_size += SUM_JOURNAL_SIZE;
1436
1437         /* Step 3: write summary entries */
1438         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1439                 unsigned short blkoff;
1440                 seg_i = CURSEG_I(sbi, i);
1441                 if (sbi->ckpt->alloc_type[i] == SSR)
1442                         blkoff = sbi->blocks_per_seg;
1443                 else
1444                         blkoff = curseg_blkoff(sbi, i);
1445
1446                 for (j = 0; j < blkoff; j++) {
1447                         if (!page) {
1448                                 page = grab_meta_page(sbi, blkaddr++);
1449                                 kaddr = (unsigned char *)page_address(page);
1450                                 written_size = 0;
1451                         }
1452                         summary = (struct f2fs_summary *)(kaddr + written_size);
1453                         *summary = seg_i->sum_blk->entries[j];
1454                         written_size += SUMMARY_SIZE;
1455
1456                         if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1457                                                         SUM_FOOTER_SIZE)
1458                                 continue;
1459
1460                         set_page_dirty(page);
1461                         f2fs_put_page(page, 1);
1462                         page = NULL;
1463                 }
1464         }
1465         if (page) {
1466                 set_page_dirty(page);
1467                 f2fs_put_page(page, 1);
1468         }
1469 }
1470
1471 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1472                                         block_t blkaddr, int type)
1473 {
1474         int i, end;
1475         if (IS_DATASEG(type))
1476                 end = type + NR_CURSEG_DATA_TYPE;
1477         else
1478                 end = type + NR_CURSEG_NODE_TYPE;
1479
1480         for (i = type; i < end; i++) {
1481                 struct curseg_info *sum = CURSEG_I(sbi, i);
1482                 mutex_lock(&sum->curseg_mutex);
1483                 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1484                 mutex_unlock(&sum->curseg_mutex);
1485         }
1486 }
1487
1488 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1489 {
1490         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1491                 write_compacted_summaries(sbi, start_blk);
1492         else
1493                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1494 }
1495
1496 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1497 {
1498         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1499                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1500 }
1501
1502 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1503                                         unsigned int val, int alloc)
1504 {
1505         int i;
1506
1507         if (type == NAT_JOURNAL) {
1508                 for (i = 0; i < nats_in_cursum(sum); i++) {
1509                         if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1510                                 return i;
1511                 }
1512                 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1513                         return update_nats_in_cursum(sum, 1);
1514         } else if (type == SIT_JOURNAL) {
1515                 for (i = 0; i < sits_in_cursum(sum); i++)
1516                         if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1517                                 return i;
1518                 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1519                         return update_sits_in_cursum(sum, 1);
1520         }
1521         return -1;
1522 }
1523
1524 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1525                                         unsigned int segno)
1526 {
1527         struct sit_info *sit_i = SIT_I(sbi);
1528         unsigned int offset = SIT_BLOCK_OFFSET(segno);
1529         block_t blk_addr = sit_i->sit_base_addr + offset;
1530
1531         check_seg_range(sbi, segno);
1532
1533         /* calculate sit block address */
1534         if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1535                 blk_addr += sit_i->sit_blocks;
1536
1537         return get_meta_page(sbi, blk_addr);
1538 }
1539
1540 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1541                                         unsigned int start)
1542 {
1543         struct sit_info *sit_i = SIT_I(sbi);
1544         struct page *src_page, *dst_page;
1545         pgoff_t src_off, dst_off;
1546         void *src_addr, *dst_addr;
1547
1548         src_off = current_sit_addr(sbi, start);
1549         dst_off = next_sit_addr(sbi, src_off);
1550
1551         /* get current sit block page without lock */
1552         src_page = get_meta_page(sbi, src_off);
1553         dst_page = grab_meta_page(sbi, dst_off);
1554         f2fs_bug_on(sbi, PageDirty(src_page));
1555
1556         src_addr = page_address(src_page);
1557         dst_addr = page_address(dst_page);
1558         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1559
1560         set_page_dirty(dst_page);
1561         f2fs_put_page(src_page, 1);
1562
1563         set_to_next_sit(sit_i, start);
1564
1565         return dst_page;
1566 }
1567
1568 static struct sit_entry_set *grab_sit_entry_set(void)
1569 {
1570         struct sit_entry_set *ses =
1571                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_ATOMIC);
1572
1573         ses->entry_cnt = 0;
1574         INIT_LIST_HEAD(&ses->set_list);
1575         return ses;
1576 }
1577
1578 static void release_sit_entry_set(struct sit_entry_set *ses)
1579 {
1580         list_del(&ses->set_list);
1581         kmem_cache_free(sit_entry_set_slab, ses);
1582 }
1583
1584 static void adjust_sit_entry_set(struct sit_entry_set *ses,
1585                                                 struct list_head *head)
1586 {
1587         struct sit_entry_set *next = ses;
1588
1589         if (list_is_last(&ses->set_list, head))
1590                 return;
1591
1592         list_for_each_entry_continue(next, head, set_list)
1593                 if (ses->entry_cnt <= next->entry_cnt)
1594                         break;
1595
1596         list_move_tail(&ses->set_list, &next->set_list);
1597 }
1598
1599 static void add_sit_entry(unsigned int segno, struct list_head *head)
1600 {
1601         struct sit_entry_set *ses;
1602         unsigned int start_segno = START_SEGNO(segno);
1603
1604         list_for_each_entry(ses, head, set_list) {
1605                 if (ses->start_segno == start_segno) {
1606                         ses->entry_cnt++;
1607                         adjust_sit_entry_set(ses, head);
1608                         return;
1609                 }
1610         }
1611
1612         ses = grab_sit_entry_set();
1613
1614         ses->start_segno = start_segno;
1615         ses->entry_cnt++;
1616         list_add(&ses->set_list, head);
1617 }
1618
1619 static void add_sits_in_set(struct f2fs_sb_info *sbi)
1620 {
1621         struct f2fs_sm_info *sm_info = SM_I(sbi);
1622         struct list_head *set_list = &sm_info->sit_entry_set;
1623         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
1624         unsigned int segno;
1625
1626         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
1627                 add_sit_entry(segno, set_list);
1628 }
1629
1630 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
1631 {
1632         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1633         struct f2fs_summary_block *sum = curseg->sum_blk;
1634         int i;
1635
1636         for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1637                 unsigned int segno;
1638                 bool dirtied;
1639
1640                 segno = le32_to_cpu(segno_in_journal(sum, i));
1641                 dirtied = __mark_sit_entry_dirty(sbi, segno);
1642
1643                 if (!dirtied)
1644                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
1645         }
1646         update_sits_in_cursum(sum, -sits_in_cursum(sum));
1647 }
1648
1649 /*
1650  * CP calls this function, which flushes SIT entries including sit_journal,
1651  * and moves prefree segs to free segs.
1652  */
1653 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1654 {
1655         struct sit_info *sit_i = SIT_I(sbi);
1656         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1657         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1658         struct f2fs_summary_block *sum = curseg->sum_blk;
1659         struct sit_entry_set *ses, *tmp;
1660         struct list_head *head = &SM_I(sbi)->sit_entry_set;
1661         bool to_journal = true;
1662         struct seg_entry *se;
1663
1664         mutex_lock(&curseg->curseg_mutex);
1665         mutex_lock(&sit_i->sentry_lock);
1666
1667         /*
1668          * add and account sit entries of dirty bitmap in sit entry
1669          * set temporarily
1670          */
1671         add_sits_in_set(sbi);
1672
1673         /*
1674          * if there are no enough space in journal to store dirty sit
1675          * entries, remove all entries from journal and add and account
1676          * them in sit entry set.
1677          */
1678         if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL))
1679                 remove_sits_in_journal(sbi);
1680
1681         if (!sit_i->dirty_sentries)
1682                 goto out;
1683
1684         /*
1685          * there are two steps to flush sit entries:
1686          * #1, flush sit entries to journal in current cold data summary block.
1687          * #2, flush sit entries to sit page.
1688          */
1689         list_for_each_entry_safe(ses, tmp, head, set_list) {
1690                 struct page *page;
1691                 struct f2fs_sit_block *raw_sit = NULL;
1692                 unsigned int start_segno = ses->start_segno;
1693                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
1694                                                 (unsigned long)MAIN_SEGS(sbi));
1695                 unsigned int segno = start_segno;
1696
1697                 if (to_journal &&
1698                         !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL))
1699                         to_journal = false;
1700
1701                 if (!to_journal) {
1702                         page = get_next_sit_page(sbi, start_segno);
1703                         raw_sit = page_address(page);
1704                 }
1705
1706                 /* flush dirty sit entries in region of current sit set */
1707                 for_each_set_bit_from(segno, bitmap, end) {
1708                         int offset, sit_offset;
1709
1710                         se = get_seg_entry(sbi, segno);
1711
1712                         /* add discard candidates */
1713                         if (SM_I(sbi)->nr_discards < SM_I(sbi)->max_discards) {
1714                                 cpc->trim_start = segno;
1715                                 add_discard_addrs(sbi, cpc);
1716                         }
1717
1718                         if (to_journal) {
1719                                 offset = lookup_journal_in_cursum(sum,
1720                                                         SIT_JOURNAL, segno, 1);
1721                                 f2fs_bug_on(sbi, offset < 0);
1722                                 segno_in_journal(sum, offset) =
1723                                                         cpu_to_le32(segno);
1724                                 seg_info_to_raw_sit(se,
1725                                                 &sit_in_journal(sum, offset));
1726                         } else {
1727                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1728                                 seg_info_to_raw_sit(se,
1729                                                 &raw_sit->entries[sit_offset]);
1730                         }
1731
1732                         __clear_bit(segno, bitmap);
1733                         sit_i->dirty_sentries--;
1734                         ses->entry_cnt--;
1735                 }
1736
1737                 if (!to_journal)
1738                         f2fs_put_page(page, 1);
1739
1740                 f2fs_bug_on(sbi, ses->entry_cnt);
1741                 release_sit_entry_set(ses);
1742         }
1743
1744         f2fs_bug_on(sbi, !list_empty(head));
1745         f2fs_bug_on(sbi, sit_i->dirty_sentries);
1746 out:
1747         if (cpc->reason == CP_DISCARD) {
1748                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
1749                         add_discard_addrs(sbi, cpc);
1750         }
1751         mutex_unlock(&sit_i->sentry_lock);
1752         mutex_unlock(&curseg->curseg_mutex);
1753
1754         set_prefree_as_free_segments(sbi);
1755 }
1756
1757 static int build_sit_info(struct f2fs_sb_info *sbi)
1758 {
1759         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1760         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1761         struct sit_info *sit_i;
1762         unsigned int sit_segs, start;
1763         char *src_bitmap, *dst_bitmap;
1764         unsigned int bitmap_size;
1765
1766         /* allocate memory for SIT information */
1767         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1768         if (!sit_i)
1769                 return -ENOMEM;
1770
1771         SM_I(sbi)->sit_info = sit_i;
1772
1773         sit_i->sentries = vzalloc(MAIN_SEGS(sbi) * sizeof(struct seg_entry));
1774         if (!sit_i->sentries)
1775                 return -ENOMEM;
1776
1777         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
1778         sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1779         if (!sit_i->dirty_sentries_bitmap)
1780                 return -ENOMEM;
1781
1782         for (start = 0; start < MAIN_SEGS(sbi); start++) {
1783                 sit_i->sentries[start].cur_valid_map
1784                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1785                 sit_i->sentries[start].ckpt_valid_map
1786                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1787                 if (!sit_i->sentries[start].cur_valid_map
1788                                 || !sit_i->sentries[start].ckpt_valid_map)
1789                         return -ENOMEM;
1790         }
1791
1792         if (sbi->segs_per_sec > 1) {
1793                 sit_i->sec_entries = vzalloc(MAIN_SECS(sbi) *
1794                                         sizeof(struct sec_entry));
1795                 if (!sit_i->sec_entries)
1796                         return -ENOMEM;
1797         }
1798
1799         /* get information related with SIT */
1800         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1801
1802         /* setup SIT bitmap from ckeckpoint pack */
1803         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1804         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1805
1806         dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1807         if (!dst_bitmap)
1808                 return -ENOMEM;
1809
1810         /* init SIT information */
1811         sit_i->s_ops = &default_salloc_ops;
1812
1813         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1814         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1815         sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1816         sit_i->sit_bitmap = dst_bitmap;
1817         sit_i->bitmap_size = bitmap_size;
1818         sit_i->dirty_sentries = 0;
1819         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1820         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1821         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1822         mutex_init(&sit_i->sentry_lock);
1823         return 0;
1824 }
1825
1826 static int build_free_segmap(struct f2fs_sb_info *sbi)
1827 {
1828         struct free_segmap_info *free_i;
1829         unsigned int bitmap_size, sec_bitmap_size;
1830
1831         /* allocate memory for free segmap information */
1832         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1833         if (!free_i)
1834                 return -ENOMEM;
1835
1836         SM_I(sbi)->free_info = free_i;
1837
1838         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
1839         free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1840         if (!free_i->free_segmap)
1841                 return -ENOMEM;
1842
1843         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
1844         free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1845         if (!free_i->free_secmap)
1846                 return -ENOMEM;
1847
1848         /* set all segments as dirty temporarily */
1849         memset(free_i->free_segmap, 0xff, bitmap_size);
1850         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1851
1852         /* init free segmap information */
1853         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
1854         free_i->free_segments = 0;
1855         free_i->free_sections = 0;
1856         rwlock_init(&free_i->segmap_lock);
1857         return 0;
1858 }
1859
1860 static int build_curseg(struct f2fs_sb_info *sbi)
1861 {
1862         struct curseg_info *array;
1863         int i;
1864
1865         array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
1866         if (!array)
1867                 return -ENOMEM;
1868
1869         SM_I(sbi)->curseg_array = array;
1870
1871         for (i = 0; i < NR_CURSEG_TYPE; i++) {
1872                 mutex_init(&array[i].curseg_mutex);
1873                 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1874                 if (!array[i].sum_blk)
1875                         return -ENOMEM;
1876                 array[i].segno = NULL_SEGNO;
1877                 array[i].next_blkoff = 0;
1878         }
1879         return restore_curseg_summaries(sbi);
1880 }
1881
1882 static void build_sit_entries(struct f2fs_sb_info *sbi)
1883 {
1884         struct sit_info *sit_i = SIT_I(sbi);
1885         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1886         struct f2fs_summary_block *sum = curseg->sum_blk;
1887         int sit_blk_cnt = SIT_BLK_CNT(sbi);
1888         unsigned int i, start, end;
1889         unsigned int readed, start_blk = 0;
1890         int nrpages = MAX_BIO_BLOCKS(sbi);
1891
1892         do {
1893                 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT);
1894
1895                 start = start_blk * sit_i->sents_per_block;
1896                 end = (start_blk + readed) * sit_i->sents_per_block;
1897
1898                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
1899                         struct seg_entry *se = &sit_i->sentries[start];
1900                         struct f2fs_sit_block *sit_blk;
1901                         struct f2fs_sit_entry sit;
1902                         struct page *page;
1903
1904                         mutex_lock(&curseg->curseg_mutex);
1905                         for (i = 0; i < sits_in_cursum(sum); i++) {
1906                                 if (le32_to_cpu(segno_in_journal(sum, i))
1907                                                                 == start) {
1908                                         sit = sit_in_journal(sum, i);
1909                                         mutex_unlock(&curseg->curseg_mutex);
1910                                         goto got_it;
1911                                 }
1912                         }
1913                         mutex_unlock(&curseg->curseg_mutex);
1914
1915                         page = get_current_sit_page(sbi, start);
1916                         sit_blk = (struct f2fs_sit_block *)page_address(page);
1917                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1918                         f2fs_put_page(page, 1);
1919 got_it:
1920                         check_block_count(sbi, start, &sit);
1921                         seg_info_from_raw_sit(se, &sit);
1922                         if (sbi->segs_per_sec > 1) {
1923                                 struct sec_entry *e = get_sec_entry(sbi, start);
1924                                 e->valid_blocks += se->valid_blocks;
1925                         }
1926                 }
1927                 start_blk += readed;
1928         } while (start_blk < sit_blk_cnt);
1929 }
1930
1931 static void init_free_segmap(struct f2fs_sb_info *sbi)
1932 {
1933         unsigned int start;
1934         int type;
1935
1936         for (start = 0; start < MAIN_SEGS(sbi); start++) {
1937                 struct seg_entry *sentry = get_seg_entry(sbi, start);
1938                 if (!sentry->valid_blocks)
1939                         __set_free(sbi, start);
1940         }
1941
1942         /* set use the current segments */
1943         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1944                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1945                 __set_test_and_inuse(sbi, curseg_t->segno);
1946         }
1947 }
1948
1949 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1950 {
1951         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1952         struct free_segmap_info *free_i = FREE_I(sbi);
1953         unsigned int segno = 0, offset = 0;
1954         unsigned short valid_blocks;
1955
1956         while (1) {
1957                 /* find dirty segment based on free segmap */
1958                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
1959                 if (segno >= MAIN_SEGS(sbi))
1960                         break;
1961                 offset = segno + 1;
1962                 valid_blocks = get_valid_blocks(sbi, segno, 0);
1963                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
1964                         continue;
1965                 if (valid_blocks > sbi->blocks_per_seg) {
1966                         f2fs_bug_on(sbi, 1);
1967                         continue;
1968                 }
1969                 mutex_lock(&dirty_i->seglist_lock);
1970                 __locate_dirty_segment(sbi, segno, DIRTY);
1971                 mutex_unlock(&dirty_i->seglist_lock);
1972         }
1973 }
1974
1975 static int init_victim_secmap(struct f2fs_sb_info *sbi)
1976 {
1977         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1978         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
1979
1980         dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
1981         if (!dirty_i->victim_secmap)
1982                 return -ENOMEM;
1983         return 0;
1984 }
1985
1986 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1987 {
1988         struct dirty_seglist_info *dirty_i;
1989         unsigned int bitmap_size, i;
1990
1991         /* allocate memory for dirty segments list information */
1992         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1993         if (!dirty_i)
1994                 return -ENOMEM;
1995
1996         SM_I(sbi)->dirty_info = dirty_i;
1997         mutex_init(&dirty_i->seglist_lock);
1998
1999         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2000
2001         for (i = 0; i < NR_DIRTY_TYPE; i++) {
2002                 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
2003                 if (!dirty_i->dirty_segmap[i])
2004                         return -ENOMEM;
2005         }
2006
2007         init_dirty_segmap(sbi);
2008         return init_victim_secmap(sbi);
2009 }
2010
2011 /*
2012  * Update min, max modified time for cost-benefit GC algorithm
2013  */
2014 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2015 {
2016         struct sit_info *sit_i = SIT_I(sbi);
2017         unsigned int segno;
2018
2019         mutex_lock(&sit_i->sentry_lock);
2020
2021         sit_i->min_mtime = LLONG_MAX;
2022
2023         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2024                 unsigned int i;
2025                 unsigned long long mtime = 0;
2026
2027                 for (i = 0; i < sbi->segs_per_sec; i++)
2028                         mtime += get_seg_entry(sbi, segno + i)->mtime;
2029
2030                 mtime = div_u64(mtime, sbi->segs_per_sec);
2031
2032                 if (sit_i->min_mtime > mtime)
2033                         sit_i->min_mtime = mtime;
2034         }
2035         sit_i->max_mtime = get_mtime(sbi);
2036         mutex_unlock(&sit_i->sentry_lock);
2037 }
2038
2039 int build_segment_manager(struct f2fs_sb_info *sbi)
2040 {
2041         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2042         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2043         struct f2fs_sm_info *sm_info;
2044         int err;
2045
2046         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2047         if (!sm_info)
2048                 return -ENOMEM;
2049
2050         /* init sm info */
2051         sbi->sm_info = sm_info;
2052         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2053         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2054         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2055         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2056         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2057         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2058         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2059         sm_info->rec_prefree_segments = sm_info->main_segments *
2060                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2061         sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2062         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2063         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2064
2065         INIT_LIST_HEAD(&sm_info->discard_list);
2066         sm_info->nr_discards = 0;
2067         sm_info->max_discards = 0;
2068
2069         INIT_LIST_HEAD(&sm_info->sit_entry_set);
2070
2071         if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2072                 err = create_flush_cmd_control(sbi);
2073                 if (err)
2074                         return err;
2075         }
2076
2077         err = build_sit_info(sbi);
2078         if (err)
2079                 return err;
2080         err = build_free_segmap(sbi);
2081         if (err)
2082                 return err;
2083         err = build_curseg(sbi);
2084         if (err)
2085                 return err;
2086
2087         /* reinit free segmap based on SIT */
2088         build_sit_entries(sbi);
2089
2090         init_free_segmap(sbi);
2091         err = build_dirty_segmap(sbi);
2092         if (err)
2093                 return err;
2094
2095         init_min_max_mtime(sbi);
2096         return 0;
2097 }
2098
2099 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2100                 enum dirty_type dirty_type)
2101 {
2102         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2103
2104         mutex_lock(&dirty_i->seglist_lock);
2105         kfree(dirty_i->dirty_segmap[dirty_type]);
2106         dirty_i->nr_dirty[dirty_type] = 0;
2107         mutex_unlock(&dirty_i->seglist_lock);
2108 }
2109
2110 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2111 {
2112         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2113         kfree(dirty_i->victim_secmap);
2114 }
2115
2116 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2117 {
2118         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2119         int i;
2120
2121         if (!dirty_i)
2122                 return;
2123
2124         /* discard pre-free/dirty segments list */
2125         for (i = 0; i < NR_DIRTY_TYPE; i++)
2126                 discard_dirty_segmap(sbi, i);
2127
2128         destroy_victim_secmap(sbi);
2129         SM_I(sbi)->dirty_info = NULL;
2130         kfree(dirty_i);
2131 }
2132
2133 static void destroy_curseg(struct f2fs_sb_info *sbi)
2134 {
2135         struct curseg_info *array = SM_I(sbi)->curseg_array;
2136         int i;
2137
2138         if (!array)
2139                 return;
2140         SM_I(sbi)->curseg_array = NULL;
2141         for (i = 0; i < NR_CURSEG_TYPE; i++)
2142                 kfree(array[i].sum_blk);
2143         kfree(array);
2144 }
2145
2146 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2147 {
2148         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2149         if (!free_i)
2150                 return;
2151         SM_I(sbi)->free_info = NULL;
2152         kfree(free_i->free_segmap);
2153         kfree(free_i->free_secmap);
2154         kfree(free_i);
2155 }
2156
2157 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2158 {
2159         struct sit_info *sit_i = SIT_I(sbi);
2160         unsigned int start;
2161
2162         if (!sit_i)
2163                 return;
2164
2165         if (sit_i->sentries) {
2166                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2167                         kfree(sit_i->sentries[start].cur_valid_map);
2168                         kfree(sit_i->sentries[start].ckpt_valid_map);
2169                 }
2170         }
2171         vfree(sit_i->sentries);
2172         vfree(sit_i->sec_entries);
2173         kfree(sit_i->dirty_sentries_bitmap);
2174
2175         SM_I(sbi)->sit_info = NULL;
2176         kfree(sit_i->sit_bitmap);
2177         kfree(sit_i);
2178 }
2179
2180 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2181 {
2182         struct f2fs_sm_info *sm_info = SM_I(sbi);
2183
2184         if (!sm_info)
2185                 return;
2186         destroy_flush_cmd_control(sbi);
2187         destroy_dirty_segmap(sbi);
2188         destroy_curseg(sbi);
2189         destroy_free_segmap(sbi);
2190         destroy_sit_info(sbi);
2191         sbi->sm_info = NULL;
2192         kfree(sm_info);
2193 }
2194
2195 int __init create_segment_manager_caches(void)
2196 {
2197         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2198                         sizeof(struct discard_entry));
2199         if (!discard_entry_slab)
2200                 goto fail;
2201
2202         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2203                         sizeof(struct nat_entry_set));
2204         if (!sit_entry_set_slab)
2205                 goto destory_discard_entry;
2206
2207         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2208                         sizeof(struct inmem_pages));
2209         if (!inmem_entry_slab)
2210                 goto destroy_sit_entry_set;
2211         return 0;
2212
2213 destroy_sit_entry_set:
2214         kmem_cache_destroy(sit_entry_set_slab);
2215 destory_discard_entry:
2216         kmem_cache_destroy(discard_entry_slab);
2217 fail:
2218         return -ENOMEM;
2219 }
2220
2221 void destroy_segment_manager_caches(void)
2222 {
2223         kmem_cache_destroy(sit_entry_set_slab);
2224         kmem_cache_destroy(discard_entry_slab);
2225         kmem_cache_destroy(inmem_entry_slab);
2226 }