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