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/swap.h>
  18 #include <linux/timer.h>
  19 #include <linux/freezer.h>
  20
  21 #include "f2fs.h"
  22 #include "segment.h"
  23 #include "node.h"
  24 #include "trace.h"
  25 #include <trace/events/f2fs.h>
  26
  27 #define __reverse_ffz(x) __reverse_ffs(~(x))
  28
  29 static struct kmem_cache *discard_entry_slab;
  30 static struct kmem_cache *discard_cmd_slab;
  31 static struct kmem_cache *sit_entry_set_slab;
  32 static struct kmem_cache *inmem_entry_slab;
  33
  34 static unsigned long __reverse_ulong(unsigned char *str)
  35 {
  36         unsigned long tmp = 0;
  37         int shift = 24, idx = 0;
  38
  39 #if BITS_PER_LONG == 64
  40         shift = 56;
  41 #endif
  42         while (shift >= 0) {
  43                 tmp |= (unsigned long)str[idx++] << shift;
  44                 shift -= BITS_PER_BYTE;
  45         }
  46         return tmp;
  47 }
  48
  49 /*
  50  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  51  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  52  */
  53 static inline unsigned long __reverse_ffs(unsigned long word)
  54 {
  55         int num = 0;
  56
  57 #if BITS_PER_LONG == 64
  58         if ((word & 0xffffffff00000000UL) == 0)
  59                 num += 32;
  60         else
  61                 word >>= 32;
  62 #endif
  63         if ((word & 0xffff0000) == 0)
  64                 num += 16;
  65         else
  66                 word >>= 16;
  67
  68         if ((word & 0xff00) == 0)
  69                 num += 8;
  70         else
  71                 word >>= 8;
  72
  73         if ((word & 0xf0) == 0)
  74                 num += 4;
  75         else
  76                 word >>= 4;
  77
  78         if ((word & 0xc) == 0)
  79                 num += 2;
  80         else
  81                 word >>= 2;
  82
  83         if ((word & 0x2) == 0)
  84                 num += 1;
  85         return num;
  86 }
  87
  88 /*
  89  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  90  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  91  * @size must be integral times of unsigned long.
  92  * Example:
  93  *                             MSB <--> LSB
  94  *   f2fs_set_bit(0, bitmap) => 1000 0000
  95  *   f2fs_set_bit(7, bitmap) => 0000 0001
  96  */
  97 static unsigned long __find_rev_next_bit(const unsigned long *addr,
  98                         unsigned long size, unsigned long offset)
  99 {
 100         const unsigned long *p = addr + BIT_WORD(offset);
 101         unsigned long result = size;
 102         unsigned long tmp;
 103
 104         if (offset >= size)
 105                 return size;
 106
 107         size -= (offset & ~(BITS_PER_LONG - 1));
 108         offset %= BITS_PER_LONG;
 109
 110         while (1) {
 111                 if (*p == 0)
 112                         goto pass;
 113
 114                 tmp = __reverse_ulong((unsigned char *)p);
 115
 116                 tmp &= ~0UL >> offset;
 117                 if (size < BITS_PER_LONG)
 118                         tmp &= (~0UL << (BITS_PER_LONG - size));
 119                 if (tmp)
 120                         goto found;
 121 pass:
 122                 if (size <= BITS_PER_LONG)
 123                         break;
 124                 size -= BITS_PER_LONG;
 125                 offset = 0;
 126                 p++;
 127         }
 128         return result;
 129 found:
 130         return result - size + __reverse_ffs(tmp);
 131 }
 132
 133 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 134                         unsigned long size, unsigned long offset)
 135 {
 136         const unsigned long *p = addr + BIT_WORD(offset);
 137         unsigned long result = size;
 138         unsigned long tmp;
 139
 140         if (offset >= size)
 141                 return size;
 142
 143         size -= (offset & ~(BITS_PER_LONG - 1));
 144         offset %= BITS_PER_LONG;
 145
 146         while (1) {
 147                 if (*p == ~0UL)
 148                         goto pass;
 149
 150                 tmp = __reverse_ulong((unsigned char *)p);
 151
 152                 if (offset)
 153                         tmp |= ~0UL << (BITS_PER_LONG - offset);
 154                 if (size < BITS_PER_LONG)
 155                         tmp |= ~0UL >> size;
 156                 if (tmp != ~0UL)
 157                         goto found;
 158 pass:
 159                 if (size <= BITS_PER_LONG)
 160                         break;
 161                 size -= BITS_PER_LONG;
 162                 offset = 0;
 163                 p++;
 164         }
 165         return result;
 166 found:
 167         return result - size + __reverse_ffz(tmp);
 168 }
 169
 170 void register_inmem_page(struct inode *inode, struct page *page)
 171 {
 172         struct f2fs_inode_info *fi = F2FS_I(inode);
 173         struct inmem_pages *new;
 174
 175         f2fs_trace_pid(page);
 176
 177         set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
 178         SetPagePrivate(page);
 179
 180         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 181
 182         /* add atomic page indices to the list */
 183         new->page = page;
 184         INIT_LIST_HEAD(&new->list);
 185
 186         /* increase reference count with clean state */
 187         mutex_lock(&fi->inmem_lock);
 188         get_page(page);
 189         list_add_tail(&new->list, &fi->inmem_pages);
 190         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 191         mutex_unlock(&fi->inmem_lock);
 192
 193         trace_f2fs_register_inmem_page(page, INMEM);
 194 }
 195
 196 static int __revoke_inmem_pages(struct inode *inode,
 197                                 struct list_head *head, bool drop, bool recover)
 198 {
 199         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 200         struct inmem_pages *cur, *tmp;
 201         int err = 0;
 202
 203         list_for_each_entry_safe(cur, tmp, head, list) {
 204                 struct page *page = cur->page;
 205
 206                 if (drop)
 207                         trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 208
 209                 lock_page(page);
 210
 211                 if (recover) {
 212                         struct dnode_of_data dn;
 213                         struct node_info ni;
 214
 215                         trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 216
 217                         set_new_dnode(&dn, inode, NULL, NULL, 0);
 218                         if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) {
 219                                 err = -EAGAIN;
 220                                 goto next;
 221                         }
 222                         get_node_info(sbi, dn.nid, &ni);
 223                         f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 224                                         cur->old_addr, ni.version, true, true);
 225                         f2fs_put_dnode(&dn);
 226                 }
 227 next:
 228                 /* we don't need to invalidate this in the sccessful status */
 229                 if (drop || recover)
 230                         ClearPageUptodate(page);
 231                 set_page_private(page, 0);
 232                 ClearPagePrivate(page);
 233                 f2fs_put_page(page, 1);
 234
 235                 list_del(&cur->list);
 236                 kmem_cache_free(inmem_entry_slab, cur);
 237                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 238         }
 239         return err;
 240 }
 241
 242 void drop_inmem_pages(struct inode *inode)
 243 {
 244         struct f2fs_inode_info *fi = F2FS_I(inode);
 245
 246         mutex_lock(&fi->inmem_lock);
 247         __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 248         mutex_unlock(&fi->inmem_lock);
 249
 250         clear_inode_flag(inode, FI_ATOMIC_FILE);
 251         stat_dec_atomic_write(inode);
 252 }
 253
 254 void drop_inmem_page(struct inode *inode, struct page *page)
 255 {
 256         struct f2fs_inode_info *fi = F2FS_I(inode);
 257         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 258         struct list_head *head = &fi->inmem_pages;
 259         struct inmem_pages *cur = NULL;
 260
 261         f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
 262
 263         mutex_lock(&fi->inmem_lock);
 264         list_for_each_entry(cur, head, list) {
 265                 if (cur->page == page)
 266                         break;
 267         }
 268
 269         f2fs_bug_on(sbi, !cur || cur->page != page);
 270         list_del(&cur->list);
 271         mutex_unlock(&fi->inmem_lock);
 272
 273         dec_page_count(sbi, F2FS_INMEM_PAGES);
 274         kmem_cache_free(inmem_entry_slab, cur);
 275
 276         ClearPageUptodate(page);
 277         set_page_private(page, 0);
 278         ClearPagePrivate(page);
 279         f2fs_put_page(page, 0);
 280
 281         trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 282 }
 283
 284 static int __commit_inmem_pages(struct inode *inode,
 285                                         struct list_head *revoke_list)
 286 {
 287         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 288         struct f2fs_inode_info *fi = F2FS_I(inode);
 289         struct inmem_pages *cur, *tmp;
 290         struct f2fs_io_info fio = {
 291                 .sbi = sbi,
 292                 .type = DATA,
 293                 .op = REQ_OP_WRITE,
 294                 .op_flags = REQ_SYNC | REQ_PRIO,
 295         };
 296         pgoff_t last_idx = ULONG_MAX;
 297         int err = 0;
 298
 299         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 300                 struct page *page = cur->page;
 301
 302                 lock_page(page);
 303                 if (page->mapping == inode->i_mapping) {
 304                         trace_f2fs_commit_inmem_page(page, INMEM);
 305
 306                         set_page_dirty(page);
 307                         f2fs_wait_on_page_writeback(page, DATA, true);
 308                         if (clear_page_dirty_for_io(page)) {
 309                                 inode_dec_dirty_pages(inode);
 310                                 remove_dirty_inode(inode);
 311                         }
 312
 313                         fio.page = page;
 314                         fio.old_blkaddr = NULL_ADDR;
 315                         fio.encrypted_page = NULL;
 316                         fio.need_lock = LOCK_DONE;
 317                         err = do_write_data_page(&fio);
 318                         if (err) {
 319                                 unlock_page(page);
 320                                 break;
 321                         }
 322
 323                         /* record old blkaddr for revoking */
 324                         cur->old_addr = fio.old_blkaddr;
 325                         last_idx = page->index;
 326                 }
 327                 unlock_page(page);
 328                 list_move_tail(&cur->list, revoke_list);
 329         }
 330
 331         if (last_idx != ULONG_MAX)
 332                 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
 333
 334         if (!err)
 335                 __revoke_inmem_pages(inode, revoke_list, false, false);
 336
 337         return err;
 338 }
 339
 340 int commit_inmem_pages(struct inode *inode)
 341 {
 342         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 343         struct f2fs_inode_info *fi = F2FS_I(inode);
 344         struct list_head revoke_list;
 345         int err;
 346
 347         INIT_LIST_HEAD(&revoke_list);
 348         f2fs_balance_fs(sbi, true);
 349         f2fs_lock_op(sbi);
 350
 351         set_inode_flag(inode, FI_ATOMIC_COMMIT);
 352
 353         mutex_lock(&fi->inmem_lock);
 354         err = __commit_inmem_pages(inode, &revoke_list);
 355         if (err) {
 356                 int ret;
 357                 /*
 358                  * try to revoke all committed pages, but still we could fail
 359                  * due to no memory or other reason, if that happened, EAGAIN
 360                  * will be returned, which means in such case, transaction is
 361                  * already not integrity, caller should use journal to do the
 362                  * recovery or rewrite & commit last transaction. For other
 363                  * error number, revoking was done by filesystem itself.
 364                  */
 365                 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
 366                 if (ret)
 367                         err = ret;
 368
 369                 /* drop all uncommitted pages */
 370                 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 371         }
 372         mutex_unlock(&fi->inmem_lock);
 373
 374         clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 375
 376         f2fs_unlock_op(sbi);
 377         return err;
 378 }
 379
 380 /*
 381  * This function balances dirty node and dentry pages.
 382  * In addition, it controls garbage collection.
 383  */
 384 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 385 {
 386 #ifdef CONFIG_F2FS_FAULT_INJECTION
 387         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 388                 f2fs_show_injection_info(FAULT_CHECKPOINT);
 389                 f2fs_stop_checkpoint(sbi, false);
 390         }
 391 #endif
 392
 393         /* balance_fs_bg is able to be pending */
 394         if (need && excess_cached_nats(sbi))
 395                 f2fs_balance_fs_bg(sbi);
 396
 397         /*
 398          * We should do GC or end up with checkpoint, if there are so many dirty
 399          * dir/node pages without enough free segments.
 400          */
 401         if (has_not_enough_free_secs(sbi, 0, 0)) {
 402                 mutex_lock(&sbi->gc_mutex);
 403                 f2fs_gc(sbi, false, false, NULL_SEGNO);
 404         }
 405 }
 406
 407 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 408 {
 409         /* try to shrink extent cache when there is no enough memory */
 410         if (!available_free_memory(sbi, EXTENT_CACHE))
 411                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 412
 413         /* check the # of cached NAT entries */
 414         if (!available_free_memory(sbi, NAT_ENTRIES))
 415                 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 416
 417         if (!available_free_memory(sbi, FREE_NIDS))
 418                 try_to_free_nids(sbi, MAX_FREE_NIDS);
 419         else
 420                 build_free_nids(sbi, false, false);
 421
 422         if (!is_idle(sbi) && !excess_dirty_nats(sbi))
 423                 return;
 424
 425         /* checkpoint is the only way to shrink partial cached entries */
 426         if (!available_free_memory(sbi, NAT_ENTRIES) ||
 427                         !available_free_memory(sbi, INO_ENTRIES) ||
 428                         excess_prefree_segs(sbi) ||
 429                         excess_dirty_nats(sbi) ||
 430                         f2fs_time_over(sbi, CP_TIME)) {
 431                 if (test_opt(sbi, DATA_FLUSH)) {
 432                         struct blk_plug plug;
 433
 434                         blk_start_plug(&plug);
 435                         sync_dirty_inodes(sbi, FILE_INODE);
 436                         blk_finish_plug(&plug);
 437                 }
 438                 f2fs_sync_fs(sbi->sb, true);
 439                 stat_inc_bg_cp_count(sbi->stat_info);
 440         }
 441 }
 442
 443 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 444                                 struct block_device *bdev)
 445 {
 446         struct bio *bio = f2fs_bio_alloc(0);
 447         int ret;
 448
 449         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
 450         bio->bi_bdev = bdev;
 451         ret = submit_bio_wait(bio);
 452         bio_put(bio);
 453
 454         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 455                                 test_opt(sbi, FLUSH_MERGE), ret);
 456         return ret;
 457 }
 458
 459 static int submit_flush_wait(struct f2fs_sb_info *sbi)
 460 {
 461         int ret = __submit_flush_wait(sbi, sbi->sb->s_bdev);
 462         int i;
 463
 464         if (!sbi->s_ndevs || ret)
 465                 return ret;
 466
 467         for (i = 1; i < sbi->s_ndevs; i++) {
 468                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 469                 if (ret)
 470                         break;
 471         }
 472         return ret;
 473 }
 474
 475 static int issue_flush_thread(void *data)
 476 {
 477         struct f2fs_sb_info *sbi = data;
 478         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 479         wait_queue_head_t *q = &fcc->flush_wait_queue;
 480 repeat:
 481         if (kthread_should_stop())
 482                 return 0;
 483
 484         if (!llist_empty(&fcc->issue_list)) {
 485                 struct flush_cmd *cmd, *next;
 486                 int ret;
 487
 488                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 489                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 490
 491                 ret = submit_flush_wait(sbi);
 492                 atomic_inc(&fcc->issued_flush);
 493
 494                 llist_for_each_entry_safe(cmd, next,
 495                                           fcc->dispatch_list, llnode) {
 496                         cmd->ret = ret;
 497                         complete(&cmd->wait);
 498                 }
 499                 fcc->dispatch_list = NULL;
 500         }
 501
 502         wait_event_interruptible(*q,
 503                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 504         goto repeat;
 505 }
 506
 507 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
 508 {
 509         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 510         struct flush_cmd cmd;
 511         int ret;
 512
 513         if (test_opt(sbi, NOBARRIER))
 514                 return 0;
 515
 516         if (!test_opt(sbi, FLUSH_MERGE)) {
 517                 ret = submit_flush_wait(sbi);
 518                 atomic_inc(&fcc->issued_flush);
 519                 return ret;
 520         }
 521
 522         if (!atomic_read(&fcc->issing_flush)) {
 523                 atomic_inc(&fcc->issing_flush);
 524                 ret = submit_flush_wait(sbi);
 525                 atomic_dec(&fcc->issing_flush);
 526
 527                 atomic_inc(&fcc->issued_flush);
 528                 return ret;
 529         }
 530
 531         init_completion(&cmd.wait);
 532
 533         atomic_inc(&fcc->issing_flush);
 534         llist_add(&cmd.llnode, &fcc->issue_list);
 535
 536         if (!fcc->dispatch_list)
 537                 wake_up(&fcc->flush_wait_queue);
 538
 539         if (fcc->f2fs_issue_flush) {
 540                 wait_for_completion(&cmd.wait);
 541                 atomic_dec(&fcc->issing_flush);
 542         } else {
 543                 llist_del_all(&fcc->issue_list);
 544                 atomic_set(&fcc->issing_flush, 0);
 545         }
 546
 547         return cmd.ret;
 548 }
 549
 550 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
 551 {
 552         dev_t dev = sbi->sb->s_bdev->bd_dev;
 553         struct flush_cmd_control *fcc;
 554         int err = 0;
 555
 556         if (SM_I(sbi)->fcc_info) {
 557                 fcc = SM_I(sbi)->fcc_info;
 558                 if (fcc->f2fs_issue_flush)
 559                         return err;
 560                 goto init_thread;
 561         }
 562
 563         fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
 564         if (!fcc)
 565                 return -ENOMEM;
 566         atomic_set(&fcc->issued_flush, 0);
 567         atomic_set(&fcc->issing_flush, 0);
 568         init_waitqueue_head(&fcc->flush_wait_queue);
 569         init_llist_head(&fcc->issue_list);
 570         SM_I(sbi)->fcc_info = fcc;
 571         if (!test_opt(sbi, FLUSH_MERGE))
 572                 return err;
 573
 574 init_thread:
 575         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 576                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 577         if (IS_ERR(fcc->f2fs_issue_flush)) {
 578                 err = PTR_ERR(fcc->f2fs_issue_flush);
 579                 kfree(fcc);
 580                 SM_I(sbi)->fcc_info = NULL;
 581                 return err;
 582         }
 583
 584         return err;
 585 }
 586
 587 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 588 {
 589         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 590
 591         if (fcc && fcc->f2fs_issue_flush) {
 592                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 593
 594                 fcc->f2fs_issue_flush = NULL;
 595                 kthread_stop(flush_thread);
 596         }
 597         if (free) {
 598                 kfree(fcc);
 599                 SM_I(sbi)->fcc_info = NULL;
 600         }
 601 }
 602
 603 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 604                 enum dirty_type dirty_type)
 605 {
 606         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 607
 608         /* need not be added */
 609         if (IS_CURSEG(sbi, segno))
 610                 return;
 611
 612         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 613                 dirty_i->nr_dirty[dirty_type]++;
 614
 615         if (dirty_type == DIRTY) {
 616                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 617                 enum dirty_type t = sentry->type;
 618
 619                 if (unlikely(t >= DIRTY)) {
 620                         f2fs_bug_on(sbi, 1);
 621                         return;
 622                 }
 623                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 624                         dirty_i->nr_dirty[t]++;
 625         }
 626 }
 627
 628 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 629                 enum dirty_type dirty_type)
 630 {
 631         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 632
 633         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 634                 dirty_i->nr_dirty[dirty_type]--;
 635
 636         if (dirty_type == DIRTY) {
 637                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 638                 enum dirty_type t = sentry->type;
 639
 640                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 641                         dirty_i->nr_dirty[t]--;
 642
 643                 if (get_valid_blocks(sbi, segno, true) == 0)
 644                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 645                                                 dirty_i->victim_secmap);
 646         }
 647 }
 648
 649 /*
 650  * Should not occur error such as -ENOMEM.
 651  * Adding dirty entry into seglist is not critical operation.
 652  * If a given segment is one of current working segments, it won't be added.
 653  */
 654 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 655 {
 656         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 657         unsigned short valid_blocks;
 658
 659         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 660                 return;
 661
 662         mutex_lock(&dirty_i->seglist_lock);
 663
 664         valid_blocks = get_valid_blocks(sbi, segno, false);
 665
 666         if (valid_blocks == 0) {
 667                 __locate_dirty_segment(sbi, segno, PRE);
 668                 __remove_dirty_segment(sbi, segno, DIRTY);
 669         } else if (valid_blocks < sbi->blocks_per_seg) {
 670                 __locate_dirty_segment(sbi, segno, DIRTY);
 671         } else {
 672                 /* Recovery routine with SSR needs this */
 673                 __remove_dirty_segment(sbi, segno, DIRTY);
 674         }
 675
 676         mutex_unlock(&dirty_i->seglist_lock);
 677 }
 678
 679 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 680                 struct block_device *bdev, block_t lstart,
 681                 block_t start, block_t len)
 682 {
 683         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 684         struct list_head *pend_list;
 685         struct discard_cmd *dc;
 686
 687         f2fs_bug_on(sbi, !len);
 688
 689         pend_list = &dcc->pend_list[plist_idx(len)];
 690
 691         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 692         INIT_LIST_HEAD(&dc->list);
 693         dc->bdev = bdev;
 694         dc->lstart = lstart;
 695         dc->start = start;
 696         dc->len = len;
 697         dc->ref = 0;
 698         dc->state = D_PREP;
 699         dc->error = 0;
 700         init_completion(&dc->wait);
 701         list_add_tail(&dc->list, pend_list);
 702         atomic_inc(&dcc->discard_cmd_cnt);
 703         dcc->undiscard_blks += len;
 704
 705         return dc;
 706 }
 707
 708 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 709                                 struct block_device *bdev, block_t lstart,
 710                                 block_t start, block_t len,
 711                                 struct rb_node *parent, struct rb_node **p)
 712 {
 713         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 714         struct discard_cmd *dc;
 715
 716         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
 717
 718         rb_link_node(&dc->rb_node, parent, p);
 719         rb_insert_color(&dc->rb_node, &dcc->root);
 720
 721         return dc;
 722 }
 723
 724 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
 725                                                         struct discard_cmd *dc)
 726 {
 727         if (dc->state == D_DONE)
 728                 atomic_dec(&dcc->issing_discard);
 729
 730         list_del(&dc->list);
 731         rb_erase(&dc->rb_node, &dcc->root);
 732         dcc->undiscard_blks -= dc->len;
 733
 734         kmem_cache_free(discard_cmd_slab, dc);
 735
 736         atomic_dec(&dcc->discard_cmd_cnt);
 737 }
 738
 739 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
 740                                                         struct discard_cmd *dc)
 741 {
 742         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 743
 744         f2fs_bug_on(sbi, dc->ref);
 745
 746         if (dc->error == -EOPNOTSUPP)
 747                 dc->error = 0;
 748
 749         if (dc->error)
 750                 f2fs_msg(sbi->sb, KERN_INFO,
 751                         "Issue discard(%u, %u, %u) failed, ret: %d",
 752                         dc->lstart, dc->start, dc->len, dc->error);
 753         __detach_discard_cmd(dcc, dc);
 754 }
 755
 756 static void f2fs_submit_discard_endio(struct bio *bio)
 757 {
 758         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
 759
 760         dc->error = blk_status_to_errno(bio->bi_status);
 761         dc->state = D_DONE;
 762         complete_all(&dc->wait);
 763         bio_put(bio);
 764 }
 765
 766 void __check_sit_bitmap(struct f2fs_sb_info *sbi,
 767                                 block_t start, block_t end)
 768 {
 769 #ifdef CONFIG_F2FS_CHECK_FS
 770         struct seg_entry *sentry;
 771         unsigned int segno;
 772         block_t blk = start;
 773         unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
 774         unsigned long *map;
 775
 776         while (blk < end) {
 777                 segno = GET_SEGNO(sbi, blk);
 778                 sentry = get_seg_entry(sbi, segno);
 779                 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
 780
 781                 size = min((unsigned long)(end - blk), max_blocks);
 782                 map = (unsigned long *)(sentry->cur_valid_map);
 783                 offset = __find_rev_next_bit(map, size, offset);
 784                 f2fs_bug_on(sbi, offset != size);
 785                 blk += size;
 786         }
 787 #endif
 788 }
 789
 790 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
 791 static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
 792                                 struct discard_cmd *dc)
 793 {
 794         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 795         struct bio *bio = NULL;
 796
 797         if (dc->state != D_PREP)
 798                 return;
 799
 800         trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
 801
 802         dc->error = __blkdev_issue_discard(dc->bdev,
 803                                 SECTOR_FROM_BLOCK(dc->start),
 804                                 SECTOR_FROM_BLOCK(dc->len),
 805                                 GFP_NOFS, 0, &bio);
 806         if (!dc->error) {
 807                 /* should keep before submission to avoid D_DONE right away */
 808                 dc->state = D_SUBMIT;
 809                 atomic_inc(&dcc->issued_discard);
 810                 atomic_inc(&dcc->issing_discard);
 811                 if (bio) {
 812                         bio->bi_private = dc;
 813                         bio->bi_end_io = f2fs_submit_discard_endio;
 814                         bio->bi_opf |= REQ_SYNC;
 815                         submit_bio(bio);
 816                         list_move_tail(&dc->list, &dcc->wait_list);
 817                         __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
 818                 }
 819         } else {
 820                 __remove_discard_cmd(sbi, dc);
 821         }
 822 }
 823
 824 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
 825                                 struct block_device *bdev, block_t lstart,
 826                                 block_t start, block_t len,
 827                                 struct rb_node **insert_p,
 828                                 struct rb_node *insert_parent)
 829 {
 830         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 831         struct rb_node **p = &dcc->root.rb_node;
 832         struct rb_node *parent = NULL;
 833         struct discard_cmd *dc = NULL;
 834
 835         if (insert_p && insert_parent) {
 836                 parent = insert_parent;
 837                 p = insert_p;
 838                 goto do_insert;
 839         }
 840
 841         p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
 842 do_insert:
 843         dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
 844         if (!dc)
 845                 return NULL;
 846
 847         return dc;
 848 }
 849
 850 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
 851                                                 struct discard_cmd *dc)
 852 {
 853         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
 854 }
 855
 856 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
 857                                 struct discard_cmd *dc, block_t blkaddr)
 858 {
 859         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 860         struct discard_info di = dc->di;
 861         bool modified = false;
 862
 863         if (dc->state == D_DONE || dc->len == 1) {
 864                 __remove_discard_cmd(sbi, dc);
 865                 return;
 866         }
 867
 868         dcc->undiscard_blks -= di.len;
 869
 870         if (blkaddr > di.lstart) {
 871                 dc->len = blkaddr - dc->lstart;
 872                 dcc->undiscard_blks += dc->len;
 873                 __relocate_discard_cmd(dcc, dc);
 874                 modified = true;
 875         }
 876
 877         if (blkaddr < di.lstart + di.len - 1) {
 878                 if (modified) {
 879                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
 880                                         di.start + blkaddr + 1 - di.lstart,
 881                                         di.lstart + di.len - 1 - blkaddr,
 882                                         NULL, NULL);
 883                 } else {
 884                         dc->lstart++;
 885                         dc->len--;
 886                         dc->start++;
 887                         dcc->undiscard_blks += dc->len;
 888                         __relocate_discard_cmd(dcc, dc);
 889                 }
 890         }
 891 }
 892
 893 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
 894                                 struct block_device *bdev, block_t lstart,
 895                                 block_t start, block_t len)
 896 {
 897         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 898         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
 899         struct discard_cmd *dc;
 900         struct discard_info di = {0};
 901         struct rb_node **insert_p = NULL, *insert_parent = NULL;
 902         block_t end = lstart + len;
 903
 904         mutex_lock(&dcc->cmd_lock);
 905
 906         dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
 907                                         NULL, lstart,
 908                                         (struct rb_entry **)&prev_dc,
 909                                         (struct rb_entry **)&next_dc,
 910                                         &insert_p, &insert_parent, true);
 911         if (dc)
 912                 prev_dc = dc;
 913
 914         if (!prev_dc) {
 915                 di.lstart = lstart;
 916                 di.len = next_dc ? next_dc->lstart - lstart : len;
 917                 di.len = min(di.len, len);
 918                 di.start = start;
 919         }
 920
 921         while (1) {
 922                 struct rb_node *node;
 923                 bool merged = false;
 924                 struct discard_cmd *tdc = NULL;
 925
 926                 if (prev_dc) {
 927                         di.lstart = prev_dc->lstart + prev_dc->len;
 928                         if (di.lstart < lstart)
 929                                 di.lstart = lstart;
 930                         if (di.lstart >= end)
 931                                 break;
 932
 933                         if (!next_dc || next_dc->lstart > end)
 934                                 di.len = end - di.lstart;
 935                         else
 936                                 di.len = next_dc->lstart - di.lstart;
 937                         di.start = start + di.lstart - lstart;
 938                 }
 939
 940                 if (!di.len)
 941                         goto next;
 942
 943                 if (prev_dc && prev_dc->state == D_PREP &&
 944                         prev_dc->bdev == bdev &&
 945                         __is_discard_back_mergeable(&di, &prev_dc->di)) {
 946                         prev_dc->di.len += di.len;
 947                         dcc->undiscard_blks += di.len;
 948                         __relocate_discard_cmd(dcc, prev_dc);
 949                         di = prev_dc->di;
 950                         tdc = prev_dc;
 951                         merged = true;
 952                 }
 953
 954                 if (next_dc && next_dc->state == D_PREP &&
 955                         next_dc->bdev == bdev &&
 956                         __is_discard_front_mergeable(&di, &next_dc->di)) {
 957                         next_dc->di.lstart = di.lstart;
 958                         next_dc->di.len += di.len;
 959                         next_dc->di.start = di.start;
 960                         dcc->undiscard_blks += di.len;
 961                         __relocate_discard_cmd(dcc, next_dc);
 962                         if (tdc)
 963                                 __remove_discard_cmd(sbi, tdc);
 964                         merged = true;
 965                 }
 966
 967                 if (!merged) {
 968                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
 969                                                         di.len, NULL, NULL);
 970                 }
 971  next:
 972                 prev_dc = next_dc;
 973                 if (!prev_dc)
 974                         break;
 975
 976                 node = rb_next(&prev_dc->rb_node);
 977                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
 978         }
 979
 980         mutex_unlock(&dcc->cmd_lock);
 981 }
 982
 983 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
 984                 struct block_device *bdev, block_t blkstart, block_t blklen)
 985 {
 986         block_t lblkstart = blkstart;
 987
 988         trace_f2fs_queue_discard(bdev, blkstart, blklen);
 989
 990         if (sbi->s_ndevs) {
 991                 int devi = f2fs_target_device_index(sbi, blkstart);
 992
 993                 blkstart -= FDEV(devi).start_blk;
 994         }
 995         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
 996         return 0;
 997 }
 998
 999 static void __issue_discard_cmd(struct f2fs_sb_info *sbi, bool issue_cond)
1000 {
1001         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1002         struct list_head *pend_list;
1003         struct discard_cmd *dc, *tmp;
1004         struct blk_plug plug;
1005         int i, iter = 0;
1006
1007         mutex_lock(&dcc->cmd_lock);
1008         f2fs_bug_on(sbi,
1009                 !__check_rb_tree_consistence(sbi, &dcc->root));
1010         blk_start_plug(&plug);
1011         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1012                 pend_list = &dcc->pend_list[i];
1013                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1014                         f2fs_bug_on(sbi, dc->state != D_PREP);
1015
1016                         if (!issue_cond || is_idle(sbi))
1017                                 __submit_discard_cmd(sbi, dc);
1018                         if (issue_cond && iter++ > DISCARD_ISSUE_RATE)
1019                                 goto out;
1020                 }
1021         }
1022 out:
1023         blk_finish_plug(&plug);
1024         mutex_unlock(&dcc->cmd_lock);
1025 }
1026
1027 static void __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1028                                                         struct discard_cmd *dc)
1029 {
1030         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1031
1032         wait_for_completion_io(&dc->wait);
1033         mutex_lock(&dcc->cmd_lock);
1034         f2fs_bug_on(sbi, dc->state != D_DONE);
1035         dc->ref--;
1036         if (!dc->ref)
1037                 __remove_discard_cmd(sbi, dc);
1038         mutex_unlock(&dcc->cmd_lock);
1039 }
1040
1041 static void __wait_discard_cmd(struct f2fs_sb_info *sbi, bool wait_cond)
1042 {
1043         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1044         struct list_head *wait_list = &(dcc->wait_list);
1045         struct discard_cmd *dc, *tmp;
1046         bool need_wait;
1047
1048 next:
1049         need_wait = false;
1050
1051         mutex_lock(&dcc->cmd_lock);
1052         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1053                 if (!wait_cond || (dc->state == D_DONE && !dc->ref)) {
1054                         wait_for_completion_io(&dc->wait);
1055                         __remove_discard_cmd(sbi, dc);
1056                 } else {
1057                         dc->ref++;
1058                         need_wait = true;
1059                         break;
1060                 }
1061         }
1062         mutex_unlock(&dcc->cmd_lock);
1063
1064         if (need_wait) {
1065                 __wait_one_discard_bio(sbi, dc);
1066                 goto next;
1067         }
1068 }
1069
1070 /* This should be covered by global mutex, &sit_i->sentry_lock */
1071 void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1072 {
1073         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1074         struct discard_cmd *dc;
1075         bool need_wait = false;
1076
1077         mutex_lock(&dcc->cmd_lock);
1078         dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1079         if (dc) {
1080                 if (dc->state == D_PREP) {
1081                         __punch_discard_cmd(sbi, dc, blkaddr);
1082                 } else {
1083                         dc->ref++;
1084                         need_wait = true;
1085                 }
1086         }
1087         mutex_unlock(&dcc->cmd_lock);
1088
1089         if (need_wait)
1090                 __wait_one_discard_bio(sbi, dc);
1091 }
1092
1093 void stop_discard_thread(struct f2fs_sb_info *sbi)
1094 {
1095         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1096
1097         if (dcc && dcc->f2fs_issue_discard) {
1098                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1099
1100                 dcc->f2fs_issue_discard = NULL;
1101                 kthread_stop(discard_thread);
1102         }
1103 }
1104
1105 /* This comes from f2fs_put_super */
1106 void f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1107 {
1108         __issue_discard_cmd(sbi, false);
1109         __wait_discard_cmd(sbi, false);
1110 }
1111
1112 static int issue_discard_thread(void *data)
1113 {
1114         struct f2fs_sb_info *sbi = data;
1115         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1116         wait_queue_head_t *q = &dcc->discard_wait_queue;
1117
1118         set_freezable();
1119
1120         do {
1121                 wait_event_interruptible(*q, kthread_should_stop() ||
1122                                         freezing(current) ||
1123                                         atomic_read(&dcc->discard_cmd_cnt));
1124                 if (try_to_freeze())
1125                         continue;
1126                 if (kthread_should_stop())
1127                         return 0;
1128
1129                 __issue_discard_cmd(sbi, true);
1130                 __wait_discard_cmd(sbi, true);
1131
1132                 congestion_wait(BLK_RW_SYNC, HZ/50);
1133         } while (!kthread_should_stop());
1134         return 0;
1135 }
1136
1137 #ifdef CONFIG_BLK_DEV_ZONED
1138 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1139                 struct block_device *bdev, block_t blkstart, block_t blklen)
1140 {
1141         sector_t sector, nr_sects;
1142         block_t lblkstart = blkstart;
1143         int devi = 0;
1144
1145         if (sbi->s_ndevs) {
1146                 devi = f2fs_target_device_index(sbi, blkstart);
1147                 blkstart -= FDEV(devi).start_blk;
1148         }
1149
1150         /*
1151          * We need to know the type of the zone: for conventional zones,
1152          * use regular discard if the drive supports it. For sequential
1153          * zones, reset the zone write pointer.
1154          */
1155         switch (get_blkz_type(sbi, bdev, blkstart)) {
1156
1157         case BLK_ZONE_TYPE_CONVENTIONAL:
1158                 if (!blk_queue_discard(bdev_get_queue(bdev)))
1159                         return 0;
1160                 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1161         case BLK_ZONE_TYPE_SEQWRITE_REQ:
1162         case BLK_ZONE_TYPE_SEQWRITE_PREF:
1163                 sector = SECTOR_FROM_BLOCK(blkstart);
1164                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1165
1166                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1167                                 nr_sects != bdev_zone_sectors(bdev)) {
1168                         f2fs_msg(sbi->sb, KERN_INFO,
1169                                 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1170                                 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1171                                 blkstart, blklen);
1172                         return -EIO;
1173                 }
1174                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1175                 return blkdev_reset_zones(bdev, sector,
1176                                           nr_sects, GFP_NOFS);
1177         default:
1178                 /* Unknown zone type: broken device ? */
1179                 return -EIO;
1180         }
1181 }
1182 #endif
1183
1184 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1185                 struct block_device *bdev, block_t blkstart, block_t blklen)
1186 {
1187 #ifdef CONFIG_BLK_DEV_ZONED
1188         if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
1189                                 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1190                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1191 #endif
1192         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1193 }
1194
1195 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1196                                 block_t blkstart, block_t blklen)
1197 {
1198         sector_t start = blkstart, len = 0;
1199         struct block_device *bdev;
1200         struct seg_entry *se;
1201         unsigned int offset;
1202         block_t i;
1203         int err = 0;
1204
1205         bdev = f2fs_target_device(sbi, blkstart, NULL);
1206
1207         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1208                 if (i != start) {
1209                         struct block_device *bdev2 =
1210                                 f2fs_target_device(sbi, i, NULL);
1211
1212                         if (bdev2 != bdev) {
1213                                 err = __issue_discard_async(sbi, bdev,
1214                                                 start, len);
1215                                 if (err)
1216                                         return err;
1217                                 bdev = bdev2;
1218                                 start = i;
1219                                 len = 0;
1220                         }
1221                 }
1222
1223                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1224                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1225
1226                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1227                         sbi->discard_blks--;
1228         }
1229
1230         if (len)
1231                 err = __issue_discard_async(sbi, bdev, start, len);
1232         return err;
1233 }
1234
1235 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1236                                                         bool check_only)
1237 {
1238         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1239         int max_blocks = sbi->blocks_per_seg;
1240         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1241         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1242         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1243         unsigned long *discard_map = (unsigned long *)se->discard_map;
1244         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1245         unsigned int start = 0, end = -1;
1246         bool force = (cpc->reason & CP_DISCARD);
1247         struct discard_entry *de = NULL;
1248         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1249         int i;
1250
1251         if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
1252                 return false;
1253
1254         if (!force) {
1255                 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
1256                         SM_I(sbi)->dcc_info->nr_discards >=
1257                                 SM_I(sbi)->dcc_info->max_discards)
1258                         return false;
1259         }
1260
1261         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1262         for (i = 0; i < entries; i++)
1263                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1264                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1265
1266         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1267                                 SM_I(sbi)->dcc_info->max_discards) {
1268                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1269                 if (start >= max_blocks)
1270                         break;
1271
1272                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1273                 if (force && start && end != max_blocks
1274                                         && (end - start) < cpc->trim_minlen)
1275                         continue;
1276
1277                 if (check_only)
1278                         return true;
1279
1280                 if (!de) {
1281                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1282                                                                 GFP_F2FS_ZERO);
1283                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1284                         list_add_tail(&de->list, head);
1285                 }
1286
1287                 for (i = start; i < end; i++)
1288                         __set_bit_le(i, (void *)de->discard_map);
1289
1290                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1291         }
1292         return false;
1293 }
1294
1295 void release_discard_addrs(struct f2fs_sb_info *sbi)
1296 {
1297         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1298         struct discard_entry *entry, *this;
1299
1300         /* drop caches */
1301         list_for_each_entry_safe(entry, this, head, list) {
1302                 list_del(&entry->list);
1303                 kmem_cache_free(discard_entry_slab, entry);
1304         }
1305 }
1306
1307 /*
1308  * Should call clear_prefree_segments after checkpoint is done.
1309  */
1310 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1311 {
1312         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1313         unsigned int segno;
1314
1315         mutex_lock(&dirty_i->seglist_lock);
1316         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1317                 __set_test_and_free(sbi, segno);
1318         mutex_unlock(&dirty_i->seglist_lock);
1319 }
1320
1321 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1322 {
1323         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1324         struct discard_entry *entry, *this;
1325         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1326         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1327         unsigned int start = 0, end = -1;
1328         unsigned int secno, start_segno;
1329         bool force = (cpc->reason & CP_DISCARD);
1330
1331         mutex_lock(&dirty_i->seglist_lock);
1332
1333         while (1) {
1334                 int i;
1335                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1336                 if (start >= MAIN_SEGS(sbi))
1337                         break;
1338                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1339                                                                 start + 1);
1340
1341                 for (i = start; i < end; i++)
1342                         clear_bit(i, prefree_map);
1343
1344                 dirty_i->nr_dirty[PRE] -= end - start;
1345
1346                 if (!test_opt(sbi, DISCARD))
1347                         continue;
1348
1349                 if (force && start >= cpc->trim_start &&
1350                                         (end - 1) <= cpc->trim_end)
1351                                 continue;
1352
1353                 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1354                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1355                                 (end - start) << sbi->log_blocks_per_seg);
1356                         continue;
1357                 }
1358 next:
1359                 secno = GET_SEC_FROM_SEG(sbi, start);
1360                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1361                 if (!IS_CURSEC(sbi, secno) &&
1362                         !get_valid_blocks(sbi, start, true))
1363                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1364                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1365
1366                 start = start_segno + sbi->segs_per_sec;
1367                 if (start < end)
1368                         goto next;
1369                 else
1370                         end = start - 1;
1371         }
1372         mutex_unlock(&dirty_i->seglist_lock);
1373
1374         /* send small discards */
1375         list_for_each_entry_safe(entry, this, head, list) {
1376                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1377                 bool is_valid = test_bit_le(0, entry->discard_map);
1378
1379 find_next:
1380                 if (is_valid) {
1381                         next_pos = find_next_zero_bit_le(entry->discard_map,
1382                                         sbi->blocks_per_seg, cur_pos);
1383                         len = next_pos - cur_pos;
1384
1385                         if (f2fs_sb_mounted_blkzoned(sbi->sb) ||
1386                             (force && len < cpc->trim_minlen))
1387                                 goto skip;
1388
1389                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1390                                                                         len);
1391                         cpc->trimmed += len;
1392                         total_len += len;
1393                 } else {
1394                         next_pos = find_next_bit_le(entry->discard_map,
1395                                         sbi->blocks_per_seg, cur_pos);
1396                 }
1397 skip:
1398                 cur_pos = next_pos;
1399                 is_valid = !is_valid;
1400
1401                 if (cur_pos < sbi->blocks_per_seg)
1402                         goto find_next;
1403
1404                 list_del(&entry->list);
1405                 SM_I(sbi)->dcc_info->nr_discards -= total_len;
1406                 kmem_cache_free(discard_entry_slab, entry);
1407         }
1408
1409         wake_up(&SM_I(sbi)->dcc_info->discard_wait_queue);
1410 }
1411
1412 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1413 {
1414         dev_t dev = sbi->sb->s_bdev->bd_dev;
1415         struct discard_cmd_control *dcc;
1416         int err = 0, i;
1417
1418         if (SM_I(sbi)->dcc_info) {
1419                 dcc = SM_I(sbi)->dcc_info;
1420                 goto init_thread;
1421         }
1422
1423         dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1424         if (!dcc)
1425                 return -ENOMEM;
1426
1427         INIT_LIST_HEAD(&dcc->entry_list);
1428         for (i = 0; i < MAX_PLIST_NUM; i++)
1429                 INIT_LIST_HEAD(&dcc->pend_list[i]);
1430         INIT_LIST_HEAD(&dcc->wait_list);
1431         mutex_init(&dcc->cmd_lock);
1432         atomic_set(&dcc->issued_discard, 0);
1433         atomic_set(&dcc->issing_discard, 0);
1434         atomic_set(&dcc->discard_cmd_cnt, 0);
1435         dcc->nr_discards = 0;
1436         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1437         dcc->undiscard_blks = 0;
1438         dcc->root = RB_ROOT;
1439
1440         init_waitqueue_head(&dcc->discard_wait_queue);
1441         SM_I(sbi)->dcc_info = dcc;
1442 init_thread:
1443         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1444                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1445         if (IS_ERR(dcc->f2fs_issue_discard)) {
1446                 err = PTR_ERR(dcc->f2fs_issue_discard);
1447                 kfree(dcc);
1448                 SM_I(sbi)->dcc_info = NULL;
1449                 return err;
1450         }
1451
1452         return err;
1453 }
1454
1455 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1456 {
1457         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1458
1459         if (!dcc)
1460                 return;
1461
1462         stop_discard_thread(sbi);
1463
1464         kfree(dcc);
1465         SM_I(sbi)->dcc_info = NULL;
1466 }
1467
1468 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1469 {
1470         struct sit_info *sit_i = SIT_I(sbi);
1471
1472         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1473                 sit_i->dirty_sentries++;
1474                 return false;
1475         }
1476
1477         return true;
1478 }
1479
1480 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1481                                         unsigned int segno, int modified)
1482 {
1483         struct seg_entry *se = get_seg_entry(sbi, segno);
1484         se->type = type;
1485         if (modified)
1486                 __mark_sit_entry_dirty(sbi, segno);
1487 }
1488
1489 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1490 {
1491         struct seg_entry *se;
1492         unsigned int segno, offset;
1493         long int new_vblocks;
1494
1495         segno = GET_SEGNO(sbi, blkaddr);
1496
1497         se = get_seg_entry(sbi, segno);
1498         new_vblocks = se->valid_blocks + del;
1499         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1500
1501         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
1502                                 (new_vblocks > sbi->blocks_per_seg)));
1503
1504         se->valid_blocks = new_vblocks;
1505         se->mtime = get_mtime(sbi);
1506         SIT_I(sbi)->max_mtime = se->mtime;
1507
1508         /* Update valid block bitmap */
1509         if (del > 0) {
1510                 if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) {
1511 #ifdef CONFIG_F2FS_CHECK_FS
1512                         if (f2fs_test_and_set_bit(offset,
1513                                                 se->cur_valid_map_mir))
1514                                 f2fs_bug_on(sbi, 1);
1515                         else
1516                                 WARN_ON(1);
1517 #else
1518                         f2fs_bug_on(sbi, 1);
1519 #endif
1520                 }
1521                 if (f2fs_discard_en(sbi) &&
1522                         !f2fs_test_and_set_bit(offset, se->discard_map))
1523                         sbi->discard_blks--;
1524
1525                 /* don't overwrite by SSR to keep node chain */
1526                 if (se->type == CURSEG_WARM_NODE) {
1527                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1528                                 se->ckpt_valid_blocks++;
1529                 }
1530         } else {
1531                 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) {
1532 #ifdef CONFIG_F2FS_CHECK_FS
1533                         if (!f2fs_test_and_clear_bit(offset,
1534                                                 se->cur_valid_map_mir))
1535                                 f2fs_bug_on(sbi, 1);
1536                         else
1537                                 WARN_ON(1);
1538 #else
1539                         f2fs_bug_on(sbi, 1);
1540 #endif
1541                 }
1542                 if (f2fs_discard_en(sbi) &&
1543                         f2fs_test_and_clear_bit(offset, se->discard_map))
1544                         sbi->discard_blks++;
1545         }
1546         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1547                 se->ckpt_valid_blocks += del;
1548
1549         __mark_sit_entry_dirty(sbi, segno);
1550
1551         /* update total number of valid blocks to be written in ckpt area */
1552         SIT_I(sbi)->written_valid_blocks += del;
1553
1554         if (sbi->segs_per_sec > 1)
1555                 get_sec_entry(sbi, segno)->valid_blocks += del;
1556 }
1557
1558 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
1559 {
1560         update_sit_entry(sbi, new, 1);
1561         if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1562                 update_sit_entry(sbi, old, -1);
1563
1564         locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1565         locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
1566 }
1567
1568 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1569 {
1570         unsigned int segno = GET_SEGNO(sbi, addr);
1571         struct sit_info *sit_i = SIT_I(sbi);
1572
1573         f2fs_bug_on(sbi, addr == NULL_ADDR);
1574         if (addr == NEW_ADDR)
1575                 return;
1576
1577         /* add it into sit main buffer */
1578         mutex_lock(&sit_i->sentry_lock);
1579
1580         update_sit_entry(sbi, addr, -1);
1581
1582         /* add it into dirty seglist */
1583         locate_dirty_segment(sbi, segno);
1584
1585         mutex_unlock(&sit_i->sentry_lock);
1586 }
1587
1588 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1589 {
1590         struct sit_info *sit_i = SIT_I(sbi);
1591         unsigned int segno, offset;
1592         struct seg_entry *se;
1593         bool is_cp = false;
1594
1595         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1596                 return true;
1597
1598         mutex_lock(&sit_i->sentry_lock);
1599
1600         segno = GET_SEGNO(sbi, blkaddr);
1601         se = get_seg_entry(sbi, segno);
1602         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1603
1604         if (f2fs_test_bit(offset, se->ckpt_valid_map))
1605                 is_cp = true;
1606
1607         mutex_unlock(&sit_i->sentry_lock);
1608
1609         return is_cp;
1610 }
1611
1612 /*
1613  * This function should be resided under the curseg_mutex lock
1614  */
1615 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1616                                         struct f2fs_summary *sum)
1617 {
1618         struct curseg_info *curseg = CURSEG_I(sbi, type);
1619         void *addr = curseg->sum_blk;
1620         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1621         memcpy(addr, sum, sizeof(struct f2fs_summary));
1622 }
1623
1624 /*
1625  * Calculate the number of current summary pages for writing
1626  */
1627 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1628 {
1629         int valid_sum_count = 0;
1630         int i, sum_in_page;
1631
1632         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1633                 if (sbi->ckpt->alloc_type[i] == SSR)
1634                         valid_sum_count += sbi->blocks_per_seg;
1635                 else {
1636                         if (for_ra)
1637                                 valid_sum_count += le16_to_cpu(
1638                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1639                         else
1640                                 valid_sum_count += curseg_blkoff(sbi, i);
1641                 }
1642         }
1643
1644         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1645                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1646         if (valid_sum_count <= sum_in_page)
1647                 return 1;
1648         else if ((valid_sum_count - sum_in_page) <=
1649                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1650                 return 2;
1651         return 3;
1652 }
1653
1654 /*
1655  * Caller should put this summary page
1656  */
1657 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1658 {
1659         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1660 }
1661
1662 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1663 {
1664         struct page *page = grab_meta_page(sbi, blk_addr);
1665         void *dst = page_address(page);
1666
1667         if (src)
1668                 memcpy(dst, src, PAGE_SIZE);
1669         else
1670                 memset(dst, 0, PAGE_SIZE);
1671         set_page_dirty(page);
1672         f2fs_put_page(page, 1);
1673 }
1674
1675 static void write_sum_page(struct f2fs_sb_info *sbi,
1676                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
1677 {
1678         update_meta_page(sbi, (void *)sum_blk, blk_addr);
1679 }
1680
1681 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1682                                                 int type, block_t blk_addr)
1683 {
1684         struct curseg_info *curseg = CURSEG_I(sbi, type);
1685         struct page *page = grab_meta_page(sbi, blk_addr);
1686         struct f2fs_summary_block *src = curseg->sum_blk;
1687         struct f2fs_summary_block *dst;
1688
1689         dst = (struct f2fs_summary_block *)page_address(page);
1690
1691         mutex_lock(&curseg->curseg_mutex);
1692
1693         down_read(&curseg->journal_rwsem);
1694         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1695         up_read(&curseg->journal_rwsem);
1696
1697         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1698         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1699
1700         mutex_unlock(&curseg->curseg_mutex);
1701
1702         set_page_dirty(page);
1703         f2fs_put_page(page, 1);
1704 }
1705
1706 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1707 {
1708         struct curseg_info *curseg = CURSEG_I(sbi, type);
1709         unsigned int segno = curseg->segno + 1;
1710         struct free_segmap_info *free_i = FREE_I(sbi);
1711
1712         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
1713                 return !test_bit(segno, free_i->free_segmap);
1714         return 0;
1715 }
1716
1717 /*
1718  * Find a new segment from the free segments bitmap to right order
1719  * This function should be returned with success, otherwise BUG
1720  */
1721 static void get_new_segment(struct f2fs_sb_info *sbi,
1722                         unsigned int *newseg, bool new_sec, int dir)
1723 {
1724         struct free_segmap_info *free_i = FREE_I(sbi);
1725         unsigned int segno, secno, zoneno;
1726         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
1727         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
1728         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
1729         unsigned int left_start = hint;
1730         bool init = true;
1731         int go_left = 0;
1732         int i;
1733
1734         spin_lock(&free_i->segmap_lock);
1735
1736         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1737                 segno = find_next_zero_bit(free_i->free_segmap,
1738                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
1739                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
1740                         goto got_it;
1741         }
1742 find_other_zone:
1743         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1744         if (secno >= MAIN_SECS(sbi)) {
1745                 if (dir == ALLOC_RIGHT) {
1746                         secno = find_next_zero_bit(free_i->free_secmap,
1747                                                         MAIN_SECS(sbi), 0);
1748                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1749                 } else {
1750                         go_left = 1;
1751                         left_start = hint - 1;
1752                 }
1753         }
1754         if (go_left == 0)
1755                 goto skip_left;
1756
1757         while (test_bit(left_start, free_i->free_secmap)) {
1758                 if (left_start > 0) {
1759                         left_start--;
1760                         continue;
1761                 }
1762                 left_start = find_next_zero_bit(free_i->free_secmap,
1763                                                         MAIN_SECS(sbi), 0);
1764                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1765                 break;
1766         }
1767         secno = left_start;
1768 skip_left:
1769         hint = secno;
1770         segno = GET_SEG_FROM_SEC(sbi, secno);
1771         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
1772
1773         /* give up on finding another zone */
1774         if (!init)
1775                 goto got_it;
1776         if (sbi->secs_per_zone == 1)
1777                 goto got_it;
1778         if (zoneno == old_zoneno)
1779                 goto got_it;
1780         if (dir == ALLOC_LEFT) {
1781                 if (!go_left && zoneno + 1 >= total_zones)
1782                         goto got_it;
1783                 if (go_left && zoneno == 0)
1784                         goto got_it;
1785         }
1786         for (i = 0; i < NR_CURSEG_TYPE; i++)
1787                 if (CURSEG_I(sbi, i)->zone == zoneno)
1788                         break;
1789
1790         if (i < NR_CURSEG_TYPE) {
1791                 /* zone is in user, try another */
1792                 if (go_left)
1793                         hint = zoneno * sbi->secs_per_zone - 1;
1794                 else if (zoneno + 1 >= total_zones)
1795                         hint = 0;
1796                 else
1797                         hint = (zoneno + 1) * sbi->secs_per_zone;
1798                 init = false;
1799                 goto find_other_zone;
1800         }
1801 got_it:
1802         /* set it as dirty segment in free segmap */
1803         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1804         __set_inuse(sbi, segno);
1805         *newseg = segno;
1806         spin_unlock(&free_i->segmap_lock);
1807 }
1808
1809 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1810 {
1811         struct curseg_info *curseg = CURSEG_I(sbi, type);
1812         struct summary_footer *sum_footer;
1813
1814         curseg->segno = curseg->next_segno;
1815         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
1816         curseg->next_blkoff = 0;
1817         curseg->next_segno = NULL_SEGNO;
1818
1819         sum_footer = &(curseg->sum_blk->footer);
1820         memset(sum_footer, 0, sizeof(struct summary_footer));
1821         if (IS_DATASEG(type))
1822                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1823         if (IS_NODESEG(type))
1824                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1825         __set_sit_entry_type(sbi, type, curseg->segno, modified);
1826 }
1827
1828 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
1829 {
1830         /* if segs_per_sec is large than 1, we need to keep original policy. */
1831         if (sbi->segs_per_sec != 1)
1832                 return CURSEG_I(sbi, type)->segno;
1833
1834         if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
1835                 return 0;
1836
1837         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
1838                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
1839         return CURSEG_I(sbi, type)->segno;
1840 }
1841
1842 /*
1843  * Allocate a current working segment.
1844  * This function always allocates a free segment in LFS manner.
1845  */
1846 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1847 {
1848         struct curseg_info *curseg = CURSEG_I(sbi, type);
1849         unsigned int segno = curseg->segno;
1850         int dir = ALLOC_LEFT;
1851
1852         write_sum_page(sbi, curseg->sum_blk,
1853                                 GET_SUM_BLOCK(sbi, segno));
1854         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1855                 dir = ALLOC_RIGHT;
1856
1857         if (test_opt(sbi, NOHEAP))
1858                 dir = ALLOC_RIGHT;
1859
1860         segno = __get_next_segno(sbi, type);
1861         get_new_segment(sbi, &segno, new_sec, dir);
1862         curseg->next_segno = segno;
1863         reset_curseg(sbi, type, 1);
1864         curseg->alloc_type = LFS;
1865 }
1866
1867 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1868                         struct curseg_info *seg, block_t start)
1869 {
1870         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1871         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1872         unsigned long *target_map = SIT_I(sbi)->tmp_map;
1873         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1874         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1875         int i, pos;
1876
1877         for (i = 0; i < entries; i++)
1878                 target_map[i] = ckpt_map[i] | cur_map[i];
1879
1880         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1881
1882         seg->next_blkoff = pos;
1883 }
1884
1885 /*
1886  * If a segment is written by LFS manner, next block offset is just obtained
1887  * by increasing the current block offset. However, if a segment is written by
1888  * SSR manner, next block offset obtained by calling __next_free_blkoff
1889  */
1890 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1891                                 struct curseg_info *seg)
1892 {
1893         if (seg->alloc_type == SSR)
1894                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1895         else
1896                 seg->next_blkoff++;
1897 }
1898
1899 /*
1900  * This function always allocates a used segment(from dirty seglist) by SSR
1901  * manner, so it should recover the existing segment information of valid blocks
1902  */
1903 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1904 {
1905         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1906         struct curseg_info *curseg = CURSEG_I(sbi, type);
1907         unsigned int new_segno = curseg->next_segno;
1908         struct f2fs_summary_block *sum_node;
1909         struct page *sum_page;
1910
1911         write_sum_page(sbi, curseg->sum_blk,
1912                                 GET_SUM_BLOCK(sbi, curseg->segno));
1913         __set_test_and_inuse(sbi, new_segno);
1914
1915         mutex_lock(&dirty_i->seglist_lock);
1916         __remove_dirty_segment(sbi, new_segno, PRE);
1917         __remove_dirty_segment(sbi, new_segno, DIRTY);
1918         mutex_unlock(&dirty_i->seglist_lock);
1919
1920         reset_curseg(sbi, type, 1);
1921         curseg->alloc_type = SSR;
1922         __next_free_blkoff(sbi, curseg, 0);
1923
1924         if (reuse) {
1925                 sum_page = get_sum_page(sbi, new_segno);
1926                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1927                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1928                 f2fs_put_page(sum_page, 1);
1929         }
1930 }
1931
1932 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1933 {
1934         struct curseg_info *curseg = CURSEG_I(sbi, type);
1935         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1936         unsigned segno = NULL_SEGNO;
1937         int i, cnt;
1938         bool reversed = false;
1939
1940         /* need_SSR() already forces to do this */
1941         if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
1942                 curseg->next_segno = segno;
1943                 return 1;
1944         }
1945
1946         /* For node segments, let's do SSR more intensively */
1947         if (IS_NODESEG(type)) {
1948                 if (type >= CURSEG_WARM_NODE) {
1949                         reversed = true;
1950                         i = CURSEG_COLD_NODE;
1951                 } else {
1952                         i = CURSEG_HOT_NODE;
1953                 }
1954                 cnt = NR_CURSEG_NODE_TYPE;
1955         } else {
1956                 if (type >= CURSEG_WARM_DATA) {
1957                         reversed = true;
1958                         i = CURSEG_COLD_DATA;
1959                 } else {
1960                         i = CURSEG_HOT_DATA;
1961                 }
1962                 cnt = NR_CURSEG_DATA_TYPE;
1963         }
1964
1965         for (; cnt-- > 0; reversed ? i-- : i++) {
1966                 if (i == type)
1967                         continue;
1968                 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
1969                         curseg->next_segno = segno;
1970                         return 1;
1971                 }
1972         }
1973         return 0;
1974 }
1975
1976 /*
1977  * flush out current segment and replace it with new segment
1978  * This function should be returned with success, otherwise BUG
1979  */
1980 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1981                                                 int type, bool force)
1982 {
1983         struct curseg_info *curseg = CURSEG_I(sbi, type);
1984
1985         if (force)
1986                 new_curseg(sbi, type, true);
1987         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
1988                                         type == CURSEG_WARM_NODE)
1989                 new_curseg(sbi, type, false);
1990         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1991                 new_curseg(sbi, type, false);
1992         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1993                 change_curseg(sbi, type, true);
1994         else
1995                 new_curseg(sbi, type, false);
1996
1997         stat_inc_seg_type(sbi, curseg);
1998 }
1999
2000 void allocate_new_segments(struct f2fs_sb_info *sbi)
2001 {
2002         struct curseg_info *curseg;
2003         unsigned int old_segno;
2004         int i;
2005
2006         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2007                 curseg = CURSEG_I(sbi, i);
2008                 old_segno = curseg->segno;
2009                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2010                 locate_dirty_segment(sbi, old_segno);
2011         }
2012 }
2013
2014 static const struct segment_allocation default_salloc_ops = {
2015         .allocate_segment = allocate_segment_by_default,
2016 };
2017
2018 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2019 {
2020         __u64 trim_start = cpc->trim_start;
2021         bool has_candidate = false;
2022
2023         mutex_lock(&SIT_I(sbi)->sentry_lock);
2024         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2025                 if (add_discard_addrs(sbi, cpc, true)) {
2026                         has_candidate = true;
2027                         break;
2028                 }
2029         }
2030         mutex_unlock(&SIT_I(sbi)->sentry_lock);
2031
2032         cpc->trim_start = trim_start;
2033         return has_candidate;
2034 }
2035
2036 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2037 {
2038         __u64 start = F2FS_BYTES_TO_BLK(range->start);
2039         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2040         unsigned int start_segno, end_segno;
2041         struct cp_control cpc;
2042         int err = 0;
2043
2044         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2045                 return -EINVAL;
2046
2047         cpc.trimmed = 0;
2048         if (end <= MAIN_BLKADDR(sbi))
2049                 goto out;
2050
2051         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2052                 f2fs_msg(sbi->sb, KERN_WARNING,
2053                         "Found FS corruption, run fsck to fix.");
2054                 goto out;
2055         }
2056
2057         /* start/end segment number in main_area */
2058         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2059         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2060                                                 GET_SEGNO(sbi, end);
2061         cpc.reason = CP_DISCARD;
2062         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2063
2064         /* do checkpoint to issue discard commands safely */
2065         for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
2066                 cpc.trim_start = start_segno;
2067
2068                 if (sbi->discard_blks == 0)
2069                         break;
2070                 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
2071                         cpc.trim_end = end_segno;
2072                 else
2073                         cpc.trim_end = min_t(unsigned int,
2074                                 rounddown(start_segno +
2075                                 BATCHED_TRIM_SEGMENTS(sbi),
2076                                 sbi->segs_per_sec) - 1, end_segno);
2077
2078                 mutex_lock(&sbi->gc_mutex);
2079                 err = write_checkpoint(sbi, &cpc);
2080                 mutex_unlock(&sbi->gc_mutex);
2081                 if (err)
2082                         break;
2083
2084                 schedule();
2085         }
2086 out:
2087         range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
2088         return err;
2089 }
2090
2091 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2092 {
2093         struct curseg_info *curseg = CURSEG_I(sbi, type);
2094         if (curseg->next_blkoff < sbi->blocks_per_seg)
2095                 return true;
2096         return false;
2097 }
2098
2099 static int __get_segment_type_2(struct f2fs_io_info *fio)
2100 {
2101         if (fio->type == DATA)
2102                 return CURSEG_HOT_DATA;
2103         else
2104                 return CURSEG_HOT_NODE;
2105 }
2106
2107 static int __get_segment_type_4(struct f2fs_io_info *fio)
2108 {
2109         if (fio->type == DATA) {
2110                 struct inode *inode = fio->page->mapping->host;
2111
2112                 if (S_ISDIR(inode->i_mode))
2113                         return CURSEG_HOT_DATA;
2114                 else
2115                         return CURSEG_COLD_DATA;
2116         } else {
2117                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2118                         return CURSEG_WARM_NODE;
2119                 else
2120                         return CURSEG_COLD_NODE;
2121         }
2122 }
2123
2124 static int __get_segment_type_6(struct f2fs_io_info *fio)
2125 {
2126         if (fio->type == DATA) {
2127                 struct inode *inode = fio->page->mapping->host;
2128
2129                 if (is_cold_data(fio->page) || file_is_cold(inode))
2130                         return CURSEG_COLD_DATA;
2131                 if (is_inode_flag_set(inode, FI_HOT_DATA))
2132                         return CURSEG_HOT_DATA;
2133                 return CURSEG_WARM_DATA;
2134         } else {
2135                 if (IS_DNODE(fio->page))
2136                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2137                                                 CURSEG_HOT_NODE;
2138                 return CURSEG_COLD_NODE;
2139         }
2140 }
2141
2142 static int __get_segment_type(struct f2fs_io_info *fio)
2143 {
2144         int type = 0;
2145
2146         switch (fio->sbi->active_logs) {
2147         case 2:
2148                 type = __get_segment_type_2(fio);
2149                 break;
2150         case 4:
2151                 type = __get_segment_type_4(fio);
2152                 break;
2153         case 6:
2154                 type = __get_segment_type_6(fio);
2155                 break;
2156         default:
2157                 f2fs_bug_on(fio->sbi, true);
2158         }
2159
2160         if (IS_HOT(type))
2161                 fio->temp = HOT;
2162         else if (IS_WARM(type))
2163                 fio->temp = WARM;
2164         else
2165                 fio->temp = COLD;
2166         return type;
2167 }
2168
2169 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2170                 block_t old_blkaddr, block_t *new_blkaddr,
2171                 struct f2fs_summary *sum, int type,
2172                 struct f2fs_io_info *fio, bool add_list)
2173 {
2174         struct sit_info *sit_i = SIT_I(sbi);
2175         struct curseg_info *curseg = CURSEG_I(sbi, type);
2176
2177         mutex_lock(&curseg->curseg_mutex);
2178         mutex_lock(&sit_i->sentry_lock);
2179
2180         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2181
2182         f2fs_wait_discard_bio(sbi, *new_blkaddr);
2183
2184         /*
2185          * __add_sum_entry should be resided under the curseg_mutex
2186          * because, this function updates a summary entry in the
2187          * current summary block.
2188          */
2189         __add_sum_entry(sbi, type, sum);
2190
2191         __refresh_next_blkoff(sbi, curseg);
2192
2193         stat_inc_block_count(sbi, curseg);
2194
2195         if (!__has_curseg_space(sbi, type))
2196                 sit_i->s_ops->allocate_segment(sbi, type, false);
2197         /*
2198          * SIT information should be updated after segment allocation,
2199          * since we need to keep dirty segments precisely under SSR.
2200          */
2201         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
2202
2203         mutex_unlock(&sit_i->sentry_lock);
2204
2205         if (page && IS_NODESEG(type))
2206                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2207
2208         if (add_list) {
2209                 struct f2fs_bio_info *io;
2210
2211                 INIT_LIST_HEAD(&fio->list);
2212                 fio->in_list = true;
2213                 io = sbi->write_io[fio->type] + fio->temp;
2214                 spin_lock(&io->io_lock);
2215                 list_add_tail(&fio->list, &io->io_list);
2216                 spin_unlock(&io->io_lock);
2217         }
2218
2219         mutex_unlock(&curseg->curseg_mutex);
2220 }
2221
2222 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2223 {
2224         int type = __get_segment_type(fio);
2225         int err;
2226
2227 reallocate:
2228         allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
2229                         &fio->new_blkaddr, sum, type, fio, true);
2230
2231         /* writeout dirty page into bdev */
2232         err = f2fs_submit_page_write(fio);
2233         if (err == -EAGAIN) {
2234                 fio->old_blkaddr = fio->new_blkaddr;
2235                 goto reallocate;
2236         }
2237 }
2238
2239 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
2240 {
2241         struct f2fs_io_info fio = {
2242                 .sbi = sbi,
2243                 .type = META,
2244                 .op = REQ_OP_WRITE,
2245                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
2246                 .old_blkaddr = page->index,
2247                 .new_blkaddr = page->index,
2248                 .page = page,
2249                 .encrypted_page = NULL,
2250                 .in_list = false,
2251         };
2252
2253         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
2254                 fio.op_flags &= ~REQ_META;
2255
2256         set_page_writeback(page);
2257         f2fs_submit_page_write(&fio);
2258 }
2259
2260 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
2261 {
2262         struct f2fs_summary sum;
2263
2264         set_summary(&sum, nid, 0, 0);
2265         do_write_page(&sum, fio);
2266 }
2267
2268 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
2269 {
2270         struct f2fs_sb_info *sbi = fio->sbi;
2271         struct f2fs_summary sum;
2272         struct node_info ni;
2273
2274         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
2275         get_node_info(sbi, dn->nid, &ni);
2276         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
2277         do_write_page(&sum, fio);
2278         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
2279 }
2280
2281 int rewrite_data_page(struct f2fs_io_info *fio)
2282 {
2283         fio->new_blkaddr = fio->old_blkaddr;
2284         stat_inc_inplace_blocks(fio->sbi);
2285         return f2fs_submit_page_bio(fio);
2286 }
2287
2288 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
2289                                 block_t old_blkaddr, block_t new_blkaddr,
2290                                 bool recover_curseg, bool recover_newaddr)
2291 {
2292         struct sit_info *sit_i = SIT_I(sbi);
2293         struct curseg_info *curseg;
2294         unsigned int segno, old_cursegno;
2295         struct seg_entry *se;
2296         int type;
2297         unsigned short old_blkoff;
2298
2299         segno = GET_SEGNO(sbi, new_blkaddr);
2300         se = get_seg_entry(sbi, segno);
2301         type = se->type;
2302
2303         if (!recover_curseg) {
2304                 /* for recovery flow */
2305                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2306                         if (old_blkaddr == NULL_ADDR)
2307                                 type = CURSEG_COLD_DATA;
2308                         else
2309                                 type = CURSEG_WARM_DATA;
2310                 }
2311         } else {
2312                 if (!IS_CURSEG(sbi, segno))
2313                         type = CURSEG_WARM_DATA;
2314         }
2315
2316         curseg = CURSEG_I(sbi, type);
2317
2318         mutex_lock(&curseg->curseg_mutex);
2319         mutex_lock(&sit_i->sentry_lock);
2320
2321         old_cursegno = curseg->segno;
2322         old_blkoff = curseg->next_blkoff;
2323
2324         /* change the current segment */
2325         if (segno != curseg->segno) {
2326                 curseg->next_segno = segno;
2327                 change_curseg(sbi, type, true);
2328         }
2329
2330         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
2331         __add_sum_entry(sbi, type, sum);
2332
2333         if (!recover_curseg || recover_newaddr)
2334                 update_sit_entry(sbi, new_blkaddr, 1);
2335         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2336                 update_sit_entry(sbi, old_blkaddr, -1);
2337
2338         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2339         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2340
2341         locate_dirty_segment(sbi, old_cursegno);
2342
2343         if (recover_curseg) {
2344                 if (old_cursegno != curseg->segno) {
2345                         curseg->next_segno = old_cursegno;
2346                         change_curseg(sbi, type, true);
2347                 }
2348                 curseg->next_blkoff = old_blkoff;
2349         }
2350
2351         mutex_unlock(&sit_i->sentry_lock);
2352         mutex_unlock(&curseg->curseg_mutex);
2353 }
2354
2355 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2356                                 block_t old_addr, block_t new_addr,
2357                                 unsigned char version, bool recover_curseg,
2358                                 bool recover_newaddr)
2359 {
2360         struct f2fs_summary sum;
2361
2362         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2363
2364         __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2365                                         recover_curseg, recover_newaddr);
2366
2367         f2fs_update_data_blkaddr(dn, new_addr);
2368 }
2369
2370 void f2fs_wait_on_page_writeback(struct page *page,
2371                                 enum page_type type, bool ordered)
2372 {
2373         if (PageWriteback(page)) {
2374                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2375
2376                 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2377                                                 0, page->index, type);
2378                 if (ordered)
2379                         wait_on_page_writeback(page);
2380                 else
2381                         wait_for_stable_page(page);
2382         }
2383 }
2384
2385 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
2386                                                         block_t blkaddr)
2387 {
2388         struct page *cpage;
2389
2390         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
2391                 return;
2392
2393         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2394         if (cpage) {
2395                 f2fs_wait_on_page_writeback(cpage, DATA, true);
2396                 f2fs_put_page(cpage, 1);
2397         }
2398 }
2399
2400 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
2401 {
2402         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2403         struct curseg_info *seg_i;
2404         unsigned char *kaddr;
2405         struct page *page;
2406         block_t start;
2407         int i, j, offset;
2408
2409         start = start_sum_block(sbi);
2410
2411         page = get_meta_page(sbi, start++);
2412         kaddr = (unsigned char *)page_address(page);
2413
2414         /* Step 1: restore nat cache */
2415         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2416         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
2417
2418         /* Step 2: restore sit cache */
2419         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2420         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
2421         offset = 2 * SUM_JOURNAL_SIZE;
2422
2423         /* Step 3: restore summary entries */
2424         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2425                 unsigned short blk_off;
2426                 unsigned int segno;
2427
2428                 seg_i = CURSEG_I(sbi, i);
2429                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2430                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2431                 seg_i->next_segno = segno;
2432                 reset_curseg(sbi, i, 0);
2433                 seg_i->alloc_type = ckpt->alloc_type[i];
2434                 seg_i->next_blkoff = blk_off;
2435
2436                 if (seg_i->alloc_type == SSR)
2437                         blk_off = sbi->blocks_per_seg;
2438
2439                 for (j = 0; j < blk_off; j++) {
2440                         struct f2fs_summary *s;
2441                         s = (struct f2fs_summary *)(kaddr + offset);
2442                         seg_i->sum_blk->entries[j] = *s;
2443                         offset += SUMMARY_SIZE;
2444                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
2445                                                 SUM_FOOTER_SIZE)
2446                                 continue;
2447
2448                         f2fs_put_page(page, 1);
2449                         page = NULL;
2450
2451                         page = get_meta_page(sbi, start++);
2452                         kaddr = (unsigned char *)page_address(page);
2453                         offset = 0;
2454                 }
2455         }
2456         f2fs_put_page(page, 1);
2457         return 0;
2458 }
2459
2460 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2461 {
2462         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2463         struct f2fs_summary_block *sum;
2464         struct curseg_info *curseg;
2465         struct page *new;
2466         unsigned short blk_off;
2467         unsigned int segno = 0;
2468         block_t blk_addr = 0;
2469
2470         /* get segment number and block addr */
2471         if (IS_DATASEG(type)) {
2472                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2473                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2474                                                         CURSEG_HOT_DATA]);
2475                 if (__exist_node_summaries(sbi))
2476                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2477                 else
2478                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2479         } else {
2480                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2481                                                         CURSEG_HOT_NODE]);
2482                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2483                                                         CURSEG_HOT_NODE]);
2484                 if (__exist_node_summaries(sbi))
2485                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2486                                                         type - CURSEG_HOT_NODE);
2487                 else
2488                         blk_addr = GET_SUM_BLOCK(sbi, segno);
2489         }
2490
2491         new = get_meta_page(sbi, blk_addr);
2492         sum = (struct f2fs_summary_block *)page_address(new);
2493
2494         if (IS_NODESEG(type)) {
2495                 if (__exist_node_summaries(sbi)) {
2496                         struct f2fs_summary *ns = &sum->entries[0];
2497                         int i;
2498                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2499                                 ns->version = 0;
2500                                 ns->ofs_in_node = 0;
2501                         }
2502                 } else {
2503                         int err;
2504
2505                         err = restore_node_summary(sbi, segno, sum);
2506                         if (err) {
2507                                 f2fs_put_page(new, 1);
2508                                 return err;
2509                         }
2510                 }
2511         }
2512
2513         /* set uncompleted segment to curseg */
2514         curseg = CURSEG_I(sbi, type);
2515         mutex_lock(&curseg->curseg_mutex);
2516
2517         /* update journal info */
2518         down_write(&curseg->journal_rwsem);
2519         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2520         up_write(&curseg->journal_rwsem);
2521
2522         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2523         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
2524         curseg->next_segno = segno;
2525         reset_curseg(sbi, type, 0);
2526         curseg->alloc_type = ckpt->alloc_type[type];
2527         curseg->next_blkoff = blk_off;
2528         mutex_unlock(&curseg->curseg_mutex);
2529         f2fs_put_page(new, 1);
2530         return 0;
2531 }
2532
2533 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2534 {
2535         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
2536         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
2537         int type = CURSEG_HOT_DATA;
2538         int err;
2539
2540         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
2541                 int npages = npages_for_summary_flush(sbi, true);
2542
2543                 if (npages >= 2)
2544                         ra_meta_pages(sbi, start_sum_block(sbi), npages,
2545                                                         META_CP, true);
2546
2547                 /* restore for compacted data summary */
2548                 if (read_compacted_summaries(sbi))
2549                         return -EINVAL;
2550                 type = CURSEG_HOT_NODE;
2551         }
2552
2553         if (__exist_node_summaries(sbi))
2554                 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
2555                                         NR_CURSEG_TYPE - type, META_CP, true);
2556
2557         for (; type <= CURSEG_COLD_NODE; type++) {
2558                 err = read_normal_summaries(sbi, type);
2559                 if (err)
2560                         return err;
2561         }
2562
2563         /* sanity check for summary blocks */
2564         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
2565                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
2566                 return -EINVAL;
2567
2568         return 0;
2569 }
2570
2571 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2572 {
2573         struct page *page;
2574         unsigned char *kaddr;
2575         struct f2fs_summary *summary;
2576         struct curseg_info *seg_i;
2577         int written_size = 0;
2578         int i, j;
2579
2580         page = grab_meta_page(sbi, blkaddr++);
2581         kaddr = (unsigned char *)page_address(page);
2582
2583         /* Step 1: write nat cache */
2584         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2585         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
2586         written_size += SUM_JOURNAL_SIZE;
2587
2588         /* Step 2: write sit cache */
2589         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2590         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
2591         written_size += SUM_JOURNAL_SIZE;
2592
2593         /* Step 3: write summary entries */
2594         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2595                 unsigned short blkoff;
2596                 seg_i = CURSEG_I(sbi, i);
2597                 if (sbi->ckpt->alloc_type[i] == SSR)
2598                         blkoff = sbi->blocks_per_seg;
2599                 else
2600                         blkoff = curseg_blkoff(sbi, i);
2601
2602                 for (j = 0; j < blkoff; j++) {
2603                         if (!page) {
2604                                 page = grab_meta_page(sbi, blkaddr++);
2605                                 kaddr = (unsigned char *)page_address(page);
2606                                 written_size = 0;
2607                         }
2608                         summary = (struct f2fs_summary *)(kaddr + written_size);
2609                         *summary = seg_i->sum_blk->entries[j];
2610                         written_size += SUMMARY_SIZE;
2611
2612                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
2613                                                         SUM_FOOTER_SIZE)
2614                                 continue;
2615
2616                         set_page_dirty(page);
2617                         f2fs_put_page(page, 1);
2618                         page = NULL;
2619                 }
2620         }
2621         if (page) {
2622                 set_page_dirty(page);
2623                 f2fs_put_page(page, 1);
2624         }
2625 }
2626
2627 static void write_normal_summaries(struct f2fs_sb_info *sbi,
2628                                         block_t blkaddr, int type)
2629 {
2630         int i, end;
2631         if (IS_DATASEG(type))
2632                 end = type + NR_CURSEG_DATA_TYPE;
2633         else
2634                 end = type + NR_CURSEG_NODE_TYPE;
2635
2636         for (i = type; i < end; i++)
2637                 write_current_sum_page(sbi, i, blkaddr + (i - type));
2638 }
2639
2640 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2641 {
2642         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
2643                 write_compacted_summaries(sbi, start_blk);
2644         else
2645                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2646 }
2647
2648 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2649 {
2650         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
2651 }
2652
2653 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
2654                                         unsigned int val, int alloc)
2655 {
2656         int i;
2657
2658         if (type == NAT_JOURNAL) {
2659                 for (i = 0; i < nats_in_cursum(journal); i++) {
2660                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
2661                                 return i;
2662                 }
2663                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2664                         return update_nats_in_cursum(journal, 1);
2665         } else if (type == SIT_JOURNAL) {
2666                 for (i = 0; i < sits_in_cursum(journal); i++)
2667                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
2668                                 return i;
2669                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2670                         return update_sits_in_cursum(journal, 1);
2671         }
2672         return -1;
2673 }
2674
2675 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2676                                         unsigned int segno)
2677 {
2678         return get_meta_page(sbi, current_sit_addr(sbi, segno));
2679 }
2680
2681 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2682                                         unsigned int start)
2683 {
2684         struct sit_info *sit_i = SIT_I(sbi);
2685         struct page *src_page, *dst_page;
2686         pgoff_t src_off, dst_off;
2687         void *src_addr, *dst_addr;
2688
2689         src_off = current_sit_addr(sbi, start);
2690         dst_off = next_sit_addr(sbi, src_off);
2691
2692         /* get current sit block page without lock */
2693         src_page = get_meta_page(sbi, src_off);
2694         dst_page = grab_meta_page(sbi, dst_off);
2695         f2fs_bug_on(sbi, PageDirty(src_page));
2696
2697         src_addr = page_address(src_page);
2698         dst_addr = page_address(dst_page);
2699         memcpy(dst_addr, src_addr, PAGE_SIZE);
2700
2701         set_page_dirty(dst_page);
2702         f2fs_put_page(src_page, 1);
2703
2704         set_to_next_sit(sit_i, start);
2705
2706         return dst_page;
2707 }
2708
2709 static struct sit_entry_set *grab_sit_entry_set(void)
2710 {
2711         struct sit_entry_set *ses =
2712                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
2713
2714         ses->entry_cnt = 0;
2715         INIT_LIST_HEAD(&ses->set_list);
2716         return ses;
2717 }
2718
2719 static void release_sit_entry_set(struct sit_entry_set *ses)
2720 {
2721         list_del(&ses->set_list);
2722         kmem_cache_free(sit_entry_set_slab, ses);
2723 }
2724
2725 static void adjust_sit_entry_set(struct sit_entry_set *ses,
2726                                                 struct list_head *head)
2727 {
2728         struct sit_entry_set *next = ses;
2729
2730         if (list_is_last(&ses->set_list, head))
2731                 return;
2732
2733         list_for_each_entry_continue(next, head, set_list)
2734                 if (ses->entry_cnt <= next->entry_cnt)
2735                         break;
2736
2737         list_move_tail(&ses->set_list, &next->set_list);
2738 }
2739
2740 static void add_sit_entry(unsigned int segno, struct list_head *head)
2741 {
2742         struct sit_entry_set *ses;
2743         unsigned int start_segno = START_SEGNO(segno);
2744
2745         list_for_each_entry(ses, head, set_list) {
2746                 if (ses->start_segno == start_segno) {
2747                         ses->entry_cnt++;
2748                         adjust_sit_entry_set(ses, head);
2749                         return;
2750                 }
2751         }
2752
2753         ses = grab_sit_entry_set();
2754
2755         ses->start_segno = start_segno;
2756         ses->entry_cnt++;
2757         list_add(&ses->set_list, head);
2758 }
2759
2760 static void add_sits_in_set(struct f2fs_sb_info *sbi)
2761 {
2762         struct f2fs_sm_info *sm_info = SM_I(sbi);
2763         struct list_head *set_list = &sm_info->sit_entry_set;
2764         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
2765         unsigned int segno;
2766
2767         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
2768                 add_sit_entry(segno, set_list);
2769 }
2770
2771 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
2772 {
2773         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2774         struct f2fs_journal *journal = curseg->journal;
2775         int i;
2776
2777         down_write(&curseg->journal_rwsem);
2778         for (i = 0; i < sits_in_cursum(journal); i++) {
2779                 unsigned int segno;
2780                 bool dirtied;
2781
2782                 segno = le32_to_cpu(segno_in_journal(journal, i));
2783                 dirtied = __mark_sit_entry_dirty(sbi, segno);
2784
2785                 if (!dirtied)
2786                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
2787         }
2788         update_sits_in_cursum(journal, -i);
2789         up_write(&curseg->journal_rwsem);
2790 }
2791
2792 /*
2793  * CP calls this function, which flushes SIT entries including sit_journal,
2794  * and moves prefree segs to free segs.
2795  */
2796 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2797 {
2798         struct sit_info *sit_i = SIT_I(sbi);
2799         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2800         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2801         struct f2fs_journal *journal = curseg->journal;
2802         struct sit_entry_set *ses, *tmp;
2803         struct list_head *head = &SM_I(sbi)->sit_entry_set;
2804         bool to_journal = true;
2805         struct seg_entry *se;
2806
2807         mutex_lock(&sit_i->sentry_lock);
2808
2809         if (!sit_i->dirty_sentries)
2810                 goto out;
2811
2812         /*
2813          * add and account sit entries of dirty bitmap in sit entry
2814          * set temporarily
2815          */
2816         add_sits_in_set(sbi);
2817
2818         /*
2819          * if there are no enough space in journal to store dirty sit
2820          * entries, remove all entries from journal and add and account
2821          * them in sit entry set.
2822          */
2823         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
2824                 remove_sits_in_journal(sbi);
2825
2826         /*
2827          * there are two steps to flush sit entries:
2828          * #1, flush sit entries to journal in current cold data summary block.
2829          * #2, flush sit entries to sit page.
2830          */
2831         list_for_each_entry_safe(ses, tmp, head, set_list) {
2832                 struct page *page = NULL;
2833                 struct f2fs_sit_block *raw_sit = NULL;
2834                 unsigned int start_segno = ses->start_segno;
2835                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
2836                                                 (unsigned long)MAIN_SEGS(sbi));
2837                 unsigned int segno = start_segno;
2838
2839                 if (to_journal &&
2840                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
2841                         to_journal = false;
2842
2843                 if (to_journal) {
2844                         down_write(&curseg->journal_rwsem);
2845                 } else {
2846                         page = get_next_sit_page(sbi, start_segno);
2847                         raw_sit = page_address(page);
2848                 }
2849
2850                 /* flush dirty sit entries in region of current sit set */
2851                 for_each_set_bit_from(segno, bitmap, end) {
2852                         int offset, sit_offset;
2853
2854                         se = get_seg_entry(sbi, segno);
2855
2856                         /* add discard candidates */
2857                         if (!(cpc->reason & CP_DISCARD)) {
2858                                 cpc->trim_start = segno;
2859                                 add_discard_addrs(sbi, cpc, false);
2860                         }
2861
2862                         if (to_journal) {
2863                                 offset = lookup_journal_in_cursum(journal,
2864                                                         SIT_JOURNAL, segno, 1);
2865                                 f2fs_bug_on(sbi, offset < 0);
2866                                 segno_in_journal(journal, offset) =
2867                                                         cpu_to_le32(segno);
2868                                 seg_info_to_raw_sit(se,
2869                                         &sit_in_journal(journal, offset));
2870                         } else {
2871                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2872                                 seg_info_to_raw_sit(se,
2873                                                 &raw_sit->entries[sit_offset]);
2874                         }
2875
2876                         __clear_bit(segno, bitmap);
2877                         sit_i->dirty_sentries--;
2878                         ses->entry_cnt--;
2879                 }
2880
2881                 if (to_journal)
2882                         up_write(&curseg->journal_rwsem);
2883                 else
2884                         f2fs_put_page(page, 1);
2885
2886                 f2fs_bug_on(sbi, ses->entry_cnt);
2887                 release_sit_entry_set(ses);
2888         }
2889
2890         f2fs_bug_on(sbi, !list_empty(head));
2891         f2fs_bug_on(sbi, sit_i->dirty_sentries);
2892 out:
2893         if (cpc->reason & CP_DISCARD) {
2894                 __u64 trim_start = cpc->trim_start;
2895
2896                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2897                         add_discard_addrs(sbi, cpc, false);
2898
2899                 cpc->trim_start = trim_start;
2900         }
2901         mutex_unlock(&sit_i->sentry_lock);
2902
2903         set_prefree_as_free_segments(sbi);
2904 }
2905
2906 static int build_sit_info(struct f2fs_sb_info *sbi)
2907 {
2908         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2909         struct sit_info *sit_i;
2910         unsigned int sit_segs, start;
2911         char *src_bitmap;
2912         unsigned int bitmap_size;
2913
2914         /* allocate memory for SIT information */
2915         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2916         if (!sit_i)
2917                 return -ENOMEM;
2918
2919         SM_I(sbi)->sit_info = sit_i;
2920
2921         sit_i->sentries = kvzalloc(MAIN_SEGS(sbi) *
2922                                         sizeof(struct seg_entry), GFP_KERNEL);
2923         if (!sit_i->sentries)
2924                 return -ENOMEM;
2925
2926         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2927         sit_i->dirty_sentries_bitmap = kvzalloc(bitmap_size, GFP_KERNEL);
2928         if (!sit_i->dirty_sentries_bitmap)
2929                 return -ENOMEM;
2930
2931         for (start = 0; start < MAIN_SEGS(sbi); start++) {
2932                 sit_i->sentries[start].cur_valid_map
2933                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2934                 sit_i->sentries[start].ckpt_valid_map
2935                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2936                 if (!sit_i->sentries[start].cur_valid_map ||
2937                                 !sit_i->sentries[start].ckpt_valid_map)
2938                         return -ENOMEM;
2939
2940 #ifdef CONFIG_F2FS_CHECK_FS
2941                 sit_i->sentries[start].cur_valid_map_mir
2942                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2943                 if (!sit_i->sentries[start].cur_valid_map_mir)
2944                         return -ENOMEM;
2945 #endif
2946
2947                 if (f2fs_discard_en(sbi)) {
2948                         sit_i->sentries[start].discard_map
2949                                 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2950                         if (!sit_i->sentries[start].discard_map)
2951                                 return -ENOMEM;
2952                 }
2953         }
2954
2955         sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2956         if (!sit_i->tmp_map)
2957                 return -ENOMEM;
2958
2959         if (sbi->segs_per_sec > 1) {
2960                 sit_i->sec_entries = kvzalloc(MAIN_SECS(sbi) *
2961                                         sizeof(struct sec_entry), GFP_KERNEL);
2962                 if (!sit_i->sec_entries)
2963                         return -ENOMEM;
2964         }
2965
2966         /* get information related with SIT */
2967         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2968
2969         /* setup SIT bitmap from ckeckpoint pack */
2970         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2971         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2972
2973         sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2974         if (!sit_i->sit_bitmap)
2975                 return -ENOMEM;
2976
2977 #ifdef CONFIG_F2FS_CHECK_FS
2978         sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2979         if (!sit_i->sit_bitmap_mir)
2980                 return -ENOMEM;
2981 #endif
2982
2983         /* init SIT information */
2984         sit_i->s_ops = &default_salloc_ops;
2985
2986         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2987         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2988         sit_i->written_valid_blocks = 0;
2989         sit_i->bitmap_size = bitmap_size;
2990         sit_i->dirty_sentries = 0;
2991         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2992         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2993         sit_i->mounted_time = ktime_get_real_seconds();
2994         mutex_init(&sit_i->sentry_lock);
2995         return 0;
2996 }
2997
2998 static int build_free_segmap(struct f2fs_sb_info *sbi)
2999 {
3000         struct free_segmap_info *free_i;
3001         unsigned int bitmap_size, sec_bitmap_size;
3002
3003         /* allocate memory for free segmap information */
3004         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
3005         if (!free_i)
3006                 return -ENOMEM;
3007
3008         SM_I(sbi)->free_info = free_i;
3009
3010         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3011         free_i->free_segmap = kvmalloc(bitmap_size, GFP_KERNEL);
3012         if (!free_i->free_segmap)
3013                 return -ENOMEM;
3014
3015         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3016         free_i->free_secmap = kvmalloc(sec_bitmap_size, GFP_KERNEL);
3017         if (!free_i->free_secmap)
3018                 return -ENOMEM;
3019
3020         /* set all segments as dirty temporarily */
3021         memset(free_i->free_segmap, 0xff, bitmap_size);
3022         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3023
3024         /* init free segmap information */
3025         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
3026         free_i->free_segments = 0;
3027         free_i->free_sections = 0;
3028         spin_lock_init(&free_i->segmap_lock);
3029         return 0;
3030 }
3031
3032 static int build_curseg(struct f2fs_sb_info *sbi)
3033 {
3034         struct curseg_info *array;
3035         int i;
3036
3037         array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
3038         if (!array)
3039                 return -ENOMEM;
3040
3041         SM_I(sbi)->curseg_array = array;
3042
3043         for (i = 0; i < NR_CURSEG_TYPE; i++) {
3044                 mutex_init(&array[i].curseg_mutex);
3045                 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
3046                 if (!array[i].sum_blk)
3047                         return -ENOMEM;
3048                 init_rwsem(&array[i].journal_rwsem);
3049                 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
3050                                                         GFP_KERNEL);
3051                 if (!array[i].journal)
3052                         return -ENOMEM;
3053                 array[i].segno = NULL_SEGNO;
3054                 array[i].next_blkoff = 0;
3055         }
3056         return restore_curseg_summaries(sbi);
3057 }
3058
3059 static void build_sit_entries(struct f2fs_sb_info *sbi)
3060 {
3061         struct sit_info *sit_i = SIT_I(sbi);
3062         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3063         struct f2fs_journal *journal = curseg->journal;
3064         struct seg_entry *se;
3065         struct f2fs_sit_entry sit;
3066         int sit_blk_cnt = SIT_BLK_CNT(sbi);
3067         unsigned int i, start, end;
3068         unsigned int readed, start_blk = 0;
3069
3070         do {
3071                 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3072                                                         META_SIT, true);
3073
3074                 start = start_blk * sit_i->sents_per_block;
3075                 end = (start_blk + readed) * sit_i->sents_per_block;
3076
3077                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
3078                         struct f2fs_sit_block *sit_blk;
3079                         struct page *page;
3080
3081                         se = &sit_i->sentries[start];
3082                         page = get_current_sit_page(sbi, start);
3083                         sit_blk = (struct f2fs_sit_block *)page_address(page);
3084                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3085                         f2fs_put_page(page, 1);
3086
3087                         check_block_count(sbi, start, &sit);
3088                         seg_info_from_raw_sit(se, &sit);
3089
3090                         /* build discard map only one time */
3091                         if (f2fs_discard_en(sbi)) {
3092                                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3093                                         memset(se->discard_map, 0xff,
3094                                                 SIT_VBLOCK_MAP_SIZE);
3095                                 } else {
3096                                         memcpy(se->discard_map,
3097                                                 se->cur_valid_map,
3098                                                 SIT_VBLOCK_MAP_SIZE);
3099                                         sbi->discard_blks +=
3100                                                 sbi->blocks_per_seg -
3101                                                 se->valid_blocks;
3102                                 }
3103                         }
3104
3105                         if (sbi->segs_per_sec > 1)
3106                                 get_sec_entry(sbi, start)->valid_blocks +=
3107                                                         se->valid_blocks;
3108                 }
3109                 start_blk += readed;
3110         } while (start_blk < sit_blk_cnt);
3111
3112         down_read(&curseg->journal_rwsem);
3113         for (i = 0; i < sits_in_cursum(journal); i++) {
3114                 unsigned int old_valid_blocks;
3115
3116                 start = le32_to_cpu(segno_in_journal(journal, i));
3117                 se = &sit_i->sentries[start];
3118                 sit = sit_in_journal(journal, i);
3119
3120                 old_valid_blocks = se->valid_blocks;
3121
3122                 check_block_count(sbi, start, &sit);
3123                 seg_info_from_raw_sit(se, &sit);
3124
3125                 if (f2fs_discard_en(sbi)) {
3126                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3127                                 memset(se->discard_map, 0xff,
3128                                                         SIT_VBLOCK_MAP_SIZE);
3129                         } else {
3130                                 memcpy(se->discard_map, se->cur_valid_map,
3131                                                         SIT_VBLOCK_MAP_SIZE);
3132                                 sbi->discard_blks += old_valid_blocks -
3133                                                         se->valid_blocks;
3134                         }
3135                 }
3136
3137                 if (sbi->segs_per_sec > 1)
3138                         get_sec_entry(sbi, start)->valid_blocks +=
3139                                 se->valid_blocks - old_valid_blocks;
3140         }
3141         up_read(&curseg->journal_rwsem);
3142 }
3143
3144 static void init_free_segmap(struct f2fs_sb_info *sbi)
3145 {
3146         unsigned int start;
3147         int type;
3148
3149         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3150                 struct seg_entry *sentry = get_seg_entry(sbi, start);
3151                 if (!sentry->valid_blocks)
3152                         __set_free(sbi, start);
3153                 else
3154                         SIT_I(sbi)->written_valid_blocks +=
3155                                                 sentry->valid_blocks;
3156         }
3157
3158         /* set use the current segments */
3159         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3160                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3161                 __set_test_and_inuse(sbi, curseg_t->segno);
3162         }
3163 }
3164
3165 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3166 {
3167         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3168         struct free_segmap_info *free_i = FREE_I(sbi);
3169         unsigned int segno = 0, offset = 0;
3170         unsigned short valid_blocks;
3171
3172         while (1) {
3173                 /* find dirty segment based on free segmap */
3174                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3175                 if (segno >= MAIN_SEGS(sbi))
3176                         break;
3177                 offset = segno + 1;
3178                 valid_blocks = get_valid_blocks(sbi, segno, false);
3179                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
3180                         continue;
3181                 if (valid_blocks > sbi->blocks_per_seg) {
3182                         f2fs_bug_on(sbi, 1);
3183                         continue;
3184                 }
3185                 mutex_lock(&dirty_i->seglist_lock);
3186                 __locate_dirty_segment(sbi, segno, DIRTY);
3187                 mutex_unlock(&dirty_i->seglist_lock);
3188         }
3189 }
3190
3191 static int init_victim_secmap(struct f2fs_sb_info *sbi)
3192 {
3193         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3194         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3195
3196         dirty_i->victim_secmap = kvzalloc(bitmap_size, GFP_KERNEL);
3197         if (!dirty_i->victim_secmap)
3198                 return -ENOMEM;
3199         return 0;
3200 }
3201
3202 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3203 {
3204         struct dirty_seglist_info *dirty_i;
3205         unsigned int bitmap_size, i;
3206
3207         /* allocate memory for dirty segments list information */
3208         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
3209         if (!dirty_i)
3210                 return -ENOMEM;
3211
3212         SM_I(sbi)->dirty_info = dirty_i;
3213         mutex_init(&dirty_i->seglist_lock);
3214
3215         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3216
3217         for (i = 0; i < NR_DIRTY_TYPE; i++) {
3218                 dirty_i->dirty_segmap[i] = kvzalloc(bitmap_size, GFP_KERNEL);
3219                 if (!dirty_i->dirty_segmap[i])
3220                         return -ENOMEM;
3221         }
3222
3223         init_dirty_segmap(sbi);
3224         return init_victim_secmap(sbi);
3225 }
3226
3227 /*
3228  * Update min, max modified time for cost-benefit GC algorithm
3229  */
3230 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3231 {
3232         struct sit_info *sit_i = SIT_I(sbi);
3233         unsigned int segno;
3234
3235         mutex_lock(&sit_i->sentry_lock);
3236
3237         sit_i->min_mtime = LLONG_MAX;
3238
3239         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
3240                 unsigned int i;
3241                 unsigned long long mtime = 0;
3242
3243                 for (i = 0; i < sbi->segs_per_sec; i++)
3244                         mtime += get_seg_entry(sbi, segno + i)->mtime;
3245
3246                 mtime = div_u64(mtime, sbi->segs_per_sec);
3247
3248                 if (sit_i->min_mtime > mtime)
3249                         sit_i->min_mtime = mtime;
3250         }
3251         sit_i->max_mtime = get_mtime(sbi);
3252         mutex_unlock(&sit_i->sentry_lock);
3253 }
3254
3255 int build_segment_manager(struct f2fs_sb_info *sbi)
3256 {
3257         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3258         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3259         struct f2fs_sm_info *sm_info;
3260         int err;
3261
3262         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
3263         if (!sm_info)
3264                 return -ENOMEM;
3265
3266         /* init sm info */
3267         sbi->sm_info = sm_info;
3268         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3269         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3270         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3271         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3272         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3273         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3274         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
3275         sm_info->rec_prefree_segments = sm_info->main_segments *
3276                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
3277         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3278                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3279
3280         if (!test_opt(sbi, LFS))
3281                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
3282         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
3283         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
3284         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
3285
3286         sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
3287
3288         INIT_LIST_HEAD(&sm_info->sit_entry_set);
3289
3290         if (!f2fs_readonly(sbi->sb)) {
3291                 err = create_flush_cmd_control(sbi);
3292                 if (err)
3293                         return err;
3294         }
3295
3296         err = create_discard_cmd_control(sbi);
3297         if (err)
3298                 return err;
3299
3300         err = build_sit_info(sbi);
3301         if (err)
3302                 return err;
3303         err = build_free_segmap(sbi);
3304         if (err)
3305                 return err;
3306         err = build_curseg(sbi);
3307         if (err)
3308                 return err;
3309
3310         /* reinit free segmap based on SIT */
3311         build_sit_entries(sbi);
3312
3313         init_free_segmap(sbi);
3314         err = build_dirty_segmap(sbi);
3315         if (err)
3316                 return err;
3317
3318         init_min_max_mtime(sbi);
3319         return 0;
3320 }
3321
3322 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3323                 enum dirty_type dirty_type)
3324 {
3325         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3326
3327         mutex_lock(&dirty_i->seglist_lock);
3328         kvfree(dirty_i->dirty_segmap[dirty_type]);
3329         dirty_i->nr_dirty[dirty_type] = 0;
3330         mutex_unlock(&dirty_i->seglist_lock);
3331 }
3332
3333 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
3334 {
3335         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3336         kvfree(dirty_i->victim_secmap);
3337 }
3338
3339 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3340 {
3341         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3342         int i;
3343
3344         if (!dirty_i)
3345                 return;
3346
3347         /* discard pre-free/dirty segments list */
3348         for (i = 0; i < NR_DIRTY_TYPE; i++)
3349                 discard_dirty_segmap(sbi, i);
3350
3351         destroy_victim_secmap(sbi);
3352         SM_I(sbi)->dirty_info = NULL;
3353         kfree(dirty_i);
3354 }
3355
3356 static void destroy_curseg(struct f2fs_sb_info *sbi)
3357 {
3358         struct curseg_info *array = SM_I(sbi)->curseg_array;
3359         int i;
3360
3361         if (!array)
3362                 return;
3363         SM_I(sbi)->curseg_array = NULL;
3364         for (i = 0; i < NR_CURSEG_TYPE; i++) {
3365                 kfree(array[i].sum_blk);
3366                 kfree(array[i].journal);
3367         }
3368         kfree(array);
3369 }
3370
3371 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3372 {
3373         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3374         if (!free_i)
3375                 return;
3376         SM_I(sbi)->free_info = NULL;
3377         kvfree(free_i->free_segmap);
3378         kvfree(free_i->free_secmap);
3379         kfree(free_i);
3380 }
3381
3382 static void destroy_sit_info(struct f2fs_sb_info *sbi)
3383 {
3384         struct sit_info *sit_i = SIT_I(sbi);
3385         unsigned int start;
3386
3387         if (!sit_i)
3388                 return;
3389
3390         if (sit_i->sentries) {
3391                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3392                         kfree(sit_i->sentries[start].cur_valid_map);
3393 #ifdef CONFIG_F2FS_CHECK_FS
3394                         kfree(sit_i->sentries[start].cur_valid_map_mir);
3395 #endif
3396                         kfree(sit_i->sentries[start].ckpt_valid_map);
3397                         kfree(sit_i->sentries[start].discard_map);
3398                 }
3399         }
3400         kfree(sit_i->tmp_map);
3401
3402         kvfree(sit_i->sentries);
3403         kvfree(sit_i->sec_entries);
3404         kvfree(sit_i->dirty_sentries_bitmap);
3405
3406         SM_I(sbi)->sit_info = NULL;
3407         kfree(sit_i->sit_bitmap);
3408 #ifdef CONFIG_F2FS_CHECK_FS
3409         kfree(sit_i->sit_bitmap_mir);
3410 #endif
3411         kfree(sit_i);
3412 }
3413
3414 void destroy_segment_manager(struct f2fs_sb_info *sbi)
3415 {
3416         struct f2fs_sm_info *sm_info = SM_I(sbi);
3417
3418         if (!sm_info)
3419                 return;
3420         destroy_flush_cmd_control(sbi, true);
3421         destroy_discard_cmd_control(sbi);
3422         destroy_dirty_segmap(sbi);
3423         destroy_curseg(sbi);
3424         destroy_free_segmap(sbi);
3425         destroy_sit_info(sbi);
3426         sbi->sm_info = NULL;
3427         kfree(sm_info);
3428 }
3429
3430 int __init create_segment_manager_caches(void)
3431 {
3432         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
3433                         sizeof(struct discard_entry));
3434         if (!discard_entry_slab)
3435                 goto fail;
3436
3437         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3438                         sizeof(struct discard_cmd));
3439         if (!discard_cmd_slab)
3440                 goto destroy_discard_entry;
3441
3442         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
3443                         sizeof(struct sit_entry_set));
3444         if (!sit_entry_set_slab)
3445                 goto destroy_discard_cmd;
3446
3447         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3448                         sizeof(struct inmem_pages));
3449         if (!inmem_entry_slab)
3450                 goto destroy_sit_entry_set;
3451         return 0;
3452
3453 destroy_sit_entry_set:
3454         kmem_cache_destroy(sit_entry_set_slab);
3455 destroy_discard_cmd:
3456         kmem_cache_destroy(discard_cmd_slab);
3457 destroy_discard_entry:
3458         kmem_cache_destroy(discard_entry_slab);
3459 fail:
3460         return -ENOMEM;
3461 }
3462
3463 void destroy_segment_manager_caches(void)
3464 {
3465         kmem_cache_destroy(sit_entry_set_slab);
3466         kmem_cache_destroy(discard_cmd_slab);
3467         kmem_cache_destroy(discard_entry_slab);
3468         kmem_cache_destroy(inmem_entry_slab);
3469 }