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                 goto init_thread;
 559         }
 560
 561         fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
 562         if (!fcc)
 563                 return -ENOMEM;
 564         atomic_set(&fcc->issued_flush, 0);
 565         atomic_set(&fcc->issing_flush, 0);
 566         init_waitqueue_head(&fcc->flush_wait_queue);
 567         init_llist_head(&fcc->issue_list);
 568         SM_I(sbi)->fcc_info = fcc;
 569 init_thread:
 570         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 571                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 572         if (IS_ERR(fcc->f2fs_issue_flush)) {
 573                 err = PTR_ERR(fcc->f2fs_issue_flush);
 574                 kfree(fcc);
 575                 SM_I(sbi)->fcc_info = NULL;
 576                 return err;
 577         }
 578
 579         return err;
 580 }
 581
 582 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 583 {
 584         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 585
 586         if (fcc && fcc->f2fs_issue_flush) {
 587                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 588
 589                 fcc->f2fs_issue_flush = NULL;
 590                 kthread_stop(flush_thread);
 591         }
 592         if (free) {
 593                 kfree(fcc);
 594                 SM_I(sbi)->fcc_info = NULL;
 595         }
 596 }
 597
 598 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 599                 enum dirty_type dirty_type)
 600 {
 601         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 602
 603         /* need not be added */
 604         if (IS_CURSEG(sbi, segno))
 605                 return;
 606
 607         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 608                 dirty_i->nr_dirty[dirty_type]++;
 609
 610         if (dirty_type == DIRTY) {
 611                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 612                 enum dirty_type t = sentry->type;
 613
 614                 if (unlikely(t >= DIRTY)) {
 615                         f2fs_bug_on(sbi, 1);
 616                         return;
 617                 }
 618                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 619                         dirty_i->nr_dirty[t]++;
 620         }
 621 }
 622
 623 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 624                 enum dirty_type dirty_type)
 625 {
 626         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 627
 628         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 629                 dirty_i->nr_dirty[dirty_type]--;
 630
 631         if (dirty_type == DIRTY) {
 632                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 633                 enum dirty_type t = sentry->type;
 634
 635                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 636                         dirty_i->nr_dirty[t]--;
 637
 638                 if (get_valid_blocks(sbi, segno, true) == 0)
 639                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 640                                                 dirty_i->victim_secmap);
 641         }
 642 }
 643
 644 /*
 645  * Should not occur error such as -ENOMEM.
 646  * Adding dirty entry into seglist is not critical operation.
 647  * If a given segment is one of current working segments, it won't be added.
 648  */
 649 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 650 {
 651         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 652         unsigned short valid_blocks;
 653
 654         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 655                 return;
 656
 657         mutex_lock(&dirty_i->seglist_lock);
 658
 659         valid_blocks = get_valid_blocks(sbi, segno, false);
 660
 661         if (valid_blocks == 0) {
 662                 __locate_dirty_segment(sbi, segno, PRE);
 663                 __remove_dirty_segment(sbi, segno, DIRTY);
 664         } else if (valid_blocks < sbi->blocks_per_seg) {
 665                 __locate_dirty_segment(sbi, segno, DIRTY);
 666         } else {
 667                 /* Recovery routine with SSR needs this */
 668                 __remove_dirty_segment(sbi, segno, DIRTY);
 669         }
 670
 671         mutex_unlock(&dirty_i->seglist_lock);
 672 }
 673
 674 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 675                 struct block_device *bdev, block_t lstart,
 676                 block_t start, block_t len)
 677 {
 678         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 679         struct list_head *pend_list;
 680         struct discard_cmd *dc;
 681
 682         f2fs_bug_on(sbi, !len);
 683
 684         pend_list = &dcc->pend_list[plist_idx(len)];
 685
 686         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 687         INIT_LIST_HEAD(&dc->list);
 688         dc->bdev = bdev;
 689         dc->lstart = lstart;
 690         dc->start = start;
 691         dc->len = len;
 692         dc->ref = 0;
 693         dc->state = D_PREP;
 694         dc->error = 0;
 695         init_completion(&dc->wait);
 696         list_add_tail(&dc->list, pend_list);
 697         atomic_inc(&dcc->discard_cmd_cnt);
 698         dcc->undiscard_blks += len;
 699
 700         return dc;
 701 }
 702
 703 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 704                                 struct block_device *bdev, block_t lstart,
 705                                 block_t start, block_t len,
 706                                 struct rb_node *parent, struct rb_node **p)
 707 {
 708         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 709         struct discard_cmd *dc;
 710
 711         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
 712
 713         rb_link_node(&dc->rb_node, parent, p);
 714         rb_insert_color(&dc->rb_node, &dcc->root);
 715
 716         return dc;
 717 }
 718
 719 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
 720                                                         struct discard_cmd *dc)
 721 {
 722         if (dc->state == D_DONE)
 723                 atomic_dec(&dcc->issing_discard);
 724
 725         list_del(&dc->list);
 726         rb_erase(&dc->rb_node, &dcc->root);
 727         dcc->undiscard_blks -= dc->len;
 728
 729         kmem_cache_free(discard_cmd_slab, dc);
 730
 731         atomic_dec(&dcc->discard_cmd_cnt);
 732 }
 733
 734 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
 735                                                         struct discard_cmd *dc)
 736 {
 737         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 738
 739         if (dc->error == -EOPNOTSUPP)
 740                 dc->error = 0;
 741
 742         if (dc->error)
 743                 f2fs_msg(sbi->sb, KERN_INFO,
 744                         "Issue discard(%u, %u, %u) failed, ret: %d",
 745                         dc->lstart, dc->start, dc->len, dc->error);
 746         __detach_discard_cmd(dcc, dc);
 747 }
 748
 749 static void f2fs_submit_discard_endio(struct bio *bio)
 750 {
 751         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
 752
 753         dc->error = bio->bi_error;
 754         dc->state = D_DONE;
 755         complete_all(&dc->wait);
 756         bio_put(bio);
 757 }
 758
 759 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
 760 static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
 761                                 struct discard_cmd *dc)
 762 {
 763         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 764         struct bio *bio = NULL;
 765
 766         if (dc->state != D_PREP)
 767                 return;
 768
 769         trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
 770
 771         dc->error = __blkdev_issue_discard(dc->bdev,
 772                                 SECTOR_FROM_BLOCK(dc->start),
 773                                 SECTOR_FROM_BLOCK(dc->len),
 774                                 GFP_NOFS, 0, &bio);
 775         if (!dc->error) {
 776                 /* should keep before submission to avoid D_DONE right away */
 777                 dc->state = D_SUBMIT;
 778                 atomic_inc(&dcc->issued_discard);
 779                 atomic_inc(&dcc->issing_discard);
 780                 if (bio) {
 781                         bio->bi_private = dc;
 782                         bio->bi_end_io = f2fs_submit_discard_endio;
 783                         bio->bi_opf |= REQ_SYNC;
 784                         submit_bio(bio);
 785                         list_move_tail(&dc->list, &dcc->wait_list);
 786                 }
 787         } else {
 788                 __remove_discard_cmd(sbi, dc);
 789         }
 790 }
 791
 792 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
 793                                 struct block_device *bdev, block_t lstart,
 794                                 block_t start, block_t len,
 795                                 struct rb_node **insert_p,
 796                                 struct rb_node *insert_parent)
 797 {
 798         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 799         struct rb_node **p = &dcc->root.rb_node;
 800         struct rb_node *parent = NULL;
 801         struct discard_cmd *dc = NULL;
 802
 803         if (insert_p && insert_parent) {
 804                 parent = insert_parent;
 805                 p = insert_p;
 806                 goto do_insert;
 807         }
 808
 809         p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
 810 do_insert:
 811         dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
 812         if (!dc)
 813                 return NULL;
 814
 815         return dc;
 816 }
 817
 818 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
 819                                                 struct discard_cmd *dc)
 820 {
 821         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
 822 }
 823
 824 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
 825                                 struct discard_cmd *dc, block_t blkaddr)
 826 {
 827         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 828         struct discard_info di = dc->di;
 829         bool modified = false;
 830
 831         if (dc->state == D_DONE || dc->len == 1) {
 832                 __remove_discard_cmd(sbi, dc);
 833                 return;
 834         }
 835
 836         dcc->undiscard_blks -= di.len;
 837
 838         if (blkaddr > di.lstart) {
 839                 dc->len = blkaddr - dc->lstart;
 840                 dcc->undiscard_blks += dc->len;
 841                 __relocate_discard_cmd(dcc, dc);
 842                 modified = true;
 843         }
 844
 845         if (blkaddr < di.lstart + di.len - 1) {
 846                 if (modified) {
 847                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
 848                                         di.start + blkaddr + 1 - di.lstart,
 849                                         di.lstart + di.len - 1 - blkaddr,
 850                                         NULL, NULL);
 851                 } else {
 852                         dc->lstart++;
 853                         dc->len--;
 854                         dc->start++;
 855                         dcc->undiscard_blks += dc->len;
 856                         __relocate_discard_cmd(dcc, dc);
 857                 }
 858         }
 859 }
 860
 861 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
 862                                 struct block_device *bdev, block_t lstart,
 863                                 block_t start, block_t len)
 864 {
 865         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 866         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
 867         struct discard_cmd *dc;
 868         struct discard_info di = {0};
 869         struct rb_node **insert_p = NULL, *insert_parent = NULL;
 870         block_t end = lstart + len;
 871
 872         mutex_lock(&dcc->cmd_lock);
 873
 874         dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
 875                                         NULL, lstart,
 876                                         (struct rb_entry **)&prev_dc,
 877                                         (struct rb_entry **)&next_dc,
 878                                         &insert_p, &insert_parent, true);
 879         if (dc)
 880                 prev_dc = dc;
 881
 882         if (!prev_dc) {
 883                 di.lstart = lstart;
 884                 di.len = next_dc ? next_dc->lstart - lstart : len;
 885                 di.len = min(di.len, len);
 886                 di.start = start;
 887         }
 888
 889         while (1) {
 890                 struct rb_node *node;
 891                 bool merged = false;
 892                 struct discard_cmd *tdc = NULL;
 893
 894                 if (prev_dc) {
 895                         di.lstart = prev_dc->lstart + prev_dc->len;
 896                         if (di.lstart < lstart)
 897                                 di.lstart = lstart;
 898                         if (di.lstart >= end)
 899                                 break;
 900
 901                         if (!next_dc || next_dc->lstart > end)
 902                                 di.len = end - di.lstart;
 903                         else
 904                                 di.len = next_dc->lstart - di.lstart;
 905                         di.start = start + di.lstart - lstart;
 906                 }
 907
 908                 if (!di.len)
 909                         goto next;
 910
 911                 if (prev_dc && prev_dc->state == D_PREP &&
 912                         prev_dc->bdev == bdev &&
 913                         __is_discard_back_mergeable(&di, &prev_dc->di)) {
 914                         prev_dc->di.len += di.len;
 915                         dcc->undiscard_blks += di.len;
 916                         __relocate_discard_cmd(dcc, prev_dc);
 917                         di = prev_dc->di;
 918                         tdc = prev_dc;
 919                         merged = true;
 920                 }
 921
 922                 if (next_dc && next_dc->state == D_PREP &&
 923                         next_dc->bdev == bdev &&
 924                         __is_discard_front_mergeable(&di, &next_dc->di)) {
 925                         next_dc->di.lstart = di.lstart;
 926                         next_dc->di.len += di.len;
 927                         next_dc->di.start = di.start;
 928                         dcc->undiscard_blks += di.len;
 929                         __relocate_discard_cmd(dcc, next_dc);
 930                         if (tdc)
 931                                 __remove_discard_cmd(sbi, tdc);
 932                         merged = true;
 933                 }
 934
 935                 if (!merged) {
 936                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
 937                                                         di.len, NULL, NULL);
 938                 }
 939  next:
 940                 prev_dc = next_dc;
 941                 if (!prev_dc)
 942                         break;
 943
 944                 node = rb_next(&prev_dc->rb_node);
 945                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
 946         }
 947
 948         mutex_unlock(&dcc->cmd_lock);
 949 }
 950
 951 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
 952                 struct block_device *bdev, block_t blkstart, block_t blklen)
 953 {
 954         block_t lblkstart = blkstart;
 955
 956         trace_f2fs_queue_discard(bdev, blkstart, blklen);
 957
 958         if (sbi->s_ndevs) {
 959                 int devi = f2fs_target_device_index(sbi, blkstart);
 960
 961                 blkstart -= FDEV(devi).start_blk;
 962         }
 963         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
 964         return 0;
 965 }
 966
 967 static void __issue_discard_cmd(struct f2fs_sb_info *sbi, bool issue_cond)
 968 {
 969         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 970         struct list_head *pend_list;
 971         struct discard_cmd *dc, *tmp;
 972         struct blk_plug plug;
 973         int i, iter = 0;
 974
 975         mutex_lock(&dcc->cmd_lock);
 976         f2fs_bug_on(sbi,
 977                 !__check_rb_tree_consistence(sbi, &dcc->root));
 978         blk_start_plug(&plug);
 979         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
 980                 pend_list = &dcc->pend_list[i];
 981                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
 982                         f2fs_bug_on(sbi, dc->state != D_PREP);
 983
 984                         if (!issue_cond || is_idle(sbi))
 985                                 __submit_discard_cmd(sbi, dc);
 986                         if (issue_cond && iter++ > DISCARD_ISSUE_RATE)
 987                                 goto out;
 988                 }
 989         }
 990 out:
 991         blk_finish_plug(&plug);
 992         mutex_unlock(&dcc->cmd_lock);
 993 }
 994
 995 static void __wait_discard_cmd(struct f2fs_sb_info *sbi, bool wait_cond)
 996 {
 997         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 998         struct list_head *wait_list = &(dcc->wait_list);
 999         struct discard_cmd *dc, *tmp;
1000         bool need_wait;
1001
1002 next:
1003         need_wait = false;
1004
1005         mutex_lock(&dcc->cmd_lock);
1006         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1007                 if (!wait_cond || (dc->state == D_DONE && !dc->ref)) {
1008                         wait_for_completion_io(&dc->wait);
1009                         __remove_discard_cmd(sbi, dc);
1010                 } else {
1011                         dc->ref++;
1012                         need_wait = true;
1013                         break;
1014                 }
1015         }
1016         mutex_unlock(&dcc->cmd_lock);
1017
1018         if (need_wait) {
1019                 wait_for_completion_io(&dc->wait);
1020                 mutex_lock(&dcc->cmd_lock);
1021                 f2fs_bug_on(sbi, dc->state != D_DONE);
1022                 dc->ref--;
1023                 if (!dc->ref)
1024                         __remove_discard_cmd(sbi, dc);
1025                 mutex_unlock(&dcc->cmd_lock);
1026                 goto next;
1027         }
1028 }
1029
1030 /* This should be covered by global mutex, &sit_i->sentry_lock */
1031 void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1032 {
1033         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1034         struct discard_cmd *dc;
1035         bool need_wait = false;
1036
1037         mutex_lock(&dcc->cmd_lock);
1038         dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1039         if (dc) {
1040                 if (dc->state == D_PREP) {
1041                         __punch_discard_cmd(sbi, dc, blkaddr);
1042                 } else {
1043                         dc->ref++;
1044                         need_wait = true;
1045                 }
1046         }
1047         mutex_unlock(&dcc->cmd_lock);
1048
1049         if (need_wait) {
1050                 wait_for_completion_io(&dc->wait);
1051                 mutex_lock(&dcc->cmd_lock);
1052                 f2fs_bug_on(sbi, dc->state != D_DONE);
1053                 dc->ref--;
1054                 if (!dc->ref)
1055                         __remove_discard_cmd(sbi, dc);
1056                 mutex_unlock(&dcc->cmd_lock);
1057         }
1058 }
1059
1060 /* This comes from f2fs_put_super */
1061 void f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1062 {
1063         __issue_discard_cmd(sbi, false);
1064         __wait_discard_cmd(sbi, false);
1065 }
1066
1067 static int issue_discard_thread(void *data)
1068 {
1069         struct f2fs_sb_info *sbi = data;
1070         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1071         wait_queue_head_t *q = &dcc->discard_wait_queue;
1072
1073         set_freezable();
1074
1075         do {
1076                 wait_event_interruptible(*q, kthread_should_stop() ||
1077                                         freezing(current) ||
1078                                         atomic_read(&dcc->discard_cmd_cnt));
1079                 if (try_to_freeze())
1080                         continue;
1081                 if (kthread_should_stop())
1082                         return 0;
1083
1084                 __issue_discard_cmd(sbi, true);
1085                 __wait_discard_cmd(sbi, true);
1086
1087                 congestion_wait(BLK_RW_SYNC, HZ/50);
1088         } while (!kthread_should_stop());
1089         return 0;
1090 }
1091
1092 #ifdef CONFIG_BLK_DEV_ZONED
1093 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1094                 struct block_device *bdev, block_t blkstart, block_t blklen)
1095 {
1096         sector_t sector, nr_sects;
1097         block_t lblkstart = blkstart;
1098         int devi = 0;
1099
1100         if (sbi->s_ndevs) {
1101                 devi = f2fs_target_device_index(sbi, blkstart);
1102                 blkstart -= FDEV(devi).start_blk;
1103         }
1104
1105         /*
1106          * We need to know the type of the zone: for conventional zones,
1107          * use regular discard if the drive supports it. For sequential
1108          * zones, reset the zone write pointer.
1109          */
1110         switch (get_blkz_type(sbi, bdev, blkstart)) {
1111
1112         case BLK_ZONE_TYPE_CONVENTIONAL:
1113                 if (!blk_queue_discard(bdev_get_queue(bdev)))
1114                         return 0;
1115                 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1116         case BLK_ZONE_TYPE_SEQWRITE_REQ:
1117         case BLK_ZONE_TYPE_SEQWRITE_PREF:
1118                 sector = SECTOR_FROM_BLOCK(blkstart);
1119                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1120
1121                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1122                                 nr_sects != bdev_zone_sectors(bdev)) {
1123                         f2fs_msg(sbi->sb, KERN_INFO,
1124                                 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1125                                 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1126                                 blkstart, blklen);
1127                         return -EIO;
1128                 }
1129                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1130                 return blkdev_reset_zones(bdev, sector,
1131                                           nr_sects, GFP_NOFS);
1132         default:
1133                 /* Unknown zone type: broken device ? */
1134                 return -EIO;
1135         }
1136 }
1137 #endif
1138
1139 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1140                 struct block_device *bdev, block_t blkstart, block_t blklen)
1141 {
1142 #ifdef CONFIG_BLK_DEV_ZONED
1143         if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
1144                                 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1145                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1146 #endif
1147         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1148 }
1149
1150 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1151                                 block_t blkstart, block_t blklen)
1152 {
1153         sector_t start = blkstart, len = 0;
1154         struct block_device *bdev;
1155         struct seg_entry *se;
1156         unsigned int offset;
1157         block_t i;
1158         int err = 0;
1159
1160         bdev = f2fs_target_device(sbi, blkstart, NULL);
1161
1162         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1163                 if (i != start) {
1164                         struct block_device *bdev2 =
1165                                 f2fs_target_device(sbi, i, NULL);
1166
1167                         if (bdev2 != bdev) {
1168                                 err = __issue_discard_async(sbi, bdev,
1169                                                 start, len);
1170                                 if (err)
1171                                         return err;
1172                                 bdev = bdev2;
1173                                 start = i;
1174                                 len = 0;
1175                         }
1176                 }
1177
1178                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1179                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1180
1181                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1182                         sbi->discard_blks--;
1183         }
1184
1185         if (len)
1186                 err = __issue_discard_async(sbi, bdev, start, len);
1187         return err;
1188 }
1189
1190 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1191                                                         bool check_only)
1192 {
1193         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1194         int max_blocks = sbi->blocks_per_seg;
1195         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1196         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1197         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1198         unsigned long *discard_map = (unsigned long *)se->discard_map;
1199         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1200         unsigned int start = 0, end = -1;
1201         bool force = (cpc->reason & CP_DISCARD);
1202         struct discard_entry *de = NULL;
1203         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1204         int i;
1205
1206         if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
1207                 return false;
1208
1209         if (!force) {
1210                 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
1211                         SM_I(sbi)->dcc_info->nr_discards >=
1212                                 SM_I(sbi)->dcc_info->max_discards)
1213                         return false;
1214         }
1215
1216         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1217         for (i = 0; i < entries; i++)
1218                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1219                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1220
1221         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1222                                 SM_I(sbi)->dcc_info->max_discards) {
1223                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1224                 if (start >= max_blocks)
1225                         break;
1226
1227                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1228                 if (force && start && end != max_blocks
1229                                         && (end - start) < cpc->trim_minlen)
1230                         continue;
1231
1232                 if (check_only)
1233                         return true;
1234
1235                 if (!de) {
1236                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1237                                                                 GFP_F2FS_ZERO);
1238                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1239                         list_add_tail(&de->list, head);
1240                 }
1241
1242                 for (i = start; i < end; i++)
1243                         __set_bit_le(i, (void *)de->discard_map);
1244
1245                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1246         }
1247         return false;
1248 }
1249
1250 void release_discard_addrs(struct f2fs_sb_info *sbi)
1251 {
1252         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1253         struct discard_entry *entry, *this;
1254
1255         /* drop caches */
1256         list_for_each_entry_safe(entry, this, head, list) {
1257                 list_del(&entry->list);
1258                 kmem_cache_free(discard_entry_slab, entry);
1259         }
1260 }
1261
1262 /*
1263  * Should call clear_prefree_segments after checkpoint is done.
1264  */
1265 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1266 {
1267         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1268         unsigned int segno;
1269
1270         mutex_lock(&dirty_i->seglist_lock);
1271         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1272                 __set_test_and_free(sbi, segno);
1273         mutex_unlock(&dirty_i->seglist_lock);
1274 }
1275
1276 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1277 {
1278         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1279         struct discard_entry *entry, *this;
1280         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1281         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1282         unsigned int start = 0, end = -1;
1283         unsigned int secno, start_segno;
1284         bool force = (cpc->reason & CP_DISCARD);
1285
1286         mutex_lock(&dirty_i->seglist_lock);
1287
1288         while (1) {
1289                 int i;
1290                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1291                 if (start >= MAIN_SEGS(sbi))
1292                         break;
1293                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1294                                                                 start + 1);
1295
1296                 for (i = start; i < end; i++)
1297                         clear_bit(i, prefree_map);
1298
1299                 dirty_i->nr_dirty[PRE] -= end - start;
1300
1301                 if (!test_opt(sbi, DISCARD))
1302                         continue;
1303
1304                 if (force && start >= cpc->trim_start &&
1305                                         (end - 1) <= cpc->trim_end)
1306                                 continue;
1307
1308                 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1309                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1310                                 (end - start) << sbi->log_blocks_per_seg);
1311                         continue;
1312                 }
1313 next:
1314                 secno = GET_SEC_FROM_SEG(sbi, start);
1315                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1316                 if (!IS_CURSEC(sbi, secno) &&
1317                         !get_valid_blocks(sbi, start, true))
1318                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1319                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1320
1321                 start = start_segno + sbi->segs_per_sec;
1322                 if (start < end)
1323                         goto next;
1324                 else
1325                         end = start - 1;
1326         }
1327         mutex_unlock(&dirty_i->seglist_lock);
1328
1329         /* send small discards */
1330         list_for_each_entry_safe(entry, this, head, list) {
1331                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1332                 bool is_valid = test_bit_le(0, entry->discard_map);
1333
1334 find_next:
1335                 if (is_valid) {
1336                         next_pos = find_next_zero_bit_le(entry->discard_map,
1337                                         sbi->blocks_per_seg, cur_pos);
1338                         len = next_pos - cur_pos;
1339
1340                         if (f2fs_sb_mounted_blkzoned(sbi->sb) ||
1341                             (force && len < cpc->trim_minlen))
1342                                 goto skip;
1343
1344                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1345                                                                         len);
1346                         cpc->trimmed += len;
1347                         total_len += len;
1348                 } else {
1349                         next_pos = find_next_bit_le(entry->discard_map,
1350                                         sbi->blocks_per_seg, cur_pos);
1351                 }
1352 skip:
1353                 cur_pos = next_pos;
1354                 is_valid = !is_valid;
1355
1356                 if (cur_pos < sbi->blocks_per_seg)
1357                         goto find_next;
1358
1359                 list_del(&entry->list);
1360                 SM_I(sbi)->dcc_info->nr_discards -= total_len;
1361                 kmem_cache_free(discard_entry_slab, entry);
1362         }
1363
1364         wake_up(&SM_I(sbi)->dcc_info->discard_wait_queue);
1365 }
1366
1367 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1368 {
1369         dev_t dev = sbi->sb->s_bdev->bd_dev;
1370         struct discard_cmd_control *dcc;
1371         int err = 0, i;
1372
1373         if (SM_I(sbi)->dcc_info) {
1374                 dcc = SM_I(sbi)->dcc_info;
1375                 goto init_thread;
1376         }
1377
1378         dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1379         if (!dcc)
1380                 return -ENOMEM;
1381
1382         INIT_LIST_HEAD(&dcc->entry_list);
1383         for (i = 0; i < MAX_PLIST_NUM; i++)
1384                 INIT_LIST_HEAD(&dcc->pend_list[i]);
1385         INIT_LIST_HEAD(&dcc->wait_list);
1386         mutex_init(&dcc->cmd_lock);
1387         atomic_set(&dcc->issued_discard, 0);
1388         atomic_set(&dcc->issing_discard, 0);
1389         atomic_set(&dcc->discard_cmd_cnt, 0);
1390         dcc->nr_discards = 0;
1391         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1392         dcc->undiscard_blks = 0;
1393         dcc->root = RB_ROOT;
1394
1395         init_waitqueue_head(&dcc->discard_wait_queue);
1396         SM_I(sbi)->dcc_info = dcc;
1397 init_thread:
1398         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1399                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1400         if (IS_ERR(dcc->f2fs_issue_discard)) {
1401                 err = PTR_ERR(dcc->f2fs_issue_discard);
1402                 kfree(dcc);
1403                 SM_I(sbi)->dcc_info = NULL;
1404                 return err;
1405         }
1406
1407         return err;
1408 }
1409
1410 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1411 {
1412         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1413
1414         if (!dcc)
1415                 return;
1416
1417         if (dcc->f2fs_issue_discard) {
1418                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1419
1420                 dcc->f2fs_issue_discard = NULL;
1421                 kthread_stop(discard_thread);
1422         }
1423
1424         kfree(dcc);
1425         SM_I(sbi)->dcc_info = NULL;
1426 }
1427
1428 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1429 {
1430         struct sit_info *sit_i = SIT_I(sbi);
1431
1432         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1433                 sit_i->dirty_sentries++;
1434                 return false;
1435         }
1436
1437         return true;
1438 }
1439
1440 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1441                                         unsigned int segno, int modified)
1442 {
1443         struct seg_entry *se = get_seg_entry(sbi, segno);
1444         se->type = type;
1445         if (modified)
1446                 __mark_sit_entry_dirty(sbi, segno);
1447 }
1448
1449 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1450 {
1451         struct seg_entry *se;
1452         unsigned int segno, offset;
1453         long int new_vblocks;
1454
1455         segno = GET_SEGNO(sbi, blkaddr);
1456
1457         se = get_seg_entry(sbi, segno);
1458         new_vblocks = se->valid_blocks + del;
1459         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1460
1461         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
1462                                 (new_vblocks > sbi->blocks_per_seg)));
1463
1464         se->valid_blocks = new_vblocks;
1465         se->mtime = get_mtime(sbi);
1466         SIT_I(sbi)->max_mtime = se->mtime;
1467
1468         /* Update valid block bitmap */
1469         if (del > 0) {
1470                 if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) {
1471 #ifdef CONFIG_F2FS_CHECK_FS
1472                         if (f2fs_test_and_set_bit(offset,
1473                                                 se->cur_valid_map_mir))
1474                                 f2fs_bug_on(sbi, 1);
1475                         else
1476                                 WARN_ON(1);
1477 #else
1478                         f2fs_bug_on(sbi, 1);
1479 #endif
1480                 }
1481                 if (f2fs_discard_en(sbi) &&
1482                         !f2fs_test_and_set_bit(offset, se->discard_map))
1483                         sbi->discard_blks--;
1484
1485                 /* don't overwrite by SSR to keep node chain */
1486                 if (se->type == CURSEG_WARM_NODE) {
1487                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1488                                 se->ckpt_valid_blocks++;
1489                 }
1490         } else {
1491                 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) {
1492 #ifdef CONFIG_F2FS_CHECK_FS
1493                         if (!f2fs_test_and_clear_bit(offset,
1494                                                 se->cur_valid_map_mir))
1495                                 f2fs_bug_on(sbi, 1);
1496                         else
1497                                 WARN_ON(1);
1498 #else
1499                         f2fs_bug_on(sbi, 1);
1500 #endif
1501                 }
1502                 if (f2fs_discard_en(sbi) &&
1503                         f2fs_test_and_clear_bit(offset, se->discard_map))
1504                         sbi->discard_blks++;
1505         }
1506         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1507                 se->ckpt_valid_blocks += del;
1508
1509         __mark_sit_entry_dirty(sbi, segno);
1510
1511         /* update total number of valid blocks to be written in ckpt area */
1512         SIT_I(sbi)->written_valid_blocks += del;
1513
1514         if (sbi->segs_per_sec > 1)
1515                 get_sec_entry(sbi, segno)->valid_blocks += del;
1516 }
1517
1518 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
1519 {
1520         update_sit_entry(sbi, new, 1);
1521         if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1522                 update_sit_entry(sbi, old, -1);
1523
1524         locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1525         locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
1526 }
1527
1528 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1529 {
1530         unsigned int segno = GET_SEGNO(sbi, addr);
1531         struct sit_info *sit_i = SIT_I(sbi);
1532
1533         f2fs_bug_on(sbi, addr == NULL_ADDR);
1534         if (addr == NEW_ADDR)
1535                 return;
1536
1537         /* add it into sit main buffer */
1538         mutex_lock(&sit_i->sentry_lock);
1539
1540         update_sit_entry(sbi, addr, -1);
1541
1542         /* add it into dirty seglist */
1543         locate_dirty_segment(sbi, segno);
1544
1545         mutex_unlock(&sit_i->sentry_lock);
1546 }
1547
1548 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1549 {
1550         struct sit_info *sit_i = SIT_I(sbi);
1551         unsigned int segno, offset;
1552         struct seg_entry *se;
1553         bool is_cp = false;
1554
1555         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1556                 return true;
1557
1558         mutex_lock(&sit_i->sentry_lock);
1559
1560         segno = GET_SEGNO(sbi, blkaddr);
1561         se = get_seg_entry(sbi, segno);
1562         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1563
1564         if (f2fs_test_bit(offset, se->ckpt_valid_map))
1565                 is_cp = true;
1566
1567         mutex_unlock(&sit_i->sentry_lock);
1568
1569         return is_cp;
1570 }
1571
1572 /*
1573  * This function should be resided under the curseg_mutex lock
1574  */
1575 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1576                                         struct f2fs_summary *sum)
1577 {
1578         struct curseg_info *curseg = CURSEG_I(sbi, type);
1579         void *addr = curseg->sum_blk;
1580         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1581         memcpy(addr, sum, sizeof(struct f2fs_summary));
1582 }
1583
1584 /*
1585  * Calculate the number of current summary pages for writing
1586  */
1587 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1588 {
1589         int valid_sum_count = 0;
1590         int i, sum_in_page;
1591
1592         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1593                 if (sbi->ckpt->alloc_type[i] == SSR)
1594                         valid_sum_count += sbi->blocks_per_seg;
1595                 else {
1596                         if (for_ra)
1597                                 valid_sum_count += le16_to_cpu(
1598                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1599                         else
1600                                 valid_sum_count += curseg_blkoff(sbi, i);
1601                 }
1602         }
1603
1604         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1605                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1606         if (valid_sum_count <= sum_in_page)
1607                 return 1;
1608         else if ((valid_sum_count - sum_in_page) <=
1609                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1610                 return 2;
1611         return 3;
1612 }
1613
1614 /*
1615  * Caller should put this summary page
1616  */
1617 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1618 {
1619         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1620 }
1621
1622 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1623 {
1624         struct page *page = grab_meta_page(sbi, blk_addr);
1625         void *dst = page_address(page);
1626
1627         if (src)
1628                 memcpy(dst, src, PAGE_SIZE);
1629         else
1630                 memset(dst, 0, PAGE_SIZE);
1631         set_page_dirty(page);
1632         f2fs_put_page(page, 1);
1633 }
1634
1635 static void write_sum_page(struct f2fs_sb_info *sbi,
1636                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
1637 {
1638         update_meta_page(sbi, (void *)sum_blk, blk_addr);
1639 }
1640
1641 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1642                                                 int type, block_t blk_addr)
1643 {
1644         struct curseg_info *curseg = CURSEG_I(sbi, type);
1645         struct page *page = grab_meta_page(sbi, blk_addr);
1646         struct f2fs_summary_block *src = curseg->sum_blk;
1647         struct f2fs_summary_block *dst;
1648
1649         dst = (struct f2fs_summary_block *)page_address(page);
1650
1651         mutex_lock(&curseg->curseg_mutex);
1652
1653         down_read(&curseg->journal_rwsem);
1654         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1655         up_read(&curseg->journal_rwsem);
1656
1657         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1658         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1659
1660         mutex_unlock(&curseg->curseg_mutex);
1661
1662         set_page_dirty(page);
1663         f2fs_put_page(page, 1);
1664 }
1665
1666 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1667 {
1668         struct curseg_info *curseg = CURSEG_I(sbi, type);
1669         unsigned int segno = curseg->segno + 1;
1670         struct free_segmap_info *free_i = FREE_I(sbi);
1671
1672         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
1673                 return !test_bit(segno, free_i->free_segmap);
1674         return 0;
1675 }
1676
1677 /*
1678  * Find a new segment from the free segments bitmap to right order
1679  * This function should be returned with success, otherwise BUG
1680  */
1681 static void get_new_segment(struct f2fs_sb_info *sbi,
1682                         unsigned int *newseg, bool new_sec, int dir)
1683 {
1684         struct free_segmap_info *free_i = FREE_I(sbi);
1685         unsigned int segno, secno, zoneno;
1686         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
1687         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
1688         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
1689         unsigned int left_start = hint;
1690         bool init = true;
1691         int go_left = 0;
1692         int i;
1693
1694         spin_lock(&free_i->segmap_lock);
1695
1696         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1697                 segno = find_next_zero_bit(free_i->free_segmap,
1698                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
1699                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
1700                         goto got_it;
1701         }
1702 find_other_zone:
1703         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1704         if (secno >= MAIN_SECS(sbi)) {
1705                 if (dir == ALLOC_RIGHT) {
1706                         secno = find_next_zero_bit(free_i->free_secmap,
1707                                                         MAIN_SECS(sbi), 0);
1708                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1709                 } else {
1710                         go_left = 1;
1711                         left_start = hint - 1;
1712                 }
1713         }
1714         if (go_left == 0)
1715                 goto skip_left;
1716
1717         while (test_bit(left_start, free_i->free_secmap)) {
1718                 if (left_start > 0) {
1719                         left_start--;
1720                         continue;
1721                 }
1722                 left_start = find_next_zero_bit(free_i->free_secmap,
1723                                                         MAIN_SECS(sbi), 0);
1724                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1725                 break;
1726         }
1727         secno = left_start;
1728 skip_left:
1729         hint = secno;
1730         segno = GET_SEG_FROM_SEC(sbi, secno);
1731         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
1732
1733         /* give up on finding another zone */
1734         if (!init)
1735                 goto got_it;
1736         if (sbi->secs_per_zone == 1)
1737                 goto got_it;
1738         if (zoneno == old_zoneno)
1739                 goto got_it;
1740         if (dir == ALLOC_LEFT) {
1741                 if (!go_left && zoneno + 1 >= total_zones)
1742                         goto got_it;
1743                 if (go_left && zoneno == 0)
1744                         goto got_it;
1745         }
1746         for (i = 0; i < NR_CURSEG_TYPE; i++)
1747                 if (CURSEG_I(sbi, i)->zone == zoneno)
1748                         break;
1749
1750         if (i < NR_CURSEG_TYPE) {
1751                 /* zone is in user, try another */
1752                 if (go_left)
1753                         hint = zoneno * sbi->secs_per_zone - 1;
1754                 else if (zoneno + 1 >= total_zones)
1755                         hint = 0;
1756                 else
1757                         hint = (zoneno + 1) * sbi->secs_per_zone;
1758                 init = false;
1759                 goto find_other_zone;
1760         }
1761 got_it:
1762         /* set it as dirty segment in free segmap */
1763         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1764         __set_inuse(sbi, segno);
1765         *newseg = segno;
1766         spin_unlock(&free_i->segmap_lock);
1767 }
1768
1769 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1770 {
1771         struct curseg_info *curseg = CURSEG_I(sbi, type);
1772         struct summary_footer *sum_footer;
1773
1774         curseg->segno = curseg->next_segno;
1775         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
1776         curseg->next_blkoff = 0;
1777         curseg->next_segno = NULL_SEGNO;
1778
1779         sum_footer = &(curseg->sum_blk->footer);
1780         memset(sum_footer, 0, sizeof(struct summary_footer));
1781         if (IS_DATASEG(type))
1782                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1783         if (IS_NODESEG(type))
1784                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1785         __set_sit_entry_type(sbi, type, curseg->segno, modified);
1786 }
1787
1788 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
1789 {
1790         /* if segs_per_sec is large than 1, we need to keep original policy. */
1791         if (sbi->segs_per_sec != 1)
1792                 return CURSEG_I(sbi, type)->segno;
1793
1794         if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
1795                 return 0;
1796
1797         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
1798                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
1799         return CURSEG_I(sbi, type)->segno;
1800 }
1801
1802 /*
1803  * Allocate a current working segment.
1804  * This function always allocates a free segment in LFS manner.
1805  */
1806 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1807 {
1808         struct curseg_info *curseg = CURSEG_I(sbi, type);
1809         unsigned int segno = curseg->segno;
1810         int dir = ALLOC_LEFT;
1811
1812         write_sum_page(sbi, curseg->sum_blk,
1813                                 GET_SUM_BLOCK(sbi, segno));
1814         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1815                 dir = ALLOC_RIGHT;
1816
1817         if (test_opt(sbi, NOHEAP))
1818                 dir = ALLOC_RIGHT;
1819
1820         segno = __get_next_segno(sbi, type);
1821         get_new_segment(sbi, &segno, new_sec, dir);
1822         curseg->next_segno = segno;
1823         reset_curseg(sbi, type, 1);
1824         curseg->alloc_type = LFS;
1825 }
1826
1827 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1828                         struct curseg_info *seg, block_t start)
1829 {
1830         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1831         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1832         unsigned long *target_map = SIT_I(sbi)->tmp_map;
1833         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1834         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1835         int i, pos;
1836
1837         for (i = 0; i < entries; i++)
1838                 target_map[i] = ckpt_map[i] | cur_map[i];
1839
1840         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1841
1842         seg->next_blkoff = pos;
1843 }
1844
1845 /*
1846  * If a segment is written by LFS manner, next block offset is just obtained
1847  * by increasing the current block offset. However, if a segment is written by
1848  * SSR manner, next block offset obtained by calling __next_free_blkoff
1849  */
1850 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1851                                 struct curseg_info *seg)
1852 {
1853         if (seg->alloc_type == SSR)
1854                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1855         else
1856                 seg->next_blkoff++;
1857 }
1858
1859 /*
1860  * This function always allocates a used segment(from dirty seglist) by SSR
1861  * manner, so it should recover the existing segment information of valid blocks
1862  */
1863 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1864 {
1865         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1866         struct curseg_info *curseg = CURSEG_I(sbi, type);
1867         unsigned int new_segno = curseg->next_segno;
1868         struct f2fs_summary_block *sum_node;
1869         struct page *sum_page;
1870
1871         write_sum_page(sbi, curseg->sum_blk,
1872                                 GET_SUM_BLOCK(sbi, curseg->segno));
1873         __set_test_and_inuse(sbi, new_segno);
1874
1875         mutex_lock(&dirty_i->seglist_lock);
1876         __remove_dirty_segment(sbi, new_segno, PRE);
1877         __remove_dirty_segment(sbi, new_segno, DIRTY);
1878         mutex_unlock(&dirty_i->seglist_lock);
1879
1880         reset_curseg(sbi, type, 1);
1881         curseg->alloc_type = SSR;
1882         __next_free_blkoff(sbi, curseg, 0);
1883
1884         if (reuse) {
1885                 sum_page = get_sum_page(sbi, new_segno);
1886                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1887                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1888                 f2fs_put_page(sum_page, 1);
1889         }
1890 }
1891
1892 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1893 {
1894         struct curseg_info *curseg = CURSEG_I(sbi, type);
1895         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1896         unsigned segno = NULL_SEGNO;
1897         int i, cnt;
1898         bool reversed = false;
1899
1900         /* need_SSR() already forces to do this */
1901         if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
1902                 curseg->next_segno = segno;
1903                 return 1;
1904         }
1905
1906         /* For node segments, let's do SSR more intensively */
1907         if (IS_NODESEG(type)) {
1908                 if (type >= CURSEG_WARM_NODE) {
1909                         reversed = true;
1910                         i = CURSEG_COLD_NODE;
1911                 } else {
1912                         i = CURSEG_HOT_NODE;
1913                 }
1914                 cnt = NR_CURSEG_NODE_TYPE;
1915         } else {
1916                 if (type >= CURSEG_WARM_DATA) {
1917                         reversed = true;
1918                         i = CURSEG_COLD_DATA;
1919                 } else {
1920                         i = CURSEG_HOT_DATA;
1921                 }
1922                 cnt = NR_CURSEG_DATA_TYPE;
1923         }
1924
1925         for (; cnt-- > 0; reversed ? i-- : i++) {
1926                 if (i == type)
1927                         continue;
1928                 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
1929                         curseg->next_segno = segno;
1930                         return 1;
1931                 }
1932         }
1933         return 0;
1934 }
1935
1936 /*
1937  * flush out current segment and replace it with new segment
1938  * This function should be returned with success, otherwise BUG
1939  */
1940 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1941                                                 int type, bool force)
1942 {
1943         struct curseg_info *curseg = CURSEG_I(sbi, type);
1944
1945         if (force)
1946                 new_curseg(sbi, type, true);
1947         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
1948                                         type == CURSEG_WARM_NODE)
1949                 new_curseg(sbi, type, false);
1950         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1951                 new_curseg(sbi, type, false);
1952         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1953                 change_curseg(sbi, type, true);
1954         else
1955                 new_curseg(sbi, type, false);
1956
1957         stat_inc_seg_type(sbi, curseg);
1958 }
1959
1960 void allocate_new_segments(struct f2fs_sb_info *sbi)
1961 {
1962         struct curseg_info *curseg;
1963         unsigned int old_segno;
1964         int i;
1965
1966         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1967                 curseg = CURSEG_I(sbi, i);
1968                 old_segno = curseg->segno;
1969                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
1970                 locate_dirty_segment(sbi, old_segno);
1971         }
1972 }
1973
1974 static const struct segment_allocation default_salloc_ops = {
1975         .allocate_segment = allocate_segment_by_default,
1976 };
1977
1978 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1979 {
1980         __u64 trim_start = cpc->trim_start;
1981         bool has_candidate = false;
1982
1983         mutex_lock(&SIT_I(sbi)->sentry_lock);
1984         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
1985                 if (add_discard_addrs(sbi, cpc, true)) {
1986                         has_candidate = true;
1987                         break;
1988                 }
1989         }
1990         mutex_unlock(&SIT_I(sbi)->sentry_lock);
1991
1992         cpc->trim_start = trim_start;
1993         return has_candidate;
1994 }
1995
1996 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1997 {
1998         __u64 start = F2FS_BYTES_TO_BLK(range->start);
1999         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2000         unsigned int start_segno, end_segno;
2001         struct cp_control cpc;
2002         int err = 0;
2003
2004         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2005                 return -EINVAL;
2006
2007         cpc.trimmed = 0;
2008         if (end <= MAIN_BLKADDR(sbi))
2009                 goto out;
2010
2011         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2012                 f2fs_msg(sbi->sb, KERN_WARNING,
2013                         "Found FS corruption, run fsck to fix.");
2014                 goto out;
2015         }
2016
2017         /* start/end segment number in main_area */
2018         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2019         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2020                                                 GET_SEGNO(sbi, end);
2021         cpc.reason = CP_DISCARD;
2022         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2023
2024         /* do checkpoint to issue discard commands safely */
2025         for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
2026                 cpc.trim_start = start_segno;
2027
2028                 if (sbi->discard_blks == 0)
2029                         break;
2030                 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
2031                         cpc.trim_end = end_segno;
2032                 else
2033                         cpc.trim_end = min_t(unsigned int,
2034                                 rounddown(start_segno +
2035                                 BATCHED_TRIM_SEGMENTS(sbi),
2036                                 sbi->segs_per_sec) - 1, end_segno);
2037
2038                 mutex_lock(&sbi->gc_mutex);
2039                 err = write_checkpoint(sbi, &cpc);
2040                 mutex_unlock(&sbi->gc_mutex);
2041                 if (err)
2042                         break;
2043
2044                 schedule();
2045         }
2046 out:
2047         range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
2048         return err;
2049 }
2050
2051 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2052 {
2053         struct curseg_info *curseg = CURSEG_I(sbi, type);
2054         if (curseg->next_blkoff < sbi->blocks_per_seg)
2055                 return true;
2056         return false;
2057 }
2058
2059 static int __get_segment_type_2(struct f2fs_io_info *fio)
2060 {
2061         if (fio->type == DATA)
2062                 return CURSEG_HOT_DATA;
2063         else
2064                 return CURSEG_HOT_NODE;
2065 }
2066
2067 static int __get_segment_type_4(struct f2fs_io_info *fio)
2068 {
2069         if (fio->type == DATA) {
2070                 struct inode *inode = fio->page->mapping->host;
2071
2072                 if (S_ISDIR(inode->i_mode))
2073                         return CURSEG_HOT_DATA;
2074                 else
2075                         return CURSEG_COLD_DATA;
2076         } else {
2077                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2078                         return CURSEG_WARM_NODE;
2079                 else
2080                         return CURSEG_COLD_NODE;
2081         }
2082 }
2083
2084 static int __get_segment_type_6(struct f2fs_io_info *fio)
2085 {
2086         if (fio->type == DATA) {
2087                 struct inode *inode = fio->page->mapping->host;
2088
2089                 if (is_cold_data(fio->page) || file_is_cold(inode))
2090                         return CURSEG_COLD_DATA;
2091                 if (is_inode_flag_set(inode, FI_HOT_DATA))
2092                         return CURSEG_HOT_DATA;
2093                 return CURSEG_WARM_DATA;
2094         } else {
2095                 if (IS_DNODE(fio->page))
2096                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2097                                                 CURSEG_HOT_NODE;
2098                 return CURSEG_COLD_NODE;
2099         }
2100 }
2101
2102 static int __get_segment_type(struct f2fs_io_info *fio)
2103 {
2104         int type = 0;
2105
2106         switch (fio->sbi->active_logs) {
2107         case 2:
2108                 type = __get_segment_type_2(fio);
2109                 break;
2110         case 4:
2111                 type = __get_segment_type_4(fio);
2112                 break;
2113         case 6:
2114                 type = __get_segment_type_6(fio);
2115                 break;
2116         default:
2117                 f2fs_bug_on(fio->sbi, true);
2118         }
2119
2120         if (IS_HOT(type))
2121                 fio->temp = HOT;
2122         else if (IS_WARM(type))
2123                 fio->temp = WARM;
2124         else
2125                 fio->temp = COLD;
2126         return type;
2127 }
2128
2129 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2130                 block_t old_blkaddr, block_t *new_blkaddr,
2131                 struct f2fs_summary *sum, int type,
2132                 struct f2fs_io_info *fio, bool add_list)
2133 {
2134         struct sit_info *sit_i = SIT_I(sbi);
2135         struct curseg_info *curseg = CURSEG_I(sbi, type);
2136
2137         mutex_lock(&curseg->curseg_mutex);
2138         mutex_lock(&sit_i->sentry_lock);
2139
2140         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2141
2142         f2fs_wait_discard_bio(sbi, *new_blkaddr);
2143
2144         /*
2145          * __add_sum_entry should be resided under the curseg_mutex
2146          * because, this function updates a summary entry in the
2147          * current summary block.
2148          */
2149         __add_sum_entry(sbi, type, sum);
2150
2151         __refresh_next_blkoff(sbi, curseg);
2152
2153         stat_inc_block_count(sbi, curseg);
2154
2155         if (!__has_curseg_space(sbi, type))
2156                 sit_i->s_ops->allocate_segment(sbi, type, false);
2157         /*
2158          * SIT information should be updated after segment allocation,
2159          * since we need to keep dirty segments precisely under SSR.
2160          */
2161         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
2162
2163         mutex_unlock(&sit_i->sentry_lock);
2164
2165         if (page && IS_NODESEG(type))
2166                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2167
2168         if (add_list) {
2169                 struct f2fs_bio_info *io;
2170
2171                 INIT_LIST_HEAD(&fio->list);
2172                 fio->in_list = true;
2173                 io = sbi->write_io[fio->type] + fio->temp;
2174                 spin_lock(&io->io_lock);
2175                 list_add_tail(&fio->list, &io->io_list);
2176                 spin_unlock(&io->io_lock);
2177         }
2178
2179         mutex_unlock(&curseg->curseg_mutex);
2180 }
2181
2182 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2183 {
2184         int type = __get_segment_type(fio);
2185         int err;
2186
2187 reallocate:
2188         allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
2189                         &fio->new_blkaddr, sum, type, fio, true);
2190
2191         /* writeout dirty page into bdev */
2192         err = f2fs_submit_page_write(fio);
2193         if (err == -EAGAIN) {
2194                 fio->old_blkaddr = fio->new_blkaddr;
2195                 goto reallocate;
2196         }
2197 }
2198
2199 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
2200 {
2201         struct f2fs_io_info fio = {
2202                 .sbi = sbi,
2203                 .type = META,
2204                 .op = REQ_OP_WRITE,
2205                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
2206                 .old_blkaddr = page->index,
2207                 .new_blkaddr = page->index,
2208                 .page = page,
2209                 .encrypted_page = NULL,
2210                 .in_list = false,
2211         };
2212
2213         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
2214                 fio.op_flags &= ~REQ_META;
2215
2216         set_page_writeback(page);
2217         f2fs_submit_page_write(&fio);
2218 }
2219
2220 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
2221 {
2222         struct f2fs_summary sum;
2223
2224         set_summary(&sum, nid, 0, 0);
2225         do_write_page(&sum, fio);
2226 }
2227
2228 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
2229 {
2230         struct f2fs_sb_info *sbi = fio->sbi;
2231         struct f2fs_summary sum;
2232         struct node_info ni;
2233
2234         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
2235         get_node_info(sbi, dn->nid, &ni);
2236         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
2237         do_write_page(&sum, fio);
2238         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
2239 }
2240
2241 int rewrite_data_page(struct f2fs_io_info *fio)
2242 {
2243         fio->new_blkaddr = fio->old_blkaddr;
2244         stat_inc_inplace_blocks(fio->sbi);
2245         return f2fs_submit_page_bio(fio);
2246 }
2247
2248 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
2249                                 block_t old_blkaddr, block_t new_blkaddr,
2250                                 bool recover_curseg, bool recover_newaddr)
2251 {
2252         struct sit_info *sit_i = SIT_I(sbi);
2253         struct curseg_info *curseg;
2254         unsigned int segno, old_cursegno;
2255         struct seg_entry *se;
2256         int type;
2257         unsigned short old_blkoff;
2258
2259         segno = GET_SEGNO(sbi, new_blkaddr);
2260         se = get_seg_entry(sbi, segno);
2261         type = se->type;
2262
2263         if (!recover_curseg) {
2264                 /* for recovery flow */
2265                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2266                         if (old_blkaddr == NULL_ADDR)
2267                                 type = CURSEG_COLD_DATA;
2268                         else
2269                                 type = CURSEG_WARM_DATA;
2270                 }
2271         } else {
2272                 if (!IS_CURSEG(sbi, segno))
2273                         type = CURSEG_WARM_DATA;
2274         }
2275
2276         curseg = CURSEG_I(sbi, type);
2277
2278         mutex_lock(&curseg->curseg_mutex);
2279         mutex_lock(&sit_i->sentry_lock);
2280
2281         old_cursegno = curseg->segno;
2282         old_blkoff = curseg->next_blkoff;
2283
2284         /* change the current segment */
2285         if (segno != curseg->segno) {
2286                 curseg->next_segno = segno;
2287                 change_curseg(sbi, type, true);
2288         }
2289
2290         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
2291         __add_sum_entry(sbi, type, sum);
2292
2293         if (!recover_curseg || recover_newaddr)
2294                 update_sit_entry(sbi, new_blkaddr, 1);
2295         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2296                 update_sit_entry(sbi, old_blkaddr, -1);
2297
2298         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2299         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2300
2301         locate_dirty_segment(sbi, old_cursegno);
2302
2303         if (recover_curseg) {
2304                 if (old_cursegno != curseg->segno) {
2305                         curseg->next_segno = old_cursegno;
2306                         change_curseg(sbi, type, true);
2307                 }
2308                 curseg->next_blkoff = old_blkoff;
2309         }
2310
2311         mutex_unlock(&sit_i->sentry_lock);
2312         mutex_unlock(&curseg->curseg_mutex);
2313 }
2314
2315 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2316                                 block_t old_addr, block_t new_addr,
2317                                 unsigned char version, bool recover_curseg,
2318                                 bool recover_newaddr)
2319 {
2320         struct f2fs_summary sum;
2321
2322         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2323
2324         __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2325                                         recover_curseg, recover_newaddr);
2326
2327         f2fs_update_data_blkaddr(dn, new_addr);
2328 }
2329
2330 void f2fs_wait_on_page_writeback(struct page *page,
2331                                 enum page_type type, bool ordered)
2332 {
2333         if (PageWriteback(page)) {
2334                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2335
2336                 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2337                                                 0, page->index, type);
2338                 if (ordered)
2339                         wait_on_page_writeback(page);
2340                 else
2341                         wait_for_stable_page(page);
2342         }
2343 }
2344
2345 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
2346                                                         block_t blkaddr)
2347 {
2348         struct page *cpage;
2349
2350         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
2351                 return;
2352
2353         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2354         if (cpage) {
2355                 f2fs_wait_on_page_writeback(cpage, DATA, true);
2356                 f2fs_put_page(cpage, 1);
2357         }
2358 }
2359
2360 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
2361 {
2362         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2363         struct curseg_info *seg_i;
2364         unsigned char *kaddr;
2365         struct page *page;
2366         block_t start;
2367         int i, j, offset;
2368
2369         start = start_sum_block(sbi);
2370
2371         page = get_meta_page(sbi, start++);
2372         kaddr = (unsigned char *)page_address(page);
2373
2374         /* Step 1: restore nat cache */
2375         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2376         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
2377
2378         /* Step 2: restore sit cache */
2379         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2380         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
2381         offset = 2 * SUM_JOURNAL_SIZE;
2382
2383         /* Step 3: restore summary entries */
2384         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2385                 unsigned short blk_off;
2386                 unsigned int segno;
2387
2388                 seg_i = CURSEG_I(sbi, i);
2389                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2390                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2391                 seg_i->next_segno = segno;
2392                 reset_curseg(sbi, i, 0);
2393                 seg_i->alloc_type = ckpt->alloc_type[i];
2394                 seg_i->next_blkoff = blk_off;
2395
2396                 if (seg_i->alloc_type == SSR)
2397                         blk_off = sbi->blocks_per_seg;
2398
2399                 for (j = 0; j < blk_off; j++) {
2400                         struct f2fs_summary *s;
2401                         s = (struct f2fs_summary *)(kaddr + offset);
2402                         seg_i->sum_blk->entries[j] = *s;
2403                         offset += SUMMARY_SIZE;
2404                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
2405                                                 SUM_FOOTER_SIZE)
2406                                 continue;
2407
2408                         f2fs_put_page(page, 1);
2409                         page = NULL;
2410
2411                         page = get_meta_page(sbi, start++);
2412                         kaddr = (unsigned char *)page_address(page);
2413                         offset = 0;
2414                 }
2415         }
2416         f2fs_put_page(page, 1);
2417         return 0;
2418 }
2419
2420 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2421 {
2422         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2423         struct f2fs_summary_block *sum;
2424         struct curseg_info *curseg;
2425         struct page *new;
2426         unsigned short blk_off;
2427         unsigned int segno = 0;
2428         block_t blk_addr = 0;
2429
2430         /* get segment number and block addr */
2431         if (IS_DATASEG(type)) {
2432                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2433                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2434                                                         CURSEG_HOT_DATA]);
2435                 if (__exist_node_summaries(sbi))
2436                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2437                 else
2438                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2439         } else {
2440                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2441                                                         CURSEG_HOT_NODE]);
2442                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2443                                                         CURSEG_HOT_NODE]);
2444                 if (__exist_node_summaries(sbi))
2445                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2446                                                         type - CURSEG_HOT_NODE);
2447                 else
2448                         blk_addr = GET_SUM_BLOCK(sbi, segno);
2449         }
2450
2451         new = get_meta_page(sbi, blk_addr);
2452         sum = (struct f2fs_summary_block *)page_address(new);
2453
2454         if (IS_NODESEG(type)) {
2455                 if (__exist_node_summaries(sbi)) {
2456                         struct f2fs_summary *ns = &sum->entries[0];
2457                         int i;
2458                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2459                                 ns->version = 0;
2460                                 ns->ofs_in_node = 0;
2461                         }
2462                 } else {
2463                         int err;
2464
2465                         err = restore_node_summary(sbi, segno, sum);
2466                         if (err) {
2467                                 f2fs_put_page(new, 1);
2468                                 return err;
2469                         }
2470                 }
2471         }
2472
2473         /* set uncompleted segment to curseg */
2474         curseg = CURSEG_I(sbi, type);
2475         mutex_lock(&curseg->curseg_mutex);
2476
2477         /* update journal info */
2478         down_write(&curseg->journal_rwsem);
2479         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2480         up_write(&curseg->journal_rwsem);
2481
2482         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2483         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
2484         curseg->next_segno = segno;
2485         reset_curseg(sbi, type, 0);
2486         curseg->alloc_type = ckpt->alloc_type[type];
2487         curseg->next_blkoff = blk_off;
2488         mutex_unlock(&curseg->curseg_mutex);
2489         f2fs_put_page(new, 1);
2490         return 0;
2491 }
2492
2493 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2494 {
2495         int type = CURSEG_HOT_DATA;
2496         int err;
2497
2498         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
2499                 int npages = npages_for_summary_flush(sbi, true);
2500
2501                 if (npages >= 2)
2502                         ra_meta_pages(sbi, start_sum_block(sbi), npages,
2503                                                         META_CP, true);
2504
2505                 /* restore for compacted data summary */
2506                 if (read_compacted_summaries(sbi))
2507                         return -EINVAL;
2508                 type = CURSEG_HOT_NODE;
2509         }
2510
2511         if (__exist_node_summaries(sbi))
2512                 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
2513                                         NR_CURSEG_TYPE - type, META_CP, true);
2514
2515         for (; type <= CURSEG_COLD_NODE; type++) {
2516                 err = read_normal_summaries(sbi, type);
2517                 if (err)
2518                         return err;
2519         }
2520
2521         return 0;
2522 }
2523
2524 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2525 {
2526         struct page *page;
2527         unsigned char *kaddr;
2528         struct f2fs_summary *summary;
2529         struct curseg_info *seg_i;
2530         int written_size = 0;
2531         int i, j;
2532
2533         page = grab_meta_page(sbi, blkaddr++);
2534         kaddr = (unsigned char *)page_address(page);
2535
2536         /* Step 1: write nat cache */
2537         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2538         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
2539         written_size += SUM_JOURNAL_SIZE;
2540
2541         /* Step 2: write sit cache */
2542         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2543         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
2544         written_size += SUM_JOURNAL_SIZE;
2545
2546         /* Step 3: write summary entries */
2547         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2548                 unsigned short blkoff;
2549                 seg_i = CURSEG_I(sbi, i);
2550                 if (sbi->ckpt->alloc_type[i] == SSR)
2551                         blkoff = sbi->blocks_per_seg;
2552                 else
2553                         blkoff = curseg_blkoff(sbi, i);
2554
2555                 for (j = 0; j < blkoff; j++) {
2556                         if (!page) {
2557                                 page = grab_meta_page(sbi, blkaddr++);
2558                                 kaddr = (unsigned char *)page_address(page);
2559                                 written_size = 0;
2560                         }
2561                         summary = (struct f2fs_summary *)(kaddr + written_size);
2562                         *summary = seg_i->sum_blk->entries[j];
2563                         written_size += SUMMARY_SIZE;
2564
2565                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
2566                                                         SUM_FOOTER_SIZE)
2567                                 continue;
2568
2569                         set_page_dirty(page);
2570                         f2fs_put_page(page, 1);
2571                         page = NULL;
2572                 }
2573         }
2574         if (page) {
2575                 set_page_dirty(page);
2576                 f2fs_put_page(page, 1);
2577         }
2578 }
2579
2580 static void write_normal_summaries(struct f2fs_sb_info *sbi,
2581                                         block_t blkaddr, int type)
2582 {
2583         int i, end;
2584         if (IS_DATASEG(type))
2585                 end = type + NR_CURSEG_DATA_TYPE;
2586         else
2587                 end = type + NR_CURSEG_NODE_TYPE;
2588
2589         for (i = type; i < end; i++)
2590                 write_current_sum_page(sbi, i, blkaddr + (i - type));
2591 }
2592
2593 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2594 {
2595         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
2596                 write_compacted_summaries(sbi, start_blk);
2597         else
2598                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2599 }
2600
2601 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2602 {
2603         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
2604 }
2605
2606 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
2607                                         unsigned int val, int alloc)
2608 {
2609         int i;
2610
2611         if (type == NAT_JOURNAL) {
2612                 for (i = 0; i < nats_in_cursum(journal); i++) {
2613                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
2614                                 return i;
2615                 }
2616                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2617                         return update_nats_in_cursum(journal, 1);
2618         } else if (type == SIT_JOURNAL) {
2619                 for (i = 0; i < sits_in_cursum(journal); i++)
2620                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
2621                                 return i;
2622                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2623                         return update_sits_in_cursum(journal, 1);
2624         }
2625         return -1;
2626 }
2627
2628 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2629                                         unsigned int segno)
2630 {
2631         return get_meta_page(sbi, current_sit_addr(sbi, segno));
2632 }
2633
2634 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2635                                         unsigned int start)
2636 {
2637         struct sit_info *sit_i = SIT_I(sbi);
2638         struct page *src_page, *dst_page;
2639         pgoff_t src_off, dst_off;
2640         void *src_addr, *dst_addr;
2641
2642         src_off = current_sit_addr(sbi, start);
2643         dst_off = next_sit_addr(sbi, src_off);
2644
2645         /* get current sit block page without lock */
2646         src_page = get_meta_page(sbi, src_off);
2647         dst_page = grab_meta_page(sbi, dst_off);
2648         f2fs_bug_on(sbi, PageDirty(src_page));
2649
2650         src_addr = page_address(src_page);
2651         dst_addr = page_address(dst_page);
2652         memcpy(dst_addr, src_addr, PAGE_SIZE);
2653
2654         set_page_dirty(dst_page);
2655         f2fs_put_page(src_page, 1);
2656
2657         set_to_next_sit(sit_i, start);
2658
2659         return dst_page;
2660 }
2661
2662 static struct sit_entry_set *grab_sit_entry_set(void)
2663 {
2664         struct sit_entry_set *ses =
2665                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
2666
2667         ses->entry_cnt = 0;
2668         INIT_LIST_HEAD(&ses->set_list);
2669         return ses;
2670 }
2671
2672 static void release_sit_entry_set(struct sit_entry_set *ses)
2673 {
2674         list_del(&ses->set_list);
2675         kmem_cache_free(sit_entry_set_slab, ses);
2676 }
2677
2678 static void adjust_sit_entry_set(struct sit_entry_set *ses,
2679                                                 struct list_head *head)
2680 {
2681         struct sit_entry_set *next = ses;
2682
2683         if (list_is_last(&ses->set_list, head))
2684                 return;
2685
2686         list_for_each_entry_continue(next, head, set_list)
2687                 if (ses->entry_cnt <= next->entry_cnt)
2688                         break;
2689
2690         list_move_tail(&ses->set_list, &next->set_list);
2691 }
2692
2693 static void add_sit_entry(unsigned int segno, struct list_head *head)
2694 {
2695         struct sit_entry_set *ses;
2696         unsigned int start_segno = START_SEGNO(segno);
2697
2698         list_for_each_entry(ses, head, set_list) {
2699                 if (ses->start_segno == start_segno) {
2700                         ses->entry_cnt++;
2701                         adjust_sit_entry_set(ses, head);
2702                         return;
2703                 }
2704         }
2705
2706         ses = grab_sit_entry_set();
2707
2708         ses->start_segno = start_segno;
2709         ses->entry_cnt++;
2710         list_add(&ses->set_list, head);
2711 }
2712
2713 static void add_sits_in_set(struct f2fs_sb_info *sbi)
2714 {
2715         struct f2fs_sm_info *sm_info = SM_I(sbi);
2716         struct list_head *set_list = &sm_info->sit_entry_set;
2717         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
2718         unsigned int segno;
2719
2720         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
2721                 add_sit_entry(segno, set_list);
2722 }
2723
2724 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
2725 {
2726         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2727         struct f2fs_journal *journal = curseg->journal;
2728         int i;
2729
2730         down_write(&curseg->journal_rwsem);
2731         for (i = 0; i < sits_in_cursum(journal); i++) {
2732                 unsigned int segno;
2733                 bool dirtied;
2734
2735                 segno = le32_to_cpu(segno_in_journal(journal, i));
2736                 dirtied = __mark_sit_entry_dirty(sbi, segno);
2737
2738                 if (!dirtied)
2739                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
2740         }
2741         update_sits_in_cursum(journal, -i);
2742         up_write(&curseg->journal_rwsem);
2743 }
2744
2745 /*
2746  * CP calls this function, which flushes SIT entries including sit_journal,
2747  * and moves prefree segs to free segs.
2748  */
2749 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2750 {
2751         struct sit_info *sit_i = SIT_I(sbi);
2752         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2753         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2754         struct f2fs_journal *journal = curseg->journal;
2755         struct sit_entry_set *ses, *tmp;
2756         struct list_head *head = &SM_I(sbi)->sit_entry_set;
2757         bool to_journal = true;
2758         struct seg_entry *se;
2759
2760         mutex_lock(&sit_i->sentry_lock);
2761
2762         if (!sit_i->dirty_sentries)
2763                 goto out;
2764
2765         /*
2766          * add and account sit entries of dirty bitmap in sit entry
2767          * set temporarily
2768          */
2769         add_sits_in_set(sbi);
2770
2771         /*
2772          * if there are no enough space in journal to store dirty sit
2773          * entries, remove all entries from journal and add and account
2774          * them in sit entry set.
2775          */
2776         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
2777                 remove_sits_in_journal(sbi);
2778
2779         /*
2780          * there are two steps to flush sit entries:
2781          * #1, flush sit entries to journal in current cold data summary block.
2782          * #2, flush sit entries to sit page.
2783          */
2784         list_for_each_entry_safe(ses, tmp, head, set_list) {
2785                 struct page *page = NULL;
2786                 struct f2fs_sit_block *raw_sit = NULL;
2787                 unsigned int start_segno = ses->start_segno;
2788                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
2789                                                 (unsigned long)MAIN_SEGS(sbi));
2790                 unsigned int segno = start_segno;
2791
2792                 if (to_journal &&
2793                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
2794                         to_journal = false;
2795
2796                 if (to_journal) {
2797                         down_write(&curseg->journal_rwsem);
2798                 } else {
2799                         page = get_next_sit_page(sbi, start_segno);
2800                         raw_sit = page_address(page);
2801                 }
2802
2803                 /* flush dirty sit entries in region of current sit set */
2804                 for_each_set_bit_from(segno, bitmap, end) {
2805                         int offset, sit_offset;
2806
2807                         se = get_seg_entry(sbi, segno);
2808
2809                         /* add discard candidates */
2810                         if (!(cpc->reason & CP_DISCARD)) {
2811                                 cpc->trim_start = segno;
2812                                 add_discard_addrs(sbi, cpc, false);
2813                         }
2814
2815                         if (to_journal) {
2816                                 offset = lookup_journal_in_cursum(journal,
2817                                                         SIT_JOURNAL, segno, 1);
2818                                 f2fs_bug_on(sbi, offset < 0);
2819                                 segno_in_journal(journal, offset) =
2820                                                         cpu_to_le32(segno);
2821                                 seg_info_to_raw_sit(se,
2822                                         &sit_in_journal(journal, offset));
2823                         } else {
2824                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2825                                 seg_info_to_raw_sit(se,
2826                                                 &raw_sit->entries[sit_offset]);
2827                         }
2828
2829                         __clear_bit(segno, bitmap);
2830                         sit_i->dirty_sentries--;
2831                         ses->entry_cnt--;
2832                 }
2833
2834                 if (to_journal)
2835                         up_write(&curseg->journal_rwsem);
2836                 else
2837                         f2fs_put_page(page, 1);
2838
2839                 f2fs_bug_on(sbi, ses->entry_cnt);
2840                 release_sit_entry_set(ses);
2841         }
2842
2843         f2fs_bug_on(sbi, !list_empty(head));
2844         f2fs_bug_on(sbi, sit_i->dirty_sentries);
2845 out:
2846         if (cpc->reason & CP_DISCARD) {
2847                 __u64 trim_start = cpc->trim_start;
2848
2849                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2850                         add_discard_addrs(sbi, cpc, false);
2851
2852                 cpc->trim_start = trim_start;
2853         }
2854         mutex_unlock(&sit_i->sentry_lock);
2855
2856         set_prefree_as_free_segments(sbi);
2857 }
2858
2859 static int build_sit_info(struct f2fs_sb_info *sbi)
2860 {
2861         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2862         struct sit_info *sit_i;
2863         unsigned int sit_segs, start;
2864         char *src_bitmap;
2865         unsigned int bitmap_size;
2866
2867         /* allocate memory for SIT information */
2868         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2869         if (!sit_i)
2870                 return -ENOMEM;
2871
2872         SM_I(sbi)->sit_info = sit_i;
2873
2874         sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2875                                         sizeof(struct seg_entry), GFP_KERNEL);
2876         if (!sit_i->sentries)
2877                 return -ENOMEM;
2878
2879         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2880         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2881         if (!sit_i->dirty_sentries_bitmap)
2882                 return -ENOMEM;
2883
2884         for (start = 0; start < MAIN_SEGS(sbi); start++) {
2885                 sit_i->sentries[start].cur_valid_map
2886                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2887                 sit_i->sentries[start].ckpt_valid_map
2888                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2889                 if (!sit_i->sentries[start].cur_valid_map ||
2890                                 !sit_i->sentries[start].ckpt_valid_map)
2891                         return -ENOMEM;
2892
2893 #ifdef CONFIG_F2FS_CHECK_FS
2894                 sit_i->sentries[start].cur_valid_map_mir
2895                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2896                 if (!sit_i->sentries[start].cur_valid_map_mir)
2897                         return -ENOMEM;
2898 #endif
2899
2900                 if (f2fs_discard_en(sbi)) {
2901                         sit_i->sentries[start].discard_map
2902                                 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2903                         if (!sit_i->sentries[start].discard_map)
2904                                 return -ENOMEM;
2905                 }
2906         }
2907
2908         sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2909         if (!sit_i->tmp_map)
2910                 return -ENOMEM;
2911
2912         if (sbi->segs_per_sec > 1) {
2913                 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2914                                         sizeof(struct sec_entry), GFP_KERNEL);
2915                 if (!sit_i->sec_entries)
2916                         return -ENOMEM;
2917         }
2918
2919         /* get information related with SIT */
2920         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2921
2922         /* setup SIT bitmap from ckeckpoint pack */
2923         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2924         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2925
2926         sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2927         if (!sit_i->sit_bitmap)
2928                 return -ENOMEM;
2929
2930 #ifdef CONFIG_F2FS_CHECK_FS
2931         sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2932         if (!sit_i->sit_bitmap_mir)
2933                 return -ENOMEM;
2934 #endif
2935
2936         /* init SIT information */
2937         sit_i->s_ops = &default_salloc_ops;
2938
2939         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2940         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2941         sit_i->written_valid_blocks = 0;
2942         sit_i->bitmap_size = bitmap_size;
2943         sit_i->dirty_sentries = 0;
2944         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2945         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2946         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2947         mutex_init(&sit_i->sentry_lock);
2948         return 0;
2949 }
2950
2951 static int build_free_segmap(struct f2fs_sb_info *sbi)
2952 {
2953         struct free_segmap_info *free_i;
2954         unsigned int bitmap_size, sec_bitmap_size;
2955
2956         /* allocate memory for free segmap information */
2957         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2958         if (!free_i)
2959                 return -ENOMEM;
2960
2961         SM_I(sbi)->free_info = free_i;
2962
2963         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2964         free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2965         if (!free_i->free_segmap)
2966                 return -ENOMEM;
2967
2968         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2969         free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2970         if (!free_i->free_secmap)
2971                 return -ENOMEM;
2972
2973         /* set all segments as dirty temporarily */
2974         memset(free_i->free_segmap, 0xff, bitmap_size);
2975         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2976
2977         /* init free segmap information */
2978         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2979         free_i->free_segments = 0;
2980         free_i->free_sections = 0;
2981         spin_lock_init(&free_i->segmap_lock);
2982         return 0;
2983 }
2984
2985 static int build_curseg(struct f2fs_sb_info *sbi)
2986 {
2987         struct curseg_info *array;
2988         int i;
2989
2990         array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2991         if (!array)
2992                 return -ENOMEM;
2993
2994         SM_I(sbi)->curseg_array = array;
2995
2996         for (i = 0; i < NR_CURSEG_TYPE; i++) {
2997                 mutex_init(&array[i].curseg_mutex);
2998                 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2999                 if (!array[i].sum_blk)
3000                         return -ENOMEM;
3001                 init_rwsem(&array[i].journal_rwsem);
3002                 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
3003                                                         GFP_KERNEL);
3004                 if (!array[i].journal)
3005                         return -ENOMEM;
3006                 array[i].segno = NULL_SEGNO;
3007                 array[i].next_blkoff = 0;
3008         }
3009         return restore_curseg_summaries(sbi);
3010 }
3011
3012 static void build_sit_entries(struct f2fs_sb_info *sbi)
3013 {
3014         struct sit_info *sit_i = SIT_I(sbi);
3015         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3016         struct f2fs_journal *journal = curseg->journal;
3017         struct seg_entry *se;
3018         struct f2fs_sit_entry sit;
3019         int sit_blk_cnt = SIT_BLK_CNT(sbi);
3020         unsigned int i, start, end;
3021         unsigned int readed, start_blk = 0;
3022
3023         do {
3024                 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3025                                                         META_SIT, true);
3026
3027                 start = start_blk * sit_i->sents_per_block;
3028                 end = (start_blk + readed) * sit_i->sents_per_block;
3029
3030                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
3031                         struct f2fs_sit_block *sit_blk;
3032                         struct page *page;
3033
3034                         se = &sit_i->sentries[start];
3035                         page = get_current_sit_page(sbi, start);
3036                         sit_blk = (struct f2fs_sit_block *)page_address(page);
3037                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3038                         f2fs_put_page(page, 1);
3039
3040                         check_block_count(sbi, start, &sit);
3041                         seg_info_from_raw_sit(se, &sit);
3042
3043                         /* build discard map only one time */
3044                         if (f2fs_discard_en(sbi)) {
3045                                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3046                                         memset(se->discard_map, 0xff,
3047                                                 SIT_VBLOCK_MAP_SIZE);
3048                                 } else {
3049                                         memcpy(se->discard_map,
3050                                                 se->cur_valid_map,
3051                                                 SIT_VBLOCK_MAP_SIZE);
3052                                         sbi->discard_blks +=
3053                                                 sbi->blocks_per_seg -
3054                                                 se->valid_blocks;
3055                                 }
3056                         }
3057
3058                         if (sbi->segs_per_sec > 1)
3059                                 get_sec_entry(sbi, start)->valid_blocks +=
3060                                                         se->valid_blocks;
3061                 }
3062                 start_blk += readed;
3063         } while (start_blk < sit_blk_cnt);
3064
3065         down_read(&curseg->journal_rwsem);
3066         for (i = 0; i < sits_in_cursum(journal); i++) {
3067                 unsigned int old_valid_blocks;
3068
3069                 start = le32_to_cpu(segno_in_journal(journal, i));
3070                 se = &sit_i->sentries[start];
3071                 sit = sit_in_journal(journal, i);
3072
3073                 old_valid_blocks = se->valid_blocks;
3074
3075                 check_block_count(sbi, start, &sit);
3076                 seg_info_from_raw_sit(se, &sit);
3077
3078                 if (f2fs_discard_en(sbi)) {
3079                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3080                                 memset(se->discard_map, 0xff,
3081                                                         SIT_VBLOCK_MAP_SIZE);
3082                         } else {
3083                                 memcpy(se->discard_map, se->cur_valid_map,
3084                                                         SIT_VBLOCK_MAP_SIZE);
3085                                 sbi->discard_blks += old_valid_blocks -
3086                                                         se->valid_blocks;
3087                         }
3088                 }
3089
3090                 if (sbi->segs_per_sec > 1)
3091                         get_sec_entry(sbi, start)->valid_blocks +=
3092                                 se->valid_blocks - old_valid_blocks;
3093         }
3094         up_read(&curseg->journal_rwsem);
3095 }
3096
3097 static void init_free_segmap(struct f2fs_sb_info *sbi)
3098 {
3099         unsigned int start;
3100         int type;
3101
3102         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3103                 struct seg_entry *sentry = get_seg_entry(sbi, start);
3104                 if (!sentry->valid_blocks)
3105                         __set_free(sbi, start);
3106                 else
3107                         SIT_I(sbi)->written_valid_blocks +=
3108                                                 sentry->valid_blocks;
3109         }
3110
3111         /* set use the current segments */
3112         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3113                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3114                 __set_test_and_inuse(sbi, curseg_t->segno);
3115         }
3116 }
3117
3118 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3119 {
3120         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3121         struct free_segmap_info *free_i = FREE_I(sbi);
3122         unsigned int segno = 0, offset = 0;
3123         unsigned short valid_blocks;
3124
3125         while (1) {
3126                 /* find dirty segment based on free segmap */
3127                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3128                 if (segno >= MAIN_SEGS(sbi))
3129                         break;
3130                 offset = segno + 1;
3131                 valid_blocks = get_valid_blocks(sbi, segno, false);
3132                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
3133                         continue;
3134                 if (valid_blocks > sbi->blocks_per_seg) {
3135                         f2fs_bug_on(sbi, 1);
3136                         continue;
3137                 }
3138                 mutex_lock(&dirty_i->seglist_lock);
3139                 __locate_dirty_segment(sbi, segno, DIRTY);
3140                 mutex_unlock(&dirty_i->seglist_lock);
3141         }
3142 }
3143
3144 static int init_victim_secmap(struct f2fs_sb_info *sbi)
3145 {
3146         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3147         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3148
3149         dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
3150         if (!dirty_i->victim_secmap)
3151                 return -ENOMEM;
3152         return 0;
3153 }
3154
3155 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3156 {
3157         struct dirty_seglist_info *dirty_i;
3158         unsigned int bitmap_size, i;
3159
3160         /* allocate memory for dirty segments list information */
3161         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
3162         if (!dirty_i)
3163                 return -ENOMEM;
3164
3165         SM_I(sbi)->dirty_info = dirty_i;
3166         mutex_init(&dirty_i->seglist_lock);
3167
3168         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3169
3170         for (i = 0; i < NR_DIRTY_TYPE; i++) {
3171                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
3172                 if (!dirty_i->dirty_segmap[i])
3173                         return -ENOMEM;
3174         }
3175
3176         init_dirty_segmap(sbi);
3177         return init_victim_secmap(sbi);
3178 }
3179
3180 /*
3181  * Update min, max modified time for cost-benefit GC algorithm
3182  */
3183 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3184 {
3185         struct sit_info *sit_i = SIT_I(sbi);
3186         unsigned int segno;
3187
3188         mutex_lock(&sit_i->sentry_lock);
3189
3190         sit_i->min_mtime = LLONG_MAX;
3191
3192         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
3193                 unsigned int i;
3194                 unsigned long long mtime = 0;
3195
3196                 for (i = 0; i < sbi->segs_per_sec; i++)
3197                         mtime += get_seg_entry(sbi, segno + i)->mtime;
3198
3199                 mtime = div_u64(mtime, sbi->segs_per_sec);
3200
3201                 if (sit_i->min_mtime > mtime)
3202                         sit_i->min_mtime = mtime;
3203         }
3204         sit_i->max_mtime = get_mtime(sbi);
3205         mutex_unlock(&sit_i->sentry_lock);
3206 }
3207
3208 int build_segment_manager(struct f2fs_sb_info *sbi)
3209 {
3210         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3211         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3212         struct f2fs_sm_info *sm_info;
3213         int err;
3214
3215         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
3216         if (!sm_info)
3217                 return -ENOMEM;
3218
3219         /* init sm info */
3220         sbi->sm_info = sm_info;
3221         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3222         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3223         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3224         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3225         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3226         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3227         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
3228         sm_info->rec_prefree_segments = sm_info->main_segments *
3229                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
3230         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3231                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3232
3233         if (!test_opt(sbi, LFS))
3234                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
3235         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
3236         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
3237         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
3238
3239         sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
3240
3241         INIT_LIST_HEAD(&sm_info->sit_entry_set);
3242
3243         if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
3244                 err = create_flush_cmd_control(sbi);
3245                 if (err)
3246                         return err;
3247         }
3248
3249         err = create_discard_cmd_control(sbi);
3250         if (err)
3251                 return err;
3252
3253         err = build_sit_info(sbi);
3254         if (err)
3255                 return err;
3256         err = build_free_segmap(sbi);
3257         if (err)
3258                 return err;
3259         err = build_curseg(sbi);
3260         if (err)
3261                 return err;
3262
3263         /* reinit free segmap based on SIT */
3264         build_sit_entries(sbi);
3265
3266         init_free_segmap(sbi);
3267         err = build_dirty_segmap(sbi);
3268         if (err)
3269                 return err;
3270
3271         init_min_max_mtime(sbi);
3272         return 0;
3273 }
3274
3275 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3276                 enum dirty_type dirty_type)
3277 {
3278         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3279
3280         mutex_lock(&dirty_i->seglist_lock);
3281         kvfree(dirty_i->dirty_segmap[dirty_type]);
3282         dirty_i->nr_dirty[dirty_type] = 0;
3283         mutex_unlock(&dirty_i->seglist_lock);
3284 }
3285
3286 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
3287 {
3288         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3289         kvfree(dirty_i->victim_secmap);
3290 }
3291
3292 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3293 {
3294         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3295         int i;
3296
3297         if (!dirty_i)
3298                 return;
3299
3300         /* discard pre-free/dirty segments list */
3301         for (i = 0; i < NR_DIRTY_TYPE; i++)
3302                 discard_dirty_segmap(sbi, i);
3303
3304         destroy_victim_secmap(sbi);
3305         SM_I(sbi)->dirty_info = NULL;
3306         kfree(dirty_i);
3307 }
3308
3309 static void destroy_curseg(struct f2fs_sb_info *sbi)
3310 {
3311         struct curseg_info *array = SM_I(sbi)->curseg_array;
3312         int i;
3313
3314         if (!array)
3315                 return;
3316         SM_I(sbi)->curseg_array = NULL;
3317         for (i = 0; i < NR_CURSEG_TYPE; i++) {
3318                 kfree(array[i].sum_blk);
3319                 kfree(array[i].journal);
3320         }
3321         kfree(array);
3322 }
3323
3324 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3325 {
3326         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3327         if (!free_i)
3328                 return;
3329         SM_I(sbi)->free_info = NULL;
3330         kvfree(free_i->free_segmap);
3331         kvfree(free_i->free_secmap);
3332         kfree(free_i);
3333 }
3334
3335 static void destroy_sit_info(struct f2fs_sb_info *sbi)
3336 {
3337         struct sit_info *sit_i = SIT_I(sbi);
3338         unsigned int start;
3339
3340         if (!sit_i)
3341                 return;
3342
3343         if (sit_i->sentries) {
3344                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3345                         kfree(sit_i->sentries[start].cur_valid_map);
3346 #ifdef CONFIG_F2FS_CHECK_FS
3347                         kfree(sit_i->sentries[start].cur_valid_map_mir);
3348 #endif
3349                         kfree(sit_i->sentries[start].ckpt_valid_map);
3350                         kfree(sit_i->sentries[start].discard_map);
3351                 }
3352         }
3353         kfree(sit_i->tmp_map);
3354
3355         kvfree(sit_i->sentries);
3356         kvfree(sit_i->sec_entries);
3357         kvfree(sit_i->dirty_sentries_bitmap);
3358
3359         SM_I(sbi)->sit_info = NULL;
3360         kfree(sit_i->sit_bitmap);
3361 #ifdef CONFIG_F2FS_CHECK_FS
3362         kfree(sit_i->sit_bitmap_mir);
3363 #endif
3364         kfree(sit_i);
3365 }
3366
3367 void destroy_segment_manager(struct f2fs_sb_info *sbi)
3368 {
3369         struct f2fs_sm_info *sm_info = SM_I(sbi);
3370
3371         if (!sm_info)
3372                 return;
3373         destroy_flush_cmd_control(sbi, true);
3374         destroy_discard_cmd_control(sbi);
3375         destroy_dirty_segmap(sbi);
3376         destroy_curseg(sbi);
3377         destroy_free_segmap(sbi);
3378         destroy_sit_info(sbi);
3379         sbi->sm_info = NULL;
3380         kfree(sm_info);
3381 }
3382
3383 int __init create_segment_manager_caches(void)
3384 {
3385         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
3386                         sizeof(struct discard_entry));
3387         if (!discard_entry_slab)
3388                 goto fail;
3389
3390         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3391                         sizeof(struct discard_cmd));
3392         if (!discard_cmd_slab)
3393                 goto destroy_discard_entry;
3394
3395         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
3396                         sizeof(struct sit_entry_set));
3397         if (!sit_entry_set_slab)
3398                 goto destroy_discard_cmd;
3399
3400         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3401                         sizeof(struct inmem_pages));
3402         if (!inmem_entry_slab)
3403                 goto destroy_sit_entry_set;
3404         return 0;
3405
3406 destroy_sit_entry_set:
3407         kmem_cache_destroy(sit_entry_set_slab);
3408 destroy_discard_cmd:
3409         kmem_cache_destroy(discard_cmd_slab);
3410 destroy_discard_entry:
3411         kmem_cache_destroy(discard_entry_slab);
3412 fail:
3413         return -ENOMEM;
3414 }
3415
3416 void destroy_segment_manager_caches(void)
3417 {
3418         kmem_cache_destroy(sit_entry_set_slab);
3419         kmem_cache_destroy(discard_cmd_slab);
3420         kmem_cache_destroy(discard_entry_slab);
3421         kmem_cache_destroy(inmem_entry_slab);
3422 }