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
  20 #include "f2fs.h"
  21 #include "segment.h"
  22 #include "node.h"
  23 #include "trace.h"
  24 #include <trace/events/f2fs.h>
  25
  26 #define __reverse_ffz(x) __reverse_ffs(~(x))
  27
  28 static struct kmem_cache *discard_entry_slab;
  29 static struct kmem_cache *discard_cmd_slab;
  30 static struct kmem_cache *sit_entry_set_slab;
  31 static struct kmem_cache *inmem_entry_slab;
  32
  33 static unsigned long __reverse_ulong(unsigned char *str)
  34 {
  35         unsigned long tmp = 0;
  36         int shift = 24, idx = 0;
  37
  38 #if BITS_PER_LONG == 64
  39         shift = 56;
  40 #endif
  41         while (shift >= 0) {
  42                 tmp |= (unsigned long)str[idx++] << shift;
  43                 shift -= BITS_PER_BYTE;
  44         }
  45         return tmp;
  46 }
  47
  48 /*
  49  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  50  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  51  */
  52 static inline unsigned long __reverse_ffs(unsigned long word)
  53 {
  54         int num = 0;
  55
  56 #if BITS_PER_LONG == 64
  57         if ((word & 0xffffffff00000000UL) == 0)
  58                 num += 32;
  59         else
  60                 word >>= 32;
  61 #endif
  62         if ((word & 0xffff0000) == 0)
  63                 num += 16;
  64         else
  65                 word >>= 16;
  66
  67         if ((word & 0xff00) == 0)
  68                 num += 8;
  69         else
  70                 word >>= 8;
  71
  72         if ((word & 0xf0) == 0)
  73                 num += 4;
  74         else
  75                 word >>= 4;
  76
  77         if ((word & 0xc) == 0)
  78                 num += 2;
  79         else
  80                 word >>= 2;
  81
  82         if ((word & 0x2) == 0)
  83                 num += 1;
  84         return num;
  85 }
  86
  87 /*
  88  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  89  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  90  * @size must be integral times of unsigned long.
  91  * Example:
  92  *                             MSB <--> LSB
  93  *   f2fs_set_bit(0, bitmap) => 1000 0000
  94  *   f2fs_set_bit(7, bitmap) => 0000 0001
  95  */
  96 static unsigned long __find_rev_next_bit(const unsigned long *addr,
  97                         unsigned long size, unsigned long offset)
  98 {
  99         const unsigned long *p = addr + BIT_WORD(offset);
 100         unsigned long result = size;
 101         unsigned long tmp;
 102
 103         if (offset >= size)
 104                 return size;
 105
 106         size -= (offset & ~(BITS_PER_LONG - 1));
 107         offset %= BITS_PER_LONG;
 108
 109         while (1) {
 110                 if (*p == 0)
 111                         goto pass;
 112
 113                 tmp = __reverse_ulong((unsigned char *)p);
 114
 115                 tmp &= ~0UL >> offset;
 116                 if (size < BITS_PER_LONG)
 117                         tmp &= (~0UL << (BITS_PER_LONG - size));
 118                 if (tmp)
 119                         goto found;
 120 pass:
 121                 if (size <= BITS_PER_LONG)
 122                         break;
 123                 size -= BITS_PER_LONG;
 124                 offset = 0;
 125                 p++;
 126         }
 127         return result;
 128 found:
 129         return result - size + __reverse_ffs(tmp);
 130 }
 131
 132 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 133                         unsigned long size, unsigned long offset)
 134 {
 135         const unsigned long *p = addr + BIT_WORD(offset);
 136         unsigned long result = size;
 137         unsigned long tmp;
 138
 139         if (offset >= size)
 140                 return size;
 141
 142         size -= (offset & ~(BITS_PER_LONG - 1));
 143         offset %= BITS_PER_LONG;
 144
 145         while (1) {
 146                 if (*p == ~0UL)
 147                         goto pass;
 148
 149                 tmp = __reverse_ulong((unsigned char *)p);
 150
 151                 if (offset)
 152                         tmp |= ~0UL << (BITS_PER_LONG - offset);
 153                 if (size < BITS_PER_LONG)
 154                         tmp |= ~0UL >> size;
 155                 if (tmp != ~0UL)
 156                         goto found;
 157 pass:
 158                 if (size <= BITS_PER_LONG)
 159                         break;
 160                 size -= BITS_PER_LONG;
 161                 offset = 0;
 162                 p++;
 163         }
 164         return result;
 165 found:
 166         return result - size + __reverse_ffz(tmp);
 167 }
 168
 169 void register_inmem_page(struct inode *inode, struct page *page)
 170 {
 171         struct f2fs_inode_info *fi = F2FS_I(inode);
 172         struct inmem_pages *new;
 173
 174         f2fs_trace_pid(page);
 175
 176         set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
 177         SetPagePrivate(page);
 178
 179         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 180
 181         /* add atomic page indices to the list */
 182         new->page = page;
 183         INIT_LIST_HEAD(&new->list);
 184
 185         /* increase reference count with clean state */
 186         mutex_lock(&fi->inmem_lock);
 187         get_page(page);
 188         list_add_tail(&new->list, &fi->inmem_pages);
 189         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 190         mutex_unlock(&fi->inmem_lock);
 191
 192         trace_f2fs_register_inmem_page(page, INMEM);
 193 }
 194
 195 static int __revoke_inmem_pages(struct inode *inode,
 196                                 struct list_head *head, bool drop, bool recover)
 197 {
 198         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 199         struct inmem_pages *cur, *tmp;
 200         int err = 0;
 201
 202         list_for_each_entry_safe(cur, tmp, head, list) {
 203                 struct page *page = cur->page;
 204
 205                 if (drop)
 206                         trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 207
 208                 lock_page(page);
 209
 210                 if (recover) {
 211                         struct dnode_of_data dn;
 212                         struct node_info ni;
 213
 214                         trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 215
 216                         set_new_dnode(&dn, inode, NULL, NULL, 0);
 217                         if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) {
 218                                 err = -EAGAIN;
 219                                 goto next;
 220                         }
 221                         get_node_info(sbi, dn.nid, &ni);
 222                         f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 223                                         cur->old_addr, ni.version, true, true);
 224                         f2fs_put_dnode(&dn);
 225                 }
 226 next:
 227                 /* we don't need to invalidate this in the sccessful status */
 228                 if (drop || recover)
 229                         ClearPageUptodate(page);
 230                 set_page_private(page, 0);
 231                 ClearPagePrivate(page);
 232                 f2fs_put_page(page, 1);
 233
 234                 list_del(&cur->list);
 235                 kmem_cache_free(inmem_entry_slab, cur);
 236                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 237         }
 238         return err;
 239 }
 240
 241 void drop_inmem_pages(struct inode *inode)
 242 {
 243         struct f2fs_inode_info *fi = F2FS_I(inode);
 244
 245         mutex_lock(&fi->inmem_lock);
 246         __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 247         mutex_unlock(&fi->inmem_lock);
 248
 249         clear_inode_flag(inode, FI_ATOMIC_FILE);
 250         stat_dec_atomic_write(inode);
 251 }
 252
 253 void drop_inmem_page(struct inode *inode, struct page *page)
 254 {
 255         struct f2fs_inode_info *fi = F2FS_I(inode);
 256         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 257         struct list_head *head = &fi->inmem_pages;
 258         struct inmem_pages *cur = NULL;
 259
 260         f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
 261
 262         mutex_lock(&fi->inmem_lock);
 263         list_for_each_entry(cur, head, list) {
 264                 if (cur->page == page)
 265                         break;
 266         }
 267
 268         f2fs_bug_on(sbi, !cur || cur->page != page);
 269         list_del(&cur->list);
 270         mutex_unlock(&fi->inmem_lock);
 271
 272         dec_page_count(sbi, F2FS_INMEM_PAGES);
 273         kmem_cache_free(inmem_entry_slab, cur);
 274
 275         ClearPageUptodate(page);
 276         set_page_private(page, 0);
 277         ClearPagePrivate(page);
 278         f2fs_put_page(page, 0);
 279
 280         trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 281 }
 282
 283 static int __commit_inmem_pages(struct inode *inode,
 284                                         struct list_head *revoke_list)
 285 {
 286         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 287         struct f2fs_inode_info *fi = F2FS_I(inode);
 288         struct inmem_pages *cur, *tmp;
 289         struct f2fs_io_info fio = {
 290                 .sbi = sbi,
 291                 .type = DATA,
 292                 .op = REQ_OP_WRITE,
 293                 .op_flags = REQ_SYNC | REQ_PRIO,
 294         };
 295         pgoff_t last_idx = ULONG_MAX;
 296         int err = 0;
 297
 298         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 299                 struct page *page = cur->page;
 300
 301                 lock_page(page);
 302                 if (page->mapping == inode->i_mapping) {
 303                         trace_f2fs_commit_inmem_page(page, INMEM);
 304
 305                         set_page_dirty(page);
 306                         f2fs_wait_on_page_writeback(page, DATA, true);
 307                         if (clear_page_dirty_for_io(page)) {
 308                                 inode_dec_dirty_pages(inode);
 309                                 remove_dirty_inode(inode);
 310                         }
 311
 312                         fio.page = page;
 313                         fio.old_blkaddr = NULL_ADDR;
 314                         fio.encrypted_page = NULL;
 315                         fio.need_lock = false,
 316                         err = do_write_data_page(&fio);
 317                         if (err) {
 318                                 unlock_page(page);
 319                                 break;
 320                         }
 321
 322                         /* record old blkaddr for revoking */
 323                         cur->old_addr = fio.old_blkaddr;
 324                         last_idx = page->index;
 325                 }
 326                 unlock_page(page);
 327                 list_move_tail(&cur->list, revoke_list);
 328         }
 329
 330         if (last_idx != ULONG_MAX)
 331                 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
 332
 333         if (!err)
 334                 __revoke_inmem_pages(inode, revoke_list, false, false);
 335
 336         return err;
 337 }
 338
 339 int commit_inmem_pages(struct inode *inode)
 340 {
 341         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 342         struct f2fs_inode_info *fi = F2FS_I(inode);
 343         struct list_head revoke_list;
 344         int err;
 345
 346         INIT_LIST_HEAD(&revoke_list);
 347         f2fs_balance_fs(sbi, true);
 348         f2fs_lock_op(sbi);
 349
 350         set_inode_flag(inode, FI_ATOMIC_COMMIT);
 351
 352         mutex_lock(&fi->inmem_lock);
 353         err = __commit_inmem_pages(inode, &revoke_list);
 354         if (err) {
 355                 int ret;
 356                 /*
 357                  * try to revoke all committed pages, but still we could fail
 358                  * due to no memory or other reason, if that happened, EAGAIN
 359                  * will be returned, which means in such case, transaction is
 360                  * already not integrity, caller should use journal to do the
 361                  * recovery or rewrite & commit last transaction. For other
 362                  * error number, revoking was done by filesystem itself.
 363                  */
 364                 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
 365                 if (ret)
 366                         err = ret;
 367
 368                 /* drop all uncommitted pages */
 369                 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 370         }
 371         mutex_unlock(&fi->inmem_lock);
 372
 373         clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 374
 375         f2fs_unlock_op(sbi);
 376         return err;
 377 }
 378
 379 /*
 380  * This function balances dirty node and dentry pages.
 381  * In addition, it controls garbage collection.
 382  */
 383 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 384 {
 385 #ifdef CONFIG_F2FS_FAULT_INJECTION
 386         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 387                 f2fs_show_injection_info(FAULT_CHECKPOINT);
 388                 f2fs_stop_checkpoint(sbi, false);
 389         }
 390 #endif
 391
 392         /* balance_fs_bg is able to be pending */
 393         if (need && excess_cached_nats(sbi))
 394                 f2fs_balance_fs_bg(sbi);
 395
 396         /*
 397          * We should do GC or end up with checkpoint, if there are so many dirty
 398          * dir/node pages without enough free segments.
 399          */
 400         if (has_not_enough_free_secs(sbi, 0, 0)) {
 401                 mutex_lock(&sbi->gc_mutex);
 402                 f2fs_gc(sbi, false, false, NULL_SEGNO);
 403         }
 404 }
 405
 406 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 407 {
 408         /* try to shrink extent cache when there is no enough memory */
 409         if (!available_free_memory(sbi, EXTENT_CACHE))
 410                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 411
 412         /* check the # of cached NAT entries */
 413         if (!available_free_memory(sbi, NAT_ENTRIES))
 414                 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 415
 416         if (!available_free_memory(sbi, FREE_NIDS))
 417                 try_to_free_nids(sbi, MAX_FREE_NIDS);
 418         else
 419                 build_free_nids(sbi, false, false);
 420
 421         if (!is_idle(sbi) && !excess_dirty_nats(sbi))
 422                 return;
 423
 424         /* checkpoint is the only way to shrink partial cached entries */
 425         if (!available_free_memory(sbi, NAT_ENTRIES) ||
 426                         !available_free_memory(sbi, INO_ENTRIES) ||
 427                         excess_prefree_segs(sbi) ||
 428                         excess_dirty_nats(sbi) ||
 429                         f2fs_time_over(sbi, CP_TIME)) {
 430                 if (test_opt(sbi, DATA_FLUSH)) {
 431                         struct blk_plug plug;
 432
 433                         blk_start_plug(&plug);
 434                         sync_dirty_inodes(sbi, FILE_INODE);
 435                         blk_finish_plug(&plug);
 436                 }
 437                 f2fs_sync_fs(sbi->sb, true);
 438                 stat_inc_bg_cp_count(sbi->stat_info);
 439         }
 440 }
 441
 442 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 443                                 struct block_device *bdev)
 444 {
 445         struct bio *bio = f2fs_bio_alloc(0);
 446         int ret;
 447
 448         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
 449         bio->bi_bdev = bdev;
 450         ret = submit_bio_wait(bio);
 451         bio_put(bio);
 452
 453         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 454                                 test_opt(sbi, FLUSH_MERGE), ret);
 455         return ret;
 456 }
 457
 458 static int submit_flush_wait(struct f2fs_sb_info *sbi)
 459 {
 460         int ret = __submit_flush_wait(sbi, sbi->sb->s_bdev);
 461         int i;
 462
 463         if (!sbi->s_ndevs || ret)
 464                 return ret;
 465
 466         for (i = 1; i < sbi->s_ndevs; i++) {
 467                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 468                 if (ret)
 469                         break;
 470         }
 471         return ret;
 472 }
 473
 474 static int issue_flush_thread(void *data)
 475 {
 476         struct f2fs_sb_info *sbi = data;
 477         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 478         wait_queue_head_t *q = &fcc->flush_wait_queue;
 479 repeat:
 480         if (kthread_should_stop())
 481                 return 0;
 482
 483         if (!llist_empty(&fcc->issue_list)) {
 484                 struct flush_cmd *cmd, *next;
 485                 int ret;
 486
 487                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 488                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 489
 490                 ret = submit_flush_wait(sbi);
 491                 atomic_inc(&fcc->issued_flush);
 492
 493                 llist_for_each_entry_safe(cmd, next,
 494                                           fcc->dispatch_list, llnode) {
 495                         cmd->ret = ret;
 496                         complete(&cmd->wait);
 497                 }
 498                 fcc->dispatch_list = NULL;
 499         }
 500
 501         wait_event_interruptible(*q,
 502                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 503         goto repeat;
 504 }
 505
 506 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
 507 {
 508         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 509         struct flush_cmd cmd;
 510         int ret;
 511
 512         if (test_opt(sbi, NOBARRIER))
 513                 return 0;
 514
 515         if (!test_opt(sbi, FLUSH_MERGE)) {
 516                 ret = submit_flush_wait(sbi);
 517                 atomic_inc(&fcc->issued_flush);
 518                 return ret;
 519         }
 520
 521         if (!atomic_read(&fcc->issing_flush)) {
 522                 atomic_inc(&fcc->issing_flush);
 523                 ret = submit_flush_wait(sbi);
 524                 atomic_dec(&fcc->issing_flush);
 525
 526                 atomic_inc(&fcc->issued_flush);
 527                 return ret;
 528         }
 529
 530         init_completion(&cmd.wait);
 531
 532         atomic_inc(&fcc->issing_flush);
 533         llist_add(&cmd.llnode, &fcc->issue_list);
 534
 535         if (!fcc->dispatch_list)
 536                 wake_up(&fcc->flush_wait_queue);
 537
 538         if (fcc->f2fs_issue_flush) {
 539                 wait_for_completion(&cmd.wait);
 540                 atomic_dec(&fcc->issing_flush);
 541         } else {
 542                 llist_del_all(&fcc->issue_list);
 543                 atomic_set(&fcc->issing_flush, 0);
 544         }
 545
 546         return cmd.ret;
 547 }
 548
 549 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
 550 {
 551         dev_t dev = sbi->sb->s_bdev->bd_dev;
 552         struct flush_cmd_control *fcc;
 553         int err = 0;
 554
 555         if (SM_I(sbi)->fcc_info) {
 556                 fcc = SM_I(sbi)->fcc_info;
 557                 goto init_thread;
 558         }
 559
 560         fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
 561         if (!fcc)
 562                 return -ENOMEM;
 563         atomic_set(&fcc->issued_flush, 0);
 564         atomic_set(&fcc->issing_flush, 0);
 565         init_waitqueue_head(&fcc->flush_wait_queue);
 566         init_llist_head(&fcc->issue_list);
 567         SM_I(sbi)->fcc_info = fcc;
 568 init_thread:
 569         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 570                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 571         if (IS_ERR(fcc->f2fs_issue_flush)) {
 572                 err = PTR_ERR(fcc->f2fs_issue_flush);
 573                 kfree(fcc);
 574                 SM_I(sbi)->fcc_info = NULL;
 575                 return err;
 576         }
 577
 578         return err;
 579 }
 580
 581 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 582 {
 583         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 584
 585         if (fcc && fcc->f2fs_issue_flush) {
 586                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 587
 588                 fcc->f2fs_issue_flush = NULL;
 589                 kthread_stop(flush_thread);
 590         }
 591         if (free) {
 592                 kfree(fcc);
 593                 SM_I(sbi)->fcc_info = NULL;
 594         }
 595 }
 596
 597 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 598                 enum dirty_type dirty_type)
 599 {
 600         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 601
 602         /* need not be added */
 603         if (IS_CURSEG(sbi, segno))
 604                 return;
 605
 606         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 607                 dirty_i->nr_dirty[dirty_type]++;
 608
 609         if (dirty_type == DIRTY) {
 610                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 611                 enum dirty_type t = sentry->type;
 612
 613                 if (unlikely(t >= DIRTY)) {
 614                         f2fs_bug_on(sbi, 1);
 615                         return;
 616                 }
 617                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 618                         dirty_i->nr_dirty[t]++;
 619         }
 620 }
 621
 622 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 623                 enum dirty_type dirty_type)
 624 {
 625         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 626
 627         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 628                 dirty_i->nr_dirty[dirty_type]--;
 629
 630         if (dirty_type == DIRTY) {
 631                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 632                 enum dirty_type t = sentry->type;
 633
 634                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 635                         dirty_i->nr_dirty[t]--;
 636
 637                 if (get_valid_blocks(sbi, segno, true) == 0)
 638                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 639                                                 dirty_i->victim_secmap);
 640         }
 641 }
 642
 643 /*
 644  * Should not occur error such as -ENOMEM.
 645  * Adding dirty entry into seglist is not critical operation.
 646  * If a given segment is one of current working segments, it won't be added.
 647  */
 648 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 649 {
 650         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 651         unsigned short valid_blocks;
 652
 653         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 654                 return;
 655
 656         mutex_lock(&dirty_i->seglist_lock);
 657
 658         valid_blocks = get_valid_blocks(sbi, segno, false);
 659
 660         if (valid_blocks == 0) {
 661                 __locate_dirty_segment(sbi, segno, PRE);
 662                 __remove_dirty_segment(sbi, segno, DIRTY);
 663         } else if (valid_blocks < sbi->blocks_per_seg) {
 664                 __locate_dirty_segment(sbi, segno, DIRTY);
 665         } else {
 666                 /* Recovery routine with SSR needs this */
 667                 __remove_dirty_segment(sbi, segno, DIRTY);
 668         }
 669
 670         mutex_unlock(&dirty_i->seglist_lock);
 671 }
 672
 673 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 674                 struct block_device *bdev, block_t lstart,
 675                 block_t start, block_t len)
 676 {
 677         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 678         struct list_head *pend_list;
 679         struct discard_cmd *dc;
 680
 681         f2fs_bug_on(sbi, !len);
 682
 683         pend_list = &dcc->pend_list[plist_idx(len)];
 684
 685         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 686         INIT_LIST_HEAD(&dc->list);
 687         dc->bdev = bdev;
 688         dc->lstart = lstart;
 689         dc->start = start;
 690         dc->len = len;
 691         dc->ref = 0;
 692         dc->state = D_PREP;
 693         dc->error = 0;
 694         init_completion(&dc->wait);
 695         list_add_tail(&dc->list, pend_list);
 696         atomic_inc(&dcc->discard_cmd_cnt);
 697         dcc->undiscard_blks += len;
 698
 699         return dc;
 700 }
 701
 702 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 703                                 struct block_device *bdev, block_t lstart,
 704                                 block_t start, block_t len,
 705                                 struct rb_node *parent, struct rb_node **p)
 706 {
 707         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 708         struct discard_cmd *dc;
 709
 710         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
 711
 712         rb_link_node(&dc->rb_node, parent, p);
 713         rb_insert_color(&dc->rb_node, &dcc->root);
 714
 715         return dc;
 716 }
 717
 718 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
 719                                                         struct discard_cmd *dc)
 720 {
 721         if (dc->state == D_DONE)
 722                 atomic_dec(&dcc->issing_discard);
 723
 724         list_del(&dc->list);
 725         rb_erase(&dc->rb_node, &dcc->root);
 726         dcc->undiscard_blks -= dc->len;
 727
 728         kmem_cache_free(discard_cmd_slab, dc);
 729
 730         atomic_dec(&dcc->discard_cmd_cnt);
 731 }
 732
 733 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
 734                                                         struct discard_cmd *dc)
 735 {
 736         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 737
 738         if (dc->error == -EOPNOTSUPP)
 739                 dc->error = 0;
 740
 741         if (dc->error)
 742                 f2fs_msg(sbi->sb, KERN_INFO,
 743                                 "Issue discard failed, ret: %d", dc->error);
 744         __detach_discard_cmd(dcc, dc);
 745 }
 746
 747 static void f2fs_submit_discard_endio(struct bio *bio)
 748 {
 749         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
 750
 751         dc->error = bio->bi_error;
 752         dc->state = D_DONE;
 753         complete(&dc->wait);
 754         bio_put(bio);
 755 }
 756
 757 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
 758 static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
 759                                 struct discard_cmd *dc)
 760 {
 761         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 762         struct bio *bio = NULL;
 763
 764         if (dc->state != D_PREP)
 765                 return;
 766
 767         trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
 768
 769         dc->error = __blkdev_issue_discard(dc->bdev,
 770                                 SECTOR_FROM_BLOCK(dc->start),
 771                                 SECTOR_FROM_BLOCK(dc->len),
 772                                 GFP_NOFS, 0, &bio);
 773         if (!dc->error) {
 774                 /* should keep before submission to avoid D_DONE right away */
 775                 dc->state = D_SUBMIT;
 776                 atomic_inc(&dcc->issued_discard);
 777                 atomic_inc(&dcc->issing_discard);
 778                 if (bio) {
 779                         bio->bi_private = dc;
 780                         bio->bi_end_io = f2fs_submit_discard_endio;
 781                         bio->bi_opf |= REQ_SYNC;
 782                         submit_bio(bio);
 783                         list_move_tail(&dc->list, &dcc->wait_list);
 784                 }
 785         } else {
 786                 __remove_discard_cmd(sbi, dc);
 787         }
 788 }
 789
 790 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
 791                                 struct block_device *bdev, block_t lstart,
 792                                 block_t start, block_t len,
 793                                 struct rb_node **insert_p,
 794                                 struct rb_node *insert_parent)
 795 {
 796         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 797         struct rb_node **p = &dcc->root.rb_node;
 798         struct rb_node *parent = NULL;
 799         struct discard_cmd *dc = NULL;
 800
 801         if (insert_p && insert_parent) {
 802                 parent = insert_parent;
 803                 p = insert_p;
 804                 goto do_insert;
 805         }
 806
 807         p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
 808 do_insert:
 809         dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
 810         if (!dc)
 811                 return NULL;
 812
 813         return dc;
 814 }
 815
 816 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
 817                                                 struct discard_cmd *dc)
 818 {
 819         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
 820 }
 821
 822 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
 823                                 struct discard_cmd *dc, block_t blkaddr)
 824 {
 825         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 826         struct discard_info di = dc->di;
 827         bool modified = false;
 828
 829         if (dc->state == D_DONE || dc->len == 1) {
 830                 __remove_discard_cmd(sbi, dc);
 831                 return;
 832         }
 833
 834         dcc->undiscard_blks -= di.len;
 835
 836         if (blkaddr > di.lstart) {
 837                 dc->len = blkaddr - dc->lstart;
 838                 dcc->undiscard_blks += dc->len;
 839                 __relocate_discard_cmd(dcc, dc);
 840                 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
 841                 modified = true;
 842         }
 843
 844         if (blkaddr < di.lstart + di.len - 1) {
 845                 if (modified) {
 846                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
 847                                         di.start + blkaddr + 1 - di.lstart,
 848                                         di.lstart + di.len - 1 - blkaddr,
 849                                         NULL, NULL);
 850                         f2fs_bug_on(sbi,
 851                                 !__check_rb_tree_consistence(sbi, &dcc->root));
 852                 } else {
 853                         dc->lstart++;
 854                         dc->len--;
 855                         dc->start++;
 856                         dcc->undiscard_blks += dc->len;
 857                         __relocate_discard_cmd(dcc, dc);
 858                         f2fs_bug_on(sbi,
 859                                 !__check_rb_tree_consistence(sbi, &dcc->root));
 860                 }
 861         }
 862 }
 863
 864 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
 865                                 struct block_device *bdev, block_t lstart,
 866                                 block_t start, block_t len)
 867 {
 868         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 869         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
 870         struct discard_cmd *dc;
 871         struct discard_info di = {0};
 872         struct rb_node **insert_p = NULL, *insert_parent = NULL;
 873         block_t end = lstart + len;
 874
 875         mutex_lock(&dcc->cmd_lock);
 876
 877         dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
 878                                         NULL, lstart,
 879                                         (struct rb_entry **)&prev_dc,
 880                                         (struct rb_entry **)&next_dc,
 881                                         &insert_p, &insert_parent, true);
 882         if (dc)
 883                 prev_dc = dc;
 884
 885         if (!prev_dc) {
 886                 di.lstart = lstart;
 887                 di.len = next_dc ? next_dc->lstart - lstart : len;
 888                 di.len = min(di.len, len);
 889                 di.start = start;
 890         }
 891
 892         while (1) {
 893                 struct rb_node *node;
 894                 bool merged = false;
 895                 struct discard_cmd *tdc = NULL;
 896
 897                 if (prev_dc) {
 898                         di.lstart = prev_dc->lstart + prev_dc->len;
 899                         if (di.lstart < lstart)
 900                                 di.lstart = lstart;
 901                         if (di.lstart >= end)
 902                                 break;
 903
 904                         if (!next_dc || next_dc->lstart > end)
 905                                 di.len = end - di.lstart;
 906                         else
 907                                 di.len = next_dc->lstart - di.lstart;
 908                         di.start = start + di.lstart - lstart;
 909                 }
 910
 911                 if (!di.len)
 912                         goto next;
 913
 914                 if (prev_dc && prev_dc->state == D_PREP &&
 915                         prev_dc->bdev == bdev &&
 916                         __is_discard_back_mergeable(&di, &prev_dc->di)) {
 917                         prev_dc->di.len += di.len;
 918                         dcc->undiscard_blks += di.len;
 919                         __relocate_discard_cmd(dcc, prev_dc);
 920                         f2fs_bug_on(sbi,
 921                                 !__check_rb_tree_consistence(sbi, &dcc->root));
 922                         di = prev_dc->di;
 923                         tdc = prev_dc;
 924                         merged = true;
 925                 }
 926
 927                 if (next_dc && next_dc->state == D_PREP &&
 928                         next_dc->bdev == bdev &&
 929                         __is_discard_front_mergeable(&di, &next_dc->di)) {
 930                         next_dc->di.lstart = di.lstart;
 931                         next_dc->di.len += di.len;
 932                         next_dc->di.start = di.start;
 933                         dcc->undiscard_blks += di.len;
 934                         __relocate_discard_cmd(dcc, next_dc);
 935                         if (tdc)
 936                                 __remove_discard_cmd(sbi, tdc);
 937                         f2fs_bug_on(sbi,
 938                                 !__check_rb_tree_consistence(sbi, &dcc->root));
 939                         merged = true;
 940                 }
 941
 942                 if (!merged) {
 943                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
 944                                                         di.len, NULL, NULL);
 945                         f2fs_bug_on(sbi,
 946                                 !__check_rb_tree_consistence(sbi, &dcc->root));
 947                 }
 948  next:
 949                 prev_dc = next_dc;
 950                 if (!prev_dc)
 951                         break;
 952
 953                 node = rb_next(&prev_dc->rb_node);
 954                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
 955         }
 956
 957         mutex_unlock(&dcc->cmd_lock);
 958 }
 959
 960 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
 961                 struct block_device *bdev, block_t blkstart, block_t blklen)
 962 {
 963         block_t lblkstart = blkstart;
 964
 965         trace_f2fs_queue_discard(bdev, blkstart, blklen);
 966
 967         if (sbi->s_ndevs) {
 968                 int devi = f2fs_target_device_index(sbi, blkstart);
 969
 970                 blkstart -= FDEV(devi).start_blk;
 971         }
 972         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
 973         return 0;
 974 }
 975
 976 static void __issue_discard_cmd(struct f2fs_sb_info *sbi, bool issue_cond)
 977 {
 978         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 979         struct list_head *pend_list;
 980         struct discard_cmd *dc, *tmp;
 981         struct blk_plug plug;
 982         int i, iter = 0;
 983
 984         mutex_lock(&dcc->cmd_lock);
 985         blk_start_plug(&plug);
 986         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
 987                 pend_list = &dcc->pend_list[i];
 988                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
 989                         f2fs_bug_on(sbi, dc->state != D_PREP);
 990
 991                         if (!issue_cond || is_idle(sbi))
 992                                 __submit_discard_cmd(sbi, dc);
 993                         if (issue_cond && iter++ > DISCARD_ISSUE_RATE)
 994                                 goto out;
 995                 }
 996         }
 997 out:
 998         blk_finish_plug(&plug);
 999         mutex_unlock(&dcc->cmd_lock);
1000 }
1001
1002 static void __wait_discard_cmd(struct f2fs_sb_info *sbi, bool wait_cond)
1003 {
1004         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1005         struct list_head *wait_list = &(dcc->wait_list);
1006         struct discard_cmd *dc, *tmp;
1007
1008         mutex_lock(&dcc->cmd_lock);
1009         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1010                 if (!wait_cond || dc->state == D_DONE) {
1011                         if (dc->ref)
1012                                 continue;
1013                         wait_for_completion_io(&dc->wait);
1014                         __remove_discard_cmd(sbi, dc);
1015                 }
1016         }
1017         mutex_unlock(&dcc->cmd_lock);
1018 }
1019
1020 /* This should be covered by global mutex, &sit_i->sentry_lock */
1021 void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1022 {
1023         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1024         struct discard_cmd *dc;
1025         bool need_wait = false;
1026
1027         mutex_lock(&dcc->cmd_lock);
1028         dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1029         if (dc) {
1030                 if (dc->state == D_PREP) {
1031                         __punch_discard_cmd(sbi, dc, blkaddr);
1032                 } else {
1033                         dc->ref++;
1034                         need_wait = true;
1035                 }
1036         }
1037         mutex_unlock(&dcc->cmd_lock);
1038
1039         if (need_wait) {
1040                 wait_for_completion_io(&dc->wait);
1041                 mutex_lock(&dcc->cmd_lock);
1042                 f2fs_bug_on(sbi, dc->state != D_DONE);
1043                 dc->ref--;
1044                 if (!dc->ref)
1045                         __remove_discard_cmd(sbi, dc);
1046                 mutex_unlock(&dcc->cmd_lock);
1047         }
1048 }
1049
1050 /* This comes from f2fs_put_super */
1051 void f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1052 {
1053         __issue_discard_cmd(sbi, false);
1054         __wait_discard_cmd(sbi, false);
1055 }
1056
1057 static int issue_discard_thread(void *data)
1058 {
1059         struct f2fs_sb_info *sbi = data;
1060         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1061         wait_queue_head_t *q = &dcc->discard_wait_queue;
1062 repeat:
1063         if (kthread_should_stop())
1064                 return 0;
1065
1066         __issue_discard_cmd(sbi, true);
1067         __wait_discard_cmd(sbi, true);
1068
1069         congestion_wait(BLK_RW_SYNC, HZ/50);
1070
1071         wait_event_interruptible(*q, kthread_should_stop() ||
1072                                 atomic_read(&dcc->discard_cmd_cnt));
1073         goto repeat;
1074 }
1075
1076 #ifdef CONFIG_BLK_DEV_ZONED
1077 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1078                 struct block_device *bdev, block_t blkstart, block_t blklen)
1079 {
1080         sector_t sector, nr_sects;
1081         block_t lblkstart = blkstart;
1082         int devi = 0;
1083
1084         if (sbi->s_ndevs) {
1085                 devi = f2fs_target_device_index(sbi, blkstart);
1086                 blkstart -= FDEV(devi).start_blk;
1087         }
1088
1089         /*
1090          * We need to know the type of the zone: for conventional zones,
1091          * use regular discard if the drive supports it. For sequential
1092          * zones, reset the zone write pointer.
1093          */
1094         switch (get_blkz_type(sbi, bdev, blkstart)) {
1095
1096         case BLK_ZONE_TYPE_CONVENTIONAL:
1097                 if (!blk_queue_discard(bdev_get_queue(bdev)))
1098                         return 0;
1099                 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1100         case BLK_ZONE_TYPE_SEQWRITE_REQ:
1101         case BLK_ZONE_TYPE_SEQWRITE_PREF:
1102                 sector = SECTOR_FROM_BLOCK(blkstart);
1103                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1104
1105                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1106                                 nr_sects != bdev_zone_sectors(bdev)) {
1107                         f2fs_msg(sbi->sb, KERN_INFO,
1108                                 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1109                                 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1110                                 blkstart, blklen);
1111                         return -EIO;
1112                 }
1113                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1114                 return blkdev_reset_zones(bdev, sector,
1115                                           nr_sects, GFP_NOFS);
1116         default:
1117                 /* Unknown zone type: broken device ? */
1118                 return -EIO;
1119         }
1120 }
1121 #endif
1122
1123 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1124                 struct block_device *bdev, block_t blkstart, block_t blklen)
1125 {
1126 #ifdef CONFIG_BLK_DEV_ZONED
1127         if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
1128                                 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1129                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1130 #endif
1131         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1132 }
1133
1134 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1135                                 block_t blkstart, block_t blklen)
1136 {
1137         sector_t start = blkstart, len = 0;
1138         struct block_device *bdev;
1139         struct seg_entry *se;
1140         unsigned int offset;
1141         block_t i;
1142         int err = 0;
1143
1144         bdev = f2fs_target_device(sbi, blkstart, NULL);
1145
1146         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1147                 if (i != start) {
1148                         struct block_device *bdev2 =
1149                                 f2fs_target_device(sbi, i, NULL);
1150
1151                         if (bdev2 != bdev) {
1152                                 err = __issue_discard_async(sbi, bdev,
1153                                                 start, len);
1154                                 if (err)
1155                                         return err;
1156                                 bdev = bdev2;
1157                                 start = i;
1158                                 len = 0;
1159                         }
1160                 }
1161
1162                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1163                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1164
1165                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1166                         sbi->discard_blks--;
1167         }
1168
1169         if (len)
1170                 err = __issue_discard_async(sbi, bdev, start, len);
1171         return err;
1172 }
1173
1174 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1175                                                         bool check_only)
1176 {
1177         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1178         int max_blocks = sbi->blocks_per_seg;
1179         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1180         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1181         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1182         unsigned long *discard_map = (unsigned long *)se->discard_map;
1183         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1184         unsigned int start = 0, end = -1;
1185         bool force = (cpc->reason & CP_DISCARD);
1186         struct discard_entry *de = NULL;
1187         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1188         int i;
1189
1190         if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
1191                 return false;
1192
1193         if (!force) {
1194                 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
1195                         SM_I(sbi)->dcc_info->nr_discards >=
1196                                 SM_I(sbi)->dcc_info->max_discards)
1197                         return false;
1198         }
1199
1200         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1201         for (i = 0; i < entries; i++)
1202                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1203                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1204
1205         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1206                                 SM_I(sbi)->dcc_info->max_discards) {
1207                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1208                 if (start >= max_blocks)
1209                         break;
1210
1211                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1212                 if (force && start && end != max_blocks
1213                                         && (end - start) < cpc->trim_minlen)
1214                         continue;
1215
1216                 if (check_only)
1217                         return true;
1218
1219                 if (!de) {
1220                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1221                                                                 GFP_F2FS_ZERO);
1222                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1223                         list_add_tail(&de->list, head);
1224                 }
1225
1226                 for (i = start; i < end; i++)
1227                         __set_bit_le(i, (void *)de->discard_map);
1228
1229                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1230         }
1231         return false;
1232 }
1233
1234 void release_discard_addrs(struct f2fs_sb_info *sbi)
1235 {
1236         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1237         struct discard_entry *entry, *this;
1238
1239         /* drop caches */
1240         list_for_each_entry_safe(entry, this, head, list) {
1241                 list_del(&entry->list);
1242                 kmem_cache_free(discard_entry_slab, entry);
1243         }
1244 }
1245
1246 /*
1247  * Should call clear_prefree_segments after checkpoint is done.
1248  */
1249 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1250 {
1251         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1252         unsigned int segno;
1253
1254         mutex_lock(&dirty_i->seglist_lock);
1255         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1256                 __set_test_and_free(sbi, segno);
1257         mutex_unlock(&dirty_i->seglist_lock);
1258 }
1259
1260 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1261 {
1262         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1263         struct discard_entry *entry, *this;
1264         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1265         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1266         unsigned int start = 0, end = -1;
1267         unsigned int secno, start_segno;
1268         bool force = (cpc->reason & CP_DISCARD);
1269
1270         mutex_lock(&dirty_i->seglist_lock);
1271
1272         while (1) {
1273                 int i;
1274                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1275                 if (start >= MAIN_SEGS(sbi))
1276                         break;
1277                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1278                                                                 start + 1);
1279
1280                 for (i = start; i < end; i++)
1281                         clear_bit(i, prefree_map);
1282
1283                 dirty_i->nr_dirty[PRE] -= end - start;
1284
1285                 if (!test_opt(sbi, DISCARD))
1286                         continue;
1287
1288                 if (force && start >= cpc->trim_start &&
1289                                         (end - 1) <= cpc->trim_end)
1290                                 continue;
1291
1292                 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1293                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1294                                 (end - start) << sbi->log_blocks_per_seg);
1295                         continue;
1296                 }
1297 next:
1298                 secno = GET_SEC_FROM_SEG(sbi, start);
1299                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1300                 if (!IS_CURSEC(sbi, secno) &&
1301                         !get_valid_blocks(sbi, start, true))
1302                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1303                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1304
1305                 start = start_segno + sbi->segs_per_sec;
1306                 if (start < end)
1307                         goto next;
1308                 else
1309                         end = start - 1;
1310         }
1311         mutex_unlock(&dirty_i->seglist_lock);
1312
1313         /* send small discards */
1314         list_for_each_entry_safe(entry, this, head, list) {
1315                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1316                 bool is_valid = test_bit_le(0, entry->discard_map);
1317
1318 find_next:
1319                 if (is_valid) {
1320                         next_pos = find_next_zero_bit_le(entry->discard_map,
1321                                         sbi->blocks_per_seg, cur_pos);
1322                         len = next_pos - cur_pos;
1323
1324                         if (force && len < cpc->trim_minlen)
1325                                 goto skip;
1326
1327                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1328                                                                         len);
1329                         cpc->trimmed += len;
1330                         total_len += len;
1331                 } else {
1332                         next_pos = find_next_bit_le(entry->discard_map,
1333                                         sbi->blocks_per_seg, cur_pos);
1334                 }
1335 skip:
1336                 cur_pos = next_pos;
1337                 is_valid = !is_valid;
1338
1339                 if (cur_pos < sbi->blocks_per_seg)
1340                         goto find_next;
1341
1342                 list_del(&entry->list);
1343                 SM_I(sbi)->dcc_info->nr_discards -= total_len;
1344                 kmem_cache_free(discard_entry_slab, entry);
1345         }
1346
1347         wake_up(&SM_I(sbi)->dcc_info->discard_wait_queue);
1348 }
1349
1350 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1351 {
1352         dev_t dev = sbi->sb->s_bdev->bd_dev;
1353         struct discard_cmd_control *dcc;
1354         int err = 0, i;
1355
1356         if (SM_I(sbi)->dcc_info) {
1357                 dcc = SM_I(sbi)->dcc_info;
1358                 goto init_thread;
1359         }
1360
1361         dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1362         if (!dcc)
1363                 return -ENOMEM;
1364
1365         INIT_LIST_HEAD(&dcc->entry_list);
1366         for (i = 0; i < MAX_PLIST_NUM; i++)
1367                 INIT_LIST_HEAD(&dcc->pend_list[i]);
1368         INIT_LIST_HEAD(&dcc->wait_list);
1369         mutex_init(&dcc->cmd_lock);
1370         atomic_set(&dcc->issued_discard, 0);
1371         atomic_set(&dcc->issing_discard, 0);
1372         atomic_set(&dcc->discard_cmd_cnt, 0);
1373         dcc->nr_discards = 0;
1374         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1375         dcc->undiscard_blks = 0;
1376         dcc->root = RB_ROOT;
1377
1378         init_waitqueue_head(&dcc->discard_wait_queue);
1379         SM_I(sbi)->dcc_info = dcc;
1380 init_thread:
1381         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1382                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1383         if (IS_ERR(dcc->f2fs_issue_discard)) {
1384                 err = PTR_ERR(dcc->f2fs_issue_discard);
1385                 kfree(dcc);
1386                 SM_I(sbi)->dcc_info = NULL;
1387                 return err;
1388         }
1389
1390         return err;
1391 }
1392
1393 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1394 {
1395         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1396
1397         if (!dcc)
1398                 return;
1399
1400         if (dcc->f2fs_issue_discard) {
1401                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1402
1403                 dcc->f2fs_issue_discard = NULL;
1404                 kthread_stop(discard_thread);
1405         }
1406
1407         kfree(dcc);
1408         SM_I(sbi)->dcc_info = NULL;
1409 }
1410
1411 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1412 {
1413         struct sit_info *sit_i = SIT_I(sbi);
1414
1415         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1416                 sit_i->dirty_sentries++;
1417                 return false;
1418         }
1419
1420         return true;
1421 }
1422
1423 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1424                                         unsigned int segno, int modified)
1425 {
1426         struct seg_entry *se = get_seg_entry(sbi, segno);
1427         se->type = type;
1428         if (modified)
1429                 __mark_sit_entry_dirty(sbi, segno);
1430 }
1431
1432 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1433 {
1434         struct seg_entry *se;
1435         unsigned int segno, offset;
1436         long int new_vblocks;
1437
1438         segno = GET_SEGNO(sbi, blkaddr);
1439
1440         se = get_seg_entry(sbi, segno);
1441         new_vblocks = se->valid_blocks + del;
1442         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1443
1444         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
1445                                 (new_vblocks > sbi->blocks_per_seg)));
1446
1447         se->valid_blocks = new_vblocks;
1448         se->mtime = get_mtime(sbi);
1449         SIT_I(sbi)->max_mtime = se->mtime;
1450
1451         /* Update valid block bitmap */
1452         if (del > 0) {
1453                 if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) {
1454 #ifdef CONFIG_F2FS_CHECK_FS
1455                         if (f2fs_test_and_set_bit(offset,
1456                                                 se->cur_valid_map_mir))
1457                                 f2fs_bug_on(sbi, 1);
1458                         else
1459                                 WARN_ON(1);
1460 #else
1461                         f2fs_bug_on(sbi, 1);
1462 #endif
1463                 }
1464                 if (f2fs_discard_en(sbi) &&
1465                         !f2fs_test_and_set_bit(offset, se->discard_map))
1466                         sbi->discard_blks--;
1467
1468                 /* don't overwrite by SSR to keep node chain */
1469                 if (se->type == CURSEG_WARM_NODE) {
1470                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1471                                 se->ckpt_valid_blocks++;
1472                 }
1473         } else {
1474                 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) {
1475 #ifdef CONFIG_F2FS_CHECK_FS
1476                         if (!f2fs_test_and_clear_bit(offset,
1477                                                 se->cur_valid_map_mir))
1478                                 f2fs_bug_on(sbi, 1);
1479                         else
1480                                 WARN_ON(1);
1481 #else
1482                         f2fs_bug_on(sbi, 1);
1483 #endif
1484                 }
1485                 if (f2fs_discard_en(sbi) &&
1486                         f2fs_test_and_clear_bit(offset, se->discard_map))
1487                         sbi->discard_blks++;
1488         }
1489         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1490                 se->ckpt_valid_blocks += del;
1491
1492         __mark_sit_entry_dirty(sbi, segno);
1493
1494         /* update total number of valid blocks to be written in ckpt area */
1495         SIT_I(sbi)->written_valid_blocks += del;
1496
1497         if (sbi->segs_per_sec > 1)
1498                 get_sec_entry(sbi, segno)->valid_blocks += del;
1499 }
1500
1501 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
1502 {
1503         update_sit_entry(sbi, new, 1);
1504         if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1505                 update_sit_entry(sbi, old, -1);
1506
1507         locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1508         locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
1509 }
1510
1511 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1512 {
1513         unsigned int segno = GET_SEGNO(sbi, addr);
1514         struct sit_info *sit_i = SIT_I(sbi);
1515
1516         f2fs_bug_on(sbi, addr == NULL_ADDR);
1517         if (addr == NEW_ADDR)
1518                 return;
1519
1520         /* add it into sit main buffer */
1521         mutex_lock(&sit_i->sentry_lock);
1522
1523         update_sit_entry(sbi, addr, -1);
1524
1525         /* add it into dirty seglist */
1526         locate_dirty_segment(sbi, segno);
1527
1528         mutex_unlock(&sit_i->sentry_lock);
1529 }
1530
1531 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1532 {
1533         struct sit_info *sit_i = SIT_I(sbi);
1534         unsigned int segno, offset;
1535         struct seg_entry *se;
1536         bool is_cp = false;
1537
1538         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1539                 return true;
1540
1541         mutex_lock(&sit_i->sentry_lock);
1542
1543         segno = GET_SEGNO(sbi, blkaddr);
1544         se = get_seg_entry(sbi, segno);
1545         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1546
1547         if (f2fs_test_bit(offset, se->ckpt_valid_map))
1548                 is_cp = true;
1549
1550         mutex_unlock(&sit_i->sentry_lock);
1551
1552         return is_cp;
1553 }
1554
1555 /*
1556  * This function should be resided under the curseg_mutex lock
1557  */
1558 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1559                                         struct f2fs_summary *sum)
1560 {
1561         struct curseg_info *curseg = CURSEG_I(sbi, type);
1562         void *addr = curseg->sum_blk;
1563         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1564         memcpy(addr, sum, sizeof(struct f2fs_summary));
1565 }
1566
1567 /*
1568  * Calculate the number of current summary pages for writing
1569  */
1570 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1571 {
1572         int valid_sum_count = 0;
1573         int i, sum_in_page;
1574
1575         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1576                 if (sbi->ckpt->alloc_type[i] == SSR)
1577                         valid_sum_count += sbi->blocks_per_seg;
1578                 else {
1579                         if (for_ra)
1580                                 valid_sum_count += le16_to_cpu(
1581                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1582                         else
1583                                 valid_sum_count += curseg_blkoff(sbi, i);
1584                 }
1585         }
1586
1587         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1588                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1589         if (valid_sum_count <= sum_in_page)
1590                 return 1;
1591         else if ((valid_sum_count - sum_in_page) <=
1592                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1593                 return 2;
1594         return 3;
1595 }
1596
1597 /*
1598  * Caller should put this summary page
1599  */
1600 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1601 {
1602         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1603 }
1604
1605 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1606 {
1607         struct page *page = grab_meta_page(sbi, blk_addr);
1608         void *dst = page_address(page);
1609
1610         if (src)
1611                 memcpy(dst, src, PAGE_SIZE);
1612         else
1613                 memset(dst, 0, PAGE_SIZE);
1614         set_page_dirty(page);
1615         f2fs_put_page(page, 1);
1616 }
1617
1618 static void write_sum_page(struct f2fs_sb_info *sbi,
1619                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
1620 {
1621         update_meta_page(sbi, (void *)sum_blk, blk_addr);
1622 }
1623
1624 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1625                                                 int type, block_t blk_addr)
1626 {
1627         struct curseg_info *curseg = CURSEG_I(sbi, type);
1628         struct page *page = grab_meta_page(sbi, blk_addr);
1629         struct f2fs_summary_block *src = curseg->sum_blk;
1630         struct f2fs_summary_block *dst;
1631
1632         dst = (struct f2fs_summary_block *)page_address(page);
1633
1634         mutex_lock(&curseg->curseg_mutex);
1635
1636         down_read(&curseg->journal_rwsem);
1637         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1638         up_read(&curseg->journal_rwsem);
1639
1640         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1641         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1642
1643         mutex_unlock(&curseg->curseg_mutex);
1644
1645         set_page_dirty(page);
1646         f2fs_put_page(page, 1);
1647 }
1648
1649 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1650 {
1651         struct curseg_info *curseg = CURSEG_I(sbi, type);
1652         unsigned int segno = curseg->segno + 1;
1653         struct free_segmap_info *free_i = FREE_I(sbi);
1654
1655         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
1656                 return !test_bit(segno, free_i->free_segmap);
1657         return 0;
1658 }
1659
1660 /*
1661  * Find a new segment from the free segments bitmap to right order
1662  * This function should be returned with success, otherwise BUG
1663  */
1664 static void get_new_segment(struct f2fs_sb_info *sbi,
1665                         unsigned int *newseg, bool new_sec, int dir)
1666 {
1667         struct free_segmap_info *free_i = FREE_I(sbi);
1668         unsigned int segno, secno, zoneno;
1669         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
1670         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
1671         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
1672         unsigned int left_start = hint;
1673         bool init = true;
1674         int go_left = 0;
1675         int i;
1676
1677         spin_lock(&free_i->segmap_lock);
1678
1679         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1680                 segno = find_next_zero_bit(free_i->free_segmap,
1681                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
1682                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
1683                         goto got_it;
1684         }
1685 find_other_zone:
1686         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1687         if (secno >= MAIN_SECS(sbi)) {
1688                 if (dir == ALLOC_RIGHT) {
1689                         secno = find_next_zero_bit(free_i->free_secmap,
1690                                                         MAIN_SECS(sbi), 0);
1691                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1692                 } else {
1693                         go_left = 1;
1694                         left_start = hint - 1;
1695                 }
1696         }
1697         if (go_left == 0)
1698                 goto skip_left;
1699
1700         while (test_bit(left_start, free_i->free_secmap)) {
1701                 if (left_start > 0) {
1702                         left_start--;
1703                         continue;
1704                 }
1705                 left_start = find_next_zero_bit(free_i->free_secmap,
1706                                                         MAIN_SECS(sbi), 0);
1707                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1708                 break;
1709         }
1710         secno = left_start;
1711 skip_left:
1712         hint = secno;
1713         segno = GET_SEG_FROM_SEC(sbi, secno);
1714         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
1715
1716         /* give up on finding another zone */
1717         if (!init)
1718                 goto got_it;
1719         if (sbi->secs_per_zone == 1)
1720                 goto got_it;
1721         if (zoneno == old_zoneno)
1722                 goto got_it;
1723         if (dir == ALLOC_LEFT) {
1724                 if (!go_left && zoneno + 1 >= total_zones)
1725                         goto got_it;
1726                 if (go_left && zoneno == 0)
1727                         goto got_it;
1728         }
1729         for (i = 0; i < NR_CURSEG_TYPE; i++)
1730                 if (CURSEG_I(sbi, i)->zone == zoneno)
1731                         break;
1732
1733         if (i < NR_CURSEG_TYPE) {
1734                 /* zone is in user, try another */
1735                 if (go_left)
1736                         hint = zoneno * sbi->secs_per_zone - 1;
1737                 else if (zoneno + 1 >= total_zones)
1738                         hint = 0;
1739                 else
1740                         hint = (zoneno + 1) * sbi->secs_per_zone;
1741                 init = false;
1742                 goto find_other_zone;
1743         }
1744 got_it:
1745         /* set it as dirty segment in free segmap */
1746         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1747         __set_inuse(sbi, segno);
1748         *newseg = segno;
1749         spin_unlock(&free_i->segmap_lock);
1750 }
1751
1752 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1753 {
1754         struct curseg_info *curseg = CURSEG_I(sbi, type);
1755         struct summary_footer *sum_footer;
1756
1757         curseg->segno = curseg->next_segno;
1758         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
1759         curseg->next_blkoff = 0;
1760         curseg->next_segno = NULL_SEGNO;
1761
1762         sum_footer = &(curseg->sum_blk->footer);
1763         memset(sum_footer, 0, sizeof(struct summary_footer));
1764         if (IS_DATASEG(type))
1765                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1766         if (IS_NODESEG(type))
1767                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1768         __set_sit_entry_type(sbi, type, curseg->segno, modified);
1769 }
1770
1771 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
1772 {
1773         /* if segs_per_sec is large than 1, we need to keep original policy. */
1774         if (sbi->segs_per_sec != 1)
1775                 return CURSEG_I(sbi, type)->segno;
1776
1777         if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
1778                 return 0;
1779
1780         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
1781                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
1782         return CURSEG_I(sbi, type)->segno;
1783 }
1784
1785 /*
1786  * Allocate a current working segment.
1787  * This function always allocates a free segment in LFS manner.
1788  */
1789 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1790 {
1791         struct curseg_info *curseg = CURSEG_I(sbi, type);
1792         unsigned int segno = curseg->segno;
1793         int dir = ALLOC_LEFT;
1794
1795         write_sum_page(sbi, curseg->sum_blk,
1796                                 GET_SUM_BLOCK(sbi, segno));
1797         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1798                 dir = ALLOC_RIGHT;
1799
1800         if (test_opt(sbi, NOHEAP))
1801                 dir = ALLOC_RIGHT;
1802
1803         segno = __get_next_segno(sbi, type);
1804         get_new_segment(sbi, &segno, new_sec, dir);
1805         curseg->next_segno = segno;
1806         reset_curseg(sbi, type, 1);
1807         curseg->alloc_type = LFS;
1808 }
1809
1810 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1811                         struct curseg_info *seg, block_t start)
1812 {
1813         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1814         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1815         unsigned long *target_map = SIT_I(sbi)->tmp_map;
1816         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1817         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1818         int i, pos;
1819
1820         for (i = 0; i < entries; i++)
1821                 target_map[i] = ckpt_map[i] | cur_map[i];
1822
1823         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1824
1825         seg->next_blkoff = pos;
1826 }
1827
1828 /*
1829  * If a segment is written by LFS manner, next block offset is just obtained
1830  * by increasing the current block offset. However, if a segment is written by
1831  * SSR manner, next block offset obtained by calling __next_free_blkoff
1832  */
1833 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1834                                 struct curseg_info *seg)
1835 {
1836         if (seg->alloc_type == SSR)
1837                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1838         else
1839                 seg->next_blkoff++;
1840 }
1841
1842 /*
1843  * This function always allocates a used segment(from dirty seglist) by SSR
1844  * manner, so it should recover the existing segment information of valid blocks
1845  */
1846 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1847 {
1848         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1849         struct curseg_info *curseg = CURSEG_I(sbi, type);
1850         unsigned int new_segno = curseg->next_segno;
1851         struct f2fs_summary_block *sum_node;
1852         struct page *sum_page;
1853
1854         write_sum_page(sbi, curseg->sum_blk,
1855                                 GET_SUM_BLOCK(sbi, curseg->segno));
1856         __set_test_and_inuse(sbi, new_segno);
1857
1858         mutex_lock(&dirty_i->seglist_lock);
1859         __remove_dirty_segment(sbi, new_segno, PRE);
1860         __remove_dirty_segment(sbi, new_segno, DIRTY);
1861         mutex_unlock(&dirty_i->seglist_lock);
1862
1863         reset_curseg(sbi, type, 1);
1864         curseg->alloc_type = SSR;
1865         __next_free_blkoff(sbi, curseg, 0);
1866
1867         if (reuse) {
1868                 sum_page = get_sum_page(sbi, new_segno);
1869                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1870                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1871                 f2fs_put_page(sum_page, 1);
1872         }
1873 }
1874
1875 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1876 {
1877         struct curseg_info *curseg = CURSEG_I(sbi, type);
1878         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1879         unsigned segno = NULL_SEGNO;
1880         int i, cnt;
1881         bool reversed = false;
1882
1883         /* need_SSR() already forces to do this */
1884         if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
1885                 curseg->next_segno = segno;
1886                 return 1;
1887         }
1888
1889         /* For node segments, let's do SSR more intensively */
1890         if (IS_NODESEG(type)) {
1891                 if (type >= CURSEG_WARM_NODE) {
1892                         reversed = true;
1893                         i = CURSEG_COLD_NODE;
1894                 } else {
1895                         i = CURSEG_HOT_NODE;
1896                 }
1897                 cnt = NR_CURSEG_NODE_TYPE;
1898         } else {
1899                 if (type >= CURSEG_WARM_DATA) {
1900                         reversed = true;
1901                         i = CURSEG_COLD_DATA;
1902                 } else {
1903                         i = CURSEG_HOT_DATA;
1904                 }
1905                 cnt = NR_CURSEG_DATA_TYPE;
1906         }
1907
1908         for (; cnt-- > 0; reversed ? i-- : i++) {
1909                 if (i == type)
1910                         continue;
1911                 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
1912                         curseg->next_segno = segno;
1913                         return 1;
1914                 }
1915         }
1916         return 0;
1917 }
1918
1919 /*
1920  * flush out current segment and replace it with new segment
1921  * This function should be returned with success, otherwise BUG
1922  */
1923 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1924                                                 int type, bool force)
1925 {
1926         struct curseg_info *curseg = CURSEG_I(sbi, type);
1927
1928         if (force)
1929                 new_curseg(sbi, type, true);
1930         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
1931                                         type == CURSEG_WARM_NODE)
1932                 new_curseg(sbi, type, false);
1933         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1934                 new_curseg(sbi, type, false);
1935         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1936                 change_curseg(sbi, type, true);
1937         else
1938                 new_curseg(sbi, type, false);
1939
1940         stat_inc_seg_type(sbi, curseg);
1941 }
1942
1943 void allocate_new_segments(struct f2fs_sb_info *sbi)
1944 {
1945         struct curseg_info *curseg;
1946         unsigned int old_segno;
1947         int i;
1948
1949         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1950                 curseg = CURSEG_I(sbi, i);
1951                 old_segno = curseg->segno;
1952                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
1953                 locate_dirty_segment(sbi, old_segno);
1954         }
1955 }
1956
1957 static const struct segment_allocation default_salloc_ops = {
1958         .allocate_segment = allocate_segment_by_default,
1959 };
1960
1961 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1962 {
1963         __u64 trim_start = cpc->trim_start;
1964         bool has_candidate = false;
1965
1966         mutex_lock(&SIT_I(sbi)->sentry_lock);
1967         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
1968                 if (add_discard_addrs(sbi, cpc, true)) {
1969                         has_candidate = true;
1970                         break;
1971                 }
1972         }
1973         mutex_unlock(&SIT_I(sbi)->sentry_lock);
1974
1975         cpc->trim_start = trim_start;
1976         return has_candidate;
1977 }
1978
1979 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1980 {
1981         __u64 start = F2FS_BYTES_TO_BLK(range->start);
1982         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1983         unsigned int start_segno, end_segno;
1984         struct cp_control cpc;
1985         int err = 0;
1986
1987         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1988                 return -EINVAL;
1989
1990         cpc.trimmed = 0;
1991         if (end <= MAIN_BLKADDR(sbi))
1992                 goto out;
1993
1994         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1995                 f2fs_msg(sbi->sb, KERN_WARNING,
1996                         "Found FS corruption, run fsck to fix.");
1997                 goto out;
1998         }
1999
2000         /* start/end segment number in main_area */
2001         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2002         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2003                                                 GET_SEGNO(sbi, end);
2004         cpc.reason = CP_DISCARD;
2005         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2006
2007         /* do checkpoint to issue discard commands safely */
2008         for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
2009                 cpc.trim_start = start_segno;
2010
2011                 if (sbi->discard_blks == 0)
2012                         break;
2013                 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
2014                         cpc.trim_end = end_segno;
2015                 else
2016                         cpc.trim_end = min_t(unsigned int,
2017                                 rounddown(start_segno +
2018                                 BATCHED_TRIM_SEGMENTS(sbi),
2019                                 sbi->segs_per_sec) - 1, end_segno);
2020
2021                 mutex_lock(&sbi->gc_mutex);
2022                 err = write_checkpoint(sbi, &cpc);
2023                 mutex_unlock(&sbi->gc_mutex);
2024                 if (err)
2025                         break;
2026
2027                 schedule();
2028         }
2029 out:
2030         range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
2031         return err;
2032 }
2033
2034 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2035 {
2036         struct curseg_info *curseg = CURSEG_I(sbi, type);
2037         if (curseg->next_blkoff < sbi->blocks_per_seg)
2038                 return true;
2039         return false;
2040 }
2041
2042 static int __get_segment_type_2(struct page *page, enum page_type p_type)
2043 {
2044         if (p_type == DATA)
2045                 return CURSEG_HOT_DATA;
2046         else
2047                 return CURSEG_HOT_NODE;
2048 }
2049
2050 static int __get_segment_type_4(struct page *page, enum page_type p_type)
2051 {
2052         if (p_type == DATA) {
2053                 struct inode *inode = page->mapping->host;
2054
2055                 if (S_ISDIR(inode->i_mode))
2056                         return CURSEG_HOT_DATA;
2057                 else
2058                         return CURSEG_COLD_DATA;
2059         } else {
2060                 if (IS_DNODE(page) && is_cold_node(page))
2061                         return CURSEG_WARM_NODE;
2062                 else
2063                         return CURSEG_COLD_NODE;
2064         }
2065 }
2066
2067 static int __get_segment_type_6(struct page *page, enum page_type p_type)
2068 {
2069         if (p_type == DATA) {
2070                 struct inode *inode = page->mapping->host;
2071
2072                 if (is_cold_data(page) || file_is_cold(inode))
2073                         return CURSEG_COLD_DATA;
2074                 if (is_inode_flag_set(inode, FI_HOT_DATA))
2075                         return CURSEG_HOT_DATA;
2076                 return CURSEG_WARM_DATA;
2077         } else {
2078                 if (IS_DNODE(page))
2079                         return is_cold_node(page) ? CURSEG_WARM_NODE :
2080                                                 CURSEG_HOT_NODE;
2081                 return CURSEG_COLD_NODE;
2082         }
2083 }
2084
2085 static int __get_segment_type(struct page *page, enum page_type p_type)
2086 {
2087         switch (F2FS_P_SB(page)->active_logs) {
2088         case 2:
2089                 return __get_segment_type_2(page, p_type);
2090         case 4:
2091                 return __get_segment_type_4(page, p_type);
2092         }
2093         /* NR_CURSEG_TYPE(6) logs by default */
2094         f2fs_bug_on(F2FS_P_SB(page),
2095                 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
2096         return __get_segment_type_6(page, p_type);
2097 }
2098
2099 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2100                 block_t old_blkaddr, block_t *new_blkaddr,
2101                 struct f2fs_summary *sum, int type)
2102 {
2103         struct sit_info *sit_i = SIT_I(sbi);
2104         struct curseg_info *curseg = CURSEG_I(sbi, type);
2105
2106         mutex_lock(&curseg->curseg_mutex);
2107         mutex_lock(&sit_i->sentry_lock);
2108
2109         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2110
2111         f2fs_wait_discard_bio(sbi, *new_blkaddr);
2112
2113         /*
2114          * __add_sum_entry should be resided under the curseg_mutex
2115          * because, this function updates a summary entry in the
2116          * current summary block.
2117          */
2118         __add_sum_entry(sbi, type, sum);
2119
2120         __refresh_next_blkoff(sbi, curseg);
2121
2122         stat_inc_block_count(sbi, curseg);
2123
2124         if (!__has_curseg_space(sbi, type))
2125                 sit_i->s_ops->allocate_segment(sbi, type, false);
2126         /*
2127          * SIT information should be updated after segment allocation,
2128          * since we need to keep dirty segments precisely under SSR.
2129          */
2130         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
2131
2132         mutex_unlock(&sit_i->sentry_lock);
2133
2134         if (page && IS_NODESEG(type))
2135                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2136
2137         mutex_unlock(&curseg->curseg_mutex);
2138 }
2139
2140 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2141 {
2142         int type = __get_segment_type(fio->page, fio->type);
2143         int err;
2144
2145         if (fio->type == NODE || fio->type == DATA)
2146                 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
2147 reallocate:
2148         allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
2149                                         &fio->new_blkaddr, sum, type);
2150
2151         /* writeout dirty page into bdev */
2152         err = f2fs_submit_page_write(fio);
2153         if (err == -EAGAIN) {
2154                 fio->old_blkaddr = fio->new_blkaddr;
2155                 goto reallocate;
2156         }
2157
2158         if (fio->type == NODE || fio->type == DATA)
2159                 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
2160 }
2161
2162 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
2163 {
2164         struct f2fs_io_info fio = {
2165                 .sbi = sbi,
2166                 .type = META,
2167                 .op = REQ_OP_WRITE,
2168                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
2169                 .old_blkaddr = page->index,
2170                 .new_blkaddr = page->index,
2171                 .page = page,
2172                 .encrypted_page = NULL,
2173         };
2174
2175         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
2176                 fio.op_flags &= ~REQ_META;
2177
2178         set_page_writeback(page);
2179         f2fs_submit_page_write(&fio);
2180 }
2181
2182 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
2183 {
2184         struct f2fs_summary sum;
2185
2186         set_summary(&sum, nid, 0, 0);
2187         do_write_page(&sum, fio);
2188 }
2189
2190 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
2191 {
2192         struct f2fs_sb_info *sbi = fio->sbi;
2193         struct f2fs_summary sum;
2194         struct node_info ni;
2195
2196         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
2197         get_node_info(sbi, dn->nid, &ni);
2198         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
2199         do_write_page(&sum, fio);
2200         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
2201 }
2202
2203 int rewrite_data_page(struct f2fs_io_info *fio)
2204 {
2205         fio->new_blkaddr = fio->old_blkaddr;
2206         stat_inc_inplace_blocks(fio->sbi);
2207         return f2fs_submit_page_bio(fio);
2208 }
2209
2210 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
2211                                 block_t old_blkaddr, block_t new_blkaddr,
2212                                 bool recover_curseg, bool recover_newaddr)
2213 {
2214         struct sit_info *sit_i = SIT_I(sbi);
2215         struct curseg_info *curseg;
2216         unsigned int segno, old_cursegno;
2217         struct seg_entry *se;
2218         int type;
2219         unsigned short old_blkoff;
2220
2221         segno = GET_SEGNO(sbi, new_blkaddr);
2222         se = get_seg_entry(sbi, segno);
2223         type = se->type;
2224
2225         if (!recover_curseg) {
2226                 /* for recovery flow */
2227                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2228                         if (old_blkaddr == NULL_ADDR)
2229                                 type = CURSEG_COLD_DATA;
2230                         else
2231                                 type = CURSEG_WARM_DATA;
2232                 }
2233         } else {
2234                 if (!IS_CURSEG(sbi, segno))
2235                         type = CURSEG_WARM_DATA;
2236         }
2237
2238         curseg = CURSEG_I(sbi, type);
2239
2240         mutex_lock(&curseg->curseg_mutex);
2241         mutex_lock(&sit_i->sentry_lock);
2242
2243         old_cursegno = curseg->segno;
2244         old_blkoff = curseg->next_blkoff;
2245
2246         /* change the current segment */
2247         if (segno != curseg->segno) {
2248                 curseg->next_segno = segno;
2249                 change_curseg(sbi, type, true);
2250         }
2251
2252         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
2253         __add_sum_entry(sbi, type, sum);
2254
2255         if (!recover_curseg || recover_newaddr)
2256                 update_sit_entry(sbi, new_blkaddr, 1);
2257         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2258                 update_sit_entry(sbi, old_blkaddr, -1);
2259
2260         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2261         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2262
2263         locate_dirty_segment(sbi, old_cursegno);
2264
2265         if (recover_curseg) {
2266                 if (old_cursegno != curseg->segno) {
2267                         curseg->next_segno = old_cursegno;
2268                         change_curseg(sbi, type, true);
2269                 }
2270                 curseg->next_blkoff = old_blkoff;
2271         }
2272
2273         mutex_unlock(&sit_i->sentry_lock);
2274         mutex_unlock(&curseg->curseg_mutex);
2275 }
2276
2277 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2278                                 block_t old_addr, block_t new_addr,
2279                                 unsigned char version, bool recover_curseg,
2280                                 bool recover_newaddr)
2281 {
2282         struct f2fs_summary sum;
2283
2284         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2285
2286         __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2287                                         recover_curseg, recover_newaddr);
2288
2289         f2fs_update_data_blkaddr(dn, new_addr);
2290 }
2291
2292 void f2fs_wait_on_page_writeback(struct page *page,
2293                                 enum page_type type, bool ordered)
2294 {
2295         if (PageWriteback(page)) {
2296                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2297
2298                 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2299                                                 0, page->index, type);
2300                 if (ordered)
2301                         wait_on_page_writeback(page);
2302                 else
2303                         wait_for_stable_page(page);
2304         }
2305 }
2306
2307 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
2308                                                         block_t blkaddr)
2309 {
2310         struct page *cpage;
2311
2312         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
2313                 return;
2314
2315         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2316         if (cpage) {
2317                 f2fs_wait_on_page_writeback(cpage, DATA, true);
2318                 f2fs_put_page(cpage, 1);
2319         }
2320 }
2321
2322 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
2323 {
2324         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2325         struct curseg_info *seg_i;
2326         unsigned char *kaddr;
2327         struct page *page;
2328         block_t start;
2329         int i, j, offset;
2330
2331         start = start_sum_block(sbi);
2332
2333         page = get_meta_page(sbi, start++);
2334         kaddr = (unsigned char *)page_address(page);
2335
2336         /* Step 1: restore nat cache */
2337         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2338         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
2339
2340         /* Step 2: restore sit cache */
2341         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2342         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
2343         offset = 2 * SUM_JOURNAL_SIZE;
2344
2345         /* Step 3: restore summary entries */
2346         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2347                 unsigned short blk_off;
2348                 unsigned int segno;
2349
2350                 seg_i = CURSEG_I(sbi, i);
2351                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2352                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2353                 seg_i->next_segno = segno;
2354                 reset_curseg(sbi, i, 0);
2355                 seg_i->alloc_type = ckpt->alloc_type[i];
2356                 seg_i->next_blkoff = blk_off;
2357
2358                 if (seg_i->alloc_type == SSR)
2359                         blk_off = sbi->blocks_per_seg;
2360
2361                 for (j = 0; j < blk_off; j++) {
2362                         struct f2fs_summary *s;
2363                         s = (struct f2fs_summary *)(kaddr + offset);
2364                         seg_i->sum_blk->entries[j] = *s;
2365                         offset += SUMMARY_SIZE;
2366                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
2367                                                 SUM_FOOTER_SIZE)
2368                                 continue;
2369
2370                         f2fs_put_page(page, 1);
2371                         page = NULL;
2372
2373                         page = get_meta_page(sbi, start++);
2374                         kaddr = (unsigned char *)page_address(page);
2375                         offset = 0;
2376                 }
2377         }
2378         f2fs_put_page(page, 1);
2379         return 0;
2380 }
2381
2382 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2383 {
2384         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2385         struct f2fs_summary_block *sum;
2386         struct curseg_info *curseg;
2387         struct page *new;
2388         unsigned short blk_off;
2389         unsigned int segno = 0;
2390         block_t blk_addr = 0;
2391
2392         /* get segment number and block addr */
2393         if (IS_DATASEG(type)) {
2394                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2395                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2396                                                         CURSEG_HOT_DATA]);
2397                 if (__exist_node_summaries(sbi))
2398                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2399                 else
2400                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2401         } else {
2402                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2403                                                         CURSEG_HOT_NODE]);
2404                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2405                                                         CURSEG_HOT_NODE]);
2406                 if (__exist_node_summaries(sbi))
2407                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2408                                                         type - CURSEG_HOT_NODE);
2409                 else
2410                         blk_addr = GET_SUM_BLOCK(sbi, segno);
2411         }
2412
2413         new = get_meta_page(sbi, blk_addr);
2414         sum = (struct f2fs_summary_block *)page_address(new);
2415
2416         if (IS_NODESEG(type)) {
2417                 if (__exist_node_summaries(sbi)) {
2418                         struct f2fs_summary *ns = &sum->entries[0];
2419                         int i;
2420                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2421                                 ns->version = 0;
2422                                 ns->ofs_in_node = 0;
2423                         }
2424                 } else {
2425                         int err;
2426
2427                         err = restore_node_summary(sbi, segno, sum);
2428                         if (err) {
2429                                 f2fs_put_page(new, 1);
2430                                 return err;
2431                         }
2432                 }
2433         }
2434
2435         /* set uncompleted segment to curseg */
2436         curseg = CURSEG_I(sbi, type);
2437         mutex_lock(&curseg->curseg_mutex);
2438
2439         /* update journal info */
2440         down_write(&curseg->journal_rwsem);
2441         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2442         up_write(&curseg->journal_rwsem);
2443
2444         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2445         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
2446         curseg->next_segno = segno;
2447         reset_curseg(sbi, type, 0);
2448         curseg->alloc_type = ckpt->alloc_type[type];
2449         curseg->next_blkoff = blk_off;
2450         mutex_unlock(&curseg->curseg_mutex);
2451         f2fs_put_page(new, 1);
2452         return 0;
2453 }
2454
2455 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2456 {
2457         int type = CURSEG_HOT_DATA;
2458         int err;
2459
2460         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
2461                 int npages = npages_for_summary_flush(sbi, true);
2462
2463                 if (npages >= 2)
2464                         ra_meta_pages(sbi, start_sum_block(sbi), npages,
2465                                                         META_CP, true);
2466
2467                 /* restore for compacted data summary */
2468                 if (read_compacted_summaries(sbi))
2469                         return -EINVAL;
2470                 type = CURSEG_HOT_NODE;
2471         }
2472
2473         if (__exist_node_summaries(sbi))
2474                 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
2475                                         NR_CURSEG_TYPE - type, META_CP, true);
2476
2477         for (; type <= CURSEG_COLD_NODE; type++) {
2478                 err = read_normal_summaries(sbi, type);
2479                 if (err)
2480                         return err;
2481         }
2482
2483         return 0;
2484 }
2485
2486 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2487 {
2488         struct page *page;
2489         unsigned char *kaddr;
2490         struct f2fs_summary *summary;
2491         struct curseg_info *seg_i;
2492         int written_size = 0;
2493         int i, j;
2494
2495         page = grab_meta_page(sbi, blkaddr++);
2496         kaddr = (unsigned char *)page_address(page);
2497
2498         /* Step 1: write nat cache */
2499         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2500         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
2501         written_size += SUM_JOURNAL_SIZE;
2502
2503         /* Step 2: write sit cache */
2504         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2505         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
2506         written_size += SUM_JOURNAL_SIZE;
2507
2508         /* Step 3: write summary entries */
2509         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2510                 unsigned short blkoff;
2511                 seg_i = CURSEG_I(sbi, i);
2512                 if (sbi->ckpt->alloc_type[i] == SSR)
2513                         blkoff = sbi->blocks_per_seg;
2514                 else
2515                         blkoff = curseg_blkoff(sbi, i);
2516
2517                 for (j = 0; j < blkoff; j++) {
2518                         if (!page) {
2519                                 page = grab_meta_page(sbi, blkaddr++);
2520                                 kaddr = (unsigned char *)page_address(page);
2521                                 written_size = 0;
2522                         }
2523                         summary = (struct f2fs_summary *)(kaddr + written_size);
2524                         *summary = seg_i->sum_blk->entries[j];
2525                         written_size += SUMMARY_SIZE;
2526
2527                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
2528                                                         SUM_FOOTER_SIZE)
2529                                 continue;
2530
2531                         set_page_dirty(page);
2532                         f2fs_put_page(page, 1);
2533                         page = NULL;
2534                 }
2535         }
2536         if (page) {
2537                 set_page_dirty(page);
2538                 f2fs_put_page(page, 1);
2539         }
2540 }
2541
2542 static void write_normal_summaries(struct f2fs_sb_info *sbi,
2543                                         block_t blkaddr, int type)
2544 {
2545         int i, end;
2546         if (IS_DATASEG(type))
2547                 end = type + NR_CURSEG_DATA_TYPE;
2548         else
2549                 end = type + NR_CURSEG_NODE_TYPE;
2550
2551         for (i = type; i < end; i++)
2552                 write_current_sum_page(sbi, i, blkaddr + (i - type));
2553 }
2554
2555 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2556 {
2557         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
2558                 write_compacted_summaries(sbi, start_blk);
2559         else
2560                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2561 }
2562
2563 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2564 {
2565         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
2566 }
2567
2568 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
2569                                         unsigned int val, int alloc)
2570 {
2571         int i;
2572
2573         if (type == NAT_JOURNAL) {
2574                 for (i = 0; i < nats_in_cursum(journal); i++) {
2575                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
2576                                 return i;
2577                 }
2578                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2579                         return update_nats_in_cursum(journal, 1);
2580         } else if (type == SIT_JOURNAL) {
2581                 for (i = 0; i < sits_in_cursum(journal); i++)
2582                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
2583                                 return i;
2584                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2585                         return update_sits_in_cursum(journal, 1);
2586         }
2587         return -1;
2588 }
2589
2590 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2591                                         unsigned int segno)
2592 {
2593         return get_meta_page(sbi, current_sit_addr(sbi, segno));
2594 }
2595
2596 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2597                                         unsigned int start)
2598 {
2599         struct sit_info *sit_i = SIT_I(sbi);
2600         struct page *src_page, *dst_page;
2601         pgoff_t src_off, dst_off;
2602         void *src_addr, *dst_addr;
2603
2604         src_off = current_sit_addr(sbi, start);
2605         dst_off = next_sit_addr(sbi, src_off);
2606
2607         /* get current sit block page without lock */
2608         src_page = get_meta_page(sbi, src_off);
2609         dst_page = grab_meta_page(sbi, dst_off);
2610         f2fs_bug_on(sbi, PageDirty(src_page));
2611
2612         src_addr = page_address(src_page);
2613         dst_addr = page_address(dst_page);
2614         memcpy(dst_addr, src_addr, PAGE_SIZE);
2615
2616         set_page_dirty(dst_page);
2617         f2fs_put_page(src_page, 1);
2618
2619         set_to_next_sit(sit_i, start);
2620
2621         return dst_page;
2622 }
2623
2624 static struct sit_entry_set *grab_sit_entry_set(void)
2625 {
2626         struct sit_entry_set *ses =
2627                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
2628
2629         ses->entry_cnt = 0;
2630         INIT_LIST_HEAD(&ses->set_list);
2631         return ses;
2632 }
2633
2634 static void release_sit_entry_set(struct sit_entry_set *ses)
2635 {
2636         list_del(&ses->set_list);
2637         kmem_cache_free(sit_entry_set_slab, ses);
2638 }
2639
2640 static void adjust_sit_entry_set(struct sit_entry_set *ses,
2641                                                 struct list_head *head)
2642 {
2643         struct sit_entry_set *next = ses;
2644
2645         if (list_is_last(&ses->set_list, head))
2646                 return;
2647
2648         list_for_each_entry_continue(next, head, set_list)
2649                 if (ses->entry_cnt <= next->entry_cnt)
2650                         break;
2651
2652         list_move_tail(&ses->set_list, &next->set_list);
2653 }
2654
2655 static void add_sit_entry(unsigned int segno, struct list_head *head)
2656 {
2657         struct sit_entry_set *ses;
2658         unsigned int start_segno = START_SEGNO(segno);
2659
2660         list_for_each_entry(ses, head, set_list) {
2661                 if (ses->start_segno == start_segno) {
2662                         ses->entry_cnt++;
2663                         adjust_sit_entry_set(ses, head);
2664                         return;
2665                 }
2666         }
2667
2668         ses = grab_sit_entry_set();
2669
2670         ses->start_segno = start_segno;
2671         ses->entry_cnt++;
2672         list_add(&ses->set_list, head);
2673 }
2674
2675 static void add_sits_in_set(struct f2fs_sb_info *sbi)
2676 {
2677         struct f2fs_sm_info *sm_info = SM_I(sbi);
2678         struct list_head *set_list = &sm_info->sit_entry_set;
2679         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
2680         unsigned int segno;
2681
2682         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
2683                 add_sit_entry(segno, set_list);
2684 }
2685
2686 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
2687 {
2688         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2689         struct f2fs_journal *journal = curseg->journal;
2690         int i;
2691
2692         down_write(&curseg->journal_rwsem);
2693         for (i = 0; i < sits_in_cursum(journal); i++) {
2694                 unsigned int segno;
2695                 bool dirtied;
2696
2697                 segno = le32_to_cpu(segno_in_journal(journal, i));
2698                 dirtied = __mark_sit_entry_dirty(sbi, segno);
2699
2700                 if (!dirtied)
2701                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
2702         }
2703         update_sits_in_cursum(journal, -i);
2704         up_write(&curseg->journal_rwsem);
2705 }
2706
2707 /*
2708  * CP calls this function, which flushes SIT entries including sit_journal,
2709  * and moves prefree segs to free segs.
2710  */
2711 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2712 {
2713         struct sit_info *sit_i = SIT_I(sbi);
2714         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2715         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2716         struct f2fs_journal *journal = curseg->journal;
2717         struct sit_entry_set *ses, *tmp;
2718         struct list_head *head = &SM_I(sbi)->sit_entry_set;
2719         bool to_journal = true;
2720         struct seg_entry *se;
2721
2722         mutex_lock(&sit_i->sentry_lock);
2723
2724         if (!sit_i->dirty_sentries)
2725                 goto out;
2726
2727         /*
2728          * add and account sit entries of dirty bitmap in sit entry
2729          * set temporarily
2730          */
2731         add_sits_in_set(sbi);
2732
2733         /*
2734          * if there are no enough space in journal to store dirty sit
2735          * entries, remove all entries from journal and add and account
2736          * them in sit entry set.
2737          */
2738         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
2739                 remove_sits_in_journal(sbi);
2740
2741         /*
2742          * there are two steps to flush sit entries:
2743          * #1, flush sit entries to journal in current cold data summary block.
2744          * #2, flush sit entries to sit page.
2745          */
2746         list_for_each_entry_safe(ses, tmp, head, set_list) {
2747                 struct page *page = NULL;
2748                 struct f2fs_sit_block *raw_sit = NULL;
2749                 unsigned int start_segno = ses->start_segno;
2750                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
2751                                                 (unsigned long)MAIN_SEGS(sbi));
2752                 unsigned int segno = start_segno;
2753
2754                 if (to_journal &&
2755                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
2756                         to_journal = false;
2757
2758                 if (to_journal) {
2759                         down_write(&curseg->journal_rwsem);
2760                 } else {
2761                         page = get_next_sit_page(sbi, start_segno);
2762                         raw_sit = page_address(page);
2763                 }
2764
2765                 /* flush dirty sit entries in region of current sit set */
2766                 for_each_set_bit_from(segno, bitmap, end) {
2767                         int offset, sit_offset;
2768
2769                         se = get_seg_entry(sbi, segno);
2770
2771                         /* add discard candidates */
2772                         if (!(cpc->reason & CP_DISCARD)) {
2773                                 cpc->trim_start = segno;
2774                                 add_discard_addrs(sbi, cpc, false);
2775                         }
2776
2777                         if (to_journal) {
2778                                 offset = lookup_journal_in_cursum(journal,
2779                                                         SIT_JOURNAL, segno, 1);
2780                                 f2fs_bug_on(sbi, offset < 0);
2781                                 segno_in_journal(journal, offset) =
2782                                                         cpu_to_le32(segno);
2783                                 seg_info_to_raw_sit(se,
2784                                         &sit_in_journal(journal, offset));
2785                         } else {
2786                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2787                                 seg_info_to_raw_sit(se,
2788                                                 &raw_sit->entries[sit_offset]);
2789                         }
2790
2791                         __clear_bit(segno, bitmap);
2792                         sit_i->dirty_sentries--;
2793                         ses->entry_cnt--;
2794                 }
2795
2796                 if (to_journal)
2797                         up_write(&curseg->journal_rwsem);
2798                 else
2799                         f2fs_put_page(page, 1);
2800
2801                 f2fs_bug_on(sbi, ses->entry_cnt);
2802                 release_sit_entry_set(ses);
2803         }
2804
2805         f2fs_bug_on(sbi, !list_empty(head));
2806         f2fs_bug_on(sbi, sit_i->dirty_sentries);
2807 out:
2808         if (cpc->reason & CP_DISCARD) {
2809                 __u64 trim_start = cpc->trim_start;
2810
2811                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2812                         add_discard_addrs(sbi, cpc, false);
2813
2814                 cpc->trim_start = trim_start;
2815         }
2816         mutex_unlock(&sit_i->sentry_lock);
2817
2818         set_prefree_as_free_segments(sbi);
2819 }
2820
2821 static int build_sit_info(struct f2fs_sb_info *sbi)
2822 {
2823         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2824         struct sit_info *sit_i;
2825         unsigned int sit_segs, start;
2826         char *src_bitmap;
2827         unsigned int bitmap_size;
2828
2829         /* allocate memory for SIT information */
2830         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2831         if (!sit_i)
2832                 return -ENOMEM;
2833
2834         SM_I(sbi)->sit_info = sit_i;
2835
2836         sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2837                                         sizeof(struct seg_entry), GFP_KERNEL);
2838         if (!sit_i->sentries)
2839                 return -ENOMEM;
2840
2841         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2842         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2843         if (!sit_i->dirty_sentries_bitmap)
2844                 return -ENOMEM;
2845
2846         for (start = 0; start < MAIN_SEGS(sbi); start++) {
2847                 sit_i->sentries[start].cur_valid_map
2848                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2849                 sit_i->sentries[start].ckpt_valid_map
2850                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2851                 if (!sit_i->sentries[start].cur_valid_map ||
2852                                 !sit_i->sentries[start].ckpt_valid_map)
2853                         return -ENOMEM;
2854
2855 #ifdef CONFIG_F2FS_CHECK_FS
2856                 sit_i->sentries[start].cur_valid_map_mir
2857                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2858                 if (!sit_i->sentries[start].cur_valid_map_mir)
2859                         return -ENOMEM;
2860 #endif
2861
2862                 if (f2fs_discard_en(sbi)) {
2863                         sit_i->sentries[start].discard_map
2864                                 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2865                         if (!sit_i->sentries[start].discard_map)
2866                                 return -ENOMEM;
2867                 }
2868         }
2869
2870         sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2871         if (!sit_i->tmp_map)
2872                 return -ENOMEM;
2873
2874         if (sbi->segs_per_sec > 1) {
2875                 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2876                                         sizeof(struct sec_entry), GFP_KERNEL);
2877                 if (!sit_i->sec_entries)
2878                         return -ENOMEM;
2879         }
2880
2881         /* get information related with SIT */
2882         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2883
2884         /* setup SIT bitmap from ckeckpoint pack */
2885         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2886         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2887
2888         sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2889         if (!sit_i->sit_bitmap)
2890                 return -ENOMEM;
2891
2892 #ifdef CONFIG_F2FS_CHECK_FS
2893         sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2894         if (!sit_i->sit_bitmap_mir)
2895                 return -ENOMEM;
2896 #endif
2897
2898         /* init SIT information */
2899         sit_i->s_ops = &default_salloc_ops;
2900
2901         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2902         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2903         sit_i->written_valid_blocks = 0;
2904         sit_i->bitmap_size = bitmap_size;
2905         sit_i->dirty_sentries = 0;
2906         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2907         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2908         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2909         mutex_init(&sit_i->sentry_lock);
2910         return 0;
2911 }
2912
2913 static int build_free_segmap(struct f2fs_sb_info *sbi)
2914 {
2915         struct free_segmap_info *free_i;
2916         unsigned int bitmap_size, sec_bitmap_size;
2917
2918         /* allocate memory for free segmap information */
2919         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2920         if (!free_i)
2921                 return -ENOMEM;
2922
2923         SM_I(sbi)->free_info = free_i;
2924
2925         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2926         free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2927         if (!free_i->free_segmap)
2928                 return -ENOMEM;
2929
2930         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2931         free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2932         if (!free_i->free_secmap)
2933                 return -ENOMEM;
2934
2935         /* set all segments as dirty temporarily */
2936         memset(free_i->free_segmap, 0xff, bitmap_size);
2937         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2938
2939         /* init free segmap information */
2940         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2941         free_i->free_segments = 0;
2942         free_i->free_sections = 0;
2943         spin_lock_init(&free_i->segmap_lock);
2944         return 0;
2945 }
2946
2947 static int build_curseg(struct f2fs_sb_info *sbi)
2948 {
2949         struct curseg_info *array;
2950         int i;
2951
2952         array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2953         if (!array)
2954                 return -ENOMEM;
2955
2956         SM_I(sbi)->curseg_array = array;
2957
2958         for (i = 0; i < NR_CURSEG_TYPE; i++) {
2959                 mutex_init(&array[i].curseg_mutex);
2960                 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2961                 if (!array[i].sum_blk)
2962                         return -ENOMEM;
2963                 init_rwsem(&array[i].journal_rwsem);
2964                 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2965                                                         GFP_KERNEL);
2966                 if (!array[i].journal)
2967                         return -ENOMEM;
2968                 array[i].segno = NULL_SEGNO;
2969                 array[i].next_blkoff = 0;
2970         }
2971         return restore_curseg_summaries(sbi);
2972 }
2973
2974 static void build_sit_entries(struct f2fs_sb_info *sbi)
2975 {
2976         struct sit_info *sit_i = SIT_I(sbi);
2977         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2978         struct f2fs_journal *journal = curseg->journal;
2979         struct seg_entry *se;
2980         struct f2fs_sit_entry sit;
2981         int sit_blk_cnt = SIT_BLK_CNT(sbi);
2982         unsigned int i, start, end;
2983         unsigned int readed, start_blk = 0;
2984
2985         do {
2986                 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
2987                                                         META_SIT, true);
2988
2989                 start = start_blk * sit_i->sents_per_block;
2990                 end = (start_blk + readed) * sit_i->sents_per_block;
2991
2992                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2993                         struct f2fs_sit_block *sit_blk;
2994                         struct page *page;
2995
2996                         se = &sit_i->sentries[start];
2997                         page = get_current_sit_page(sbi, start);
2998                         sit_blk = (struct f2fs_sit_block *)page_address(page);
2999                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3000                         f2fs_put_page(page, 1);
3001
3002                         check_block_count(sbi, start, &sit);
3003                         seg_info_from_raw_sit(se, &sit);
3004
3005                         /* build discard map only one time */
3006                         if (f2fs_discard_en(sbi)) {
3007                                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3008                                         memset(se->discard_map, 0xff,
3009                                                 SIT_VBLOCK_MAP_SIZE);
3010                                 } else {
3011                                         memcpy(se->discard_map,
3012                                                 se->cur_valid_map,
3013                                                 SIT_VBLOCK_MAP_SIZE);
3014                                         sbi->discard_blks +=
3015                                                 sbi->blocks_per_seg -
3016                                                 se->valid_blocks;
3017                                 }
3018                         }
3019
3020                         if (sbi->segs_per_sec > 1)
3021                                 get_sec_entry(sbi, start)->valid_blocks +=
3022                                                         se->valid_blocks;
3023                 }
3024                 start_blk += readed;
3025         } while (start_blk < sit_blk_cnt);
3026
3027         down_read(&curseg->journal_rwsem);
3028         for (i = 0; i < sits_in_cursum(journal); i++) {
3029                 unsigned int old_valid_blocks;
3030
3031                 start = le32_to_cpu(segno_in_journal(journal, i));
3032                 se = &sit_i->sentries[start];
3033                 sit = sit_in_journal(journal, i);
3034
3035                 old_valid_blocks = se->valid_blocks;
3036
3037                 check_block_count(sbi, start, &sit);
3038                 seg_info_from_raw_sit(se, &sit);
3039
3040                 if (f2fs_discard_en(sbi)) {
3041                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3042                                 memset(se->discard_map, 0xff,
3043                                                         SIT_VBLOCK_MAP_SIZE);
3044                         } else {
3045                                 memcpy(se->discard_map, se->cur_valid_map,
3046                                                         SIT_VBLOCK_MAP_SIZE);
3047                                 sbi->discard_blks += old_valid_blocks -
3048                                                         se->valid_blocks;
3049                         }
3050                 }
3051
3052                 if (sbi->segs_per_sec > 1)
3053                         get_sec_entry(sbi, start)->valid_blocks +=
3054                                 se->valid_blocks - old_valid_blocks;
3055         }
3056         up_read(&curseg->journal_rwsem);
3057 }
3058
3059 static void init_free_segmap(struct f2fs_sb_info *sbi)
3060 {
3061         unsigned int start;
3062         int type;
3063
3064         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3065                 struct seg_entry *sentry = get_seg_entry(sbi, start);
3066                 if (!sentry->valid_blocks)
3067                         __set_free(sbi, start);
3068                 else
3069                         SIT_I(sbi)->written_valid_blocks +=
3070                                                 sentry->valid_blocks;
3071         }
3072
3073         /* set use the current segments */
3074         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3075                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3076                 __set_test_and_inuse(sbi, curseg_t->segno);
3077         }
3078 }
3079
3080 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3081 {
3082         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3083         struct free_segmap_info *free_i = FREE_I(sbi);
3084         unsigned int segno = 0, offset = 0;
3085         unsigned short valid_blocks;
3086
3087         while (1) {
3088                 /* find dirty segment based on free segmap */
3089                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3090                 if (segno >= MAIN_SEGS(sbi))
3091                         break;
3092                 offset = segno + 1;
3093                 valid_blocks = get_valid_blocks(sbi, segno, false);
3094                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
3095                         continue;
3096                 if (valid_blocks > sbi->blocks_per_seg) {
3097                         f2fs_bug_on(sbi, 1);
3098                         continue;
3099                 }
3100                 mutex_lock(&dirty_i->seglist_lock);
3101                 __locate_dirty_segment(sbi, segno, DIRTY);
3102                 mutex_unlock(&dirty_i->seglist_lock);
3103         }
3104 }
3105
3106 static int init_victim_secmap(struct f2fs_sb_info *sbi)
3107 {
3108         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3109         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3110
3111         dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
3112         if (!dirty_i->victim_secmap)
3113                 return -ENOMEM;
3114         return 0;
3115 }
3116
3117 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3118 {
3119         struct dirty_seglist_info *dirty_i;
3120         unsigned int bitmap_size, i;
3121
3122         /* allocate memory for dirty segments list information */
3123         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
3124         if (!dirty_i)
3125                 return -ENOMEM;
3126
3127         SM_I(sbi)->dirty_info = dirty_i;
3128         mutex_init(&dirty_i->seglist_lock);
3129
3130         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3131
3132         for (i = 0; i < NR_DIRTY_TYPE; i++) {
3133                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
3134                 if (!dirty_i->dirty_segmap[i])
3135                         return -ENOMEM;
3136         }
3137
3138         init_dirty_segmap(sbi);
3139         return init_victim_secmap(sbi);
3140 }
3141
3142 /*
3143  * Update min, max modified time for cost-benefit GC algorithm
3144  */
3145 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3146 {
3147         struct sit_info *sit_i = SIT_I(sbi);
3148         unsigned int segno;
3149
3150         mutex_lock(&sit_i->sentry_lock);
3151
3152         sit_i->min_mtime = LLONG_MAX;
3153
3154         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
3155                 unsigned int i;
3156                 unsigned long long mtime = 0;
3157
3158                 for (i = 0; i < sbi->segs_per_sec; i++)
3159                         mtime += get_seg_entry(sbi, segno + i)->mtime;
3160
3161                 mtime = div_u64(mtime, sbi->segs_per_sec);
3162
3163                 if (sit_i->min_mtime > mtime)
3164                         sit_i->min_mtime = mtime;
3165         }
3166         sit_i->max_mtime = get_mtime(sbi);
3167         mutex_unlock(&sit_i->sentry_lock);
3168 }
3169
3170 int build_segment_manager(struct f2fs_sb_info *sbi)
3171 {
3172         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3173         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3174         struct f2fs_sm_info *sm_info;
3175         int err;
3176
3177         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
3178         if (!sm_info)
3179                 return -ENOMEM;
3180
3181         /* init sm info */
3182         sbi->sm_info = sm_info;
3183         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3184         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3185         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3186         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3187         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3188         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3189         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
3190         sm_info->rec_prefree_segments = sm_info->main_segments *
3191                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
3192         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3193                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3194
3195         if (!test_opt(sbi, LFS))
3196                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
3197         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
3198         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
3199         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
3200
3201         sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
3202
3203         INIT_LIST_HEAD(&sm_info->sit_entry_set);
3204
3205         if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
3206                 err = create_flush_cmd_control(sbi);
3207                 if (err)
3208                         return err;
3209         }
3210
3211         err = create_discard_cmd_control(sbi);
3212         if (err)
3213                 return err;
3214
3215         err = build_sit_info(sbi);
3216         if (err)
3217                 return err;
3218         err = build_free_segmap(sbi);
3219         if (err)
3220                 return err;
3221         err = build_curseg(sbi);
3222         if (err)
3223                 return err;
3224
3225         /* reinit free segmap based on SIT */
3226         build_sit_entries(sbi);
3227
3228         init_free_segmap(sbi);
3229         err = build_dirty_segmap(sbi);
3230         if (err)
3231                 return err;
3232
3233         init_min_max_mtime(sbi);
3234         return 0;
3235 }
3236
3237 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3238                 enum dirty_type dirty_type)
3239 {
3240         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3241
3242         mutex_lock(&dirty_i->seglist_lock);
3243         kvfree(dirty_i->dirty_segmap[dirty_type]);
3244         dirty_i->nr_dirty[dirty_type] = 0;
3245         mutex_unlock(&dirty_i->seglist_lock);
3246 }
3247
3248 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
3249 {
3250         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3251         kvfree(dirty_i->victim_secmap);
3252 }
3253
3254 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3255 {
3256         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3257         int i;
3258
3259         if (!dirty_i)
3260                 return;
3261
3262         /* discard pre-free/dirty segments list */
3263         for (i = 0; i < NR_DIRTY_TYPE; i++)
3264                 discard_dirty_segmap(sbi, i);
3265
3266         destroy_victim_secmap(sbi);
3267         SM_I(sbi)->dirty_info = NULL;
3268         kfree(dirty_i);
3269 }
3270
3271 static void destroy_curseg(struct f2fs_sb_info *sbi)
3272 {
3273         struct curseg_info *array = SM_I(sbi)->curseg_array;
3274         int i;
3275
3276         if (!array)
3277                 return;
3278         SM_I(sbi)->curseg_array = NULL;
3279         for (i = 0; i < NR_CURSEG_TYPE; i++) {
3280                 kfree(array[i].sum_blk);
3281                 kfree(array[i].journal);
3282         }
3283         kfree(array);
3284 }
3285
3286 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3287 {
3288         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3289         if (!free_i)
3290                 return;
3291         SM_I(sbi)->free_info = NULL;
3292         kvfree(free_i->free_segmap);
3293         kvfree(free_i->free_secmap);
3294         kfree(free_i);
3295 }
3296
3297 static void destroy_sit_info(struct f2fs_sb_info *sbi)
3298 {
3299         struct sit_info *sit_i = SIT_I(sbi);
3300         unsigned int start;
3301
3302         if (!sit_i)
3303                 return;
3304
3305         if (sit_i->sentries) {
3306                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3307                         kfree(sit_i->sentries[start].cur_valid_map);
3308 #ifdef CONFIG_F2FS_CHECK_FS
3309                         kfree(sit_i->sentries[start].cur_valid_map_mir);
3310 #endif
3311                         kfree(sit_i->sentries[start].ckpt_valid_map);
3312                         kfree(sit_i->sentries[start].discard_map);
3313                 }
3314         }
3315         kfree(sit_i->tmp_map);
3316
3317         kvfree(sit_i->sentries);
3318         kvfree(sit_i->sec_entries);
3319         kvfree(sit_i->dirty_sentries_bitmap);
3320
3321         SM_I(sbi)->sit_info = NULL;
3322         kfree(sit_i->sit_bitmap);
3323 #ifdef CONFIG_F2FS_CHECK_FS
3324         kfree(sit_i->sit_bitmap_mir);
3325 #endif
3326         kfree(sit_i);
3327 }
3328
3329 void destroy_segment_manager(struct f2fs_sb_info *sbi)
3330 {
3331         struct f2fs_sm_info *sm_info = SM_I(sbi);
3332
3333         if (!sm_info)
3334                 return;
3335         destroy_flush_cmd_control(sbi, true);
3336         destroy_discard_cmd_control(sbi);
3337         destroy_dirty_segmap(sbi);
3338         destroy_curseg(sbi);
3339         destroy_free_segmap(sbi);
3340         destroy_sit_info(sbi);
3341         sbi->sm_info = NULL;
3342         kfree(sm_info);
3343 }
3344
3345 int __init create_segment_manager_caches(void)
3346 {
3347         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
3348                         sizeof(struct discard_entry));
3349         if (!discard_entry_slab)
3350                 goto fail;
3351
3352         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3353                         sizeof(struct discard_cmd));
3354         if (!discard_cmd_slab)
3355                 goto destroy_discard_entry;
3356
3357         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
3358                         sizeof(struct sit_entry_set));
3359         if (!sit_entry_set_slab)
3360                 goto destroy_discard_cmd;
3361
3362         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3363                         sizeof(struct inmem_pages));
3364         if (!inmem_entry_slab)
3365                 goto destroy_sit_entry_set;
3366         return 0;
3367
3368 destroy_sit_entry_set:
3369         kmem_cache_destroy(sit_entry_set_slab);
3370 destroy_discard_cmd:
3371         kmem_cache_destroy(discard_cmd_slab);
3372 destroy_discard_entry:
3373         kmem_cache_destroy(discard_entry_slab);
3374 fail:
3375         return -ENOMEM;
3376 }
3377
3378 void destroy_segment_manager_caches(void)
3379 {
3380         kmem_cache_destroy(sit_entry_set_slab);
3381         kmem_cache_destroy(discard_cmd_slab);
3382         kmem_cache_destroy(discard_entry_slab);
3383         kmem_cache_destroy(inmem_entry_slab);
3384 }