fs/f2fs/segment.c

   1 /*
   2  * fs/f2fs/segment.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/bio.h>
  14 #include <linux/blkdev.h>
  15 #include <linux/prefetch.h>
  16 #include <linux/kthread.h>
  17 #include <linux/swap.h>
  18 #include <linux/timer.h>
  19 #include <linux/freezer.h>
  20 #include <linux/sched/signal.h>
  21
  22 #include "f2fs.h"
  23 #include "segment.h"
  24 #include "node.h"
  25 #include "gc.h"
  26 #include "trace.h"
  27 #include <trace/events/f2fs.h>
  28
  29 #define __reverse_ffz(x) __reverse_ffs(~(x))
  30
  31 static struct kmem_cache *discard_entry_slab;
  32 static struct kmem_cache *discard_cmd_slab;
  33 static struct kmem_cache *sit_entry_set_slab;
  34 static struct kmem_cache *inmem_entry_slab;
  35
  36 static unsigned long __reverse_ulong(unsigned char *str)
  37 {
  38         unsigned long tmp = 0;
  39         int shift = 24, idx = 0;
  40
  41 #if BITS_PER_LONG == 64
  42         shift = 56;
  43 #endif
  44         while (shift >= 0) {
  45                 tmp |= (unsigned long)str[idx++] << shift;
  46                 shift -= BITS_PER_BYTE;
  47         }
  48         return tmp;
  49 }
  50
  51 /*
  52  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  53  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  54  */
  55 static inline unsigned long __reverse_ffs(unsigned long word)
  56 {
  57         int num = 0;
  58
  59 #if BITS_PER_LONG == 64
  60         if ((word & 0xffffffff00000000UL) == 0)
  61                 num += 32;
  62         else
  63                 word >>= 32;
  64 #endif
  65         if ((word & 0xffff0000) == 0)
  66                 num += 16;
  67         else
  68                 word >>= 16;
  69
  70         if ((word & 0xff00) == 0)
  71                 num += 8;
  72         else
  73                 word >>= 8;
  74
  75         if ((word & 0xf0) == 0)
  76                 num += 4;
  77         else
  78                 word >>= 4;
  79
  80         if ((word & 0xc) == 0)
  81                 num += 2;
  82         else
  83                 word >>= 2;
  84
  85         if ((word & 0x2) == 0)
  86                 num += 1;
  87         return num;
  88 }
  89
  90 /*
  91  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  92  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  93  * @size must be integral times of unsigned long.
  94  * Example:
  95  *                             MSB <--> LSB
  96  *   f2fs_set_bit(0, bitmap) => 1000 0000
  97  *   f2fs_set_bit(7, bitmap) => 0000 0001
  98  */
  99 static unsigned long __find_rev_next_bit(const unsigned long *addr,
 100                         unsigned long size, unsigned long offset)
 101 {
 102         const unsigned long *p = addr + BIT_WORD(offset);
 103         unsigned long result = size;
 104         unsigned long tmp;
 105
 106         if (offset >= size)
 107                 return size;
 108
 109         size -= (offset & ~(BITS_PER_LONG - 1));
 110         offset %= BITS_PER_LONG;
 111
 112         while (1) {
 113                 if (*p == 0)
 114                         goto pass;
 115
 116                 tmp = __reverse_ulong((unsigned char *)p);
 117
 118                 tmp &= ~0UL >> offset;
 119                 if (size < BITS_PER_LONG)
 120                         tmp &= (~0UL << (BITS_PER_LONG - size));
 121                 if (tmp)
 122                         goto found;
 123 pass:
 124                 if (size <= BITS_PER_LONG)
 125                         break;
 126                 size -= BITS_PER_LONG;
 127                 offset = 0;
 128                 p++;
 129         }
 130         return result;
 131 found:
 132         return result - size + __reverse_ffs(tmp);
 133 }
 134
 135 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 136                         unsigned long size, unsigned long offset)
 137 {
 138         const unsigned long *p = addr + BIT_WORD(offset);
 139         unsigned long result = size;
 140         unsigned long tmp;
 141
 142         if (offset >= size)
 143                 return size;
 144
 145         size -= (offset & ~(BITS_PER_LONG - 1));
 146         offset %= BITS_PER_LONG;
 147
 148         while (1) {
 149                 if (*p == ~0UL)
 150                         goto pass;
 151
 152                 tmp = __reverse_ulong((unsigned char *)p);
 153
 154                 if (offset)
 155                         tmp |= ~0UL << (BITS_PER_LONG - offset);
 156                 if (size < BITS_PER_LONG)
 157                         tmp |= ~0UL >> size;
 158                 if (tmp != ~0UL)
 159                         goto found;
 160 pass:
 161                 if (size <= BITS_PER_LONG)
 162                         break;
 163                 size -= BITS_PER_LONG;
 164                 offset = 0;
 165                 p++;
 166         }
 167         return result;
 168 found:
 169         return result - size + __reverse_ffz(tmp);
 170 }
 171
 172 bool need_SSR(struct f2fs_sb_info *sbi)
 173 {
 174         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 175         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 176         int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
 177
 178         if (test_opt(sbi, LFS))
 179                 return false;
 180         if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
 181                 return true;
 182
 183         return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
 184                                                 2 * reserved_sections(sbi));
 185 }
 186
 187 void register_inmem_page(struct inode *inode, struct page *page)
 188 {
 189         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 190         struct f2fs_inode_info *fi = F2FS_I(inode);
 191         struct inmem_pages *new;
 192
 193         f2fs_trace_pid(page);
 194
 195         set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
 196         SetPagePrivate(page);
 197
 198         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 199
 200         /* add atomic page indices to the list */
 201         new->page = page;
 202         INIT_LIST_HEAD(&new->list);
 203
 204         /* increase reference count with clean state */
 205         mutex_lock(&fi->inmem_lock);
 206         get_page(page);
 207         list_add_tail(&new->list, &fi->inmem_pages);
 208         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 209         if (list_empty(&fi->inmem_ilist))
 210                 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
 211         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 212         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 213         mutex_unlock(&fi->inmem_lock);
 214
 215         trace_f2fs_register_inmem_page(page, INMEM);
 216 }
 217
 218 static int __revoke_inmem_pages(struct inode *inode,
 219                                 struct list_head *head, bool drop, bool recover)
 220 {
 221         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 222         struct inmem_pages *cur, *tmp;
 223         int err = 0;
 224
 225         list_for_each_entry_safe(cur, tmp, head, list) {
 226                 struct page *page = cur->page;
 227
 228                 if (drop)
 229                         trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 230
 231                 lock_page(page);
 232
 233                 if (recover) {
 234                         struct dnode_of_data dn;
 235                         struct node_info ni;
 236
 237                         trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 238 retry:
 239                         set_new_dnode(&dn, inode, NULL, NULL, 0);
 240                         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
 241                         if (err) {
 242                                 if (err == -ENOMEM) {
 243                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 244                                         cond_resched();
 245                                         goto retry;
 246                                 }
 247                                 err = -EAGAIN;
 248                                 goto next;
 249                         }
 250                         get_node_info(sbi, dn.nid, &ni);
 251                         f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 252                                         cur->old_addr, ni.version, true, true);
 253                         f2fs_put_dnode(&dn);
 254                 }
 255 next:
 256                 /* we don't need to invalidate this in the sccessful status */
 257                 if (drop || recover)
 258                         ClearPageUptodate(page);
 259                 set_page_private(page, 0);
 260                 ClearPagePrivate(page);
 261                 f2fs_put_page(page, 1);
 262
 263                 list_del(&cur->list);
 264                 kmem_cache_free(inmem_entry_slab, cur);
 265                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 266         }
 267         return err;
 268 }
 269
 270 void drop_inmem_pages_all(struct f2fs_sb_info *sbi)
 271 {
 272         struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
 273         struct inode *inode;
 274         struct f2fs_inode_info *fi;
 275 next:
 276         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 277         if (list_empty(head)) {
 278                 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 279                 return;
 280         }
 281         fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
 282         inode = igrab(&fi->vfs_inode);
 283         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 284
 285         if (inode) {
 286                 drop_inmem_pages(inode);
 287                 iput(inode);
 288         }
 289         congestion_wait(BLK_RW_ASYNC, HZ/50);
 290         cond_resched();
 291         goto next;
 292 }
 293
 294 void drop_inmem_pages(struct inode *inode)
 295 {
 296         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 297         struct f2fs_inode_info *fi = F2FS_I(inode);
 298
 299         mutex_lock(&fi->inmem_lock);
 300         __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 301         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 302         if (!list_empty(&fi->inmem_ilist))
 303                 list_del_init(&fi->inmem_ilist);
 304         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 305         mutex_unlock(&fi->inmem_lock);
 306
 307         clear_inode_flag(inode, FI_ATOMIC_FILE);
 308         clear_inode_flag(inode, FI_HOT_DATA);
 309         stat_dec_atomic_write(inode);
 310 }
 311
 312 void drop_inmem_page(struct inode *inode, struct page *page)
 313 {
 314         struct f2fs_inode_info *fi = F2FS_I(inode);
 315         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 316         struct list_head *head = &fi->inmem_pages;
 317         struct inmem_pages *cur = NULL;
 318
 319         f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
 320
 321         mutex_lock(&fi->inmem_lock);
 322         list_for_each_entry(cur, head, list) {
 323                 if (cur->page == page)
 324                         break;
 325         }
 326
 327         f2fs_bug_on(sbi, !cur || cur->page != page);
 328         list_del(&cur->list);
 329         mutex_unlock(&fi->inmem_lock);
 330
 331         dec_page_count(sbi, F2FS_INMEM_PAGES);
 332         kmem_cache_free(inmem_entry_slab, cur);
 333
 334         ClearPageUptodate(page);
 335         set_page_private(page, 0);
 336         ClearPagePrivate(page);
 337         f2fs_put_page(page, 0);
 338
 339         trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 340 }
 341
 342 static int __commit_inmem_pages(struct inode *inode,
 343                                         struct list_head *revoke_list)
 344 {
 345         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 346         struct f2fs_inode_info *fi = F2FS_I(inode);
 347         struct inmem_pages *cur, *tmp;
 348         struct f2fs_io_info fio = {
 349                 .sbi = sbi,
 350                 .ino = inode->i_ino,
 351                 .type = DATA,
 352                 .op = REQ_OP_WRITE,
 353                 .op_flags = REQ_SYNC | REQ_PRIO,
 354                 .io_type = FS_DATA_IO,
 355         };
 356         pgoff_t last_idx = ULONG_MAX;
 357         int err = 0;
 358
 359         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 360                 struct page *page = cur->page;
 361
 362                 lock_page(page);
 363                 if (page->mapping == inode->i_mapping) {
 364                         trace_f2fs_commit_inmem_page(page, INMEM);
 365
 366                         set_page_dirty(page);
 367                         f2fs_wait_on_page_writeback(page, DATA, true);
 368                         if (clear_page_dirty_for_io(page)) {
 369                                 inode_dec_dirty_pages(inode);
 370                                 remove_dirty_inode(inode);
 371                         }
 372 retry:
 373                         fio.page = page;
 374                         fio.old_blkaddr = NULL_ADDR;
 375                         fio.encrypted_page = NULL;
 376                         fio.need_lock = LOCK_DONE;
 377                         err = do_write_data_page(&fio);
 378                         if (err) {
 379                                 if (err == -ENOMEM) {
 380                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 381                                         cond_resched();
 382                                         goto retry;
 383                                 }
 384                                 unlock_page(page);
 385                                 break;
 386                         }
 387                         /* record old blkaddr for revoking */
 388                         cur->old_addr = fio.old_blkaddr;
 389                         last_idx = page->index;
 390                 }
 391                 unlock_page(page);
 392                 list_move_tail(&cur->list, revoke_list);
 393         }
 394
 395         if (last_idx != ULONG_MAX)
 396                 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
 397
 398         if (!err)
 399                 __revoke_inmem_pages(inode, revoke_list, false, false);
 400
 401         return err;
 402 }
 403
 404 int commit_inmem_pages(struct inode *inode)
 405 {
 406         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 407         struct f2fs_inode_info *fi = F2FS_I(inode);
 408         struct list_head revoke_list;
 409         int err;
 410
 411         INIT_LIST_HEAD(&revoke_list);
 412         f2fs_balance_fs(sbi, true);
 413         f2fs_lock_op(sbi);
 414
 415         set_inode_flag(inode, FI_ATOMIC_COMMIT);
 416
 417         mutex_lock(&fi->inmem_lock);
 418         err = __commit_inmem_pages(inode, &revoke_list);
 419         if (err) {
 420                 int ret;
 421                 /*
 422                  * try to revoke all committed pages, but still we could fail
 423                  * due to no memory or other reason, if that happened, EAGAIN
 424                  * will be returned, which means in such case, transaction is
 425                  * already not integrity, caller should use journal to do the
 426                  * recovery or rewrite & commit last transaction. For other
 427                  * error number, revoking was done by filesystem itself.
 428                  */
 429                 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
 430                 if (ret)
 431                         err = ret;
 432
 433                 /* drop all uncommitted pages */
 434                 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 435         }
 436         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 437         if (!list_empty(&fi->inmem_ilist))
 438                 list_del_init(&fi->inmem_ilist);
 439         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 440         mutex_unlock(&fi->inmem_lock);
 441
 442         clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 443
 444         f2fs_unlock_op(sbi);
 445         return err;
 446 }
 447
 448 /*
 449  * This function balances dirty node and dentry pages.
 450  * In addition, it controls garbage collection.
 451  */
 452 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 453 {
 454 #ifdef CONFIG_F2FS_FAULT_INJECTION
 455         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 456                 f2fs_show_injection_info(FAULT_CHECKPOINT);
 457                 f2fs_stop_checkpoint(sbi, false);
 458         }
 459 #endif
 460
 461         /* balance_fs_bg is able to be pending */
 462         if (need && excess_cached_nats(sbi))
 463                 f2fs_balance_fs_bg(sbi);
 464
 465         /*
 466          * We should do GC or end up with checkpoint, if there are so many dirty
 467          * dir/node pages without enough free segments.
 468          */
 469         if (has_not_enough_free_secs(sbi, 0, 0)) {
 470                 mutex_lock(&sbi->gc_mutex);
 471                 f2fs_gc(sbi, false, false, NULL_SEGNO);
 472         }
 473 }
 474
 475 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 476 {
 477         /* try to shrink extent cache when there is no enough memory */
 478         if (!available_free_memory(sbi, EXTENT_CACHE))
 479                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 480
 481         /* check the # of cached NAT entries */
 482         if (!available_free_memory(sbi, NAT_ENTRIES))
 483                 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 484
 485         if (!available_free_memory(sbi, FREE_NIDS))
 486                 try_to_free_nids(sbi, MAX_FREE_NIDS);
 487         else
 488                 build_free_nids(sbi, false, false);
 489
 490         if (!is_idle(sbi) && !excess_dirty_nats(sbi))
 491                 return;
 492
 493         /* checkpoint is the only way to shrink partial cached entries */
 494         if (!available_free_memory(sbi, NAT_ENTRIES) ||
 495                         !available_free_memory(sbi, INO_ENTRIES) ||
 496                         excess_prefree_segs(sbi) ||
 497                         excess_dirty_nats(sbi) ||
 498                         f2fs_time_over(sbi, CP_TIME)) {
 499                 if (test_opt(sbi, DATA_FLUSH)) {
 500                         struct blk_plug plug;
 501
 502                         blk_start_plug(&plug);
 503                         sync_dirty_inodes(sbi, FILE_INODE);
 504                         blk_finish_plug(&plug);
 505                 }
 506                 f2fs_sync_fs(sbi->sb, true);
 507                 stat_inc_bg_cp_count(sbi->stat_info);
 508         }
 509 }
 510
 511 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 512                                 struct block_device *bdev)
 513 {
 514         struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
 515         int ret;
 516
 517         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
 518         bio_set_dev(bio, bdev);
 519         ret = submit_bio_wait(bio);
 520         bio_put(bio);
 521
 522         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 523                                 test_opt(sbi, FLUSH_MERGE), ret);
 524         return ret;
 525 }
 526
 527 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 528 {
 529         int ret = 0;
 530         int i;
 531
 532         if (!sbi->s_ndevs)
 533                 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 534
 535         for (i = 0; i < sbi->s_ndevs; i++) {
 536                 if (!is_dirty_device(sbi, ino, i, FLUSH_INO))
 537                         continue;
 538                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 539                 if (ret)
 540                         break;
 541         }
 542         return ret;
 543 }
 544
 545 static int issue_flush_thread(void *data)
 546 {
 547         struct f2fs_sb_info *sbi = data;
 548         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 549         wait_queue_head_t *q = &fcc->flush_wait_queue;
 550 repeat:
 551         if (kthread_should_stop())
 552                 return 0;
 553
 554         sb_start_intwrite(sbi->sb);
 555
 556         if (!llist_empty(&fcc->issue_list)) {
 557                 struct flush_cmd *cmd, *next;
 558                 int ret;
 559
 560                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 561                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 562
 563                 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 564
 565                 ret = submit_flush_wait(sbi, cmd->ino);
 566                 atomic_inc(&fcc->issued_flush);
 567
 568                 llist_for_each_entry_safe(cmd, next,
 569                                           fcc->dispatch_list, llnode) {
 570                         cmd->ret = ret;
 571                         complete(&cmd->wait);
 572                 }
 573                 fcc->dispatch_list = NULL;
 574         }
 575
 576         sb_end_intwrite(sbi->sb);
 577
 578         wait_event_interruptible(*q,
 579                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 580         goto repeat;
 581 }
 582
 583 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 584 {
 585         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 586         struct flush_cmd cmd;
 587         int ret;
 588
 589         if (test_opt(sbi, NOBARRIER))
 590                 return 0;
 591
 592         if (!test_opt(sbi, FLUSH_MERGE)) {
 593                 ret = submit_flush_wait(sbi, ino);
 594                 atomic_inc(&fcc->issued_flush);
 595                 return ret;
 596         }
 597
 598         if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
 599                 ret = submit_flush_wait(sbi, ino);
 600                 atomic_dec(&fcc->issing_flush);
 601
 602                 atomic_inc(&fcc->issued_flush);
 603                 return ret;
 604         }
 605
 606         cmd.ino = ino;
 607         init_completion(&cmd.wait);
 608
 609         llist_add(&cmd.llnode, &fcc->issue_list);
 610
 611         /* update issue_list before we wake up issue_flush thread */
 612         smp_mb();
 613
 614         if (waitqueue_active(&fcc->flush_wait_queue))
 615                 wake_up(&fcc->flush_wait_queue);
 616
 617         if (fcc->f2fs_issue_flush) {
 618                 wait_for_completion(&cmd.wait);
 619                 atomic_dec(&fcc->issing_flush);
 620         } else {
 621                 struct llist_node *list;
 622
 623                 list = llist_del_all(&fcc->issue_list);
 624                 if (!list) {
 625                         wait_for_completion(&cmd.wait);
 626                         atomic_dec(&fcc->issing_flush);
 627                 } else {
 628                         struct flush_cmd *tmp, *next;
 629
 630                         ret = submit_flush_wait(sbi, ino);
 631
 632                         llist_for_each_entry_safe(tmp, next, list, llnode) {
 633                                 if (tmp == &cmd) {
 634                                         cmd.ret = ret;
 635                                         atomic_dec(&fcc->issing_flush);
 636                                         continue;
 637                                 }
 638                                 tmp->ret = ret;
 639                                 complete(&tmp->wait);
 640                         }
 641                 }
 642         }
 643
 644         return cmd.ret;
 645 }
 646
 647 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
 648 {
 649         dev_t dev = sbi->sb->s_bdev->bd_dev;
 650         struct flush_cmd_control *fcc;
 651         int err = 0;
 652
 653         if (SM_I(sbi)->fcc_info) {
 654                 fcc = SM_I(sbi)->fcc_info;
 655                 if (fcc->f2fs_issue_flush)
 656                         return err;
 657                 goto init_thread;
 658         }
 659
 660         fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
 661         if (!fcc)
 662                 return -ENOMEM;
 663         atomic_set(&fcc->issued_flush, 0);
 664         atomic_set(&fcc->issing_flush, 0);
 665         init_waitqueue_head(&fcc->flush_wait_queue);
 666         init_llist_head(&fcc->issue_list);
 667         SM_I(sbi)->fcc_info = fcc;
 668         if (!test_opt(sbi, FLUSH_MERGE))
 669                 return err;
 670
 671 init_thread:
 672         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 673                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 674         if (IS_ERR(fcc->f2fs_issue_flush)) {
 675                 err = PTR_ERR(fcc->f2fs_issue_flush);
 676                 kfree(fcc);
 677                 SM_I(sbi)->fcc_info = NULL;
 678                 return err;
 679         }
 680
 681         return err;
 682 }
 683
 684 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 685 {
 686         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 687
 688         if (fcc && fcc->f2fs_issue_flush) {
 689                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 690
 691                 fcc->f2fs_issue_flush = NULL;
 692                 kthread_stop(flush_thread);
 693         }
 694         if (free) {
 695                 kfree(fcc);
 696                 SM_I(sbi)->fcc_info = NULL;
 697         }
 698 }
 699
 700 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
 701 {
 702         int ret = 0, i;
 703
 704         if (!sbi->s_ndevs)
 705                 return 0;
 706
 707         for (i = 1; i < sbi->s_ndevs; i++) {
 708                 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
 709                         continue;
 710                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 711                 if (ret)
 712                         break;
 713
 714                 spin_lock(&sbi->dev_lock);
 715                 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
 716                 spin_unlock(&sbi->dev_lock);
 717         }
 718
 719         return ret;
 720 }
 721
 722 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 723                 enum dirty_type dirty_type)
 724 {
 725         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 726
 727         /* need not be added */
 728         if (IS_CURSEG(sbi, segno))
 729                 return;
 730
 731         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 732                 dirty_i->nr_dirty[dirty_type]++;
 733
 734         if (dirty_type == DIRTY) {
 735                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 736                 enum dirty_type t = sentry->type;
 737
 738                 if (unlikely(t >= DIRTY)) {
 739                         f2fs_bug_on(sbi, 1);
 740                         return;
 741                 }
 742                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 743                         dirty_i->nr_dirty[t]++;
 744         }
 745 }
 746
 747 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 748                 enum dirty_type dirty_type)
 749 {
 750         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 751
 752         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 753                 dirty_i->nr_dirty[dirty_type]--;
 754
 755         if (dirty_type == DIRTY) {
 756                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 757                 enum dirty_type t = sentry->type;
 758
 759                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 760                         dirty_i->nr_dirty[t]--;
 761
 762                 if (get_valid_blocks(sbi, segno, true) == 0)
 763                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 764                                                 dirty_i->victim_secmap);
 765         }
 766 }
 767
 768 /*
 769  * Should not occur error such as -ENOMEM.
 770  * Adding dirty entry into seglist is not critical operation.
 771  * If a given segment is one of current working segments, it won't be added.
 772  */
 773 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 774 {
 775         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 776         unsigned short valid_blocks;
 777
 778         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 779                 return;
 780
 781         mutex_lock(&dirty_i->seglist_lock);
 782
 783         valid_blocks = get_valid_blocks(sbi, segno, false);
 784
 785         if (valid_blocks == 0) {
 786                 __locate_dirty_segment(sbi, segno, PRE);
 787                 __remove_dirty_segment(sbi, segno, DIRTY);
 788         } else if (valid_blocks < sbi->blocks_per_seg) {
 789                 __locate_dirty_segment(sbi, segno, DIRTY);
 790         } else {
 791                 /* Recovery routine with SSR needs this */
 792                 __remove_dirty_segment(sbi, segno, DIRTY);
 793         }
 794
 795         mutex_unlock(&dirty_i->seglist_lock);
 796 }
 797
 798 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 799                 struct block_device *bdev, block_t lstart,
 800                 block_t start, block_t len)
 801 {
 802         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 803         struct list_head *pend_list;
 804         struct discard_cmd *dc;
 805
 806         f2fs_bug_on(sbi, !len);
 807
 808         pend_list = &dcc->pend_list[plist_idx(len)];
 809
 810         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 811         INIT_LIST_HEAD(&dc->list);
 812         dc->bdev = bdev;
 813         dc->lstart = lstart;
 814         dc->start = start;
 815         dc->len = len;
 816         dc->ref = 0;
 817         dc->state = D_PREP;
 818         dc->error = 0;
 819         init_completion(&dc->wait);
 820         list_add_tail(&dc->list, pend_list);
 821         atomic_inc(&dcc->discard_cmd_cnt);
 822         dcc->undiscard_blks += len;
 823
 824         return dc;
 825 }
 826
 827 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 828                                 struct block_device *bdev, block_t lstart,
 829                                 block_t start, block_t len,
 830                                 struct rb_node *parent, struct rb_node **p)
 831 {
 832         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 833         struct discard_cmd *dc;
 834
 835         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
 836
 837         rb_link_node(&dc->rb_node, parent, p);
 838         rb_insert_color(&dc->rb_node, &dcc->root);
 839
 840         return dc;
 841 }
 842
 843 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
 844                                                         struct discard_cmd *dc)
 845 {
 846         if (dc->state == D_DONE)
 847                 atomic_dec(&dcc->issing_discard);
 848
 849         list_del(&dc->list);
 850         rb_erase(&dc->rb_node, &dcc->root);
 851         dcc->undiscard_blks -= dc->len;
 852
 853         kmem_cache_free(discard_cmd_slab, dc);
 854
 855         atomic_dec(&dcc->discard_cmd_cnt);
 856 }
 857
 858 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
 859                                                         struct discard_cmd *dc)
 860 {
 861         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 862
 863         trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
 864
 865         f2fs_bug_on(sbi, dc->ref);
 866
 867         if (dc->error == -EOPNOTSUPP)
 868                 dc->error = 0;
 869
 870         if (dc->error)
 871                 f2fs_msg(sbi->sb, KERN_INFO,
 872                         "Issue discard(%u, %u, %u) failed, ret: %d",
 873                         dc->lstart, dc->start, dc->len, dc->error);
 874         __detach_discard_cmd(dcc, dc);
 875 }
 876
 877 static void f2fs_submit_discard_endio(struct bio *bio)
 878 {
 879         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
 880
 881         dc->error = blk_status_to_errno(bio->bi_status);
 882         dc->state = D_DONE;
 883         complete_all(&dc->wait);
 884         bio_put(bio);
 885 }
 886
 887 void __check_sit_bitmap(struct f2fs_sb_info *sbi,
 888                                 block_t start, block_t end)
 889 {
 890 #ifdef CONFIG_F2FS_CHECK_FS
 891         struct seg_entry *sentry;
 892         unsigned int segno;
 893         block_t blk = start;
 894         unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
 895         unsigned long *map;
 896
 897         while (blk < end) {
 898                 segno = GET_SEGNO(sbi, blk);
 899                 sentry = get_seg_entry(sbi, segno);
 900                 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
 901
 902                 if (end < START_BLOCK(sbi, segno + 1))
 903                         size = GET_BLKOFF_FROM_SEG0(sbi, end);
 904                 else
 905                         size = max_blocks;
 906                 map = (unsigned long *)(sentry->cur_valid_map);
 907                 offset = __find_rev_next_bit(map, size, offset);
 908                 f2fs_bug_on(sbi, offset != size);
 909                 blk = START_BLOCK(sbi, segno + 1);
 910         }
 911 #endif
 912 }
 913
 914 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
 915 static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
 916                                                 struct discard_policy *dpolicy,
 917                                                 struct discard_cmd *dc)
 918 {
 919         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 920         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
 921                                         &(dcc->fstrim_list) : &(dcc->wait_list);
 922         struct bio *bio = NULL;
 923         int flag = dpolicy->sync ? REQ_SYNC : 0;
 924
 925         if (dc->state != D_PREP)
 926                 return;
 927
 928         trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
 929
 930         dc->error = __blkdev_issue_discard(dc->bdev,
 931                                 SECTOR_FROM_BLOCK(dc->start),
 932                                 SECTOR_FROM_BLOCK(dc->len),
 933                                 GFP_NOFS, 0, &bio);
 934         if (!dc->error) {
 935                 /* should keep before submission to avoid D_DONE right away */
 936                 dc->state = D_SUBMIT;
 937                 atomic_inc(&dcc->issued_discard);
 938                 atomic_inc(&dcc->issing_discard);
 939                 if (bio) {
 940                         bio->bi_private = dc;
 941                         bio->bi_end_io = f2fs_submit_discard_endio;
 942                         bio->bi_opf |= flag;
 943                         submit_bio(bio);
 944                         list_move_tail(&dc->list, wait_list);
 945                         __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
 946
 947                         f2fs_update_iostat(sbi, FS_DISCARD, 1);
 948                 }
 949         } else {
 950                 __remove_discard_cmd(sbi, dc);
 951         }
 952 }
 953
 954 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
 955                                 struct block_device *bdev, block_t lstart,
 956                                 block_t start, block_t len,
 957                                 struct rb_node **insert_p,
 958                                 struct rb_node *insert_parent)
 959 {
 960         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 961         struct rb_node **p;
 962         struct rb_node *parent = NULL;
 963         struct discard_cmd *dc = NULL;
 964
 965         if (insert_p && insert_parent) {
 966                 parent = insert_parent;
 967                 p = insert_p;
 968                 goto do_insert;
 969         }
 970
 971         p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
 972 do_insert:
 973         dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
 974         if (!dc)
 975                 return NULL;
 976
 977         return dc;
 978 }
 979
 980 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
 981                                                 struct discard_cmd *dc)
 982 {
 983         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
 984 }
 985
 986 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
 987                                 struct discard_cmd *dc, block_t blkaddr)
 988 {
 989         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 990         struct discard_info di = dc->di;
 991         bool modified = false;
 992
 993         if (dc->state == D_DONE || dc->len == 1) {
 994                 __remove_discard_cmd(sbi, dc);
 995                 return;
 996         }
 997
 998         dcc->undiscard_blks -= di.len;
 999
1000         if (blkaddr > di.lstart) {
1001                 dc->len = blkaddr - dc->lstart;
1002                 dcc->undiscard_blks += dc->len;
1003                 __relocate_discard_cmd(dcc, dc);
1004                 modified = true;
1005         }
1006
1007         if (blkaddr < di.lstart + di.len - 1) {
1008                 if (modified) {
1009                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1010                                         di.start + blkaddr + 1 - di.lstart,
1011                                         di.lstart + di.len - 1 - blkaddr,
1012                                         NULL, NULL);
1013                 } else {
1014                         dc->lstart++;
1015                         dc->len--;
1016                         dc->start++;
1017                         dcc->undiscard_blks += dc->len;
1018                         __relocate_discard_cmd(dcc, dc);
1019                 }
1020         }
1021 }
1022
1023 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1024                                 struct block_device *bdev, block_t lstart,
1025                                 block_t start, block_t len)
1026 {
1027         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1028         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1029         struct discard_cmd *dc;
1030         struct discard_info di = {0};
1031         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1032         block_t end = lstart + len;
1033
1034         mutex_lock(&dcc->cmd_lock);
1035
1036         dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1037                                         NULL, lstart,
1038                                         (struct rb_entry **)&prev_dc,
1039                                         (struct rb_entry **)&next_dc,
1040                                         &insert_p, &insert_parent, true);
1041         if (dc)
1042                 prev_dc = dc;
1043
1044         if (!prev_dc) {
1045                 di.lstart = lstart;
1046                 di.len = next_dc ? next_dc->lstart - lstart : len;
1047                 di.len = min(di.len, len);
1048                 di.start = start;
1049         }
1050
1051         while (1) {
1052                 struct rb_node *node;
1053                 bool merged = false;
1054                 struct discard_cmd *tdc = NULL;
1055
1056                 if (prev_dc) {
1057                         di.lstart = prev_dc->lstart + prev_dc->len;
1058                         if (di.lstart < lstart)
1059                                 di.lstart = lstart;
1060                         if (di.lstart >= end)
1061                                 break;
1062
1063                         if (!next_dc || next_dc->lstart > end)
1064                                 di.len = end - di.lstart;
1065                         else
1066                                 di.len = next_dc->lstart - di.lstart;
1067                         di.start = start + di.lstart - lstart;
1068                 }
1069
1070                 if (!di.len)
1071                         goto next;
1072
1073                 if (prev_dc && prev_dc->state == D_PREP &&
1074                         prev_dc->bdev == bdev &&
1075                         __is_discard_back_mergeable(&di, &prev_dc->di)) {
1076                         prev_dc->di.len += di.len;
1077                         dcc->undiscard_blks += di.len;
1078                         __relocate_discard_cmd(dcc, prev_dc);
1079                         di = prev_dc->di;
1080                         tdc = prev_dc;
1081                         merged = true;
1082                 }
1083
1084                 if (next_dc && next_dc->state == D_PREP &&
1085                         next_dc->bdev == bdev &&
1086                         __is_discard_front_mergeable(&di, &next_dc->di)) {
1087                         next_dc->di.lstart = di.lstart;
1088                         next_dc->di.len += di.len;
1089                         next_dc->di.start = di.start;
1090                         dcc->undiscard_blks += di.len;
1091                         __relocate_discard_cmd(dcc, next_dc);
1092                         if (tdc)
1093                                 __remove_discard_cmd(sbi, tdc);
1094                         merged = true;
1095                 }
1096
1097                 if (!merged) {
1098                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1099                                                         di.len, NULL, NULL);
1100                 }
1101  next:
1102                 prev_dc = next_dc;
1103                 if (!prev_dc)
1104                         break;
1105
1106                 node = rb_next(&prev_dc->rb_node);
1107                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1108         }
1109
1110         mutex_unlock(&dcc->cmd_lock);
1111 }
1112
1113 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1114                 struct block_device *bdev, block_t blkstart, block_t blklen)
1115 {
1116         block_t lblkstart = blkstart;
1117
1118         trace_f2fs_queue_discard(bdev, blkstart, blklen);
1119
1120         if (sbi->s_ndevs) {
1121                 int devi = f2fs_target_device_index(sbi, blkstart);
1122
1123                 blkstart -= FDEV(devi).start_blk;
1124         }
1125         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1126         return 0;
1127 }
1128
1129 static void __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
1130                                         struct discard_policy *dpolicy,
1131                                         unsigned int start, unsigned int end)
1132 {
1133         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1134         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1135         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1136         struct discard_cmd *dc;
1137         struct blk_plug plug;
1138         int issued;
1139
1140 next:
1141         issued = 0;
1142
1143         mutex_lock(&dcc->cmd_lock);
1144         f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1145
1146         dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1147                                         NULL, start,
1148                                         (struct rb_entry **)&prev_dc,
1149                                         (struct rb_entry **)&next_dc,
1150                                         &insert_p, &insert_parent, true);
1151         if (!dc)
1152                 dc = next_dc;
1153
1154         blk_start_plug(&plug);
1155
1156         while (dc && dc->lstart <= end) {
1157                 struct rb_node *node;
1158
1159                 if (dc->len < dpolicy->granularity)
1160                         goto skip;
1161
1162                 if (dc->state != D_PREP) {
1163                         list_move_tail(&dc->list, &dcc->fstrim_list);
1164                         goto skip;
1165                 }
1166
1167                 __submit_discard_cmd(sbi, dpolicy, dc);
1168
1169                 if (++issued >= dpolicy->max_requests) {
1170                         start = dc->lstart + dc->len;
1171
1172                         blk_finish_plug(&plug);
1173                         mutex_unlock(&dcc->cmd_lock);
1174
1175                         schedule();
1176
1177                         goto next;
1178                 }
1179 skip:
1180                 node = rb_next(&dc->rb_node);
1181                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1182
1183                 if (fatal_signal_pending(current))
1184                         break;
1185         }
1186
1187         blk_finish_plug(&plug);
1188         mutex_unlock(&dcc->cmd_lock);
1189 }
1190
1191 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1192                                         struct discard_policy *dpolicy)
1193 {
1194         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1195         struct list_head *pend_list;
1196         struct discard_cmd *dc, *tmp;
1197         struct blk_plug plug;
1198         int i, iter = 0, issued = 0;
1199         bool io_interrupted = false;
1200
1201         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1202                 if (i + 1 < dpolicy->granularity)
1203                         break;
1204                 pend_list = &dcc->pend_list[i];
1205
1206                 mutex_lock(&dcc->cmd_lock);
1207                 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1208                 blk_start_plug(&plug);
1209                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1210                         f2fs_bug_on(sbi, dc->state != D_PREP);
1211
1212                         if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1213                                                                 !is_idle(sbi)) {
1214                                 io_interrupted = true;
1215                                 goto skip;
1216                         }
1217
1218                         __submit_discard_cmd(sbi, dpolicy, dc);
1219                         issued++;
1220 skip:
1221                         if (++iter >= dpolicy->max_requests)
1222                                 break;
1223                 }
1224                 blk_finish_plug(&plug);
1225                 mutex_unlock(&dcc->cmd_lock);
1226
1227                 if (iter >= dpolicy->max_requests)
1228                         break;
1229         }
1230
1231         if (!issued && io_interrupted)
1232                 issued = -1;
1233
1234         return issued;
1235 }
1236
1237 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1238 {
1239         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1240         struct list_head *pend_list;
1241         struct discard_cmd *dc, *tmp;
1242         int i;
1243         bool dropped = false;
1244
1245         mutex_lock(&dcc->cmd_lock);
1246         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1247                 pend_list = &dcc->pend_list[i];
1248                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1249                         f2fs_bug_on(sbi, dc->state != D_PREP);
1250                         __remove_discard_cmd(sbi, dc);
1251                         dropped = true;
1252                 }
1253         }
1254         mutex_unlock(&dcc->cmd_lock);
1255
1256         return dropped;
1257 }
1258
1259 static void __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1260                                                         struct discard_cmd *dc)
1261 {
1262         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1263
1264         wait_for_completion_io(&dc->wait);
1265         mutex_lock(&dcc->cmd_lock);
1266         f2fs_bug_on(sbi, dc->state != D_DONE);
1267         dc->ref--;
1268         if (!dc->ref)
1269                 __remove_discard_cmd(sbi, dc);
1270         mutex_unlock(&dcc->cmd_lock);
1271 }
1272
1273 static void __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1274                                                 struct discard_policy *dpolicy,
1275                                                 block_t start, block_t end)
1276 {
1277         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1278         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1279                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1280         struct discard_cmd *dc, *tmp;
1281         bool need_wait;
1282
1283 next:
1284         need_wait = false;
1285
1286         mutex_lock(&dcc->cmd_lock);
1287         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1288                 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1289                         continue;
1290                 if (dc->len < dpolicy->granularity)
1291                         continue;
1292                 if (dc->state == D_DONE && !dc->ref) {
1293                         wait_for_completion_io(&dc->wait);
1294                         __remove_discard_cmd(sbi, dc);
1295                 } else {
1296                         dc->ref++;
1297                         need_wait = true;
1298                         break;
1299                 }
1300         }
1301         mutex_unlock(&dcc->cmd_lock);
1302
1303         if (need_wait) {
1304                 __wait_one_discard_bio(sbi, dc);
1305                 goto next;
1306         }
1307 }
1308
1309 static void __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1310                                                 struct discard_policy *dpolicy)
1311 {
1312         __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1313 }
1314
1315 /* This should be covered by global mutex, &sit_i->sentry_lock */
1316 void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1317 {
1318         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1319         struct discard_cmd *dc;
1320         bool need_wait = false;
1321
1322         mutex_lock(&dcc->cmd_lock);
1323         dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1324         if (dc) {
1325                 if (dc->state == D_PREP) {
1326                         __punch_discard_cmd(sbi, dc, blkaddr);
1327                 } else {
1328                         dc->ref++;
1329                         need_wait = true;
1330                 }
1331         }
1332         mutex_unlock(&dcc->cmd_lock);
1333
1334         if (need_wait)
1335                 __wait_one_discard_bio(sbi, dc);
1336 }
1337
1338 void stop_discard_thread(struct f2fs_sb_info *sbi)
1339 {
1340         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1341
1342         if (dcc && dcc->f2fs_issue_discard) {
1343                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1344
1345                 dcc->f2fs_issue_discard = NULL;
1346                 kthread_stop(discard_thread);
1347         }
1348 }
1349
1350 /* This comes from f2fs_put_super */
1351 bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1352 {
1353         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1354         struct discard_policy dpolicy;
1355         bool dropped;
1356
1357         init_discard_policy(&dpolicy, DPOLICY_UMOUNT, dcc->discard_granularity);
1358         __issue_discard_cmd(sbi, &dpolicy);
1359         dropped = __drop_discard_cmd(sbi);
1360         __wait_all_discard_cmd(sbi, &dpolicy);
1361
1362         return dropped;
1363 }
1364
1365 static int issue_discard_thread(void *data)
1366 {
1367         struct f2fs_sb_info *sbi = data;
1368         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1369         wait_queue_head_t *q = &dcc->discard_wait_queue;
1370         struct discard_policy dpolicy;
1371         unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1372         int issued;
1373
1374         set_freezable();
1375
1376         do {
1377                 init_discard_policy(&dpolicy, DPOLICY_BG,
1378                                         dcc->discard_granularity);
1379
1380                 wait_event_interruptible_timeout(*q,
1381                                 kthread_should_stop() || freezing(current) ||
1382                                 dcc->discard_wake,
1383                                 msecs_to_jiffies(wait_ms));
1384                 if (try_to_freeze())
1385                         continue;
1386                 if (kthread_should_stop())
1387                         return 0;
1388
1389                 if (dcc->discard_wake) {
1390                         dcc->discard_wake = 0;
1391                         if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
1392                                 init_discard_policy(&dpolicy,
1393                                                         DPOLICY_FORCE, 1);
1394                 }
1395
1396                 sb_start_intwrite(sbi->sb);
1397
1398                 issued = __issue_discard_cmd(sbi, &dpolicy);
1399                 if (issued) {
1400                         __wait_all_discard_cmd(sbi, &dpolicy);
1401                         wait_ms = dpolicy.min_interval;
1402                 } else {
1403                         wait_ms = dpolicy.max_interval;
1404                 }
1405
1406                 sb_end_intwrite(sbi->sb);
1407
1408         } while (!kthread_should_stop());
1409         return 0;
1410 }
1411
1412 #ifdef CONFIG_BLK_DEV_ZONED
1413 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1414                 struct block_device *bdev, block_t blkstart, block_t blklen)
1415 {
1416         sector_t sector, nr_sects;
1417         block_t lblkstart = blkstart;
1418         int devi = 0;
1419
1420         if (sbi->s_ndevs) {
1421                 devi = f2fs_target_device_index(sbi, blkstart);
1422                 blkstart -= FDEV(devi).start_blk;
1423         }
1424
1425         /*
1426          * We need to know the type of the zone: for conventional zones,
1427          * use regular discard if the drive supports it. For sequential
1428          * zones, reset the zone write pointer.
1429          */
1430         switch (get_blkz_type(sbi, bdev, blkstart)) {
1431
1432         case BLK_ZONE_TYPE_CONVENTIONAL:
1433                 if (!blk_queue_discard(bdev_get_queue(bdev)))
1434                         return 0;
1435                 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1436         case BLK_ZONE_TYPE_SEQWRITE_REQ:
1437         case BLK_ZONE_TYPE_SEQWRITE_PREF:
1438                 sector = SECTOR_FROM_BLOCK(blkstart);
1439                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1440
1441                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1442                                 nr_sects != bdev_zone_sectors(bdev)) {
1443                         f2fs_msg(sbi->sb, KERN_INFO,
1444                                 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1445                                 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1446                                 blkstart, blklen);
1447                         return -EIO;
1448                 }
1449                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1450                 return blkdev_reset_zones(bdev, sector,
1451                                           nr_sects, GFP_NOFS);
1452         default:
1453                 /* Unknown zone type: broken device ? */
1454                 return -EIO;
1455         }
1456 }
1457 #endif
1458
1459 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1460                 struct block_device *bdev, block_t blkstart, block_t blklen)
1461 {
1462 #ifdef CONFIG_BLK_DEV_ZONED
1463         if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
1464                                 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1465                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1466 #endif
1467         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1468 }
1469
1470 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1471                                 block_t blkstart, block_t blklen)
1472 {
1473         sector_t start = blkstart, len = 0;
1474         struct block_device *bdev;
1475         struct seg_entry *se;
1476         unsigned int offset;
1477         block_t i;
1478         int err = 0;
1479
1480         bdev = f2fs_target_device(sbi, blkstart, NULL);
1481
1482         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1483                 if (i != start) {
1484                         struct block_device *bdev2 =
1485                                 f2fs_target_device(sbi, i, NULL);
1486
1487                         if (bdev2 != bdev) {
1488                                 err = __issue_discard_async(sbi, bdev,
1489                                                 start, len);
1490                                 if (err)
1491                                         return err;
1492                                 bdev = bdev2;
1493                                 start = i;
1494                                 len = 0;
1495                         }
1496                 }
1497
1498                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1499                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1500
1501                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1502                         sbi->discard_blks--;
1503         }
1504
1505         if (len)
1506                 err = __issue_discard_async(sbi, bdev, start, len);
1507         return err;
1508 }
1509
1510 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1511                                                         bool check_only)
1512 {
1513         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1514         int max_blocks = sbi->blocks_per_seg;
1515         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1516         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1517         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1518         unsigned long *discard_map = (unsigned long *)se->discard_map;
1519         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1520         unsigned int start = 0, end = -1;
1521         bool force = (cpc->reason & CP_DISCARD);
1522         struct discard_entry *de = NULL;
1523         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1524         int i;
1525
1526         if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
1527                 return false;
1528
1529         if (!force) {
1530                 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
1531                         SM_I(sbi)->dcc_info->nr_discards >=
1532                                 SM_I(sbi)->dcc_info->max_discards)
1533                         return false;
1534         }
1535
1536         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1537         for (i = 0; i < entries; i++)
1538                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1539                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1540
1541         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1542                                 SM_I(sbi)->dcc_info->max_discards) {
1543                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1544                 if (start >= max_blocks)
1545                         break;
1546
1547                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1548                 if (force && start && end != max_blocks
1549                                         && (end - start) < cpc->trim_minlen)
1550                         continue;
1551
1552                 if (check_only)
1553                         return true;
1554
1555                 if (!de) {
1556                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1557                                                                 GFP_F2FS_ZERO);
1558                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1559                         list_add_tail(&de->list, head);
1560                 }
1561
1562                 for (i = start; i < end; i++)
1563                         __set_bit_le(i, (void *)de->discard_map);
1564
1565                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1566         }
1567         return false;
1568 }
1569
1570 void release_discard_addrs(struct f2fs_sb_info *sbi)
1571 {
1572         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1573         struct discard_entry *entry, *this;
1574
1575         /* drop caches */
1576         list_for_each_entry_safe(entry, this, head, list) {
1577                 list_del(&entry->list);
1578                 kmem_cache_free(discard_entry_slab, entry);
1579         }
1580 }
1581
1582 /*
1583  * Should call clear_prefree_segments after checkpoint is done.
1584  */
1585 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1586 {
1587         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1588         unsigned int segno;
1589
1590         mutex_lock(&dirty_i->seglist_lock);
1591         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1592                 __set_test_and_free(sbi, segno);
1593         mutex_unlock(&dirty_i->seglist_lock);
1594 }
1595
1596 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1597 {
1598         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1599         struct list_head *head = &dcc->entry_list;
1600         struct discard_entry *entry, *this;
1601         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1602         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1603         unsigned int start = 0, end = -1;
1604         unsigned int secno, start_segno;
1605         bool force = (cpc->reason & CP_DISCARD);
1606
1607         mutex_lock(&dirty_i->seglist_lock);
1608
1609         while (1) {
1610                 int i;
1611                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1612                 if (start >= MAIN_SEGS(sbi))
1613                         break;
1614                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1615                                                                 start + 1);
1616
1617                 for (i = start; i < end; i++)
1618                         clear_bit(i, prefree_map);
1619
1620                 dirty_i->nr_dirty[PRE] -= end - start;
1621
1622                 if (!test_opt(sbi, DISCARD))
1623                         continue;
1624
1625                 if (force && start >= cpc->trim_start &&
1626                                         (end - 1) <= cpc->trim_end)
1627                                 continue;
1628
1629                 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1630                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1631                                 (end - start) << sbi->log_blocks_per_seg);
1632                         continue;
1633                 }
1634 next:
1635                 secno = GET_SEC_FROM_SEG(sbi, start);
1636                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1637                 if (!IS_CURSEC(sbi, secno) &&
1638                         !get_valid_blocks(sbi, start, true))
1639                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1640                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1641
1642                 start = start_segno + sbi->segs_per_sec;
1643                 if (start < end)
1644                         goto next;
1645                 else
1646                         end = start - 1;
1647         }
1648         mutex_unlock(&dirty_i->seglist_lock);
1649
1650         /* send small discards */
1651         list_for_each_entry_safe(entry, this, head, list) {
1652                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1653                 bool is_valid = test_bit_le(0, entry->discard_map);
1654
1655 find_next:
1656                 if (is_valid) {
1657                         next_pos = find_next_zero_bit_le(entry->discard_map,
1658                                         sbi->blocks_per_seg, cur_pos);
1659                         len = next_pos - cur_pos;
1660
1661                         if (f2fs_sb_mounted_blkzoned(sbi->sb) ||
1662                             (force && len < cpc->trim_minlen))
1663                                 goto skip;
1664
1665                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1666                                                                         len);
1667                         cpc->trimmed += len;
1668                         total_len += len;
1669                 } else {
1670                         next_pos = find_next_bit_le(entry->discard_map,
1671                                         sbi->blocks_per_seg, cur_pos);
1672                 }
1673 skip:
1674                 cur_pos = next_pos;
1675                 is_valid = !is_valid;
1676
1677                 if (cur_pos < sbi->blocks_per_seg)
1678                         goto find_next;
1679
1680                 list_del(&entry->list);
1681                 dcc->nr_discards -= total_len;
1682                 kmem_cache_free(discard_entry_slab, entry);
1683         }
1684
1685         wake_up_discard_thread(sbi, false);
1686 }
1687
1688 void init_discard_policy(struct discard_policy *dpolicy,
1689                                 int discard_type, unsigned int granularity)
1690 {
1691         /* common policy */
1692         dpolicy->type = discard_type;
1693         dpolicy->sync = true;
1694         dpolicy->granularity = granularity;
1695
1696         if (discard_type == DPOLICY_BG) {
1697                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1698                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1699                 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1700                 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1701                 dpolicy->io_aware = true;
1702         } else if (discard_type == DPOLICY_FORCE) {
1703                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1704                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1705                 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1706                 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1707                 dpolicy->io_aware = true;
1708         } else if (discard_type == DPOLICY_FSTRIM) {
1709                 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1710                 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1711                 dpolicy->io_aware = false;
1712         } else if (discard_type == DPOLICY_UMOUNT) {
1713                 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1714                 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1715                 dpolicy->io_aware = false;
1716         }
1717 }
1718
1719 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1720 {
1721         dev_t dev = sbi->sb->s_bdev->bd_dev;
1722         struct discard_cmd_control *dcc;
1723         int err = 0, i;
1724
1725         if (SM_I(sbi)->dcc_info) {
1726                 dcc = SM_I(sbi)->dcc_info;
1727                 goto init_thread;
1728         }
1729
1730         dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1731         if (!dcc)
1732                 return -ENOMEM;
1733
1734         dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
1735         INIT_LIST_HEAD(&dcc->entry_list);
1736         for (i = 0; i < MAX_PLIST_NUM; i++)
1737                 INIT_LIST_HEAD(&dcc->pend_list[i]);
1738         INIT_LIST_HEAD(&dcc->wait_list);
1739         INIT_LIST_HEAD(&dcc->fstrim_list);
1740         mutex_init(&dcc->cmd_lock);
1741         atomic_set(&dcc->issued_discard, 0);
1742         atomic_set(&dcc->issing_discard, 0);
1743         atomic_set(&dcc->discard_cmd_cnt, 0);
1744         dcc->nr_discards = 0;
1745         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1746         dcc->undiscard_blks = 0;
1747         dcc->root = RB_ROOT;
1748
1749         init_waitqueue_head(&dcc->discard_wait_queue);
1750         SM_I(sbi)->dcc_info = dcc;
1751 init_thread:
1752         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1753                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1754         if (IS_ERR(dcc->f2fs_issue_discard)) {
1755                 err = PTR_ERR(dcc->f2fs_issue_discard);
1756                 kfree(dcc);
1757                 SM_I(sbi)->dcc_info = NULL;
1758                 return err;
1759         }
1760
1761         return err;
1762 }
1763
1764 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1765 {
1766         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1767
1768         if (!dcc)
1769                 return;
1770
1771         stop_discard_thread(sbi);
1772
1773         kfree(dcc);
1774         SM_I(sbi)->dcc_info = NULL;
1775 }
1776
1777 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1778 {
1779         struct sit_info *sit_i = SIT_I(sbi);
1780
1781         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1782                 sit_i->dirty_sentries++;
1783                 return false;
1784         }
1785
1786         return true;
1787 }
1788
1789 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1790                                         unsigned int segno, int modified)
1791 {
1792         struct seg_entry *se = get_seg_entry(sbi, segno);
1793         se->type = type;
1794         if (modified)
1795                 __mark_sit_entry_dirty(sbi, segno);
1796 }
1797
1798 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1799 {
1800         struct seg_entry *se;
1801         unsigned int segno, offset;
1802         long int new_vblocks;
1803         bool exist;
1804 #ifdef CONFIG_F2FS_CHECK_FS
1805         bool mir_exist;
1806 #endif
1807
1808         segno = GET_SEGNO(sbi, blkaddr);
1809
1810         se = get_seg_entry(sbi, segno);
1811         new_vblocks = se->valid_blocks + del;
1812         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1813
1814         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
1815                                 (new_vblocks > sbi->blocks_per_seg)));
1816
1817         se->valid_blocks = new_vblocks;
1818         se->mtime = get_mtime(sbi);
1819         SIT_I(sbi)->max_mtime = se->mtime;
1820
1821         /* Update valid block bitmap */
1822         if (del > 0) {
1823                 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
1824 #ifdef CONFIG_F2FS_CHECK_FS
1825                 mir_exist = f2fs_test_and_set_bit(offset,
1826                                                 se->cur_valid_map_mir);
1827                 if (unlikely(exist != mir_exist)) {
1828                         f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1829                                 "when setting bitmap, blk:%u, old bit:%d",
1830                                 blkaddr, exist);
1831                         f2fs_bug_on(sbi, 1);
1832                 }
1833 #endif
1834                 if (unlikely(exist)) {
1835                         f2fs_msg(sbi->sb, KERN_ERR,
1836                                 "Bitmap was wrongly set, blk:%u", blkaddr);
1837                         f2fs_bug_on(sbi, 1);
1838                         se->valid_blocks--;
1839                         del = 0;
1840                 }
1841
1842                 if (f2fs_discard_en(sbi) &&
1843                         !f2fs_test_and_set_bit(offset, se->discard_map))
1844                         sbi->discard_blks--;
1845
1846                 /* don't overwrite by SSR to keep node chain */
1847                 if (se->type == CURSEG_WARM_NODE) {
1848                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1849                                 se->ckpt_valid_blocks++;
1850                 }
1851         } else {
1852                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
1853 #ifdef CONFIG_F2FS_CHECK_FS
1854                 mir_exist = f2fs_test_and_clear_bit(offset,
1855                                                 se->cur_valid_map_mir);
1856                 if (unlikely(exist != mir_exist)) {
1857                         f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1858                                 "when clearing bitmap, blk:%u, old bit:%d",
1859                                 blkaddr, exist);
1860                         f2fs_bug_on(sbi, 1);
1861                 }
1862 #endif
1863                 if (unlikely(!exist)) {
1864                         f2fs_msg(sbi->sb, KERN_ERR,
1865                                 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1866                         f2fs_bug_on(sbi, 1);
1867                         se->valid_blocks++;
1868                         del = 0;
1869                 }
1870
1871                 if (f2fs_discard_en(sbi) &&
1872                         f2fs_test_and_clear_bit(offset, se->discard_map))
1873                         sbi->discard_blks++;
1874         }
1875         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1876                 se->ckpt_valid_blocks += del;
1877
1878         __mark_sit_entry_dirty(sbi, segno);
1879
1880         /* update total number of valid blocks to be written in ckpt area */
1881         SIT_I(sbi)->written_valid_blocks += del;
1882
1883         if (sbi->segs_per_sec > 1)
1884                 get_sec_entry(sbi, segno)->valid_blocks += del;
1885 }
1886
1887 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
1888 {
1889         update_sit_entry(sbi, new, 1);
1890         if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1891                 update_sit_entry(sbi, old, -1);
1892
1893         locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1894         locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
1895 }
1896
1897 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1898 {
1899         unsigned int segno = GET_SEGNO(sbi, addr);
1900         struct sit_info *sit_i = SIT_I(sbi);
1901
1902         f2fs_bug_on(sbi, addr == NULL_ADDR);
1903         if (addr == NEW_ADDR)
1904                 return;
1905
1906         /* add it into sit main buffer */
1907         mutex_lock(&sit_i->sentry_lock);
1908
1909         update_sit_entry(sbi, addr, -1);
1910
1911         /* add it into dirty seglist */
1912         locate_dirty_segment(sbi, segno);
1913
1914         mutex_unlock(&sit_i->sentry_lock);
1915 }
1916
1917 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1918 {
1919         struct sit_info *sit_i = SIT_I(sbi);
1920         unsigned int segno, offset;
1921         struct seg_entry *se;
1922         bool is_cp = false;
1923
1924         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1925                 return true;
1926
1927         mutex_lock(&sit_i->sentry_lock);
1928
1929         segno = GET_SEGNO(sbi, blkaddr);
1930         se = get_seg_entry(sbi, segno);
1931         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1932
1933         if (f2fs_test_bit(offset, se->ckpt_valid_map))
1934                 is_cp = true;
1935
1936         mutex_unlock(&sit_i->sentry_lock);
1937
1938         return is_cp;
1939 }
1940
1941 /*
1942  * This function should be resided under the curseg_mutex lock
1943  */
1944 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1945                                         struct f2fs_summary *sum)
1946 {
1947         struct curseg_info *curseg = CURSEG_I(sbi, type);
1948         void *addr = curseg->sum_blk;
1949         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1950         memcpy(addr, sum, sizeof(struct f2fs_summary));
1951 }
1952
1953 /*
1954  * Calculate the number of current summary pages for writing
1955  */
1956 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1957 {
1958         int valid_sum_count = 0;
1959         int i, sum_in_page;
1960
1961         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1962                 if (sbi->ckpt->alloc_type[i] == SSR)
1963                         valid_sum_count += sbi->blocks_per_seg;
1964                 else {
1965                         if (for_ra)
1966                                 valid_sum_count += le16_to_cpu(
1967                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1968                         else
1969                                 valid_sum_count += curseg_blkoff(sbi, i);
1970                 }
1971         }
1972
1973         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1974                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1975         if (valid_sum_count <= sum_in_page)
1976                 return 1;
1977         else if ((valid_sum_count - sum_in_page) <=
1978                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1979                 return 2;
1980         return 3;
1981 }
1982
1983 /*
1984  * Caller should put this summary page
1985  */
1986 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1987 {
1988         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1989 }
1990
1991 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1992 {
1993         struct page *page = grab_meta_page(sbi, blk_addr);
1994         void *dst = page_address(page);
1995
1996         if (src)
1997                 memcpy(dst, src, PAGE_SIZE);
1998         else
1999                 memset(dst, 0, PAGE_SIZE);
2000         set_page_dirty(page);
2001         f2fs_put_page(page, 1);
2002 }
2003
2004 static void write_sum_page(struct f2fs_sb_info *sbi,
2005                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
2006 {
2007         update_meta_page(sbi, (void *)sum_blk, blk_addr);
2008 }
2009
2010 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2011                                                 int type, block_t blk_addr)
2012 {
2013         struct curseg_info *curseg = CURSEG_I(sbi, type);
2014         struct page *page = grab_meta_page(sbi, blk_addr);
2015         struct f2fs_summary_block *src = curseg->sum_blk;
2016         struct f2fs_summary_block *dst;
2017
2018         dst = (struct f2fs_summary_block *)page_address(page);
2019
2020         mutex_lock(&curseg->curseg_mutex);
2021
2022         down_read(&curseg->journal_rwsem);
2023         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2024         up_read(&curseg->journal_rwsem);
2025
2026         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2027         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2028
2029         mutex_unlock(&curseg->curseg_mutex);
2030
2031         set_page_dirty(page);
2032         f2fs_put_page(page, 1);
2033 }
2034
2035 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2036 {
2037         struct curseg_info *curseg = CURSEG_I(sbi, type);
2038         unsigned int segno = curseg->segno + 1;
2039         struct free_segmap_info *free_i = FREE_I(sbi);
2040
2041         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2042                 return !test_bit(segno, free_i->free_segmap);
2043         return 0;
2044 }
2045
2046 /*
2047  * Find a new segment from the free segments bitmap to right order
2048  * This function should be returned with success, otherwise BUG
2049  */
2050 static void get_new_segment(struct f2fs_sb_info *sbi,
2051                         unsigned int *newseg, bool new_sec, int dir)
2052 {
2053         struct free_segmap_info *free_i = FREE_I(sbi);
2054         unsigned int segno, secno, zoneno;
2055         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2056         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2057         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2058         unsigned int left_start = hint;
2059         bool init = true;
2060         int go_left = 0;
2061         int i;
2062
2063         spin_lock(&free_i->segmap_lock);
2064
2065         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2066                 segno = find_next_zero_bit(free_i->free_segmap,
2067                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2068                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2069                         goto got_it;
2070         }
2071 find_other_zone:
2072         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2073         if (secno >= MAIN_SECS(sbi)) {
2074                 if (dir == ALLOC_RIGHT) {
2075                         secno = find_next_zero_bit(free_i->free_secmap,
2076                                                         MAIN_SECS(sbi), 0);
2077                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2078                 } else {
2079                         go_left = 1;
2080                         left_start = hint - 1;
2081                 }
2082         }
2083         if (go_left == 0)
2084                 goto skip_left;
2085
2086         while (test_bit(left_start, free_i->free_secmap)) {
2087                 if (left_start > 0) {
2088                         left_start--;
2089                         continue;
2090                 }
2091                 left_start = find_next_zero_bit(free_i->free_secmap,
2092                                                         MAIN_SECS(sbi), 0);
2093                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2094                 break;
2095         }
2096         secno = left_start;
2097 skip_left:
2098         segno = GET_SEG_FROM_SEC(sbi, secno);
2099         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2100
2101         /* give up on finding another zone */
2102         if (!init)
2103                 goto got_it;
2104         if (sbi->secs_per_zone == 1)
2105                 goto got_it;
2106         if (zoneno == old_zoneno)
2107                 goto got_it;
2108         if (dir == ALLOC_LEFT) {
2109                 if (!go_left && zoneno + 1 >= total_zones)
2110                         goto got_it;
2111                 if (go_left && zoneno == 0)
2112                         goto got_it;
2113         }
2114         for (i = 0; i < NR_CURSEG_TYPE; i++)
2115                 if (CURSEG_I(sbi, i)->zone == zoneno)
2116                         break;
2117
2118         if (i < NR_CURSEG_TYPE) {
2119                 /* zone is in user, try another */
2120                 if (go_left)
2121                         hint = zoneno * sbi->secs_per_zone - 1;
2122                 else if (zoneno + 1 >= total_zones)
2123                         hint = 0;
2124                 else
2125                         hint = (zoneno + 1) * sbi->secs_per_zone;
2126                 init = false;
2127                 goto find_other_zone;
2128         }
2129 got_it:
2130         /* set it as dirty segment in free segmap */
2131         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2132         __set_inuse(sbi, segno);
2133         *newseg = segno;
2134         spin_unlock(&free_i->segmap_lock);
2135 }
2136
2137 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2138 {
2139         struct curseg_info *curseg = CURSEG_I(sbi, type);
2140         struct summary_footer *sum_footer;
2141
2142         curseg->segno = curseg->next_segno;
2143         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2144         curseg->next_blkoff = 0;
2145         curseg->next_segno = NULL_SEGNO;
2146
2147         sum_footer = &(curseg->sum_blk->footer);
2148         memset(sum_footer, 0, sizeof(struct summary_footer));
2149         if (IS_DATASEG(type))
2150                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2151         if (IS_NODESEG(type))
2152                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2153         __set_sit_entry_type(sbi, type, curseg->segno, modified);
2154 }
2155
2156 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2157 {
2158         /* if segs_per_sec is large than 1, we need to keep original policy. */
2159         if (sbi->segs_per_sec != 1)
2160                 return CURSEG_I(sbi, type)->segno;
2161
2162         if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
2163                 return 0;
2164
2165         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2166                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2167         return CURSEG_I(sbi, type)->segno;
2168 }
2169
2170 /*
2171  * Allocate a current working segment.
2172  * This function always allocates a free segment in LFS manner.
2173  */
2174 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2175 {
2176         struct curseg_info *curseg = CURSEG_I(sbi, type);
2177         unsigned int segno = curseg->segno;
2178         int dir = ALLOC_LEFT;
2179
2180         write_sum_page(sbi, curseg->sum_blk,
2181                                 GET_SUM_BLOCK(sbi, segno));
2182         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2183                 dir = ALLOC_RIGHT;
2184
2185         if (test_opt(sbi, NOHEAP))
2186                 dir = ALLOC_RIGHT;
2187
2188         segno = __get_next_segno(sbi, type);
2189         get_new_segment(sbi, &segno, new_sec, dir);
2190         curseg->next_segno = segno;
2191         reset_curseg(sbi, type, 1);
2192         curseg->alloc_type = LFS;
2193 }
2194
2195 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2196                         struct curseg_info *seg, block_t start)
2197 {
2198         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2199         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2200         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2201         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2202         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2203         int i, pos;
2204
2205         for (i = 0; i < entries; i++)
2206                 target_map[i] = ckpt_map[i] | cur_map[i];
2207
2208         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2209
2210         seg->next_blkoff = pos;
2211 }
2212
2213 /*
2214  * If a segment is written by LFS manner, next block offset is just obtained
2215  * by increasing the current block offset. However, if a segment is written by
2216  * SSR manner, next block offset obtained by calling __next_free_blkoff
2217  */
2218 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2219                                 struct curseg_info *seg)
2220 {
2221         if (seg->alloc_type == SSR)
2222                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2223         else
2224                 seg->next_blkoff++;
2225 }
2226
2227 /*
2228  * This function always allocates a used segment(from dirty seglist) by SSR
2229  * manner, so it should recover the existing segment information of valid blocks
2230  */
2231 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2232 {
2233         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2234         struct curseg_info *curseg = CURSEG_I(sbi, type);
2235         unsigned int new_segno = curseg->next_segno;
2236         struct f2fs_summary_block *sum_node;
2237         struct page *sum_page;
2238
2239         write_sum_page(sbi, curseg->sum_blk,
2240                                 GET_SUM_BLOCK(sbi, curseg->segno));
2241         __set_test_and_inuse(sbi, new_segno);
2242
2243         mutex_lock(&dirty_i->seglist_lock);
2244         __remove_dirty_segment(sbi, new_segno, PRE);
2245         __remove_dirty_segment(sbi, new_segno, DIRTY);
2246         mutex_unlock(&dirty_i->seglist_lock);
2247
2248         reset_curseg(sbi, type, 1);
2249         curseg->alloc_type = SSR;
2250         __next_free_blkoff(sbi, curseg, 0);
2251
2252         sum_page = get_sum_page(sbi, new_segno);
2253         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2254         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2255         f2fs_put_page(sum_page, 1);
2256 }
2257
2258 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2259 {
2260         struct curseg_info *curseg = CURSEG_I(sbi, type);
2261         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2262         unsigned segno = NULL_SEGNO;
2263         int i, cnt;
2264         bool reversed = false;
2265
2266         /* need_SSR() already forces to do this */
2267         if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2268                 curseg->next_segno = segno;
2269                 return 1;
2270         }
2271
2272         /* For node segments, let's do SSR more intensively */
2273         if (IS_NODESEG(type)) {
2274                 if (type >= CURSEG_WARM_NODE) {
2275                         reversed = true;
2276                         i = CURSEG_COLD_NODE;
2277                 } else {
2278                         i = CURSEG_HOT_NODE;
2279                 }
2280                 cnt = NR_CURSEG_NODE_TYPE;
2281         } else {
2282                 if (type >= CURSEG_WARM_DATA) {
2283                         reversed = true;
2284                         i = CURSEG_COLD_DATA;
2285                 } else {
2286                         i = CURSEG_HOT_DATA;
2287                 }
2288                 cnt = NR_CURSEG_DATA_TYPE;
2289         }
2290
2291         for (; cnt-- > 0; reversed ? i-- : i++) {
2292                 if (i == type)
2293                         continue;
2294                 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2295                         curseg->next_segno = segno;
2296                         return 1;
2297                 }
2298         }
2299         return 0;
2300 }
2301
2302 /*
2303  * flush out current segment and replace it with new segment
2304  * This function should be returned with success, otherwise BUG
2305  */
2306 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2307                                                 int type, bool force)
2308 {
2309         struct curseg_info *curseg = CURSEG_I(sbi, type);
2310
2311         if (force)
2312                 new_curseg(sbi, type, true);
2313         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2314                                         type == CURSEG_WARM_NODE)
2315                 new_curseg(sbi, type, false);
2316         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2317                 new_curseg(sbi, type, false);
2318         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
2319                 change_curseg(sbi, type);
2320         else
2321                 new_curseg(sbi, type, false);
2322
2323         stat_inc_seg_type(sbi, curseg);
2324 }
2325
2326 void allocate_new_segments(struct f2fs_sb_info *sbi)
2327 {
2328         struct curseg_info *curseg;
2329         unsigned int old_segno;
2330         int i;
2331
2332         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2333                 curseg = CURSEG_I(sbi, i);
2334                 old_segno = curseg->segno;
2335                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2336                 locate_dirty_segment(sbi, old_segno);
2337         }
2338 }
2339
2340 static const struct segment_allocation default_salloc_ops = {
2341         .allocate_segment = allocate_segment_by_default,
2342 };
2343
2344 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2345 {
2346         __u64 trim_start = cpc->trim_start;
2347         bool has_candidate = false;
2348
2349         mutex_lock(&SIT_I(sbi)->sentry_lock);
2350         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2351                 if (add_discard_addrs(sbi, cpc, true)) {
2352                         has_candidate = true;
2353                         break;
2354                 }
2355         }
2356         mutex_unlock(&SIT_I(sbi)->sentry_lock);
2357
2358         cpc->trim_start = trim_start;
2359         return has_candidate;
2360 }
2361
2362 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2363 {
2364         __u64 start = F2FS_BYTES_TO_BLK(range->start);
2365         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2366         unsigned int start_segno, end_segno, cur_segno;
2367         block_t start_block, end_block;
2368         struct cp_control cpc;
2369         struct discard_policy dpolicy;
2370         int err = 0;
2371
2372         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2373                 return -EINVAL;
2374
2375         cpc.trimmed = 0;
2376         if (end <= MAIN_BLKADDR(sbi))
2377                 goto out;
2378
2379         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2380                 f2fs_msg(sbi->sb, KERN_WARNING,
2381                         "Found FS corruption, run fsck to fix.");
2382                 goto out;
2383         }
2384
2385         /* start/end segment number in main_area */
2386         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2387         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2388                                                 GET_SEGNO(sbi, end);
2389
2390         cpc.reason = CP_DISCARD;
2391         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2392
2393         /* do checkpoint to issue discard commands safely */
2394         for (cur_segno = start_segno; cur_segno <= end_segno;
2395                                         cur_segno = cpc.trim_end + 1) {
2396                 cpc.trim_start = cur_segno;
2397
2398                 if (sbi->discard_blks == 0)
2399                         break;
2400                 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
2401                         cpc.trim_end = end_segno;
2402                 else
2403                         cpc.trim_end = min_t(unsigned int,
2404                                 rounddown(cur_segno +
2405                                 BATCHED_TRIM_SEGMENTS(sbi),
2406                                 sbi->segs_per_sec) - 1, end_segno);
2407
2408                 mutex_lock(&sbi->gc_mutex);
2409                 err = write_checkpoint(sbi, &cpc);
2410                 mutex_unlock(&sbi->gc_mutex);
2411                 if (err)
2412                         break;
2413
2414                 schedule();
2415         }
2416
2417         start_block = START_BLOCK(sbi, start_segno);
2418         end_block = START_BLOCK(sbi, min(cur_segno, end_segno) + 1);
2419
2420         init_discard_policy(&dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2421         __issue_discard_cmd_range(sbi, &dpolicy, start_block, end_block);
2422         __wait_discard_cmd_range(sbi, &dpolicy, start_block, end_block);
2423 out:
2424         range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
2425         return err;
2426 }
2427
2428 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2429 {
2430         struct curseg_info *curseg = CURSEG_I(sbi, type);
2431         if (curseg->next_blkoff < sbi->blocks_per_seg)
2432                 return true;
2433         return false;
2434 }
2435
2436 static int __get_segment_type_2(struct f2fs_io_info *fio)
2437 {
2438         if (fio->type == DATA)
2439                 return CURSEG_HOT_DATA;
2440         else
2441                 return CURSEG_HOT_NODE;
2442 }
2443
2444 static int __get_segment_type_4(struct f2fs_io_info *fio)
2445 {
2446         if (fio->type == DATA) {
2447                 struct inode *inode = fio->page->mapping->host;
2448
2449                 if (S_ISDIR(inode->i_mode))
2450                         return CURSEG_HOT_DATA;
2451                 else
2452                         return CURSEG_COLD_DATA;
2453         } else {
2454                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2455                         return CURSEG_WARM_NODE;
2456                 else
2457                         return CURSEG_COLD_NODE;
2458         }
2459 }
2460
2461 static int __get_segment_type_6(struct f2fs_io_info *fio)
2462 {
2463         if (fio->type == DATA) {
2464                 struct inode *inode = fio->page->mapping->host;
2465
2466                 if (is_cold_data(fio->page) || file_is_cold(inode))
2467                         return CURSEG_COLD_DATA;
2468                 if (is_inode_flag_set(inode, FI_HOT_DATA))
2469                         return CURSEG_HOT_DATA;
2470                 return CURSEG_WARM_DATA;
2471         } else {
2472                 if (IS_DNODE(fio->page))
2473                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2474                                                 CURSEG_HOT_NODE;
2475                 return CURSEG_COLD_NODE;
2476         }
2477 }
2478
2479 static int __get_segment_type(struct f2fs_io_info *fio)
2480 {
2481         int type = 0;
2482
2483         switch (fio->sbi->active_logs) {
2484         case 2:
2485                 type = __get_segment_type_2(fio);
2486                 break;
2487         case 4:
2488                 type = __get_segment_type_4(fio);
2489                 break;
2490         case 6:
2491                 type = __get_segment_type_6(fio);
2492                 break;
2493         default:
2494                 f2fs_bug_on(fio->sbi, true);
2495         }
2496
2497         if (IS_HOT(type))
2498                 fio->temp = HOT;
2499         else if (IS_WARM(type))
2500                 fio->temp = WARM;
2501         else
2502                 fio->temp = COLD;
2503         return type;
2504 }
2505
2506 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2507                 block_t old_blkaddr, block_t *new_blkaddr,
2508                 struct f2fs_summary *sum, int type,
2509                 struct f2fs_io_info *fio, bool add_list)
2510 {
2511         struct sit_info *sit_i = SIT_I(sbi);
2512         struct curseg_info *curseg = CURSEG_I(sbi, type);
2513
2514         mutex_lock(&curseg->curseg_mutex);
2515         mutex_lock(&sit_i->sentry_lock);
2516
2517         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2518
2519         f2fs_wait_discard_bio(sbi, *new_blkaddr);
2520
2521         /*
2522          * __add_sum_entry should be resided under the curseg_mutex
2523          * because, this function updates a summary entry in the
2524          * current summary block.
2525          */
2526         __add_sum_entry(sbi, type, sum);
2527
2528         __refresh_next_blkoff(sbi, curseg);
2529
2530         stat_inc_block_count(sbi, curseg);
2531
2532         if (!__has_curseg_space(sbi, type))
2533                 sit_i->s_ops->allocate_segment(sbi, type, false);
2534         /*
2535          * SIT information should be updated after segment allocation,
2536          * since we need to keep dirty segments precisely under SSR.
2537          */
2538         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
2539
2540         mutex_unlock(&sit_i->sentry_lock);
2541
2542         if (page && IS_NODESEG(type)) {
2543                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2544
2545                 f2fs_inode_chksum_set(sbi, page);
2546         }
2547
2548         if (add_list) {
2549                 struct f2fs_bio_info *io;
2550
2551                 INIT_LIST_HEAD(&fio->list);
2552                 fio->in_list = true;
2553                 io = sbi->write_io[fio->type] + fio->temp;
2554                 spin_lock(&io->io_lock);
2555                 list_add_tail(&fio->list, &io->io_list);
2556                 spin_unlock(&io->io_lock);
2557         }
2558
2559         mutex_unlock(&curseg->curseg_mutex);
2560 }
2561
2562 static void update_device_state(struct f2fs_io_info *fio)
2563 {
2564         struct f2fs_sb_info *sbi = fio->sbi;
2565         unsigned int devidx;
2566
2567         if (!sbi->s_ndevs)
2568                 return;
2569
2570         devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2571
2572         /* update device state for fsync */
2573         set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
2574
2575         /* update device state for checkpoint */
2576         if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2577                 spin_lock(&sbi->dev_lock);
2578                 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2579                 spin_unlock(&sbi->dev_lock);
2580         }
2581 }
2582
2583 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2584 {
2585         int type = __get_segment_type(fio);
2586         int err;
2587
2588 reallocate:
2589         allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
2590                         &fio->new_blkaddr, sum, type, fio, true);
2591
2592         /* writeout dirty page into bdev */
2593         err = f2fs_submit_page_write(fio);
2594         if (err == -EAGAIN) {
2595                 fio->old_blkaddr = fio->new_blkaddr;
2596                 goto reallocate;
2597         } else if (!err) {
2598                 update_device_state(fio);
2599         }
2600 }
2601
2602 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2603                                         enum iostat_type io_type)
2604 {
2605         struct f2fs_io_info fio = {
2606                 .sbi = sbi,
2607                 .type = META,
2608                 .op = REQ_OP_WRITE,
2609                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
2610                 .old_blkaddr = page->index,
2611                 .new_blkaddr = page->index,
2612                 .page = page,
2613                 .encrypted_page = NULL,
2614                 .in_list = false,
2615         };
2616
2617         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
2618                 fio.op_flags &= ~REQ_META;
2619
2620         set_page_writeback(page);
2621         f2fs_submit_page_write(&fio);
2622
2623         f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
2624 }
2625
2626 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
2627 {
2628         struct f2fs_summary sum;
2629
2630         set_summary(&sum, nid, 0, 0);
2631         do_write_page(&sum, fio);
2632
2633         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2634 }
2635
2636 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
2637 {
2638         struct f2fs_sb_info *sbi = fio->sbi;
2639         struct f2fs_summary sum;
2640         struct node_info ni;
2641
2642         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
2643         get_node_info(sbi, dn->nid, &ni);
2644         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
2645         do_write_page(&sum, fio);
2646         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
2647
2648         f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
2649 }
2650
2651 int rewrite_data_page(struct f2fs_io_info *fio)
2652 {
2653         int err;
2654
2655         fio->new_blkaddr = fio->old_blkaddr;
2656         stat_inc_inplace_blocks(fio->sbi);
2657
2658         err = f2fs_submit_page_bio(fio);
2659         if (!err)
2660                 update_device_state(fio);
2661
2662         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2663
2664         return err;
2665 }
2666
2667 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
2668                                 block_t old_blkaddr, block_t new_blkaddr,
2669                                 bool recover_curseg, bool recover_newaddr)
2670 {
2671         struct sit_info *sit_i = SIT_I(sbi);
2672         struct curseg_info *curseg;
2673         unsigned int segno, old_cursegno;
2674         struct seg_entry *se;
2675         int type;
2676         unsigned short old_blkoff;
2677
2678         segno = GET_SEGNO(sbi, new_blkaddr);
2679         se = get_seg_entry(sbi, segno);
2680         type = se->type;
2681
2682         if (!recover_curseg) {
2683                 /* for recovery flow */
2684                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2685                         if (old_blkaddr == NULL_ADDR)
2686                                 type = CURSEG_COLD_DATA;
2687                         else
2688                                 type = CURSEG_WARM_DATA;
2689                 }
2690         } else {
2691                 if (!IS_CURSEG(sbi, segno))
2692                         type = CURSEG_WARM_DATA;
2693         }
2694
2695         curseg = CURSEG_I(sbi, type);
2696
2697         mutex_lock(&curseg->curseg_mutex);
2698         mutex_lock(&sit_i->sentry_lock);
2699
2700         old_cursegno = curseg->segno;
2701         old_blkoff = curseg->next_blkoff;
2702
2703         /* change the current segment */
2704         if (segno != curseg->segno) {
2705                 curseg->next_segno = segno;
2706                 change_curseg(sbi, type);
2707         }
2708
2709         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
2710         __add_sum_entry(sbi, type, sum);
2711
2712         if (!recover_curseg || recover_newaddr)
2713                 update_sit_entry(sbi, new_blkaddr, 1);
2714         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2715                 update_sit_entry(sbi, old_blkaddr, -1);
2716
2717         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2718         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2719
2720         locate_dirty_segment(sbi, old_cursegno);
2721
2722         if (recover_curseg) {
2723                 if (old_cursegno != curseg->segno) {
2724                         curseg->next_segno = old_cursegno;
2725                         change_curseg(sbi, type);
2726                 }
2727                 curseg->next_blkoff = old_blkoff;
2728         }
2729
2730         mutex_unlock(&sit_i->sentry_lock);
2731         mutex_unlock(&curseg->curseg_mutex);
2732 }
2733
2734 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2735                                 block_t old_addr, block_t new_addr,
2736                                 unsigned char version, bool recover_curseg,
2737                                 bool recover_newaddr)
2738 {
2739         struct f2fs_summary sum;
2740
2741         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2742
2743         __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2744                                         recover_curseg, recover_newaddr);
2745
2746         f2fs_update_data_blkaddr(dn, new_addr);
2747 }
2748
2749 void f2fs_wait_on_page_writeback(struct page *page,
2750                                 enum page_type type, bool ordered)
2751 {
2752         if (PageWriteback(page)) {
2753                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2754
2755                 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2756                                                 0, page->index, type);
2757                 if (ordered)
2758                         wait_on_page_writeback(page);
2759                 else
2760                         wait_for_stable_page(page);
2761         }
2762 }
2763
2764 void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
2765 {
2766         struct page *cpage;
2767
2768         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
2769                 return;
2770
2771         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2772         if (cpage) {
2773                 f2fs_wait_on_page_writeback(cpage, DATA, true);
2774                 f2fs_put_page(cpage, 1);
2775         }
2776 }
2777
2778 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
2779 {
2780         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2781         struct curseg_info *seg_i;
2782         unsigned char *kaddr;
2783         struct page *page;
2784         block_t start;
2785         int i, j, offset;
2786
2787         start = start_sum_block(sbi);
2788
2789         page = get_meta_page(sbi, start++);
2790         kaddr = (unsigned char *)page_address(page);
2791
2792         /* Step 1: restore nat cache */
2793         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2794         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
2795
2796         /* Step 2: restore sit cache */
2797         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2798         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
2799         offset = 2 * SUM_JOURNAL_SIZE;
2800
2801         /* Step 3: restore summary entries */
2802         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2803                 unsigned short blk_off;
2804                 unsigned int segno;
2805
2806                 seg_i = CURSEG_I(sbi, i);
2807                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2808                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2809                 seg_i->next_segno = segno;
2810                 reset_curseg(sbi, i, 0);
2811                 seg_i->alloc_type = ckpt->alloc_type[i];
2812                 seg_i->next_blkoff = blk_off;
2813
2814                 if (seg_i->alloc_type == SSR)
2815                         blk_off = sbi->blocks_per_seg;
2816
2817                 for (j = 0; j < blk_off; j++) {
2818                         struct f2fs_summary *s;
2819                         s = (struct f2fs_summary *)(kaddr + offset);
2820                         seg_i->sum_blk->entries[j] = *s;
2821                         offset += SUMMARY_SIZE;
2822                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
2823                                                 SUM_FOOTER_SIZE)
2824                                 continue;
2825
2826                         f2fs_put_page(page, 1);
2827                         page = NULL;
2828
2829                         page = get_meta_page(sbi, start++);
2830                         kaddr = (unsigned char *)page_address(page);
2831                         offset = 0;
2832                 }
2833         }
2834         f2fs_put_page(page, 1);
2835         return 0;
2836 }
2837
2838 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2839 {
2840         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2841         struct f2fs_summary_block *sum;
2842         struct curseg_info *curseg;
2843         struct page *new;
2844         unsigned short blk_off;
2845         unsigned int segno = 0;
2846         block_t blk_addr = 0;
2847
2848         /* get segment number and block addr */
2849         if (IS_DATASEG(type)) {
2850                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2851                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2852                                                         CURSEG_HOT_DATA]);
2853                 if (__exist_node_summaries(sbi))
2854                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2855                 else
2856                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2857         } else {
2858                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2859                                                         CURSEG_HOT_NODE]);
2860                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2861                                                         CURSEG_HOT_NODE]);
2862                 if (__exist_node_summaries(sbi))
2863                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2864                                                         type - CURSEG_HOT_NODE);
2865                 else
2866                         blk_addr = GET_SUM_BLOCK(sbi, segno);
2867         }
2868
2869         new = get_meta_page(sbi, blk_addr);
2870         sum = (struct f2fs_summary_block *)page_address(new);
2871
2872         if (IS_NODESEG(type)) {
2873                 if (__exist_node_summaries(sbi)) {
2874                         struct f2fs_summary *ns = &sum->entries[0];
2875                         int i;
2876                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2877                                 ns->version = 0;
2878                                 ns->ofs_in_node = 0;
2879                         }
2880                 } else {
2881                         int err;
2882
2883                         err = restore_node_summary(sbi, segno, sum);
2884                         if (err) {
2885                                 f2fs_put_page(new, 1);
2886                                 return err;
2887                         }
2888                 }
2889         }
2890
2891         /* set uncompleted segment to curseg */
2892         curseg = CURSEG_I(sbi, type);
2893         mutex_lock(&curseg->curseg_mutex);
2894
2895         /* update journal info */
2896         down_write(&curseg->journal_rwsem);
2897         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2898         up_write(&curseg->journal_rwsem);
2899
2900         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2901         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
2902         curseg->next_segno = segno;
2903         reset_curseg(sbi, type, 0);
2904         curseg->alloc_type = ckpt->alloc_type[type];
2905         curseg->next_blkoff = blk_off;
2906         mutex_unlock(&curseg->curseg_mutex);
2907         f2fs_put_page(new, 1);
2908         return 0;
2909 }
2910
2911 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2912 {
2913         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
2914         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
2915         int type = CURSEG_HOT_DATA;
2916         int err;
2917
2918         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
2919                 int npages = npages_for_summary_flush(sbi, true);
2920
2921                 if (npages >= 2)
2922                         ra_meta_pages(sbi, start_sum_block(sbi), npages,
2923                                                         META_CP, true);
2924
2925                 /* restore for compacted data summary */
2926                 if (read_compacted_summaries(sbi))
2927                         return -EINVAL;
2928                 type = CURSEG_HOT_NODE;
2929         }
2930
2931         if (__exist_node_summaries(sbi))
2932                 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
2933                                         NR_CURSEG_TYPE - type, META_CP, true);
2934
2935         for (; type <= CURSEG_COLD_NODE; type++) {
2936                 err = read_normal_summaries(sbi, type);
2937                 if (err)
2938                         return err;
2939         }
2940
2941         /* sanity check for summary blocks */
2942         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
2943                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
2944                 return -EINVAL;
2945
2946         return 0;
2947 }
2948
2949 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2950 {
2951         struct page *page;
2952         unsigned char *kaddr;
2953         struct f2fs_summary *summary;
2954         struct curseg_info *seg_i;
2955         int written_size = 0;
2956         int i, j;
2957
2958         page = grab_meta_page(sbi, blkaddr++);
2959         kaddr = (unsigned char *)page_address(page);
2960
2961         /* Step 1: write nat cache */
2962         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2963         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
2964         written_size += SUM_JOURNAL_SIZE;
2965
2966         /* Step 2: write sit cache */
2967         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2968         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
2969         written_size += SUM_JOURNAL_SIZE;
2970
2971         /* Step 3: write summary entries */
2972         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2973                 unsigned short blkoff;
2974                 seg_i = CURSEG_I(sbi, i);
2975                 if (sbi->ckpt->alloc_type[i] == SSR)
2976                         blkoff = sbi->blocks_per_seg;
2977                 else
2978                         blkoff = curseg_blkoff(sbi, i);
2979
2980                 for (j = 0; j < blkoff; j++) {
2981                         if (!page) {
2982                                 page = grab_meta_page(sbi, blkaddr++);
2983                                 kaddr = (unsigned char *)page_address(page);
2984                                 written_size = 0;
2985                         }
2986                         summary = (struct f2fs_summary *)(kaddr + written_size);
2987                         *summary = seg_i->sum_blk->entries[j];
2988                         written_size += SUMMARY_SIZE;
2989
2990                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
2991                                                         SUM_FOOTER_SIZE)
2992                                 continue;
2993
2994                         set_page_dirty(page);
2995                         f2fs_put_page(page, 1);
2996                         page = NULL;
2997                 }
2998         }
2999         if (page) {
3000                 set_page_dirty(page);
3001                 f2fs_put_page(page, 1);
3002         }
3003 }
3004
3005 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3006                                         block_t blkaddr, int type)
3007 {
3008         int i, end;
3009         if (IS_DATASEG(type))
3010                 end = type + NR_CURSEG_DATA_TYPE;
3011         else
3012                 end = type + NR_CURSEG_NODE_TYPE;
3013
3014         for (i = type; i < end; i++)
3015                 write_current_sum_page(sbi, i, blkaddr + (i - type));
3016 }
3017
3018 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3019 {
3020         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3021                 write_compacted_summaries(sbi, start_blk);
3022         else
3023                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3024 }
3025
3026 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3027 {
3028         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3029 }
3030
3031 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3032                                         unsigned int val, int alloc)
3033 {
3034         int i;
3035
3036         if (type == NAT_JOURNAL) {
3037                 for (i = 0; i < nats_in_cursum(journal); i++) {
3038                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3039                                 return i;
3040                 }
3041                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3042                         return update_nats_in_cursum(journal, 1);
3043         } else if (type == SIT_JOURNAL) {
3044                 for (i = 0; i < sits_in_cursum(journal); i++)
3045                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3046                                 return i;
3047                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3048                         return update_sits_in_cursum(journal, 1);
3049         }
3050         return -1;
3051 }
3052
3053 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3054                                         unsigned int segno)
3055 {
3056         return get_meta_page(sbi, current_sit_addr(sbi, segno));
3057 }
3058
3059 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3060                                         unsigned int start)
3061 {
3062         struct sit_info *sit_i = SIT_I(sbi);
3063         struct page *src_page, *dst_page;
3064         pgoff_t src_off, dst_off;
3065         void *src_addr, *dst_addr;
3066
3067         src_off = current_sit_addr(sbi, start);
3068         dst_off = next_sit_addr(sbi, src_off);
3069
3070         /* get current sit block page without lock */
3071         src_page = get_meta_page(sbi, src_off);
3072         dst_page = grab_meta_page(sbi, dst_off);
3073         f2fs_bug_on(sbi, PageDirty(src_page));
3074
3075         src_addr = page_address(src_page);
3076         dst_addr = page_address(dst_page);
3077         memcpy(dst_addr, src_addr, PAGE_SIZE);
3078
3079         set_page_dirty(dst_page);
3080         f2fs_put_page(src_page, 1);
3081
3082         set_to_next_sit(sit_i, start);
3083
3084         return dst_page;
3085 }
3086
3087 static struct sit_entry_set *grab_sit_entry_set(void)
3088 {
3089         struct sit_entry_set *ses =
3090                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3091
3092         ses->entry_cnt = 0;
3093         INIT_LIST_HEAD(&ses->set_list);
3094         return ses;
3095 }
3096
3097 static void release_sit_entry_set(struct sit_entry_set *ses)
3098 {
3099         list_del(&ses->set_list);
3100         kmem_cache_free(sit_entry_set_slab, ses);
3101 }
3102
3103 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3104                                                 struct list_head *head)
3105 {
3106         struct sit_entry_set *next = ses;
3107
3108         if (list_is_last(&ses->set_list, head))
3109                 return;
3110
3111         list_for_each_entry_continue(next, head, set_list)
3112                 if (ses->entry_cnt <= next->entry_cnt)
3113                         break;
3114
3115         list_move_tail(&ses->set_list, &next->set_list);
3116 }
3117
3118 static void add_sit_entry(unsigned int segno, struct list_head *head)
3119 {
3120         struct sit_entry_set *ses;
3121         unsigned int start_segno = START_SEGNO(segno);
3122
3123         list_for_each_entry(ses, head, set_list) {
3124                 if (ses->start_segno == start_segno) {
3125                         ses->entry_cnt++;
3126                         adjust_sit_entry_set(ses, head);
3127                         return;
3128                 }
3129         }
3130
3131         ses = grab_sit_entry_set();
3132
3133         ses->start_segno = start_segno;
3134         ses->entry_cnt++;
3135         list_add(&ses->set_list, head);
3136 }
3137
3138 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3139 {
3140         struct f2fs_sm_info *sm_info = SM_I(sbi);
3141         struct list_head *set_list = &sm_info->sit_entry_set;
3142         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3143         unsigned int segno;
3144
3145         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3146                 add_sit_entry(segno, set_list);
3147 }
3148
3149 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3150 {
3151         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3152         struct f2fs_journal *journal = curseg->journal;
3153         int i;
3154
3155         down_write(&curseg->journal_rwsem);
3156         for (i = 0; i < sits_in_cursum(journal); i++) {
3157                 unsigned int segno;
3158                 bool dirtied;
3159
3160                 segno = le32_to_cpu(segno_in_journal(journal, i));
3161                 dirtied = __mark_sit_entry_dirty(sbi, segno);
3162
3163                 if (!dirtied)
3164                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3165         }
3166         update_sits_in_cursum(journal, -i);
3167         up_write(&curseg->journal_rwsem);
3168 }
3169
3170 /*
3171  * CP calls this function, which flushes SIT entries including sit_journal,
3172  * and moves prefree segs to free segs.
3173  */
3174 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3175 {
3176         struct sit_info *sit_i = SIT_I(sbi);
3177         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3178         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3179         struct f2fs_journal *journal = curseg->journal;
3180         struct sit_entry_set *ses, *tmp;
3181         struct list_head *head = &SM_I(sbi)->sit_entry_set;
3182         bool to_journal = true;
3183         struct seg_entry *se;
3184
3185         mutex_lock(&sit_i->sentry_lock);
3186
3187         if (!sit_i->dirty_sentries)
3188                 goto out;
3189
3190         /*
3191          * add and account sit entries of dirty bitmap in sit entry
3192          * set temporarily
3193          */
3194         add_sits_in_set(sbi);
3195
3196         /*
3197          * if there are no enough space in journal to store dirty sit
3198          * entries, remove all entries from journal and add and account
3199          * them in sit entry set.
3200          */
3201         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3202                 remove_sits_in_journal(sbi);
3203
3204         /*
3205          * there are two steps to flush sit entries:
3206          * #1, flush sit entries to journal in current cold data summary block.
3207          * #2, flush sit entries to sit page.
3208          */
3209         list_for_each_entry_safe(ses, tmp, head, set_list) {
3210                 struct page *page = NULL;
3211                 struct f2fs_sit_block *raw_sit = NULL;
3212                 unsigned int start_segno = ses->start_segno;
3213                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3214                                                 (unsigned long)MAIN_SEGS(sbi));
3215                 unsigned int segno = start_segno;
3216
3217                 if (to_journal &&
3218                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3219                         to_journal = false;
3220
3221                 if (to_journal) {
3222                         down_write(&curseg->journal_rwsem);
3223                 } else {
3224                         page = get_next_sit_page(sbi, start_segno);
3225                         raw_sit = page_address(page);
3226                 }
3227
3228                 /* flush dirty sit entries in region of current sit set */
3229                 for_each_set_bit_from(segno, bitmap, end) {
3230                         int offset, sit_offset;
3231
3232                         se = get_seg_entry(sbi, segno);
3233
3234                         /* add discard candidates */
3235                         if (!(cpc->reason & CP_DISCARD)) {
3236                                 cpc->trim_start = segno;
3237                                 add_discard_addrs(sbi, cpc, false);
3238                         }
3239
3240                         if (to_journal) {
3241                                 offset = lookup_journal_in_cursum(journal,
3242                                                         SIT_JOURNAL, segno, 1);
3243                                 f2fs_bug_on(sbi, offset < 0);
3244                                 segno_in_journal(journal, offset) =
3245                                                         cpu_to_le32(segno);
3246                                 seg_info_to_raw_sit(se,
3247                                         &sit_in_journal(journal, offset));
3248                         } else {
3249                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3250                                 seg_info_to_raw_sit(se,
3251                                                 &raw_sit->entries[sit_offset]);
3252                         }
3253
3254                         __clear_bit(segno, bitmap);
3255                         sit_i->dirty_sentries--;
3256                         ses->entry_cnt--;
3257                 }
3258
3259                 if (to_journal)
3260                         up_write(&curseg->journal_rwsem);
3261                 else
3262                         f2fs_put_page(page, 1);
3263
3264                 f2fs_bug_on(sbi, ses->entry_cnt);
3265                 release_sit_entry_set(ses);
3266         }
3267
3268         f2fs_bug_on(sbi, !list_empty(head));
3269         f2fs_bug_on(sbi, sit_i->dirty_sentries);
3270 out:
3271         if (cpc->reason & CP_DISCARD) {
3272                 __u64 trim_start = cpc->trim_start;
3273
3274                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3275                         add_discard_addrs(sbi, cpc, false);
3276
3277                 cpc->trim_start = trim_start;
3278         }
3279         mutex_unlock(&sit_i->sentry_lock);
3280
3281         set_prefree_as_free_segments(sbi);
3282 }
3283
3284 static int build_sit_info(struct f2fs_sb_info *sbi)
3285 {
3286         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3287         struct sit_info *sit_i;
3288         unsigned int sit_segs, start;
3289         char *src_bitmap;
3290         unsigned int bitmap_size;
3291
3292         /* allocate memory for SIT information */
3293         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
3294         if (!sit_i)
3295                 return -ENOMEM;
3296
3297         SM_I(sbi)->sit_info = sit_i;
3298
3299         sit_i->sentries = kvzalloc(MAIN_SEGS(sbi) *
3300                                         sizeof(struct seg_entry), GFP_KERNEL);
3301         if (!sit_i->sentries)
3302                 return -ENOMEM;
3303
3304         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3305         sit_i->dirty_sentries_bitmap = kvzalloc(bitmap_size, GFP_KERNEL);
3306         if (!sit_i->dirty_sentries_bitmap)
3307                 return -ENOMEM;
3308
3309         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3310                 sit_i->sentries[start].cur_valid_map
3311                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3312                 sit_i->sentries[start].ckpt_valid_map
3313                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3314                 if (!sit_i->sentries[start].cur_valid_map ||
3315                                 !sit_i->sentries[start].ckpt_valid_map)
3316                         return -ENOMEM;
3317
3318 #ifdef CONFIG_F2FS_CHECK_FS
3319                 sit_i->sentries[start].cur_valid_map_mir
3320                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3321                 if (!sit_i->sentries[start].cur_valid_map_mir)
3322                         return -ENOMEM;
3323 #endif
3324
3325                 if (f2fs_discard_en(sbi)) {
3326                         sit_i->sentries[start].discard_map
3327                                 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3328                         if (!sit_i->sentries[start].discard_map)
3329                                 return -ENOMEM;
3330                 }
3331         }
3332
3333         sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3334         if (!sit_i->tmp_map)
3335                 return -ENOMEM;
3336
3337         if (sbi->segs_per_sec > 1) {
3338                 sit_i->sec_entries = kvzalloc(MAIN_SECS(sbi) *
3339                                         sizeof(struct sec_entry), GFP_KERNEL);
3340                 if (!sit_i->sec_entries)
3341                         return -ENOMEM;
3342         }
3343
3344         /* get information related with SIT */
3345         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3346
3347         /* setup SIT bitmap from ckeckpoint pack */
3348         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3349         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3350
3351         sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3352         if (!sit_i->sit_bitmap)
3353                 return -ENOMEM;
3354
3355 #ifdef CONFIG_F2FS_CHECK_FS
3356         sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3357         if (!sit_i->sit_bitmap_mir)
3358                 return -ENOMEM;
3359 #endif
3360
3361         /* init SIT information */
3362         sit_i->s_ops = &default_salloc_ops;
3363
3364         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3365         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
3366         sit_i->written_valid_blocks = 0;
3367         sit_i->bitmap_size = bitmap_size;
3368         sit_i->dirty_sentries = 0;
3369         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3370         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3371         sit_i->mounted_time = ktime_get_real_seconds();
3372         mutex_init(&sit_i->sentry_lock);
3373         return 0;
3374 }
3375
3376 static int build_free_segmap(struct f2fs_sb_info *sbi)
3377 {
3378         struct free_segmap_info *free_i;
3379         unsigned int bitmap_size, sec_bitmap_size;
3380
3381         /* allocate memory for free segmap information */
3382         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
3383         if (!free_i)
3384                 return -ENOMEM;
3385
3386         SM_I(sbi)->free_info = free_i;
3387
3388         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3389         free_i->free_segmap = kvmalloc(bitmap_size, GFP_KERNEL);
3390         if (!free_i->free_segmap)
3391                 return -ENOMEM;
3392
3393         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3394         free_i->free_secmap = kvmalloc(sec_bitmap_size, GFP_KERNEL);
3395         if (!free_i->free_secmap)
3396                 return -ENOMEM;
3397
3398         /* set all segments as dirty temporarily */
3399         memset(free_i->free_segmap, 0xff, bitmap_size);
3400         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3401
3402         /* init free segmap information */
3403         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
3404         free_i->free_segments = 0;
3405         free_i->free_sections = 0;
3406         spin_lock_init(&free_i->segmap_lock);
3407         return 0;
3408 }
3409
3410 static int build_curseg(struct f2fs_sb_info *sbi)
3411 {
3412         struct curseg_info *array;
3413         int i;
3414
3415         array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
3416         if (!array)
3417                 return -ENOMEM;
3418
3419         SM_I(sbi)->curseg_array = array;
3420
3421         for (i = 0; i < NR_CURSEG_TYPE; i++) {
3422                 mutex_init(&array[i].curseg_mutex);
3423                 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
3424                 if (!array[i].sum_blk)
3425                         return -ENOMEM;
3426                 init_rwsem(&array[i].journal_rwsem);
3427                 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
3428                                                         GFP_KERNEL);
3429                 if (!array[i].journal)
3430                         return -ENOMEM;
3431                 array[i].segno = NULL_SEGNO;
3432                 array[i].next_blkoff = 0;
3433         }
3434         return restore_curseg_summaries(sbi);
3435 }
3436
3437 static void build_sit_entries(struct f2fs_sb_info *sbi)
3438 {
3439         struct sit_info *sit_i = SIT_I(sbi);
3440         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3441         struct f2fs_journal *journal = curseg->journal;
3442         struct seg_entry *se;
3443         struct f2fs_sit_entry sit;
3444         int sit_blk_cnt = SIT_BLK_CNT(sbi);
3445         unsigned int i, start, end;
3446         unsigned int readed, start_blk = 0;
3447
3448         do {
3449                 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3450                                                         META_SIT, true);
3451
3452                 start = start_blk * sit_i->sents_per_block;
3453                 end = (start_blk + readed) * sit_i->sents_per_block;
3454
3455                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
3456                         struct f2fs_sit_block *sit_blk;
3457                         struct page *page;
3458
3459                         se = &sit_i->sentries[start];
3460                         page = get_current_sit_page(sbi, start);
3461                         sit_blk = (struct f2fs_sit_block *)page_address(page);
3462                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3463                         f2fs_put_page(page, 1);
3464
3465                         check_block_count(sbi, start, &sit);
3466                         seg_info_from_raw_sit(se, &sit);
3467
3468                         /* build discard map only one time */
3469                         if (f2fs_discard_en(sbi)) {
3470                                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3471                                         memset(se->discard_map, 0xff,
3472                                                 SIT_VBLOCK_MAP_SIZE);
3473                                 } else {
3474                                         memcpy(se->discard_map,
3475                                                 se->cur_valid_map,
3476                                                 SIT_VBLOCK_MAP_SIZE);
3477                                         sbi->discard_blks +=
3478                                                 sbi->blocks_per_seg -
3479                                                 se->valid_blocks;
3480                                 }
3481                         }
3482
3483                         if (sbi->segs_per_sec > 1)
3484                                 get_sec_entry(sbi, start)->valid_blocks +=
3485                                                         se->valid_blocks;
3486                 }
3487                 start_blk += readed;
3488         } while (start_blk < sit_blk_cnt);
3489
3490         down_read(&curseg->journal_rwsem);
3491         for (i = 0; i < sits_in_cursum(journal); i++) {
3492                 unsigned int old_valid_blocks;
3493
3494                 start = le32_to_cpu(segno_in_journal(journal, i));
3495                 se = &sit_i->sentries[start];
3496                 sit = sit_in_journal(journal, i);
3497
3498                 old_valid_blocks = se->valid_blocks;
3499
3500                 check_block_count(sbi, start, &sit);
3501                 seg_info_from_raw_sit(se, &sit);
3502
3503                 if (f2fs_discard_en(sbi)) {
3504                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3505                                 memset(se->discard_map, 0xff,
3506                                                         SIT_VBLOCK_MAP_SIZE);
3507                         } else {
3508                                 memcpy(se->discard_map, se->cur_valid_map,
3509                                                         SIT_VBLOCK_MAP_SIZE);
3510                                 sbi->discard_blks += old_valid_blocks -
3511                                                         se->valid_blocks;
3512                         }
3513                 }
3514
3515                 if (sbi->segs_per_sec > 1)
3516                         get_sec_entry(sbi, start)->valid_blocks +=
3517                                 se->valid_blocks - old_valid_blocks;
3518         }
3519         up_read(&curseg->journal_rwsem);
3520 }
3521
3522 static void init_free_segmap(struct f2fs_sb_info *sbi)
3523 {
3524         unsigned int start;
3525         int type;
3526
3527         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3528                 struct seg_entry *sentry = get_seg_entry(sbi, start);
3529                 if (!sentry->valid_blocks)
3530                         __set_free(sbi, start);
3531                 else
3532                         SIT_I(sbi)->written_valid_blocks +=
3533                                                 sentry->valid_blocks;
3534         }
3535
3536         /* set use the current segments */
3537         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3538                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3539                 __set_test_and_inuse(sbi, curseg_t->segno);
3540         }
3541 }
3542
3543 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3544 {
3545         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3546         struct free_segmap_info *free_i = FREE_I(sbi);
3547         unsigned int segno = 0, offset = 0;
3548         unsigned short valid_blocks;
3549
3550         while (1) {
3551                 /* find dirty segment based on free segmap */
3552                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3553                 if (segno >= MAIN_SEGS(sbi))
3554                         break;
3555                 offset = segno + 1;
3556                 valid_blocks = get_valid_blocks(sbi, segno, false);
3557                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
3558                         continue;
3559                 if (valid_blocks > sbi->blocks_per_seg) {
3560                         f2fs_bug_on(sbi, 1);
3561                         continue;
3562                 }
3563                 mutex_lock(&dirty_i->seglist_lock);
3564                 __locate_dirty_segment(sbi, segno, DIRTY);
3565                 mutex_unlock(&dirty_i->seglist_lock);
3566         }
3567 }
3568
3569 static int init_victim_secmap(struct f2fs_sb_info *sbi)
3570 {
3571         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3572         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3573
3574         dirty_i->victim_secmap = kvzalloc(bitmap_size, GFP_KERNEL);
3575         if (!dirty_i->victim_secmap)
3576                 return -ENOMEM;
3577         return 0;
3578 }
3579
3580 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3581 {
3582         struct dirty_seglist_info *dirty_i;
3583         unsigned int bitmap_size, i;
3584
3585         /* allocate memory for dirty segments list information */
3586         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
3587         if (!dirty_i)
3588                 return -ENOMEM;
3589
3590         SM_I(sbi)->dirty_info = dirty_i;
3591         mutex_init(&dirty_i->seglist_lock);
3592
3593         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3594
3595         for (i = 0; i < NR_DIRTY_TYPE; i++) {
3596                 dirty_i->dirty_segmap[i] = kvzalloc(bitmap_size, GFP_KERNEL);
3597                 if (!dirty_i->dirty_segmap[i])
3598                         return -ENOMEM;
3599         }
3600
3601         init_dirty_segmap(sbi);
3602         return init_victim_secmap(sbi);
3603 }
3604
3605 /*
3606  * Update min, max modified time for cost-benefit GC algorithm
3607  */
3608 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3609 {
3610         struct sit_info *sit_i = SIT_I(sbi);
3611         unsigned int segno;
3612
3613         mutex_lock(&sit_i->sentry_lock);
3614
3615         sit_i->min_mtime = LLONG_MAX;
3616
3617         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
3618                 unsigned int i;
3619                 unsigned long long mtime = 0;
3620
3621                 for (i = 0; i < sbi->segs_per_sec; i++)
3622                         mtime += get_seg_entry(sbi, segno + i)->mtime;
3623
3624                 mtime = div_u64(mtime, sbi->segs_per_sec);
3625
3626                 if (sit_i->min_mtime > mtime)
3627                         sit_i->min_mtime = mtime;
3628         }
3629         sit_i->max_mtime = get_mtime(sbi);
3630         mutex_unlock(&sit_i->sentry_lock);
3631 }
3632
3633 int build_segment_manager(struct f2fs_sb_info *sbi)
3634 {
3635         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3636         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3637         struct f2fs_sm_info *sm_info;
3638         int err;
3639
3640         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
3641         if (!sm_info)
3642                 return -ENOMEM;
3643
3644         /* init sm info */
3645         sbi->sm_info = sm_info;
3646         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3647         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3648         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3649         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3650         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3651         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3652         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
3653         sm_info->rec_prefree_segments = sm_info->main_segments *
3654                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
3655         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3656                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3657
3658         if (!test_opt(sbi, LFS))
3659                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
3660         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
3661         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
3662         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
3663
3664         sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
3665
3666         INIT_LIST_HEAD(&sm_info->sit_entry_set);
3667
3668         if (!f2fs_readonly(sbi->sb)) {
3669                 err = create_flush_cmd_control(sbi);
3670                 if (err)
3671                         return err;
3672         }
3673
3674         err = create_discard_cmd_control(sbi);
3675         if (err)
3676                 return err;
3677
3678         err = build_sit_info(sbi);
3679         if (err)
3680                 return err;
3681         err = build_free_segmap(sbi);
3682         if (err)
3683                 return err;
3684         err = build_curseg(sbi);
3685         if (err)
3686                 return err;
3687
3688         /* reinit free segmap based on SIT */
3689         build_sit_entries(sbi);
3690
3691         init_free_segmap(sbi);
3692         err = build_dirty_segmap(sbi);
3693         if (err)
3694                 return err;
3695
3696         init_min_max_mtime(sbi);
3697         return 0;
3698 }
3699
3700 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3701                 enum dirty_type dirty_type)
3702 {
3703         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3704
3705         mutex_lock(&dirty_i->seglist_lock);
3706         kvfree(dirty_i->dirty_segmap[dirty_type]);
3707         dirty_i->nr_dirty[dirty_type] = 0;
3708         mutex_unlock(&dirty_i->seglist_lock);
3709 }
3710
3711 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
3712 {
3713         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3714         kvfree(dirty_i->victim_secmap);
3715 }
3716
3717 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3718 {
3719         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3720         int i;
3721
3722         if (!dirty_i)
3723                 return;
3724
3725         /* discard pre-free/dirty segments list */
3726         for (i = 0; i < NR_DIRTY_TYPE; i++)
3727                 discard_dirty_segmap(sbi, i);
3728
3729         destroy_victim_secmap(sbi);
3730         SM_I(sbi)->dirty_info = NULL;
3731         kfree(dirty_i);
3732 }
3733
3734 static void destroy_curseg(struct f2fs_sb_info *sbi)
3735 {
3736         struct curseg_info *array = SM_I(sbi)->curseg_array;
3737         int i;
3738
3739         if (!array)
3740                 return;
3741         SM_I(sbi)->curseg_array = NULL;
3742         for (i = 0; i < NR_CURSEG_TYPE; i++) {
3743                 kfree(array[i].sum_blk);
3744                 kfree(array[i].journal);
3745         }
3746         kfree(array);
3747 }
3748
3749 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3750 {
3751         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3752         if (!free_i)
3753                 return;
3754         SM_I(sbi)->free_info = NULL;
3755         kvfree(free_i->free_segmap);
3756         kvfree(free_i->free_secmap);
3757         kfree(free_i);
3758 }
3759
3760 static void destroy_sit_info(struct f2fs_sb_info *sbi)
3761 {
3762         struct sit_info *sit_i = SIT_I(sbi);
3763         unsigned int start;
3764
3765         if (!sit_i)
3766                 return;
3767
3768         if (sit_i->sentries) {
3769                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3770                         kfree(sit_i->sentries[start].cur_valid_map);
3771 #ifdef CONFIG_F2FS_CHECK_FS
3772                         kfree(sit_i->sentries[start].cur_valid_map_mir);
3773 #endif
3774                         kfree(sit_i->sentries[start].ckpt_valid_map);
3775                         kfree(sit_i->sentries[start].discard_map);
3776                 }
3777         }
3778         kfree(sit_i->tmp_map);
3779
3780         kvfree(sit_i->sentries);
3781         kvfree(sit_i->sec_entries);
3782         kvfree(sit_i->dirty_sentries_bitmap);
3783
3784         SM_I(sbi)->sit_info = NULL;
3785         kfree(sit_i->sit_bitmap);
3786 #ifdef CONFIG_F2FS_CHECK_FS
3787         kfree(sit_i->sit_bitmap_mir);
3788 #endif
3789         kfree(sit_i);
3790 }
3791
3792 void destroy_segment_manager(struct f2fs_sb_info *sbi)
3793 {
3794         struct f2fs_sm_info *sm_info = SM_I(sbi);
3795
3796         if (!sm_info)
3797                 return;
3798         destroy_flush_cmd_control(sbi, true);
3799         destroy_discard_cmd_control(sbi);
3800         destroy_dirty_segmap(sbi);
3801         destroy_curseg(sbi);
3802         destroy_free_segmap(sbi);
3803         destroy_sit_info(sbi);
3804         sbi->sm_info = NULL;
3805         kfree(sm_info);
3806 }
3807
3808 int __init create_segment_manager_caches(void)
3809 {
3810         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
3811                         sizeof(struct discard_entry));
3812         if (!discard_entry_slab)
3813                 goto fail;
3814
3815         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3816                         sizeof(struct discard_cmd));
3817         if (!discard_cmd_slab)
3818                 goto destroy_discard_entry;
3819
3820         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
3821                         sizeof(struct sit_entry_set));
3822         if (!sit_entry_set_slab)
3823                 goto destroy_discard_cmd;
3824
3825         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3826                         sizeof(struct inmem_pages));
3827         if (!inmem_entry_slab)
3828                 goto destroy_sit_entry_set;
3829         return 0;
3830
3831 destroy_sit_entry_set:
3832         kmem_cache_destroy(sit_entry_set_slab);
3833 destroy_discard_cmd:
3834         kmem_cache_destroy(discard_cmd_slab);
3835 destroy_discard_entry:
3836         kmem_cache_destroy(discard_entry_slab);
3837 fail:
3838         return -ENOMEM;
3839 }
3840
3841 void destroy_segment_manager_caches(void)
3842 {
3843         kmem_cache_destroy(sit_entry_set_slab);
3844         kmem_cache_destroy(discard_cmd_slab);
3845         kmem_cache_destroy(discard_entry_slab);
3846         kmem_cache_destroy(inmem_entry_slab);
3847 }