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 f2fs_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_mode == GC_URGENT)
 181                 return true;
 182
 183         return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
 184                         SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
 185 }
 186
 187 void f2fs_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                                 bool trylock)
 221 {
 222         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 223         struct inmem_pages *cur, *tmp;
 224         int err = 0;
 225
 226         list_for_each_entry_safe(cur, tmp, head, list) {
 227                 struct page *page = cur->page;
 228
 229                 if (drop)
 230                         trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 231
 232                 if (trylock) {
 233                         /*
 234                          * to avoid deadlock in between page lock and
 235                          * inmem_lock.
 236                          */
 237                         if (!trylock_page(page))
 238                                 continue;
 239                 } else {
 240                         lock_page(page);
 241                 }
 242
 243                 f2fs_wait_on_page_writeback(page, DATA, true);
 244
 245                 if (recover) {
 246                         struct dnode_of_data dn;
 247                         struct node_info ni;
 248
 249                         trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 250 retry:
 251                         set_new_dnode(&dn, inode, NULL, NULL, 0);
 252                         err = f2fs_get_dnode_of_data(&dn, page->index,
 253                                                                 LOOKUP_NODE);
 254                         if (err) {
 255                                 if (err == -ENOMEM) {
 256                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 257                                         cond_resched();
 258                                         goto retry;
 259                                 }
 260                                 err = -EAGAIN;
 261                                 goto next;
 262                         }
 263
 264                         err = f2fs_get_node_info(sbi, dn.nid, &ni);
 265                         if (err) {
 266                                 f2fs_put_dnode(&dn);
 267                                 return err;
 268                         }
 269
 270                         if (cur->old_addr == NEW_ADDR) {
 271                                 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
 272                                 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
 273                         } else
 274                                 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 275                                         cur->old_addr, ni.version, true, true);
 276                         f2fs_put_dnode(&dn);
 277                 }
 278 next:
 279                 /* we don't need to invalidate this in the sccessful status */
 280                 if (drop || recover) {
 281                         ClearPageUptodate(page);
 282                         clear_cold_data(page);
 283                 }
 284                 set_page_private(page, 0);
 285                 ClearPagePrivate(page);
 286                 f2fs_put_page(page, 1);
 287
 288                 list_del(&cur->list);
 289                 kmem_cache_free(inmem_entry_slab, cur);
 290                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 291         }
 292         return err;
 293 }
 294
 295 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
 296 {
 297         struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
 298         struct inode *inode;
 299         struct f2fs_inode_info *fi;
 300 next:
 301         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 302         if (list_empty(head)) {
 303                 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 304                 return;
 305         }
 306         fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
 307         inode = igrab(&fi->vfs_inode);
 308         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 309
 310         if (inode) {
 311                 if (gc_failure) {
 312                         if (fi->i_gc_failures[GC_FAILURE_ATOMIC])
 313                                 goto drop;
 314                         goto skip;
 315                 }
 316 drop:
 317                 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
 318                 f2fs_drop_inmem_pages(inode);
 319                 iput(inode);
 320         }
 321 skip:
 322         congestion_wait(BLK_RW_ASYNC, HZ/50);
 323         cond_resched();
 324         goto next;
 325 }
 326
 327 void f2fs_drop_inmem_pages(struct inode *inode)
 328 {
 329         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 330         struct f2fs_inode_info *fi = F2FS_I(inode);
 331
 332         while (!list_empty(&fi->inmem_pages)) {
 333                 mutex_lock(&fi->inmem_lock);
 334                 __revoke_inmem_pages(inode, &fi->inmem_pages,
 335                                                 true, false, true);
 336
 337                 if (list_empty(&fi->inmem_pages)) {
 338                         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 339                         if (!list_empty(&fi->inmem_ilist))
 340                                 list_del_init(&fi->inmem_ilist);
 341                         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 342                 }
 343                 mutex_unlock(&fi->inmem_lock);
 344         }
 345
 346         clear_inode_flag(inode, FI_ATOMIC_FILE);
 347         fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
 348         stat_dec_atomic_write(inode);
 349 }
 350
 351 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
 352 {
 353         struct f2fs_inode_info *fi = F2FS_I(inode);
 354         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 355         struct list_head *head = &fi->inmem_pages;
 356         struct inmem_pages *cur = NULL;
 357         struct inmem_pages *tmp;
 358
 359         f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
 360
 361         mutex_lock(&fi->inmem_lock);
 362         list_for_each_entry(tmp, head, list) {
 363                 if (tmp->page == page) {
 364                         cur = tmp;
 365                         break;
 366                 }
 367         }
 368
 369         f2fs_bug_on(sbi, !cur);
 370         list_del(&cur->list);
 371         mutex_unlock(&fi->inmem_lock);
 372
 373         dec_page_count(sbi, F2FS_INMEM_PAGES);
 374         kmem_cache_free(inmem_entry_slab, cur);
 375
 376         ClearPageUptodate(page);
 377         set_page_private(page, 0);
 378         ClearPagePrivate(page);
 379         f2fs_put_page(page, 0);
 380
 381         trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 382 }
 383
 384 static int __f2fs_commit_inmem_pages(struct inode *inode)
 385 {
 386         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 387         struct f2fs_inode_info *fi = F2FS_I(inode);
 388         struct inmem_pages *cur, *tmp;
 389         struct f2fs_io_info fio = {
 390                 .sbi = sbi,
 391                 .ino = inode->i_ino,
 392                 .type = DATA,
 393                 .op = REQ_OP_WRITE,
 394                 .op_flags = REQ_SYNC | REQ_PRIO,
 395                 .io_type = FS_DATA_IO,
 396         };
 397         struct list_head revoke_list;
 398         pgoff_t last_idx = ULONG_MAX;
 399         int err = 0;
 400
 401         INIT_LIST_HEAD(&revoke_list);
 402
 403         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 404                 struct page *page = cur->page;
 405
 406                 lock_page(page);
 407                 if (page->mapping == inode->i_mapping) {
 408                         trace_f2fs_commit_inmem_page(page, INMEM);
 409
 410                         set_page_dirty(page);
 411                         f2fs_wait_on_page_writeback(page, DATA, true);
 412                         if (clear_page_dirty_for_io(page)) {
 413                                 inode_dec_dirty_pages(inode);
 414                                 f2fs_remove_dirty_inode(inode);
 415                         }
 416 retry:
 417                         fio.page = page;
 418                         fio.old_blkaddr = NULL_ADDR;
 419                         fio.encrypted_page = NULL;
 420                         fio.need_lock = LOCK_DONE;
 421                         err = f2fs_do_write_data_page(&fio);
 422                         if (err) {
 423                                 if (err == -ENOMEM) {
 424                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 425                                         cond_resched();
 426                                         goto retry;
 427                                 }
 428                                 unlock_page(page);
 429                                 break;
 430                         }
 431                         /* record old blkaddr for revoking */
 432                         cur->old_addr = fio.old_blkaddr;
 433                         last_idx = page->index;
 434                 }
 435                 unlock_page(page);
 436                 list_move_tail(&cur->list, &revoke_list);
 437         }
 438
 439         if (last_idx != ULONG_MAX)
 440                 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
 441
 442         if (err) {
 443                 /*
 444                  * try to revoke all committed pages, but still we could fail
 445                  * due to no memory or other reason, if that happened, EAGAIN
 446                  * will be returned, which means in such case, transaction is
 447                  * already not integrity, caller should use journal to do the
 448                  * recovery or rewrite & commit last transaction. For other
 449                  * error number, revoking was done by filesystem itself.
 450                  */
 451                 err = __revoke_inmem_pages(inode, &revoke_list,
 452                                                 false, true, false);
 453
 454                 /* drop all uncommitted pages */
 455                 __revoke_inmem_pages(inode, &fi->inmem_pages,
 456                                                 true, false, false);
 457         } else {
 458                 __revoke_inmem_pages(inode, &revoke_list,
 459                                                 false, false, false);
 460         }
 461
 462         return err;
 463 }
 464
 465 int f2fs_commit_inmem_pages(struct inode *inode)
 466 {
 467         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 468         struct f2fs_inode_info *fi = F2FS_I(inode);
 469         int err;
 470
 471         f2fs_balance_fs(sbi, true);
 472
 473         down_write(&fi->i_gc_rwsem[WRITE]);
 474
 475         f2fs_lock_op(sbi);
 476         set_inode_flag(inode, FI_ATOMIC_COMMIT);
 477
 478         mutex_lock(&fi->inmem_lock);
 479         err = __f2fs_commit_inmem_pages(inode);
 480
 481         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 482         if (!list_empty(&fi->inmem_ilist))
 483                 list_del_init(&fi->inmem_ilist);
 484         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 485         mutex_unlock(&fi->inmem_lock);
 486
 487         clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 488
 489         f2fs_unlock_op(sbi);
 490         up_write(&fi->i_gc_rwsem[WRITE]);
 491
 492         return err;
 493 }
 494
 495 /*
 496  * This function balances dirty node and dentry pages.
 497  * In addition, it controls garbage collection.
 498  */
 499 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 500 {
 501         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 502                 f2fs_show_injection_info(FAULT_CHECKPOINT);
 503                 f2fs_stop_checkpoint(sbi, false);
 504         }
 505
 506         /* balance_fs_bg is able to be pending */
 507         if (need && excess_cached_nats(sbi))
 508                 f2fs_balance_fs_bg(sbi);
 509
 510         /*
 511          * We should do GC or end up with checkpoint, if there are so many dirty
 512          * dir/node pages without enough free segments.
 513          */
 514         if (has_not_enough_free_secs(sbi, 0, 0)) {
 515                 mutex_lock(&sbi->gc_mutex);
 516                 f2fs_gc(sbi, false, false, NULL_SEGNO);
 517         }
 518 }
 519
 520 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 521 {
 522         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 523                 return;
 524
 525         /* try to shrink extent cache when there is no enough memory */
 526         if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
 527                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 528
 529         /* check the # of cached NAT entries */
 530         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
 531                 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 532
 533         if (!f2fs_available_free_memory(sbi, FREE_NIDS))
 534                 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
 535         else
 536                 f2fs_build_free_nids(sbi, false, false);
 537
 538         if (!is_idle(sbi) &&
 539                 (!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
 540                 return;
 541
 542         /* checkpoint is the only way to shrink partial cached entries */
 543         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
 544                         !f2fs_available_free_memory(sbi, INO_ENTRIES) ||
 545                         excess_prefree_segs(sbi) ||
 546                         excess_dirty_nats(sbi) ||
 547                         excess_dirty_nodes(sbi) ||
 548                         f2fs_time_over(sbi, CP_TIME)) {
 549                 if (test_opt(sbi, DATA_FLUSH)) {
 550                         struct blk_plug plug;
 551
 552                         blk_start_plug(&plug);
 553                         f2fs_sync_dirty_inodes(sbi, FILE_INODE);
 554                         blk_finish_plug(&plug);
 555                 }
 556                 f2fs_sync_fs(sbi->sb, true);
 557                 stat_inc_bg_cp_count(sbi->stat_info);
 558         }
 559 }
 560
 561 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 562                                 struct block_device *bdev)
 563 {
 564         struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
 565         int ret;
 566
 567         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
 568         bio_set_dev(bio, bdev);
 569         ret = submit_bio_wait(bio);
 570         bio_put(bio);
 571
 572         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 573                                 test_opt(sbi, FLUSH_MERGE), ret);
 574         return ret;
 575 }
 576
 577 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 578 {
 579         int ret = 0;
 580         int i;
 581
 582         if (!f2fs_is_multi_device(sbi))
 583                 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 584
 585         for (i = 0; i < sbi->s_ndevs; i++) {
 586                 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
 587                         continue;
 588                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 589                 if (ret)
 590                         break;
 591         }
 592         return ret;
 593 }
 594
 595 static int issue_flush_thread(void *data)
 596 {
 597         struct f2fs_sb_info *sbi = data;
 598         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 599         wait_queue_head_t *q = &fcc->flush_wait_queue;
 600 repeat:
 601         if (kthread_should_stop())
 602                 return 0;
 603
 604         sb_start_intwrite(sbi->sb);
 605
 606         if (!llist_empty(&fcc->issue_list)) {
 607                 struct flush_cmd *cmd, *next;
 608                 int ret;
 609
 610                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 611                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 612
 613                 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 614
 615                 ret = submit_flush_wait(sbi, cmd->ino);
 616                 atomic_inc(&fcc->issued_flush);
 617
 618                 llist_for_each_entry_safe(cmd, next,
 619                                           fcc->dispatch_list, llnode) {
 620                         cmd->ret = ret;
 621                         complete(&cmd->wait);
 622                 }
 623                 fcc->dispatch_list = NULL;
 624         }
 625
 626         sb_end_intwrite(sbi->sb);
 627
 628         wait_event_interruptible(*q,
 629                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 630         goto repeat;
 631 }
 632
 633 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 634 {
 635         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 636         struct flush_cmd cmd;
 637         int ret;
 638
 639         if (test_opt(sbi, NOBARRIER))
 640                 return 0;
 641
 642         if (!test_opt(sbi, FLUSH_MERGE)) {
 643                 atomic_inc(&fcc->issing_flush);
 644                 ret = submit_flush_wait(sbi, ino);
 645                 atomic_dec(&fcc->issing_flush);
 646                 atomic_inc(&fcc->issued_flush);
 647                 return ret;
 648         }
 649
 650         if (atomic_inc_return(&fcc->issing_flush) == 1 ||
 651             f2fs_is_multi_device(sbi)) {
 652                 ret = submit_flush_wait(sbi, ino);
 653                 atomic_dec(&fcc->issing_flush);
 654
 655                 atomic_inc(&fcc->issued_flush);
 656                 return ret;
 657         }
 658
 659         cmd.ino = ino;
 660         init_completion(&cmd.wait);
 661
 662         llist_add(&cmd.llnode, &fcc->issue_list);
 663
 664         /* update issue_list before we wake up issue_flush thread */
 665         smp_mb();
 666
 667         if (waitqueue_active(&fcc->flush_wait_queue))
 668                 wake_up(&fcc->flush_wait_queue);
 669
 670         if (fcc->f2fs_issue_flush) {
 671                 wait_for_completion(&cmd.wait);
 672                 atomic_dec(&fcc->issing_flush);
 673         } else {
 674                 struct llist_node *list;
 675
 676                 list = llist_del_all(&fcc->issue_list);
 677                 if (!list) {
 678                         wait_for_completion(&cmd.wait);
 679                         atomic_dec(&fcc->issing_flush);
 680                 } else {
 681                         struct flush_cmd *tmp, *next;
 682
 683                         ret = submit_flush_wait(sbi, ino);
 684
 685                         llist_for_each_entry_safe(tmp, next, list, llnode) {
 686                                 if (tmp == &cmd) {
 687                                         cmd.ret = ret;
 688                                         atomic_dec(&fcc->issing_flush);
 689                                         continue;
 690                                 }
 691                                 tmp->ret = ret;
 692                                 complete(&tmp->wait);
 693                         }
 694                 }
 695         }
 696
 697         return cmd.ret;
 698 }
 699
 700 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
 701 {
 702         dev_t dev = sbi->sb->s_bdev->bd_dev;
 703         struct flush_cmd_control *fcc;
 704         int err = 0;
 705
 706         if (SM_I(sbi)->fcc_info) {
 707                 fcc = SM_I(sbi)->fcc_info;
 708                 if (fcc->f2fs_issue_flush)
 709                         return err;
 710                 goto init_thread;
 711         }
 712
 713         fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
 714         if (!fcc)
 715                 return -ENOMEM;
 716         atomic_set(&fcc->issued_flush, 0);
 717         atomic_set(&fcc->issing_flush, 0);
 718         init_waitqueue_head(&fcc->flush_wait_queue);
 719         init_llist_head(&fcc->issue_list);
 720         SM_I(sbi)->fcc_info = fcc;
 721         if (!test_opt(sbi, FLUSH_MERGE))
 722                 return err;
 723
 724 init_thread:
 725         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 726                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 727         if (IS_ERR(fcc->f2fs_issue_flush)) {
 728                 err = PTR_ERR(fcc->f2fs_issue_flush);
 729                 kfree(fcc);
 730                 SM_I(sbi)->fcc_info = NULL;
 731                 return err;
 732         }
 733
 734         return err;
 735 }
 736
 737 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 738 {
 739         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 740
 741         if (fcc && fcc->f2fs_issue_flush) {
 742                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 743
 744                 fcc->f2fs_issue_flush = NULL;
 745                 kthread_stop(flush_thread);
 746         }
 747         if (free) {
 748                 kfree(fcc);
 749                 SM_I(sbi)->fcc_info = NULL;
 750         }
 751 }
 752
 753 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
 754 {
 755         int ret = 0, i;
 756
 757         if (!f2fs_is_multi_device(sbi))
 758                 return 0;
 759
 760         for (i = 1; i < sbi->s_ndevs; i++) {
 761                 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
 762                         continue;
 763                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 764                 if (ret)
 765                         break;
 766
 767                 spin_lock(&sbi->dev_lock);
 768                 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
 769                 spin_unlock(&sbi->dev_lock);
 770         }
 771
 772         return ret;
 773 }
 774
 775 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 776                 enum dirty_type dirty_type)
 777 {
 778         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 779
 780         /* need not be added */
 781         if (IS_CURSEG(sbi, segno))
 782                 return;
 783
 784         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 785                 dirty_i->nr_dirty[dirty_type]++;
 786
 787         if (dirty_type == DIRTY) {
 788                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 789                 enum dirty_type t = sentry->type;
 790
 791                 if (unlikely(t >= DIRTY)) {
 792                         f2fs_bug_on(sbi, 1);
 793                         return;
 794                 }
 795                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 796                         dirty_i->nr_dirty[t]++;
 797         }
 798 }
 799
 800 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 801                 enum dirty_type dirty_type)
 802 {
 803         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 804
 805         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 806                 dirty_i->nr_dirty[dirty_type]--;
 807
 808         if (dirty_type == DIRTY) {
 809                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 810                 enum dirty_type t = sentry->type;
 811
 812                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 813                         dirty_i->nr_dirty[t]--;
 814
 815                 if (get_valid_blocks(sbi, segno, true) == 0)
 816                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 817                                                 dirty_i->victim_secmap);
 818         }
 819 }
 820
 821 /*
 822  * Should not occur error such as -ENOMEM.
 823  * Adding dirty entry into seglist is not critical operation.
 824  * If a given segment is one of current working segments, it won't be added.
 825  */
 826 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 827 {
 828         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 829         unsigned short valid_blocks;
 830
 831         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 832                 return;
 833
 834         mutex_lock(&dirty_i->seglist_lock);
 835
 836         valid_blocks = get_valid_blocks(sbi, segno, false);
 837
 838         if (valid_blocks == 0) {
 839                 __locate_dirty_segment(sbi, segno, PRE);
 840                 __remove_dirty_segment(sbi, segno, DIRTY);
 841         } else if (valid_blocks < sbi->blocks_per_seg) {
 842                 __locate_dirty_segment(sbi, segno, DIRTY);
 843         } else {
 844                 /* Recovery routine with SSR needs this */
 845                 __remove_dirty_segment(sbi, segno, DIRTY);
 846         }
 847
 848         mutex_unlock(&dirty_i->seglist_lock);
 849 }
 850
 851 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 852                 struct block_device *bdev, block_t lstart,
 853                 block_t start, block_t len)
 854 {
 855         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 856         struct list_head *pend_list;
 857         struct discard_cmd *dc;
 858
 859         f2fs_bug_on(sbi, !len);
 860
 861         pend_list = &dcc->pend_list[plist_idx(len)];
 862
 863         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 864         INIT_LIST_HEAD(&dc->list);
 865         dc->bdev = bdev;
 866         dc->lstart = lstart;
 867         dc->start = start;
 868         dc->len = len;
 869         dc->ref = 0;
 870         dc->state = D_PREP;
 871         dc->issuing = 0;
 872         dc->error = 0;
 873         init_completion(&dc->wait);
 874         list_add_tail(&dc->list, pend_list);
 875         spin_lock_init(&dc->lock);
 876         dc->bio_ref = 0;
 877         atomic_inc(&dcc->discard_cmd_cnt);
 878         dcc->undiscard_blks += len;
 879
 880         return dc;
 881 }
 882
 883 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 884                                 struct block_device *bdev, block_t lstart,
 885                                 block_t start, block_t len,
 886                                 struct rb_node *parent, struct rb_node **p)
 887 {
 888         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 889         struct discard_cmd *dc;
 890
 891         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
 892
 893         rb_link_node(&dc->rb_node, parent, p);
 894         rb_insert_color(&dc->rb_node, &dcc->root);
 895
 896         return dc;
 897 }
 898
 899 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
 900                                                         struct discard_cmd *dc)
 901 {
 902         if (dc->state == D_DONE)
 903                 atomic_sub(dc->issuing, &dcc->issing_discard);
 904
 905         list_del(&dc->list);
 906         rb_erase(&dc->rb_node, &dcc->root);
 907         dcc->undiscard_blks -= dc->len;
 908
 909         kmem_cache_free(discard_cmd_slab, dc);
 910
 911         atomic_dec(&dcc->discard_cmd_cnt);
 912 }
 913
 914 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
 915                                                         struct discard_cmd *dc)
 916 {
 917         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 918         unsigned long flags;
 919
 920         trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
 921
 922         spin_lock_irqsave(&dc->lock, flags);
 923         if (dc->bio_ref) {
 924                 spin_unlock_irqrestore(&dc->lock, flags);
 925                 return;
 926         }
 927         spin_unlock_irqrestore(&dc->lock, flags);
 928
 929         f2fs_bug_on(sbi, dc->ref);
 930
 931         if (dc->error == -EOPNOTSUPP)
 932                 dc->error = 0;
 933
 934         if (dc->error)
 935                 f2fs_msg(sbi->sb, KERN_INFO,
 936                         "Issue discard(%u, %u, %u) failed, ret: %d",
 937                         dc->lstart, dc->start, dc->len, dc->error);
 938         __detach_discard_cmd(dcc, dc);
 939 }
 940
 941 static void f2fs_submit_discard_endio(struct bio *bio)
 942 {
 943         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
 944         unsigned long flags;
 945
 946         dc->error = blk_status_to_errno(bio->bi_status);
 947
 948         spin_lock_irqsave(&dc->lock, flags);
 949         dc->bio_ref--;
 950         if (!dc->bio_ref && dc->state == D_SUBMIT) {
 951                 dc->state = D_DONE;
 952                 complete_all(&dc->wait);
 953         }
 954         spin_unlock_irqrestore(&dc->lock, flags);
 955         bio_put(bio);
 956 }
 957
 958 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
 959                                 block_t start, block_t end)
 960 {
 961 #ifdef CONFIG_F2FS_CHECK_FS
 962         struct seg_entry *sentry;
 963         unsigned int segno;
 964         block_t blk = start;
 965         unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
 966         unsigned long *map;
 967
 968         while (blk < end) {
 969                 segno = GET_SEGNO(sbi, blk);
 970                 sentry = get_seg_entry(sbi, segno);
 971                 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
 972
 973                 if (end < START_BLOCK(sbi, segno + 1))
 974                         size = GET_BLKOFF_FROM_SEG0(sbi, end);
 975                 else
 976                         size = max_blocks;
 977                 map = (unsigned long *)(sentry->cur_valid_map);
 978                 offset = __find_rev_next_bit(map, size, offset);
 979                 f2fs_bug_on(sbi, offset != size);
 980                 blk = START_BLOCK(sbi, segno + 1);
 981         }
 982 #endif
 983 }
 984
 985 static void __init_discard_policy(struct f2fs_sb_info *sbi,
 986                                 struct discard_policy *dpolicy,
 987                                 int discard_type, unsigned int granularity)
 988 {
 989         /* common policy */
 990         dpolicy->type = discard_type;
 991         dpolicy->sync = true;
 992         dpolicy->ordered = false;
 993         dpolicy->granularity = granularity;
 994
 995         dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
 996         dpolicy->io_aware_gran = MAX_PLIST_NUM;
 997
 998         if (discard_type == DPOLICY_BG) {
 999                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1000                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1001                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1002                 dpolicy->io_aware = true;
1003                 dpolicy->sync = false;
1004                 dpolicy->ordered = true;
1005                 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1006                         dpolicy->granularity = 1;
1007                         dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1008                 }
1009         } else if (discard_type == DPOLICY_FORCE) {
1010                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1011                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1012                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1013                 dpolicy->io_aware = false;
1014         } else if (discard_type == DPOLICY_FSTRIM) {
1015                 dpolicy->io_aware = false;
1016         } else if (discard_type == DPOLICY_UMOUNT) {
1017                 dpolicy->max_requests = UINT_MAX;
1018                 dpolicy->io_aware = false;
1019         }
1020 }
1021
1022 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1023                                 struct block_device *bdev, block_t lstart,
1024                                 block_t start, block_t len);
1025 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1026 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1027                                                 struct discard_policy *dpolicy,
1028                                                 struct discard_cmd *dc,
1029                                                 unsigned int *issued)
1030 {
1031         struct block_device *bdev = dc->bdev;
1032         struct request_queue *q = bdev_get_queue(bdev);
1033         unsigned int max_discard_blocks =
1034                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1035         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1036         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1037                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1038         int flag = dpolicy->sync ? REQ_SYNC : 0;
1039         block_t lstart, start, len, total_len;
1040         int err = 0;
1041
1042         if (dc->state != D_PREP)
1043                 return 0;
1044
1045         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1046                 return 0;
1047
1048         trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1049
1050         lstart = dc->lstart;
1051         start = dc->start;
1052         len = dc->len;
1053         total_len = len;
1054
1055         dc->len = 0;
1056
1057         while (total_len && *issued < dpolicy->max_requests && !err) {
1058                 struct bio *bio = NULL;
1059                 unsigned long flags;
1060                 bool last = true;
1061
1062                 if (len > max_discard_blocks) {
1063                         len = max_discard_blocks;
1064                         last = false;
1065                 }
1066
1067                 (*issued)++;
1068                 if (*issued == dpolicy->max_requests)
1069                         last = true;
1070
1071                 dc->len += len;
1072
1073                 if (time_to_inject(sbi, FAULT_DISCARD)) {
1074                         f2fs_show_injection_info(FAULT_DISCARD);
1075                         err = -EIO;
1076                         goto submit;
1077                 }
1078                 err = __blkdev_issue_discard(bdev,
1079                                         SECTOR_FROM_BLOCK(start),
1080                                         SECTOR_FROM_BLOCK(len),
1081                                         GFP_NOFS, 0, &bio);
1082 submit:
1083                 if (err) {
1084                         spin_lock_irqsave(&dc->lock, flags);
1085                         if (dc->state == D_PARTIAL)
1086                                 dc->state = D_SUBMIT;
1087                         spin_unlock_irqrestore(&dc->lock, flags);
1088
1089                         break;
1090                 }
1091
1092                 f2fs_bug_on(sbi, !bio);
1093
1094                 /*
1095                  * should keep before submission to avoid D_DONE
1096                  * right away
1097                  */
1098                 spin_lock_irqsave(&dc->lock, flags);
1099                 if (last)
1100                         dc->state = D_SUBMIT;
1101                 else
1102                         dc->state = D_PARTIAL;
1103                 dc->bio_ref++;
1104                 spin_unlock_irqrestore(&dc->lock, flags);
1105
1106                 atomic_inc(&dcc->issing_discard);
1107                 dc->issuing++;
1108                 list_move_tail(&dc->list, wait_list);
1109
1110                 /* sanity check on discard range */
1111                 __check_sit_bitmap(sbi, lstart, lstart + len);
1112
1113                 bio->bi_private = dc;
1114                 bio->bi_end_io = f2fs_submit_discard_endio;
1115                 bio->bi_opf |= flag;
1116                 submit_bio(bio);
1117
1118                 atomic_inc(&dcc->issued_discard);
1119
1120                 f2fs_update_iostat(sbi, FS_DISCARD, 1);
1121
1122                 lstart += len;
1123                 start += len;
1124                 total_len -= len;
1125                 len = total_len;
1126         }
1127
1128         if (!err && len)
1129                 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1130         return err;
1131 }
1132
1133 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1134                                 struct block_device *bdev, block_t lstart,
1135                                 block_t start, block_t len,
1136                                 struct rb_node **insert_p,
1137                                 struct rb_node *insert_parent)
1138 {
1139         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1140         struct rb_node **p;
1141         struct rb_node *parent = NULL;
1142         struct discard_cmd *dc = NULL;
1143
1144         if (insert_p && insert_parent) {
1145                 parent = insert_parent;
1146                 p = insert_p;
1147                 goto do_insert;
1148         }
1149
1150         p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
1151 do_insert:
1152         dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
1153         if (!dc)
1154                 return NULL;
1155
1156         return dc;
1157 }
1158
1159 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1160                                                 struct discard_cmd *dc)
1161 {
1162         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1163 }
1164
1165 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1166                                 struct discard_cmd *dc, block_t blkaddr)
1167 {
1168         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1169         struct discard_info di = dc->di;
1170         bool modified = false;
1171
1172         if (dc->state == D_DONE || dc->len == 1) {
1173                 __remove_discard_cmd(sbi, dc);
1174                 return;
1175         }
1176
1177         dcc->undiscard_blks -= di.len;
1178
1179         if (blkaddr > di.lstart) {
1180                 dc->len = blkaddr - dc->lstart;
1181                 dcc->undiscard_blks += dc->len;
1182                 __relocate_discard_cmd(dcc, dc);
1183                 modified = true;
1184         }
1185
1186         if (blkaddr < di.lstart + di.len - 1) {
1187                 if (modified) {
1188                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1189                                         di.start + blkaddr + 1 - di.lstart,
1190                                         di.lstart + di.len - 1 - blkaddr,
1191                                         NULL, NULL);
1192                 } else {
1193                         dc->lstart++;
1194                         dc->len--;
1195                         dc->start++;
1196                         dcc->undiscard_blks += dc->len;
1197                         __relocate_discard_cmd(dcc, dc);
1198                 }
1199         }
1200 }
1201
1202 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1203                                 struct block_device *bdev, block_t lstart,
1204                                 block_t start, block_t len)
1205 {
1206         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1207         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1208         struct discard_cmd *dc;
1209         struct discard_info di = {0};
1210         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1211         struct request_queue *q = bdev_get_queue(bdev);
1212         unsigned int max_discard_blocks =
1213                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1214         block_t end = lstart + len;
1215
1216         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1217                                         NULL, lstart,
1218                                         (struct rb_entry **)&prev_dc,
1219                                         (struct rb_entry **)&next_dc,
1220                                         &insert_p, &insert_parent, true);
1221         if (dc)
1222                 prev_dc = dc;
1223
1224         if (!prev_dc) {
1225                 di.lstart = lstart;
1226                 di.len = next_dc ? next_dc->lstart - lstart : len;
1227                 di.len = min(di.len, len);
1228                 di.start = start;
1229         }
1230
1231         while (1) {
1232                 struct rb_node *node;
1233                 bool merged = false;
1234                 struct discard_cmd *tdc = NULL;
1235
1236                 if (prev_dc) {
1237                         di.lstart = prev_dc->lstart + prev_dc->len;
1238                         if (di.lstart < lstart)
1239                                 di.lstart = lstart;
1240                         if (di.lstart >= end)
1241                                 break;
1242
1243                         if (!next_dc || next_dc->lstart > end)
1244                                 di.len = end - di.lstart;
1245                         else
1246                                 di.len = next_dc->lstart - di.lstart;
1247                         di.start = start + di.lstart - lstart;
1248                 }
1249
1250                 if (!di.len)
1251                         goto next;
1252
1253                 if (prev_dc && prev_dc->state == D_PREP &&
1254                         prev_dc->bdev == bdev &&
1255                         __is_discard_back_mergeable(&di, &prev_dc->di,
1256                                                         max_discard_blocks)) {
1257                         prev_dc->di.len += di.len;
1258                         dcc->undiscard_blks += di.len;
1259                         __relocate_discard_cmd(dcc, prev_dc);
1260                         di = prev_dc->di;
1261                         tdc = prev_dc;
1262                         merged = true;
1263                 }
1264
1265                 if (next_dc && next_dc->state == D_PREP &&
1266                         next_dc->bdev == bdev &&
1267                         __is_discard_front_mergeable(&di, &next_dc->di,
1268                                                         max_discard_blocks)) {
1269                         next_dc->di.lstart = di.lstart;
1270                         next_dc->di.len += di.len;
1271                         next_dc->di.start = di.start;
1272                         dcc->undiscard_blks += di.len;
1273                         __relocate_discard_cmd(dcc, next_dc);
1274                         if (tdc)
1275                                 __remove_discard_cmd(sbi, tdc);
1276                         merged = true;
1277                 }
1278
1279                 if (!merged) {
1280                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1281                                                         di.len, NULL, NULL);
1282                 }
1283  next:
1284                 prev_dc = next_dc;
1285                 if (!prev_dc)
1286                         break;
1287
1288                 node = rb_next(&prev_dc->rb_node);
1289                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1290         }
1291 }
1292
1293 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1294                 struct block_device *bdev, block_t blkstart, block_t blklen)
1295 {
1296         block_t lblkstart = blkstart;
1297
1298         trace_f2fs_queue_discard(bdev, blkstart, blklen);
1299
1300         if (f2fs_is_multi_device(sbi)) {
1301                 int devi = f2fs_target_device_index(sbi, blkstart);
1302
1303                 blkstart -= FDEV(devi).start_blk;
1304         }
1305         mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1306         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1307         mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1308         return 0;
1309 }
1310
1311 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1312                                         struct discard_policy *dpolicy)
1313 {
1314         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1315         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1316         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1317         struct discard_cmd *dc;
1318         struct blk_plug plug;
1319         unsigned int pos = dcc->next_pos;
1320         unsigned int issued = 0;
1321         bool io_interrupted = false;
1322
1323         mutex_lock(&dcc->cmd_lock);
1324         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1325                                         NULL, pos,
1326                                         (struct rb_entry **)&prev_dc,
1327                                         (struct rb_entry **)&next_dc,
1328                                         &insert_p, &insert_parent, true);
1329         if (!dc)
1330                 dc = next_dc;
1331
1332         blk_start_plug(&plug);
1333
1334         while (dc) {
1335                 struct rb_node *node;
1336                 int err = 0;
1337
1338                 if (dc->state != D_PREP)
1339                         goto next;
1340
1341                 if (dpolicy->io_aware && !is_idle(sbi)) {
1342                         io_interrupted = true;
1343                         break;
1344                 }
1345
1346                 dcc->next_pos = dc->lstart + dc->len;
1347                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1348
1349                 if (issued >= dpolicy->max_requests)
1350                         break;
1351 next:
1352                 node = rb_next(&dc->rb_node);
1353                 if (err)
1354                         __remove_discard_cmd(sbi, dc);
1355                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1356         }
1357
1358         blk_finish_plug(&plug);
1359
1360         if (!dc)
1361                 dcc->next_pos = 0;
1362
1363         mutex_unlock(&dcc->cmd_lock);
1364
1365         if (!issued && io_interrupted)
1366                 issued = -1;
1367
1368         return issued;
1369 }
1370
1371 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1372                                         struct discard_policy *dpolicy)
1373 {
1374         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1375         struct list_head *pend_list;
1376         struct discard_cmd *dc, *tmp;
1377         struct blk_plug plug;
1378         int i, issued = 0;
1379         bool io_interrupted = false;
1380
1381         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1382                 if (i + 1 < dpolicy->granularity)
1383                         break;
1384
1385                 if (i + 1 < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1386                         return __issue_discard_cmd_orderly(sbi, dpolicy);
1387
1388                 pend_list = &dcc->pend_list[i];
1389
1390                 mutex_lock(&dcc->cmd_lock);
1391                 if (list_empty(pend_list))
1392                         goto next;
1393                 if (unlikely(dcc->rbtree_check))
1394                         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1395                                                                 &dcc->root));
1396                 blk_start_plug(&plug);
1397                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1398                         f2fs_bug_on(sbi, dc->state != D_PREP);
1399
1400                         if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1401                                                                 !is_idle(sbi)) {
1402                                 io_interrupted = true;
1403                                 break;
1404                         }
1405
1406                         __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1407
1408                         if (issued >= dpolicy->max_requests)
1409                                 break;
1410                 }
1411                 blk_finish_plug(&plug);
1412 next:
1413                 mutex_unlock(&dcc->cmd_lock);
1414
1415                 if (issued >= dpolicy->max_requests || io_interrupted)
1416                         break;
1417         }
1418
1419         if (!issued && io_interrupted)
1420                 issued = -1;
1421
1422         return issued;
1423 }
1424
1425 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1426 {
1427         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1428         struct list_head *pend_list;
1429         struct discard_cmd *dc, *tmp;
1430         int i;
1431         bool dropped = false;
1432
1433         mutex_lock(&dcc->cmd_lock);
1434         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1435                 pend_list = &dcc->pend_list[i];
1436                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1437                         f2fs_bug_on(sbi, dc->state != D_PREP);
1438                         __remove_discard_cmd(sbi, dc);
1439                         dropped = true;
1440                 }
1441         }
1442         mutex_unlock(&dcc->cmd_lock);
1443
1444         return dropped;
1445 }
1446
1447 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1448 {
1449         __drop_discard_cmd(sbi);
1450 }
1451
1452 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1453                                                         struct discard_cmd *dc)
1454 {
1455         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1456         unsigned int len = 0;
1457
1458         wait_for_completion_io(&dc->wait);
1459         mutex_lock(&dcc->cmd_lock);
1460         f2fs_bug_on(sbi, dc->state != D_DONE);
1461         dc->ref--;
1462         if (!dc->ref) {
1463                 if (!dc->error)
1464                         len = dc->len;
1465                 __remove_discard_cmd(sbi, dc);
1466         }
1467         mutex_unlock(&dcc->cmd_lock);
1468
1469         return len;
1470 }
1471
1472 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1473                                                 struct discard_policy *dpolicy,
1474                                                 block_t start, block_t end)
1475 {
1476         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1477         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1478                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1479         struct discard_cmd *dc, *tmp;
1480         bool need_wait;
1481         unsigned int trimmed = 0;
1482
1483 next:
1484         need_wait = false;
1485
1486         mutex_lock(&dcc->cmd_lock);
1487         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1488                 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1489                         continue;
1490                 if (dc->len < dpolicy->granularity)
1491                         continue;
1492                 if (dc->state == D_DONE && !dc->ref) {
1493                         wait_for_completion_io(&dc->wait);
1494                         if (!dc->error)
1495                                 trimmed += dc->len;
1496                         __remove_discard_cmd(sbi, dc);
1497                 } else {
1498                         dc->ref++;
1499                         need_wait = true;
1500                         break;
1501                 }
1502         }
1503         mutex_unlock(&dcc->cmd_lock);
1504
1505         if (need_wait) {
1506                 trimmed += __wait_one_discard_bio(sbi, dc);
1507                 goto next;
1508         }
1509
1510         return trimmed;
1511 }
1512
1513 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1514                                                 struct discard_policy *dpolicy)
1515 {
1516         struct discard_policy dp;
1517         unsigned int discard_blks;
1518
1519         if (dpolicy)
1520                 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1521
1522         /* wait all */
1523         __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1524         discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1525         __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1526         discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1527
1528         return discard_blks;
1529 }
1530
1531 /* This should be covered by global mutex, &sit_i->sentry_lock */
1532 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1533 {
1534         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1535         struct discard_cmd *dc;
1536         bool need_wait = false;
1537
1538         mutex_lock(&dcc->cmd_lock);
1539         dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1540                                                         NULL, blkaddr);
1541         if (dc) {
1542                 if (dc->state == D_PREP) {
1543                         __punch_discard_cmd(sbi, dc, blkaddr);
1544                 } else {
1545                         dc->ref++;
1546                         need_wait = true;
1547                 }
1548         }
1549         mutex_unlock(&dcc->cmd_lock);
1550
1551         if (need_wait)
1552                 __wait_one_discard_bio(sbi, dc);
1553 }
1554
1555 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1556 {
1557         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1558
1559         if (dcc && dcc->f2fs_issue_discard) {
1560                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1561
1562                 dcc->f2fs_issue_discard = NULL;
1563                 kthread_stop(discard_thread);
1564         }
1565 }
1566
1567 /* This comes from f2fs_put_super */
1568 bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1569 {
1570         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1571         struct discard_policy dpolicy;
1572         bool dropped;
1573
1574         __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1575                                         dcc->discard_granularity);
1576         __issue_discard_cmd(sbi, &dpolicy);
1577         dropped = __drop_discard_cmd(sbi);
1578
1579         /* just to make sure there is no pending discard commands */
1580         __wait_all_discard_cmd(sbi, NULL);
1581
1582         f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1583         return dropped;
1584 }
1585
1586 static int issue_discard_thread(void *data)
1587 {
1588         struct f2fs_sb_info *sbi = data;
1589         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1590         wait_queue_head_t *q = &dcc->discard_wait_queue;
1591         struct discard_policy dpolicy;
1592         unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1593         int issued;
1594
1595         set_freezable();
1596
1597         do {
1598                 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1599                                         dcc->discard_granularity);
1600
1601                 wait_event_interruptible_timeout(*q,
1602                                 kthread_should_stop() || freezing(current) ||
1603                                 dcc->discard_wake,
1604                                 msecs_to_jiffies(wait_ms));
1605
1606                 if (dcc->discard_wake)
1607                         dcc->discard_wake = 0;
1608
1609                 if (try_to_freeze())
1610                         continue;
1611                 if (f2fs_readonly(sbi->sb))
1612                         continue;
1613                 if (kthread_should_stop())
1614                         return 0;
1615                 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1616                         wait_ms = dpolicy.max_interval;
1617                         continue;
1618                 }
1619
1620                 if (sbi->gc_mode == GC_URGENT)
1621                         __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1622
1623                 sb_start_intwrite(sbi->sb);
1624
1625                 issued = __issue_discard_cmd(sbi, &dpolicy);
1626                 if (issued > 0) {
1627                         __wait_all_discard_cmd(sbi, &dpolicy);
1628                         wait_ms = dpolicy.min_interval;
1629                 } else if (issued == -1){
1630                         wait_ms = dpolicy.mid_interval;
1631                 } else {
1632                         wait_ms = dpolicy.max_interval;
1633                 }
1634
1635                 sb_end_intwrite(sbi->sb);
1636
1637         } while (!kthread_should_stop());
1638         return 0;
1639 }
1640
1641 #ifdef CONFIG_BLK_DEV_ZONED
1642 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1643                 struct block_device *bdev, block_t blkstart, block_t blklen)
1644 {
1645         sector_t sector, nr_sects;
1646         block_t lblkstart = blkstart;
1647         int devi = 0;
1648
1649         if (f2fs_is_multi_device(sbi)) {
1650                 devi = f2fs_target_device_index(sbi, blkstart);
1651                 blkstart -= FDEV(devi).start_blk;
1652         }
1653
1654         /*
1655          * We need to know the type of the zone: for conventional zones,
1656          * use regular discard if the drive supports it. For sequential
1657          * zones, reset the zone write pointer.
1658          */
1659         switch (get_blkz_type(sbi, bdev, blkstart)) {
1660
1661         case BLK_ZONE_TYPE_CONVENTIONAL:
1662                 if (!blk_queue_discard(bdev_get_queue(bdev)))
1663                         return 0;
1664                 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1665         case BLK_ZONE_TYPE_SEQWRITE_REQ:
1666         case BLK_ZONE_TYPE_SEQWRITE_PREF:
1667                 sector = SECTOR_FROM_BLOCK(blkstart);
1668                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1669
1670                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1671                                 nr_sects != bdev_zone_sectors(bdev)) {
1672                         f2fs_msg(sbi->sb, KERN_INFO,
1673                                 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1674                                 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1675                                 blkstart, blklen);
1676                         return -EIO;
1677                 }
1678                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1679                 return blkdev_reset_zones(bdev, sector,
1680                                           nr_sects, GFP_NOFS);
1681         default:
1682                 /* Unknown zone type: broken device ? */
1683                 return -EIO;
1684         }
1685 }
1686 #endif
1687
1688 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1689                 struct block_device *bdev, block_t blkstart, block_t blklen)
1690 {
1691 #ifdef CONFIG_BLK_DEV_ZONED
1692         if (f2fs_sb_has_blkzoned(sbi->sb) &&
1693                                 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1694                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1695 #endif
1696         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1697 }
1698
1699 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1700                                 block_t blkstart, block_t blklen)
1701 {
1702         sector_t start = blkstart, len = 0;
1703         struct block_device *bdev;
1704         struct seg_entry *se;
1705         unsigned int offset;
1706         block_t i;
1707         int err = 0;
1708
1709         bdev = f2fs_target_device(sbi, blkstart, NULL);
1710
1711         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1712                 if (i != start) {
1713                         struct block_device *bdev2 =
1714                                 f2fs_target_device(sbi, i, NULL);
1715
1716                         if (bdev2 != bdev) {
1717                                 err = __issue_discard_async(sbi, bdev,
1718                                                 start, len);
1719                                 if (err)
1720                                         return err;
1721                                 bdev = bdev2;
1722                                 start = i;
1723                                 len = 0;
1724                         }
1725                 }
1726
1727                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1728                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1729
1730                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1731                         sbi->discard_blks--;
1732         }
1733
1734         if (len)
1735                 err = __issue_discard_async(sbi, bdev, start, len);
1736         return err;
1737 }
1738
1739 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1740                                                         bool check_only)
1741 {
1742         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1743         int max_blocks = sbi->blocks_per_seg;
1744         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1745         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1746         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1747         unsigned long *discard_map = (unsigned long *)se->discard_map;
1748         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1749         unsigned int start = 0, end = -1;
1750         bool force = (cpc->reason & CP_DISCARD);
1751         struct discard_entry *de = NULL;
1752         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1753         int i;
1754
1755         if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
1756                 return false;
1757
1758         if (!force) {
1759                 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1760                         SM_I(sbi)->dcc_info->nr_discards >=
1761                                 SM_I(sbi)->dcc_info->max_discards)
1762                         return false;
1763         }
1764
1765         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1766         for (i = 0; i < entries; i++)
1767                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1768                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1769
1770         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1771                                 SM_I(sbi)->dcc_info->max_discards) {
1772                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1773                 if (start >= max_blocks)
1774                         break;
1775
1776                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1777                 if (force && start && end != max_blocks
1778                                         && (end - start) < cpc->trim_minlen)
1779                         continue;
1780
1781                 if (check_only)
1782                         return true;
1783
1784                 if (!de) {
1785                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1786                                                                 GFP_F2FS_ZERO);
1787                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1788                         list_add_tail(&de->list, head);
1789                 }
1790
1791                 for (i = start; i < end; i++)
1792                         __set_bit_le(i, (void *)de->discard_map);
1793
1794                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1795         }
1796         return false;
1797 }
1798
1799 static void release_discard_addr(struct discard_entry *entry)
1800 {
1801         list_del(&entry->list);
1802         kmem_cache_free(discard_entry_slab, entry);
1803 }
1804
1805 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1806 {
1807         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1808         struct discard_entry *entry, *this;
1809
1810         /* drop caches */
1811         list_for_each_entry_safe(entry, this, head, list)
1812                 release_discard_addr(entry);
1813 }
1814
1815 /*
1816  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1817  */
1818 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1819 {
1820         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1821         unsigned int segno;
1822
1823         mutex_lock(&dirty_i->seglist_lock);
1824         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1825                 __set_test_and_free(sbi, segno);
1826         mutex_unlock(&dirty_i->seglist_lock);
1827 }
1828
1829 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1830                                                 struct cp_control *cpc)
1831 {
1832         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1833         struct list_head *head = &dcc->entry_list;
1834         struct discard_entry *entry, *this;
1835         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1836         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1837         unsigned int start = 0, end = -1;
1838         unsigned int secno, start_segno;
1839         bool force = (cpc->reason & CP_DISCARD);
1840         bool need_align = test_opt(sbi, LFS) && sbi->segs_per_sec > 1;
1841
1842         mutex_lock(&dirty_i->seglist_lock);
1843
1844         while (1) {
1845                 int i;
1846
1847                 if (need_align && end != -1)
1848                         end--;
1849                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1850                 if (start >= MAIN_SEGS(sbi))
1851                         break;
1852                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1853                                                                 start + 1);
1854
1855                 if (need_align) {
1856                         start = rounddown(start, sbi->segs_per_sec);
1857                         end = roundup(end, sbi->segs_per_sec);
1858                 }
1859
1860                 for (i = start; i < end; i++) {
1861                         if (test_and_clear_bit(i, prefree_map))
1862                                 dirty_i->nr_dirty[PRE]--;
1863                 }
1864
1865                 if (!f2fs_realtime_discard_enable(sbi))
1866                         continue;
1867
1868                 if (force && start >= cpc->trim_start &&
1869                                         (end - 1) <= cpc->trim_end)
1870                                 continue;
1871
1872                 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1873                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1874                                 (end - start) << sbi->log_blocks_per_seg);
1875                         continue;
1876                 }
1877 next:
1878                 secno = GET_SEC_FROM_SEG(sbi, start);
1879                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1880                 if (!IS_CURSEC(sbi, secno) &&
1881                         !get_valid_blocks(sbi, start, true))
1882                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1883                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1884
1885                 start = start_segno + sbi->segs_per_sec;
1886                 if (start < end)
1887                         goto next;
1888                 else
1889                         end = start - 1;
1890         }
1891         mutex_unlock(&dirty_i->seglist_lock);
1892
1893         /* send small discards */
1894         list_for_each_entry_safe(entry, this, head, list) {
1895                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1896                 bool is_valid = test_bit_le(0, entry->discard_map);
1897
1898 find_next:
1899                 if (is_valid) {
1900                         next_pos = find_next_zero_bit_le(entry->discard_map,
1901                                         sbi->blocks_per_seg, cur_pos);
1902                         len = next_pos - cur_pos;
1903
1904                         if (f2fs_sb_has_blkzoned(sbi->sb) ||
1905                             (force && len < cpc->trim_minlen))
1906                                 goto skip;
1907
1908                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1909                                                                         len);
1910                         total_len += len;
1911                 } else {
1912                         next_pos = find_next_bit_le(entry->discard_map,
1913                                         sbi->blocks_per_seg, cur_pos);
1914                 }
1915 skip:
1916                 cur_pos = next_pos;
1917                 is_valid = !is_valid;
1918
1919                 if (cur_pos < sbi->blocks_per_seg)
1920                         goto find_next;
1921
1922                 release_discard_addr(entry);
1923                 dcc->nr_discards -= total_len;
1924         }
1925
1926         wake_up_discard_thread(sbi, false);
1927 }
1928
1929 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1930 {
1931         dev_t dev = sbi->sb->s_bdev->bd_dev;
1932         struct discard_cmd_control *dcc;
1933         int err = 0, i;
1934
1935         if (SM_I(sbi)->dcc_info) {
1936                 dcc = SM_I(sbi)->dcc_info;
1937                 goto init_thread;
1938         }
1939
1940         dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
1941         if (!dcc)
1942                 return -ENOMEM;
1943
1944         dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
1945         INIT_LIST_HEAD(&dcc->entry_list);
1946         for (i = 0; i < MAX_PLIST_NUM; i++)
1947                 INIT_LIST_HEAD(&dcc->pend_list[i]);
1948         INIT_LIST_HEAD(&dcc->wait_list);
1949         INIT_LIST_HEAD(&dcc->fstrim_list);
1950         mutex_init(&dcc->cmd_lock);
1951         atomic_set(&dcc->issued_discard, 0);
1952         atomic_set(&dcc->issing_discard, 0);
1953         atomic_set(&dcc->discard_cmd_cnt, 0);
1954         dcc->nr_discards = 0;
1955         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1956         dcc->undiscard_blks = 0;
1957         dcc->next_pos = 0;
1958         dcc->root = RB_ROOT;
1959         dcc->rbtree_check = false;
1960
1961         init_waitqueue_head(&dcc->discard_wait_queue);
1962         SM_I(sbi)->dcc_info = dcc;
1963 init_thread:
1964         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1965                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1966         if (IS_ERR(dcc->f2fs_issue_discard)) {
1967                 err = PTR_ERR(dcc->f2fs_issue_discard);
1968                 kfree(dcc);
1969                 SM_I(sbi)->dcc_info = NULL;
1970                 return err;
1971         }
1972
1973         return err;
1974 }
1975
1976 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1977 {
1978         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1979
1980         if (!dcc)
1981                 return;
1982
1983         f2fs_stop_discard_thread(sbi);
1984
1985         kfree(dcc);
1986         SM_I(sbi)->dcc_info = NULL;
1987 }
1988
1989 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1990 {
1991         struct sit_info *sit_i = SIT_I(sbi);
1992
1993         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1994                 sit_i->dirty_sentries++;
1995                 return false;
1996         }
1997
1998         return true;
1999 }
2000
2001 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2002                                         unsigned int segno, int modified)
2003 {
2004         struct seg_entry *se = get_seg_entry(sbi, segno);
2005         se->type = type;
2006         if (modified)
2007                 __mark_sit_entry_dirty(sbi, segno);
2008 }
2009
2010 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2011 {
2012         struct seg_entry *se;
2013         unsigned int segno, offset;
2014         long int new_vblocks;
2015         bool exist;
2016 #ifdef CONFIG_F2FS_CHECK_FS
2017         bool mir_exist;
2018 #endif
2019
2020         segno = GET_SEGNO(sbi, blkaddr);
2021
2022         se = get_seg_entry(sbi, segno);
2023         new_vblocks = se->valid_blocks + del;
2024         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2025
2026         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
2027                                 (new_vblocks > sbi->blocks_per_seg)));
2028
2029         se->valid_blocks = new_vblocks;
2030         se->mtime = get_mtime(sbi, false);
2031         if (se->mtime > SIT_I(sbi)->max_mtime)
2032                 SIT_I(sbi)->max_mtime = se->mtime;
2033
2034         /* Update valid block bitmap */
2035         if (del > 0) {
2036                 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2037 #ifdef CONFIG_F2FS_CHECK_FS
2038                 mir_exist = f2fs_test_and_set_bit(offset,
2039                                                 se->cur_valid_map_mir);
2040                 if (unlikely(exist != mir_exist)) {
2041                         f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2042                                 "when setting bitmap, blk:%u, old bit:%d",
2043                                 blkaddr, exist);
2044                         f2fs_bug_on(sbi, 1);
2045                 }
2046 #endif
2047                 if (unlikely(exist)) {
2048                         f2fs_msg(sbi->sb, KERN_ERR,
2049                                 "Bitmap was wrongly set, blk:%u", blkaddr);
2050                         f2fs_bug_on(sbi, 1);
2051                         se->valid_blocks--;
2052                         del = 0;
2053                 }
2054
2055                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
2056                         sbi->discard_blks--;
2057
2058                 /* don't overwrite by SSR to keep node chain */
2059                 if (IS_NODESEG(se->type)) {
2060                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2061                                 se->ckpt_valid_blocks++;
2062                 }
2063         } else {
2064                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2065 #ifdef CONFIG_F2FS_CHECK_FS
2066                 mir_exist = f2fs_test_and_clear_bit(offset,
2067                                                 se->cur_valid_map_mir);
2068                 if (unlikely(exist != mir_exist)) {
2069                         f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2070                                 "when clearing bitmap, blk:%u, old bit:%d",
2071                                 blkaddr, exist);
2072                         f2fs_bug_on(sbi, 1);
2073                 }
2074 #endif
2075                 if (unlikely(!exist)) {
2076                         f2fs_msg(sbi->sb, KERN_ERR,
2077                                 "Bitmap was wrongly cleared, blk:%u", blkaddr);
2078                         f2fs_bug_on(sbi, 1);
2079                         se->valid_blocks++;
2080                         del = 0;
2081                 }
2082
2083                 if (f2fs_test_and_clear_bit(offset, se->discard_map))
2084                         sbi->discard_blks++;
2085         }
2086         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2087                 se->ckpt_valid_blocks += del;
2088
2089         __mark_sit_entry_dirty(sbi, segno);
2090
2091         /* update total number of valid blocks to be written in ckpt area */
2092         SIT_I(sbi)->written_valid_blocks += del;
2093
2094         if (sbi->segs_per_sec > 1)
2095                 get_sec_entry(sbi, segno)->valid_blocks += del;
2096 }
2097
2098 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2099 {
2100         unsigned int segno = GET_SEGNO(sbi, addr);
2101         struct sit_info *sit_i = SIT_I(sbi);
2102
2103         f2fs_bug_on(sbi, addr == NULL_ADDR);
2104         if (addr == NEW_ADDR)
2105                 return;
2106
2107         invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2108
2109         /* add it into sit main buffer */
2110         down_write(&sit_i->sentry_lock);
2111
2112         update_sit_entry(sbi, addr, -1);
2113
2114         /* add it into dirty seglist */
2115         locate_dirty_segment(sbi, segno);
2116
2117         up_write(&sit_i->sentry_lock);
2118 }
2119
2120 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2121 {
2122         struct sit_info *sit_i = SIT_I(sbi);
2123         unsigned int segno, offset;
2124         struct seg_entry *se;
2125         bool is_cp = false;
2126
2127         if (!is_valid_data_blkaddr(sbi, blkaddr))
2128                 return true;
2129
2130         down_read(&sit_i->sentry_lock);
2131
2132         segno = GET_SEGNO(sbi, blkaddr);
2133         se = get_seg_entry(sbi, segno);
2134         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2135
2136         if (f2fs_test_bit(offset, se->ckpt_valid_map))
2137                 is_cp = true;
2138
2139         up_read(&sit_i->sentry_lock);
2140
2141         return is_cp;
2142 }
2143
2144 /*
2145  * This function should be resided under the curseg_mutex lock
2146  */
2147 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2148                                         struct f2fs_summary *sum)
2149 {
2150         struct curseg_info *curseg = CURSEG_I(sbi, type);
2151         void *addr = curseg->sum_blk;
2152         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2153         memcpy(addr, sum, sizeof(struct f2fs_summary));
2154 }
2155
2156 /*
2157  * Calculate the number of current summary pages for writing
2158  */
2159 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2160 {
2161         int valid_sum_count = 0;
2162         int i, sum_in_page;
2163
2164         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2165                 if (sbi->ckpt->alloc_type[i] == SSR)
2166                         valid_sum_count += sbi->blocks_per_seg;
2167                 else {
2168                         if (for_ra)
2169                                 valid_sum_count += le16_to_cpu(
2170                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2171                         else
2172                                 valid_sum_count += curseg_blkoff(sbi, i);
2173                 }
2174         }
2175
2176         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2177                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2178         if (valid_sum_count <= sum_in_page)
2179                 return 1;
2180         else if ((valid_sum_count - sum_in_page) <=
2181                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2182                 return 2;
2183         return 3;
2184 }
2185
2186 /*
2187  * Caller should put this summary page
2188  */
2189 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2190 {
2191         return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
2192 }
2193
2194 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2195                                         void *src, block_t blk_addr)
2196 {
2197         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2198
2199         memcpy(page_address(page), src, PAGE_SIZE);
2200         set_page_dirty(page);
2201         f2fs_put_page(page, 1);
2202 }
2203
2204 static void write_sum_page(struct f2fs_sb_info *sbi,
2205                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
2206 {
2207         f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2208 }
2209
2210 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2211                                                 int type, block_t blk_addr)
2212 {
2213         struct curseg_info *curseg = CURSEG_I(sbi, type);
2214         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2215         struct f2fs_summary_block *src = curseg->sum_blk;
2216         struct f2fs_summary_block *dst;
2217
2218         dst = (struct f2fs_summary_block *)page_address(page);
2219         memset(dst, 0, PAGE_SIZE);
2220
2221         mutex_lock(&curseg->curseg_mutex);
2222
2223         down_read(&curseg->journal_rwsem);
2224         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2225         up_read(&curseg->journal_rwsem);
2226
2227         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2228         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2229
2230         mutex_unlock(&curseg->curseg_mutex);
2231
2232         set_page_dirty(page);
2233         f2fs_put_page(page, 1);
2234 }
2235
2236 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2237 {
2238         struct curseg_info *curseg = CURSEG_I(sbi, type);
2239         unsigned int segno = curseg->segno + 1;
2240         struct free_segmap_info *free_i = FREE_I(sbi);
2241
2242         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2243                 return !test_bit(segno, free_i->free_segmap);
2244         return 0;
2245 }
2246
2247 /*
2248  * Find a new segment from the free segments bitmap to right order
2249  * This function should be returned with success, otherwise BUG
2250  */
2251 static void get_new_segment(struct f2fs_sb_info *sbi,
2252                         unsigned int *newseg, bool new_sec, int dir)
2253 {
2254         struct free_segmap_info *free_i = FREE_I(sbi);
2255         unsigned int segno, secno, zoneno;
2256         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2257         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2258         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2259         unsigned int left_start = hint;
2260         bool init = true;
2261         int go_left = 0;
2262         int i;
2263
2264         spin_lock(&free_i->segmap_lock);
2265
2266         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2267                 segno = find_next_zero_bit(free_i->free_segmap,
2268                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2269                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2270                         goto got_it;
2271         }
2272 find_other_zone:
2273         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2274         if (secno >= MAIN_SECS(sbi)) {
2275                 if (dir == ALLOC_RIGHT) {
2276                         secno = find_next_zero_bit(free_i->free_secmap,
2277                                                         MAIN_SECS(sbi), 0);
2278                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2279                 } else {
2280                         go_left = 1;
2281                         left_start = hint - 1;
2282                 }
2283         }
2284         if (go_left == 0)
2285                 goto skip_left;
2286
2287         while (test_bit(left_start, free_i->free_secmap)) {
2288                 if (left_start > 0) {
2289                         left_start--;
2290                         continue;
2291                 }
2292                 left_start = find_next_zero_bit(free_i->free_secmap,
2293                                                         MAIN_SECS(sbi), 0);
2294                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2295                 break;
2296         }
2297         secno = left_start;
2298 skip_left:
2299         segno = GET_SEG_FROM_SEC(sbi, secno);
2300         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2301
2302         /* give up on finding another zone */
2303         if (!init)
2304                 goto got_it;
2305         if (sbi->secs_per_zone == 1)
2306                 goto got_it;
2307         if (zoneno == old_zoneno)
2308                 goto got_it;
2309         if (dir == ALLOC_LEFT) {
2310                 if (!go_left && zoneno + 1 >= total_zones)
2311                         goto got_it;
2312                 if (go_left && zoneno == 0)
2313                         goto got_it;
2314         }
2315         for (i = 0; i < NR_CURSEG_TYPE; i++)
2316                 if (CURSEG_I(sbi, i)->zone == zoneno)
2317                         break;
2318
2319         if (i < NR_CURSEG_TYPE) {
2320                 /* zone is in user, try another */
2321                 if (go_left)
2322                         hint = zoneno * sbi->secs_per_zone - 1;
2323                 else if (zoneno + 1 >= total_zones)
2324                         hint = 0;
2325                 else
2326                         hint = (zoneno + 1) * sbi->secs_per_zone;
2327                 init = false;
2328                 goto find_other_zone;
2329         }
2330 got_it:
2331         /* set it as dirty segment in free segmap */
2332         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2333         __set_inuse(sbi, segno);
2334         *newseg = segno;
2335         spin_unlock(&free_i->segmap_lock);
2336 }
2337
2338 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2339 {
2340         struct curseg_info *curseg = CURSEG_I(sbi, type);
2341         struct summary_footer *sum_footer;
2342
2343         curseg->segno = curseg->next_segno;
2344         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2345         curseg->next_blkoff = 0;
2346         curseg->next_segno = NULL_SEGNO;
2347
2348         sum_footer = &(curseg->sum_blk->footer);
2349         memset(sum_footer, 0, sizeof(struct summary_footer));
2350         if (IS_DATASEG(type))
2351                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2352         if (IS_NODESEG(type))
2353                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2354         __set_sit_entry_type(sbi, type, curseg->segno, modified);
2355 }
2356
2357 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2358 {
2359         /* if segs_per_sec is large than 1, we need to keep original policy. */
2360         if (sbi->segs_per_sec != 1)
2361                 return CURSEG_I(sbi, type)->segno;
2362
2363         if (test_opt(sbi, NOHEAP) &&
2364                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2365                 return 0;
2366
2367         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2368                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2369
2370         /* find segments from 0 to reuse freed segments */
2371         if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2372                 return 0;
2373
2374         return CURSEG_I(sbi, type)->segno;
2375 }
2376
2377 /*
2378  * Allocate a current working segment.
2379  * This function always allocates a free segment in LFS manner.
2380  */
2381 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2382 {
2383         struct curseg_info *curseg = CURSEG_I(sbi, type);
2384         unsigned int segno = curseg->segno;
2385         int dir = ALLOC_LEFT;
2386
2387         write_sum_page(sbi, curseg->sum_blk,
2388                                 GET_SUM_BLOCK(sbi, segno));
2389         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2390                 dir = ALLOC_RIGHT;
2391
2392         if (test_opt(sbi, NOHEAP))
2393                 dir = ALLOC_RIGHT;
2394
2395         segno = __get_next_segno(sbi, type);
2396         get_new_segment(sbi, &segno, new_sec, dir);
2397         curseg->next_segno = segno;
2398         reset_curseg(sbi, type, 1);
2399         curseg->alloc_type = LFS;
2400 }
2401
2402 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2403                         struct curseg_info *seg, block_t start)
2404 {
2405         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2406         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2407         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2408         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2409         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2410         int i, pos;
2411
2412         for (i = 0; i < entries; i++)
2413                 target_map[i] = ckpt_map[i] | cur_map[i];
2414
2415         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2416
2417         seg->next_blkoff = pos;
2418 }
2419
2420 /*
2421  * If a segment is written by LFS manner, next block offset is just obtained
2422  * by increasing the current block offset. However, if a segment is written by
2423  * SSR manner, next block offset obtained by calling __next_free_blkoff
2424  */
2425 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2426                                 struct curseg_info *seg)
2427 {
2428         if (seg->alloc_type == SSR)
2429                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2430         else
2431                 seg->next_blkoff++;
2432 }
2433
2434 /*
2435  * This function always allocates a used segment(from dirty seglist) by SSR
2436  * manner, so it should recover the existing segment information of valid blocks
2437  */
2438 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2439 {
2440         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2441         struct curseg_info *curseg = CURSEG_I(sbi, type);
2442         unsigned int new_segno = curseg->next_segno;
2443         struct f2fs_summary_block *sum_node;
2444         struct page *sum_page;
2445
2446         write_sum_page(sbi, curseg->sum_blk,
2447                                 GET_SUM_BLOCK(sbi, curseg->segno));
2448         __set_test_and_inuse(sbi, new_segno);
2449
2450         mutex_lock(&dirty_i->seglist_lock);
2451         __remove_dirty_segment(sbi, new_segno, PRE);
2452         __remove_dirty_segment(sbi, new_segno, DIRTY);
2453         mutex_unlock(&dirty_i->seglist_lock);
2454
2455         reset_curseg(sbi, type, 1);
2456         curseg->alloc_type = SSR;
2457         __next_free_blkoff(sbi, curseg, 0);
2458
2459         sum_page = f2fs_get_sum_page(sbi, new_segno);
2460         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2461         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2462         f2fs_put_page(sum_page, 1);
2463 }
2464
2465 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2466 {
2467         struct curseg_info *curseg = CURSEG_I(sbi, type);
2468         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2469         unsigned segno = NULL_SEGNO;
2470         int i, cnt;
2471         bool reversed = false;
2472
2473         /* f2fs_need_SSR() already forces to do this */
2474         if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2475                 curseg->next_segno = segno;
2476                 return 1;
2477         }
2478
2479         /* For node segments, let's do SSR more intensively */
2480         if (IS_NODESEG(type)) {
2481                 if (type >= CURSEG_WARM_NODE) {
2482                         reversed = true;
2483                         i = CURSEG_COLD_NODE;
2484                 } else {
2485                         i = CURSEG_HOT_NODE;
2486                 }
2487                 cnt = NR_CURSEG_NODE_TYPE;
2488         } else {
2489                 if (type >= CURSEG_WARM_DATA) {
2490                         reversed = true;
2491                         i = CURSEG_COLD_DATA;
2492                 } else {
2493                         i = CURSEG_HOT_DATA;
2494                 }
2495                 cnt = NR_CURSEG_DATA_TYPE;
2496         }
2497
2498         for (; cnt-- > 0; reversed ? i-- : i++) {
2499                 if (i == type)
2500                         continue;
2501                 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2502                         curseg->next_segno = segno;
2503                         return 1;
2504                 }
2505         }
2506         return 0;
2507 }
2508
2509 /*
2510  * flush out current segment and replace it with new segment
2511  * This function should be returned with success, otherwise BUG
2512  */
2513 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2514                                                 int type, bool force)
2515 {
2516         struct curseg_info *curseg = CURSEG_I(sbi, type);
2517
2518         if (force)
2519                 new_curseg(sbi, type, true);
2520         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2521                                         type == CURSEG_WARM_NODE)
2522                 new_curseg(sbi, type, false);
2523         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2524                 new_curseg(sbi, type, false);
2525         else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2526                 change_curseg(sbi, type);
2527         else
2528                 new_curseg(sbi, type, false);
2529
2530         stat_inc_seg_type(sbi, curseg);
2531 }
2532
2533 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2534 {
2535         struct curseg_info *curseg;
2536         unsigned int old_segno;
2537         int i;
2538
2539         down_write(&SIT_I(sbi)->sentry_lock);
2540
2541         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2542                 curseg = CURSEG_I(sbi, i);
2543                 old_segno = curseg->segno;
2544                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2545                 locate_dirty_segment(sbi, old_segno);
2546         }
2547
2548         up_write(&SIT_I(sbi)->sentry_lock);
2549 }
2550
2551 static const struct segment_allocation default_salloc_ops = {
2552         .allocate_segment = allocate_segment_by_default,
2553 };
2554
2555 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2556                                                 struct cp_control *cpc)
2557 {
2558         __u64 trim_start = cpc->trim_start;
2559         bool has_candidate = false;
2560
2561         down_write(&SIT_I(sbi)->sentry_lock);
2562         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2563                 if (add_discard_addrs(sbi, cpc, true)) {
2564                         has_candidate = true;
2565                         break;
2566                 }
2567         }
2568         up_write(&SIT_I(sbi)->sentry_lock);
2569
2570         cpc->trim_start = trim_start;
2571         return has_candidate;
2572 }
2573
2574 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2575                                         struct discard_policy *dpolicy,
2576                                         unsigned int start, unsigned int end)
2577 {
2578         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2579         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2580         struct rb_node **insert_p = NULL, *insert_parent = NULL;
2581         struct discard_cmd *dc;
2582         struct blk_plug plug;
2583         int issued;
2584         unsigned int trimmed = 0;
2585
2586 next:
2587         issued = 0;
2588
2589         mutex_lock(&dcc->cmd_lock);
2590         if (unlikely(dcc->rbtree_check))
2591                 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2592                                                                 &dcc->root));
2593
2594         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2595                                         NULL, start,
2596                                         (struct rb_entry **)&prev_dc,
2597                                         (struct rb_entry **)&next_dc,
2598                                         &insert_p, &insert_parent, true);
2599         if (!dc)
2600                 dc = next_dc;
2601
2602         blk_start_plug(&plug);
2603
2604         while (dc && dc->lstart <= end) {
2605                 struct rb_node *node;
2606                 int err = 0;
2607
2608                 if (dc->len < dpolicy->granularity)
2609                         goto skip;
2610
2611                 if (dc->state != D_PREP) {
2612                         list_move_tail(&dc->list, &dcc->fstrim_list);
2613                         goto skip;
2614                 }
2615
2616                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2617
2618                 if (issued >= dpolicy->max_requests) {
2619                         start = dc->lstart + dc->len;
2620
2621                         if (err)
2622                                 __remove_discard_cmd(sbi, dc);
2623
2624                         blk_finish_plug(&plug);
2625                         mutex_unlock(&dcc->cmd_lock);
2626                         trimmed += __wait_all_discard_cmd(sbi, NULL);
2627                         congestion_wait(BLK_RW_ASYNC, HZ/50);
2628                         goto next;
2629                 }
2630 skip:
2631                 node = rb_next(&dc->rb_node);
2632                 if (err)
2633                         __remove_discard_cmd(sbi, dc);
2634                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2635
2636                 if (fatal_signal_pending(current))
2637                         break;
2638         }
2639
2640         blk_finish_plug(&plug);
2641         mutex_unlock(&dcc->cmd_lock);
2642
2643         return trimmed;
2644 }
2645
2646 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2647 {
2648         __u64 start = F2FS_BYTES_TO_BLK(range->start);
2649         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2650         unsigned int start_segno, end_segno;
2651         block_t start_block, end_block;
2652         struct cp_control cpc;
2653         struct discard_policy dpolicy;
2654         unsigned long long trimmed = 0;
2655         int err = 0;
2656         bool need_align = test_opt(sbi, LFS) && sbi->segs_per_sec > 1;
2657
2658         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2659                 return -EINVAL;
2660
2661         if (end < MAIN_BLKADDR(sbi))
2662                 goto out;
2663
2664         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2665                 f2fs_msg(sbi->sb, KERN_WARNING,
2666                         "Found FS corruption, run fsck to fix.");
2667                 return -EFSCORRUPTED;
2668         }
2669
2670         /* start/end segment number in main_area */
2671         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2672         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2673                                                 GET_SEGNO(sbi, end);
2674         if (need_align) {
2675                 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2676                 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2677         }
2678
2679         cpc.reason = CP_DISCARD;
2680         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2681         cpc.trim_start = start_segno;
2682         cpc.trim_end = end_segno;
2683
2684         if (sbi->discard_blks == 0)
2685                 goto out;
2686
2687         mutex_lock(&sbi->gc_mutex);
2688         err = f2fs_write_checkpoint(sbi, &cpc);
2689         mutex_unlock(&sbi->gc_mutex);
2690         if (err)
2691                 goto out;
2692
2693         /*
2694          * We filed discard candidates, but actually we don't need to wait for
2695          * all of them, since they'll be issued in idle time along with runtime
2696          * discard option. User configuration looks like using runtime discard
2697          * or periodic fstrim instead of it.
2698          */
2699         if (f2fs_realtime_discard_enable(sbi))
2700                 goto out;
2701
2702         start_block = START_BLOCK(sbi, start_segno);
2703         end_block = START_BLOCK(sbi, end_segno + 1);
2704
2705         __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2706         trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2707                                         start_block, end_block);
2708
2709         trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
2710                                         start_block, end_block);
2711 out:
2712         if (!err)
2713                 range->len = F2FS_BLK_TO_BYTES(trimmed);
2714         return err;
2715 }
2716
2717 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2718 {
2719         struct curseg_info *curseg = CURSEG_I(sbi, type);
2720         if (curseg->next_blkoff < sbi->blocks_per_seg)
2721                 return true;
2722         return false;
2723 }
2724
2725 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
2726 {
2727         switch (hint) {
2728         case WRITE_LIFE_SHORT:
2729                 return CURSEG_HOT_DATA;
2730         case WRITE_LIFE_EXTREME:
2731                 return CURSEG_COLD_DATA;
2732         default:
2733                 return CURSEG_WARM_DATA;
2734         }
2735 }
2736
2737 /* This returns write hints for each segment type. This hints will be
2738  * passed down to block layer. There are mapping tables which depend on
2739  * the mount option 'whint_mode'.
2740  *
2741  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2742  *
2743  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2744  *
2745  * User                  F2FS                     Block
2746  * ----                  ----                     -----
2747  *                       META                     WRITE_LIFE_NOT_SET
2748  *                       HOT_NODE                 "
2749  *                       WARM_NODE                "
2750  *                       COLD_NODE                "
2751  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2752  * extension list        "                        "
2753  *
2754  * -- buffered io
2755  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2756  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2757  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2758  * WRITE_LIFE_NONE       "                        "
2759  * WRITE_LIFE_MEDIUM     "                        "
2760  * WRITE_LIFE_LONG       "                        "
2761  *
2762  * -- direct io
2763  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2764  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2765  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2766  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2767  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2768  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2769  *
2770  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2771  *
2772  * User                  F2FS                     Block
2773  * ----                  ----                     -----
2774  *                       META                     WRITE_LIFE_MEDIUM;
2775  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
2776  *                       WARM_NODE                "
2777  *                       COLD_NODE                WRITE_LIFE_NONE
2778  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2779  * extension list        "                        "
2780  *
2781  * -- buffered io
2782  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2783  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2784  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
2785  * WRITE_LIFE_NONE       "                        "
2786  * WRITE_LIFE_MEDIUM     "                        "
2787  * WRITE_LIFE_LONG       "                        "
2788  *
2789  * -- direct io
2790  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2791  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2792  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2793  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2794  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2795  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2796  */
2797
2798 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2799                                 enum page_type type, enum temp_type temp)
2800 {
2801         if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2802                 if (type == DATA) {
2803                         if (temp == WARM)
2804                                 return WRITE_LIFE_NOT_SET;
2805                         else if (temp == HOT)
2806                                 return WRITE_LIFE_SHORT;
2807                         else if (temp == COLD)
2808                                 return WRITE_LIFE_EXTREME;
2809                 } else {
2810                         return WRITE_LIFE_NOT_SET;
2811                 }
2812         } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2813                 if (type == DATA) {
2814                         if (temp == WARM)
2815                                 return WRITE_LIFE_LONG;
2816                         else if (temp == HOT)
2817                                 return WRITE_LIFE_SHORT;
2818                         else if (temp == COLD)
2819                                 return WRITE_LIFE_EXTREME;
2820                 } else if (type == NODE) {
2821                         if (temp == WARM || temp == HOT)
2822                                 return WRITE_LIFE_NOT_SET;
2823                         else if (temp == COLD)
2824                                 return WRITE_LIFE_NONE;
2825                 } else if (type == META) {
2826                         return WRITE_LIFE_MEDIUM;
2827                 }
2828         }
2829         return WRITE_LIFE_NOT_SET;
2830 }
2831
2832 static int __get_segment_type_2(struct f2fs_io_info *fio)
2833 {
2834         if (fio->type == DATA)
2835                 return CURSEG_HOT_DATA;
2836         else
2837                 return CURSEG_HOT_NODE;
2838 }
2839
2840 static int __get_segment_type_4(struct f2fs_io_info *fio)
2841 {
2842         if (fio->type == DATA) {
2843                 struct inode *inode = fio->page->mapping->host;
2844
2845                 if (S_ISDIR(inode->i_mode))
2846                         return CURSEG_HOT_DATA;
2847                 else
2848                         return CURSEG_COLD_DATA;
2849         } else {
2850                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2851                         return CURSEG_WARM_NODE;
2852                 else
2853                         return CURSEG_COLD_NODE;
2854         }
2855 }
2856
2857 static int __get_segment_type_6(struct f2fs_io_info *fio)
2858 {
2859         if (fio->type == DATA) {
2860                 struct inode *inode = fio->page->mapping->host;
2861
2862                 if (is_cold_data(fio->page) || file_is_cold(inode))
2863                         return CURSEG_COLD_DATA;
2864                 if (file_is_hot(inode) ||
2865                                 is_inode_flag_set(inode, FI_HOT_DATA) ||
2866                                 f2fs_is_atomic_file(inode) ||
2867                                 f2fs_is_volatile_file(inode))
2868                         return CURSEG_HOT_DATA;
2869                 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
2870         } else {
2871                 if (IS_DNODE(fio->page))
2872                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2873                                                 CURSEG_HOT_NODE;
2874                 return CURSEG_COLD_NODE;
2875         }
2876 }
2877
2878 static int __get_segment_type(struct f2fs_io_info *fio)
2879 {
2880         int type = 0;
2881
2882         switch (F2FS_OPTION(fio->sbi).active_logs) {
2883         case 2:
2884                 type = __get_segment_type_2(fio);
2885                 break;
2886         case 4:
2887                 type = __get_segment_type_4(fio);
2888                 break;
2889         case 6:
2890                 type = __get_segment_type_6(fio);
2891                 break;
2892         default:
2893                 f2fs_bug_on(fio->sbi, true);
2894         }
2895
2896         if (IS_HOT(type))
2897                 fio->temp = HOT;
2898         else if (IS_WARM(type))
2899                 fio->temp = WARM;
2900         else
2901                 fio->temp = COLD;
2902         return type;
2903 }
2904
2905 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2906                 block_t old_blkaddr, block_t *new_blkaddr,
2907                 struct f2fs_summary *sum, int type,
2908                 struct f2fs_io_info *fio, bool add_list)
2909 {
2910         struct sit_info *sit_i = SIT_I(sbi);
2911         struct curseg_info *curseg = CURSEG_I(sbi, type);
2912
2913         down_read(&SM_I(sbi)->curseg_lock);
2914
2915         mutex_lock(&curseg->curseg_mutex);
2916         down_write(&sit_i->sentry_lock);
2917
2918         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2919
2920         f2fs_wait_discard_bio(sbi, *new_blkaddr);
2921
2922         /*
2923          * __add_sum_entry should be resided under the curseg_mutex
2924          * because, this function updates a summary entry in the
2925          * current summary block.
2926          */
2927         __add_sum_entry(sbi, type, sum);
2928
2929         __refresh_next_blkoff(sbi, curseg);
2930
2931         stat_inc_block_count(sbi, curseg);
2932
2933         /*
2934          * SIT information should be updated before segment allocation,
2935          * since SSR needs latest valid block information.
2936          */
2937         update_sit_entry(sbi, *new_blkaddr, 1);
2938         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2939                 update_sit_entry(sbi, old_blkaddr, -1);
2940
2941         if (!__has_curseg_space(sbi, type))
2942                 sit_i->s_ops->allocate_segment(sbi, type, false);
2943
2944         /*
2945          * segment dirty status should be updated after segment allocation,
2946          * so we just need to update status only one time after previous
2947          * segment being closed.
2948          */
2949         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2950         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
2951
2952         up_write(&sit_i->sentry_lock);
2953
2954         if (page && IS_NODESEG(type)) {
2955                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2956
2957                 f2fs_inode_chksum_set(sbi, page);
2958         }
2959
2960         if (add_list) {
2961                 struct f2fs_bio_info *io;
2962
2963                 INIT_LIST_HEAD(&fio->list);
2964                 fio->in_list = true;
2965                 fio->retry = false;
2966                 io = sbi->write_io[fio->type] + fio->temp;
2967                 spin_lock(&io->io_lock);
2968                 list_add_tail(&fio->list, &io->io_list);
2969                 spin_unlock(&io->io_lock);
2970         }
2971
2972         mutex_unlock(&curseg->curseg_mutex);
2973
2974         up_read(&SM_I(sbi)->curseg_lock);
2975 }
2976
2977 static void update_device_state(struct f2fs_io_info *fio)
2978 {
2979         struct f2fs_sb_info *sbi = fio->sbi;
2980         unsigned int devidx;
2981
2982         if (!f2fs_is_multi_device(sbi))
2983                 return;
2984
2985         devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2986
2987         /* update device state for fsync */
2988         f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
2989
2990         /* update device state for checkpoint */
2991         if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2992                 spin_lock(&sbi->dev_lock);
2993                 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2994                 spin_unlock(&sbi->dev_lock);
2995         }
2996 }
2997
2998 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2999 {
3000         int type = __get_segment_type(fio);
3001         bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
3002
3003         if (keep_order)
3004                 down_read(&fio->sbi->io_order_lock);
3005 reallocate:
3006         f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3007                         &fio->new_blkaddr, sum, type, fio, true);
3008         if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3009                 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3010                                         fio->old_blkaddr, fio->old_blkaddr);
3011
3012         /* writeout dirty page into bdev */
3013         f2fs_submit_page_write(fio);
3014         if (fio->retry) {
3015                 fio->old_blkaddr = fio->new_blkaddr;
3016                 goto reallocate;
3017         }
3018
3019         update_device_state(fio);
3020
3021         if (keep_order)
3022                 up_read(&fio->sbi->io_order_lock);
3023 }
3024
3025 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3026                                         enum iostat_type io_type)
3027 {
3028         struct f2fs_io_info fio = {
3029                 .sbi = sbi,
3030                 .type = META,
3031                 .temp = HOT,
3032                 .op = REQ_OP_WRITE,
3033                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3034                 .old_blkaddr = page->index,
3035                 .new_blkaddr = page->index,
3036                 .page = page,
3037                 .encrypted_page = NULL,
3038                 .in_list = false,
3039         };
3040
3041         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3042                 fio.op_flags &= ~REQ_META;
3043
3044         set_page_writeback(page);
3045         ClearPageError(page);
3046         f2fs_submit_page_write(&fio);
3047
3048         f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3049 }
3050
3051 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3052 {
3053         struct f2fs_summary sum;
3054
3055         set_summary(&sum, nid, 0, 0);
3056         do_write_page(&sum, fio);
3057
3058         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3059 }
3060
3061 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3062                                         struct f2fs_io_info *fio)
3063 {
3064         struct f2fs_sb_info *sbi = fio->sbi;
3065         struct f2fs_summary sum;
3066
3067         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3068         set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3069         do_write_page(&sum, fio);
3070         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3071
3072         f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3073 }
3074
3075 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3076 {
3077         int err;
3078         struct f2fs_sb_info *sbi = fio->sbi;
3079         unsigned int segno;
3080
3081         fio->new_blkaddr = fio->old_blkaddr;
3082         /* i/o temperature is needed for passing down write hints */
3083         __get_segment_type(fio);
3084
3085         segno = GET_SEGNO(sbi, fio->new_blkaddr);
3086
3087         if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3088                 set_sbi_flag(sbi, SBI_NEED_FSCK);
3089                 return -EFSCORRUPTED;
3090         }
3091
3092         stat_inc_inplace_blocks(fio->sbi);
3093
3094         err = f2fs_submit_page_bio(fio);
3095         if (!err)
3096                 update_device_state(fio);
3097
3098         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3099
3100         return err;
3101 }
3102
3103 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3104                                                 unsigned int segno)
3105 {
3106         int i;
3107
3108         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3109                 if (CURSEG_I(sbi, i)->segno == segno)
3110                         break;
3111         }
3112         return i;
3113 }
3114
3115 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3116                                 block_t old_blkaddr, block_t new_blkaddr,
3117                                 bool recover_curseg, bool recover_newaddr)
3118 {
3119         struct sit_info *sit_i = SIT_I(sbi);
3120         struct curseg_info *curseg;
3121         unsigned int segno, old_cursegno;
3122         struct seg_entry *se;
3123         int type;
3124         unsigned short old_blkoff;
3125
3126         segno = GET_SEGNO(sbi, new_blkaddr);
3127         se = get_seg_entry(sbi, segno);
3128         type = se->type;
3129
3130         down_write(&SM_I(sbi)->curseg_lock);
3131
3132         if (!recover_curseg) {
3133                 /* for recovery flow */
3134                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3135                         if (old_blkaddr == NULL_ADDR)
3136                                 type = CURSEG_COLD_DATA;
3137                         else
3138                                 type = CURSEG_WARM_DATA;
3139                 }
3140         } else {
3141                 if (IS_CURSEG(sbi, segno)) {
3142                         /* se->type is volatile as SSR allocation */
3143                         type = __f2fs_get_curseg(sbi, segno);
3144                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3145                 } else {
3146                         type = CURSEG_WARM_DATA;
3147                 }
3148         }
3149
3150         f2fs_bug_on(sbi, !IS_DATASEG(type));
3151         curseg = CURSEG_I(sbi, type);
3152
3153         mutex_lock(&curseg->curseg_mutex);
3154         down_write(&sit_i->sentry_lock);
3155
3156         old_cursegno = curseg->segno;
3157         old_blkoff = curseg->next_blkoff;
3158
3159         /* change the current segment */
3160         if (segno != curseg->segno) {
3161                 curseg->next_segno = segno;
3162                 change_curseg(sbi, type);
3163         }
3164
3165         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3166         __add_sum_entry(sbi, type, sum);
3167
3168         if (!recover_curseg || recover_newaddr)
3169                 update_sit_entry(sbi, new_blkaddr, 1);
3170         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3171                 invalidate_mapping_pages(META_MAPPING(sbi),
3172                                         old_blkaddr, old_blkaddr);
3173                 update_sit_entry(sbi, old_blkaddr, -1);
3174         }
3175
3176         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3177         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3178
3179         locate_dirty_segment(sbi, old_cursegno);
3180
3181         if (recover_curseg) {
3182                 if (old_cursegno != curseg->segno) {
3183                         curseg->next_segno = old_cursegno;
3184                         change_curseg(sbi, type);
3185                 }
3186                 curseg->next_blkoff = old_blkoff;
3187         }
3188
3189         up_write(&sit_i->sentry_lock);
3190         mutex_unlock(&curseg->curseg_mutex);
3191         up_write(&SM_I(sbi)->curseg_lock);
3192 }
3193
3194 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3195                                 block_t old_addr, block_t new_addr,
3196                                 unsigned char version, bool recover_curseg,
3197                                 bool recover_newaddr)
3198 {
3199         struct f2fs_summary sum;
3200
3201         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3202
3203         f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3204                                         recover_curseg, recover_newaddr);
3205
3206         f2fs_update_data_blkaddr(dn, new_addr);
3207 }
3208
3209 void f2fs_wait_on_page_writeback(struct page *page,
3210                                 enum page_type type, bool ordered)
3211 {
3212         if (PageWriteback(page)) {
3213                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3214
3215                 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
3216                                                 0, page->index, type);
3217                 if (ordered)
3218                         wait_on_page_writeback(page);
3219                 else
3220                         wait_for_stable_page(page);
3221         }
3222 }
3223
3224 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3225 {
3226         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3227         struct page *cpage;
3228
3229         if (!f2fs_post_read_required(inode))
3230                 return;
3231
3232         if (!is_valid_data_blkaddr(sbi, blkaddr))
3233                 return;
3234
3235         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3236         if (cpage) {
3237                 f2fs_wait_on_page_writeback(cpage, DATA, true);
3238                 f2fs_put_page(cpage, 1);
3239         }
3240 }
3241
3242 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3243 {
3244         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3245         struct curseg_info *seg_i;
3246         unsigned char *kaddr;
3247         struct page *page;
3248         block_t start;
3249         int i, j, offset;
3250
3251         start = start_sum_block(sbi);
3252
3253         page = f2fs_get_meta_page(sbi, start++);
3254         if (IS_ERR(page))
3255                 return PTR_ERR(page);
3256         kaddr = (unsigned char *)page_address(page);
3257
3258         /* Step 1: restore nat cache */
3259         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3260         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3261
3262         /* Step 2: restore sit cache */
3263         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3264         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3265         offset = 2 * SUM_JOURNAL_SIZE;
3266
3267         /* Step 3: restore summary entries */
3268         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3269                 unsigned short blk_off;
3270                 unsigned int segno;
3271
3272                 seg_i = CURSEG_I(sbi, i);
3273                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3274                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3275                 seg_i->next_segno = segno;
3276                 reset_curseg(sbi, i, 0);
3277                 seg_i->alloc_type = ckpt->alloc_type[i];
3278                 seg_i->next_blkoff = blk_off;
3279
3280                 if (seg_i->alloc_type == SSR)
3281                         blk_off = sbi->blocks_per_seg;
3282
3283                 for (j = 0; j < blk_off; j++) {
3284                         struct f2fs_summary *s;
3285                         s = (struct f2fs_summary *)(kaddr + offset);
3286                         seg_i->sum_blk->entries[j] = *s;
3287                         offset += SUMMARY_SIZE;
3288                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3289                                                 SUM_FOOTER_SIZE)
3290                                 continue;
3291
3292                         f2fs_put_page(page, 1);
3293                         page = NULL;
3294
3295                         page = f2fs_get_meta_page(sbi, start++);
3296                         if (IS_ERR(page))
3297                                 return PTR_ERR(page);
3298                         kaddr = (unsigned char *)page_address(page);
3299                         offset = 0;
3300                 }
3301         }
3302         f2fs_put_page(page, 1);
3303         return 0;
3304 }
3305
3306 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3307 {
3308         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3309         struct f2fs_summary_block *sum;
3310         struct curseg_info *curseg;
3311         struct page *new;
3312         unsigned short blk_off;
3313         unsigned int segno = 0;
3314         block_t blk_addr = 0;
3315         int err = 0;
3316
3317         /* get segment number and block addr */
3318         if (IS_DATASEG(type)) {
3319                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3320                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3321                                                         CURSEG_HOT_DATA]);
3322                 if (__exist_node_summaries(sbi))
3323                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3324                 else
3325                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3326         } else {
3327                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3328                                                         CURSEG_HOT_NODE]);
3329                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3330                                                         CURSEG_HOT_NODE]);
3331                 if (__exist_node_summaries(sbi))
3332                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3333                                                         type - CURSEG_HOT_NODE);
3334                 else
3335                         blk_addr = GET_SUM_BLOCK(sbi, segno);
3336         }
3337
3338         new = f2fs_get_meta_page(sbi, blk_addr);
3339         if (IS_ERR(new))
3340                 return PTR_ERR(new);
3341         sum = (struct f2fs_summary_block *)page_address(new);
3342
3343         if (IS_NODESEG(type)) {
3344                 if (__exist_node_summaries(sbi)) {
3345                         struct f2fs_summary *ns = &sum->entries[0];
3346                         int i;
3347                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3348                                 ns->version = 0;
3349                                 ns->ofs_in_node = 0;
3350                         }
3351                 } else {
3352                         err = f2fs_restore_node_summary(sbi, segno, sum);
3353                         if (err)
3354                                 goto out;
3355                 }
3356         }
3357
3358         /* set uncompleted segment to curseg */
3359         curseg = CURSEG_I(sbi, type);
3360         mutex_lock(&curseg->curseg_mutex);
3361
3362         /* update journal info */
3363         down_write(&curseg->journal_rwsem);
3364         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3365         up_write(&curseg->journal_rwsem);
3366
3367         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3368         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3369         curseg->next_segno = segno;
3370         reset_curseg(sbi, type, 0);
3371         curseg->alloc_type = ckpt->alloc_type[type];
3372         curseg->next_blkoff = blk_off;
3373         mutex_unlock(&curseg->curseg_mutex);
3374 out:
3375         f2fs_put_page(new, 1);
3376         return err;
3377 }
3378
3379 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3380 {
3381         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3382         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3383         int type = CURSEG_HOT_DATA;
3384         int err;
3385
3386         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3387                 int npages = f2fs_npages_for_summary_flush(sbi, true);
3388
3389                 if (npages >= 2)
3390                         f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3391                                                         META_CP, true);
3392
3393                 /* restore for compacted data summary */
3394                 err = read_compacted_summaries(sbi);
3395                 if (err)
3396                         return err;
3397                 type = CURSEG_HOT_NODE;
3398         }
3399
3400         if (__exist_node_summaries(sbi))
3401                 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3402                                         NR_CURSEG_TYPE - type, META_CP, true);
3403
3404         for (; type <= CURSEG_COLD_NODE; type++) {
3405                 err = read_normal_summaries(sbi, type);
3406                 if (err)
3407                         return err;
3408         }
3409
3410         /* sanity check for summary blocks */
3411         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3412                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3413                 return -EINVAL;
3414
3415         return 0;
3416 }
3417
3418 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3419 {
3420         struct page *page;
3421         unsigned char *kaddr;
3422         struct f2fs_summary *summary;
3423         struct curseg_info *seg_i;
3424         int written_size = 0;
3425         int i, j;
3426
3427         page = f2fs_grab_meta_page(sbi, blkaddr++);
3428         kaddr = (unsigned char *)page_address(page);
3429         memset(kaddr, 0, PAGE_SIZE);
3430
3431         /* Step 1: write nat cache */
3432         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3433         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3434         written_size += SUM_JOURNAL_SIZE;
3435
3436         /* Step 2: write sit cache */
3437         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3438         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3439         written_size += SUM_JOURNAL_SIZE;
3440
3441         /* Step 3: write summary entries */
3442         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3443                 unsigned short blkoff;
3444                 seg_i = CURSEG_I(sbi, i);
3445                 if (sbi->ckpt->alloc_type[i] == SSR)
3446                         blkoff = sbi->blocks_per_seg;
3447                 else
3448                         blkoff = curseg_blkoff(sbi, i);
3449
3450                 for (j = 0; j < blkoff; j++) {
3451                         if (!page) {
3452                                 page = f2fs_grab_meta_page(sbi, blkaddr++);
3453                                 kaddr = (unsigned char *)page_address(page);
3454                                 memset(kaddr, 0, PAGE_SIZE);
3455                                 written_size = 0;
3456                         }
3457                         summary = (struct f2fs_summary *)(kaddr + written_size);
3458                         *summary = seg_i->sum_blk->entries[j];
3459                         written_size += SUMMARY_SIZE;
3460
3461                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3462                                                         SUM_FOOTER_SIZE)
3463                                 continue;
3464
3465                         set_page_dirty(page);
3466                         f2fs_put_page(page, 1);
3467                         page = NULL;
3468                 }
3469         }
3470         if (page) {
3471                 set_page_dirty(page);
3472                 f2fs_put_page(page, 1);
3473         }
3474 }
3475
3476 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3477                                         block_t blkaddr, int type)
3478 {
3479         int i, end;
3480         if (IS_DATASEG(type))
3481                 end = type + NR_CURSEG_DATA_TYPE;
3482         else
3483                 end = type + NR_CURSEG_NODE_TYPE;
3484
3485         for (i = type; i < end; i++)
3486                 write_current_sum_page(sbi, i, blkaddr + (i - type));
3487 }
3488
3489 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3490 {
3491         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3492                 write_compacted_summaries(sbi, start_blk);
3493         else
3494                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3495 }
3496
3497 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3498 {
3499         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3500 }
3501
3502 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3503                                         unsigned int val, int alloc)
3504 {
3505         int i;
3506
3507         if (type == NAT_JOURNAL) {
3508                 for (i = 0; i < nats_in_cursum(journal); i++) {
3509                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3510                                 return i;
3511                 }
3512                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3513                         return update_nats_in_cursum(journal, 1);
3514         } else if (type == SIT_JOURNAL) {
3515                 for (i = 0; i < sits_in_cursum(journal); i++)
3516                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3517                                 return i;
3518                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3519                         return update_sits_in_cursum(journal, 1);
3520         }
3521         return -1;
3522 }
3523
3524 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3525                                         unsigned int segno)
3526 {
3527         return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
3528 }
3529
3530 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3531                                         unsigned int start)
3532 {
3533         struct sit_info *sit_i = SIT_I(sbi);
3534         struct page *page;
3535         pgoff_t src_off, dst_off;
3536
3537         src_off = current_sit_addr(sbi, start);
3538         dst_off = next_sit_addr(sbi, src_off);
3539
3540         page = f2fs_grab_meta_page(sbi, dst_off);
3541         seg_info_to_sit_page(sbi, page, start);
3542
3543         set_page_dirty(page);
3544         set_to_next_sit(sit_i, start);
3545
3546         return page;
3547 }
3548
3549 static struct sit_entry_set *grab_sit_entry_set(void)
3550 {
3551         struct sit_entry_set *ses =
3552                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3553
3554         ses->entry_cnt = 0;
3555         INIT_LIST_HEAD(&ses->set_list);
3556         return ses;
3557 }
3558
3559 static void release_sit_entry_set(struct sit_entry_set *ses)
3560 {
3561         list_del(&ses->set_list);
3562         kmem_cache_free(sit_entry_set_slab, ses);
3563 }
3564
3565 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3566                                                 struct list_head *head)
3567 {
3568         struct sit_entry_set *next = ses;
3569
3570         if (list_is_last(&ses->set_list, head))
3571                 return;
3572
3573         list_for_each_entry_continue(next, head, set_list)
3574                 if (ses->entry_cnt <= next->entry_cnt)
3575                         break;
3576
3577         list_move_tail(&ses->set_list, &next->set_list);
3578 }
3579
3580 static void add_sit_entry(unsigned int segno, struct list_head *head)
3581 {
3582         struct sit_entry_set *ses;
3583         unsigned int start_segno = START_SEGNO(segno);
3584
3585         list_for_each_entry(ses, head, set_list) {
3586                 if (ses->start_segno == start_segno) {
3587                         ses->entry_cnt++;
3588                         adjust_sit_entry_set(ses, head);
3589                         return;
3590                 }
3591         }
3592
3593         ses = grab_sit_entry_set();
3594
3595         ses->start_segno = start_segno;
3596         ses->entry_cnt++;
3597         list_add(&ses->set_list, head);
3598 }
3599
3600 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3601 {
3602         struct f2fs_sm_info *sm_info = SM_I(sbi);
3603         struct list_head *set_list = &sm_info->sit_entry_set;
3604         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3605         unsigned int segno;
3606
3607         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3608                 add_sit_entry(segno, set_list);
3609 }
3610
3611 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3612 {
3613         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3614         struct f2fs_journal *journal = curseg->journal;
3615         int i;
3616
3617         down_write(&curseg->journal_rwsem);
3618         for (i = 0; i < sits_in_cursum(journal); i++) {
3619                 unsigned int segno;
3620                 bool dirtied;
3621
3622                 segno = le32_to_cpu(segno_in_journal(journal, i));
3623                 dirtied = __mark_sit_entry_dirty(sbi, segno);
3624
3625                 if (!dirtied)
3626                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3627         }
3628         update_sits_in_cursum(journal, -i);
3629         up_write(&curseg->journal_rwsem);
3630 }
3631
3632 /*
3633  * CP calls this function, which flushes SIT entries including sit_journal,
3634  * and moves prefree segs to free segs.
3635  */
3636 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3637 {
3638         struct sit_info *sit_i = SIT_I(sbi);
3639         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3640         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3641         struct f2fs_journal *journal = curseg->journal;
3642         struct sit_entry_set *ses, *tmp;
3643         struct list_head *head = &SM_I(sbi)->sit_entry_set;
3644         bool to_journal = true;
3645         struct seg_entry *se;
3646
3647         down_write(&sit_i->sentry_lock);
3648
3649         if (!sit_i->dirty_sentries)
3650                 goto out;
3651
3652         /*
3653          * add and account sit entries of dirty bitmap in sit entry
3654          * set temporarily
3655          */
3656         add_sits_in_set(sbi);
3657
3658         /*
3659          * if there are no enough space in journal to store dirty sit
3660          * entries, remove all entries from journal and add and account
3661          * them in sit entry set.
3662          */
3663         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3664                 remove_sits_in_journal(sbi);
3665
3666         /*
3667          * there are two steps to flush sit entries:
3668          * #1, flush sit entries to journal in current cold data summary block.
3669          * #2, flush sit entries to sit page.
3670          */
3671         list_for_each_entry_safe(ses, tmp, head, set_list) {
3672                 struct page *page = NULL;
3673                 struct f2fs_sit_block *raw_sit = NULL;
3674                 unsigned int start_segno = ses->start_segno;
3675                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3676                                                 (unsigned long)MAIN_SEGS(sbi));
3677                 unsigned int segno = start_segno;
3678
3679                 if (to_journal &&
3680                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3681                         to_journal = false;
3682
3683                 if (to_journal) {
3684                         down_write(&curseg->journal_rwsem);
3685                 } else {
3686                         page = get_next_sit_page(sbi, start_segno);
3687                         raw_sit = page_address(page);
3688                 }
3689
3690                 /* flush dirty sit entries in region of current sit set */
3691                 for_each_set_bit_from(segno, bitmap, end) {
3692                         int offset, sit_offset;
3693
3694                         se = get_seg_entry(sbi, segno);
3695 #ifdef CONFIG_F2FS_CHECK_FS
3696                         if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3697                                                 SIT_VBLOCK_MAP_SIZE))
3698                                 f2fs_bug_on(sbi, 1);
3699 #endif
3700
3701                         /* add discard candidates */
3702                         if (!(cpc->reason & CP_DISCARD)) {
3703                                 cpc->trim_start = segno;
3704                                 add_discard_addrs(sbi, cpc, false);
3705                         }
3706
3707                         if (to_journal) {
3708                                 offset = f2fs_lookup_journal_in_cursum(journal,
3709                                                         SIT_JOURNAL, segno, 1);
3710                                 f2fs_bug_on(sbi, offset < 0);
3711                                 segno_in_journal(journal, offset) =
3712                                                         cpu_to_le32(segno);
3713                                 seg_info_to_raw_sit(se,
3714                                         &sit_in_journal(journal, offset));
3715                                 check_block_count(sbi, segno,
3716                                         &sit_in_journal(journal, offset));
3717                         } else {
3718                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3719                                 seg_info_to_raw_sit(se,
3720                                                 &raw_sit->entries[sit_offset]);
3721                                 check_block_count(sbi, segno,
3722                                                 &raw_sit->entries[sit_offset]);
3723                         }
3724
3725                         __clear_bit(segno, bitmap);
3726                         sit_i->dirty_sentries--;
3727                         ses->entry_cnt--;
3728                 }
3729
3730                 if (to_journal)
3731                         up_write(&curseg->journal_rwsem);
3732                 else
3733                         f2fs_put_page(page, 1);
3734
3735                 f2fs_bug_on(sbi, ses->entry_cnt);
3736                 release_sit_entry_set(ses);
3737         }
3738
3739         f2fs_bug_on(sbi, !list_empty(head));
3740         f2fs_bug_on(sbi, sit_i->dirty_sentries);
3741 out:
3742         if (cpc->reason & CP_DISCARD) {
3743                 __u64 trim_start = cpc->trim_start;
3744
3745                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3746                         add_discard_addrs(sbi, cpc, false);
3747
3748                 cpc->trim_start = trim_start;
3749         }
3750         up_write(&sit_i->sentry_lock);
3751
3752         set_prefree_as_free_segments(sbi);
3753 }
3754
3755 static int build_sit_info(struct f2fs_sb_info *sbi)
3756 {
3757         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3758         struct sit_info *sit_i;
3759         unsigned int sit_segs, start;
3760         char *src_bitmap;
3761         unsigned int bitmap_size;
3762
3763         /* allocate memory for SIT information */
3764         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3765         if (!sit_i)
3766                 return -ENOMEM;
3767
3768         SM_I(sbi)->sit_info = sit_i;
3769
3770         sit_i->sentries =
3771                 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3772                                               MAIN_SEGS(sbi)),
3773                               GFP_KERNEL);
3774         if (!sit_i->sentries)
3775                 return -ENOMEM;
3776
3777         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3778         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3779                                                                 GFP_KERNEL);
3780         if (!sit_i->dirty_sentries_bitmap)
3781                 return -ENOMEM;
3782
3783         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3784                 sit_i->sentries[start].cur_valid_map
3785                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3786                 sit_i->sentries[start].ckpt_valid_map
3787                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3788                 if (!sit_i->sentries[start].cur_valid_map ||
3789                                 !sit_i->sentries[start].ckpt_valid_map)
3790                         return -ENOMEM;
3791
3792 #ifdef CONFIG_F2FS_CHECK_FS
3793                 sit_i->sentries[start].cur_valid_map_mir
3794                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3795                 if (!sit_i->sentries[start].cur_valid_map_mir)
3796                         return -ENOMEM;
3797 #endif
3798
3799                 sit_i->sentries[start].discard_map
3800                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3801                                                         GFP_KERNEL);
3802                 if (!sit_i->sentries[start].discard_map)
3803                         return -ENOMEM;
3804         }
3805
3806         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3807         if (!sit_i->tmp_map)
3808                 return -ENOMEM;
3809
3810         if (sbi->segs_per_sec > 1) {
3811                 sit_i->sec_entries =
3812                         f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
3813                                                       MAIN_SECS(sbi)),
3814                                       GFP_KERNEL);
3815                 if (!sit_i->sec_entries)
3816                         return -ENOMEM;
3817         }
3818
3819         /* get information related with SIT */
3820         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3821
3822         /* setup SIT bitmap from ckeckpoint pack */
3823         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3824         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3825
3826         sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3827         if (!sit_i->sit_bitmap)
3828                 return -ENOMEM;
3829
3830 #ifdef CONFIG_F2FS_CHECK_FS
3831         sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3832         if (!sit_i->sit_bitmap_mir)
3833                 return -ENOMEM;
3834 #endif
3835
3836         /* init SIT information */
3837         sit_i->s_ops = &default_salloc_ops;
3838
3839         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3840         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
3841         sit_i->written_valid_blocks = 0;
3842         sit_i->bitmap_size = bitmap_size;
3843         sit_i->dirty_sentries = 0;
3844         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3845         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3846         sit_i->mounted_time = ktime_get_real_seconds();
3847         init_rwsem(&sit_i->sentry_lock);
3848         return 0;
3849 }
3850
3851 static int build_free_segmap(struct f2fs_sb_info *sbi)
3852 {
3853         struct free_segmap_info *free_i;
3854         unsigned int bitmap_size, sec_bitmap_size;
3855
3856         /* allocate memory for free segmap information */
3857         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
3858         if (!free_i)
3859                 return -ENOMEM;
3860
3861         SM_I(sbi)->free_info = free_i;
3862
3863         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3864         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
3865         if (!free_i->free_segmap)
3866                 return -ENOMEM;
3867
3868         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3869         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
3870         if (!free_i->free_secmap)
3871                 return -ENOMEM;
3872
3873         /* set all segments as dirty temporarily */
3874         memset(free_i->free_segmap, 0xff, bitmap_size);
3875         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3876
3877         /* init free segmap information */
3878         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
3879         free_i->free_segments = 0;
3880         free_i->free_sections = 0;
3881         spin_lock_init(&free_i->segmap_lock);
3882         return 0;
3883 }
3884
3885 static int build_curseg(struct f2fs_sb_info *sbi)
3886 {
3887         struct curseg_info *array;
3888         int i;
3889
3890         array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
3891                              GFP_KERNEL);
3892         if (!array)
3893                 return -ENOMEM;
3894
3895         SM_I(sbi)->curseg_array = array;
3896
3897         for (i = 0; i < NR_CURSEG_TYPE; i++) {
3898                 mutex_init(&array[i].curseg_mutex);
3899                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
3900                 if (!array[i].sum_blk)
3901                         return -ENOMEM;
3902                 init_rwsem(&array[i].journal_rwsem);
3903                 array[i].journal = f2fs_kzalloc(sbi,
3904                                 sizeof(struct f2fs_journal), GFP_KERNEL);
3905                 if (!array[i].journal)
3906                         return -ENOMEM;
3907                 array[i].segno = NULL_SEGNO;
3908                 array[i].next_blkoff = 0;
3909         }
3910         return restore_curseg_summaries(sbi);
3911 }
3912
3913 static int build_sit_entries(struct f2fs_sb_info *sbi)
3914 {
3915         struct sit_info *sit_i = SIT_I(sbi);
3916         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3917         struct f2fs_journal *journal = curseg->journal;
3918         struct seg_entry *se;
3919         struct f2fs_sit_entry sit;
3920         int sit_blk_cnt = SIT_BLK_CNT(sbi);
3921         unsigned int i, start, end;
3922         unsigned int readed, start_blk = 0;
3923         int err = 0;
3924         block_t total_node_blocks = 0;
3925
3926         do {
3927                 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3928                                                         META_SIT, true);
3929
3930                 start = start_blk * sit_i->sents_per_block;
3931                 end = (start_blk + readed) * sit_i->sents_per_block;
3932
3933                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
3934                         struct f2fs_sit_block *sit_blk;
3935                         struct page *page;
3936
3937                         se = &sit_i->sentries[start];
3938                         page = get_current_sit_page(sbi, start);
3939                         sit_blk = (struct f2fs_sit_block *)page_address(page);
3940                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3941                         f2fs_put_page(page, 1);
3942
3943                         err = check_block_count(sbi, start, &sit);
3944                         if (err)
3945                                 return err;
3946                         seg_info_from_raw_sit(se, &sit);
3947                         if (IS_NODESEG(se->type))
3948                                 total_node_blocks += se->valid_blocks;
3949
3950                         /* build discard map only one time */
3951                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3952                                 memset(se->discard_map, 0xff,
3953                                         SIT_VBLOCK_MAP_SIZE);
3954                         } else {
3955                                 memcpy(se->discard_map,
3956                                         se->cur_valid_map,
3957                                         SIT_VBLOCK_MAP_SIZE);
3958                                 sbi->discard_blks +=
3959                                         sbi->blocks_per_seg -
3960                                         se->valid_blocks;
3961                         }
3962
3963                         if (sbi->segs_per_sec > 1)
3964                                 get_sec_entry(sbi, start)->valid_blocks +=
3965                                                         se->valid_blocks;
3966                 }
3967                 start_blk += readed;
3968         } while (start_blk < sit_blk_cnt);
3969
3970         down_read(&curseg->journal_rwsem);
3971         for (i = 0; i < sits_in_cursum(journal); i++) {
3972                 unsigned int old_valid_blocks;
3973
3974                 start = le32_to_cpu(segno_in_journal(journal, i));
3975                 if (start >= MAIN_SEGS(sbi)) {
3976                         f2fs_msg(sbi->sb, KERN_ERR,
3977                                         "Wrong journal entry on segno %u",
3978                                         start);
3979                         set_sbi_flag(sbi, SBI_NEED_FSCK);
3980                         err = -EFSCORRUPTED;
3981                         break;
3982                 }
3983
3984                 se = &sit_i->sentries[start];
3985                 sit = sit_in_journal(journal, i);
3986
3987                 old_valid_blocks = se->valid_blocks;
3988                 if (IS_NODESEG(se->type))
3989                         total_node_blocks -= old_valid_blocks;
3990
3991                 err = check_block_count(sbi, start, &sit);
3992                 if (err)
3993                         break;
3994                 seg_info_from_raw_sit(se, &sit);
3995                 if (IS_NODESEG(se->type))
3996                         total_node_blocks += se->valid_blocks;
3997
3998                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3999                         memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4000                 } else {
4001                         memcpy(se->discard_map, se->cur_valid_map,
4002                                                 SIT_VBLOCK_MAP_SIZE);
4003                         sbi->discard_blks += old_valid_blocks;
4004                         sbi->discard_blks -= se->valid_blocks;
4005                 }
4006
4007                 if (sbi->segs_per_sec > 1) {
4008                         get_sec_entry(sbi, start)->valid_blocks +=
4009                                                         se->valid_blocks;
4010                         get_sec_entry(sbi, start)->valid_blocks -=
4011                                                         old_valid_blocks;
4012                 }
4013         }
4014         up_read(&curseg->journal_rwsem);
4015
4016         if (!err && total_node_blocks != valid_node_count(sbi)) {
4017                 f2fs_msg(sbi->sb, KERN_ERR,
4018                         "SIT is corrupted node# %u vs %u",
4019                         total_node_blocks, valid_node_count(sbi));
4020                 set_sbi_flag(sbi, SBI_NEED_FSCK);
4021                 err = -EFSCORRUPTED;
4022         }
4023
4024         return err;
4025 }
4026
4027 static void init_free_segmap(struct f2fs_sb_info *sbi)
4028 {
4029         unsigned int start;
4030         int type;
4031
4032         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4033                 struct seg_entry *sentry = get_seg_entry(sbi, start);
4034                 if (!sentry->valid_blocks)
4035                         __set_free(sbi, start);
4036                 else
4037                         SIT_I(sbi)->written_valid_blocks +=
4038                                                 sentry->valid_blocks;
4039         }
4040
4041         /* set use the current segments */
4042         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4043                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4044                 __set_test_and_inuse(sbi, curseg_t->segno);
4045         }
4046 }
4047
4048 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4049 {
4050         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4051         struct free_segmap_info *free_i = FREE_I(sbi);
4052         unsigned int segno = 0, offset = 0;
4053         unsigned short valid_blocks;
4054
4055         while (1) {
4056                 /* find dirty segment based on free segmap */
4057                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4058                 if (segno >= MAIN_SEGS(sbi))
4059                         break;
4060                 offset = segno + 1;
4061                 valid_blocks = get_valid_blocks(sbi, segno, false);
4062                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
4063                         continue;
4064                 if (valid_blocks > sbi->blocks_per_seg) {
4065                         f2fs_bug_on(sbi, 1);
4066                         continue;
4067                 }
4068                 mutex_lock(&dirty_i->seglist_lock);
4069                 __locate_dirty_segment(sbi, segno, DIRTY);
4070                 mutex_unlock(&dirty_i->seglist_lock);
4071         }
4072 }
4073
4074 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4075 {
4076         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4077         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4078
4079         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4080         if (!dirty_i->victim_secmap)
4081                 return -ENOMEM;
4082         return 0;
4083 }
4084
4085 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4086 {
4087         struct dirty_seglist_info *dirty_i;
4088         unsigned int bitmap_size, i;
4089
4090         /* allocate memory for dirty segments list information */
4091         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4092                                                                 GFP_KERNEL);
4093         if (!dirty_i)
4094                 return -ENOMEM;
4095
4096         SM_I(sbi)->dirty_info = dirty_i;
4097         mutex_init(&dirty_i->seglist_lock);
4098
4099         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4100
4101         for (i = 0; i < NR_DIRTY_TYPE; i++) {
4102                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4103                                                                 GFP_KERNEL);
4104                 if (!dirty_i->dirty_segmap[i])
4105                         return -ENOMEM;
4106         }
4107
4108         init_dirty_segmap(sbi);
4109         return init_victim_secmap(sbi);
4110 }
4111
4112 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4113 {
4114         int i;
4115
4116         /*
4117          * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4118          * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4119          */
4120         for (i = 0; i < NO_CHECK_TYPE; i++) {
4121                 struct curseg_info *curseg = CURSEG_I(sbi, i);
4122                 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4123                 unsigned int blkofs = curseg->next_blkoff;
4124
4125                 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4126                         goto out;
4127
4128                 if (curseg->alloc_type == SSR)
4129                         continue;
4130
4131                 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4132                         if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4133                                 continue;
4134 out:
4135                         f2fs_msg(sbi->sb, KERN_ERR,
4136                                 "Current segment's next free block offset is "
4137                                 "inconsistent with bitmap, logtype:%u, "
4138                                 "segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4139                                 i, curseg->segno, curseg->alloc_type,
4140                                 curseg->next_blkoff, blkofs);
4141                         return -EFSCORRUPTED;
4142                 }
4143         }
4144         return 0;
4145 }
4146
4147 /*
4148  * Update min, max modified time for cost-benefit GC algorithm
4149  */
4150 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4151 {
4152         struct sit_info *sit_i = SIT_I(sbi);
4153         unsigned int segno;
4154
4155         down_write(&sit_i->sentry_lock);
4156
4157         sit_i->min_mtime = ULLONG_MAX;
4158
4159         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4160                 unsigned int i;
4161                 unsigned long long mtime = 0;
4162
4163                 for (i = 0; i < sbi->segs_per_sec; i++)
4164                         mtime += get_seg_entry(sbi, segno + i)->mtime;
4165
4166                 mtime = div_u64(mtime, sbi->segs_per_sec);
4167
4168                 if (sit_i->min_mtime > mtime)
4169                         sit_i->min_mtime = mtime;
4170         }
4171         sit_i->max_mtime = get_mtime(sbi, false);
4172         up_write(&sit_i->sentry_lock);
4173 }
4174
4175 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
4176 {
4177         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4178         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4179         struct f2fs_sm_info *sm_info;
4180         int err;
4181
4182         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
4183         if (!sm_info)
4184                 return -ENOMEM;
4185
4186         /* init sm info */
4187         sbi->sm_info = sm_info;
4188         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4189         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4190         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4191         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4192         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4193         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4194         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
4195         sm_info->rec_prefree_segments = sm_info->main_segments *
4196                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
4197         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4198                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4199
4200         if (!test_opt(sbi, LFS))
4201                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
4202         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
4203         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
4204         sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
4205         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
4206         sm_info->min_ssr_sections = reserved_sections(sbi);
4207
4208         INIT_LIST_HEAD(&sm_info->sit_entry_set);
4209
4210         init_rwsem(&sm_info->curseg_lock);
4211
4212         if (!f2fs_readonly(sbi->sb)) {
4213                 err = f2fs_create_flush_cmd_control(sbi);
4214                 if (err)
4215                         return err;
4216         }
4217
4218         err = create_discard_cmd_control(sbi);
4219         if (err)
4220                 return err;
4221
4222         err = build_sit_info(sbi);
4223         if (err)
4224                 return err;
4225         err = build_free_segmap(sbi);
4226         if (err)
4227                 return err;
4228         err = build_curseg(sbi);
4229         if (err)
4230                 return err;
4231
4232         /* reinit free segmap based on SIT */
4233         err = build_sit_entries(sbi);
4234         if (err)
4235                 return err;
4236
4237         init_free_segmap(sbi);
4238         err = build_dirty_segmap(sbi);
4239         if (err)
4240                 return err;
4241
4242         err = sanity_check_curseg(sbi);
4243         if (err)
4244                 return err;
4245
4246         init_min_max_mtime(sbi);
4247         return 0;
4248 }
4249
4250 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4251                 enum dirty_type dirty_type)
4252 {
4253         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4254
4255         mutex_lock(&dirty_i->seglist_lock);
4256         kvfree(dirty_i->dirty_segmap[dirty_type]);
4257         dirty_i->nr_dirty[dirty_type] = 0;
4258         mutex_unlock(&dirty_i->seglist_lock);
4259 }
4260
4261 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
4262 {
4263         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4264         kvfree(dirty_i->victim_secmap);
4265 }
4266
4267 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4268 {
4269         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4270         int i;
4271
4272         if (!dirty_i)
4273                 return;
4274
4275         /* discard pre-free/dirty segments list */
4276         for (i = 0; i < NR_DIRTY_TYPE; i++)
4277                 discard_dirty_segmap(sbi, i);
4278
4279         destroy_victim_secmap(sbi);
4280         SM_I(sbi)->dirty_info = NULL;
4281         kfree(dirty_i);
4282 }
4283
4284 static void destroy_curseg(struct f2fs_sb_info *sbi)
4285 {
4286         struct curseg_info *array = SM_I(sbi)->curseg_array;
4287         int i;
4288
4289         if (!array)
4290                 return;
4291         SM_I(sbi)->curseg_array = NULL;
4292         for (i = 0; i < NR_CURSEG_TYPE; i++) {
4293                 kfree(array[i].sum_blk);
4294                 kfree(array[i].journal);
4295         }
4296         kfree(array);
4297 }
4298
4299 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4300 {
4301         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4302         if (!free_i)
4303                 return;
4304         SM_I(sbi)->free_info = NULL;
4305         kvfree(free_i->free_segmap);
4306         kvfree(free_i->free_secmap);
4307         kfree(free_i);
4308 }
4309
4310 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4311 {
4312         struct sit_info *sit_i = SIT_I(sbi);
4313         unsigned int start;
4314
4315         if (!sit_i)
4316                 return;
4317
4318         if (sit_i->sentries) {
4319                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4320                         kfree(sit_i->sentries[start].cur_valid_map);
4321 #ifdef CONFIG_F2FS_CHECK_FS
4322                         kfree(sit_i->sentries[start].cur_valid_map_mir);
4323 #endif
4324                         kfree(sit_i->sentries[start].ckpt_valid_map);
4325                         kfree(sit_i->sentries[start].discard_map);
4326                 }
4327         }
4328         kfree(sit_i->tmp_map);
4329
4330         kvfree(sit_i->sentries);
4331         kvfree(sit_i->sec_entries);
4332         kvfree(sit_i->dirty_sentries_bitmap);
4333
4334         SM_I(sbi)->sit_info = NULL;
4335         kfree(sit_i->sit_bitmap);
4336 #ifdef CONFIG_F2FS_CHECK_FS
4337         kfree(sit_i->sit_bitmap_mir);
4338 #endif
4339         kfree(sit_i);
4340 }
4341
4342 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
4343 {
4344         struct f2fs_sm_info *sm_info = SM_I(sbi);
4345
4346         if (!sm_info)
4347                 return;
4348         f2fs_destroy_flush_cmd_control(sbi, true);
4349         destroy_discard_cmd_control(sbi);
4350         destroy_dirty_segmap(sbi);
4351         destroy_curseg(sbi);
4352         destroy_free_segmap(sbi);
4353         destroy_sit_info(sbi);
4354         sbi->sm_info = NULL;
4355         kfree(sm_info);
4356 }
4357
4358 int __init f2fs_create_segment_manager_caches(void)
4359 {
4360         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4361                         sizeof(struct discard_entry));
4362         if (!discard_entry_slab)
4363                 goto fail;
4364
4365         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4366                         sizeof(struct discard_cmd));
4367         if (!discard_cmd_slab)
4368                 goto destroy_discard_entry;
4369
4370         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4371                         sizeof(struct sit_entry_set));
4372         if (!sit_entry_set_slab)
4373                 goto destroy_discard_cmd;
4374
4375         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4376                         sizeof(struct inmem_pages));
4377         if (!inmem_entry_slab)
4378                 goto destroy_sit_entry_set;
4379         return 0;
4380
4381 destroy_sit_entry_set:
4382         kmem_cache_destroy(sit_entry_set_slab);
4383 destroy_discard_cmd:
4384         kmem_cache_destroy(discard_cmd_slab);
4385 destroy_discard_entry:
4386         kmem_cache_destroy(discard_entry_slab);
4387 fail:
4388         return -ENOMEM;
4389 }
4390
4391 void f2fs_destroy_segment_manager_caches(void)
4392 {
4393         kmem_cache_destroy(sit_entry_set_slab);
4394         kmem_cache_destroy(discard_cmd_slab);
4395         kmem_cache_destroy(discard_entry_slab);
4396         kmem_cache_destroy(inmem_entry_slab);
4397 }