fs/f2fs/segment.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/f2fs/segment.c
   4  *
   5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6  *             http://www.samsung.com/
   7  */
   8 #include <linux/fs.h>
   9 #include <linux/f2fs_fs.h>
  10 #include <linux/bio.h>
  11 #include <linux/blkdev.h>
  12 #include <linux/prefetch.h>
  13 #include <linux/kthread.h>
  14 #include <linux/swap.h>
  15 #include <linux/timer.h>
  16 #include <linux/freezer.h>
  17 #include <linux/sched/signal.h>
  18
  19 #include "f2fs.h"
  20 #include "segment.h"
  21 #include "node.h"
  22 #include "gc.h"
  23 #include "trace.h"
  24 #include <trace/events/f2fs.h>
  25
  26 #define __reverse_ffz(x) __reverse_ffs(~(x))
  27
  28 static struct kmem_cache *discard_entry_slab;
  29 static struct kmem_cache *discard_cmd_slab;
  30 static struct kmem_cache *sit_entry_set_slab;
  31 static struct kmem_cache *inmem_entry_slab;
  32
  33 static unsigned long __reverse_ulong(unsigned char *str)
  34 {
  35         unsigned long tmp = 0;
  36         int shift = 24, idx = 0;
  37
  38 #if BITS_PER_LONG == 64
  39         shift = 56;
  40 #endif
  41         while (shift >= 0) {
  42                 tmp |= (unsigned long)str[idx++] << shift;
  43                 shift -= BITS_PER_BYTE;
  44         }
  45         return tmp;
  46 }
  47
  48 /*
  49  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  50  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  51  */
  52 static inline unsigned long __reverse_ffs(unsigned long word)
  53 {
  54         int num = 0;
  55
  56 #if BITS_PER_LONG == 64
  57         if ((word & 0xffffffff00000000UL) == 0)
  58                 num += 32;
  59         else
  60                 word >>= 32;
  61 #endif
  62         if ((word & 0xffff0000) == 0)
  63                 num += 16;
  64         else
  65                 word >>= 16;
  66
  67         if ((word & 0xff00) == 0)
  68                 num += 8;
  69         else
  70                 word >>= 8;
  71
  72         if ((word & 0xf0) == 0)
  73                 num += 4;
  74         else
  75                 word >>= 4;
  76
  77         if ((word & 0xc) == 0)
  78                 num += 2;
  79         else
  80                 word >>= 2;
  81
  82         if ((word & 0x2) == 0)
  83                 num += 1;
  84         return num;
  85 }
  86
  87 /*
  88  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  89  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  90  * @size must be integral times of unsigned long.
  91  * Example:
  92  *                             MSB <--> LSB
  93  *   f2fs_set_bit(0, bitmap) => 1000 0000
  94  *   f2fs_set_bit(7, bitmap) => 0000 0001
  95  */
  96 static unsigned long __find_rev_next_bit(const unsigned long *addr,
  97                         unsigned long size, unsigned long offset)
  98 {
  99         const unsigned long *p = addr + BIT_WORD(offset);
 100         unsigned long result = size;
 101         unsigned long tmp;
 102
 103         if (offset >= size)
 104                 return size;
 105
 106         size -= (offset & ~(BITS_PER_LONG - 1));
 107         offset %= BITS_PER_LONG;
 108
 109         while (1) {
 110                 if (*p == 0)
 111                         goto pass;
 112
 113                 tmp = __reverse_ulong((unsigned char *)p);
 114
 115                 tmp &= ~0UL >> offset;
 116                 if (size < BITS_PER_LONG)
 117                         tmp &= (~0UL << (BITS_PER_LONG - size));
 118                 if (tmp)
 119                         goto found;
 120 pass:
 121                 if (size <= BITS_PER_LONG)
 122                         break;
 123                 size -= BITS_PER_LONG;
 124                 offset = 0;
 125                 p++;
 126         }
 127         return result;
 128 found:
 129         return result - size + __reverse_ffs(tmp);
 130 }
 131
 132 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 133                         unsigned long size, unsigned long offset)
 134 {
 135         const unsigned long *p = addr + BIT_WORD(offset);
 136         unsigned long result = size;
 137         unsigned long tmp;
 138
 139         if (offset >= size)
 140                 return size;
 141
 142         size -= (offset & ~(BITS_PER_LONG - 1));
 143         offset %= BITS_PER_LONG;
 144
 145         while (1) {
 146                 if (*p == ~0UL)
 147                         goto pass;
 148
 149                 tmp = __reverse_ulong((unsigned char *)p);
 150
 151                 if (offset)
 152                         tmp |= ~0UL << (BITS_PER_LONG - offset);
 153                 if (size < BITS_PER_LONG)
 154                         tmp |= ~0UL >> size;
 155                 if (tmp != ~0UL)
 156                         goto found;
 157 pass:
 158                 if (size <= BITS_PER_LONG)
 159                         break;
 160                 size -= BITS_PER_LONG;
 161                 offset = 0;
 162                 p++;
 163         }
 164         return result;
 165 found:
 166         return result - size + __reverse_ffz(tmp);
 167 }
 168
 169 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
 170 {
 171         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 172         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 173         int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
 174
 175         if (test_opt(sbi, LFS))
 176                 return false;
 177         if (sbi->gc_mode == GC_URGENT)
 178                 return true;
 179         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
 180                 return true;
 181
 182         return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
 183                         SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
 184 }
 185
 186 void f2fs_register_inmem_page(struct inode *inode, struct page *page)
 187 {
 188         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 189         struct f2fs_inode_info *fi = F2FS_I(inode);
 190         struct inmem_pages *new;
 191
 192         f2fs_trace_pid(page);
 193
 194         f2fs_set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
 195
 196         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 197
 198         /* add atomic page indices to the list */
 199         new->page = page;
 200         INIT_LIST_HEAD(&new->list);
 201
 202         /* increase reference count with clean state */
 203         mutex_lock(&fi->inmem_lock);
 204         get_page(page);
 205         list_add_tail(&new->list, &fi->inmem_pages);
 206         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 207         if (list_empty(&fi->inmem_ilist))
 208                 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
 209         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 210         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 211         mutex_unlock(&fi->inmem_lock);
 212
 213         trace_f2fs_register_inmem_page(page, INMEM);
 214 }
 215
 216 static int __revoke_inmem_pages(struct inode *inode,
 217                                 struct list_head *head, bool drop, bool recover,
 218                                 bool trylock)
 219 {
 220         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 221         struct inmem_pages *cur, *tmp;
 222         int err = 0;
 223
 224         list_for_each_entry_safe(cur, tmp, head, list) {
 225                 struct page *page = cur->page;
 226
 227                 if (drop)
 228                         trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 229
 230                 if (trylock) {
 231                         /*
 232                          * to avoid deadlock in between page lock and
 233                          * inmem_lock.
 234                          */
 235                         if (!trylock_page(page))
 236                                 continue;
 237                 } else {
 238                         lock_page(page);
 239                 }
 240
 241                 f2fs_wait_on_page_writeback(page, DATA, true, true);
 242
 243                 if (recover) {
 244                         struct dnode_of_data dn;
 245                         struct node_info ni;
 246
 247                         trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 248 retry:
 249                         set_new_dnode(&dn, inode, NULL, NULL, 0);
 250                         err = f2fs_get_dnode_of_data(&dn, page->index,
 251                                                                 LOOKUP_NODE);
 252                         if (err) {
 253                                 if (err == -ENOMEM) {
 254                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 255                                         cond_resched();
 256                                         goto retry;
 257                                 }
 258                                 err = -EAGAIN;
 259                                 goto next;
 260                         }
 261
 262                         err = f2fs_get_node_info(sbi, dn.nid, &ni);
 263                         if (err) {
 264                                 f2fs_put_dnode(&dn);
 265                                 return err;
 266                         }
 267
 268                         if (cur->old_addr == NEW_ADDR) {
 269                                 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
 270                                 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
 271                         } else
 272                                 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 273                                         cur->old_addr, ni.version, true, true);
 274                         f2fs_put_dnode(&dn);
 275                 }
 276 next:
 277                 /* we don't need to invalidate this in the sccessful status */
 278                 if (drop || recover) {
 279                         ClearPageUptodate(page);
 280                         clear_cold_data(page);
 281                 }
 282                 f2fs_clear_page_private(page);
 283                 f2fs_put_page(page, 1);
 284
 285                 list_del(&cur->list);
 286                 kmem_cache_free(inmem_entry_slab, cur);
 287                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 288         }
 289         return err;
 290 }
 291
 292 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
 293 {
 294         struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
 295         struct inode *inode;
 296         struct f2fs_inode_info *fi;
 297 next:
 298         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 299         if (list_empty(head)) {
 300                 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 301                 return;
 302         }
 303         fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
 304         inode = igrab(&fi->vfs_inode);
 305         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 306
 307         if (inode) {
 308                 if (gc_failure) {
 309                         if (fi->i_gc_failures[GC_FAILURE_ATOMIC])
 310                                 goto drop;
 311                         goto skip;
 312                 }
 313 drop:
 314                 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
 315                 f2fs_drop_inmem_pages(inode);
 316                 iput(inode);
 317         }
 318 skip:
 319         congestion_wait(BLK_RW_ASYNC, HZ/50);
 320         cond_resched();
 321         goto next;
 322 }
 323
 324 void f2fs_drop_inmem_pages(struct inode *inode)
 325 {
 326         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 327         struct f2fs_inode_info *fi = F2FS_I(inode);
 328
 329         while (!list_empty(&fi->inmem_pages)) {
 330                 mutex_lock(&fi->inmem_lock);
 331                 __revoke_inmem_pages(inode, &fi->inmem_pages,
 332                                                 true, false, true);
 333
 334                 if (list_empty(&fi->inmem_pages)) {
 335                         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 336                         if (!list_empty(&fi->inmem_ilist))
 337                                 list_del_init(&fi->inmem_ilist);
 338                         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 339                 }
 340                 mutex_unlock(&fi->inmem_lock);
 341         }
 342
 343         clear_inode_flag(inode, FI_ATOMIC_FILE);
 344         fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
 345         stat_dec_atomic_write(inode);
 346 }
 347
 348 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
 349 {
 350         struct f2fs_inode_info *fi = F2FS_I(inode);
 351         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 352         struct list_head *head = &fi->inmem_pages;
 353         struct inmem_pages *cur = NULL;
 354
 355         f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
 356
 357         mutex_lock(&fi->inmem_lock);
 358         list_for_each_entry(cur, head, list) {
 359                 if (cur->page == page)
 360                         break;
 361         }
 362
 363         f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
 364         list_del(&cur->list);
 365         mutex_unlock(&fi->inmem_lock);
 366
 367         dec_page_count(sbi, F2FS_INMEM_PAGES);
 368         kmem_cache_free(inmem_entry_slab, cur);
 369
 370         ClearPageUptodate(page);
 371         f2fs_clear_page_private(page);
 372         f2fs_put_page(page, 0);
 373
 374         trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 375 }
 376
 377 static int __f2fs_commit_inmem_pages(struct inode *inode)
 378 {
 379         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 380         struct f2fs_inode_info *fi = F2FS_I(inode);
 381         struct inmem_pages *cur, *tmp;
 382         struct f2fs_io_info fio = {
 383                 .sbi = sbi,
 384                 .ino = inode->i_ino,
 385                 .type = DATA,
 386                 .op = REQ_OP_WRITE,
 387                 .op_flags = REQ_SYNC | REQ_PRIO,
 388                 .io_type = FS_DATA_IO,
 389         };
 390         struct list_head revoke_list;
 391         bool submit_bio = false;
 392         int err = 0;
 393
 394         INIT_LIST_HEAD(&revoke_list);
 395
 396         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 397                 struct page *page = cur->page;
 398
 399                 lock_page(page);
 400                 if (page->mapping == inode->i_mapping) {
 401                         trace_f2fs_commit_inmem_page(page, INMEM);
 402
 403                         f2fs_wait_on_page_writeback(page, DATA, true, true);
 404
 405                         set_page_dirty(page);
 406                         if (clear_page_dirty_for_io(page)) {
 407                                 inode_dec_dirty_pages(inode);
 408                                 f2fs_remove_dirty_inode(inode);
 409                         }
 410 retry:
 411                         fio.page = page;
 412                         fio.old_blkaddr = NULL_ADDR;
 413                         fio.encrypted_page = NULL;
 414                         fio.need_lock = LOCK_DONE;
 415                         err = f2fs_do_write_data_page(&fio);
 416                         if (err) {
 417                                 if (err == -ENOMEM) {
 418                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 419                                         cond_resched();
 420                                         goto retry;
 421                                 }
 422                                 unlock_page(page);
 423                                 break;
 424                         }
 425                         /* record old blkaddr for revoking */
 426                         cur->old_addr = fio.old_blkaddr;
 427                         submit_bio = true;
 428                 }
 429                 unlock_page(page);
 430                 list_move_tail(&cur->list, &revoke_list);
 431         }
 432
 433         if (submit_bio)
 434                 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
 435
 436         if (err) {
 437                 /*
 438                  * try to revoke all committed pages, but still we could fail
 439                  * due to no memory or other reason, if that happened, EAGAIN
 440                  * will be returned, which means in such case, transaction is
 441                  * already not integrity, caller should use journal to do the
 442                  * recovery or rewrite & commit last transaction. For other
 443                  * error number, revoking was done by filesystem itself.
 444                  */
 445                 err = __revoke_inmem_pages(inode, &revoke_list,
 446                                                 false, true, false);
 447
 448                 /* drop all uncommitted pages */
 449                 __revoke_inmem_pages(inode, &fi->inmem_pages,
 450                                                 true, false, false);
 451         } else {
 452                 __revoke_inmem_pages(inode, &revoke_list,
 453                                                 false, false, false);
 454         }
 455
 456         return err;
 457 }
 458
 459 int f2fs_commit_inmem_pages(struct inode *inode)
 460 {
 461         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 462         struct f2fs_inode_info *fi = F2FS_I(inode);
 463         int err;
 464
 465         f2fs_balance_fs(sbi, true);
 466
 467         down_write(&fi->i_gc_rwsem[WRITE]);
 468
 469         f2fs_lock_op(sbi);
 470         set_inode_flag(inode, FI_ATOMIC_COMMIT);
 471
 472         mutex_lock(&fi->inmem_lock);
 473         err = __f2fs_commit_inmem_pages(inode);
 474
 475         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 476         if (!list_empty(&fi->inmem_ilist))
 477                 list_del_init(&fi->inmem_ilist);
 478         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 479         mutex_unlock(&fi->inmem_lock);
 480
 481         clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 482
 483         f2fs_unlock_op(sbi);
 484         up_write(&fi->i_gc_rwsem[WRITE]);
 485
 486         return err;
 487 }
 488
 489 /*
 490  * This function balances dirty node and dentry pages.
 491  * In addition, it controls garbage collection.
 492  */
 493 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 494 {
 495         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 496                 f2fs_show_injection_info(FAULT_CHECKPOINT);
 497                 f2fs_stop_checkpoint(sbi, false);
 498         }
 499
 500         /* balance_fs_bg is able to be pending */
 501         if (need && excess_cached_nats(sbi))
 502                 f2fs_balance_fs_bg(sbi);
 503
 504         if (f2fs_is_checkpoint_ready(sbi))
 505                 return;
 506
 507         /*
 508          * We should do GC or end up with checkpoint, if there are so many dirty
 509          * dir/node pages without enough free segments.
 510          */
 511         if (has_not_enough_free_secs(sbi, 0, 0)) {
 512                 mutex_lock(&sbi->gc_mutex);
 513                 f2fs_gc(sbi, false, false, NULL_SEGNO);
 514         }
 515 }
 516
 517 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 518 {
 519         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 520                 return;
 521
 522         /* try to shrink extent cache when there is no enough memory */
 523         if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
 524                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 525
 526         /* check the # of cached NAT entries */
 527         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
 528                 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 529
 530         if (!f2fs_available_free_memory(sbi, FREE_NIDS))
 531                 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
 532         else
 533                 f2fs_build_free_nids(sbi, false, false);
 534
 535         if (!is_idle(sbi, REQ_TIME) &&
 536                 (!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
 537                 return;
 538
 539         /* checkpoint is the only way to shrink partial cached entries */
 540         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
 541                         !f2fs_available_free_memory(sbi, INO_ENTRIES) ||
 542                         excess_prefree_segs(sbi) ||
 543                         excess_dirty_nats(sbi) ||
 544                         excess_dirty_nodes(sbi) ||
 545                         f2fs_time_over(sbi, CP_TIME)) {
 546                 if (test_opt(sbi, DATA_FLUSH)) {
 547                         struct blk_plug plug;
 548
 549                         blk_start_plug(&plug);
 550                         f2fs_sync_dirty_inodes(sbi, FILE_INODE);
 551                         blk_finish_plug(&plug);
 552                 }
 553                 f2fs_sync_fs(sbi->sb, true);
 554                 stat_inc_bg_cp_count(sbi->stat_info);
 555         }
 556 }
 557
 558 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 559                                 struct block_device *bdev)
 560 {
 561         struct bio *bio;
 562         int ret;
 563
 564         bio = f2fs_bio_alloc(sbi, 0, false);
 565         if (!bio)
 566                 return -ENOMEM;
 567
 568         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
 569         bio_set_dev(bio, bdev);
 570         ret = submit_bio_wait(bio);
 571         bio_put(bio);
 572
 573         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 574                                 test_opt(sbi, FLUSH_MERGE), ret);
 575         return ret;
 576 }
 577
 578 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 579 {
 580         int ret = 0;
 581         int i;
 582
 583         if (!sbi->s_ndevs)
 584                 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 585
 586         for (i = 0; i < sbi->s_ndevs; i++) {
 587                 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
 588                         continue;
 589                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 590                 if (ret)
 591                         break;
 592         }
 593         return ret;
 594 }
 595
 596 static int issue_flush_thread(void *data)
 597 {
 598         struct f2fs_sb_info *sbi = data;
 599         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 600         wait_queue_head_t *q = &fcc->flush_wait_queue;
 601 repeat:
 602         if (kthread_should_stop())
 603                 return 0;
 604
 605         sb_start_intwrite(sbi->sb);
 606
 607         if (!llist_empty(&fcc->issue_list)) {
 608                 struct flush_cmd *cmd, *next;
 609                 int ret;
 610
 611                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 612                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 613
 614                 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 615
 616                 ret = submit_flush_wait(sbi, cmd->ino);
 617                 atomic_inc(&fcc->issued_flush);
 618
 619                 llist_for_each_entry_safe(cmd, next,
 620                                           fcc->dispatch_list, llnode) {
 621                         cmd->ret = ret;
 622                         complete(&cmd->wait);
 623                 }
 624                 fcc->dispatch_list = NULL;
 625         }
 626
 627         sb_end_intwrite(sbi->sb);
 628
 629         wait_event_interruptible(*q,
 630                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 631         goto repeat;
 632 }
 633
 634 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 635 {
 636         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 637         struct flush_cmd cmd;
 638         int ret;
 639
 640         if (test_opt(sbi, NOBARRIER))
 641                 return 0;
 642
 643         if (!test_opt(sbi, FLUSH_MERGE)) {
 644                 atomic_inc(&fcc->queued_flush);
 645                 ret = submit_flush_wait(sbi, ino);
 646                 atomic_dec(&fcc->queued_flush);
 647                 atomic_inc(&fcc->issued_flush);
 648                 return ret;
 649         }
 650
 651         if (atomic_inc_return(&fcc->queued_flush) == 1 || sbi->s_ndevs > 1) {
 652                 ret = submit_flush_wait(sbi, ino);
 653                 atomic_dec(&fcc->queued_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->queued_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->queued_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->queued_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->queued_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                 kvfree(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                 kvfree(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 (!sbi->s_ndevs)
 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, ckpt_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         ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno);
 838
 839         if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
 840                                 ckpt_valid_blocks == sbi->blocks_per_seg)) {
 841                 __locate_dirty_segment(sbi, segno, PRE);
 842                 __remove_dirty_segment(sbi, segno, DIRTY);
 843         } else if (valid_blocks < sbi->blocks_per_seg) {
 844                 __locate_dirty_segment(sbi, segno, DIRTY);
 845         } else {
 846                 /* Recovery routine with SSR needs this */
 847                 __remove_dirty_segment(sbi, segno, DIRTY);
 848         }
 849
 850         mutex_unlock(&dirty_i->seglist_lock);
 851 }
 852
 853 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
 854 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
 855 {
 856         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 857         unsigned int segno;
 858
 859         mutex_lock(&dirty_i->seglist_lock);
 860         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 861                 if (get_valid_blocks(sbi, segno, false))
 862                         continue;
 863                 if (IS_CURSEG(sbi, segno))
 864                         continue;
 865                 __locate_dirty_segment(sbi, segno, PRE);
 866                 __remove_dirty_segment(sbi, segno, DIRTY);
 867         }
 868         mutex_unlock(&dirty_i->seglist_lock);
 869 }
 870
 871 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi)
 872 {
 873         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 874         block_t ovp = overprovision_segments(sbi) << sbi->log_blocks_per_seg;
 875         block_t holes[2] = {0, 0};      /* DATA and NODE */
 876         struct seg_entry *se;
 877         unsigned int segno;
 878
 879         mutex_lock(&dirty_i->seglist_lock);
 880         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 881                 se = get_seg_entry(sbi, segno);
 882                 if (IS_NODESEG(se->type))
 883                         holes[NODE] += sbi->blocks_per_seg - se->valid_blocks;
 884                 else
 885                         holes[DATA] += sbi->blocks_per_seg - se->valid_blocks;
 886         }
 887         mutex_unlock(&dirty_i->seglist_lock);
 888
 889         if (holes[DATA] > ovp || holes[NODE] > ovp)
 890                 return -EAGAIN;
 891         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
 892                 dirty_segments(sbi) > overprovision_segments(sbi))
 893                 return -EAGAIN;
 894         return 0;
 895 }
 896
 897 /* This is only used by SBI_CP_DISABLED */
 898 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
 899 {
 900         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 901         unsigned int segno = 0;
 902
 903         mutex_lock(&dirty_i->seglist_lock);
 904         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 905                 if (get_valid_blocks(sbi, segno, false))
 906                         continue;
 907                 if (get_ckpt_valid_blocks(sbi, segno))
 908                         continue;
 909                 mutex_unlock(&dirty_i->seglist_lock);
 910                 return segno;
 911         }
 912         mutex_unlock(&dirty_i->seglist_lock);
 913         return NULL_SEGNO;
 914 }
 915
 916 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 917                 struct block_device *bdev, block_t lstart,
 918                 block_t start, block_t len)
 919 {
 920         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 921         struct list_head *pend_list;
 922         struct discard_cmd *dc;
 923
 924         f2fs_bug_on(sbi, !len);
 925
 926         pend_list = &dcc->pend_list[plist_idx(len)];
 927
 928         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 929         INIT_LIST_HEAD(&dc->list);
 930         dc->bdev = bdev;
 931         dc->lstart = lstart;
 932         dc->start = start;
 933         dc->len = len;
 934         dc->ref = 0;
 935         dc->state = D_PREP;
 936         dc->queued = 0;
 937         dc->error = 0;
 938         init_completion(&dc->wait);
 939         list_add_tail(&dc->list, pend_list);
 940         spin_lock_init(&dc->lock);
 941         dc->bio_ref = 0;
 942         atomic_inc(&dcc->discard_cmd_cnt);
 943         dcc->undiscard_blks += len;
 944
 945         return dc;
 946 }
 947
 948 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 949                                 struct block_device *bdev, block_t lstart,
 950                                 block_t start, block_t len,
 951                                 struct rb_node *parent, struct rb_node **p,
 952                                 bool leftmost)
 953 {
 954         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 955         struct discard_cmd *dc;
 956
 957         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
 958
 959         rb_link_node(&dc->rb_node, parent, p);
 960         rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
 961
 962         return dc;
 963 }
 964
 965 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
 966                                                         struct discard_cmd *dc)
 967 {
 968         if (dc->state == D_DONE)
 969                 atomic_sub(dc->queued, &dcc->queued_discard);
 970
 971         list_del(&dc->list);
 972         rb_erase_cached(&dc->rb_node, &dcc->root);
 973         dcc->undiscard_blks -= dc->len;
 974
 975         kmem_cache_free(discard_cmd_slab, dc);
 976
 977         atomic_dec(&dcc->discard_cmd_cnt);
 978 }
 979
 980 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
 981                                                         struct discard_cmd *dc)
 982 {
 983         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 984         unsigned long flags;
 985
 986         trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
 987
 988         spin_lock_irqsave(&dc->lock, flags);
 989         if (dc->bio_ref) {
 990                 spin_unlock_irqrestore(&dc->lock, flags);
 991                 return;
 992         }
 993         spin_unlock_irqrestore(&dc->lock, flags);
 994
 995         f2fs_bug_on(sbi, dc->ref);
 996
 997         if (dc->error == -EOPNOTSUPP)
 998                 dc->error = 0;
 999
1000         if (dc->error)
1001                 printk_ratelimited(
1002                         "%sF2FS-fs: Issue discard(%u, %u, %u) failed, ret: %d",
1003                         KERN_INFO, dc->lstart, dc->start, dc->len, dc->error);
1004         __detach_discard_cmd(dcc, dc);
1005 }
1006
1007 static void f2fs_submit_discard_endio(struct bio *bio)
1008 {
1009         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1010         unsigned long flags;
1011
1012         dc->error = blk_status_to_errno(bio->bi_status);
1013
1014         spin_lock_irqsave(&dc->lock, flags);
1015         dc->bio_ref--;
1016         if (!dc->bio_ref && dc->state == D_SUBMIT) {
1017                 dc->state = D_DONE;
1018                 complete_all(&dc->wait);
1019         }
1020         spin_unlock_irqrestore(&dc->lock, flags);
1021         bio_put(bio);
1022 }
1023
1024 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1025                                 block_t start, block_t end)
1026 {
1027 #ifdef CONFIG_F2FS_CHECK_FS
1028         struct seg_entry *sentry;
1029         unsigned int segno;
1030         block_t blk = start;
1031         unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1032         unsigned long *map;
1033
1034         while (blk < end) {
1035                 segno = GET_SEGNO(sbi, blk);
1036                 sentry = get_seg_entry(sbi, segno);
1037                 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1038
1039                 if (end < START_BLOCK(sbi, segno + 1))
1040                         size = GET_BLKOFF_FROM_SEG0(sbi, end);
1041                 else
1042                         size = max_blocks;
1043                 map = (unsigned long *)(sentry->cur_valid_map);
1044                 offset = __find_rev_next_bit(map, size, offset);
1045                 f2fs_bug_on(sbi, offset != size);
1046                 blk = START_BLOCK(sbi, segno + 1);
1047         }
1048 #endif
1049 }
1050
1051 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1052                                 struct discard_policy *dpolicy,
1053                                 int discard_type, unsigned int granularity)
1054 {
1055         /* common policy */
1056         dpolicy->type = discard_type;
1057         dpolicy->sync = true;
1058         dpolicy->ordered = false;
1059         dpolicy->granularity = granularity;
1060
1061         dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1062         dpolicy->io_aware_gran = MAX_PLIST_NUM;
1063         dpolicy->timeout = 0;
1064
1065         if (discard_type == DPOLICY_BG) {
1066                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1067                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1068                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1069                 dpolicy->io_aware = true;
1070                 dpolicy->sync = false;
1071                 dpolicy->ordered = true;
1072                 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1073                         dpolicy->granularity = 1;
1074                         dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1075                 }
1076         } else if (discard_type == DPOLICY_FORCE) {
1077                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1078                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1079                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1080                 dpolicy->io_aware = false;
1081         } else if (discard_type == DPOLICY_FSTRIM) {
1082                 dpolicy->io_aware = false;
1083         } else if (discard_type == DPOLICY_UMOUNT) {
1084                 dpolicy->max_requests = UINT_MAX;
1085                 dpolicy->io_aware = false;
1086                 /* we need to issue all to keep CP_TRIMMED_FLAG */
1087                 dpolicy->granularity = 1;
1088         }
1089 }
1090
1091 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1092                                 struct block_device *bdev, block_t lstart,
1093                                 block_t start, block_t len);
1094 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1095 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1096                                                 struct discard_policy *dpolicy,
1097                                                 struct discard_cmd *dc,
1098                                                 unsigned int *issued)
1099 {
1100         struct block_device *bdev = dc->bdev;
1101         struct request_queue *q = bdev_get_queue(bdev);
1102         unsigned int max_discard_blocks =
1103                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1104         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1105         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1106                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1107         int flag = dpolicy->sync ? REQ_SYNC : 0;
1108         block_t lstart, start, len, total_len;
1109         int err = 0;
1110
1111         if (dc->state != D_PREP)
1112                 return 0;
1113
1114         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1115                 return 0;
1116
1117         trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1118
1119         lstart = dc->lstart;
1120         start = dc->start;
1121         len = dc->len;
1122         total_len = len;
1123
1124         dc->len = 0;
1125
1126         while (total_len && *issued < dpolicy->max_requests && !err) {
1127                 struct bio *bio = NULL;
1128                 unsigned long flags;
1129                 bool last = true;
1130
1131                 if (len > max_discard_blocks) {
1132                         len = max_discard_blocks;
1133                         last = false;
1134                 }
1135
1136                 (*issued)++;
1137                 if (*issued == dpolicy->max_requests)
1138                         last = true;
1139
1140                 dc->len += len;
1141
1142                 if (time_to_inject(sbi, FAULT_DISCARD)) {
1143                         f2fs_show_injection_info(FAULT_DISCARD);
1144                         err = -EIO;
1145                         goto submit;
1146                 }
1147                 err = __blkdev_issue_discard(bdev,
1148                                         SECTOR_FROM_BLOCK(start),
1149                                         SECTOR_FROM_BLOCK(len),
1150                                         GFP_NOFS, 0, &bio);
1151 submit:
1152                 if (err) {
1153                         spin_lock_irqsave(&dc->lock, flags);
1154                         if (dc->state == D_PARTIAL)
1155                                 dc->state = D_SUBMIT;
1156                         spin_unlock_irqrestore(&dc->lock, flags);
1157
1158                         break;
1159                 }
1160
1161                 f2fs_bug_on(sbi, !bio);
1162
1163                 /*
1164                  * should keep before submission to avoid D_DONE
1165                  * right away
1166                  */
1167                 spin_lock_irqsave(&dc->lock, flags);
1168                 if (last)
1169                         dc->state = D_SUBMIT;
1170                 else
1171                         dc->state = D_PARTIAL;
1172                 dc->bio_ref++;
1173                 spin_unlock_irqrestore(&dc->lock, flags);
1174
1175                 atomic_inc(&dcc->queued_discard);
1176                 dc->queued++;
1177                 list_move_tail(&dc->list, wait_list);
1178
1179                 /* sanity check on discard range */
1180                 __check_sit_bitmap(sbi, lstart, lstart + len);
1181
1182                 bio->bi_private = dc;
1183                 bio->bi_end_io = f2fs_submit_discard_endio;
1184                 bio->bi_opf |= flag;
1185                 submit_bio(bio);
1186
1187                 atomic_inc(&dcc->issued_discard);
1188
1189                 f2fs_update_iostat(sbi, FS_DISCARD, 1);
1190
1191                 lstart += len;
1192                 start += len;
1193                 total_len -= len;
1194                 len = total_len;
1195         }
1196
1197         if (!err && len)
1198                 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1199         return err;
1200 }
1201
1202 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1203                                 struct block_device *bdev, block_t lstart,
1204                                 block_t start, block_t len,
1205                                 struct rb_node **insert_p,
1206                                 struct rb_node *insert_parent)
1207 {
1208         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1209         struct rb_node **p;
1210         struct rb_node *parent = NULL;
1211         struct discard_cmd *dc = NULL;
1212         bool leftmost = true;
1213
1214         if (insert_p && insert_parent) {
1215                 parent = insert_parent;
1216                 p = insert_p;
1217                 goto do_insert;
1218         }
1219
1220         p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1221                                                         lstart, &leftmost);
1222 do_insert:
1223         dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1224                                                                 p, leftmost);
1225         if (!dc)
1226                 return NULL;
1227
1228         return dc;
1229 }
1230
1231 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1232                                                 struct discard_cmd *dc)
1233 {
1234         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1235 }
1236
1237 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1238                                 struct discard_cmd *dc, block_t blkaddr)
1239 {
1240         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1241         struct discard_info di = dc->di;
1242         bool modified = false;
1243
1244         if (dc->state == D_DONE || dc->len == 1) {
1245                 __remove_discard_cmd(sbi, dc);
1246                 return;
1247         }
1248
1249         dcc->undiscard_blks -= di.len;
1250
1251         if (blkaddr > di.lstart) {
1252                 dc->len = blkaddr - dc->lstart;
1253                 dcc->undiscard_blks += dc->len;
1254                 __relocate_discard_cmd(dcc, dc);
1255                 modified = true;
1256         }
1257
1258         if (blkaddr < di.lstart + di.len - 1) {
1259                 if (modified) {
1260                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1261                                         di.start + blkaddr + 1 - di.lstart,
1262                                         di.lstart + di.len - 1 - blkaddr,
1263                                         NULL, NULL);
1264                 } else {
1265                         dc->lstart++;
1266                         dc->len--;
1267                         dc->start++;
1268                         dcc->undiscard_blks += dc->len;
1269                         __relocate_discard_cmd(dcc, dc);
1270                 }
1271         }
1272 }
1273
1274 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1275                                 struct block_device *bdev, block_t lstart,
1276                                 block_t start, block_t len)
1277 {
1278         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1279         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1280         struct discard_cmd *dc;
1281         struct discard_info di = {0};
1282         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1283         struct request_queue *q = bdev_get_queue(bdev);
1284         unsigned int max_discard_blocks =
1285                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1286         block_t end = lstart + len;
1287
1288         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1289                                         NULL, lstart,
1290                                         (struct rb_entry **)&prev_dc,
1291                                         (struct rb_entry **)&next_dc,
1292                                         &insert_p, &insert_parent, true, NULL);
1293         if (dc)
1294                 prev_dc = dc;
1295
1296         if (!prev_dc) {
1297                 di.lstart = lstart;
1298                 di.len = next_dc ? next_dc->lstart - lstart : len;
1299                 di.len = min(di.len, len);
1300                 di.start = start;
1301         }
1302
1303         while (1) {
1304                 struct rb_node *node;
1305                 bool merged = false;
1306                 struct discard_cmd *tdc = NULL;
1307
1308                 if (prev_dc) {
1309                         di.lstart = prev_dc->lstart + prev_dc->len;
1310                         if (di.lstart < lstart)
1311                                 di.lstart = lstart;
1312                         if (di.lstart >= end)
1313                                 break;
1314
1315                         if (!next_dc || next_dc->lstart > end)
1316                                 di.len = end - di.lstart;
1317                         else
1318                                 di.len = next_dc->lstart - di.lstart;
1319                         di.start = start + di.lstart - lstart;
1320                 }
1321
1322                 if (!di.len)
1323                         goto next;
1324
1325                 if (prev_dc && prev_dc->state == D_PREP &&
1326                         prev_dc->bdev == bdev &&
1327                         __is_discard_back_mergeable(&di, &prev_dc->di,
1328                                                         max_discard_blocks)) {
1329                         prev_dc->di.len += di.len;
1330                         dcc->undiscard_blks += di.len;
1331                         __relocate_discard_cmd(dcc, prev_dc);
1332                         di = prev_dc->di;
1333                         tdc = prev_dc;
1334                         merged = true;
1335                 }
1336
1337                 if (next_dc && next_dc->state == D_PREP &&
1338                         next_dc->bdev == bdev &&
1339                         __is_discard_front_mergeable(&di, &next_dc->di,
1340                                                         max_discard_blocks)) {
1341                         next_dc->di.lstart = di.lstart;
1342                         next_dc->di.len += di.len;
1343                         next_dc->di.start = di.start;
1344                         dcc->undiscard_blks += di.len;
1345                         __relocate_discard_cmd(dcc, next_dc);
1346                         if (tdc)
1347                                 __remove_discard_cmd(sbi, tdc);
1348                         merged = true;
1349                 }
1350
1351                 if (!merged) {
1352                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1353                                                         di.len, NULL, NULL);
1354                 }
1355  next:
1356                 prev_dc = next_dc;
1357                 if (!prev_dc)
1358                         break;
1359
1360                 node = rb_next(&prev_dc->rb_node);
1361                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1362         }
1363 }
1364
1365 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1366                 struct block_device *bdev, block_t blkstart, block_t blklen)
1367 {
1368         block_t lblkstart = blkstart;
1369
1370         trace_f2fs_queue_discard(bdev, blkstart, blklen);
1371
1372         if (sbi->s_ndevs) {
1373                 int devi = f2fs_target_device_index(sbi, blkstart);
1374
1375                 blkstart -= FDEV(devi).start_blk;
1376         }
1377         mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1378         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1379         mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1380         return 0;
1381 }
1382
1383 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1384                                         struct discard_policy *dpolicy)
1385 {
1386         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1387         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1388         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1389         struct discard_cmd *dc;
1390         struct blk_plug plug;
1391         unsigned int pos = dcc->next_pos;
1392         unsigned int issued = 0;
1393         bool io_interrupted = false;
1394
1395         mutex_lock(&dcc->cmd_lock);
1396         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1397                                         NULL, pos,
1398                                         (struct rb_entry **)&prev_dc,
1399                                         (struct rb_entry **)&next_dc,
1400                                         &insert_p, &insert_parent, true, NULL);
1401         if (!dc)
1402                 dc = next_dc;
1403
1404         blk_start_plug(&plug);
1405
1406         while (dc) {
1407                 struct rb_node *node;
1408                 int err = 0;
1409
1410                 if (dc->state != D_PREP)
1411                         goto next;
1412
1413                 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1414                         io_interrupted = true;
1415                         break;
1416                 }
1417
1418                 dcc->next_pos = dc->lstart + dc->len;
1419                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1420
1421                 if (issued >= dpolicy->max_requests)
1422                         break;
1423 next:
1424                 node = rb_next(&dc->rb_node);
1425                 if (err)
1426                         __remove_discard_cmd(sbi, dc);
1427                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1428         }
1429
1430         blk_finish_plug(&plug);
1431
1432         if (!dc)
1433                 dcc->next_pos = 0;
1434
1435         mutex_unlock(&dcc->cmd_lock);
1436
1437         if (!issued && io_interrupted)
1438                 issued = -1;
1439
1440         return issued;
1441 }
1442
1443 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1444                                         struct discard_policy *dpolicy)
1445 {
1446         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1447         struct list_head *pend_list;
1448         struct discard_cmd *dc, *tmp;
1449         struct blk_plug plug;
1450         int i, issued = 0;
1451         bool io_interrupted = false;
1452
1453         if (dpolicy->timeout != 0)
1454                 f2fs_update_time(sbi, dpolicy->timeout);
1455
1456         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1457                 if (dpolicy->timeout != 0 &&
1458                                 f2fs_time_over(sbi, dpolicy->timeout))
1459                         break;
1460
1461                 if (i + 1 < dpolicy->granularity)
1462                         break;
1463
1464                 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1465                         return __issue_discard_cmd_orderly(sbi, dpolicy);
1466
1467                 pend_list = &dcc->pend_list[i];
1468
1469                 mutex_lock(&dcc->cmd_lock);
1470                 if (list_empty(pend_list))
1471                         goto next;
1472                 if (unlikely(dcc->rbtree_check))
1473                         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1474                                                                 &dcc->root));
1475                 blk_start_plug(&plug);
1476                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1477                         f2fs_bug_on(sbi, dc->state != D_PREP);
1478
1479                         if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1480                                                 !is_idle(sbi, DISCARD_TIME)) {
1481                                 io_interrupted = true;
1482                                 break;
1483                         }
1484
1485                         __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1486
1487                         if (issued >= dpolicy->max_requests)
1488                                 break;
1489                 }
1490                 blk_finish_plug(&plug);
1491 next:
1492                 mutex_unlock(&dcc->cmd_lock);
1493
1494                 if (issued >= dpolicy->max_requests || io_interrupted)
1495                         break;
1496         }
1497
1498         if (!issued && io_interrupted)
1499                 issued = -1;
1500
1501         return issued;
1502 }
1503
1504 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1505 {
1506         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1507         struct list_head *pend_list;
1508         struct discard_cmd *dc, *tmp;
1509         int i;
1510         bool dropped = false;
1511
1512         mutex_lock(&dcc->cmd_lock);
1513         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1514                 pend_list = &dcc->pend_list[i];
1515                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1516                         f2fs_bug_on(sbi, dc->state != D_PREP);
1517                         __remove_discard_cmd(sbi, dc);
1518                         dropped = true;
1519                 }
1520         }
1521         mutex_unlock(&dcc->cmd_lock);
1522
1523         return dropped;
1524 }
1525
1526 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1527 {
1528         __drop_discard_cmd(sbi);
1529 }
1530
1531 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1532                                                         struct discard_cmd *dc)
1533 {
1534         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1535         unsigned int len = 0;
1536
1537         wait_for_completion_io(&dc->wait);
1538         mutex_lock(&dcc->cmd_lock);
1539         f2fs_bug_on(sbi, dc->state != D_DONE);
1540         dc->ref--;
1541         if (!dc->ref) {
1542                 if (!dc->error)
1543                         len = dc->len;
1544                 __remove_discard_cmd(sbi, dc);
1545         }
1546         mutex_unlock(&dcc->cmd_lock);
1547
1548         return len;
1549 }
1550
1551 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1552                                                 struct discard_policy *dpolicy,
1553                                                 block_t start, block_t end)
1554 {
1555         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1556         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1557                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1558         struct discard_cmd *dc, *tmp;
1559         bool need_wait;
1560         unsigned int trimmed = 0;
1561
1562 next:
1563         need_wait = false;
1564
1565         mutex_lock(&dcc->cmd_lock);
1566         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1567                 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1568                         continue;
1569                 if (dc->len < dpolicy->granularity)
1570                         continue;
1571                 if (dc->state == D_DONE && !dc->ref) {
1572                         wait_for_completion_io(&dc->wait);
1573                         if (!dc->error)
1574                                 trimmed += dc->len;
1575                         __remove_discard_cmd(sbi, dc);
1576                 } else {
1577                         dc->ref++;
1578                         need_wait = true;
1579                         break;
1580                 }
1581         }
1582         mutex_unlock(&dcc->cmd_lock);
1583
1584         if (need_wait) {
1585                 trimmed += __wait_one_discard_bio(sbi, dc);
1586                 goto next;
1587         }
1588
1589         return trimmed;
1590 }
1591
1592 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1593                                                 struct discard_policy *dpolicy)
1594 {
1595         struct discard_policy dp;
1596         unsigned int discard_blks;
1597
1598         if (dpolicy)
1599                 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1600
1601         /* wait all */
1602         __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1603         discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1604         __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1605         discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1606
1607         return discard_blks;
1608 }
1609
1610 /* This should be covered by global mutex, &sit_i->sentry_lock */
1611 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1612 {
1613         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1614         struct discard_cmd *dc;
1615         bool need_wait = false;
1616
1617         mutex_lock(&dcc->cmd_lock);
1618         dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1619                                                         NULL, blkaddr);
1620         if (dc) {
1621                 if (dc->state == D_PREP) {
1622                         __punch_discard_cmd(sbi, dc, blkaddr);
1623                 } else {
1624                         dc->ref++;
1625                         need_wait = true;
1626                 }
1627         }
1628         mutex_unlock(&dcc->cmd_lock);
1629
1630         if (need_wait)
1631                 __wait_one_discard_bio(sbi, dc);
1632 }
1633
1634 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1635 {
1636         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1637
1638         if (dcc && dcc->f2fs_issue_discard) {
1639                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1640
1641                 dcc->f2fs_issue_discard = NULL;
1642                 kthread_stop(discard_thread);
1643         }
1644 }
1645
1646 /* This comes from f2fs_put_super */
1647 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1648 {
1649         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1650         struct discard_policy dpolicy;
1651         bool dropped;
1652
1653         __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1654                                         dcc->discard_granularity);
1655         dpolicy.timeout = UMOUNT_DISCARD_TIMEOUT;
1656         __issue_discard_cmd(sbi, &dpolicy);
1657         dropped = __drop_discard_cmd(sbi);
1658
1659         /* just to make sure there is no pending discard commands */
1660         __wait_all_discard_cmd(sbi, NULL);
1661
1662         f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1663         return dropped;
1664 }
1665
1666 static int issue_discard_thread(void *data)
1667 {
1668         struct f2fs_sb_info *sbi = data;
1669         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1670         wait_queue_head_t *q = &dcc->discard_wait_queue;
1671         struct discard_policy dpolicy;
1672         unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1673         int issued;
1674
1675         set_freezable();
1676
1677         do {
1678                 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1679                                         dcc->discard_granularity);
1680
1681                 wait_event_interruptible_timeout(*q,
1682                                 kthread_should_stop() || freezing(current) ||
1683                                 dcc->discard_wake,
1684                                 msecs_to_jiffies(wait_ms));
1685
1686                 if (dcc->discard_wake)
1687                         dcc->discard_wake = 0;
1688
1689                 /* clean up pending candidates before going to sleep */
1690                 if (atomic_read(&dcc->queued_discard))
1691                         __wait_all_discard_cmd(sbi, NULL);
1692
1693                 if (try_to_freeze())
1694                         continue;
1695                 if (f2fs_readonly(sbi->sb))
1696                         continue;
1697                 if (kthread_should_stop())
1698                         return 0;
1699                 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1700                         wait_ms = dpolicy.max_interval;
1701                         continue;
1702                 }
1703
1704                 if (sbi->gc_mode == GC_URGENT)
1705                         __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1706
1707                 sb_start_intwrite(sbi->sb);
1708
1709                 issued = __issue_discard_cmd(sbi, &dpolicy);
1710                 if (issued > 0) {
1711                         __wait_all_discard_cmd(sbi, &dpolicy);
1712                         wait_ms = dpolicy.min_interval;
1713                 } else if (issued == -1){
1714                         wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1715                         if (!wait_ms)
1716                                 wait_ms = dpolicy.mid_interval;
1717                 } else {
1718                         wait_ms = dpolicy.max_interval;
1719                 }
1720
1721                 sb_end_intwrite(sbi->sb);
1722
1723         } while (!kthread_should_stop());
1724         return 0;
1725 }
1726
1727 #ifdef CONFIG_BLK_DEV_ZONED
1728 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1729                 struct block_device *bdev, block_t blkstart, block_t blklen)
1730 {
1731         sector_t sector, nr_sects;
1732         block_t lblkstart = blkstart;
1733         int devi = 0;
1734
1735         if (sbi->s_ndevs) {
1736                 devi = f2fs_target_device_index(sbi, blkstart);
1737                 blkstart -= FDEV(devi).start_blk;
1738         }
1739
1740         /*
1741          * We need to know the type of the zone: for conventional zones,
1742          * use regular discard if the drive supports it. For sequential
1743          * zones, reset the zone write pointer.
1744          */
1745         switch (get_blkz_type(sbi, bdev, blkstart)) {
1746
1747         case BLK_ZONE_TYPE_CONVENTIONAL:
1748                 if (!blk_queue_discard(bdev_get_queue(bdev)))
1749                         return 0;
1750                 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1751         case BLK_ZONE_TYPE_SEQWRITE_REQ:
1752         case BLK_ZONE_TYPE_SEQWRITE_PREF:
1753                 sector = SECTOR_FROM_BLOCK(blkstart);
1754                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1755
1756                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1757                                 nr_sects != bdev_zone_sectors(bdev)) {
1758                         f2fs_msg(sbi->sb, KERN_INFO,
1759                                 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1760                                 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1761                                 blkstart, blklen);
1762                         return -EIO;
1763                 }
1764                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1765                 return blkdev_reset_zones(bdev, sector,
1766                                           nr_sects, GFP_NOFS);
1767         default:
1768                 /* Unknown zone type: broken device ? */
1769                 return -EIO;
1770         }
1771 }
1772 #endif
1773
1774 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1775                 struct block_device *bdev, block_t blkstart, block_t blklen)
1776 {
1777 #ifdef CONFIG_BLK_DEV_ZONED
1778         if (f2fs_sb_has_blkzoned(sbi) &&
1779                                 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1780                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1781 #endif
1782         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1783 }
1784
1785 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1786                                 block_t blkstart, block_t blklen)
1787 {
1788         sector_t start = blkstart, len = 0;
1789         struct block_device *bdev;
1790         struct seg_entry *se;
1791         unsigned int offset;
1792         block_t i;
1793         int err = 0;
1794
1795         bdev = f2fs_target_device(sbi, blkstart, NULL);
1796
1797         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1798                 if (i != start) {
1799                         struct block_device *bdev2 =
1800                                 f2fs_target_device(sbi, i, NULL);
1801
1802                         if (bdev2 != bdev) {
1803                                 err = __issue_discard_async(sbi, bdev,
1804                                                 start, len);
1805                                 if (err)
1806                                         return err;
1807                                 bdev = bdev2;
1808                                 start = i;
1809                                 len = 0;
1810                         }
1811                 }
1812
1813                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1814                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1815
1816                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1817                         sbi->discard_blks--;
1818         }
1819
1820         if (len)
1821                 err = __issue_discard_async(sbi, bdev, start, len);
1822         return err;
1823 }
1824
1825 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1826                                                         bool check_only)
1827 {
1828         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1829         int max_blocks = sbi->blocks_per_seg;
1830         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1831         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1832         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1833         unsigned long *discard_map = (unsigned long *)se->discard_map;
1834         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1835         unsigned int start = 0, end = -1;
1836         bool force = (cpc->reason & CP_DISCARD);
1837         struct discard_entry *de = NULL;
1838         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1839         int i;
1840
1841         if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
1842                 return false;
1843
1844         if (!force) {
1845                 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1846                         SM_I(sbi)->dcc_info->nr_discards >=
1847                                 SM_I(sbi)->dcc_info->max_discards)
1848                         return false;
1849         }
1850
1851         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1852         for (i = 0; i < entries; i++)
1853                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1854                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1855
1856         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1857                                 SM_I(sbi)->dcc_info->max_discards) {
1858                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1859                 if (start >= max_blocks)
1860                         break;
1861
1862                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1863                 if (force && start && end != max_blocks
1864                                         && (end - start) < cpc->trim_minlen)
1865                         continue;
1866
1867                 if (check_only)
1868                         return true;
1869
1870                 if (!de) {
1871                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1872                                                                 GFP_F2FS_ZERO);
1873                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1874                         list_add_tail(&de->list, head);
1875                 }
1876
1877                 for (i = start; i < end; i++)
1878                         __set_bit_le(i, (void *)de->discard_map);
1879
1880                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1881         }
1882         return false;
1883 }
1884
1885 static void release_discard_addr(struct discard_entry *entry)
1886 {
1887         list_del(&entry->list);
1888         kmem_cache_free(discard_entry_slab, entry);
1889 }
1890
1891 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1892 {
1893         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1894         struct discard_entry *entry, *this;
1895
1896         /* drop caches */
1897         list_for_each_entry_safe(entry, this, head, list)
1898                 release_discard_addr(entry);
1899 }
1900
1901 /*
1902  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1903  */
1904 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1905 {
1906         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1907         unsigned int segno;
1908
1909         mutex_lock(&dirty_i->seglist_lock);
1910         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1911                 __set_test_and_free(sbi, segno);
1912         mutex_unlock(&dirty_i->seglist_lock);
1913 }
1914
1915 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1916                                                 struct cp_control *cpc)
1917 {
1918         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1919         struct list_head *head = &dcc->entry_list;
1920         struct discard_entry *entry, *this;
1921         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1922         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1923         unsigned int start = 0, end = -1;
1924         unsigned int secno, start_segno;
1925         bool force = (cpc->reason & CP_DISCARD);
1926         bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
1927
1928         mutex_lock(&dirty_i->seglist_lock);
1929
1930         while (1) {
1931                 int i;
1932
1933                 if (need_align && end != -1)
1934                         end--;
1935                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1936                 if (start >= MAIN_SEGS(sbi))
1937                         break;
1938                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1939                                                                 start + 1);
1940
1941                 if (need_align) {
1942                         start = rounddown(start, sbi->segs_per_sec);
1943                         end = roundup(end, sbi->segs_per_sec);
1944                 }
1945
1946                 for (i = start; i < end; i++) {
1947                         if (test_and_clear_bit(i, prefree_map))
1948                                 dirty_i->nr_dirty[PRE]--;
1949                 }
1950
1951                 if (!f2fs_realtime_discard_enable(sbi))
1952                         continue;
1953
1954                 if (force && start >= cpc->trim_start &&
1955                                         (end - 1) <= cpc->trim_end)
1956                                 continue;
1957
1958                 if (!test_opt(sbi, LFS) || !__is_large_section(sbi)) {
1959                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1960                                 (end - start) << sbi->log_blocks_per_seg);
1961                         continue;
1962                 }
1963 next:
1964                 secno = GET_SEC_FROM_SEG(sbi, start);
1965                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1966                 if (!IS_CURSEC(sbi, secno) &&
1967                         !get_valid_blocks(sbi, start, true))
1968                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1969                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1970
1971                 start = start_segno + sbi->segs_per_sec;
1972                 if (start < end)
1973                         goto next;
1974                 else
1975                         end = start - 1;
1976         }
1977         mutex_unlock(&dirty_i->seglist_lock);
1978
1979         /* send small discards */
1980         list_for_each_entry_safe(entry, this, head, list) {
1981                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1982                 bool is_valid = test_bit_le(0, entry->discard_map);
1983
1984 find_next:
1985                 if (is_valid) {
1986                         next_pos = find_next_zero_bit_le(entry->discard_map,
1987                                         sbi->blocks_per_seg, cur_pos);
1988                         len = next_pos - cur_pos;
1989
1990                         if (f2fs_sb_has_blkzoned(sbi) ||
1991                             (force && len < cpc->trim_minlen))
1992                                 goto skip;
1993
1994                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1995                                                                         len);
1996                         total_len += len;
1997                 } else {
1998                         next_pos = find_next_bit_le(entry->discard_map,
1999                                         sbi->blocks_per_seg, cur_pos);
2000                 }
2001 skip:
2002                 cur_pos = next_pos;
2003                 is_valid = !is_valid;
2004
2005                 if (cur_pos < sbi->blocks_per_seg)
2006                         goto find_next;
2007
2008                 release_discard_addr(entry);
2009                 dcc->nr_discards -= total_len;
2010         }
2011
2012         wake_up_discard_thread(sbi, false);
2013 }
2014
2015 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2016 {
2017         dev_t dev = sbi->sb->s_bdev->bd_dev;
2018         struct discard_cmd_control *dcc;
2019         int err = 0, i;
2020
2021         if (SM_I(sbi)->dcc_info) {
2022                 dcc = SM_I(sbi)->dcc_info;
2023                 goto init_thread;
2024         }
2025
2026         dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2027         if (!dcc)
2028                 return -ENOMEM;
2029
2030         dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2031         INIT_LIST_HEAD(&dcc->entry_list);
2032         for (i = 0; i < MAX_PLIST_NUM; i++)
2033                 INIT_LIST_HEAD(&dcc->pend_list[i]);
2034         INIT_LIST_HEAD(&dcc->wait_list);
2035         INIT_LIST_HEAD(&dcc->fstrim_list);
2036         mutex_init(&dcc->cmd_lock);
2037         atomic_set(&dcc->issued_discard, 0);
2038         atomic_set(&dcc->queued_discard, 0);
2039         atomic_set(&dcc->discard_cmd_cnt, 0);
2040         dcc->nr_discards = 0;
2041         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2042         dcc->undiscard_blks = 0;
2043         dcc->next_pos = 0;
2044         dcc->root = RB_ROOT_CACHED;
2045         dcc->rbtree_check = false;
2046
2047         init_waitqueue_head(&dcc->discard_wait_queue);
2048         SM_I(sbi)->dcc_info = dcc;
2049 init_thread:
2050         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2051                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2052         if (IS_ERR(dcc->f2fs_issue_discard)) {
2053                 err = PTR_ERR(dcc->f2fs_issue_discard);
2054                 kvfree(dcc);
2055                 SM_I(sbi)->dcc_info = NULL;
2056                 return err;
2057         }
2058
2059         return err;
2060 }
2061
2062 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2063 {
2064         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2065
2066         if (!dcc)
2067                 return;
2068
2069         f2fs_stop_discard_thread(sbi);
2070
2071         kvfree(dcc);
2072         SM_I(sbi)->dcc_info = NULL;
2073 }
2074
2075 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2076 {
2077         struct sit_info *sit_i = SIT_I(sbi);
2078
2079         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2080                 sit_i->dirty_sentries++;
2081                 return false;
2082         }
2083
2084         return true;
2085 }
2086
2087 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2088                                         unsigned int segno, int modified)
2089 {
2090         struct seg_entry *se = get_seg_entry(sbi, segno);
2091         se->type = type;
2092         if (modified)
2093                 __mark_sit_entry_dirty(sbi, segno);
2094 }
2095
2096 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2097 {
2098         struct seg_entry *se;
2099         unsigned int segno, offset;
2100         long int new_vblocks;
2101         bool exist;
2102 #ifdef CONFIG_F2FS_CHECK_FS
2103         bool mir_exist;
2104 #endif
2105
2106         segno = GET_SEGNO(sbi, blkaddr);
2107
2108         se = get_seg_entry(sbi, segno);
2109         new_vblocks = se->valid_blocks + del;
2110         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2111
2112         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
2113                                 (new_vblocks > sbi->blocks_per_seg)));
2114
2115         se->valid_blocks = new_vblocks;
2116         se->mtime = get_mtime(sbi, false);
2117         if (se->mtime > SIT_I(sbi)->max_mtime)
2118                 SIT_I(sbi)->max_mtime = se->mtime;
2119
2120         /* Update valid block bitmap */
2121         if (del > 0) {
2122                 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2123 #ifdef CONFIG_F2FS_CHECK_FS
2124                 mir_exist = f2fs_test_and_set_bit(offset,
2125                                                 se->cur_valid_map_mir);
2126                 if (unlikely(exist != mir_exist)) {
2127                         f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2128                                 "when setting bitmap, blk:%u, old bit:%d",
2129                                 blkaddr, exist);
2130                         f2fs_bug_on(sbi, 1);
2131                 }
2132 #endif
2133                 if (unlikely(exist)) {
2134                         f2fs_msg(sbi->sb, KERN_ERR,
2135                                 "Bitmap was wrongly set, blk:%u", blkaddr);
2136                         f2fs_bug_on(sbi, 1);
2137                         se->valid_blocks--;
2138                         del = 0;
2139                 }
2140
2141                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
2142                         sbi->discard_blks--;
2143
2144                 /* don't overwrite by SSR to keep node chain */
2145                 if (IS_NODESEG(se->type) &&
2146                                 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2147                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2148                                 se->ckpt_valid_blocks++;
2149                 }
2150         } else {
2151                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2152 #ifdef CONFIG_F2FS_CHECK_FS
2153                 mir_exist = f2fs_test_and_clear_bit(offset,
2154                                                 se->cur_valid_map_mir);
2155                 if (unlikely(exist != mir_exist)) {
2156                         f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2157                                 "when clearing bitmap, blk:%u, old bit:%d",
2158                                 blkaddr, exist);
2159                         f2fs_bug_on(sbi, 1);
2160                 }
2161 #endif
2162                 if (unlikely(!exist)) {
2163                         f2fs_msg(sbi->sb, KERN_ERR,
2164                                 "Bitmap was wrongly cleared, blk:%u", blkaddr);
2165                         f2fs_bug_on(sbi, 1);
2166                         se->valid_blocks++;
2167                         del = 0;
2168                 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2169                         /*
2170                          * If checkpoints are off, we must not reuse data that
2171                          * was used in the previous checkpoint. If it was used
2172                          * before, we must track that to know how much space we
2173                          * really have.
2174                          */
2175                         if (f2fs_test_bit(offset, se->ckpt_valid_map))
2176                                 sbi->unusable_block_count++;
2177                 }
2178
2179                 if (f2fs_test_and_clear_bit(offset, se->discard_map))
2180                         sbi->discard_blks++;
2181         }
2182         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2183                 se->ckpt_valid_blocks += del;
2184
2185         __mark_sit_entry_dirty(sbi, segno);
2186
2187         /* update total number of valid blocks to be written in ckpt area */
2188         SIT_I(sbi)->written_valid_blocks += del;
2189
2190         if (__is_large_section(sbi))
2191                 get_sec_entry(sbi, segno)->valid_blocks += del;
2192 }
2193
2194 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2195 {
2196         unsigned int segno = GET_SEGNO(sbi, addr);
2197         struct sit_info *sit_i = SIT_I(sbi);
2198
2199         f2fs_bug_on(sbi, addr == NULL_ADDR);
2200         if (addr == NEW_ADDR)
2201                 return;
2202
2203         invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2204
2205         /* add it into sit main buffer */
2206         down_write(&sit_i->sentry_lock);
2207
2208         update_sit_entry(sbi, addr, -1);
2209
2210         /* add it into dirty seglist */
2211         locate_dirty_segment(sbi, segno);
2212
2213         up_write(&sit_i->sentry_lock);
2214 }
2215
2216 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2217 {
2218         struct sit_info *sit_i = SIT_I(sbi);
2219         unsigned int segno, offset;
2220         struct seg_entry *se;
2221         bool is_cp = false;
2222
2223         if (!is_valid_data_blkaddr(sbi, blkaddr))
2224                 return true;
2225
2226         down_read(&sit_i->sentry_lock);
2227
2228         segno = GET_SEGNO(sbi, blkaddr);
2229         se = get_seg_entry(sbi, segno);
2230         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2231
2232         if (f2fs_test_bit(offset, se->ckpt_valid_map))
2233                 is_cp = true;
2234
2235         up_read(&sit_i->sentry_lock);
2236
2237         return is_cp;
2238 }
2239
2240 /*
2241  * This function should be resided under the curseg_mutex lock
2242  */
2243 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2244                                         struct f2fs_summary *sum)
2245 {
2246         struct curseg_info *curseg = CURSEG_I(sbi, type);
2247         void *addr = curseg->sum_blk;
2248         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2249         memcpy(addr, sum, sizeof(struct f2fs_summary));
2250 }
2251
2252 /*
2253  * Calculate the number of current summary pages for writing
2254  */
2255 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2256 {
2257         int valid_sum_count = 0;
2258         int i, sum_in_page;
2259
2260         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2261                 if (sbi->ckpt->alloc_type[i] == SSR)
2262                         valid_sum_count += sbi->blocks_per_seg;
2263                 else {
2264                         if (for_ra)
2265                                 valid_sum_count += le16_to_cpu(
2266                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2267                         else
2268                                 valid_sum_count += curseg_blkoff(sbi, i);
2269                 }
2270         }
2271
2272         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2273                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2274         if (valid_sum_count <= sum_in_page)
2275                 return 1;
2276         else if ((valid_sum_count - sum_in_page) <=
2277                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2278                 return 2;
2279         return 3;
2280 }
2281
2282 /*
2283  * Caller should put this summary page
2284  */
2285 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2286 {
2287         return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
2288 }
2289
2290 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2291                                         void *src, block_t blk_addr)
2292 {
2293         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2294
2295         memcpy(page_address(page), src, PAGE_SIZE);
2296         set_page_dirty(page);
2297         f2fs_put_page(page, 1);
2298 }
2299
2300 static void write_sum_page(struct f2fs_sb_info *sbi,
2301                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
2302 {
2303         f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2304 }
2305
2306 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2307                                                 int type, block_t blk_addr)
2308 {
2309         struct curseg_info *curseg = CURSEG_I(sbi, type);
2310         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2311         struct f2fs_summary_block *src = curseg->sum_blk;
2312         struct f2fs_summary_block *dst;
2313
2314         dst = (struct f2fs_summary_block *)page_address(page);
2315         memset(dst, 0, PAGE_SIZE);
2316
2317         mutex_lock(&curseg->curseg_mutex);
2318
2319         down_read(&curseg->journal_rwsem);
2320         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2321         up_read(&curseg->journal_rwsem);
2322
2323         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2324         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2325
2326         mutex_unlock(&curseg->curseg_mutex);
2327
2328         set_page_dirty(page);
2329         f2fs_put_page(page, 1);
2330 }
2331
2332 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2333 {
2334         struct curseg_info *curseg = CURSEG_I(sbi, type);
2335         unsigned int segno = curseg->segno + 1;
2336         struct free_segmap_info *free_i = FREE_I(sbi);
2337
2338         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2339                 return !test_bit(segno, free_i->free_segmap);
2340         return 0;
2341 }
2342
2343 /*
2344  * Find a new segment from the free segments bitmap to right order
2345  * This function should be returned with success, otherwise BUG
2346  */
2347 static void get_new_segment(struct f2fs_sb_info *sbi,
2348                         unsigned int *newseg, bool new_sec, int dir)
2349 {
2350         struct free_segmap_info *free_i = FREE_I(sbi);
2351         unsigned int segno, secno, zoneno;
2352         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2353         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2354         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2355         unsigned int left_start = hint;
2356         bool init = true;
2357         int go_left = 0;
2358         int i;
2359
2360         spin_lock(&free_i->segmap_lock);
2361
2362         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2363                 segno = find_next_zero_bit(free_i->free_segmap,
2364                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2365                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2366                         goto got_it;
2367         }
2368 find_other_zone:
2369         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2370         if (secno >= MAIN_SECS(sbi)) {
2371                 if (dir == ALLOC_RIGHT) {
2372                         secno = find_next_zero_bit(free_i->free_secmap,
2373                                                         MAIN_SECS(sbi), 0);
2374                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2375                 } else {
2376                         go_left = 1;
2377                         left_start = hint - 1;
2378                 }
2379         }
2380         if (go_left == 0)
2381                 goto skip_left;
2382
2383         while (test_bit(left_start, free_i->free_secmap)) {
2384                 if (left_start > 0) {
2385                         left_start--;
2386                         continue;
2387                 }
2388                 left_start = find_next_zero_bit(free_i->free_secmap,
2389                                                         MAIN_SECS(sbi), 0);
2390                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2391                 break;
2392         }
2393         secno = left_start;
2394 skip_left:
2395         segno = GET_SEG_FROM_SEC(sbi, secno);
2396         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2397
2398         /* give up on finding another zone */
2399         if (!init)
2400                 goto got_it;
2401         if (sbi->secs_per_zone == 1)
2402                 goto got_it;
2403         if (zoneno == old_zoneno)
2404                 goto got_it;
2405         if (dir == ALLOC_LEFT) {
2406                 if (!go_left && zoneno + 1 >= total_zones)
2407                         goto got_it;
2408                 if (go_left && zoneno == 0)
2409                         goto got_it;
2410         }
2411         for (i = 0; i < NR_CURSEG_TYPE; i++)
2412                 if (CURSEG_I(sbi, i)->zone == zoneno)
2413                         break;
2414
2415         if (i < NR_CURSEG_TYPE) {
2416                 /* zone is in user, try another */
2417                 if (go_left)
2418                         hint = zoneno * sbi->secs_per_zone - 1;
2419                 else if (zoneno + 1 >= total_zones)
2420                         hint = 0;
2421                 else
2422                         hint = (zoneno + 1) * sbi->secs_per_zone;
2423                 init = false;
2424                 goto find_other_zone;
2425         }
2426 got_it:
2427         /* set it as dirty segment in free segmap */
2428         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2429         __set_inuse(sbi, segno);
2430         *newseg = segno;
2431         spin_unlock(&free_i->segmap_lock);
2432 }
2433
2434 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2435 {
2436         struct curseg_info *curseg = CURSEG_I(sbi, type);
2437         struct summary_footer *sum_footer;
2438
2439         curseg->segno = curseg->next_segno;
2440         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2441         curseg->next_blkoff = 0;
2442         curseg->next_segno = NULL_SEGNO;
2443
2444         sum_footer = &(curseg->sum_blk->footer);
2445         memset(sum_footer, 0, sizeof(struct summary_footer));
2446         if (IS_DATASEG(type))
2447                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2448         if (IS_NODESEG(type))
2449                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2450         __set_sit_entry_type(sbi, type, curseg->segno, modified);
2451 }
2452
2453 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2454 {
2455         /* if segs_per_sec is large than 1, we need to keep original policy. */
2456         if (__is_large_section(sbi))
2457                 return CURSEG_I(sbi, type)->segno;
2458
2459         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2460                 return 0;
2461
2462         if (test_opt(sbi, NOHEAP) &&
2463                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2464                 return 0;
2465
2466         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2467                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2468
2469         /* find segments from 0 to reuse freed segments */
2470         if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2471                 return 0;
2472
2473         return CURSEG_I(sbi, type)->segno;
2474 }
2475
2476 /*
2477  * Allocate a current working segment.
2478  * This function always allocates a free segment in LFS manner.
2479  */
2480 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2481 {
2482         struct curseg_info *curseg = CURSEG_I(sbi, type);
2483         unsigned int segno = curseg->segno;
2484         int dir = ALLOC_LEFT;
2485
2486         write_sum_page(sbi, curseg->sum_blk,
2487                                 GET_SUM_BLOCK(sbi, segno));
2488         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2489                 dir = ALLOC_RIGHT;
2490
2491         if (test_opt(sbi, NOHEAP))
2492                 dir = ALLOC_RIGHT;
2493
2494         segno = __get_next_segno(sbi, type);
2495         get_new_segment(sbi, &segno, new_sec, dir);
2496         curseg->next_segno = segno;
2497         reset_curseg(sbi, type, 1);
2498         curseg->alloc_type = LFS;
2499 }
2500
2501 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2502                         struct curseg_info *seg, block_t start)
2503 {
2504         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2505         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2506         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2507         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2508         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2509         int i, pos;
2510
2511         for (i = 0; i < entries; i++)
2512                 target_map[i] = ckpt_map[i] | cur_map[i];
2513
2514         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2515
2516         seg->next_blkoff = pos;
2517 }
2518
2519 /*
2520  * If a segment is written by LFS manner, next block offset is just obtained
2521  * by increasing the current block offset. However, if a segment is written by
2522  * SSR manner, next block offset obtained by calling __next_free_blkoff
2523  */
2524 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2525                                 struct curseg_info *seg)
2526 {
2527         if (seg->alloc_type == SSR)
2528                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2529         else
2530                 seg->next_blkoff++;
2531 }
2532
2533 /*
2534  * This function always allocates a used segment(from dirty seglist) by SSR
2535  * manner, so it should recover the existing segment information of valid blocks
2536  */
2537 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2538 {
2539         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2540         struct curseg_info *curseg = CURSEG_I(sbi, type);
2541         unsigned int new_segno = curseg->next_segno;
2542         struct f2fs_summary_block *sum_node;
2543         struct page *sum_page;
2544
2545         write_sum_page(sbi, curseg->sum_blk,
2546                                 GET_SUM_BLOCK(sbi, curseg->segno));
2547         __set_test_and_inuse(sbi, new_segno);
2548
2549         mutex_lock(&dirty_i->seglist_lock);
2550         __remove_dirty_segment(sbi, new_segno, PRE);
2551         __remove_dirty_segment(sbi, new_segno, DIRTY);
2552         mutex_unlock(&dirty_i->seglist_lock);
2553
2554         reset_curseg(sbi, type, 1);
2555         curseg->alloc_type = SSR;
2556         __next_free_blkoff(sbi, curseg, 0);
2557
2558         sum_page = f2fs_get_sum_page(sbi, new_segno);
2559         f2fs_bug_on(sbi, IS_ERR(sum_page));
2560         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2561         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2562         f2fs_put_page(sum_page, 1);
2563 }
2564
2565 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2566 {
2567         struct curseg_info *curseg = CURSEG_I(sbi, type);
2568         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2569         unsigned segno = NULL_SEGNO;
2570         int i, cnt;
2571         bool reversed = false;
2572
2573         /* f2fs_need_SSR() already forces to do this */
2574         if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2575                 curseg->next_segno = segno;
2576                 return 1;
2577         }
2578
2579         /* For node segments, let's do SSR more intensively */
2580         if (IS_NODESEG(type)) {
2581                 if (type >= CURSEG_WARM_NODE) {
2582                         reversed = true;
2583                         i = CURSEG_COLD_NODE;
2584                 } else {
2585                         i = CURSEG_HOT_NODE;
2586                 }
2587                 cnt = NR_CURSEG_NODE_TYPE;
2588         } else {
2589                 if (type >= CURSEG_WARM_DATA) {
2590                         reversed = true;
2591                         i = CURSEG_COLD_DATA;
2592                 } else {
2593                         i = CURSEG_HOT_DATA;
2594                 }
2595                 cnt = NR_CURSEG_DATA_TYPE;
2596         }
2597
2598         for (; cnt-- > 0; reversed ? i-- : i++) {
2599                 if (i == type)
2600                         continue;
2601                 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2602                         curseg->next_segno = segno;
2603                         return 1;
2604                 }
2605         }
2606
2607         /* find valid_blocks=0 in dirty list */
2608         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2609                 segno = get_free_segment(sbi);
2610                 if (segno != NULL_SEGNO) {
2611                         curseg->next_segno = segno;
2612                         return 1;
2613                 }
2614         }
2615         return 0;
2616 }
2617
2618 /*
2619  * flush out current segment and replace it with new segment
2620  * This function should be returned with success, otherwise BUG
2621  */
2622 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2623                                                 int type, bool force)
2624 {
2625         struct curseg_info *curseg = CURSEG_I(sbi, type);
2626
2627         if (force)
2628                 new_curseg(sbi, type, true);
2629         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2630                                         type == CURSEG_WARM_NODE)
2631                 new_curseg(sbi, type, false);
2632         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2633                         likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2634                 new_curseg(sbi, type, false);
2635         else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2636                 change_curseg(sbi, type);
2637         else
2638                 new_curseg(sbi, type, false);
2639
2640         stat_inc_seg_type(sbi, curseg);
2641 }
2642
2643 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2644 {
2645         struct curseg_info *curseg;
2646         unsigned int old_segno;
2647         int i;
2648
2649         down_write(&SIT_I(sbi)->sentry_lock);
2650
2651         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2652                 curseg = CURSEG_I(sbi, i);
2653                 old_segno = curseg->segno;
2654                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2655                 locate_dirty_segment(sbi, old_segno);
2656         }
2657
2658         up_write(&SIT_I(sbi)->sentry_lock);
2659 }
2660
2661 static const struct segment_allocation default_salloc_ops = {
2662         .allocate_segment = allocate_segment_by_default,
2663 };
2664
2665 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2666                                                 struct cp_control *cpc)
2667 {
2668         __u64 trim_start = cpc->trim_start;
2669         bool has_candidate = false;
2670
2671         down_write(&SIT_I(sbi)->sentry_lock);
2672         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2673                 if (add_discard_addrs(sbi, cpc, true)) {
2674                         has_candidate = true;
2675                         break;
2676                 }
2677         }
2678         up_write(&SIT_I(sbi)->sentry_lock);
2679
2680         cpc->trim_start = trim_start;
2681         return has_candidate;
2682 }
2683
2684 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2685                                         struct discard_policy *dpolicy,
2686                                         unsigned int start, unsigned int end)
2687 {
2688         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2689         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2690         struct rb_node **insert_p = NULL, *insert_parent = NULL;
2691         struct discard_cmd *dc;
2692         struct blk_plug plug;
2693         int issued;
2694         unsigned int trimmed = 0;
2695
2696 next:
2697         issued = 0;
2698
2699         mutex_lock(&dcc->cmd_lock);
2700         if (unlikely(dcc->rbtree_check))
2701                 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2702                                                                 &dcc->root));
2703
2704         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2705                                         NULL, start,
2706                                         (struct rb_entry **)&prev_dc,
2707                                         (struct rb_entry **)&next_dc,
2708                                         &insert_p, &insert_parent, true, NULL);
2709         if (!dc)
2710                 dc = next_dc;
2711
2712         blk_start_plug(&plug);
2713
2714         while (dc && dc->lstart <= end) {
2715                 struct rb_node *node;
2716                 int err = 0;
2717
2718                 if (dc->len < dpolicy->granularity)
2719                         goto skip;
2720
2721                 if (dc->state != D_PREP) {
2722                         list_move_tail(&dc->list, &dcc->fstrim_list);
2723                         goto skip;
2724                 }
2725
2726                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2727
2728                 if (issued >= dpolicy->max_requests) {
2729                         start = dc->lstart + dc->len;
2730
2731                         if (err)
2732                                 __remove_discard_cmd(sbi, dc);
2733
2734                         blk_finish_plug(&plug);
2735                         mutex_unlock(&dcc->cmd_lock);
2736                         trimmed += __wait_all_discard_cmd(sbi, NULL);
2737                         congestion_wait(BLK_RW_ASYNC, HZ/50);
2738                         goto next;
2739                 }
2740 skip:
2741                 node = rb_next(&dc->rb_node);
2742                 if (err)
2743                         __remove_discard_cmd(sbi, dc);
2744                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2745
2746                 if (fatal_signal_pending(current))
2747                         break;
2748         }
2749
2750         blk_finish_plug(&plug);
2751         mutex_unlock(&dcc->cmd_lock);
2752
2753         return trimmed;
2754 }
2755
2756 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2757 {
2758         __u64 start = F2FS_BYTES_TO_BLK(range->start);
2759         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2760         unsigned int start_segno, end_segno;
2761         block_t start_block, end_block;
2762         struct cp_control cpc;
2763         struct discard_policy dpolicy;
2764         unsigned long long trimmed = 0;
2765         int err = 0;
2766         bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
2767
2768         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2769                 return -EINVAL;
2770
2771         if (end < MAIN_BLKADDR(sbi))
2772                 goto out;
2773
2774         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2775                 f2fs_msg(sbi->sb, KERN_WARNING,
2776                         "Found FS corruption, run fsck to fix.");
2777                 return -EIO;
2778         }
2779
2780         /* start/end segment number in main_area */
2781         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2782         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2783                                                 GET_SEGNO(sbi, end);
2784         if (need_align) {
2785                 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2786                 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2787         }
2788
2789         cpc.reason = CP_DISCARD;
2790         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2791         cpc.trim_start = start_segno;
2792         cpc.trim_end = end_segno;
2793
2794         if (sbi->discard_blks == 0)
2795                 goto out;
2796
2797         mutex_lock(&sbi->gc_mutex);
2798         err = f2fs_write_checkpoint(sbi, &cpc);
2799         mutex_unlock(&sbi->gc_mutex);
2800         if (err)
2801                 goto out;
2802
2803         /*
2804          * We filed discard candidates, but actually we don't need to wait for
2805          * all of them, since they'll be issued in idle time along with runtime
2806          * discard option. User configuration looks like using runtime discard
2807          * or periodic fstrim instead of it.
2808          */
2809         if (f2fs_realtime_discard_enable(sbi))
2810                 goto out;
2811
2812         start_block = START_BLOCK(sbi, start_segno);
2813         end_block = START_BLOCK(sbi, end_segno + 1);
2814
2815         __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2816         trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2817                                         start_block, end_block);
2818
2819         trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
2820                                         start_block, end_block);
2821 out:
2822         if (!err)
2823                 range->len = F2FS_BLK_TO_BYTES(trimmed);
2824         return err;
2825 }
2826
2827 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2828 {
2829         struct curseg_info *curseg = CURSEG_I(sbi, type);
2830         if (curseg->next_blkoff < sbi->blocks_per_seg)
2831                 return true;
2832         return false;
2833 }
2834
2835 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
2836 {
2837         switch (hint) {
2838         case WRITE_LIFE_SHORT:
2839                 return CURSEG_HOT_DATA;
2840         case WRITE_LIFE_EXTREME:
2841                 return CURSEG_COLD_DATA;
2842         default:
2843                 return CURSEG_WARM_DATA;
2844         }
2845 }
2846
2847 /* This returns write hints for each segment type. This hints will be
2848  * passed down to block layer. There are mapping tables which depend on
2849  * the mount option 'whint_mode'.
2850  *
2851  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2852  *
2853  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2854  *
2855  * User                  F2FS                     Block
2856  * ----                  ----                     -----
2857  *                       META                     WRITE_LIFE_NOT_SET
2858  *                       HOT_NODE                 "
2859  *                       WARM_NODE                "
2860  *                       COLD_NODE                "
2861  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2862  * extension list        "                        "
2863  *
2864  * -- buffered io
2865  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2866  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2867  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2868  * WRITE_LIFE_NONE       "                        "
2869  * WRITE_LIFE_MEDIUM     "                        "
2870  * WRITE_LIFE_LONG       "                        "
2871  *
2872  * -- direct io
2873  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2874  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2875  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2876  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2877  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2878  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2879  *
2880  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2881  *
2882  * User                  F2FS                     Block
2883  * ----                  ----                     -----
2884  *                       META                     WRITE_LIFE_MEDIUM;
2885  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
2886  *                       WARM_NODE                "
2887  *                       COLD_NODE                WRITE_LIFE_NONE
2888  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2889  * extension list        "                        "
2890  *
2891  * -- buffered io
2892  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2893  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2894  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
2895  * WRITE_LIFE_NONE       "                        "
2896  * WRITE_LIFE_MEDIUM     "                        "
2897  * WRITE_LIFE_LONG       "                        "
2898  *
2899  * -- direct io
2900  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2901  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2902  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2903  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2904  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2905  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2906  */
2907
2908 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2909                                 enum page_type type, enum temp_type temp)
2910 {
2911         if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2912                 if (type == DATA) {
2913                         if (temp == WARM)
2914                                 return WRITE_LIFE_NOT_SET;
2915                         else if (temp == HOT)
2916                                 return WRITE_LIFE_SHORT;
2917                         else if (temp == COLD)
2918                                 return WRITE_LIFE_EXTREME;
2919                 } else {
2920                         return WRITE_LIFE_NOT_SET;
2921                 }
2922         } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2923                 if (type == DATA) {
2924                         if (temp == WARM)
2925                                 return WRITE_LIFE_LONG;
2926                         else if (temp == HOT)
2927                                 return WRITE_LIFE_SHORT;
2928                         else if (temp == COLD)
2929                                 return WRITE_LIFE_EXTREME;
2930                 } else if (type == NODE) {
2931                         if (temp == WARM || temp == HOT)
2932                                 return WRITE_LIFE_NOT_SET;
2933                         else if (temp == COLD)
2934                                 return WRITE_LIFE_NONE;
2935                 } else if (type == META) {
2936                         return WRITE_LIFE_MEDIUM;
2937                 }
2938         }
2939         return WRITE_LIFE_NOT_SET;
2940 }
2941
2942 static int __get_segment_type_2(struct f2fs_io_info *fio)
2943 {
2944         if (fio->type == DATA)
2945                 return CURSEG_HOT_DATA;
2946         else
2947                 return CURSEG_HOT_NODE;
2948 }
2949
2950 static int __get_segment_type_4(struct f2fs_io_info *fio)
2951 {
2952         if (fio->type == DATA) {
2953                 struct inode *inode = fio->page->mapping->host;
2954
2955                 if (S_ISDIR(inode->i_mode))
2956                         return CURSEG_HOT_DATA;
2957                 else
2958                         return CURSEG_COLD_DATA;
2959         } else {
2960                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2961                         return CURSEG_WARM_NODE;
2962                 else
2963                         return CURSEG_COLD_NODE;
2964         }
2965 }
2966
2967 static int __get_segment_type_6(struct f2fs_io_info *fio)
2968 {
2969         if (fio->type == DATA) {
2970                 struct inode *inode = fio->page->mapping->host;
2971
2972                 if (is_cold_data(fio->page) || file_is_cold(inode))
2973                         return CURSEG_COLD_DATA;
2974                 if (file_is_hot(inode) ||
2975                                 is_inode_flag_set(inode, FI_HOT_DATA) ||
2976                                 f2fs_is_atomic_file(inode) ||
2977                                 f2fs_is_volatile_file(inode))
2978                         return CURSEG_HOT_DATA;
2979                 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
2980         } else {
2981                 if (IS_DNODE(fio->page))
2982                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2983                                                 CURSEG_HOT_NODE;
2984                 return CURSEG_COLD_NODE;
2985         }
2986 }
2987
2988 static int __get_segment_type(struct f2fs_io_info *fio)
2989 {
2990         int type = 0;
2991
2992         switch (F2FS_OPTION(fio->sbi).active_logs) {
2993         case 2:
2994                 type = __get_segment_type_2(fio);
2995                 break;
2996         case 4:
2997                 type = __get_segment_type_4(fio);
2998                 break;
2999         case 6:
3000                 type = __get_segment_type_6(fio);
3001                 break;
3002         default:
3003                 f2fs_bug_on(fio->sbi, true);
3004         }
3005
3006         if (IS_HOT(type))
3007                 fio->temp = HOT;
3008         else if (IS_WARM(type))
3009                 fio->temp = WARM;
3010         else
3011                 fio->temp = COLD;
3012         return type;
3013 }
3014
3015 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3016                 block_t old_blkaddr, block_t *new_blkaddr,
3017                 struct f2fs_summary *sum, int type,
3018                 struct f2fs_io_info *fio, bool add_list)
3019 {
3020         struct sit_info *sit_i = SIT_I(sbi);
3021         struct curseg_info *curseg = CURSEG_I(sbi, type);
3022
3023         down_read(&SM_I(sbi)->curseg_lock);
3024
3025         mutex_lock(&curseg->curseg_mutex);
3026         down_write(&sit_i->sentry_lock);
3027
3028         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3029
3030         f2fs_wait_discard_bio(sbi, *new_blkaddr);
3031
3032         /*
3033          * __add_sum_entry should be resided under the curseg_mutex
3034          * because, this function updates a summary entry in the
3035          * current summary block.
3036          */
3037         __add_sum_entry(sbi, type, sum);
3038
3039         __refresh_next_blkoff(sbi, curseg);
3040
3041         stat_inc_block_count(sbi, curseg);
3042
3043         /*
3044          * SIT information should be updated before segment allocation,
3045          * since SSR needs latest valid block information.
3046          */
3047         update_sit_entry(sbi, *new_blkaddr, 1);
3048         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3049                 update_sit_entry(sbi, old_blkaddr, -1);
3050
3051         if (!__has_curseg_space(sbi, type))
3052                 sit_i->s_ops->allocate_segment(sbi, type, false);
3053
3054         /*
3055          * segment dirty status should be updated after segment allocation,
3056          * so we just need to update status only one time after previous
3057          * segment being closed.
3058          */
3059         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3060         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3061
3062         up_write(&sit_i->sentry_lock);
3063
3064         if (page && IS_NODESEG(type)) {
3065                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3066
3067                 f2fs_inode_chksum_set(sbi, page);
3068         }
3069
3070         if (add_list) {
3071                 struct f2fs_bio_info *io;
3072
3073                 INIT_LIST_HEAD(&fio->list);
3074                 fio->in_list = true;
3075                 fio->retry = false;
3076                 io = sbi->write_io[fio->type] + fio->temp;
3077                 spin_lock(&io->io_lock);
3078                 list_add_tail(&fio->list, &io->io_list);
3079                 spin_unlock(&io->io_lock);
3080         }
3081
3082         mutex_unlock(&curseg->curseg_mutex);
3083
3084         up_read(&SM_I(sbi)->curseg_lock);
3085 }
3086
3087 static void update_device_state(struct f2fs_io_info *fio)
3088 {
3089         struct f2fs_sb_info *sbi = fio->sbi;
3090         unsigned int devidx;
3091
3092         if (!sbi->s_ndevs)
3093                 return;
3094
3095         devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3096
3097         /* update device state for fsync */
3098         f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
3099
3100         /* update device state for checkpoint */
3101         if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3102                 spin_lock(&sbi->dev_lock);
3103                 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3104                 spin_unlock(&sbi->dev_lock);
3105         }
3106 }
3107
3108 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3109 {
3110         int type = __get_segment_type(fio);
3111         bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
3112
3113         if (keep_order)
3114                 down_read(&fio->sbi->io_order_lock);
3115 reallocate:
3116         f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3117                         &fio->new_blkaddr, sum, type, fio, true);
3118         if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3119                 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3120                                         fio->old_blkaddr, fio->old_blkaddr);
3121
3122         /* writeout dirty page into bdev */
3123         f2fs_submit_page_write(fio);
3124         if (fio->retry) {
3125                 fio->old_blkaddr = fio->new_blkaddr;
3126                 goto reallocate;
3127         }
3128
3129         update_device_state(fio);
3130
3131         if (keep_order)
3132                 up_read(&fio->sbi->io_order_lock);
3133 }
3134
3135 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3136                                         enum iostat_type io_type)
3137 {
3138         struct f2fs_io_info fio = {
3139                 .sbi = sbi,
3140                 .type = META,
3141                 .temp = HOT,
3142                 .op = REQ_OP_WRITE,
3143                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3144                 .old_blkaddr = page->index,
3145                 .new_blkaddr = page->index,
3146                 .page = page,
3147                 .encrypted_page = NULL,
3148                 .in_list = false,
3149         };
3150
3151         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3152                 fio.op_flags &= ~REQ_META;
3153
3154         set_page_writeback(page);
3155         ClearPageError(page);
3156         f2fs_submit_page_write(&fio);
3157
3158         stat_inc_meta_count(sbi, page->index);
3159         f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3160 }
3161
3162 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3163 {
3164         struct f2fs_summary sum;
3165
3166         set_summary(&sum, nid, 0, 0);
3167         do_write_page(&sum, fio);
3168
3169         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3170 }
3171
3172 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3173                                         struct f2fs_io_info *fio)
3174 {
3175         struct f2fs_sb_info *sbi = fio->sbi;
3176         struct f2fs_summary sum;
3177
3178         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3179         set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3180         do_write_page(&sum, fio);
3181         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3182
3183         f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3184 }
3185
3186 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3187 {
3188         int err;
3189         struct f2fs_sb_info *sbi = fio->sbi;
3190
3191         fio->new_blkaddr = fio->old_blkaddr;
3192         /* i/o temperature is needed for passing down write hints */
3193         __get_segment_type(fio);
3194
3195         f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
3196                         GET_SEGNO(sbi, fio->new_blkaddr))->type));
3197
3198         stat_inc_inplace_blocks(fio->sbi);
3199
3200         err = f2fs_submit_page_bio(fio);
3201         if (!err) {
3202                 update_device_state(fio);
3203                 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3204         }
3205
3206         return err;
3207 }
3208
3209 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3210                                                 unsigned int segno)
3211 {
3212         int i;
3213
3214         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3215                 if (CURSEG_I(sbi, i)->segno == segno)
3216                         break;
3217         }
3218         return i;
3219 }
3220
3221 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3222                                 block_t old_blkaddr, block_t new_blkaddr,
3223                                 bool recover_curseg, bool recover_newaddr)
3224 {
3225         struct sit_info *sit_i = SIT_I(sbi);
3226         struct curseg_info *curseg;
3227         unsigned int segno, old_cursegno;
3228         struct seg_entry *se;
3229         int type;
3230         unsigned short old_blkoff;
3231
3232         segno = GET_SEGNO(sbi, new_blkaddr);
3233         se = get_seg_entry(sbi, segno);
3234         type = se->type;
3235
3236         down_write(&SM_I(sbi)->curseg_lock);
3237
3238         if (!recover_curseg) {
3239                 /* for recovery flow */
3240                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3241                         if (old_blkaddr == NULL_ADDR)
3242                                 type = CURSEG_COLD_DATA;
3243                         else
3244                                 type = CURSEG_WARM_DATA;
3245                 }
3246         } else {
3247                 if (IS_CURSEG(sbi, segno)) {
3248                         /* se->type is volatile as SSR allocation */
3249                         type = __f2fs_get_curseg(sbi, segno);
3250                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3251                 } else {
3252                         type = CURSEG_WARM_DATA;
3253                 }
3254         }
3255
3256         f2fs_bug_on(sbi, !IS_DATASEG(type));
3257         curseg = CURSEG_I(sbi, type);
3258
3259         mutex_lock(&curseg->curseg_mutex);
3260         down_write(&sit_i->sentry_lock);
3261
3262         old_cursegno = curseg->segno;
3263         old_blkoff = curseg->next_blkoff;
3264
3265         /* change the current segment */
3266         if (segno != curseg->segno) {
3267                 curseg->next_segno = segno;
3268                 change_curseg(sbi, type);
3269         }
3270
3271         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3272         __add_sum_entry(sbi, type, sum);
3273
3274         if (!recover_curseg || recover_newaddr)
3275                 update_sit_entry(sbi, new_blkaddr, 1);
3276         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3277                 invalidate_mapping_pages(META_MAPPING(sbi),
3278                                         old_blkaddr, old_blkaddr);
3279                 update_sit_entry(sbi, old_blkaddr, -1);
3280         }
3281
3282         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3283         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3284
3285         locate_dirty_segment(sbi, old_cursegno);
3286
3287         if (recover_curseg) {
3288                 if (old_cursegno != curseg->segno) {
3289                         curseg->next_segno = old_cursegno;
3290                         change_curseg(sbi, type);
3291                 }
3292                 curseg->next_blkoff = old_blkoff;
3293         }
3294
3295         up_write(&sit_i->sentry_lock);
3296         mutex_unlock(&curseg->curseg_mutex);
3297         up_write(&SM_I(sbi)->curseg_lock);
3298 }
3299
3300 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3301                                 block_t old_addr, block_t new_addr,
3302                                 unsigned char version, bool recover_curseg,
3303                                 bool recover_newaddr)
3304 {
3305         struct f2fs_summary sum;
3306
3307         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3308
3309         f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3310                                         recover_curseg, recover_newaddr);
3311
3312         f2fs_update_data_blkaddr(dn, new_addr);
3313 }
3314
3315 void f2fs_wait_on_page_writeback(struct page *page,
3316                                 enum page_type type, bool ordered, bool locked)
3317 {
3318         if (PageWriteback(page)) {
3319                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3320
3321                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3322                 if (ordered) {
3323                         wait_on_page_writeback(page);
3324                         f2fs_bug_on(sbi, locked && PageWriteback(page));
3325                 } else {
3326                         wait_for_stable_page(page);
3327                 }
3328         }
3329 }
3330
3331 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3332 {
3333         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3334         struct page *cpage;
3335
3336         if (!f2fs_post_read_required(inode))
3337                 return;
3338
3339         if (!is_valid_data_blkaddr(sbi, blkaddr))
3340                 return;
3341
3342         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3343         if (cpage) {
3344                 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3345                 f2fs_put_page(cpage, 1);
3346         }
3347 }
3348
3349 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3350                                                                 block_t len)
3351 {
3352         block_t i;
3353
3354         for (i = 0; i < len; i++)
3355                 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3356 }
3357
3358 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3359 {
3360         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3361         struct curseg_info *seg_i;
3362         unsigned char *kaddr;
3363         struct page *page;
3364         block_t start;
3365         int i, j, offset;
3366
3367         start = start_sum_block(sbi);
3368
3369         page = f2fs_get_meta_page(sbi, start++);
3370         if (IS_ERR(page))
3371                 return PTR_ERR(page);
3372         kaddr = (unsigned char *)page_address(page);
3373
3374         /* Step 1: restore nat cache */
3375         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3376         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3377
3378         /* Step 2: restore sit cache */
3379         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3380         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3381         offset = 2 * SUM_JOURNAL_SIZE;
3382
3383         /* Step 3: restore summary entries */
3384         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3385                 unsigned short blk_off;
3386                 unsigned int segno;
3387
3388                 seg_i = CURSEG_I(sbi, i);
3389                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3390                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3391                 seg_i->next_segno = segno;
3392                 reset_curseg(sbi, i, 0);
3393                 seg_i->alloc_type = ckpt->alloc_type[i];
3394                 seg_i->next_blkoff = blk_off;
3395
3396                 if (seg_i->alloc_type == SSR)
3397                         blk_off = sbi->blocks_per_seg;
3398
3399                 for (j = 0; j < blk_off; j++) {
3400                         struct f2fs_summary *s;
3401                         s = (struct f2fs_summary *)(kaddr + offset);
3402                         seg_i->sum_blk->entries[j] = *s;
3403                         offset += SUMMARY_SIZE;
3404                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3405                                                 SUM_FOOTER_SIZE)
3406                                 continue;
3407
3408                         f2fs_put_page(page, 1);
3409                         page = NULL;
3410
3411                         page = f2fs_get_meta_page(sbi, start++);
3412                         if (IS_ERR(page))
3413                                 return PTR_ERR(page);
3414                         kaddr = (unsigned char *)page_address(page);
3415                         offset = 0;
3416                 }
3417         }
3418         f2fs_put_page(page, 1);
3419         return 0;
3420 }
3421
3422 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3423 {
3424         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3425         struct f2fs_summary_block *sum;
3426         struct curseg_info *curseg;
3427         struct page *new;
3428         unsigned short blk_off;
3429         unsigned int segno = 0;
3430         block_t blk_addr = 0;
3431         int err = 0;
3432
3433         /* get segment number and block addr */
3434         if (IS_DATASEG(type)) {
3435                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3436                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3437                                                         CURSEG_HOT_DATA]);
3438                 if (__exist_node_summaries(sbi))
3439                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3440                 else
3441                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3442         } else {
3443                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3444                                                         CURSEG_HOT_NODE]);
3445                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3446                                                         CURSEG_HOT_NODE]);
3447                 if (__exist_node_summaries(sbi))
3448                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3449                                                         type - CURSEG_HOT_NODE);
3450                 else
3451                         blk_addr = GET_SUM_BLOCK(sbi, segno);
3452         }
3453
3454         new = f2fs_get_meta_page(sbi, blk_addr);
3455         if (IS_ERR(new))
3456                 return PTR_ERR(new);
3457         sum = (struct f2fs_summary_block *)page_address(new);
3458
3459         if (IS_NODESEG(type)) {
3460                 if (__exist_node_summaries(sbi)) {
3461                         struct f2fs_summary *ns = &sum->entries[0];
3462                         int i;
3463                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3464                                 ns->version = 0;
3465                                 ns->ofs_in_node = 0;
3466                         }
3467                 } else {
3468                         err = f2fs_restore_node_summary(sbi, segno, sum);
3469                         if (err)
3470                                 goto out;
3471                 }
3472         }
3473
3474         /* set uncompleted segment to curseg */
3475         curseg = CURSEG_I(sbi, type);
3476         mutex_lock(&curseg->curseg_mutex);
3477
3478         /* update journal info */
3479         down_write(&curseg->journal_rwsem);
3480         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3481         up_write(&curseg->journal_rwsem);
3482
3483         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3484         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3485         curseg->next_segno = segno;
3486         reset_curseg(sbi, type, 0);
3487         curseg->alloc_type = ckpt->alloc_type[type];
3488         curseg->next_blkoff = blk_off;
3489         mutex_unlock(&curseg->curseg_mutex);
3490 out:
3491         f2fs_put_page(new, 1);
3492         return err;
3493 }
3494
3495 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3496 {
3497         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3498         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3499         int type = CURSEG_HOT_DATA;
3500         int err;
3501
3502         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3503                 int npages = f2fs_npages_for_summary_flush(sbi, true);
3504
3505                 if (npages >= 2)
3506                         f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3507                                                         META_CP, true);
3508
3509                 /* restore for compacted data summary */
3510                 err = read_compacted_summaries(sbi);
3511                 if (err)
3512                         return err;
3513                 type = CURSEG_HOT_NODE;
3514         }
3515
3516         if (__exist_node_summaries(sbi))
3517                 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3518                                         NR_CURSEG_TYPE - type, META_CP, true);
3519
3520         for (; type <= CURSEG_COLD_NODE; type++) {
3521                 err = read_normal_summaries(sbi, type);
3522                 if (err)
3523                         return err;
3524         }
3525
3526         /* sanity check for summary blocks */
3527         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3528                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3529                 return -EINVAL;
3530
3531         return 0;
3532 }
3533
3534 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3535 {
3536         struct page *page;
3537         unsigned char *kaddr;
3538         struct f2fs_summary *summary;
3539         struct curseg_info *seg_i;
3540         int written_size = 0;
3541         int i, j;
3542
3543         page = f2fs_grab_meta_page(sbi, blkaddr++);
3544         kaddr = (unsigned char *)page_address(page);
3545         memset(kaddr, 0, PAGE_SIZE);
3546
3547         /* Step 1: write nat cache */
3548         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3549         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3550         written_size += SUM_JOURNAL_SIZE;
3551
3552         /* Step 2: write sit cache */
3553         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3554         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3555         written_size += SUM_JOURNAL_SIZE;
3556
3557         /* Step 3: write summary entries */
3558         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3559                 unsigned short blkoff;
3560                 seg_i = CURSEG_I(sbi, i);
3561                 if (sbi->ckpt->alloc_type[i] == SSR)
3562                         blkoff = sbi->blocks_per_seg;
3563                 else
3564                         blkoff = curseg_blkoff(sbi, i);
3565
3566                 for (j = 0; j < blkoff; j++) {
3567                         if (!page) {
3568                                 page = f2fs_grab_meta_page(sbi, blkaddr++);
3569                                 kaddr = (unsigned char *)page_address(page);
3570                                 memset(kaddr, 0, PAGE_SIZE);
3571                                 written_size = 0;
3572                         }
3573                         summary = (struct f2fs_summary *)(kaddr + written_size);
3574                         *summary = seg_i->sum_blk->entries[j];
3575                         written_size += SUMMARY_SIZE;
3576
3577                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3578                                                         SUM_FOOTER_SIZE)
3579                                 continue;
3580
3581                         set_page_dirty(page);
3582                         f2fs_put_page(page, 1);
3583                         page = NULL;
3584                 }
3585         }
3586         if (page) {
3587                 set_page_dirty(page);
3588                 f2fs_put_page(page, 1);
3589         }
3590 }
3591
3592 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3593                                         block_t blkaddr, int type)
3594 {
3595         int i, end;
3596         if (IS_DATASEG(type))
3597                 end = type + NR_CURSEG_DATA_TYPE;
3598         else
3599                 end = type + NR_CURSEG_NODE_TYPE;
3600
3601         for (i = type; i < end; i++)
3602                 write_current_sum_page(sbi, i, blkaddr + (i - type));
3603 }
3604
3605 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3606 {
3607         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3608                 write_compacted_summaries(sbi, start_blk);
3609         else
3610                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3611 }
3612
3613 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3614 {
3615         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3616 }
3617
3618 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3619                                         unsigned int val, int alloc)
3620 {
3621         int i;
3622
3623         if (type == NAT_JOURNAL) {
3624                 for (i = 0; i < nats_in_cursum(journal); i++) {
3625                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3626                                 return i;
3627                 }
3628                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3629                         return update_nats_in_cursum(journal, 1);
3630         } else if (type == SIT_JOURNAL) {
3631                 for (i = 0; i < sits_in_cursum(journal); i++)
3632                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3633                                 return i;
3634                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3635                         return update_sits_in_cursum(journal, 1);
3636         }
3637         return -1;
3638 }
3639
3640 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3641                                         unsigned int segno)
3642 {
3643         return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
3644 }
3645
3646 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3647                                         unsigned int start)
3648 {
3649         struct sit_info *sit_i = SIT_I(sbi);
3650         struct page *page;
3651         pgoff_t src_off, dst_off;
3652
3653         src_off = current_sit_addr(sbi, start);
3654         dst_off = next_sit_addr(sbi, src_off);
3655
3656         page = f2fs_grab_meta_page(sbi, dst_off);
3657         seg_info_to_sit_page(sbi, page, start);
3658
3659         set_page_dirty(page);
3660         set_to_next_sit(sit_i, start);
3661
3662         return page;
3663 }
3664
3665 static struct sit_entry_set *grab_sit_entry_set(void)
3666 {
3667         struct sit_entry_set *ses =
3668                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3669
3670         ses->entry_cnt = 0;
3671         INIT_LIST_HEAD(&ses->set_list);
3672         return ses;
3673 }
3674
3675 static void release_sit_entry_set(struct sit_entry_set *ses)
3676 {
3677         list_del(&ses->set_list);
3678         kmem_cache_free(sit_entry_set_slab, ses);
3679 }
3680
3681 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3682                                                 struct list_head *head)
3683 {
3684         struct sit_entry_set *next = ses;
3685
3686         if (list_is_last(&ses->set_list, head))
3687                 return;
3688
3689         list_for_each_entry_continue(next, head, set_list)
3690                 if (ses->entry_cnt <= next->entry_cnt)
3691                         break;
3692
3693         list_move_tail(&ses->set_list, &next->set_list);
3694 }
3695
3696 static void add_sit_entry(unsigned int segno, struct list_head *head)
3697 {
3698         struct sit_entry_set *ses;
3699         unsigned int start_segno = START_SEGNO(segno);
3700
3701         list_for_each_entry(ses, head, set_list) {
3702                 if (ses->start_segno == start_segno) {
3703                         ses->entry_cnt++;
3704                         adjust_sit_entry_set(ses, head);
3705                         return;
3706                 }
3707         }
3708
3709         ses = grab_sit_entry_set();
3710
3711         ses->start_segno = start_segno;
3712         ses->entry_cnt++;
3713         list_add(&ses->set_list, head);
3714 }
3715
3716 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3717 {
3718         struct f2fs_sm_info *sm_info = SM_I(sbi);
3719         struct list_head *set_list = &sm_info->sit_entry_set;
3720         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3721         unsigned int segno;
3722
3723         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3724                 add_sit_entry(segno, set_list);
3725 }
3726
3727 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3728 {
3729         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3730         struct f2fs_journal *journal = curseg->journal;
3731         int i;
3732
3733         down_write(&curseg->journal_rwsem);
3734         for (i = 0; i < sits_in_cursum(journal); i++) {
3735                 unsigned int segno;
3736                 bool dirtied;
3737
3738                 segno = le32_to_cpu(segno_in_journal(journal, i));
3739                 dirtied = __mark_sit_entry_dirty(sbi, segno);
3740
3741                 if (!dirtied)
3742                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3743         }
3744         update_sits_in_cursum(journal, -i);
3745         up_write(&curseg->journal_rwsem);
3746 }
3747
3748 /*
3749  * CP calls this function, which flushes SIT entries including sit_journal,
3750  * and moves prefree segs to free segs.
3751  */
3752 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3753 {
3754         struct sit_info *sit_i = SIT_I(sbi);
3755         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3756         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3757         struct f2fs_journal *journal = curseg->journal;
3758         struct sit_entry_set *ses, *tmp;
3759         struct list_head *head = &SM_I(sbi)->sit_entry_set;
3760         bool to_journal = true;
3761         struct seg_entry *se;
3762
3763         down_write(&sit_i->sentry_lock);
3764
3765         if (!sit_i->dirty_sentries)
3766                 goto out;
3767
3768         /*
3769          * add and account sit entries of dirty bitmap in sit entry
3770          * set temporarily
3771          */
3772         add_sits_in_set(sbi);
3773
3774         /*
3775          * if there are no enough space in journal to store dirty sit
3776          * entries, remove all entries from journal and add and account
3777          * them in sit entry set.
3778          */
3779         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3780                 remove_sits_in_journal(sbi);
3781
3782         /*
3783          * there are two steps to flush sit entries:
3784          * #1, flush sit entries to journal in current cold data summary block.
3785          * #2, flush sit entries to sit page.
3786          */
3787         list_for_each_entry_safe(ses, tmp, head, set_list) {
3788                 struct page *page = NULL;
3789                 struct f2fs_sit_block *raw_sit = NULL;
3790                 unsigned int start_segno = ses->start_segno;
3791                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3792                                                 (unsigned long)MAIN_SEGS(sbi));
3793                 unsigned int segno = start_segno;
3794
3795                 if (to_journal &&
3796                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3797                         to_journal = false;
3798
3799                 if (to_journal) {
3800                         down_write(&curseg->journal_rwsem);
3801                 } else {
3802                         page = get_next_sit_page(sbi, start_segno);
3803                         raw_sit = page_address(page);
3804                 }
3805
3806                 /* flush dirty sit entries in region of current sit set */
3807                 for_each_set_bit_from(segno, bitmap, end) {
3808                         int offset, sit_offset;
3809
3810                         se = get_seg_entry(sbi, segno);
3811 #ifdef CONFIG_F2FS_CHECK_FS
3812                         if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3813                                                 SIT_VBLOCK_MAP_SIZE))
3814                                 f2fs_bug_on(sbi, 1);
3815 #endif
3816
3817                         /* add discard candidates */
3818                         if (!(cpc->reason & CP_DISCARD)) {
3819                                 cpc->trim_start = segno;
3820                                 add_discard_addrs(sbi, cpc, false);
3821                         }
3822
3823                         if (to_journal) {
3824                                 offset = f2fs_lookup_journal_in_cursum(journal,
3825                                                         SIT_JOURNAL, segno, 1);
3826                                 f2fs_bug_on(sbi, offset < 0);
3827                                 segno_in_journal(journal, offset) =
3828                                                         cpu_to_le32(segno);
3829                                 seg_info_to_raw_sit(se,
3830                                         &sit_in_journal(journal, offset));
3831                                 check_block_count(sbi, segno,
3832                                         &sit_in_journal(journal, offset));
3833                         } else {
3834                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3835                                 seg_info_to_raw_sit(se,
3836                                                 &raw_sit->entries[sit_offset]);
3837                                 check_block_count(sbi, segno,
3838                                                 &raw_sit->entries[sit_offset]);
3839                         }
3840
3841                         __clear_bit(segno, bitmap);
3842                         sit_i->dirty_sentries--;
3843                         ses->entry_cnt--;
3844                 }
3845
3846                 if (to_journal)
3847                         up_write(&curseg->journal_rwsem);
3848                 else
3849                         f2fs_put_page(page, 1);
3850
3851                 f2fs_bug_on(sbi, ses->entry_cnt);
3852                 release_sit_entry_set(ses);
3853         }
3854
3855         f2fs_bug_on(sbi, !list_empty(head));
3856         f2fs_bug_on(sbi, sit_i->dirty_sentries);
3857 out:
3858         if (cpc->reason & CP_DISCARD) {
3859                 __u64 trim_start = cpc->trim_start;
3860
3861                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3862                         add_discard_addrs(sbi, cpc, false);
3863
3864                 cpc->trim_start = trim_start;
3865         }
3866         up_write(&sit_i->sentry_lock);
3867
3868         set_prefree_as_free_segments(sbi);
3869 }
3870
3871 static int build_sit_info(struct f2fs_sb_info *sbi)
3872 {
3873         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3874         struct sit_info *sit_i;
3875         unsigned int sit_segs, start;
3876         char *src_bitmap;
3877         unsigned int bitmap_size;
3878
3879         /* allocate memory for SIT information */
3880         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3881         if (!sit_i)
3882                 return -ENOMEM;
3883
3884         SM_I(sbi)->sit_info = sit_i;
3885
3886         sit_i->sentries =
3887                 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3888                                               MAIN_SEGS(sbi)),
3889                               GFP_KERNEL);
3890         if (!sit_i->sentries)
3891                 return -ENOMEM;
3892
3893         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3894         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3895                                                                 GFP_KERNEL);
3896         if (!sit_i->dirty_sentries_bitmap)
3897                 return -ENOMEM;
3898
3899         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3900                 sit_i->sentries[start].cur_valid_map
3901                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3902                 sit_i->sentries[start].ckpt_valid_map
3903                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3904                 if (!sit_i->sentries[start].cur_valid_map ||
3905                                 !sit_i->sentries[start].ckpt_valid_map)
3906                         return -ENOMEM;
3907
3908 #ifdef CONFIG_F2FS_CHECK_FS
3909                 sit_i->sentries[start].cur_valid_map_mir
3910                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3911                 if (!sit_i->sentries[start].cur_valid_map_mir)
3912                         return -ENOMEM;
3913 #endif
3914
3915                 sit_i->sentries[start].discard_map
3916                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3917                                                         GFP_KERNEL);
3918                 if (!sit_i->sentries[start].discard_map)
3919                         return -ENOMEM;
3920         }
3921
3922         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3923         if (!sit_i->tmp_map)
3924                 return -ENOMEM;
3925
3926         if (__is_large_section(sbi)) {
3927                 sit_i->sec_entries =
3928                         f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
3929                                                       MAIN_SECS(sbi)),
3930                                       GFP_KERNEL);
3931                 if (!sit_i->sec_entries)
3932                         return -ENOMEM;
3933         }
3934
3935         /* get information related with SIT */
3936         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3937
3938         /* setup SIT bitmap from ckeckpoint pack */
3939         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3940         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3941
3942         sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3943         if (!sit_i->sit_bitmap)
3944                 return -ENOMEM;
3945
3946 #ifdef CONFIG_F2FS_CHECK_FS
3947         sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3948         if (!sit_i->sit_bitmap_mir)
3949                 return -ENOMEM;
3950 #endif
3951
3952         /* init SIT information */
3953         sit_i->s_ops = &default_salloc_ops;
3954
3955         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3956         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
3957         sit_i->written_valid_blocks = 0;
3958         sit_i->bitmap_size = bitmap_size;
3959         sit_i->dirty_sentries = 0;
3960         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3961         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3962         sit_i->mounted_time = ktime_get_real_seconds();
3963         init_rwsem(&sit_i->sentry_lock);
3964         return 0;
3965 }
3966
3967 static int build_free_segmap(struct f2fs_sb_info *sbi)
3968 {
3969         struct free_segmap_info *free_i;
3970         unsigned int bitmap_size, sec_bitmap_size;
3971
3972         /* allocate memory for free segmap information */
3973         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
3974         if (!free_i)
3975                 return -ENOMEM;
3976
3977         SM_I(sbi)->free_info = free_i;
3978
3979         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3980         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
3981         if (!free_i->free_segmap)
3982                 return -ENOMEM;
3983
3984         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3985         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
3986         if (!free_i->free_secmap)
3987                 return -ENOMEM;
3988
3989         /* set all segments as dirty temporarily */
3990         memset(free_i->free_segmap, 0xff, bitmap_size);
3991         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3992
3993         /* init free segmap information */
3994         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
3995         free_i->free_segments = 0;
3996         free_i->free_sections = 0;
3997         spin_lock_init(&free_i->segmap_lock);
3998         return 0;
3999 }
4000
4001 static int build_curseg(struct f2fs_sb_info *sbi)
4002 {
4003         struct curseg_info *array;
4004         int i;
4005
4006         array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
4007                              GFP_KERNEL);
4008         if (!array)
4009                 return -ENOMEM;
4010
4011         SM_I(sbi)->curseg_array = array;
4012
4013         for (i = 0; i < NR_CURSEG_TYPE; i++) {
4014                 mutex_init(&array[i].curseg_mutex);
4015                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4016                 if (!array[i].sum_blk)
4017                         return -ENOMEM;
4018                 init_rwsem(&array[i].journal_rwsem);
4019                 array[i].journal = f2fs_kzalloc(sbi,
4020                                 sizeof(struct f2fs_journal), GFP_KERNEL);
4021                 if (!array[i].journal)
4022                         return -ENOMEM;
4023                 array[i].segno = NULL_SEGNO;
4024                 array[i].next_blkoff = 0;
4025         }
4026         return restore_curseg_summaries(sbi);
4027 }
4028
4029 static int build_sit_entries(struct f2fs_sb_info *sbi)
4030 {
4031         struct sit_info *sit_i = SIT_I(sbi);
4032         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4033         struct f2fs_journal *journal = curseg->journal;
4034         struct seg_entry *se;
4035         struct f2fs_sit_entry sit;
4036         int sit_blk_cnt = SIT_BLK_CNT(sbi);
4037         unsigned int i, start, end;
4038         unsigned int readed, start_blk = 0;
4039         int err = 0;
4040         block_t total_node_blocks = 0;
4041
4042         do {
4043                 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
4044                                                         META_SIT, true);
4045
4046                 start = start_blk * sit_i->sents_per_block;
4047                 end = (start_blk + readed) * sit_i->sents_per_block;
4048
4049                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4050                         struct f2fs_sit_block *sit_blk;
4051                         struct page *page;
4052
4053                         se = &sit_i->sentries[start];
4054                         page = get_current_sit_page(sbi, start);
4055                         if (IS_ERR(page))
4056                                 return PTR_ERR(page);
4057                         sit_blk = (struct f2fs_sit_block *)page_address(page);
4058                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4059                         f2fs_put_page(page, 1);
4060
4061                         err = check_block_count(sbi, start, &sit);
4062                         if (err)
4063                                 return err;
4064                         seg_info_from_raw_sit(se, &sit);
4065                         if (IS_NODESEG(se->type))
4066                                 total_node_blocks += se->valid_blocks;
4067
4068                         /* build discard map only one time */
4069                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4070                                 memset(se->discard_map, 0xff,
4071                                         SIT_VBLOCK_MAP_SIZE);
4072                         } else {
4073                                 memcpy(se->discard_map,
4074                                         se->cur_valid_map,
4075                                         SIT_VBLOCK_MAP_SIZE);
4076                                 sbi->discard_blks +=
4077                                         sbi->blocks_per_seg -
4078                                         se->valid_blocks;
4079                         }
4080
4081                         if (__is_large_section(sbi))
4082                                 get_sec_entry(sbi, start)->valid_blocks +=
4083                                                         se->valid_blocks;
4084                 }
4085                 start_blk += readed;
4086         } while (start_blk < sit_blk_cnt);
4087
4088         down_read(&curseg->journal_rwsem);
4089         for (i = 0; i < sits_in_cursum(journal); i++) {
4090                 unsigned int old_valid_blocks;
4091
4092                 start = le32_to_cpu(segno_in_journal(journal, i));
4093                 if (start >= MAIN_SEGS(sbi)) {
4094                         f2fs_msg(sbi->sb, KERN_ERR,
4095                                         "Wrong journal entry on segno %u",
4096                                         start);
4097                         set_sbi_flag(sbi, SBI_NEED_FSCK);
4098                         err = -EINVAL;
4099                         break;
4100                 }
4101
4102                 se = &sit_i->sentries[start];
4103                 sit = sit_in_journal(journal, i);
4104
4105                 old_valid_blocks = se->valid_blocks;
4106                 if (IS_NODESEG(se->type))
4107                         total_node_blocks -= old_valid_blocks;
4108
4109                 err = check_block_count(sbi, start, &sit);
4110                 if (err)
4111                         break;
4112                 seg_info_from_raw_sit(se, &sit);
4113                 if (IS_NODESEG(se->type))
4114                         total_node_blocks += se->valid_blocks;
4115
4116                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4117                         memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4118                 } else {
4119                         memcpy(se->discard_map, se->cur_valid_map,
4120                                                 SIT_VBLOCK_MAP_SIZE);
4121                         sbi->discard_blks += old_valid_blocks;
4122                         sbi->discard_blks -= se->valid_blocks;
4123                 }
4124
4125                 if (__is_large_section(sbi)) {
4126                         get_sec_entry(sbi, start)->valid_blocks +=
4127                                                         se->valid_blocks;
4128                         get_sec_entry(sbi, start)->valid_blocks -=
4129                                                         old_valid_blocks;
4130                 }
4131         }
4132         up_read(&curseg->journal_rwsem);
4133
4134         if (!err && total_node_blocks != valid_node_count(sbi)) {
4135                 f2fs_msg(sbi->sb, KERN_ERR,
4136                         "SIT is corrupted node# %u vs %u",
4137                         total_node_blocks, valid_node_count(sbi));
4138                 set_sbi_flag(sbi, SBI_NEED_FSCK);
4139                 err = -EINVAL;
4140         }
4141
4142         return err;
4143 }
4144
4145 static void init_free_segmap(struct f2fs_sb_info *sbi)
4146 {
4147         unsigned int start;
4148         int type;
4149
4150         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4151                 struct seg_entry *sentry = get_seg_entry(sbi, start);
4152                 if (!sentry->valid_blocks)
4153                         __set_free(sbi, start);
4154                 else
4155                         SIT_I(sbi)->written_valid_blocks +=
4156                                                 sentry->valid_blocks;
4157         }
4158
4159         /* set use the current segments */
4160         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4161                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4162                 __set_test_and_inuse(sbi, curseg_t->segno);
4163         }
4164 }
4165
4166 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4167 {
4168         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4169         struct free_segmap_info *free_i = FREE_I(sbi);
4170         unsigned int segno = 0, offset = 0;
4171         unsigned short valid_blocks;
4172
4173         while (1) {
4174                 /* find dirty segment based on free segmap */
4175                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4176                 if (segno >= MAIN_SEGS(sbi))
4177                         break;
4178                 offset = segno + 1;
4179                 valid_blocks = get_valid_blocks(sbi, segno, false);
4180                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
4181                         continue;
4182                 if (valid_blocks > sbi->blocks_per_seg) {
4183                         f2fs_bug_on(sbi, 1);
4184                         continue;
4185                 }
4186                 mutex_lock(&dirty_i->seglist_lock);
4187                 __locate_dirty_segment(sbi, segno, DIRTY);
4188                 mutex_unlock(&dirty_i->seglist_lock);
4189         }
4190 }
4191
4192 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4193 {
4194         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4195         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4196
4197         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4198         if (!dirty_i->victim_secmap)
4199                 return -ENOMEM;
4200         return 0;
4201 }
4202
4203 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4204 {
4205         struct dirty_seglist_info *dirty_i;
4206         unsigned int bitmap_size, i;
4207
4208         /* allocate memory for dirty segments list information */
4209         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4210                                                                 GFP_KERNEL);
4211         if (!dirty_i)
4212                 return -ENOMEM;
4213
4214         SM_I(sbi)->dirty_info = dirty_i;
4215         mutex_init(&dirty_i->seglist_lock);
4216
4217         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4218
4219         for (i = 0; i < NR_DIRTY_TYPE; i++) {
4220                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4221                                                                 GFP_KERNEL);
4222                 if (!dirty_i->dirty_segmap[i])
4223                         return -ENOMEM;
4224         }
4225
4226         init_dirty_segmap(sbi);
4227         return init_victim_secmap(sbi);
4228 }
4229
4230 /*
4231  * Update min, max modified time for cost-benefit GC algorithm
4232  */
4233 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4234 {
4235         struct sit_info *sit_i = SIT_I(sbi);
4236         unsigned int segno;
4237
4238         down_write(&sit_i->sentry_lock);
4239
4240         sit_i->min_mtime = ULLONG_MAX;
4241
4242         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4243                 unsigned int i;
4244                 unsigned long long mtime = 0;
4245
4246                 for (i = 0; i < sbi->segs_per_sec; i++)
4247                         mtime += get_seg_entry(sbi, segno + i)->mtime;
4248
4249                 mtime = div_u64(mtime, sbi->segs_per_sec);
4250
4251                 if (sit_i->min_mtime > mtime)
4252                         sit_i->min_mtime = mtime;
4253         }
4254         sit_i->max_mtime = get_mtime(sbi, false);
4255         up_write(&sit_i->sentry_lock);
4256 }
4257
4258 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
4259 {
4260         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4261         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4262         struct f2fs_sm_info *sm_info;
4263         int err;
4264
4265         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
4266         if (!sm_info)
4267                 return -ENOMEM;
4268
4269         /* init sm info */
4270         sbi->sm_info = sm_info;
4271         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4272         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4273         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4274         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4275         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4276         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4277         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
4278         sm_info->rec_prefree_segments = sm_info->main_segments *
4279                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
4280         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4281                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4282
4283         if (!test_opt(sbi, LFS))
4284                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
4285         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
4286         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
4287         sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
4288         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
4289         sm_info->min_ssr_sections = reserved_sections(sbi);
4290
4291         INIT_LIST_HEAD(&sm_info->sit_entry_set);
4292
4293         init_rwsem(&sm_info->curseg_lock);
4294
4295         if (!f2fs_readonly(sbi->sb)) {
4296                 err = f2fs_create_flush_cmd_control(sbi);
4297                 if (err)
4298                         return err;
4299         }
4300
4301         err = create_discard_cmd_control(sbi);
4302         if (err)
4303                 return err;
4304
4305         err = build_sit_info(sbi);
4306         if (err)
4307                 return err;
4308         err = build_free_segmap(sbi);
4309         if (err)
4310                 return err;
4311         err = build_curseg(sbi);
4312         if (err)
4313                 return err;
4314
4315         /* reinit free segmap based on SIT */
4316         err = build_sit_entries(sbi);
4317         if (err)
4318                 return err;
4319
4320         init_free_segmap(sbi);
4321         err = build_dirty_segmap(sbi);
4322         if (err)
4323                 return err;
4324
4325         init_min_max_mtime(sbi);
4326         return 0;
4327 }
4328
4329 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4330                 enum dirty_type dirty_type)
4331 {
4332         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4333
4334         mutex_lock(&dirty_i->seglist_lock);
4335         kvfree(dirty_i->dirty_segmap[dirty_type]);
4336         dirty_i->nr_dirty[dirty_type] = 0;
4337         mutex_unlock(&dirty_i->seglist_lock);
4338 }
4339
4340 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
4341 {
4342         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4343         kvfree(dirty_i->victim_secmap);
4344 }
4345
4346 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4347 {
4348         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4349         int i;
4350
4351         if (!dirty_i)
4352                 return;
4353
4354         /* discard pre-free/dirty segments list */
4355         for (i = 0; i < NR_DIRTY_TYPE; i++)
4356                 discard_dirty_segmap(sbi, i);
4357
4358         destroy_victim_secmap(sbi);
4359         SM_I(sbi)->dirty_info = NULL;
4360         kvfree(dirty_i);
4361 }
4362
4363 static void destroy_curseg(struct f2fs_sb_info *sbi)
4364 {
4365         struct curseg_info *array = SM_I(sbi)->curseg_array;
4366         int i;
4367
4368         if (!array)
4369                 return;
4370         SM_I(sbi)->curseg_array = NULL;
4371         for (i = 0; i < NR_CURSEG_TYPE; i++) {
4372                 kvfree(array[i].sum_blk);
4373                 kvfree(array[i].journal);
4374         }
4375         kvfree(array);
4376 }
4377
4378 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4379 {
4380         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4381         if (!free_i)
4382                 return;
4383         SM_I(sbi)->free_info = NULL;
4384         kvfree(free_i->free_segmap);
4385         kvfree(free_i->free_secmap);
4386         kvfree(free_i);
4387 }
4388
4389 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4390 {
4391         struct sit_info *sit_i = SIT_I(sbi);
4392         unsigned int start;
4393
4394         if (!sit_i)
4395                 return;
4396
4397         if (sit_i->sentries) {
4398                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4399                         kvfree(sit_i->sentries[start].cur_valid_map);
4400 #ifdef CONFIG_F2FS_CHECK_FS
4401                         kvfree(sit_i->sentries[start].cur_valid_map_mir);
4402 #endif
4403                         kvfree(sit_i->sentries[start].ckpt_valid_map);
4404                         kvfree(sit_i->sentries[start].discard_map);
4405                 }
4406         }
4407         kvfree(sit_i->tmp_map);
4408
4409         kvfree(sit_i->sentries);
4410         kvfree(sit_i->sec_entries);
4411         kvfree(sit_i->dirty_sentries_bitmap);
4412
4413         SM_I(sbi)->sit_info = NULL;
4414         kvfree(sit_i->sit_bitmap);
4415 #ifdef CONFIG_F2FS_CHECK_FS
4416         kvfree(sit_i->sit_bitmap_mir);
4417 #endif
4418         kvfree(sit_i);
4419 }
4420
4421 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
4422 {
4423         struct f2fs_sm_info *sm_info = SM_I(sbi);
4424
4425         if (!sm_info)
4426                 return;
4427         f2fs_destroy_flush_cmd_control(sbi, true);
4428         destroy_discard_cmd_control(sbi);
4429         destroy_dirty_segmap(sbi);
4430         destroy_curseg(sbi);
4431         destroy_free_segmap(sbi);
4432         destroy_sit_info(sbi);
4433         sbi->sm_info = NULL;
4434         kvfree(sm_info);
4435 }
4436
4437 int __init f2fs_create_segment_manager_caches(void)
4438 {
4439         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4440                         sizeof(struct discard_entry));
4441         if (!discard_entry_slab)
4442                 goto fail;
4443
4444         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4445                         sizeof(struct discard_cmd));
4446         if (!discard_cmd_slab)
4447                 goto destroy_discard_entry;
4448
4449         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4450                         sizeof(struct sit_entry_set));
4451         if (!sit_entry_set_slab)
4452                 goto destroy_discard_cmd;
4453
4454         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4455                         sizeof(struct inmem_pages));
4456         if (!inmem_entry_slab)
4457                 goto destroy_sit_entry_set;
4458         return 0;
4459
4460 destroy_sit_entry_set:
4461         kmem_cache_destroy(sit_entry_set_slab);
4462 destroy_discard_cmd:
4463         kmem_cache_destroy(discard_cmd_slab);
4464 destroy_discard_entry:
4465         kmem_cache_destroy(discard_entry_slab);
4466 fail:
4467         return -ENOMEM;
4468 }
4469
4470 void f2fs_destroy_segment_manager_caches(void)
4471 {
4472         kmem_cache_destroy(sit_entry_set_slab);
4473         kmem_cache_destroy(discard_cmd_slab);
4474         kmem_cache_destroy(discard_entry_slab);
4475         kmem_cache_destroy(inmem_entry_slab);
4476 }