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                         mutex_lock(&sbi->flush_lock);
 550
 551                         blk_start_plug(&plug);
 552                         f2fs_sync_dirty_inodes(sbi, FILE_INODE);
 553                         blk_finish_plug(&plug);
 554
 555                         mutex_unlock(&sbi->flush_lock);
 556                 }
 557                 f2fs_sync_fs(sbi->sb, true);
 558                 stat_inc_bg_cp_count(sbi->stat_info);
 559         }
 560 }
 561
 562 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 563                                 struct block_device *bdev)
 564 {
 565         struct bio *bio;
 566         int ret;
 567
 568         bio = f2fs_bio_alloc(sbi, 0, false);
 569         if (!bio)
 570                 return -ENOMEM;
 571
 572         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
 573         bio_set_dev(bio, bdev);
 574         ret = submit_bio_wait(bio);
 575         bio_put(bio);
 576
 577         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 578                                 test_opt(sbi, FLUSH_MERGE), ret);
 579         return ret;
 580 }
 581
 582 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 583 {
 584         int ret = 0;
 585         int i;
 586
 587         if (!f2fs_is_multi_device(sbi))
 588                 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 589
 590         for (i = 0; i < sbi->s_ndevs; i++) {
 591                 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
 592                         continue;
 593                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 594                 if (ret)
 595                         break;
 596         }
 597         return ret;
 598 }
 599
 600 static int issue_flush_thread(void *data)
 601 {
 602         struct f2fs_sb_info *sbi = data;
 603         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 604         wait_queue_head_t *q = &fcc->flush_wait_queue;
 605 repeat:
 606         if (kthread_should_stop())
 607                 return 0;
 608
 609         sb_start_intwrite(sbi->sb);
 610
 611         if (!llist_empty(&fcc->issue_list)) {
 612                 struct flush_cmd *cmd, *next;
 613                 int ret;
 614
 615                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 616                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 617
 618                 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 619
 620                 ret = submit_flush_wait(sbi, cmd->ino);
 621                 atomic_inc(&fcc->issued_flush);
 622
 623                 llist_for_each_entry_safe(cmd, next,
 624                                           fcc->dispatch_list, llnode) {
 625                         cmd->ret = ret;
 626                         complete(&cmd->wait);
 627                 }
 628                 fcc->dispatch_list = NULL;
 629         }
 630
 631         sb_end_intwrite(sbi->sb);
 632
 633         wait_event_interruptible(*q,
 634                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 635         goto repeat;
 636 }
 637
 638 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 639 {
 640         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 641         struct flush_cmd cmd;
 642         int ret;
 643
 644         if (test_opt(sbi, NOBARRIER))
 645                 return 0;
 646
 647         if (!test_opt(sbi, FLUSH_MERGE)) {
 648                 atomic_inc(&fcc->queued_flush);
 649                 ret = submit_flush_wait(sbi, ino);
 650                 atomic_dec(&fcc->queued_flush);
 651                 atomic_inc(&fcc->issued_flush);
 652                 return ret;
 653         }
 654
 655         if (atomic_inc_return(&fcc->queued_flush) == 1 ||
 656             f2fs_is_multi_device(sbi)) {
 657                 ret = submit_flush_wait(sbi, ino);
 658                 atomic_dec(&fcc->queued_flush);
 659
 660                 atomic_inc(&fcc->issued_flush);
 661                 return ret;
 662         }
 663
 664         cmd.ino = ino;
 665         init_completion(&cmd.wait);
 666
 667         llist_add(&cmd.llnode, &fcc->issue_list);
 668
 669         /* update issue_list before we wake up issue_flush thread */
 670         smp_mb();
 671
 672         if (waitqueue_active(&fcc->flush_wait_queue))
 673                 wake_up(&fcc->flush_wait_queue);
 674
 675         if (fcc->f2fs_issue_flush) {
 676                 wait_for_completion(&cmd.wait);
 677                 atomic_dec(&fcc->queued_flush);
 678         } else {
 679                 struct llist_node *list;
 680
 681                 list = llist_del_all(&fcc->issue_list);
 682                 if (!list) {
 683                         wait_for_completion(&cmd.wait);
 684                         atomic_dec(&fcc->queued_flush);
 685                 } else {
 686                         struct flush_cmd *tmp, *next;
 687
 688                         ret = submit_flush_wait(sbi, ino);
 689
 690                         llist_for_each_entry_safe(tmp, next, list, llnode) {
 691                                 if (tmp == &cmd) {
 692                                         cmd.ret = ret;
 693                                         atomic_dec(&fcc->queued_flush);
 694                                         continue;
 695                                 }
 696                                 tmp->ret = ret;
 697                                 complete(&tmp->wait);
 698                         }
 699                 }
 700         }
 701
 702         return cmd.ret;
 703 }
 704
 705 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
 706 {
 707         dev_t dev = sbi->sb->s_bdev->bd_dev;
 708         struct flush_cmd_control *fcc;
 709         int err = 0;
 710
 711         if (SM_I(sbi)->fcc_info) {
 712                 fcc = SM_I(sbi)->fcc_info;
 713                 if (fcc->f2fs_issue_flush)
 714                         return err;
 715                 goto init_thread;
 716         }
 717
 718         fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
 719         if (!fcc)
 720                 return -ENOMEM;
 721         atomic_set(&fcc->issued_flush, 0);
 722         atomic_set(&fcc->queued_flush, 0);
 723         init_waitqueue_head(&fcc->flush_wait_queue);
 724         init_llist_head(&fcc->issue_list);
 725         SM_I(sbi)->fcc_info = fcc;
 726         if (!test_opt(sbi, FLUSH_MERGE))
 727                 return err;
 728
 729 init_thread:
 730         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 731                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 732         if (IS_ERR(fcc->f2fs_issue_flush)) {
 733                 err = PTR_ERR(fcc->f2fs_issue_flush);
 734                 kvfree(fcc);
 735                 SM_I(sbi)->fcc_info = NULL;
 736                 return err;
 737         }
 738
 739         return err;
 740 }
 741
 742 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 743 {
 744         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 745
 746         if (fcc && fcc->f2fs_issue_flush) {
 747                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 748
 749                 fcc->f2fs_issue_flush = NULL;
 750                 kthread_stop(flush_thread);
 751         }
 752         if (free) {
 753                 kvfree(fcc);
 754                 SM_I(sbi)->fcc_info = NULL;
 755         }
 756 }
 757
 758 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
 759 {
 760         int ret = 0, i;
 761
 762         if (!f2fs_is_multi_device(sbi))
 763                 return 0;
 764
 765         for (i = 1; i < sbi->s_ndevs; i++) {
 766                 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
 767                         continue;
 768                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 769                 if (ret)
 770                         break;
 771
 772                 spin_lock(&sbi->dev_lock);
 773                 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
 774                 spin_unlock(&sbi->dev_lock);
 775         }
 776
 777         return ret;
 778 }
 779
 780 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 781                 enum dirty_type dirty_type)
 782 {
 783         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 784
 785         /* need not be added */
 786         if (IS_CURSEG(sbi, segno))
 787                 return;
 788
 789         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 790                 dirty_i->nr_dirty[dirty_type]++;
 791
 792         if (dirty_type == DIRTY) {
 793                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 794                 enum dirty_type t = sentry->type;
 795
 796                 if (unlikely(t >= DIRTY)) {
 797                         f2fs_bug_on(sbi, 1);
 798                         return;
 799                 }
 800                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 801                         dirty_i->nr_dirty[t]++;
 802         }
 803 }
 804
 805 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 806                 enum dirty_type dirty_type)
 807 {
 808         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 809
 810         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 811                 dirty_i->nr_dirty[dirty_type]--;
 812
 813         if (dirty_type == DIRTY) {
 814                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 815                 enum dirty_type t = sentry->type;
 816
 817                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 818                         dirty_i->nr_dirty[t]--;
 819
 820                 if (get_valid_blocks(sbi, segno, true) == 0) {
 821                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 822                                                 dirty_i->victim_secmap);
 823 #ifdef CONFIG_F2FS_CHECK_FS
 824                         clear_bit(segno, SIT_I(sbi)->invalid_segmap);
 825 #endif
 826                 }
 827         }
 828 }
 829
 830 /*
 831  * Should not occur error such as -ENOMEM.
 832  * Adding dirty entry into seglist is not critical operation.
 833  * If a given segment is one of current working segments, it won't be added.
 834  */
 835 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 836 {
 837         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 838         unsigned short valid_blocks, ckpt_valid_blocks;
 839
 840         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 841                 return;
 842
 843         mutex_lock(&dirty_i->seglist_lock);
 844
 845         valid_blocks = get_valid_blocks(sbi, segno, false);
 846         ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno);
 847
 848         if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
 849                                 ckpt_valid_blocks == sbi->blocks_per_seg)) {
 850                 __locate_dirty_segment(sbi, segno, PRE);
 851                 __remove_dirty_segment(sbi, segno, DIRTY);
 852         } else if (valid_blocks < sbi->blocks_per_seg) {
 853                 __locate_dirty_segment(sbi, segno, DIRTY);
 854         } else {
 855                 /* Recovery routine with SSR needs this */
 856                 __remove_dirty_segment(sbi, segno, DIRTY);
 857         }
 858
 859         mutex_unlock(&dirty_i->seglist_lock);
 860 }
 861
 862 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
 863 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
 864 {
 865         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 866         unsigned int segno;
 867
 868         mutex_lock(&dirty_i->seglist_lock);
 869         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 870                 if (get_valid_blocks(sbi, segno, false))
 871                         continue;
 872                 if (IS_CURSEG(sbi, segno))
 873                         continue;
 874                 __locate_dirty_segment(sbi, segno, PRE);
 875                 __remove_dirty_segment(sbi, segno, DIRTY);
 876         }
 877         mutex_unlock(&dirty_i->seglist_lock);
 878 }
 879
 880 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
 881 {
 882         int ovp_hole_segs =
 883                 (overprovision_segments(sbi) - reserved_segments(sbi));
 884         block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
 885         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 886         block_t holes[2] = {0, 0};      /* DATA and NODE */
 887         block_t unusable;
 888         struct seg_entry *se;
 889         unsigned int segno;
 890
 891         mutex_lock(&dirty_i->seglist_lock);
 892         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 893                 se = get_seg_entry(sbi, segno);
 894                 if (IS_NODESEG(se->type))
 895                         holes[NODE] += sbi->blocks_per_seg - se->valid_blocks;
 896                 else
 897                         holes[DATA] += sbi->blocks_per_seg - se->valid_blocks;
 898         }
 899         mutex_unlock(&dirty_i->seglist_lock);
 900
 901         unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
 902         if (unusable > ovp_holes)
 903                 return unusable - ovp_holes;
 904         return 0;
 905 }
 906
 907 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
 908 {
 909         int ovp_hole_segs =
 910                 (overprovision_segments(sbi) - reserved_segments(sbi));
 911         if (unusable > F2FS_OPTION(sbi).unusable_cap)
 912                 return -EAGAIN;
 913         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
 914                 dirty_segments(sbi) > ovp_hole_segs)
 915                 return -EAGAIN;
 916         return 0;
 917 }
 918
 919 /* This is only used by SBI_CP_DISABLED */
 920 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
 921 {
 922         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 923         unsigned int segno = 0;
 924
 925         mutex_lock(&dirty_i->seglist_lock);
 926         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 927                 if (get_valid_blocks(sbi, segno, false))
 928                         continue;
 929                 if (get_ckpt_valid_blocks(sbi, segno))
 930                         continue;
 931                 mutex_unlock(&dirty_i->seglist_lock);
 932                 return segno;
 933         }
 934         mutex_unlock(&dirty_i->seglist_lock);
 935         return NULL_SEGNO;
 936 }
 937
 938 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 939                 struct block_device *bdev, block_t lstart,
 940                 block_t start, block_t len)
 941 {
 942         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 943         struct list_head *pend_list;
 944         struct discard_cmd *dc;
 945
 946         f2fs_bug_on(sbi, !len);
 947
 948         pend_list = &dcc->pend_list[plist_idx(len)];
 949
 950         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 951         INIT_LIST_HEAD(&dc->list);
 952         dc->bdev = bdev;
 953         dc->lstart = lstart;
 954         dc->start = start;
 955         dc->len = len;
 956         dc->ref = 0;
 957         dc->state = D_PREP;
 958         dc->queued = 0;
 959         dc->error = 0;
 960         init_completion(&dc->wait);
 961         list_add_tail(&dc->list, pend_list);
 962         spin_lock_init(&dc->lock);
 963         dc->bio_ref = 0;
 964         atomic_inc(&dcc->discard_cmd_cnt);
 965         dcc->undiscard_blks += len;
 966
 967         return dc;
 968 }
 969
 970 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 971                                 struct block_device *bdev, block_t lstart,
 972                                 block_t start, block_t len,
 973                                 struct rb_node *parent, struct rb_node **p,
 974                                 bool leftmost)
 975 {
 976         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 977         struct discard_cmd *dc;
 978
 979         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
 980
 981         rb_link_node(&dc->rb_node, parent, p);
 982         rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
 983
 984         return dc;
 985 }
 986
 987 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
 988                                                         struct discard_cmd *dc)
 989 {
 990         if (dc->state == D_DONE)
 991                 atomic_sub(dc->queued, &dcc->queued_discard);
 992
 993         list_del(&dc->list);
 994         rb_erase_cached(&dc->rb_node, &dcc->root);
 995         dcc->undiscard_blks -= dc->len;
 996
 997         kmem_cache_free(discard_cmd_slab, dc);
 998
 999         atomic_dec(&dcc->discard_cmd_cnt);
1000 }
1001
1002 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1003                                                         struct discard_cmd *dc)
1004 {
1005         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1006         unsigned long flags;
1007
1008         trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
1009
1010         spin_lock_irqsave(&dc->lock, flags);
1011         if (dc->bio_ref) {
1012                 spin_unlock_irqrestore(&dc->lock, flags);
1013                 return;
1014         }
1015         spin_unlock_irqrestore(&dc->lock, flags);
1016
1017         f2fs_bug_on(sbi, dc->ref);
1018
1019         if (dc->error == -EOPNOTSUPP)
1020                 dc->error = 0;
1021
1022         if (dc->error)
1023                 printk_ratelimited(
1024                         "%sF2FS-fs: Issue discard(%u, %u, %u) failed, ret: %d",
1025                         KERN_INFO, dc->lstart, dc->start, dc->len, dc->error);
1026         __detach_discard_cmd(dcc, dc);
1027 }
1028
1029 static void f2fs_submit_discard_endio(struct bio *bio)
1030 {
1031         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1032         unsigned long flags;
1033
1034         dc->error = blk_status_to_errno(bio->bi_status);
1035
1036         spin_lock_irqsave(&dc->lock, flags);
1037         dc->bio_ref--;
1038         if (!dc->bio_ref && dc->state == D_SUBMIT) {
1039                 dc->state = D_DONE;
1040                 complete_all(&dc->wait);
1041         }
1042         spin_unlock_irqrestore(&dc->lock, flags);
1043         bio_put(bio);
1044 }
1045
1046 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1047                                 block_t start, block_t end)
1048 {
1049 #ifdef CONFIG_F2FS_CHECK_FS
1050         struct seg_entry *sentry;
1051         unsigned int segno;
1052         block_t blk = start;
1053         unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1054         unsigned long *map;
1055
1056         while (blk < end) {
1057                 segno = GET_SEGNO(sbi, blk);
1058                 sentry = get_seg_entry(sbi, segno);
1059                 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1060
1061                 if (end < START_BLOCK(sbi, segno + 1))
1062                         size = GET_BLKOFF_FROM_SEG0(sbi, end);
1063                 else
1064                         size = max_blocks;
1065                 map = (unsigned long *)(sentry->cur_valid_map);
1066                 offset = __find_rev_next_bit(map, size, offset);
1067                 f2fs_bug_on(sbi, offset != size);
1068                 blk = START_BLOCK(sbi, segno + 1);
1069         }
1070 #endif
1071 }
1072
1073 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1074                                 struct discard_policy *dpolicy,
1075                                 int discard_type, unsigned int granularity)
1076 {
1077         /* common policy */
1078         dpolicy->type = discard_type;
1079         dpolicy->sync = true;
1080         dpolicy->ordered = false;
1081         dpolicy->granularity = granularity;
1082
1083         dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1084         dpolicy->io_aware_gran = MAX_PLIST_NUM;
1085         dpolicy->timeout = 0;
1086
1087         if (discard_type == DPOLICY_BG) {
1088                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1089                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1090                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1091                 dpolicy->io_aware = true;
1092                 dpolicy->sync = false;
1093                 dpolicy->ordered = true;
1094                 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1095                         dpolicy->granularity = 1;
1096                         dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1097                 }
1098         } else if (discard_type == DPOLICY_FORCE) {
1099                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1100                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1101                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1102                 dpolicy->io_aware = false;
1103         } else if (discard_type == DPOLICY_FSTRIM) {
1104                 dpolicy->io_aware = false;
1105         } else if (discard_type == DPOLICY_UMOUNT) {
1106                 dpolicy->max_requests = UINT_MAX;
1107                 dpolicy->io_aware = false;
1108                 /* we need to issue all to keep CP_TRIMMED_FLAG */
1109                 dpolicy->granularity = 1;
1110         }
1111 }
1112
1113 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1114                                 struct block_device *bdev, block_t lstart,
1115                                 block_t start, block_t len);
1116 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1117 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1118                                                 struct discard_policy *dpolicy,
1119                                                 struct discard_cmd *dc,
1120                                                 unsigned int *issued)
1121 {
1122         struct block_device *bdev = dc->bdev;
1123         struct request_queue *q = bdev_get_queue(bdev);
1124         unsigned int max_discard_blocks =
1125                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1126         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1127         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1128                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1129         int flag = dpolicy->sync ? REQ_SYNC : 0;
1130         block_t lstart, start, len, total_len;
1131         int err = 0;
1132
1133         if (dc->state != D_PREP)
1134                 return 0;
1135
1136         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1137                 return 0;
1138
1139         trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1140
1141         lstart = dc->lstart;
1142         start = dc->start;
1143         len = dc->len;
1144         total_len = len;
1145
1146         dc->len = 0;
1147
1148         while (total_len && *issued < dpolicy->max_requests && !err) {
1149                 struct bio *bio = NULL;
1150                 unsigned long flags;
1151                 bool last = true;
1152
1153                 if (len > max_discard_blocks) {
1154                         len = max_discard_blocks;
1155                         last = false;
1156                 }
1157
1158                 (*issued)++;
1159                 if (*issued == dpolicy->max_requests)
1160                         last = true;
1161
1162                 dc->len += len;
1163
1164                 if (time_to_inject(sbi, FAULT_DISCARD)) {
1165                         f2fs_show_injection_info(FAULT_DISCARD);
1166                         err = -EIO;
1167                         goto submit;
1168                 }
1169                 err = __blkdev_issue_discard(bdev,
1170                                         SECTOR_FROM_BLOCK(start),
1171                                         SECTOR_FROM_BLOCK(len),
1172                                         GFP_NOFS, 0, &bio);
1173 submit:
1174                 if (err) {
1175                         spin_lock_irqsave(&dc->lock, flags);
1176                         if (dc->state == D_PARTIAL)
1177                                 dc->state = D_SUBMIT;
1178                         spin_unlock_irqrestore(&dc->lock, flags);
1179
1180                         break;
1181                 }
1182
1183                 f2fs_bug_on(sbi, !bio);
1184
1185                 /*
1186                  * should keep before submission to avoid D_DONE
1187                  * right away
1188                  */
1189                 spin_lock_irqsave(&dc->lock, flags);
1190                 if (last)
1191                         dc->state = D_SUBMIT;
1192                 else
1193                         dc->state = D_PARTIAL;
1194                 dc->bio_ref++;
1195                 spin_unlock_irqrestore(&dc->lock, flags);
1196
1197                 atomic_inc(&dcc->queued_discard);
1198                 dc->queued++;
1199                 list_move_tail(&dc->list, wait_list);
1200
1201                 /* sanity check on discard range */
1202                 __check_sit_bitmap(sbi, lstart, lstart + len);
1203
1204                 bio->bi_private = dc;
1205                 bio->bi_end_io = f2fs_submit_discard_endio;
1206                 bio->bi_opf |= flag;
1207                 submit_bio(bio);
1208
1209                 atomic_inc(&dcc->issued_discard);
1210
1211                 f2fs_update_iostat(sbi, FS_DISCARD, 1);
1212
1213                 lstart += len;
1214                 start += len;
1215                 total_len -= len;
1216                 len = total_len;
1217         }
1218
1219         if (!err && len)
1220                 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1221         return err;
1222 }
1223
1224 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1225                                 struct block_device *bdev, block_t lstart,
1226                                 block_t start, block_t len,
1227                                 struct rb_node **insert_p,
1228                                 struct rb_node *insert_parent)
1229 {
1230         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1231         struct rb_node **p;
1232         struct rb_node *parent = NULL;
1233         struct discard_cmd *dc = NULL;
1234         bool leftmost = true;
1235
1236         if (insert_p && insert_parent) {
1237                 parent = insert_parent;
1238                 p = insert_p;
1239                 goto do_insert;
1240         }
1241
1242         p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1243                                                         lstart, &leftmost);
1244 do_insert:
1245         dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1246                                                                 p, leftmost);
1247         if (!dc)
1248                 return NULL;
1249
1250         return dc;
1251 }
1252
1253 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1254                                                 struct discard_cmd *dc)
1255 {
1256         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1257 }
1258
1259 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1260                                 struct discard_cmd *dc, block_t blkaddr)
1261 {
1262         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1263         struct discard_info di = dc->di;
1264         bool modified = false;
1265
1266         if (dc->state == D_DONE || dc->len == 1) {
1267                 __remove_discard_cmd(sbi, dc);
1268                 return;
1269         }
1270
1271         dcc->undiscard_blks -= di.len;
1272
1273         if (blkaddr > di.lstart) {
1274                 dc->len = blkaddr - dc->lstart;
1275                 dcc->undiscard_blks += dc->len;
1276                 __relocate_discard_cmd(dcc, dc);
1277                 modified = true;
1278         }
1279
1280         if (blkaddr < di.lstart + di.len - 1) {
1281                 if (modified) {
1282                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1283                                         di.start + blkaddr + 1 - di.lstart,
1284                                         di.lstart + di.len - 1 - blkaddr,
1285                                         NULL, NULL);
1286                 } else {
1287                         dc->lstart++;
1288                         dc->len--;
1289                         dc->start++;
1290                         dcc->undiscard_blks += dc->len;
1291                         __relocate_discard_cmd(dcc, dc);
1292                 }
1293         }
1294 }
1295
1296 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1297                                 struct block_device *bdev, block_t lstart,
1298                                 block_t start, block_t len)
1299 {
1300         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1301         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1302         struct discard_cmd *dc;
1303         struct discard_info di = {0};
1304         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1305         struct request_queue *q = bdev_get_queue(bdev);
1306         unsigned int max_discard_blocks =
1307                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1308         block_t end = lstart + len;
1309
1310         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1311                                         NULL, lstart,
1312                                         (struct rb_entry **)&prev_dc,
1313                                         (struct rb_entry **)&next_dc,
1314                                         &insert_p, &insert_parent, true, NULL);
1315         if (dc)
1316                 prev_dc = dc;
1317
1318         if (!prev_dc) {
1319                 di.lstart = lstart;
1320                 di.len = next_dc ? next_dc->lstart - lstart : len;
1321                 di.len = min(di.len, len);
1322                 di.start = start;
1323         }
1324
1325         while (1) {
1326                 struct rb_node *node;
1327                 bool merged = false;
1328                 struct discard_cmd *tdc = NULL;
1329
1330                 if (prev_dc) {
1331                         di.lstart = prev_dc->lstart + prev_dc->len;
1332                         if (di.lstart < lstart)
1333                                 di.lstart = lstart;
1334                         if (di.lstart >= end)
1335                                 break;
1336
1337                         if (!next_dc || next_dc->lstart > end)
1338                                 di.len = end - di.lstart;
1339                         else
1340                                 di.len = next_dc->lstart - di.lstart;
1341                         di.start = start + di.lstart - lstart;
1342                 }
1343
1344                 if (!di.len)
1345                         goto next;
1346
1347                 if (prev_dc && prev_dc->state == D_PREP &&
1348                         prev_dc->bdev == bdev &&
1349                         __is_discard_back_mergeable(&di, &prev_dc->di,
1350                                                         max_discard_blocks)) {
1351                         prev_dc->di.len += di.len;
1352                         dcc->undiscard_blks += di.len;
1353                         __relocate_discard_cmd(dcc, prev_dc);
1354                         di = prev_dc->di;
1355                         tdc = prev_dc;
1356                         merged = true;
1357                 }
1358
1359                 if (next_dc && next_dc->state == D_PREP &&
1360                         next_dc->bdev == bdev &&
1361                         __is_discard_front_mergeable(&di, &next_dc->di,
1362                                                         max_discard_blocks)) {
1363                         next_dc->di.lstart = di.lstart;
1364                         next_dc->di.len += di.len;
1365                         next_dc->di.start = di.start;
1366                         dcc->undiscard_blks += di.len;
1367                         __relocate_discard_cmd(dcc, next_dc);
1368                         if (tdc)
1369                                 __remove_discard_cmd(sbi, tdc);
1370                         merged = true;
1371                 }
1372
1373                 if (!merged) {
1374                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1375                                                         di.len, NULL, NULL);
1376                 }
1377  next:
1378                 prev_dc = next_dc;
1379                 if (!prev_dc)
1380                         break;
1381
1382                 node = rb_next(&prev_dc->rb_node);
1383                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1384         }
1385 }
1386
1387 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1388                 struct block_device *bdev, block_t blkstart, block_t blklen)
1389 {
1390         block_t lblkstart = blkstart;
1391
1392         if (!f2fs_bdev_support_discard(bdev))
1393                 return 0;
1394
1395         trace_f2fs_queue_discard(bdev, blkstart, blklen);
1396
1397         if (f2fs_is_multi_device(sbi)) {
1398                 int devi = f2fs_target_device_index(sbi, blkstart);
1399
1400                 blkstart -= FDEV(devi).start_blk;
1401         }
1402         mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1403         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1404         mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1405         return 0;
1406 }
1407
1408 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1409                                         struct discard_policy *dpolicy)
1410 {
1411         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1412         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1413         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1414         struct discard_cmd *dc;
1415         struct blk_plug plug;
1416         unsigned int pos = dcc->next_pos;
1417         unsigned int issued = 0;
1418         bool io_interrupted = false;
1419
1420         mutex_lock(&dcc->cmd_lock);
1421         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1422                                         NULL, pos,
1423                                         (struct rb_entry **)&prev_dc,
1424                                         (struct rb_entry **)&next_dc,
1425                                         &insert_p, &insert_parent, true, NULL);
1426         if (!dc)
1427                 dc = next_dc;
1428
1429         blk_start_plug(&plug);
1430
1431         while (dc) {
1432                 struct rb_node *node;
1433                 int err = 0;
1434
1435                 if (dc->state != D_PREP)
1436                         goto next;
1437
1438                 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1439                         io_interrupted = true;
1440                         break;
1441                 }
1442
1443                 dcc->next_pos = dc->lstart + dc->len;
1444                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1445
1446                 if (issued >= dpolicy->max_requests)
1447                         break;
1448 next:
1449                 node = rb_next(&dc->rb_node);
1450                 if (err)
1451                         __remove_discard_cmd(sbi, dc);
1452                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1453         }
1454
1455         blk_finish_plug(&plug);
1456
1457         if (!dc)
1458                 dcc->next_pos = 0;
1459
1460         mutex_unlock(&dcc->cmd_lock);
1461
1462         if (!issued && io_interrupted)
1463                 issued = -1;
1464
1465         return issued;
1466 }
1467
1468 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1469                                         struct discard_policy *dpolicy)
1470 {
1471         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1472         struct list_head *pend_list;
1473         struct discard_cmd *dc, *tmp;
1474         struct blk_plug plug;
1475         int i, issued = 0;
1476         bool io_interrupted = false;
1477
1478         if (dpolicy->timeout != 0)
1479                 f2fs_update_time(sbi, dpolicy->timeout);
1480
1481         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1482                 if (dpolicy->timeout != 0 &&
1483                                 f2fs_time_over(sbi, dpolicy->timeout))
1484                         break;
1485
1486                 if (i + 1 < dpolicy->granularity)
1487                         break;
1488
1489                 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1490                         return __issue_discard_cmd_orderly(sbi, dpolicy);
1491
1492                 pend_list = &dcc->pend_list[i];
1493
1494                 mutex_lock(&dcc->cmd_lock);
1495                 if (list_empty(pend_list))
1496                         goto next;
1497                 if (unlikely(dcc->rbtree_check))
1498                         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1499                                                                 &dcc->root));
1500                 blk_start_plug(&plug);
1501                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1502                         f2fs_bug_on(sbi, dc->state != D_PREP);
1503
1504                         if (dpolicy->timeout != 0 &&
1505                                 f2fs_time_over(sbi, dpolicy->timeout))
1506                                 break;
1507
1508                         if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1509                                                 !is_idle(sbi, DISCARD_TIME)) {
1510                                 io_interrupted = true;
1511                                 break;
1512                         }
1513
1514                         __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1515
1516                         if (issued >= dpolicy->max_requests)
1517                                 break;
1518                 }
1519                 blk_finish_plug(&plug);
1520 next:
1521                 mutex_unlock(&dcc->cmd_lock);
1522
1523                 if (issued >= dpolicy->max_requests || io_interrupted)
1524                         break;
1525         }
1526
1527         if (!issued && io_interrupted)
1528                 issued = -1;
1529
1530         return issued;
1531 }
1532
1533 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1534 {
1535         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1536         struct list_head *pend_list;
1537         struct discard_cmd *dc, *tmp;
1538         int i;
1539         bool dropped = false;
1540
1541         mutex_lock(&dcc->cmd_lock);
1542         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1543                 pend_list = &dcc->pend_list[i];
1544                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1545                         f2fs_bug_on(sbi, dc->state != D_PREP);
1546                         __remove_discard_cmd(sbi, dc);
1547                         dropped = true;
1548                 }
1549         }
1550         mutex_unlock(&dcc->cmd_lock);
1551
1552         return dropped;
1553 }
1554
1555 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1556 {
1557         __drop_discard_cmd(sbi);
1558 }
1559
1560 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1561                                                         struct discard_cmd *dc)
1562 {
1563         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1564         unsigned int len = 0;
1565
1566         wait_for_completion_io(&dc->wait);
1567         mutex_lock(&dcc->cmd_lock);
1568         f2fs_bug_on(sbi, dc->state != D_DONE);
1569         dc->ref--;
1570         if (!dc->ref) {
1571                 if (!dc->error)
1572                         len = dc->len;
1573                 __remove_discard_cmd(sbi, dc);
1574         }
1575         mutex_unlock(&dcc->cmd_lock);
1576
1577         return len;
1578 }
1579
1580 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1581                                                 struct discard_policy *dpolicy,
1582                                                 block_t start, block_t end)
1583 {
1584         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1585         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1586                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1587         struct discard_cmd *dc, *tmp;
1588         bool need_wait;
1589         unsigned int trimmed = 0;
1590
1591 next:
1592         need_wait = false;
1593
1594         mutex_lock(&dcc->cmd_lock);
1595         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1596                 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1597                         continue;
1598                 if (dc->len < dpolicy->granularity)
1599                         continue;
1600                 if (dc->state == D_DONE && !dc->ref) {
1601                         wait_for_completion_io(&dc->wait);
1602                         if (!dc->error)
1603                                 trimmed += dc->len;
1604                         __remove_discard_cmd(sbi, dc);
1605                 } else {
1606                         dc->ref++;
1607                         need_wait = true;
1608                         break;
1609                 }
1610         }
1611         mutex_unlock(&dcc->cmd_lock);
1612
1613         if (need_wait) {
1614                 trimmed += __wait_one_discard_bio(sbi, dc);
1615                 goto next;
1616         }
1617
1618         return trimmed;
1619 }
1620
1621 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1622                                                 struct discard_policy *dpolicy)
1623 {
1624         struct discard_policy dp;
1625         unsigned int discard_blks;
1626
1627         if (dpolicy)
1628                 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1629
1630         /* wait all */
1631         __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1632         discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1633         __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1634         discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1635
1636         return discard_blks;
1637 }
1638
1639 /* This should be covered by global mutex, &sit_i->sentry_lock */
1640 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1641 {
1642         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1643         struct discard_cmd *dc;
1644         bool need_wait = false;
1645
1646         mutex_lock(&dcc->cmd_lock);
1647         dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1648                                                         NULL, blkaddr);
1649         if (dc) {
1650                 if (dc->state == D_PREP) {
1651                         __punch_discard_cmd(sbi, dc, blkaddr);
1652                 } else {
1653                         dc->ref++;
1654                         need_wait = true;
1655                 }
1656         }
1657         mutex_unlock(&dcc->cmd_lock);
1658
1659         if (need_wait)
1660                 __wait_one_discard_bio(sbi, dc);
1661 }
1662
1663 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1664 {
1665         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1666
1667         if (dcc && dcc->f2fs_issue_discard) {
1668                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1669
1670                 dcc->f2fs_issue_discard = NULL;
1671                 kthread_stop(discard_thread);
1672         }
1673 }
1674
1675 /* This comes from f2fs_put_super */
1676 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1677 {
1678         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1679         struct discard_policy dpolicy;
1680         bool dropped;
1681
1682         __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1683                                         dcc->discard_granularity);
1684         dpolicy.timeout = UMOUNT_DISCARD_TIMEOUT;
1685         __issue_discard_cmd(sbi, &dpolicy);
1686         dropped = __drop_discard_cmd(sbi);
1687
1688         /* just to make sure there is no pending discard commands */
1689         __wait_all_discard_cmd(sbi, NULL);
1690
1691         f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1692         return dropped;
1693 }
1694
1695 static int issue_discard_thread(void *data)
1696 {
1697         struct f2fs_sb_info *sbi = data;
1698         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1699         wait_queue_head_t *q = &dcc->discard_wait_queue;
1700         struct discard_policy dpolicy;
1701         unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1702         int issued;
1703
1704         set_freezable();
1705
1706         do {
1707                 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1708                                         dcc->discard_granularity);
1709
1710                 wait_event_interruptible_timeout(*q,
1711                                 kthread_should_stop() || freezing(current) ||
1712                                 dcc->discard_wake,
1713                                 msecs_to_jiffies(wait_ms));
1714
1715                 if (dcc->discard_wake)
1716                         dcc->discard_wake = 0;
1717
1718                 /* clean up pending candidates before going to sleep */
1719                 if (atomic_read(&dcc->queued_discard))
1720                         __wait_all_discard_cmd(sbi, NULL);
1721
1722                 if (try_to_freeze())
1723                         continue;
1724                 if (f2fs_readonly(sbi->sb))
1725                         continue;
1726                 if (kthread_should_stop())
1727                         return 0;
1728                 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1729                         wait_ms = dpolicy.max_interval;
1730                         continue;
1731                 }
1732
1733                 if (sbi->gc_mode == GC_URGENT)
1734                         __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1735
1736                 sb_start_intwrite(sbi->sb);
1737
1738                 issued = __issue_discard_cmd(sbi, &dpolicy);
1739                 if (issued > 0) {
1740                         __wait_all_discard_cmd(sbi, &dpolicy);
1741                         wait_ms = dpolicy.min_interval;
1742                 } else if (issued == -1){
1743                         wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1744                         if (!wait_ms)
1745                                 wait_ms = dpolicy.mid_interval;
1746                 } else {
1747                         wait_ms = dpolicy.max_interval;
1748                 }
1749
1750                 sb_end_intwrite(sbi->sb);
1751
1752         } while (!kthread_should_stop());
1753         return 0;
1754 }
1755
1756 #ifdef CONFIG_BLK_DEV_ZONED
1757 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1758                 struct block_device *bdev, block_t blkstart, block_t blklen)
1759 {
1760         sector_t sector, nr_sects;
1761         block_t lblkstart = blkstart;
1762         int devi = 0;
1763
1764         if (f2fs_is_multi_device(sbi)) {
1765                 devi = f2fs_target_device_index(sbi, blkstart);
1766                 if (blkstart < FDEV(devi).start_blk ||
1767                     blkstart > FDEV(devi).end_blk) {
1768                         f2fs_err(sbi, "Invalid block %x", blkstart);
1769                         return -EIO;
1770                 }
1771                 blkstart -= FDEV(devi).start_blk;
1772         }
1773
1774         /* For sequential zones, reset the zone write pointer */
1775         if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1776                 sector = SECTOR_FROM_BLOCK(blkstart);
1777                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1778
1779                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1780                                 nr_sects != bdev_zone_sectors(bdev)) {
1781                         f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1782                                  devi, sbi->s_ndevs ? FDEV(devi).path : "",
1783                                  blkstart, blklen);
1784                         return -EIO;
1785                 }
1786                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1787                 return blkdev_reset_zones(bdev, sector, nr_sects, GFP_NOFS);
1788         }
1789
1790         /* For conventional zones, use regular discard if supported */
1791         return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1792 }
1793 #endif
1794
1795 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1796                 struct block_device *bdev, block_t blkstart, block_t blklen)
1797 {
1798 #ifdef CONFIG_BLK_DEV_ZONED
1799         if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1800                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1801 #endif
1802         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1803 }
1804
1805 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1806                                 block_t blkstart, block_t blklen)
1807 {
1808         sector_t start = blkstart, len = 0;
1809         struct block_device *bdev;
1810         struct seg_entry *se;
1811         unsigned int offset;
1812         block_t i;
1813         int err = 0;
1814
1815         bdev = f2fs_target_device(sbi, blkstart, NULL);
1816
1817         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1818                 if (i != start) {
1819                         struct block_device *bdev2 =
1820                                 f2fs_target_device(sbi, i, NULL);
1821
1822                         if (bdev2 != bdev) {
1823                                 err = __issue_discard_async(sbi, bdev,
1824                                                 start, len);
1825                                 if (err)
1826                                         return err;
1827                                 bdev = bdev2;
1828                                 start = i;
1829                                 len = 0;
1830                         }
1831                 }
1832
1833                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1834                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1835
1836                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1837                         sbi->discard_blks--;
1838         }
1839
1840         if (len)
1841                 err = __issue_discard_async(sbi, bdev, start, len);
1842         return err;
1843 }
1844
1845 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1846                                                         bool check_only)
1847 {
1848         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1849         int max_blocks = sbi->blocks_per_seg;
1850         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1851         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1852         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1853         unsigned long *discard_map = (unsigned long *)se->discard_map;
1854         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1855         unsigned int start = 0, end = -1;
1856         bool force = (cpc->reason & CP_DISCARD);
1857         struct discard_entry *de = NULL;
1858         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1859         int i;
1860
1861         if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
1862                 return false;
1863
1864         if (!force) {
1865                 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1866                         SM_I(sbi)->dcc_info->nr_discards >=
1867                                 SM_I(sbi)->dcc_info->max_discards)
1868                         return false;
1869         }
1870
1871         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1872         for (i = 0; i < entries; i++)
1873                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1874                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1875
1876         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1877                                 SM_I(sbi)->dcc_info->max_discards) {
1878                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1879                 if (start >= max_blocks)
1880                         break;
1881
1882                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1883                 if (force && start && end != max_blocks
1884                                         && (end - start) < cpc->trim_minlen)
1885                         continue;
1886
1887                 if (check_only)
1888                         return true;
1889
1890                 if (!de) {
1891                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1892                                                                 GFP_F2FS_ZERO);
1893                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1894                         list_add_tail(&de->list, head);
1895                 }
1896
1897                 for (i = start; i < end; i++)
1898                         __set_bit_le(i, (void *)de->discard_map);
1899
1900                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1901         }
1902         return false;
1903 }
1904
1905 static void release_discard_addr(struct discard_entry *entry)
1906 {
1907         list_del(&entry->list);
1908         kmem_cache_free(discard_entry_slab, entry);
1909 }
1910
1911 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1912 {
1913         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1914         struct discard_entry *entry, *this;
1915
1916         /* drop caches */
1917         list_for_each_entry_safe(entry, this, head, list)
1918                 release_discard_addr(entry);
1919 }
1920
1921 /*
1922  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1923  */
1924 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1925 {
1926         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1927         unsigned int segno;
1928
1929         mutex_lock(&dirty_i->seglist_lock);
1930         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1931                 __set_test_and_free(sbi, segno);
1932         mutex_unlock(&dirty_i->seglist_lock);
1933 }
1934
1935 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1936                                                 struct cp_control *cpc)
1937 {
1938         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1939         struct list_head *head = &dcc->entry_list;
1940         struct discard_entry *entry, *this;
1941         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1942         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1943         unsigned int start = 0, end = -1;
1944         unsigned int secno, start_segno;
1945         bool force = (cpc->reason & CP_DISCARD);
1946         bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
1947
1948         mutex_lock(&dirty_i->seglist_lock);
1949
1950         while (1) {
1951                 int i;
1952
1953                 if (need_align && end != -1)
1954                         end--;
1955                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1956                 if (start >= MAIN_SEGS(sbi))
1957                         break;
1958                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1959                                                                 start + 1);
1960
1961                 if (need_align) {
1962                         start = rounddown(start, sbi->segs_per_sec);
1963                         end = roundup(end, sbi->segs_per_sec);
1964                 }
1965
1966                 for (i = start; i < end; i++) {
1967                         if (test_and_clear_bit(i, prefree_map))
1968                                 dirty_i->nr_dirty[PRE]--;
1969                 }
1970
1971                 if (!f2fs_realtime_discard_enable(sbi))
1972                         continue;
1973
1974                 if (force && start >= cpc->trim_start &&
1975                                         (end - 1) <= cpc->trim_end)
1976                                 continue;
1977
1978                 if (!test_opt(sbi, LFS) || !__is_large_section(sbi)) {
1979                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1980                                 (end - start) << sbi->log_blocks_per_seg);
1981                         continue;
1982                 }
1983 next:
1984                 secno = GET_SEC_FROM_SEG(sbi, start);
1985                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1986                 if (!IS_CURSEC(sbi, secno) &&
1987                         !get_valid_blocks(sbi, start, true))
1988                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1989                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1990
1991                 start = start_segno + sbi->segs_per_sec;
1992                 if (start < end)
1993                         goto next;
1994                 else
1995                         end = start - 1;
1996         }
1997         mutex_unlock(&dirty_i->seglist_lock);
1998
1999         /* send small discards */
2000         list_for_each_entry_safe(entry, this, head, list) {
2001                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2002                 bool is_valid = test_bit_le(0, entry->discard_map);
2003
2004 find_next:
2005                 if (is_valid) {
2006                         next_pos = find_next_zero_bit_le(entry->discard_map,
2007                                         sbi->blocks_per_seg, cur_pos);
2008                         len = next_pos - cur_pos;
2009
2010                         if (f2fs_sb_has_blkzoned(sbi) ||
2011                             (force && len < cpc->trim_minlen))
2012                                 goto skip;
2013
2014                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2015                                                                         len);
2016                         total_len += len;
2017                 } else {
2018                         next_pos = find_next_bit_le(entry->discard_map,
2019                                         sbi->blocks_per_seg, cur_pos);
2020                 }
2021 skip:
2022                 cur_pos = next_pos;
2023                 is_valid = !is_valid;
2024
2025                 if (cur_pos < sbi->blocks_per_seg)
2026                         goto find_next;
2027
2028                 release_discard_addr(entry);
2029                 dcc->nr_discards -= total_len;
2030         }
2031
2032         wake_up_discard_thread(sbi, false);
2033 }
2034
2035 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2036 {
2037         dev_t dev = sbi->sb->s_bdev->bd_dev;
2038         struct discard_cmd_control *dcc;
2039         int err = 0, i;
2040
2041         if (SM_I(sbi)->dcc_info) {
2042                 dcc = SM_I(sbi)->dcc_info;
2043                 goto init_thread;
2044         }
2045
2046         dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2047         if (!dcc)
2048                 return -ENOMEM;
2049
2050         dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2051         INIT_LIST_HEAD(&dcc->entry_list);
2052         for (i = 0; i < MAX_PLIST_NUM; i++)
2053                 INIT_LIST_HEAD(&dcc->pend_list[i]);
2054         INIT_LIST_HEAD(&dcc->wait_list);
2055         INIT_LIST_HEAD(&dcc->fstrim_list);
2056         mutex_init(&dcc->cmd_lock);
2057         atomic_set(&dcc->issued_discard, 0);
2058         atomic_set(&dcc->queued_discard, 0);
2059         atomic_set(&dcc->discard_cmd_cnt, 0);
2060         dcc->nr_discards = 0;
2061         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2062         dcc->undiscard_blks = 0;
2063         dcc->next_pos = 0;
2064         dcc->root = RB_ROOT_CACHED;
2065         dcc->rbtree_check = false;
2066
2067         init_waitqueue_head(&dcc->discard_wait_queue);
2068         SM_I(sbi)->dcc_info = dcc;
2069 init_thread:
2070         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2071                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2072         if (IS_ERR(dcc->f2fs_issue_discard)) {
2073                 err = PTR_ERR(dcc->f2fs_issue_discard);
2074                 kvfree(dcc);
2075                 SM_I(sbi)->dcc_info = NULL;
2076                 return err;
2077         }
2078
2079         return err;
2080 }
2081
2082 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2083 {
2084         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2085
2086         if (!dcc)
2087                 return;
2088
2089         f2fs_stop_discard_thread(sbi);
2090
2091         /*
2092          * Recovery can cache discard commands, so in error path of
2093          * fill_super(), it needs to give a chance to handle them.
2094          */
2095         if (unlikely(atomic_read(&dcc->discard_cmd_cnt)))
2096                 f2fs_issue_discard_timeout(sbi);
2097
2098         kvfree(dcc);
2099         SM_I(sbi)->dcc_info = NULL;
2100 }
2101
2102 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2103 {
2104         struct sit_info *sit_i = SIT_I(sbi);
2105
2106         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2107                 sit_i->dirty_sentries++;
2108                 return false;
2109         }
2110
2111         return true;
2112 }
2113
2114 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2115                                         unsigned int segno, int modified)
2116 {
2117         struct seg_entry *se = get_seg_entry(sbi, segno);
2118         se->type = type;
2119         if (modified)
2120                 __mark_sit_entry_dirty(sbi, segno);
2121 }
2122
2123 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2124 {
2125         struct seg_entry *se;
2126         unsigned int segno, offset;
2127         long int new_vblocks;
2128         bool exist;
2129 #ifdef CONFIG_F2FS_CHECK_FS
2130         bool mir_exist;
2131 #endif
2132
2133         segno = GET_SEGNO(sbi, blkaddr);
2134
2135         se = get_seg_entry(sbi, segno);
2136         new_vblocks = se->valid_blocks + del;
2137         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2138
2139         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
2140                                 (new_vblocks > sbi->blocks_per_seg)));
2141
2142         se->valid_blocks = new_vblocks;
2143         se->mtime = get_mtime(sbi, false);
2144         if (se->mtime > SIT_I(sbi)->max_mtime)
2145                 SIT_I(sbi)->max_mtime = se->mtime;
2146
2147         /* Update valid block bitmap */
2148         if (del > 0) {
2149                 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2150 #ifdef CONFIG_F2FS_CHECK_FS
2151                 mir_exist = f2fs_test_and_set_bit(offset,
2152                                                 se->cur_valid_map_mir);
2153                 if (unlikely(exist != mir_exist)) {
2154                         f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2155                                  blkaddr, exist);
2156                         f2fs_bug_on(sbi, 1);
2157                 }
2158 #endif
2159                 if (unlikely(exist)) {
2160                         f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2161                                  blkaddr);
2162                         f2fs_bug_on(sbi, 1);
2163                         se->valid_blocks--;
2164                         del = 0;
2165                 }
2166
2167                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
2168                         sbi->discard_blks--;
2169
2170                 /*
2171                  * SSR should never reuse block which is checkpointed
2172                  * or newly invalidated.
2173                  */
2174                 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2175                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2176                                 se->ckpt_valid_blocks++;
2177                 }
2178         } else {
2179                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2180 #ifdef CONFIG_F2FS_CHECK_FS
2181                 mir_exist = f2fs_test_and_clear_bit(offset,
2182                                                 se->cur_valid_map_mir);
2183                 if (unlikely(exist != mir_exist)) {
2184                         f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2185                                  blkaddr, exist);
2186                         f2fs_bug_on(sbi, 1);
2187                 }
2188 #endif
2189                 if (unlikely(!exist)) {
2190                         f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2191                                  blkaddr);
2192                         f2fs_bug_on(sbi, 1);
2193                         se->valid_blocks++;
2194                         del = 0;
2195                 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2196                         /*
2197                          * If checkpoints are off, we must not reuse data that
2198                          * was used in the previous checkpoint. If it was used
2199                          * before, we must track that to know how much space we
2200                          * really have.
2201                          */
2202                         if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2203                                 spin_lock(&sbi->stat_lock);
2204                                 sbi->unusable_block_count++;
2205                                 spin_unlock(&sbi->stat_lock);
2206                         }
2207                 }
2208
2209                 if (f2fs_test_and_clear_bit(offset, se->discard_map))
2210                         sbi->discard_blks++;
2211         }
2212         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2213                 se->ckpt_valid_blocks += del;
2214
2215         __mark_sit_entry_dirty(sbi, segno);
2216
2217         /* update total number of valid blocks to be written in ckpt area */
2218         SIT_I(sbi)->written_valid_blocks += del;
2219
2220         if (__is_large_section(sbi))
2221                 get_sec_entry(sbi, segno)->valid_blocks += del;
2222 }
2223
2224 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2225 {
2226         unsigned int segno = GET_SEGNO(sbi, addr);
2227         struct sit_info *sit_i = SIT_I(sbi);
2228
2229         f2fs_bug_on(sbi, addr == NULL_ADDR);
2230         if (addr == NEW_ADDR)
2231                 return;
2232
2233         invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2234
2235         /* add it into sit main buffer */
2236         down_write(&sit_i->sentry_lock);
2237
2238         update_sit_entry(sbi, addr, -1);
2239
2240         /* add it into dirty seglist */
2241         locate_dirty_segment(sbi, segno);
2242
2243         up_write(&sit_i->sentry_lock);
2244 }
2245
2246 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2247 {
2248         struct sit_info *sit_i = SIT_I(sbi);
2249         unsigned int segno, offset;
2250         struct seg_entry *se;
2251         bool is_cp = false;
2252
2253         if (!__is_valid_data_blkaddr(blkaddr))
2254                 return true;
2255
2256         down_read(&sit_i->sentry_lock);
2257
2258         segno = GET_SEGNO(sbi, blkaddr);
2259         se = get_seg_entry(sbi, segno);
2260         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2261
2262         if (f2fs_test_bit(offset, se->ckpt_valid_map))
2263                 is_cp = true;
2264
2265         up_read(&sit_i->sentry_lock);
2266
2267         return is_cp;
2268 }
2269
2270 /*
2271  * This function should be resided under the curseg_mutex lock
2272  */
2273 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2274                                         struct f2fs_summary *sum)
2275 {
2276         struct curseg_info *curseg = CURSEG_I(sbi, type);
2277         void *addr = curseg->sum_blk;
2278         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2279         memcpy(addr, sum, sizeof(struct f2fs_summary));
2280 }
2281
2282 /*
2283  * Calculate the number of current summary pages for writing
2284  */
2285 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2286 {
2287         int valid_sum_count = 0;
2288         int i, sum_in_page;
2289
2290         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2291                 if (sbi->ckpt->alloc_type[i] == SSR)
2292                         valid_sum_count += sbi->blocks_per_seg;
2293                 else {
2294                         if (for_ra)
2295                                 valid_sum_count += le16_to_cpu(
2296                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2297                         else
2298                                 valid_sum_count += curseg_blkoff(sbi, i);
2299                 }
2300         }
2301
2302         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2303                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2304         if (valid_sum_count <= sum_in_page)
2305                 return 1;
2306         else if ((valid_sum_count - sum_in_page) <=
2307                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2308                 return 2;
2309         return 3;
2310 }
2311
2312 /*
2313  * Caller should put this summary page
2314  */
2315 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2316 {
2317         return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
2318 }
2319
2320 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2321                                         void *src, block_t blk_addr)
2322 {
2323         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2324
2325         memcpy(page_address(page), src, PAGE_SIZE);
2326         set_page_dirty(page);
2327         f2fs_put_page(page, 1);
2328 }
2329
2330 static void write_sum_page(struct f2fs_sb_info *sbi,
2331                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
2332 {
2333         f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2334 }
2335
2336 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2337                                                 int type, block_t blk_addr)
2338 {
2339         struct curseg_info *curseg = CURSEG_I(sbi, type);
2340         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2341         struct f2fs_summary_block *src = curseg->sum_blk;
2342         struct f2fs_summary_block *dst;
2343
2344         dst = (struct f2fs_summary_block *)page_address(page);
2345         memset(dst, 0, PAGE_SIZE);
2346
2347         mutex_lock(&curseg->curseg_mutex);
2348
2349         down_read(&curseg->journal_rwsem);
2350         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2351         up_read(&curseg->journal_rwsem);
2352
2353         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2354         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2355
2356         mutex_unlock(&curseg->curseg_mutex);
2357
2358         set_page_dirty(page);
2359         f2fs_put_page(page, 1);
2360 }
2361
2362 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2363 {
2364         struct curseg_info *curseg = CURSEG_I(sbi, type);
2365         unsigned int segno = curseg->segno + 1;
2366         struct free_segmap_info *free_i = FREE_I(sbi);
2367
2368         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2369                 return !test_bit(segno, free_i->free_segmap);
2370         return 0;
2371 }
2372
2373 /*
2374  * Find a new segment from the free segments bitmap to right order
2375  * This function should be returned with success, otherwise BUG
2376  */
2377 static void get_new_segment(struct f2fs_sb_info *sbi,
2378                         unsigned int *newseg, bool new_sec, int dir)
2379 {
2380         struct free_segmap_info *free_i = FREE_I(sbi);
2381         unsigned int segno, secno, zoneno;
2382         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2383         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2384         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2385         unsigned int left_start = hint;
2386         bool init = true;
2387         int go_left = 0;
2388         int i;
2389
2390         spin_lock(&free_i->segmap_lock);
2391
2392         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2393                 segno = find_next_zero_bit(free_i->free_segmap,
2394                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2395                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2396                         goto got_it;
2397         }
2398 find_other_zone:
2399         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2400         if (secno >= MAIN_SECS(sbi)) {
2401                 if (dir == ALLOC_RIGHT) {
2402                         secno = find_next_zero_bit(free_i->free_secmap,
2403                                                         MAIN_SECS(sbi), 0);
2404                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2405                 } else {
2406                         go_left = 1;
2407                         left_start = hint - 1;
2408                 }
2409         }
2410         if (go_left == 0)
2411                 goto skip_left;
2412
2413         while (test_bit(left_start, free_i->free_secmap)) {
2414                 if (left_start > 0) {
2415                         left_start--;
2416                         continue;
2417                 }
2418                 left_start = find_next_zero_bit(free_i->free_secmap,
2419                                                         MAIN_SECS(sbi), 0);
2420                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2421                 break;
2422         }
2423         secno = left_start;
2424 skip_left:
2425         segno = GET_SEG_FROM_SEC(sbi, secno);
2426         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2427
2428         /* give up on finding another zone */
2429         if (!init)
2430                 goto got_it;
2431         if (sbi->secs_per_zone == 1)
2432                 goto got_it;
2433         if (zoneno == old_zoneno)
2434                 goto got_it;
2435         if (dir == ALLOC_LEFT) {
2436                 if (!go_left && zoneno + 1 >= total_zones)
2437                         goto got_it;
2438                 if (go_left && zoneno == 0)
2439                         goto got_it;
2440         }
2441         for (i = 0; i < NR_CURSEG_TYPE; i++)
2442                 if (CURSEG_I(sbi, i)->zone == zoneno)
2443                         break;
2444
2445         if (i < NR_CURSEG_TYPE) {
2446                 /* zone is in user, try another */
2447                 if (go_left)
2448                         hint = zoneno * sbi->secs_per_zone - 1;
2449                 else if (zoneno + 1 >= total_zones)
2450                         hint = 0;
2451                 else
2452                         hint = (zoneno + 1) * sbi->secs_per_zone;
2453                 init = false;
2454                 goto find_other_zone;
2455         }
2456 got_it:
2457         /* set it as dirty segment in free segmap */
2458         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2459         __set_inuse(sbi, segno);
2460         *newseg = segno;
2461         spin_unlock(&free_i->segmap_lock);
2462 }
2463
2464 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2465 {
2466         struct curseg_info *curseg = CURSEG_I(sbi, type);
2467         struct summary_footer *sum_footer;
2468
2469         curseg->segno = curseg->next_segno;
2470         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2471         curseg->next_blkoff = 0;
2472         curseg->next_segno = NULL_SEGNO;
2473
2474         sum_footer = &(curseg->sum_blk->footer);
2475         memset(sum_footer, 0, sizeof(struct summary_footer));
2476         if (IS_DATASEG(type))
2477                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2478         if (IS_NODESEG(type))
2479                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2480         __set_sit_entry_type(sbi, type, curseg->segno, modified);
2481 }
2482
2483 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2484 {
2485         /* if segs_per_sec is large than 1, we need to keep original policy. */
2486         if (__is_large_section(sbi))
2487                 return CURSEG_I(sbi, type)->segno;
2488
2489         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2490                 return 0;
2491
2492         if (test_opt(sbi, NOHEAP) &&
2493                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2494                 return 0;
2495
2496         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2497                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2498
2499         /* find segments from 0 to reuse freed segments */
2500         if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2501                 return 0;
2502
2503         return CURSEG_I(sbi, type)->segno;
2504 }
2505
2506 /*
2507  * Allocate a current working segment.
2508  * This function always allocates a free segment in LFS manner.
2509  */
2510 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2511 {
2512         struct curseg_info *curseg = CURSEG_I(sbi, type);
2513         unsigned int segno = curseg->segno;
2514         int dir = ALLOC_LEFT;
2515
2516         write_sum_page(sbi, curseg->sum_blk,
2517                                 GET_SUM_BLOCK(sbi, segno));
2518         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2519                 dir = ALLOC_RIGHT;
2520
2521         if (test_opt(sbi, NOHEAP))
2522                 dir = ALLOC_RIGHT;
2523
2524         segno = __get_next_segno(sbi, type);
2525         get_new_segment(sbi, &segno, new_sec, dir);
2526         curseg->next_segno = segno;
2527         reset_curseg(sbi, type, 1);
2528         curseg->alloc_type = LFS;
2529 }
2530
2531 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2532                         struct curseg_info *seg, block_t start)
2533 {
2534         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2535         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2536         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2537         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2538         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2539         int i, pos;
2540
2541         for (i = 0; i < entries; i++)
2542                 target_map[i] = ckpt_map[i] | cur_map[i];
2543
2544         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2545
2546         seg->next_blkoff = pos;
2547 }
2548
2549 /*
2550  * If a segment is written by LFS manner, next block offset is just obtained
2551  * by increasing the current block offset. However, if a segment is written by
2552  * SSR manner, next block offset obtained by calling __next_free_blkoff
2553  */
2554 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2555                                 struct curseg_info *seg)
2556 {
2557         if (seg->alloc_type == SSR)
2558                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2559         else
2560                 seg->next_blkoff++;
2561 }
2562
2563 /*
2564  * This function always allocates a used segment(from dirty seglist) by SSR
2565  * manner, so it should recover the existing segment information of valid blocks
2566  */
2567 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2568 {
2569         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2570         struct curseg_info *curseg = CURSEG_I(sbi, type);
2571         unsigned int new_segno = curseg->next_segno;
2572         struct f2fs_summary_block *sum_node;
2573         struct page *sum_page;
2574
2575         write_sum_page(sbi, curseg->sum_blk,
2576                                 GET_SUM_BLOCK(sbi, curseg->segno));
2577         __set_test_and_inuse(sbi, new_segno);
2578
2579         mutex_lock(&dirty_i->seglist_lock);
2580         __remove_dirty_segment(sbi, new_segno, PRE);
2581         __remove_dirty_segment(sbi, new_segno, DIRTY);
2582         mutex_unlock(&dirty_i->seglist_lock);
2583
2584         reset_curseg(sbi, type, 1);
2585         curseg->alloc_type = SSR;
2586         __next_free_blkoff(sbi, curseg, 0);
2587
2588         sum_page = f2fs_get_sum_page(sbi, new_segno);
2589         f2fs_bug_on(sbi, IS_ERR(sum_page));
2590         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2591         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2592         f2fs_put_page(sum_page, 1);
2593 }
2594
2595 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2596 {
2597         struct curseg_info *curseg = CURSEG_I(sbi, type);
2598         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2599         unsigned segno = NULL_SEGNO;
2600         int i, cnt;
2601         bool reversed = false;
2602
2603         /* f2fs_need_SSR() already forces to do this */
2604         if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2605                 curseg->next_segno = segno;
2606                 return 1;
2607         }
2608
2609         /* For node segments, let's do SSR more intensively */
2610         if (IS_NODESEG(type)) {
2611                 if (type >= CURSEG_WARM_NODE) {
2612                         reversed = true;
2613                         i = CURSEG_COLD_NODE;
2614                 } else {
2615                         i = CURSEG_HOT_NODE;
2616                 }
2617                 cnt = NR_CURSEG_NODE_TYPE;
2618         } else {
2619                 if (type >= CURSEG_WARM_DATA) {
2620                         reversed = true;
2621                         i = CURSEG_COLD_DATA;
2622                 } else {
2623                         i = CURSEG_HOT_DATA;
2624                 }
2625                 cnt = NR_CURSEG_DATA_TYPE;
2626         }
2627
2628         for (; cnt-- > 0; reversed ? i-- : i++) {
2629                 if (i == type)
2630                         continue;
2631                 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2632                         curseg->next_segno = segno;
2633                         return 1;
2634                 }
2635         }
2636
2637         /* find valid_blocks=0 in dirty list */
2638         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2639                 segno = get_free_segment(sbi);
2640                 if (segno != NULL_SEGNO) {
2641                         curseg->next_segno = segno;
2642                         return 1;
2643                 }
2644         }
2645         return 0;
2646 }
2647
2648 /*
2649  * flush out current segment and replace it with new segment
2650  * This function should be returned with success, otherwise BUG
2651  */
2652 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2653                                                 int type, bool force)
2654 {
2655         struct curseg_info *curseg = CURSEG_I(sbi, type);
2656
2657         if (force)
2658                 new_curseg(sbi, type, true);
2659         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2660                                         type == CURSEG_WARM_NODE)
2661                 new_curseg(sbi, type, false);
2662         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2663                         likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2664                 new_curseg(sbi, type, false);
2665         else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2666                 change_curseg(sbi, type);
2667         else
2668                 new_curseg(sbi, type, false);
2669
2670         stat_inc_seg_type(sbi, curseg);
2671 }
2672
2673 void allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2674                                         unsigned int start, unsigned int end)
2675 {
2676         struct curseg_info *curseg = CURSEG_I(sbi, type);
2677         unsigned int segno;
2678
2679         down_read(&SM_I(sbi)->curseg_lock);
2680         mutex_lock(&curseg->curseg_mutex);
2681         down_write(&SIT_I(sbi)->sentry_lock);
2682
2683         segno = CURSEG_I(sbi, type)->segno;
2684         if (segno < start || segno > end)
2685                 goto unlock;
2686
2687         if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2688                 change_curseg(sbi, type);
2689         else
2690                 new_curseg(sbi, type, true);
2691
2692         stat_inc_seg_type(sbi, curseg);
2693
2694         locate_dirty_segment(sbi, segno);
2695 unlock:
2696         up_write(&SIT_I(sbi)->sentry_lock);
2697
2698         if (segno != curseg->segno)
2699                 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2700                             type, segno, curseg->segno);
2701
2702         mutex_unlock(&curseg->curseg_mutex);
2703         up_read(&SM_I(sbi)->curseg_lock);
2704 }
2705
2706 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2707 {
2708         struct curseg_info *curseg;
2709         unsigned int old_segno;
2710         int i;
2711
2712         down_write(&SIT_I(sbi)->sentry_lock);
2713
2714         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2715                 curseg = CURSEG_I(sbi, i);
2716                 old_segno = curseg->segno;
2717                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2718                 locate_dirty_segment(sbi, old_segno);
2719         }
2720
2721         up_write(&SIT_I(sbi)->sentry_lock);
2722 }
2723
2724 static const struct segment_allocation default_salloc_ops = {
2725         .allocate_segment = allocate_segment_by_default,
2726 };
2727
2728 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2729                                                 struct cp_control *cpc)
2730 {
2731         __u64 trim_start = cpc->trim_start;
2732         bool has_candidate = false;
2733
2734         down_write(&SIT_I(sbi)->sentry_lock);
2735         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2736                 if (add_discard_addrs(sbi, cpc, true)) {
2737                         has_candidate = true;
2738                         break;
2739                 }
2740         }
2741         up_write(&SIT_I(sbi)->sentry_lock);
2742
2743         cpc->trim_start = trim_start;
2744         return has_candidate;
2745 }
2746
2747 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2748                                         struct discard_policy *dpolicy,
2749                                         unsigned int start, unsigned int end)
2750 {
2751         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2752         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2753         struct rb_node **insert_p = NULL, *insert_parent = NULL;
2754         struct discard_cmd *dc;
2755         struct blk_plug plug;
2756         int issued;
2757         unsigned int trimmed = 0;
2758
2759 next:
2760         issued = 0;
2761
2762         mutex_lock(&dcc->cmd_lock);
2763         if (unlikely(dcc->rbtree_check))
2764                 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2765                                                                 &dcc->root));
2766
2767         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2768                                         NULL, start,
2769                                         (struct rb_entry **)&prev_dc,
2770                                         (struct rb_entry **)&next_dc,
2771                                         &insert_p, &insert_parent, true, NULL);
2772         if (!dc)
2773                 dc = next_dc;
2774
2775         blk_start_plug(&plug);
2776
2777         while (dc && dc->lstart <= end) {
2778                 struct rb_node *node;
2779                 int err = 0;
2780
2781                 if (dc->len < dpolicy->granularity)
2782                         goto skip;
2783
2784                 if (dc->state != D_PREP) {
2785                         list_move_tail(&dc->list, &dcc->fstrim_list);
2786                         goto skip;
2787                 }
2788
2789                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2790
2791                 if (issued >= dpolicy->max_requests) {
2792                         start = dc->lstart + dc->len;
2793
2794                         if (err)
2795                                 __remove_discard_cmd(sbi, dc);
2796
2797                         blk_finish_plug(&plug);
2798                         mutex_unlock(&dcc->cmd_lock);
2799                         trimmed += __wait_all_discard_cmd(sbi, NULL);
2800                         congestion_wait(BLK_RW_ASYNC, HZ/50);
2801                         goto next;
2802                 }
2803 skip:
2804                 node = rb_next(&dc->rb_node);
2805                 if (err)
2806                         __remove_discard_cmd(sbi, dc);
2807                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2808
2809                 if (fatal_signal_pending(current))
2810                         break;
2811         }
2812
2813         blk_finish_plug(&plug);
2814         mutex_unlock(&dcc->cmd_lock);
2815
2816         return trimmed;
2817 }
2818
2819 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2820 {
2821         __u64 start = F2FS_BYTES_TO_BLK(range->start);
2822         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2823         unsigned int start_segno, end_segno;
2824         block_t start_block, end_block;
2825         struct cp_control cpc;
2826         struct discard_policy dpolicy;
2827         unsigned long long trimmed = 0;
2828         int err = 0;
2829         bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
2830
2831         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2832                 return -EINVAL;
2833
2834         if (end < MAIN_BLKADDR(sbi))
2835                 goto out;
2836
2837         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2838                 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
2839                 return -EFSCORRUPTED;
2840         }
2841
2842         /* start/end segment number in main_area */
2843         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2844         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2845                                                 GET_SEGNO(sbi, end);
2846         if (need_align) {
2847                 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2848                 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2849         }
2850
2851         cpc.reason = CP_DISCARD;
2852         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2853         cpc.trim_start = start_segno;
2854         cpc.trim_end = end_segno;
2855
2856         if (sbi->discard_blks == 0)
2857                 goto out;
2858
2859         mutex_lock(&sbi->gc_mutex);
2860         err = f2fs_write_checkpoint(sbi, &cpc);
2861         mutex_unlock(&sbi->gc_mutex);
2862         if (err)
2863                 goto out;
2864
2865         /*
2866          * We filed discard candidates, but actually we don't need to wait for
2867          * all of them, since they'll be issued in idle time along with runtime
2868          * discard option. User configuration looks like using runtime discard
2869          * or periodic fstrim instead of it.
2870          */
2871         if (f2fs_realtime_discard_enable(sbi))
2872                 goto out;
2873
2874         start_block = START_BLOCK(sbi, start_segno);
2875         end_block = START_BLOCK(sbi, end_segno + 1);
2876
2877         __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2878         trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2879                                         start_block, end_block);
2880
2881         trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
2882                                         start_block, end_block);
2883 out:
2884         if (!err)
2885                 range->len = F2FS_BLK_TO_BYTES(trimmed);
2886         return err;
2887 }
2888
2889 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2890 {
2891         struct curseg_info *curseg = CURSEG_I(sbi, type);
2892         if (curseg->next_blkoff < sbi->blocks_per_seg)
2893                 return true;
2894         return false;
2895 }
2896
2897 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
2898 {
2899         switch (hint) {
2900         case WRITE_LIFE_SHORT:
2901                 return CURSEG_HOT_DATA;
2902         case WRITE_LIFE_EXTREME:
2903                 return CURSEG_COLD_DATA;
2904         default:
2905                 return CURSEG_WARM_DATA;
2906         }
2907 }
2908
2909 /* This returns write hints for each segment type. This hints will be
2910  * passed down to block layer. There are mapping tables which depend on
2911  * the mount option 'whint_mode'.
2912  *
2913  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2914  *
2915  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2916  *
2917  * User                  F2FS                     Block
2918  * ----                  ----                     -----
2919  *                       META                     WRITE_LIFE_NOT_SET
2920  *                       HOT_NODE                 "
2921  *                       WARM_NODE                "
2922  *                       COLD_NODE                "
2923  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2924  * extension list        "                        "
2925  *
2926  * -- buffered io
2927  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2928  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2929  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2930  * WRITE_LIFE_NONE       "                        "
2931  * WRITE_LIFE_MEDIUM     "                        "
2932  * WRITE_LIFE_LONG       "                        "
2933  *
2934  * -- direct io
2935  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2936  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2937  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2938  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2939  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2940  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2941  *
2942  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2943  *
2944  * User                  F2FS                     Block
2945  * ----                  ----                     -----
2946  *                       META                     WRITE_LIFE_MEDIUM;
2947  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
2948  *                       WARM_NODE                "
2949  *                       COLD_NODE                WRITE_LIFE_NONE
2950  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2951  * extension list        "                        "
2952  *
2953  * -- buffered io
2954  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2955  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2956  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
2957  * WRITE_LIFE_NONE       "                        "
2958  * WRITE_LIFE_MEDIUM     "                        "
2959  * WRITE_LIFE_LONG       "                        "
2960  *
2961  * -- direct io
2962  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2963  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2964  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2965  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2966  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2967  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2968  */
2969
2970 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2971                                 enum page_type type, enum temp_type temp)
2972 {
2973         if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2974                 if (type == DATA) {
2975                         if (temp == WARM)
2976                                 return WRITE_LIFE_NOT_SET;
2977                         else if (temp == HOT)
2978                                 return WRITE_LIFE_SHORT;
2979                         else if (temp == COLD)
2980                                 return WRITE_LIFE_EXTREME;
2981                 } else {
2982                         return WRITE_LIFE_NOT_SET;
2983                 }
2984         } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2985                 if (type == DATA) {
2986                         if (temp == WARM)
2987                                 return WRITE_LIFE_LONG;
2988                         else if (temp == HOT)
2989                                 return WRITE_LIFE_SHORT;
2990                         else if (temp == COLD)
2991                                 return WRITE_LIFE_EXTREME;
2992                 } else if (type == NODE) {
2993                         if (temp == WARM || temp == HOT)
2994                                 return WRITE_LIFE_NOT_SET;
2995                         else if (temp == COLD)
2996                                 return WRITE_LIFE_NONE;
2997                 } else if (type == META) {
2998                         return WRITE_LIFE_MEDIUM;
2999                 }
3000         }
3001         return WRITE_LIFE_NOT_SET;
3002 }
3003
3004 static int __get_segment_type_2(struct f2fs_io_info *fio)
3005 {
3006         if (fio->type == DATA)
3007                 return CURSEG_HOT_DATA;
3008         else
3009                 return CURSEG_HOT_NODE;
3010 }
3011
3012 static int __get_segment_type_4(struct f2fs_io_info *fio)
3013 {
3014         if (fio->type == DATA) {
3015                 struct inode *inode = fio->page->mapping->host;
3016
3017                 if (S_ISDIR(inode->i_mode))
3018                         return CURSEG_HOT_DATA;
3019                 else
3020                         return CURSEG_COLD_DATA;
3021         } else {
3022                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3023                         return CURSEG_WARM_NODE;
3024                 else
3025                         return CURSEG_COLD_NODE;
3026         }
3027 }
3028
3029 static int __get_segment_type_6(struct f2fs_io_info *fio)
3030 {
3031         if (fio->type == DATA) {
3032                 struct inode *inode = fio->page->mapping->host;
3033
3034                 if (is_cold_data(fio->page) || file_is_cold(inode))
3035                         return CURSEG_COLD_DATA;
3036                 if (file_is_hot(inode) ||
3037                                 is_inode_flag_set(inode, FI_HOT_DATA) ||
3038                                 f2fs_is_atomic_file(inode) ||
3039                                 f2fs_is_volatile_file(inode))
3040                         return CURSEG_HOT_DATA;
3041                 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3042         } else {
3043                 if (IS_DNODE(fio->page))
3044                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3045                                                 CURSEG_HOT_NODE;
3046                 return CURSEG_COLD_NODE;
3047         }
3048 }
3049
3050 static int __get_segment_type(struct f2fs_io_info *fio)
3051 {
3052         int type = 0;
3053
3054         switch (F2FS_OPTION(fio->sbi).active_logs) {
3055         case 2:
3056                 type = __get_segment_type_2(fio);
3057                 break;
3058         case 4:
3059                 type = __get_segment_type_4(fio);
3060                 break;
3061         case 6:
3062                 type = __get_segment_type_6(fio);
3063                 break;
3064         default:
3065                 f2fs_bug_on(fio->sbi, true);
3066         }
3067
3068         if (IS_HOT(type))
3069                 fio->temp = HOT;
3070         else if (IS_WARM(type))
3071                 fio->temp = WARM;
3072         else
3073                 fio->temp = COLD;
3074         return type;
3075 }
3076
3077 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3078                 block_t old_blkaddr, block_t *new_blkaddr,
3079                 struct f2fs_summary *sum, int type,
3080                 struct f2fs_io_info *fio, bool add_list)
3081 {
3082         struct sit_info *sit_i = SIT_I(sbi);
3083         struct curseg_info *curseg = CURSEG_I(sbi, type);
3084
3085         down_read(&SM_I(sbi)->curseg_lock);
3086
3087         mutex_lock(&curseg->curseg_mutex);
3088         down_write(&sit_i->sentry_lock);
3089
3090         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3091
3092         f2fs_wait_discard_bio(sbi, *new_blkaddr);
3093
3094         /*
3095          * __add_sum_entry should be resided under the curseg_mutex
3096          * because, this function updates a summary entry in the
3097          * current summary block.
3098          */
3099         __add_sum_entry(sbi, type, sum);
3100
3101         __refresh_next_blkoff(sbi, curseg);
3102
3103         stat_inc_block_count(sbi, curseg);
3104
3105         /*
3106          * SIT information should be updated before segment allocation,
3107          * since SSR needs latest valid block information.
3108          */
3109         update_sit_entry(sbi, *new_blkaddr, 1);
3110         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3111                 update_sit_entry(sbi, old_blkaddr, -1);
3112
3113         if (!__has_curseg_space(sbi, type))
3114                 sit_i->s_ops->allocate_segment(sbi, type, false);
3115
3116         /*
3117          * segment dirty status should be updated after segment allocation,
3118          * so we just need to update status only one time after previous
3119          * segment being closed.
3120          */
3121         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3122         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3123
3124         up_write(&sit_i->sentry_lock);
3125
3126         if (page && IS_NODESEG(type)) {
3127                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3128
3129                 f2fs_inode_chksum_set(sbi, page);
3130         }
3131
3132         if (add_list) {
3133                 struct f2fs_bio_info *io;
3134
3135                 INIT_LIST_HEAD(&fio->list);
3136                 fio->in_list = true;
3137                 fio->retry = false;
3138                 io = sbi->write_io[fio->type] + fio->temp;
3139                 spin_lock(&io->io_lock);
3140                 list_add_tail(&fio->list, &io->io_list);
3141                 spin_unlock(&io->io_lock);
3142         }
3143
3144         mutex_unlock(&curseg->curseg_mutex);
3145
3146         up_read(&SM_I(sbi)->curseg_lock);
3147 }
3148
3149 static void update_device_state(struct f2fs_io_info *fio)
3150 {
3151         struct f2fs_sb_info *sbi = fio->sbi;
3152         unsigned int devidx;
3153
3154         if (!f2fs_is_multi_device(sbi))
3155                 return;
3156
3157         devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3158
3159         /* update device state for fsync */
3160         f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
3161
3162         /* update device state for checkpoint */
3163         if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3164                 spin_lock(&sbi->dev_lock);
3165                 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3166                 spin_unlock(&sbi->dev_lock);
3167         }
3168 }
3169
3170 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3171 {
3172         int type = __get_segment_type(fio);
3173         bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
3174
3175         if (keep_order)
3176                 down_read(&fio->sbi->io_order_lock);
3177 reallocate:
3178         f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3179                         &fio->new_blkaddr, sum, type, fio, true);
3180         if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3181                 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3182                                         fio->old_blkaddr, fio->old_blkaddr);
3183
3184         /* writeout dirty page into bdev */
3185         f2fs_submit_page_write(fio);
3186         if (fio->retry) {
3187                 fio->old_blkaddr = fio->new_blkaddr;
3188                 goto reallocate;
3189         }
3190
3191         update_device_state(fio);
3192
3193         if (keep_order)
3194                 up_read(&fio->sbi->io_order_lock);
3195 }
3196
3197 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3198                                         enum iostat_type io_type)
3199 {
3200         struct f2fs_io_info fio = {
3201                 .sbi = sbi,
3202                 .type = META,
3203                 .temp = HOT,
3204                 .op = REQ_OP_WRITE,
3205                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3206                 .old_blkaddr = page->index,
3207                 .new_blkaddr = page->index,
3208                 .page = page,
3209                 .encrypted_page = NULL,
3210                 .in_list = false,
3211         };
3212
3213         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3214                 fio.op_flags &= ~REQ_META;
3215
3216         set_page_writeback(page);
3217         ClearPageError(page);
3218         f2fs_submit_page_write(&fio);
3219
3220         stat_inc_meta_count(sbi, page->index);
3221         f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3222 }
3223
3224 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3225 {
3226         struct f2fs_summary sum;
3227
3228         set_summary(&sum, nid, 0, 0);
3229         do_write_page(&sum, fio);
3230
3231         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3232 }
3233
3234 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3235                                         struct f2fs_io_info *fio)
3236 {
3237         struct f2fs_sb_info *sbi = fio->sbi;
3238         struct f2fs_summary sum;
3239
3240         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3241         set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3242         do_write_page(&sum, fio);
3243         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3244
3245         f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3246 }
3247
3248 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3249 {
3250         int err;
3251         struct f2fs_sb_info *sbi = fio->sbi;
3252         unsigned int segno;
3253
3254         fio->new_blkaddr = fio->old_blkaddr;
3255         /* i/o temperature is needed for passing down write hints */
3256         __get_segment_type(fio);
3257
3258         segno = GET_SEGNO(sbi, fio->new_blkaddr);
3259
3260         if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3261                 set_sbi_flag(sbi, SBI_NEED_FSCK);
3262                 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3263                           __func__, segno);
3264                 return -EFSCORRUPTED;
3265         }
3266
3267         stat_inc_inplace_blocks(fio->sbi);
3268
3269         if (fio->bio)
3270                 err = f2fs_merge_page_bio(fio);
3271         else
3272                 err = f2fs_submit_page_bio(fio);
3273         if (!err) {
3274                 update_device_state(fio);
3275                 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3276         }
3277
3278         return err;
3279 }
3280
3281 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3282                                                 unsigned int segno)
3283 {
3284         int i;
3285
3286         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3287                 if (CURSEG_I(sbi, i)->segno == segno)
3288                         break;
3289         }
3290         return i;
3291 }
3292
3293 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3294                                 block_t old_blkaddr, block_t new_blkaddr,
3295                                 bool recover_curseg, bool recover_newaddr)
3296 {
3297         struct sit_info *sit_i = SIT_I(sbi);
3298         struct curseg_info *curseg;
3299         unsigned int segno, old_cursegno;
3300         struct seg_entry *se;
3301         int type;
3302         unsigned short old_blkoff;
3303
3304         segno = GET_SEGNO(sbi, new_blkaddr);
3305         se = get_seg_entry(sbi, segno);
3306         type = se->type;
3307
3308         down_write(&SM_I(sbi)->curseg_lock);
3309
3310         if (!recover_curseg) {
3311                 /* for recovery flow */
3312                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3313                         if (old_blkaddr == NULL_ADDR)
3314                                 type = CURSEG_COLD_DATA;
3315                         else
3316                                 type = CURSEG_WARM_DATA;
3317                 }
3318         } else {
3319                 if (IS_CURSEG(sbi, segno)) {
3320                         /* se->type is volatile as SSR allocation */
3321                         type = __f2fs_get_curseg(sbi, segno);
3322                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3323                 } else {
3324                         type = CURSEG_WARM_DATA;
3325                 }
3326         }
3327
3328         f2fs_bug_on(sbi, !IS_DATASEG(type));
3329         curseg = CURSEG_I(sbi, type);
3330
3331         mutex_lock(&curseg->curseg_mutex);
3332         down_write(&sit_i->sentry_lock);
3333
3334         old_cursegno = curseg->segno;
3335         old_blkoff = curseg->next_blkoff;
3336
3337         /* change the current segment */
3338         if (segno != curseg->segno) {
3339                 curseg->next_segno = segno;
3340                 change_curseg(sbi, type);
3341         }
3342
3343         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3344         __add_sum_entry(sbi, type, sum);
3345
3346         if (!recover_curseg || recover_newaddr)
3347                 update_sit_entry(sbi, new_blkaddr, 1);
3348         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3349                 invalidate_mapping_pages(META_MAPPING(sbi),
3350                                         old_blkaddr, old_blkaddr);
3351                 update_sit_entry(sbi, old_blkaddr, -1);
3352         }
3353
3354         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3355         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3356
3357         locate_dirty_segment(sbi, old_cursegno);
3358
3359         if (recover_curseg) {
3360                 if (old_cursegno != curseg->segno) {
3361                         curseg->next_segno = old_cursegno;
3362                         change_curseg(sbi, type);
3363                 }
3364                 curseg->next_blkoff = old_blkoff;
3365         }
3366
3367         up_write(&sit_i->sentry_lock);
3368         mutex_unlock(&curseg->curseg_mutex);
3369         up_write(&SM_I(sbi)->curseg_lock);
3370 }
3371
3372 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3373                                 block_t old_addr, block_t new_addr,
3374                                 unsigned char version, bool recover_curseg,
3375                                 bool recover_newaddr)
3376 {
3377         struct f2fs_summary sum;
3378
3379         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3380
3381         f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3382                                         recover_curseg, recover_newaddr);
3383
3384         f2fs_update_data_blkaddr(dn, new_addr);
3385 }
3386
3387 void f2fs_wait_on_page_writeback(struct page *page,
3388                                 enum page_type type, bool ordered, bool locked)
3389 {
3390         if (PageWriteback(page)) {
3391                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3392
3393                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3394                 if (ordered) {
3395                         wait_on_page_writeback(page);
3396                         f2fs_bug_on(sbi, locked && PageWriteback(page));
3397                 } else {
3398                         wait_for_stable_page(page);
3399                 }
3400         }
3401 }
3402
3403 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3404 {
3405         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3406         struct page *cpage;
3407
3408         if (!f2fs_post_read_required(inode))
3409                 return;
3410
3411         if (!__is_valid_data_blkaddr(blkaddr))
3412                 return;
3413
3414         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3415         if (cpage) {
3416                 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3417                 f2fs_put_page(cpage, 1);
3418         }
3419 }
3420
3421 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3422                                                                 block_t len)
3423 {
3424         block_t i;
3425
3426         for (i = 0; i < len; i++)
3427                 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3428 }
3429
3430 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3431 {
3432         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3433         struct curseg_info *seg_i;
3434         unsigned char *kaddr;
3435         struct page *page;
3436         block_t start;
3437         int i, j, offset;
3438
3439         start = start_sum_block(sbi);
3440
3441         page = f2fs_get_meta_page(sbi, start++);
3442         if (IS_ERR(page))
3443                 return PTR_ERR(page);
3444         kaddr = (unsigned char *)page_address(page);
3445
3446         /* Step 1: restore nat cache */
3447         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3448         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3449
3450         /* Step 2: restore sit cache */
3451         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3452         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3453         offset = 2 * SUM_JOURNAL_SIZE;
3454
3455         /* Step 3: restore summary entries */
3456         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3457                 unsigned short blk_off;
3458                 unsigned int segno;
3459
3460                 seg_i = CURSEG_I(sbi, i);
3461                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3462                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3463                 seg_i->next_segno = segno;
3464                 reset_curseg(sbi, i, 0);
3465                 seg_i->alloc_type = ckpt->alloc_type[i];
3466                 seg_i->next_blkoff = blk_off;
3467
3468                 if (seg_i->alloc_type == SSR)
3469                         blk_off = sbi->blocks_per_seg;
3470
3471                 for (j = 0; j < blk_off; j++) {
3472                         struct f2fs_summary *s;
3473                         s = (struct f2fs_summary *)(kaddr + offset);
3474                         seg_i->sum_blk->entries[j] = *s;
3475                         offset += SUMMARY_SIZE;
3476                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3477                                                 SUM_FOOTER_SIZE)
3478                                 continue;
3479
3480                         f2fs_put_page(page, 1);
3481                         page = NULL;
3482
3483                         page = f2fs_get_meta_page(sbi, start++);
3484                         if (IS_ERR(page))
3485                                 return PTR_ERR(page);
3486                         kaddr = (unsigned char *)page_address(page);
3487                         offset = 0;
3488                 }
3489         }
3490         f2fs_put_page(page, 1);
3491         return 0;
3492 }
3493
3494 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3495 {
3496         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3497         struct f2fs_summary_block *sum;
3498         struct curseg_info *curseg;
3499         struct page *new;
3500         unsigned short blk_off;
3501         unsigned int segno = 0;
3502         block_t blk_addr = 0;
3503         int err = 0;
3504
3505         /* get segment number and block addr */
3506         if (IS_DATASEG(type)) {
3507                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3508                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3509                                                         CURSEG_HOT_DATA]);
3510                 if (__exist_node_summaries(sbi))
3511                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3512                 else
3513                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3514         } else {
3515                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3516                                                         CURSEG_HOT_NODE]);
3517                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3518                                                         CURSEG_HOT_NODE]);
3519                 if (__exist_node_summaries(sbi))
3520                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3521                                                         type - CURSEG_HOT_NODE);
3522                 else
3523                         blk_addr = GET_SUM_BLOCK(sbi, segno);
3524         }
3525
3526         new = f2fs_get_meta_page(sbi, blk_addr);
3527         if (IS_ERR(new))
3528                 return PTR_ERR(new);
3529         sum = (struct f2fs_summary_block *)page_address(new);
3530
3531         if (IS_NODESEG(type)) {
3532                 if (__exist_node_summaries(sbi)) {
3533                         struct f2fs_summary *ns = &sum->entries[0];
3534                         int i;
3535                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3536                                 ns->version = 0;
3537                                 ns->ofs_in_node = 0;
3538                         }
3539                 } else {
3540                         err = f2fs_restore_node_summary(sbi, segno, sum);
3541                         if (err)
3542                                 goto out;
3543                 }
3544         }
3545
3546         /* set uncompleted segment to curseg */
3547         curseg = CURSEG_I(sbi, type);
3548         mutex_lock(&curseg->curseg_mutex);
3549
3550         /* update journal info */
3551         down_write(&curseg->journal_rwsem);
3552         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3553         up_write(&curseg->journal_rwsem);
3554
3555         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3556         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3557         curseg->next_segno = segno;
3558         reset_curseg(sbi, type, 0);
3559         curseg->alloc_type = ckpt->alloc_type[type];
3560         curseg->next_blkoff = blk_off;
3561         mutex_unlock(&curseg->curseg_mutex);
3562 out:
3563         f2fs_put_page(new, 1);
3564         return err;
3565 }
3566
3567 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3568 {
3569         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3570         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3571         int type = CURSEG_HOT_DATA;
3572         int err;
3573
3574         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3575                 int npages = f2fs_npages_for_summary_flush(sbi, true);
3576
3577                 if (npages >= 2)
3578                         f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3579                                                         META_CP, true);
3580
3581                 /* restore for compacted data summary */
3582                 err = read_compacted_summaries(sbi);
3583                 if (err)
3584                         return err;
3585                 type = CURSEG_HOT_NODE;
3586         }
3587
3588         if (__exist_node_summaries(sbi))
3589                 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3590                                         NR_CURSEG_TYPE - type, META_CP, true);
3591
3592         for (; type <= CURSEG_COLD_NODE; type++) {
3593                 err = read_normal_summaries(sbi, type);
3594                 if (err)
3595                         return err;
3596         }
3597
3598         /* sanity check for summary blocks */
3599         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3600                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3601                 f2fs_err(sbi, "invalid journal entries nats %u sits %u\n",
3602                          nats_in_cursum(nat_j), sits_in_cursum(sit_j));
3603                 return -EINVAL;
3604         }
3605
3606         return 0;
3607 }
3608
3609 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3610 {
3611         struct page *page;
3612         unsigned char *kaddr;
3613         struct f2fs_summary *summary;
3614         struct curseg_info *seg_i;
3615         int written_size = 0;
3616         int i, j;
3617
3618         page = f2fs_grab_meta_page(sbi, blkaddr++);
3619         kaddr = (unsigned char *)page_address(page);
3620         memset(kaddr, 0, PAGE_SIZE);
3621
3622         /* Step 1: write nat cache */
3623         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3624         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3625         written_size += SUM_JOURNAL_SIZE;
3626
3627         /* Step 2: write sit cache */
3628         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3629         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3630         written_size += SUM_JOURNAL_SIZE;
3631
3632         /* Step 3: write summary entries */
3633         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3634                 unsigned short blkoff;
3635                 seg_i = CURSEG_I(sbi, i);
3636                 if (sbi->ckpt->alloc_type[i] == SSR)
3637                         blkoff = sbi->blocks_per_seg;
3638                 else
3639                         blkoff = curseg_blkoff(sbi, i);
3640
3641                 for (j = 0; j < blkoff; j++) {
3642                         if (!page) {
3643                                 page = f2fs_grab_meta_page(sbi, blkaddr++);
3644                                 kaddr = (unsigned char *)page_address(page);
3645                                 memset(kaddr, 0, PAGE_SIZE);
3646                                 written_size = 0;
3647                         }
3648                         summary = (struct f2fs_summary *)(kaddr + written_size);
3649                         *summary = seg_i->sum_blk->entries[j];
3650                         written_size += SUMMARY_SIZE;
3651
3652                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3653                                                         SUM_FOOTER_SIZE)
3654                                 continue;
3655
3656                         set_page_dirty(page);
3657                         f2fs_put_page(page, 1);
3658                         page = NULL;
3659                 }
3660         }
3661         if (page) {
3662                 set_page_dirty(page);
3663                 f2fs_put_page(page, 1);
3664         }
3665 }
3666
3667 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3668                                         block_t blkaddr, int type)
3669 {
3670         int i, end;
3671         if (IS_DATASEG(type))
3672                 end = type + NR_CURSEG_DATA_TYPE;
3673         else
3674                 end = type + NR_CURSEG_NODE_TYPE;
3675
3676         for (i = type; i < end; i++)
3677                 write_current_sum_page(sbi, i, blkaddr + (i - type));
3678 }
3679
3680 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3681 {
3682         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3683                 write_compacted_summaries(sbi, start_blk);
3684         else
3685                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3686 }
3687
3688 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3689 {
3690         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3691 }
3692
3693 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3694                                         unsigned int val, int alloc)
3695 {
3696         int i;
3697
3698         if (type == NAT_JOURNAL) {
3699                 for (i = 0; i < nats_in_cursum(journal); i++) {
3700                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3701                                 return i;
3702                 }
3703                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3704                         return update_nats_in_cursum(journal, 1);
3705         } else if (type == SIT_JOURNAL) {
3706                 for (i = 0; i < sits_in_cursum(journal); i++)
3707                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3708                                 return i;
3709                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3710                         return update_sits_in_cursum(journal, 1);
3711         }
3712         return -1;
3713 }
3714
3715 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3716                                         unsigned int segno)
3717 {
3718         return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
3719 }
3720
3721 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3722                                         unsigned int start)
3723 {
3724         struct sit_info *sit_i = SIT_I(sbi);
3725         struct page *page;
3726         pgoff_t src_off, dst_off;
3727
3728         src_off = current_sit_addr(sbi, start);
3729         dst_off = next_sit_addr(sbi, src_off);
3730
3731         page = f2fs_grab_meta_page(sbi, dst_off);
3732         seg_info_to_sit_page(sbi, page, start);
3733
3734         set_page_dirty(page);
3735         set_to_next_sit(sit_i, start);
3736
3737         return page;
3738 }
3739
3740 static struct sit_entry_set *grab_sit_entry_set(void)
3741 {
3742         struct sit_entry_set *ses =
3743                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3744
3745         ses->entry_cnt = 0;
3746         INIT_LIST_HEAD(&ses->set_list);
3747         return ses;
3748 }
3749
3750 static void release_sit_entry_set(struct sit_entry_set *ses)
3751 {
3752         list_del(&ses->set_list);
3753         kmem_cache_free(sit_entry_set_slab, ses);
3754 }
3755
3756 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3757                                                 struct list_head *head)
3758 {
3759         struct sit_entry_set *next = ses;
3760
3761         if (list_is_last(&ses->set_list, head))
3762                 return;
3763
3764         list_for_each_entry_continue(next, head, set_list)
3765                 if (ses->entry_cnt <= next->entry_cnt)
3766                         break;
3767
3768         list_move_tail(&ses->set_list, &next->set_list);
3769 }
3770
3771 static void add_sit_entry(unsigned int segno, struct list_head *head)
3772 {
3773         struct sit_entry_set *ses;
3774         unsigned int start_segno = START_SEGNO(segno);
3775
3776         list_for_each_entry(ses, head, set_list) {
3777                 if (ses->start_segno == start_segno) {
3778                         ses->entry_cnt++;
3779                         adjust_sit_entry_set(ses, head);
3780                         return;
3781                 }
3782         }
3783
3784         ses = grab_sit_entry_set();
3785
3786         ses->start_segno = start_segno;
3787         ses->entry_cnt++;
3788         list_add(&ses->set_list, head);
3789 }
3790
3791 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3792 {
3793         struct f2fs_sm_info *sm_info = SM_I(sbi);
3794         struct list_head *set_list = &sm_info->sit_entry_set;
3795         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3796         unsigned int segno;
3797
3798         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3799                 add_sit_entry(segno, set_list);
3800 }
3801
3802 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3803 {
3804         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3805         struct f2fs_journal *journal = curseg->journal;
3806         int i;
3807
3808         down_write(&curseg->journal_rwsem);
3809         for (i = 0; i < sits_in_cursum(journal); i++) {
3810                 unsigned int segno;
3811                 bool dirtied;
3812
3813                 segno = le32_to_cpu(segno_in_journal(journal, i));
3814                 dirtied = __mark_sit_entry_dirty(sbi, segno);
3815
3816                 if (!dirtied)
3817                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3818         }
3819         update_sits_in_cursum(journal, -i);
3820         up_write(&curseg->journal_rwsem);
3821 }
3822
3823 /*
3824  * CP calls this function, which flushes SIT entries including sit_journal,
3825  * and moves prefree segs to free segs.
3826  */
3827 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3828 {
3829         struct sit_info *sit_i = SIT_I(sbi);
3830         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3831         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3832         struct f2fs_journal *journal = curseg->journal;
3833         struct sit_entry_set *ses, *tmp;
3834         struct list_head *head = &SM_I(sbi)->sit_entry_set;
3835         bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
3836         struct seg_entry *se;
3837
3838         down_write(&sit_i->sentry_lock);
3839
3840         if (!sit_i->dirty_sentries)
3841                 goto out;
3842
3843         /*
3844          * add and account sit entries of dirty bitmap in sit entry
3845          * set temporarily
3846          */
3847         add_sits_in_set(sbi);
3848
3849         /*
3850          * if there are no enough space in journal to store dirty sit
3851          * entries, remove all entries from journal and add and account
3852          * them in sit entry set.
3853          */
3854         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
3855                                                                 !to_journal)
3856                 remove_sits_in_journal(sbi);
3857
3858         /*
3859          * there are two steps to flush sit entries:
3860          * #1, flush sit entries to journal in current cold data summary block.
3861          * #2, flush sit entries to sit page.
3862          */
3863         list_for_each_entry_safe(ses, tmp, head, set_list) {
3864                 struct page *page = NULL;
3865                 struct f2fs_sit_block *raw_sit = NULL;
3866                 unsigned int start_segno = ses->start_segno;
3867                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3868                                                 (unsigned long)MAIN_SEGS(sbi));
3869                 unsigned int segno = start_segno;
3870
3871                 if (to_journal &&
3872                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3873                         to_journal = false;
3874
3875                 if (to_journal) {
3876                         down_write(&curseg->journal_rwsem);
3877                 } else {
3878                         page = get_next_sit_page(sbi, start_segno);
3879                         raw_sit = page_address(page);
3880                 }
3881
3882                 /* flush dirty sit entries in region of current sit set */
3883                 for_each_set_bit_from(segno, bitmap, end) {
3884                         int offset, sit_offset;
3885
3886                         se = get_seg_entry(sbi, segno);
3887 #ifdef CONFIG_F2FS_CHECK_FS
3888                         if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3889                                                 SIT_VBLOCK_MAP_SIZE))
3890                                 f2fs_bug_on(sbi, 1);
3891 #endif
3892
3893                         /* add discard candidates */
3894                         if (!(cpc->reason & CP_DISCARD)) {
3895                                 cpc->trim_start = segno;
3896                                 add_discard_addrs(sbi, cpc, false);
3897                         }
3898
3899                         if (to_journal) {
3900                                 offset = f2fs_lookup_journal_in_cursum(journal,
3901                                                         SIT_JOURNAL, segno, 1);
3902                                 f2fs_bug_on(sbi, offset < 0);
3903                                 segno_in_journal(journal, offset) =
3904                                                         cpu_to_le32(segno);
3905                                 seg_info_to_raw_sit(se,
3906                                         &sit_in_journal(journal, offset));
3907                                 check_block_count(sbi, segno,
3908                                         &sit_in_journal(journal, offset));
3909                         } else {
3910                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3911                                 seg_info_to_raw_sit(se,
3912                                                 &raw_sit->entries[sit_offset]);
3913                                 check_block_count(sbi, segno,
3914                                                 &raw_sit->entries[sit_offset]);
3915                         }
3916
3917                         __clear_bit(segno, bitmap);
3918                         sit_i->dirty_sentries--;
3919                         ses->entry_cnt--;
3920                 }
3921
3922                 if (to_journal)
3923                         up_write(&curseg->journal_rwsem);
3924                 else
3925                         f2fs_put_page(page, 1);
3926
3927                 f2fs_bug_on(sbi, ses->entry_cnt);
3928                 release_sit_entry_set(ses);
3929         }
3930
3931         f2fs_bug_on(sbi, !list_empty(head));
3932         f2fs_bug_on(sbi, sit_i->dirty_sentries);
3933 out:
3934         if (cpc->reason & CP_DISCARD) {
3935                 __u64 trim_start = cpc->trim_start;
3936
3937                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3938                         add_discard_addrs(sbi, cpc, false);
3939
3940                 cpc->trim_start = trim_start;
3941         }
3942         up_write(&sit_i->sentry_lock);
3943
3944         set_prefree_as_free_segments(sbi);
3945 }
3946
3947 static int build_sit_info(struct f2fs_sb_info *sbi)
3948 {
3949         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3950         struct sit_info *sit_i;
3951         unsigned int sit_segs, start;
3952         char *src_bitmap, *bitmap;
3953         unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
3954
3955         /* allocate memory for SIT information */
3956         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3957         if (!sit_i)
3958                 return -ENOMEM;
3959
3960         SM_I(sbi)->sit_info = sit_i;
3961
3962         sit_i->sentries =
3963                 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3964                                               MAIN_SEGS(sbi)),
3965                               GFP_KERNEL);
3966         if (!sit_i->sentries)
3967                 return -ENOMEM;
3968
3969         main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3970         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
3971                                                                 GFP_KERNEL);
3972         if (!sit_i->dirty_sentries_bitmap)
3973                 return -ENOMEM;
3974
3975 #ifdef CONFIG_F2FS_CHECK_FS
3976         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 4;
3977 #else
3978         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 3;
3979 #endif
3980         sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
3981         if (!sit_i->bitmap)
3982                 return -ENOMEM;
3983
3984         bitmap = sit_i->bitmap;
3985
3986         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3987                 sit_i->sentries[start].cur_valid_map = bitmap;
3988                 bitmap += SIT_VBLOCK_MAP_SIZE;
3989
3990                 sit_i->sentries[start].ckpt_valid_map = bitmap;
3991                 bitmap += SIT_VBLOCK_MAP_SIZE;
3992
3993 #ifdef CONFIG_F2FS_CHECK_FS
3994                 sit_i->sentries[start].cur_valid_map_mir = bitmap;
3995                 bitmap += SIT_VBLOCK_MAP_SIZE;
3996 #endif
3997
3998                 sit_i->sentries[start].discard_map = bitmap;
3999                 bitmap += SIT_VBLOCK_MAP_SIZE;
4000         }
4001
4002         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4003         if (!sit_i->tmp_map)
4004                 return -ENOMEM;
4005
4006         if (__is_large_section(sbi)) {
4007                 sit_i->sec_entries =
4008                         f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4009                                                       MAIN_SECS(sbi)),
4010                                       GFP_KERNEL);
4011                 if (!sit_i->sec_entries)
4012                         return -ENOMEM;
4013         }
4014
4015         /* get information related with SIT */
4016         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4017
4018         /* setup SIT bitmap from ckeckpoint pack */
4019         sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4020         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4021
4022         sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4023         if (!sit_i->sit_bitmap)
4024                 return -ENOMEM;
4025
4026 #ifdef CONFIG_F2FS_CHECK_FS
4027         sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4028                                         sit_bitmap_size, GFP_KERNEL);
4029         if (!sit_i->sit_bitmap_mir)
4030                 return -ENOMEM;
4031
4032         sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4033                                         main_bitmap_size, GFP_KERNEL);
4034         if (!sit_i->invalid_segmap)
4035                 return -ENOMEM;
4036 #endif
4037
4038         /* init SIT information */
4039         sit_i->s_ops = &default_salloc_ops;
4040
4041         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4042         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4043         sit_i->written_valid_blocks = 0;
4044         sit_i->bitmap_size = sit_bitmap_size;
4045         sit_i->dirty_sentries = 0;
4046         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4047         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4048         sit_i->mounted_time = ktime_get_real_seconds();
4049         init_rwsem(&sit_i->sentry_lock);
4050         return 0;
4051 }
4052
4053 static int build_free_segmap(struct f2fs_sb_info *sbi)
4054 {
4055         struct free_segmap_info *free_i;
4056         unsigned int bitmap_size, sec_bitmap_size;
4057
4058         /* allocate memory for free segmap information */
4059         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4060         if (!free_i)
4061                 return -ENOMEM;
4062
4063         SM_I(sbi)->free_info = free_i;
4064
4065         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4066         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4067         if (!free_i->free_segmap)
4068                 return -ENOMEM;
4069
4070         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4071         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4072         if (!free_i->free_secmap)
4073                 return -ENOMEM;
4074
4075         /* set all segments as dirty temporarily */
4076         memset(free_i->free_segmap, 0xff, bitmap_size);
4077         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4078
4079         /* init free segmap information */
4080         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4081         free_i->free_segments = 0;
4082         free_i->free_sections = 0;
4083         spin_lock_init(&free_i->segmap_lock);
4084         return 0;
4085 }
4086
4087 static int build_curseg(struct f2fs_sb_info *sbi)
4088 {
4089         struct curseg_info *array;
4090         int i;
4091
4092         array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
4093                              GFP_KERNEL);
4094         if (!array)
4095                 return -ENOMEM;
4096
4097         SM_I(sbi)->curseg_array = array;
4098
4099         for (i = 0; i < NR_CURSEG_TYPE; i++) {
4100                 mutex_init(&array[i].curseg_mutex);
4101                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4102                 if (!array[i].sum_blk)
4103                         return -ENOMEM;
4104                 init_rwsem(&array[i].journal_rwsem);
4105                 array[i].journal = f2fs_kzalloc(sbi,
4106                                 sizeof(struct f2fs_journal), GFP_KERNEL);
4107                 if (!array[i].journal)
4108                         return -ENOMEM;
4109                 array[i].segno = NULL_SEGNO;
4110                 array[i].next_blkoff = 0;
4111         }
4112         return restore_curseg_summaries(sbi);
4113 }
4114
4115 static int build_sit_entries(struct f2fs_sb_info *sbi)
4116 {
4117         struct sit_info *sit_i = SIT_I(sbi);
4118         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4119         struct f2fs_journal *journal = curseg->journal;
4120         struct seg_entry *se;
4121         struct f2fs_sit_entry sit;
4122         int sit_blk_cnt = SIT_BLK_CNT(sbi);
4123         unsigned int i, start, end;
4124         unsigned int readed, start_blk = 0;
4125         int err = 0;
4126         block_t total_node_blocks = 0;
4127
4128         do {
4129                 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
4130                                                         META_SIT, true);
4131
4132                 start = start_blk * sit_i->sents_per_block;
4133                 end = (start_blk + readed) * sit_i->sents_per_block;
4134
4135                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4136                         struct f2fs_sit_block *sit_blk;
4137                         struct page *page;
4138
4139                         se = &sit_i->sentries[start];
4140                         page = get_current_sit_page(sbi, start);
4141                         if (IS_ERR(page))
4142                                 return PTR_ERR(page);
4143                         sit_blk = (struct f2fs_sit_block *)page_address(page);
4144                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4145                         f2fs_put_page(page, 1);
4146
4147                         err = check_block_count(sbi, start, &sit);
4148                         if (err)
4149                                 return err;
4150                         seg_info_from_raw_sit(se, &sit);
4151                         if (IS_NODESEG(se->type))
4152                                 total_node_blocks += se->valid_blocks;
4153
4154                         /* build discard map only one time */
4155                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4156                                 memset(se->discard_map, 0xff,
4157                                         SIT_VBLOCK_MAP_SIZE);
4158                         } else {
4159                                 memcpy(se->discard_map,
4160                                         se->cur_valid_map,
4161                                         SIT_VBLOCK_MAP_SIZE);
4162                                 sbi->discard_blks +=
4163                                         sbi->blocks_per_seg -
4164                                         se->valid_blocks;
4165                         }
4166
4167                         if (__is_large_section(sbi))
4168                                 get_sec_entry(sbi, start)->valid_blocks +=
4169                                                         se->valid_blocks;
4170                 }
4171                 start_blk += readed;
4172         } while (start_blk < sit_blk_cnt);
4173
4174         down_read(&curseg->journal_rwsem);
4175         for (i = 0; i < sits_in_cursum(journal); i++) {
4176                 unsigned int old_valid_blocks;
4177
4178                 start = le32_to_cpu(segno_in_journal(journal, i));
4179                 if (start >= MAIN_SEGS(sbi)) {
4180                         f2fs_err(sbi, "Wrong journal entry on segno %u",
4181                                  start);
4182                         err = -EFSCORRUPTED;
4183                         break;
4184                 }
4185
4186                 se = &sit_i->sentries[start];
4187                 sit = sit_in_journal(journal, i);
4188
4189                 old_valid_blocks = se->valid_blocks;
4190                 if (IS_NODESEG(se->type))
4191                         total_node_blocks -= old_valid_blocks;
4192
4193                 err = check_block_count(sbi, start, &sit);
4194                 if (err)
4195                         break;
4196                 seg_info_from_raw_sit(se, &sit);
4197                 if (IS_NODESEG(se->type))
4198                         total_node_blocks += se->valid_blocks;
4199
4200                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4201                         memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4202                 } else {
4203                         memcpy(se->discard_map, se->cur_valid_map,
4204                                                 SIT_VBLOCK_MAP_SIZE);
4205                         sbi->discard_blks += old_valid_blocks;
4206                         sbi->discard_blks -= se->valid_blocks;
4207                 }
4208
4209                 if (__is_large_section(sbi)) {
4210                         get_sec_entry(sbi, start)->valid_blocks +=
4211                                                         se->valid_blocks;
4212                         get_sec_entry(sbi, start)->valid_blocks -=
4213                                                         old_valid_blocks;
4214                 }
4215         }
4216         up_read(&curseg->journal_rwsem);
4217
4218         if (!err && total_node_blocks != valid_node_count(sbi)) {
4219                 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4220                          total_node_blocks, valid_node_count(sbi));
4221                 err = -EFSCORRUPTED;
4222         }
4223
4224         return err;
4225 }
4226
4227 static void init_free_segmap(struct f2fs_sb_info *sbi)
4228 {
4229         unsigned int start;
4230         int type;
4231
4232         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4233                 struct seg_entry *sentry = get_seg_entry(sbi, start);
4234                 if (!sentry->valid_blocks)
4235                         __set_free(sbi, start);
4236                 else
4237                         SIT_I(sbi)->written_valid_blocks +=
4238                                                 sentry->valid_blocks;
4239         }
4240
4241         /* set use the current segments */
4242         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4243                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4244                 __set_test_and_inuse(sbi, curseg_t->segno);
4245         }
4246 }
4247
4248 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4249 {
4250         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4251         struct free_segmap_info *free_i = FREE_I(sbi);
4252         unsigned int segno = 0, offset = 0;
4253         unsigned short valid_blocks;
4254
4255         while (1) {
4256                 /* find dirty segment based on free segmap */
4257                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4258                 if (segno >= MAIN_SEGS(sbi))
4259                         break;
4260                 offset = segno + 1;
4261                 valid_blocks = get_valid_blocks(sbi, segno, false);
4262                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
4263                         continue;
4264                 if (valid_blocks > sbi->blocks_per_seg) {
4265                         f2fs_bug_on(sbi, 1);
4266                         continue;
4267                 }
4268                 mutex_lock(&dirty_i->seglist_lock);
4269                 __locate_dirty_segment(sbi, segno, DIRTY);
4270                 mutex_unlock(&dirty_i->seglist_lock);
4271         }
4272 }
4273
4274 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4275 {
4276         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4277         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4278
4279         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4280         if (!dirty_i->victim_secmap)
4281                 return -ENOMEM;
4282         return 0;
4283 }
4284
4285 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4286 {
4287         struct dirty_seglist_info *dirty_i;
4288         unsigned int bitmap_size, i;
4289
4290         /* allocate memory for dirty segments list information */
4291         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4292                                                                 GFP_KERNEL);
4293         if (!dirty_i)
4294                 return -ENOMEM;
4295
4296         SM_I(sbi)->dirty_info = dirty_i;
4297         mutex_init(&dirty_i->seglist_lock);
4298
4299         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4300
4301         for (i = 0; i < NR_DIRTY_TYPE; i++) {
4302                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4303                                                                 GFP_KERNEL);
4304                 if (!dirty_i->dirty_segmap[i])
4305                         return -ENOMEM;
4306         }
4307
4308         init_dirty_segmap(sbi);
4309         return init_victim_secmap(sbi);
4310 }
4311
4312 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4313 {
4314         int i;
4315
4316         /*
4317          * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4318          * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4319          */
4320         for (i = 0; i < NO_CHECK_TYPE; i++) {
4321                 struct curseg_info *curseg = CURSEG_I(sbi, i);
4322                 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4323                 unsigned int blkofs = curseg->next_blkoff;
4324
4325                 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4326                         goto out;
4327
4328                 if (curseg->alloc_type == SSR)
4329                         continue;
4330
4331                 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4332                         if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4333                                 continue;
4334 out:
4335                         f2fs_err(sbi,
4336                                  "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4337                                  i, curseg->segno, curseg->alloc_type,
4338                                  curseg->next_blkoff, blkofs);
4339                         return -EFSCORRUPTED;
4340                 }
4341         }
4342         return 0;
4343 }
4344
4345 /*
4346  * Update min, max modified time for cost-benefit GC algorithm
4347  */
4348 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4349 {
4350         struct sit_info *sit_i = SIT_I(sbi);
4351         unsigned int segno;
4352
4353         down_write(&sit_i->sentry_lock);
4354
4355         sit_i->min_mtime = ULLONG_MAX;
4356
4357         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4358                 unsigned int i;
4359                 unsigned long long mtime = 0;
4360
4361                 for (i = 0; i < sbi->segs_per_sec; i++)
4362                         mtime += get_seg_entry(sbi, segno + i)->mtime;
4363
4364                 mtime = div_u64(mtime, sbi->segs_per_sec);
4365
4366                 if (sit_i->min_mtime > mtime)
4367                         sit_i->min_mtime = mtime;
4368         }
4369         sit_i->max_mtime = get_mtime(sbi, false);
4370         up_write(&sit_i->sentry_lock);
4371 }
4372
4373 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
4374 {
4375         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4376         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4377         struct f2fs_sm_info *sm_info;
4378         int err;
4379
4380         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
4381         if (!sm_info)
4382                 return -ENOMEM;
4383
4384         /* init sm info */
4385         sbi->sm_info = sm_info;
4386         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4387         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4388         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4389         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4390         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4391         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4392         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
4393         sm_info->rec_prefree_segments = sm_info->main_segments *
4394                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
4395         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4396                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4397
4398         if (!test_opt(sbi, LFS))
4399                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
4400         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
4401         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
4402         sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
4403         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
4404         sm_info->min_ssr_sections = reserved_sections(sbi);
4405
4406         INIT_LIST_HEAD(&sm_info->sit_entry_set);
4407
4408         init_rwsem(&sm_info->curseg_lock);
4409
4410         if (!f2fs_readonly(sbi->sb)) {
4411                 err = f2fs_create_flush_cmd_control(sbi);
4412                 if (err)
4413                         return err;
4414         }
4415
4416         err = create_discard_cmd_control(sbi);
4417         if (err)
4418                 return err;
4419
4420         err = build_sit_info(sbi);
4421         if (err)
4422                 return err;
4423         err = build_free_segmap(sbi);
4424         if (err)
4425                 return err;
4426         err = build_curseg(sbi);
4427         if (err)
4428                 return err;
4429
4430         /* reinit free segmap based on SIT */
4431         err = build_sit_entries(sbi);
4432         if (err)
4433                 return err;
4434
4435         init_free_segmap(sbi);
4436         err = build_dirty_segmap(sbi);
4437         if (err)
4438                 return err;
4439
4440         err = sanity_check_curseg(sbi);
4441         if (err)
4442                 return err;
4443
4444         init_min_max_mtime(sbi);
4445         return 0;
4446 }
4447
4448 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4449                 enum dirty_type dirty_type)
4450 {
4451         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4452
4453         mutex_lock(&dirty_i->seglist_lock);
4454         kvfree(dirty_i->dirty_segmap[dirty_type]);
4455         dirty_i->nr_dirty[dirty_type] = 0;
4456         mutex_unlock(&dirty_i->seglist_lock);
4457 }
4458
4459 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
4460 {
4461         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4462         kvfree(dirty_i->victim_secmap);
4463 }
4464
4465 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4466 {
4467         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4468         int i;
4469
4470         if (!dirty_i)
4471                 return;
4472
4473         /* discard pre-free/dirty segments list */
4474         for (i = 0; i < NR_DIRTY_TYPE; i++)
4475                 discard_dirty_segmap(sbi, i);
4476
4477         destroy_victim_secmap(sbi);
4478         SM_I(sbi)->dirty_info = NULL;
4479         kvfree(dirty_i);
4480 }
4481
4482 static void destroy_curseg(struct f2fs_sb_info *sbi)
4483 {
4484         struct curseg_info *array = SM_I(sbi)->curseg_array;
4485         int i;
4486
4487         if (!array)
4488                 return;
4489         SM_I(sbi)->curseg_array = NULL;
4490         for (i = 0; i < NR_CURSEG_TYPE; i++) {
4491                 kvfree(array[i].sum_blk);
4492                 kvfree(array[i].journal);
4493         }
4494         kvfree(array);
4495 }
4496
4497 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4498 {
4499         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4500         if (!free_i)
4501                 return;
4502         SM_I(sbi)->free_info = NULL;
4503         kvfree(free_i->free_segmap);
4504         kvfree(free_i->free_secmap);
4505         kvfree(free_i);
4506 }
4507
4508 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4509 {
4510         struct sit_info *sit_i = SIT_I(sbi);
4511
4512         if (!sit_i)
4513                 return;
4514
4515         if (sit_i->sentries)
4516                 kvfree(sit_i->bitmap);
4517         kvfree(sit_i->tmp_map);
4518
4519         kvfree(sit_i->sentries);
4520         kvfree(sit_i->sec_entries);
4521         kvfree(sit_i->dirty_sentries_bitmap);
4522
4523         SM_I(sbi)->sit_info = NULL;
4524         kvfree(sit_i->sit_bitmap);
4525 #ifdef CONFIG_F2FS_CHECK_FS
4526         kvfree(sit_i->sit_bitmap_mir);
4527         kvfree(sit_i->invalid_segmap);
4528 #endif
4529         kvfree(sit_i);
4530 }
4531
4532 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
4533 {
4534         struct f2fs_sm_info *sm_info = SM_I(sbi);
4535
4536         if (!sm_info)
4537                 return;
4538         f2fs_destroy_flush_cmd_control(sbi, true);
4539         destroy_discard_cmd_control(sbi);
4540         destroy_dirty_segmap(sbi);
4541         destroy_curseg(sbi);
4542         destroy_free_segmap(sbi);
4543         destroy_sit_info(sbi);
4544         sbi->sm_info = NULL;
4545         kvfree(sm_info);
4546 }
4547
4548 int __init f2fs_create_segment_manager_caches(void)
4549 {
4550         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4551                         sizeof(struct discard_entry));
4552         if (!discard_entry_slab)
4553                 goto fail;
4554
4555         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4556                         sizeof(struct discard_cmd));
4557         if (!discard_cmd_slab)
4558                 goto destroy_discard_entry;
4559
4560         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4561                         sizeof(struct sit_entry_set));
4562         if (!sit_entry_set_slab)
4563                 goto destroy_discard_cmd;
4564
4565         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4566                         sizeof(struct inmem_pages));
4567         if (!inmem_entry_slab)
4568                 goto destroy_sit_entry_set;
4569         return 0;
4570
4571 destroy_sit_entry_set:
4572         kmem_cache_destroy(sit_entry_set_slab);
4573 destroy_discard_cmd:
4574         kmem_cache_destroy(discard_cmd_slab);
4575 destroy_discard_entry:
4576         kmem_cache_destroy(discard_entry_slab);
4577 fail:
4578         return -ENOMEM;
4579 }
4580
4581 void f2fs_destroy_segment_manager_caches(void)
4582 {
4583         kmem_cache_destroy(sit_entry_set_slab);
4584         kmem_cache_destroy(discard_cmd_slab);
4585         kmem_cache_destroy(discard_entry_slab);
4586         kmem_cache_destroy(inmem_entry_slab);
4587 }