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 inmem_pages *new;
 189
 190         f2fs_trace_pid(page);
 191
 192         f2fs_set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
 193
 194         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 195
 196         /* add atomic page indices to the list */
 197         new->page = page;
 198         INIT_LIST_HEAD(&new->list);
 199
 200         /* increase reference count with clean state */
 201         get_page(page);
 202         mutex_lock(&F2FS_I(inode)->inmem_lock);
 203         list_add_tail(&new->list, &F2FS_I(inode)->inmem_pages);
 204         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 205         mutex_unlock(&F2FS_I(inode)->inmem_lock);
 206
 207         trace_f2fs_register_inmem_page(page, INMEM);
 208 }
 209
 210 static int __revoke_inmem_pages(struct inode *inode,
 211                                 struct list_head *head, bool drop, bool recover,
 212                                 bool trylock)
 213 {
 214         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 215         struct inmem_pages *cur, *tmp;
 216         int err = 0;
 217
 218         list_for_each_entry_safe(cur, tmp, head, list) {
 219                 struct page *page = cur->page;
 220
 221                 if (drop)
 222                         trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 223
 224                 if (trylock) {
 225                         /*
 226                          * to avoid deadlock in between page lock and
 227                          * inmem_lock.
 228                          */
 229                         if (!trylock_page(page))
 230                                 continue;
 231                 } else {
 232                         lock_page(page);
 233                 }
 234
 235                 f2fs_wait_on_page_writeback(page, DATA, true, true);
 236
 237                 if (recover) {
 238                         struct dnode_of_data dn;
 239                         struct node_info ni;
 240
 241                         trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 242 retry:
 243                         set_new_dnode(&dn, inode, NULL, NULL, 0);
 244                         err = f2fs_get_dnode_of_data(&dn, page->index,
 245                                                                 LOOKUP_NODE);
 246                         if (err) {
 247                                 if (err == -ENOMEM) {
 248                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 249                                         cond_resched();
 250                                         goto retry;
 251                                 }
 252                                 err = -EAGAIN;
 253                                 goto next;
 254                         }
 255
 256                         err = f2fs_get_node_info(sbi, dn.nid, &ni);
 257                         if (err) {
 258                                 f2fs_put_dnode(&dn);
 259                                 return err;
 260                         }
 261
 262                         if (cur->old_addr == NEW_ADDR) {
 263                                 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
 264                                 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
 265                         } else
 266                                 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 267                                         cur->old_addr, ni.version, true, true);
 268                         f2fs_put_dnode(&dn);
 269                 }
 270 next:
 271                 /* we don't need to invalidate this in the sccessful status */
 272                 if (drop || recover) {
 273                         ClearPageUptodate(page);
 274                         clear_cold_data(page);
 275                 }
 276                 f2fs_clear_page_private(page);
 277                 f2fs_put_page(page, 1);
 278
 279                 list_del(&cur->list);
 280                 kmem_cache_free(inmem_entry_slab, cur);
 281                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 282         }
 283         return err;
 284 }
 285
 286 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
 287 {
 288         struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
 289         struct inode *inode;
 290         struct f2fs_inode_info *fi;
 291         unsigned int count = sbi->atomic_files;
 292         unsigned int looped = 0;
 293 next:
 294         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 295         if (list_empty(head)) {
 296                 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 297                 return;
 298         }
 299         fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
 300         inode = igrab(&fi->vfs_inode);
 301         if (inode)
 302                 list_move_tail(&fi->inmem_ilist, head);
 303         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 304
 305         if (inode) {
 306                 if (gc_failure) {
 307                         if (!fi->i_gc_failures[GC_FAILURE_ATOMIC])
 308                                 goto skip;
 309                 }
 310                 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
 311                 f2fs_drop_inmem_pages(inode);
 312 skip:
 313                 iput(inode);
 314         }
 315         congestion_wait(BLK_RW_ASYNC, HZ/50);
 316         cond_resched();
 317         if (gc_failure) {
 318                 if (++looped >= count)
 319                         return;
 320         }
 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                 mutex_unlock(&fi->inmem_lock);
 334         }
 335
 336         fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
 337         stat_dec_atomic_write(inode);
 338
 339         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 340         if (!list_empty(&fi->inmem_ilist))
 341                 list_del_init(&fi->inmem_ilist);
 342         if (f2fs_is_atomic_file(inode)) {
 343                 clear_inode_flag(inode, FI_ATOMIC_FILE);
 344                 sbi->atomic_files--;
 345         }
 346         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 347 }
 348
 349 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
 350 {
 351         struct f2fs_inode_info *fi = F2FS_I(inode);
 352         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 353         struct list_head *head = &fi->inmem_pages;
 354         struct inmem_pages *cur = NULL;
 355         struct inmem_pages *tmp;
 356
 357         f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
 358
 359         mutex_lock(&fi->inmem_lock);
 360         list_for_each_entry(tmp, head, list) {
 361                 if (tmp->page == page) {
 362                         cur = tmp;
 363                         break;
 364                 }
 365         }
 366
 367         f2fs_bug_on(sbi, !cur);
 368         list_del(&cur->list);
 369         mutex_unlock(&fi->inmem_lock);
 370
 371         dec_page_count(sbi, F2FS_INMEM_PAGES);
 372         kmem_cache_free(inmem_entry_slab, cur);
 373
 374         ClearPageUptodate(page);
 375         f2fs_clear_page_private(page);
 376         f2fs_put_page(page, 0);
 377
 378         trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 379 }
 380
 381 static int __f2fs_commit_inmem_pages(struct inode *inode)
 382 {
 383         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 384         struct f2fs_inode_info *fi = F2FS_I(inode);
 385         struct inmem_pages *cur, *tmp;
 386         struct f2fs_io_info fio = {
 387                 .sbi = sbi,
 388                 .ino = inode->i_ino,
 389                 .type = DATA,
 390                 .op = REQ_OP_WRITE,
 391                 .op_flags = REQ_SYNC | REQ_PRIO,
 392                 .io_type = FS_DATA_IO,
 393         };
 394         struct list_head revoke_list;
 395         bool submit_bio = false;
 396         int err = 0;
 397
 398         INIT_LIST_HEAD(&revoke_list);
 399
 400         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 401                 struct page *page = cur->page;
 402
 403                 lock_page(page);
 404                 if (page->mapping == inode->i_mapping) {
 405                         trace_f2fs_commit_inmem_page(page, INMEM);
 406
 407                         f2fs_wait_on_page_writeback(page, DATA, true, true);
 408
 409                         set_page_dirty(page);
 410                         if (clear_page_dirty_for_io(page)) {
 411                                 inode_dec_dirty_pages(inode);
 412                                 f2fs_remove_dirty_inode(inode);
 413                         }
 414 retry:
 415                         fio.page = page;
 416                         fio.old_blkaddr = NULL_ADDR;
 417                         fio.encrypted_page = NULL;
 418                         fio.need_lock = LOCK_DONE;
 419                         err = f2fs_do_write_data_page(&fio);
 420                         if (err) {
 421                                 if (err == -ENOMEM) {
 422                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 423                                         cond_resched();
 424                                         goto retry;
 425                                 }
 426                                 unlock_page(page);
 427                                 break;
 428                         }
 429                         /* record old blkaddr for revoking */
 430                         cur->old_addr = fio.old_blkaddr;
 431                         submit_bio = true;
 432                 }
 433                 unlock_page(page);
 434                 list_move_tail(&cur->list, &revoke_list);
 435         }
 436
 437         if (submit_bio)
 438                 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
 439
 440         if (err) {
 441                 /*
 442                  * try to revoke all committed pages, but still we could fail
 443                  * due to no memory or other reason, if that happened, EAGAIN
 444                  * will be returned, which means in such case, transaction is
 445                  * already not integrity, caller should use journal to do the
 446                  * recovery or rewrite & commit last transaction. For other
 447                  * error number, revoking was done by filesystem itself.
 448                  */
 449                 err = __revoke_inmem_pages(inode, &revoke_list,
 450                                                 false, true, false);
 451
 452                 /* drop all uncommitted pages */
 453                 __revoke_inmem_pages(inode, &fi->inmem_pages,
 454                                                 true, false, false);
 455         } else {
 456                 __revoke_inmem_pages(inode, &revoke_list,
 457                                                 false, false, false);
 458         }
 459
 460         return err;
 461 }
 462
 463 int f2fs_commit_inmem_pages(struct inode *inode)
 464 {
 465         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 466         struct f2fs_inode_info *fi = F2FS_I(inode);
 467         int err;
 468
 469         f2fs_balance_fs(sbi, true);
 470
 471         down_write(&fi->i_gc_rwsem[WRITE]);
 472
 473         f2fs_lock_op(sbi);
 474         set_inode_flag(inode, FI_ATOMIC_COMMIT);
 475
 476         mutex_lock(&fi->inmem_lock);
 477         err = __f2fs_commit_inmem_pages(inode);
 478         mutex_unlock(&fi->inmem_lock);
 479
 480         clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 481
 482         f2fs_unlock_op(sbi);
 483         up_write(&fi->i_gc_rwsem[WRITE]);
 484
 485         return err;
 486 }
 487
 488 /*
 489  * This function balances dirty node and dentry pages.
 490  * In addition, it controls garbage collection.
 491  */
 492 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 493 {
 494         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 495                 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
 496                 f2fs_stop_checkpoint(sbi, false);
 497         }
 498
 499         /* balance_fs_bg is able to be pending */
 500         if (need && excess_cached_nats(sbi))
 501                 f2fs_balance_fs_bg(sbi);
 502
 503         if (!f2fs_is_checkpoint_ready(sbi))
 504                 return;
 505
 506         /*
 507          * We should do GC or end up with checkpoint, if there are so many dirty
 508          * dir/node pages without enough free segments.
 509          */
 510         if (has_not_enough_free_secs(sbi, 0, 0)) {
 511                 mutex_lock(&sbi->gc_mutex);
 512                 f2fs_gc(sbi, false, false, NULL_SEGNO);
 513         }
 514 }
 515
 516 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 517 {
 518         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 519                 return;
 520
 521         /* try to shrink extent cache when there is no enough memory */
 522         if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
 523                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 524
 525         /* check the # of cached NAT entries */
 526         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
 527                 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 528
 529         if (!f2fs_available_free_memory(sbi, FREE_NIDS))
 530                 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
 531         else
 532                 f2fs_build_free_nids(sbi, false, false);
 533
 534         if (!is_idle(sbi, REQ_TIME) &&
 535                 (!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
 536                 return;
 537
 538         /* checkpoint is the only way to shrink partial cached entries */
 539         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
 540                         !f2fs_available_free_memory(sbi, INO_ENTRIES) ||
 541                         excess_prefree_segs(sbi) ||
 542                         excess_dirty_nats(sbi) ||
 543                         excess_dirty_nodes(sbi) ||
 544                         f2fs_time_over(sbi, CP_TIME)) {
 545                 if (test_opt(sbi, DATA_FLUSH)) {
 546                         struct blk_plug plug;
 547
 548                         mutex_lock(&sbi->flush_lock);
 549
 550                         blk_start_plug(&plug);
 551                         f2fs_sync_dirty_inodes(sbi, FILE_INODE);
 552                         blk_finish_plug(&plug);
 553
 554                         mutex_unlock(&sbi->flush_lock);
 555                 }
 556                 f2fs_sync_fs(sbi->sb, true);
 557                 stat_inc_bg_cp_count(sbi->stat_info);
 558         }
 559 }
 560
 561 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 562                                 struct block_device *bdev)
 563 {
 564         struct bio *bio;
 565         int ret;
 566
 567         bio = f2fs_bio_alloc(sbi, 0, false);
 568         if (!bio)
 569                 return -ENOMEM;
 570
 571         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
 572         bio_set_dev(bio, bdev);
 573         ret = submit_bio_wait(bio);
 574         bio_put(bio);
 575
 576         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 577                                 test_opt(sbi, FLUSH_MERGE), ret);
 578         return ret;
 579 }
 580
 581 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 582 {
 583         int ret = 0;
 584         int i;
 585
 586         if (!f2fs_is_multi_device(sbi))
 587                 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 588
 589         for (i = 0; i < sbi->s_ndevs; i++) {
 590                 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
 591                         continue;
 592                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 593                 if (ret)
 594                         break;
 595         }
 596         return ret;
 597 }
 598
 599 static int issue_flush_thread(void *data)
 600 {
 601         struct f2fs_sb_info *sbi = data;
 602         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 603         wait_queue_head_t *q = &fcc->flush_wait_queue;
 604 repeat:
 605         if (kthread_should_stop())
 606                 return 0;
 607
 608         sb_start_intwrite(sbi->sb);
 609
 610         if (!llist_empty(&fcc->issue_list)) {
 611                 struct flush_cmd *cmd, *next;
 612                 int ret;
 613
 614                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 615                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 616
 617                 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 618
 619                 ret = submit_flush_wait(sbi, cmd->ino);
 620                 atomic_inc(&fcc->issued_flush);
 621
 622                 llist_for_each_entry_safe(cmd, next,
 623                                           fcc->dispatch_list, llnode) {
 624                         cmd->ret = ret;
 625                         complete(&cmd->wait);
 626                 }
 627                 fcc->dispatch_list = NULL;
 628         }
 629
 630         sb_end_intwrite(sbi->sb);
 631
 632         wait_event_interruptible(*q,
 633                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 634         goto repeat;
 635 }
 636
 637 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 638 {
 639         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 640         struct flush_cmd cmd;
 641         int ret;
 642
 643         if (test_opt(sbi, NOBARRIER))
 644                 return 0;
 645
 646         if (!test_opt(sbi, FLUSH_MERGE)) {
 647                 atomic_inc(&fcc->queued_flush);
 648                 ret = submit_flush_wait(sbi, ino);
 649                 atomic_dec(&fcc->queued_flush);
 650                 atomic_inc(&fcc->issued_flush);
 651                 return ret;
 652         }
 653
 654         if (atomic_inc_return(&fcc->queued_flush) == 1 ||
 655             f2fs_is_multi_device(sbi)) {
 656                 ret = submit_flush_wait(sbi, ino);
 657                 atomic_dec(&fcc->queued_flush);
 658
 659                 atomic_inc(&fcc->issued_flush);
 660                 return ret;
 661         }
 662
 663         cmd.ino = ino;
 664         init_completion(&cmd.wait);
 665
 666         llist_add(&cmd.llnode, &fcc->issue_list);
 667
 668         /* update issue_list before we wake up issue_flush thread */
 669         smp_mb();
 670
 671         if (waitqueue_active(&fcc->flush_wait_queue))
 672                 wake_up(&fcc->flush_wait_queue);
 673
 674         if (fcc->f2fs_issue_flush) {
 675                 wait_for_completion(&cmd.wait);
 676                 atomic_dec(&fcc->queued_flush);
 677         } else {
 678                 struct llist_node *list;
 679
 680                 list = llist_del_all(&fcc->issue_list);
 681                 if (!list) {
 682                         wait_for_completion(&cmd.wait);
 683                         atomic_dec(&fcc->queued_flush);
 684                 } else {
 685                         struct flush_cmd *tmp, *next;
 686
 687                         ret = submit_flush_wait(sbi, ino);
 688
 689                         llist_for_each_entry_safe(tmp, next, list, llnode) {
 690                                 if (tmp == &cmd) {
 691                                         cmd.ret = ret;
 692                                         atomic_dec(&fcc->queued_flush);
 693                                         continue;
 694                                 }
 695                                 tmp->ret = ret;
 696                                 complete(&tmp->wait);
 697                         }
 698                 }
 699         }
 700
 701         return cmd.ret;
 702 }
 703
 704 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
 705 {
 706         dev_t dev = sbi->sb->s_bdev->bd_dev;
 707         struct flush_cmd_control *fcc;
 708         int err = 0;
 709
 710         if (SM_I(sbi)->fcc_info) {
 711                 fcc = SM_I(sbi)->fcc_info;
 712                 if (fcc->f2fs_issue_flush)
 713                         return err;
 714                 goto init_thread;
 715         }
 716
 717         fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
 718         if (!fcc)
 719                 return -ENOMEM;
 720         atomic_set(&fcc->issued_flush, 0);
 721         atomic_set(&fcc->queued_flush, 0);
 722         init_waitqueue_head(&fcc->flush_wait_queue);
 723         init_llist_head(&fcc->issue_list);
 724         SM_I(sbi)->fcc_info = fcc;
 725         if (!test_opt(sbi, FLUSH_MERGE))
 726                 return err;
 727
 728 init_thread:
 729         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 730                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 731         if (IS_ERR(fcc->f2fs_issue_flush)) {
 732                 err = PTR_ERR(fcc->f2fs_issue_flush);
 733                 kvfree(fcc);
 734                 SM_I(sbi)->fcc_info = NULL;
 735                 return err;
 736         }
 737
 738         return err;
 739 }
 740
 741 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 742 {
 743         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 744
 745         if (fcc && fcc->f2fs_issue_flush) {
 746                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 747
 748                 fcc->f2fs_issue_flush = NULL;
 749                 kthread_stop(flush_thread);
 750         }
 751         if (free) {
 752                 kvfree(fcc);
 753                 SM_I(sbi)->fcc_info = NULL;
 754         }
 755 }
 756
 757 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
 758 {
 759         int ret = 0, i;
 760
 761         if (!f2fs_is_multi_device(sbi))
 762                 return 0;
 763
 764         for (i = 1; i < sbi->s_ndevs; i++) {
 765                 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
 766                         continue;
 767                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 768                 if (ret)
 769                         break;
 770
 771                 spin_lock(&sbi->dev_lock);
 772                 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
 773                 spin_unlock(&sbi->dev_lock);
 774         }
 775
 776         return ret;
 777 }
 778
 779 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 780                 enum dirty_type dirty_type)
 781 {
 782         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 783
 784         /* need not be added */
 785         if (IS_CURSEG(sbi, segno))
 786                 return;
 787
 788         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 789                 dirty_i->nr_dirty[dirty_type]++;
 790
 791         if (dirty_type == DIRTY) {
 792                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 793                 enum dirty_type t = sentry->type;
 794
 795                 if (unlikely(t >= DIRTY)) {
 796                         f2fs_bug_on(sbi, 1);
 797                         return;
 798                 }
 799                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 800                         dirty_i->nr_dirty[t]++;
 801         }
 802 }
 803
 804 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 805                 enum dirty_type dirty_type)
 806 {
 807         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 808
 809         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 810                 dirty_i->nr_dirty[dirty_type]--;
 811
 812         if (dirty_type == DIRTY) {
 813                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 814                 enum dirty_type t = sentry->type;
 815
 816                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 817                         dirty_i->nr_dirty[t]--;
 818
 819                 if (get_valid_blocks(sbi, segno, true) == 0) {
 820                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 821                                                 dirty_i->victim_secmap);
 822 #ifdef CONFIG_F2FS_CHECK_FS
 823                         clear_bit(segno, SIT_I(sbi)->invalid_segmap);
 824 #endif
 825                 }
 826         }
 827 }
 828
 829 /*
 830  * Should not occur error such as -ENOMEM.
 831  * Adding dirty entry into seglist is not critical operation.
 832  * If a given segment is one of current working segments, it won't be added.
 833  */
 834 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 835 {
 836         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 837         unsigned short valid_blocks, ckpt_valid_blocks;
 838
 839         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 840                 return;
 841
 842         mutex_lock(&dirty_i->seglist_lock);
 843
 844         valid_blocks = get_valid_blocks(sbi, segno, false);
 845         ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno);
 846
 847         if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
 848                                 ckpt_valid_blocks == sbi->blocks_per_seg)) {
 849                 __locate_dirty_segment(sbi, segno, PRE);
 850                 __remove_dirty_segment(sbi, segno, DIRTY);
 851         } else if (valid_blocks < sbi->blocks_per_seg) {
 852                 __locate_dirty_segment(sbi, segno, DIRTY);
 853         } else {
 854                 /* Recovery routine with SSR needs this */
 855                 __remove_dirty_segment(sbi, segno, DIRTY);
 856         }
 857
 858         mutex_unlock(&dirty_i->seglist_lock);
 859 }
 860
 861 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
 862 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
 863 {
 864         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 865         unsigned int segno;
 866
 867         mutex_lock(&dirty_i->seglist_lock);
 868         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 869                 if (get_valid_blocks(sbi, segno, false))
 870                         continue;
 871                 if (IS_CURSEG(sbi, segno))
 872                         continue;
 873                 __locate_dirty_segment(sbi, segno, PRE);
 874                 __remove_dirty_segment(sbi, segno, DIRTY);
 875         }
 876         mutex_unlock(&dirty_i->seglist_lock);
 877 }
 878
 879 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
 880 {
 881         int ovp_hole_segs =
 882                 (overprovision_segments(sbi) - reserved_segments(sbi));
 883         block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
 884         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 885         block_t holes[2] = {0, 0};      /* DATA and NODE */
 886         block_t unusable;
 887         struct seg_entry *se;
 888         unsigned int segno;
 889
 890         mutex_lock(&dirty_i->seglist_lock);
 891         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 892                 se = get_seg_entry(sbi, segno);
 893                 if (IS_NODESEG(se->type))
 894                         holes[NODE] += sbi->blocks_per_seg - se->valid_blocks;
 895                 else
 896                         holes[DATA] += sbi->blocks_per_seg - se->valid_blocks;
 897         }
 898         mutex_unlock(&dirty_i->seglist_lock);
 899
 900         unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
 901         if (unusable > ovp_holes)
 902                 return unusable - ovp_holes;
 903         return 0;
 904 }
 905
 906 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
 907 {
 908         int ovp_hole_segs =
 909                 (overprovision_segments(sbi) - reserved_segments(sbi));
 910         if (unusable > F2FS_OPTION(sbi).unusable_cap)
 911                 return -EAGAIN;
 912         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
 913                 dirty_segments(sbi) > ovp_hole_segs)
 914                 return -EAGAIN;
 915         return 0;
 916 }
 917
 918 /* This is only used by SBI_CP_DISABLED */
 919 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
 920 {
 921         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 922         unsigned int segno = 0;
 923
 924         mutex_lock(&dirty_i->seglist_lock);
 925         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 926                 if (get_valid_blocks(sbi, segno, false))
 927                         continue;
 928                 if (get_ckpt_valid_blocks(sbi, segno))
 929                         continue;
 930                 mutex_unlock(&dirty_i->seglist_lock);
 931                 return segno;
 932         }
 933         mutex_unlock(&dirty_i->seglist_lock);
 934         return NULL_SEGNO;
 935 }
 936
 937 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 938                 struct block_device *bdev, block_t lstart,
 939                 block_t start, block_t len)
 940 {
 941         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 942         struct list_head *pend_list;
 943         struct discard_cmd *dc;
 944
 945         f2fs_bug_on(sbi, !len);
 946
 947         pend_list = &dcc->pend_list[plist_idx(len)];
 948
 949         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 950         INIT_LIST_HEAD(&dc->list);
 951         dc->bdev = bdev;
 952         dc->lstart = lstart;
 953         dc->start = start;
 954         dc->len = len;
 955         dc->ref = 0;
 956         dc->state = D_PREP;
 957         dc->queued = 0;
 958         dc->error = 0;
 959         init_completion(&dc->wait);
 960         list_add_tail(&dc->list, pend_list);
 961         spin_lock_init(&dc->lock);
 962         dc->bio_ref = 0;
 963         atomic_inc(&dcc->discard_cmd_cnt);
 964         dcc->undiscard_blks += len;
 965
 966         return dc;
 967 }
 968
 969 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 970                                 struct block_device *bdev, block_t lstart,
 971                                 block_t start, block_t len,
 972                                 struct rb_node *parent, struct rb_node **p,
 973                                 bool leftmost)
 974 {
 975         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 976         struct discard_cmd *dc;
 977
 978         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
 979
 980         rb_link_node(&dc->rb_node, parent, p);
 981         rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
 982
 983         return dc;
 984 }
 985
 986 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
 987                                                         struct discard_cmd *dc)
 988 {
 989         if (dc->state == D_DONE)
 990                 atomic_sub(dc->queued, &dcc->queued_discard);
 991
 992         list_del(&dc->list);
 993         rb_erase_cached(&dc->rb_node, &dcc->root);
 994         dcc->undiscard_blks -= dc->len;
 995
 996         kmem_cache_free(discard_cmd_slab, dc);
 997
 998         atomic_dec(&dcc->discard_cmd_cnt);
 999 }
1000
1001 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1002                                                         struct discard_cmd *dc)
1003 {
1004         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1005         unsigned long flags;
1006
1007         trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
1008
1009         spin_lock_irqsave(&dc->lock, flags);
1010         if (dc->bio_ref) {
1011                 spin_unlock_irqrestore(&dc->lock, flags);
1012                 return;
1013         }
1014         spin_unlock_irqrestore(&dc->lock, flags);
1015
1016         f2fs_bug_on(sbi, dc->ref);
1017
1018         if (dc->error == -EOPNOTSUPP)
1019                 dc->error = 0;
1020
1021         if (dc->error)
1022                 printk_ratelimited(
1023                         "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
1024                         KERN_INFO, sbi->sb->s_id,
1025                         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(sbi, 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 + 1 < 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         if (unlikely(f2fs_cp_error(sbi)))
2318                 return ERR_PTR(-EIO);
2319         return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno));
2320 }
2321
2322 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2323                                         void *src, block_t blk_addr)
2324 {
2325         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2326
2327         memcpy(page_address(page), src, PAGE_SIZE);
2328         set_page_dirty(page);
2329         f2fs_put_page(page, 1);
2330 }
2331
2332 static void write_sum_page(struct f2fs_sb_info *sbi,
2333                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
2334 {
2335         f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2336 }
2337
2338 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2339                                                 int type, block_t blk_addr)
2340 {
2341         struct curseg_info *curseg = CURSEG_I(sbi, type);
2342         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2343         struct f2fs_summary_block *src = curseg->sum_blk;
2344         struct f2fs_summary_block *dst;
2345
2346         dst = (struct f2fs_summary_block *)page_address(page);
2347         memset(dst, 0, PAGE_SIZE);
2348
2349         mutex_lock(&curseg->curseg_mutex);
2350
2351         down_read(&curseg->journal_rwsem);
2352         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2353         up_read(&curseg->journal_rwsem);
2354
2355         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2356         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2357
2358         mutex_unlock(&curseg->curseg_mutex);
2359
2360         set_page_dirty(page);
2361         f2fs_put_page(page, 1);
2362 }
2363
2364 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2365 {
2366         struct curseg_info *curseg = CURSEG_I(sbi, type);
2367         unsigned int segno = curseg->segno + 1;
2368         struct free_segmap_info *free_i = FREE_I(sbi);
2369
2370         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2371                 return !test_bit(segno, free_i->free_segmap);
2372         return 0;
2373 }
2374
2375 /*
2376  * Find a new segment from the free segments bitmap to right order
2377  * This function should be returned with success, otherwise BUG
2378  */
2379 static void get_new_segment(struct f2fs_sb_info *sbi,
2380                         unsigned int *newseg, bool new_sec, int dir)
2381 {
2382         struct free_segmap_info *free_i = FREE_I(sbi);
2383         unsigned int segno, secno, zoneno;
2384         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2385         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2386         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2387         unsigned int left_start = hint;
2388         bool init = true;
2389         int go_left = 0;
2390         int i;
2391
2392         spin_lock(&free_i->segmap_lock);
2393
2394         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2395                 segno = find_next_zero_bit(free_i->free_segmap,
2396                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2397                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2398                         goto got_it;
2399         }
2400 find_other_zone:
2401         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2402         if (secno >= MAIN_SECS(sbi)) {
2403                 if (dir == ALLOC_RIGHT) {
2404                         secno = find_next_zero_bit(free_i->free_secmap,
2405                                                         MAIN_SECS(sbi), 0);
2406                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2407                 } else {
2408                         go_left = 1;
2409                         left_start = hint - 1;
2410                 }
2411         }
2412         if (go_left == 0)
2413                 goto skip_left;
2414
2415         while (test_bit(left_start, free_i->free_secmap)) {
2416                 if (left_start > 0) {
2417                         left_start--;
2418                         continue;
2419                 }
2420                 left_start = find_next_zero_bit(free_i->free_secmap,
2421                                                         MAIN_SECS(sbi), 0);
2422                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2423                 break;
2424         }
2425         secno = left_start;
2426 skip_left:
2427         segno = GET_SEG_FROM_SEC(sbi, secno);
2428         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2429
2430         /* give up on finding another zone */
2431         if (!init)
2432                 goto got_it;
2433         if (sbi->secs_per_zone == 1)
2434                 goto got_it;
2435         if (zoneno == old_zoneno)
2436                 goto got_it;
2437         if (dir == ALLOC_LEFT) {
2438                 if (!go_left && zoneno + 1 >= total_zones)
2439                         goto got_it;
2440                 if (go_left && zoneno == 0)
2441                         goto got_it;
2442         }
2443         for (i = 0; i < NR_CURSEG_TYPE; i++)
2444                 if (CURSEG_I(sbi, i)->zone == zoneno)
2445                         break;
2446
2447         if (i < NR_CURSEG_TYPE) {
2448                 /* zone is in user, try another */
2449                 if (go_left)
2450                         hint = zoneno * sbi->secs_per_zone - 1;
2451                 else if (zoneno + 1 >= total_zones)
2452                         hint = 0;
2453                 else
2454                         hint = (zoneno + 1) * sbi->secs_per_zone;
2455                 init = false;
2456                 goto find_other_zone;
2457         }
2458 got_it:
2459         /* set it as dirty segment in free segmap */
2460         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2461         __set_inuse(sbi, segno);
2462         *newseg = segno;
2463         spin_unlock(&free_i->segmap_lock);
2464 }
2465
2466 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2467 {
2468         struct curseg_info *curseg = CURSEG_I(sbi, type);
2469         struct summary_footer *sum_footer;
2470
2471         curseg->segno = curseg->next_segno;
2472         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2473         curseg->next_blkoff = 0;
2474         curseg->next_segno = NULL_SEGNO;
2475
2476         sum_footer = &(curseg->sum_blk->footer);
2477         memset(sum_footer, 0, sizeof(struct summary_footer));
2478         if (IS_DATASEG(type))
2479                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2480         if (IS_NODESEG(type))
2481                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2482         __set_sit_entry_type(sbi, type, curseg->segno, modified);
2483 }
2484
2485 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2486 {
2487         /* if segs_per_sec is large than 1, we need to keep original policy. */
2488         if (__is_large_section(sbi))
2489                 return CURSEG_I(sbi, type)->segno;
2490
2491         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2492                 return 0;
2493
2494         if (test_opt(sbi, NOHEAP) &&
2495                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2496                 return 0;
2497
2498         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2499                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2500
2501         /* find segments from 0 to reuse freed segments */
2502         if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2503                 return 0;
2504
2505         return CURSEG_I(sbi, type)->segno;
2506 }
2507
2508 /*
2509  * Allocate a current working segment.
2510  * This function always allocates a free segment in LFS manner.
2511  */
2512 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2513 {
2514         struct curseg_info *curseg = CURSEG_I(sbi, type);
2515         unsigned int segno = curseg->segno;
2516         int dir = ALLOC_LEFT;
2517
2518         write_sum_page(sbi, curseg->sum_blk,
2519                                 GET_SUM_BLOCK(sbi, segno));
2520         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2521                 dir = ALLOC_RIGHT;
2522
2523         if (test_opt(sbi, NOHEAP))
2524                 dir = ALLOC_RIGHT;
2525
2526         segno = __get_next_segno(sbi, type);
2527         get_new_segment(sbi, &segno, new_sec, dir);
2528         curseg->next_segno = segno;
2529         reset_curseg(sbi, type, 1);
2530         curseg->alloc_type = LFS;
2531 }
2532
2533 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2534                         struct curseg_info *seg, block_t start)
2535 {
2536         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2537         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2538         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2539         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2540         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2541         int i, pos;
2542
2543         for (i = 0; i < entries; i++)
2544                 target_map[i] = ckpt_map[i] | cur_map[i];
2545
2546         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2547
2548         seg->next_blkoff = pos;
2549 }
2550
2551 /*
2552  * If a segment is written by LFS manner, next block offset is just obtained
2553  * by increasing the current block offset. However, if a segment is written by
2554  * SSR manner, next block offset obtained by calling __next_free_blkoff
2555  */
2556 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2557                                 struct curseg_info *seg)
2558 {
2559         if (seg->alloc_type == SSR)
2560                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2561         else
2562                 seg->next_blkoff++;
2563 }
2564
2565 /*
2566  * This function always allocates a used segment(from dirty seglist) by SSR
2567  * manner, so it should recover the existing segment information of valid blocks
2568  */
2569 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2570 {
2571         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2572         struct curseg_info *curseg = CURSEG_I(sbi, type);
2573         unsigned int new_segno = curseg->next_segno;
2574         struct f2fs_summary_block *sum_node;
2575         struct page *sum_page;
2576
2577         write_sum_page(sbi, curseg->sum_blk,
2578                                 GET_SUM_BLOCK(sbi, curseg->segno));
2579         __set_test_and_inuse(sbi, new_segno);
2580
2581         mutex_lock(&dirty_i->seglist_lock);
2582         __remove_dirty_segment(sbi, new_segno, PRE);
2583         __remove_dirty_segment(sbi, new_segno, DIRTY);
2584         mutex_unlock(&dirty_i->seglist_lock);
2585
2586         reset_curseg(sbi, type, 1);
2587         curseg->alloc_type = SSR;
2588         __next_free_blkoff(sbi, curseg, 0);
2589
2590         sum_page = f2fs_get_sum_page(sbi, new_segno);
2591         if (IS_ERR(sum_page)) {
2592                 /* GC won't be able to use stale summary pages by cp_error */
2593                 memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
2594                 return;
2595         }
2596         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2597         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2598         f2fs_put_page(sum_page, 1);
2599 }
2600
2601 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2602 {
2603         struct curseg_info *curseg = CURSEG_I(sbi, type);
2604         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2605         unsigned segno = NULL_SEGNO;
2606         int i, cnt;
2607         bool reversed = false;
2608
2609         /* f2fs_need_SSR() already forces to do this */
2610         if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2611                 curseg->next_segno = segno;
2612                 return 1;
2613         }
2614
2615         /* For node segments, let's do SSR more intensively */
2616         if (IS_NODESEG(type)) {
2617                 if (type >= CURSEG_WARM_NODE) {
2618                         reversed = true;
2619                         i = CURSEG_COLD_NODE;
2620                 } else {
2621                         i = CURSEG_HOT_NODE;
2622                 }
2623                 cnt = NR_CURSEG_NODE_TYPE;
2624         } else {
2625                 if (type >= CURSEG_WARM_DATA) {
2626                         reversed = true;
2627                         i = CURSEG_COLD_DATA;
2628                 } else {
2629                         i = CURSEG_HOT_DATA;
2630                 }
2631                 cnt = NR_CURSEG_DATA_TYPE;
2632         }
2633
2634         for (; cnt-- > 0; reversed ? i-- : i++) {
2635                 if (i == type)
2636                         continue;
2637                 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2638                         curseg->next_segno = segno;
2639                         return 1;
2640                 }
2641         }
2642
2643         /* find valid_blocks=0 in dirty list */
2644         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2645                 segno = get_free_segment(sbi);
2646                 if (segno != NULL_SEGNO) {
2647                         curseg->next_segno = segno;
2648                         return 1;
2649                 }
2650         }
2651         return 0;
2652 }
2653
2654 /*
2655  * flush out current segment and replace it with new segment
2656  * This function should be returned with success, otherwise BUG
2657  */
2658 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2659                                                 int type, bool force)
2660 {
2661         struct curseg_info *curseg = CURSEG_I(sbi, type);
2662
2663         if (force)
2664                 new_curseg(sbi, type, true);
2665         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2666                                         type == CURSEG_WARM_NODE)
2667                 new_curseg(sbi, type, false);
2668         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2669                         likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2670                 new_curseg(sbi, type, false);
2671         else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2672                 change_curseg(sbi, type);
2673         else
2674                 new_curseg(sbi, type, false);
2675
2676         stat_inc_seg_type(sbi, curseg);
2677 }
2678
2679 void allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2680                                         unsigned int start, unsigned int end)
2681 {
2682         struct curseg_info *curseg = CURSEG_I(sbi, type);
2683         unsigned int segno;
2684
2685         down_read(&SM_I(sbi)->curseg_lock);
2686         mutex_lock(&curseg->curseg_mutex);
2687         down_write(&SIT_I(sbi)->sentry_lock);
2688
2689         segno = CURSEG_I(sbi, type)->segno;
2690         if (segno < start || segno > end)
2691                 goto unlock;
2692
2693         if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2694                 change_curseg(sbi, type);
2695         else
2696                 new_curseg(sbi, type, true);
2697
2698         stat_inc_seg_type(sbi, curseg);
2699
2700         locate_dirty_segment(sbi, segno);
2701 unlock:
2702         up_write(&SIT_I(sbi)->sentry_lock);
2703
2704         if (segno != curseg->segno)
2705                 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2706                             type, segno, curseg->segno);
2707
2708         mutex_unlock(&curseg->curseg_mutex);
2709         up_read(&SM_I(sbi)->curseg_lock);
2710 }
2711
2712 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2713 {
2714         struct curseg_info *curseg;
2715         unsigned int old_segno;
2716         int i;
2717
2718         down_write(&SIT_I(sbi)->sentry_lock);
2719
2720         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2721                 curseg = CURSEG_I(sbi, i);
2722                 old_segno = curseg->segno;
2723                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2724                 locate_dirty_segment(sbi, old_segno);
2725         }
2726
2727         up_write(&SIT_I(sbi)->sentry_lock);
2728 }
2729
2730 static const struct segment_allocation default_salloc_ops = {
2731         .allocate_segment = allocate_segment_by_default,
2732 };
2733
2734 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2735                                                 struct cp_control *cpc)
2736 {
2737         __u64 trim_start = cpc->trim_start;
2738         bool has_candidate = false;
2739
2740         down_write(&SIT_I(sbi)->sentry_lock);
2741         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2742                 if (add_discard_addrs(sbi, cpc, true)) {
2743                         has_candidate = true;
2744                         break;
2745                 }
2746         }
2747         up_write(&SIT_I(sbi)->sentry_lock);
2748
2749         cpc->trim_start = trim_start;
2750         return has_candidate;
2751 }
2752
2753 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2754                                         struct discard_policy *dpolicy,
2755                                         unsigned int start, unsigned int end)
2756 {
2757         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2758         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2759         struct rb_node **insert_p = NULL, *insert_parent = NULL;
2760         struct discard_cmd *dc;
2761         struct blk_plug plug;
2762         int issued;
2763         unsigned int trimmed = 0;
2764
2765 next:
2766         issued = 0;
2767
2768         mutex_lock(&dcc->cmd_lock);
2769         if (unlikely(dcc->rbtree_check))
2770                 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2771                                                                 &dcc->root));
2772
2773         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2774                                         NULL, start,
2775                                         (struct rb_entry **)&prev_dc,
2776                                         (struct rb_entry **)&next_dc,
2777                                         &insert_p, &insert_parent, true, NULL);
2778         if (!dc)
2779                 dc = next_dc;
2780
2781         blk_start_plug(&plug);
2782
2783         while (dc && dc->lstart <= end) {
2784                 struct rb_node *node;
2785                 int err = 0;
2786
2787                 if (dc->len < dpolicy->granularity)
2788                         goto skip;
2789
2790                 if (dc->state != D_PREP) {
2791                         list_move_tail(&dc->list, &dcc->fstrim_list);
2792                         goto skip;
2793                 }
2794
2795                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2796
2797                 if (issued >= dpolicy->max_requests) {
2798                         start = dc->lstart + dc->len;
2799
2800                         if (err)
2801                                 __remove_discard_cmd(sbi, dc);
2802
2803                         blk_finish_plug(&plug);
2804                         mutex_unlock(&dcc->cmd_lock);
2805                         trimmed += __wait_all_discard_cmd(sbi, NULL);
2806                         congestion_wait(BLK_RW_ASYNC, HZ/50);
2807                         goto next;
2808                 }
2809 skip:
2810                 node = rb_next(&dc->rb_node);
2811                 if (err)
2812                         __remove_discard_cmd(sbi, dc);
2813                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2814
2815                 if (fatal_signal_pending(current))
2816                         break;
2817         }
2818
2819         blk_finish_plug(&plug);
2820         mutex_unlock(&dcc->cmd_lock);
2821
2822         return trimmed;
2823 }
2824
2825 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2826 {
2827         __u64 start = F2FS_BYTES_TO_BLK(range->start);
2828         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2829         unsigned int start_segno, end_segno;
2830         block_t start_block, end_block;
2831         struct cp_control cpc;
2832         struct discard_policy dpolicy;
2833         unsigned long long trimmed = 0;
2834         int err = 0;
2835         bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
2836
2837         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2838                 return -EINVAL;
2839
2840         if (end < MAIN_BLKADDR(sbi))
2841                 goto out;
2842
2843         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2844                 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
2845                 return -EFSCORRUPTED;
2846         }
2847
2848         /* start/end segment number in main_area */
2849         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2850         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2851                                                 GET_SEGNO(sbi, end);
2852         if (need_align) {
2853                 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2854                 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2855         }
2856
2857         cpc.reason = CP_DISCARD;
2858         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2859         cpc.trim_start = start_segno;
2860         cpc.trim_end = end_segno;
2861
2862         if (sbi->discard_blks == 0)
2863                 goto out;
2864
2865         mutex_lock(&sbi->gc_mutex);
2866         err = f2fs_write_checkpoint(sbi, &cpc);
2867         mutex_unlock(&sbi->gc_mutex);
2868         if (err)
2869                 goto out;
2870
2871         /*
2872          * We filed discard candidates, but actually we don't need to wait for
2873          * all of them, since they'll be issued in idle time along with runtime
2874          * discard option. User configuration looks like using runtime discard
2875          * or periodic fstrim instead of it.
2876          */
2877         if (f2fs_realtime_discard_enable(sbi))
2878                 goto out;
2879
2880         start_block = START_BLOCK(sbi, start_segno);
2881         end_block = START_BLOCK(sbi, end_segno + 1);
2882
2883         __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2884         trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2885                                         start_block, end_block);
2886
2887         trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
2888                                         start_block, end_block);
2889 out:
2890         if (!err)
2891                 range->len = F2FS_BLK_TO_BYTES(trimmed);
2892         return err;
2893 }
2894
2895 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2896 {
2897         struct curseg_info *curseg = CURSEG_I(sbi, type);
2898         if (curseg->next_blkoff < sbi->blocks_per_seg)
2899                 return true;
2900         return false;
2901 }
2902
2903 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
2904 {
2905         switch (hint) {
2906         case WRITE_LIFE_SHORT:
2907                 return CURSEG_HOT_DATA;
2908         case WRITE_LIFE_EXTREME:
2909                 return CURSEG_COLD_DATA;
2910         default:
2911                 return CURSEG_WARM_DATA;
2912         }
2913 }
2914
2915 /* This returns write hints for each segment type. This hints will be
2916  * passed down to block layer. There are mapping tables which depend on
2917  * the mount option 'whint_mode'.
2918  *
2919  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2920  *
2921  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2922  *
2923  * User                  F2FS                     Block
2924  * ----                  ----                     -----
2925  *                       META                     WRITE_LIFE_NOT_SET
2926  *                       HOT_NODE                 "
2927  *                       WARM_NODE                "
2928  *                       COLD_NODE                "
2929  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2930  * extension list        "                        "
2931  *
2932  * -- buffered io
2933  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2934  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2935  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2936  * WRITE_LIFE_NONE       "                        "
2937  * WRITE_LIFE_MEDIUM     "                        "
2938  * WRITE_LIFE_LONG       "                        "
2939  *
2940  * -- direct io
2941  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2942  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2943  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2944  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2945  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2946  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2947  *
2948  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2949  *
2950  * User                  F2FS                     Block
2951  * ----                  ----                     -----
2952  *                       META                     WRITE_LIFE_MEDIUM;
2953  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
2954  *                       WARM_NODE                "
2955  *                       COLD_NODE                WRITE_LIFE_NONE
2956  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2957  * extension list        "                        "
2958  *
2959  * -- buffered io
2960  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2961  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2962  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
2963  * WRITE_LIFE_NONE       "                        "
2964  * WRITE_LIFE_MEDIUM     "                        "
2965  * WRITE_LIFE_LONG       "                        "
2966  *
2967  * -- direct io
2968  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2969  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2970  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2971  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2972  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2973  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2974  */
2975
2976 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2977                                 enum page_type type, enum temp_type temp)
2978 {
2979         if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2980                 if (type == DATA) {
2981                         if (temp == WARM)
2982                                 return WRITE_LIFE_NOT_SET;
2983                         else if (temp == HOT)
2984                                 return WRITE_LIFE_SHORT;
2985                         else if (temp == COLD)
2986                                 return WRITE_LIFE_EXTREME;
2987                 } else {
2988                         return WRITE_LIFE_NOT_SET;
2989                 }
2990         } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2991                 if (type == DATA) {
2992                         if (temp == WARM)
2993                                 return WRITE_LIFE_LONG;
2994                         else if (temp == HOT)
2995                                 return WRITE_LIFE_SHORT;
2996                         else if (temp == COLD)
2997                                 return WRITE_LIFE_EXTREME;
2998                 } else if (type == NODE) {
2999                         if (temp == WARM || temp == HOT)
3000                                 return WRITE_LIFE_NOT_SET;
3001                         else if (temp == COLD)
3002                                 return WRITE_LIFE_NONE;
3003                 } else if (type == META) {
3004                         return WRITE_LIFE_MEDIUM;
3005                 }
3006         }
3007         return WRITE_LIFE_NOT_SET;
3008 }
3009
3010 static int __get_segment_type_2(struct f2fs_io_info *fio)
3011 {
3012         if (fio->type == DATA)
3013                 return CURSEG_HOT_DATA;
3014         else
3015                 return CURSEG_HOT_NODE;
3016 }
3017
3018 static int __get_segment_type_4(struct f2fs_io_info *fio)
3019 {
3020         if (fio->type == DATA) {
3021                 struct inode *inode = fio->page->mapping->host;
3022
3023                 if (S_ISDIR(inode->i_mode))
3024                         return CURSEG_HOT_DATA;
3025                 else
3026                         return CURSEG_COLD_DATA;
3027         } else {
3028                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3029                         return CURSEG_WARM_NODE;
3030                 else
3031                         return CURSEG_COLD_NODE;
3032         }
3033 }
3034
3035 static int __get_segment_type_6(struct f2fs_io_info *fio)
3036 {
3037         if (fio->type == DATA) {
3038                 struct inode *inode = fio->page->mapping->host;
3039
3040                 if (is_cold_data(fio->page) || file_is_cold(inode))
3041                         return CURSEG_COLD_DATA;
3042                 if (file_is_hot(inode) ||
3043                                 is_inode_flag_set(inode, FI_HOT_DATA) ||
3044                                 f2fs_is_atomic_file(inode) ||
3045                                 f2fs_is_volatile_file(inode))
3046                         return CURSEG_HOT_DATA;
3047                 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3048         } else {
3049                 if (IS_DNODE(fio->page))
3050                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3051                                                 CURSEG_HOT_NODE;
3052                 return CURSEG_COLD_NODE;
3053         }
3054 }
3055
3056 static int __get_segment_type(struct f2fs_io_info *fio)
3057 {
3058         int type = 0;
3059
3060         switch (F2FS_OPTION(fio->sbi).active_logs) {
3061         case 2:
3062                 type = __get_segment_type_2(fio);
3063                 break;
3064         case 4:
3065                 type = __get_segment_type_4(fio);
3066                 break;
3067         case 6:
3068                 type = __get_segment_type_6(fio);
3069                 break;
3070         default:
3071                 f2fs_bug_on(fio->sbi, true);
3072         }
3073
3074         if (IS_HOT(type))
3075                 fio->temp = HOT;
3076         else if (IS_WARM(type))
3077                 fio->temp = WARM;
3078         else
3079                 fio->temp = COLD;
3080         return type;
3081 }
3082
3083 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3084                 block_t old_blkaddr, block_t *new_blkaddr,
3085                 struct f2fs_summary *sum, int type,
3086                 struct f2fs_io_info *fio, bool add_list)
3087 {
3088         struct sit_info *sit_i = SIT_I(sbi);
3089         struct curseg_info *curseg = CURSEG_I(sbi, type);
3090
3091         down_read(&SM_I(sbi)->curseg_lock);
3092
3093         mutex_lock(&curseg->curseg_mutex);
3094         down_write(&sit_i->sentry_lock);
3095
3096         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3097
3098         f2fs_wait_discard_bio(sbi, *new_blkaddr);
3099
3100         /*
3101          * __add_sum_entry should be resided under the curseg_mutex
3102          * because, this function updates a summary entry in the
3103          * current summary block.
3104          */
3105         __add_sum_entry(sbi, type, sum);
3106
3107         __refresh_next_blkoff(sbi, curseg);
3108
3109         stat_inc_block_count(sbi, curseg);
3110
3111         /*
3112          * SIT information should be updated before segment allocation,
3113          * since SSR needs latest valid block information.
3114          */
3115         update_sit_entry(sbi, *new_blkaddr, 1);
3116         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3117                 update_sit_entry(sbi, old_blkaddr, -1);
3118
3119         if (!__has_curseg_space(sbi, type))
3120                 sit_i->s_ops->allocate_segment(sbi, type, false);
3121
3122         /*
3123          * segment dirty status should be updated after segment allocation,
3124          * so we just need to update status only one time after previous
3125          * segment being closed.
3126          */
3127         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3128         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3129
3130         up_write(&sit_i->sentry_lock);
3131
3132         if (page && IS_NODESEG(type)) {
3133                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3134
3135                 f2fs_inode_chksum_set(sbi, page);
3136         }
3137
3138         if (F2FS_IO_ALIGNED(sbi))
3139                 fio->retry = false;
3140
3141         if (add_list) {
3142                 struct f2fs_bio_info *io;
3143
3144                 INIT_LIST_HEAD(&fio->list);
3145                 fio->in_list = true;
3146                 io = sbi->write_io[fio->type] + fio->temp;
3147                 spin_lock(&io->io_lock);
3148                 list_add_tail(&fio->list, &io->io_list);
3149                 spin_unlock(&io->io_lock);
3150         }
3151
3152         mutex_unlock(&curseg->curseg_mutex);
3153
3154         up_read(&SM_I(sbi)->curseg_lock);
3155 }
3156
3157 static void update_device_state(struct f2fs_io_info *fio)
3158 {
3159         struct f2fs_sb_info *sbi = fio->sbi;
3160         unsigned int devidx;
3161
3162         if (!f2fs_is_multi_device(sbi))
3163                 return;
3164
3165         devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3166
3167         /* update device state for fsync */
3168         f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
3169
3170         /* update device state for checkpoint */
3171         if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3172                 spin_lock(&sbi->dev_lock);
3173                 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3174                 spin_unlock(&sbi->dev_lock);
3175         }
3176 }
3177
3178 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3179 {
3180         int type = __get_segment_type(fio);
3181         bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
3182
3183         if (keep_order)
3184                 down_read(&fio->sbi->io_order_lock);
3185 reallocate:
3186         f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3187                         &fio->new_blkaddr, sum, type, fio, true);
3188         if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3189                 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3190                                         fio->old_blkaddr, fio->old_blkaddr);
3191
3192         /* writeout dirty page into bdev */
3193         f2fs_submit_page_write(fio);
3194         if (fio->retry) {
3195                 fio->old_blkaddr = fio->new_blkaddr;
3196                 goto reallocate;
3197         }
3198
3199         update_device_state(fio);
3200
3201         if (keep_order)
3202                 up_read(&fio->sbi->io_order_lock);
3203 }
3204
3205 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3206                                         enum iostat_type io_type)
3207 {
3208         struct f2fs_io_info fio = {
3209                 .sbi = sbi,
3210                 .type = META,
3211                 .temp = HOT,
3212                 .op = REQ_OP_WRITE,
3213                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3214                 .old_blkaddr = page->index,
3215                 .new_blkaddr = page->index,
3216                 .page = page,
3217                 .encrypted_page = NULL,
3218                 .in_list = false,
3219         };
3220
3221         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3222                 fio.op_flags &= ~REQ_META;
3223
3224         set_page_writeback(page);
3225         ClearPageError(page);
3226         f2fs_submit_page_write(&fio);
3227
3228         stat_inc_meta_count(sbi, page->index);
3229         f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3230 }
3231
3232 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3233 {
3234         struct f2fs_summary sum;
3235
3236         set_summary(&sum, nid, 0, 0);
3237         do_write_page(&sum, fio);
3238
3239         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3240 }
3241
3242 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3243                                         struct f2fs_io_info *fio)
3244 {
3245         struct f2fs_sb_info *sbi = fio->sbi;
3246         struct f2fs_summary sum;
3247
3248         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3249         set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3250         do_write_page(&sum, fio);
3251         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3252
3253         f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3254 }
3255
3256 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3257 {
3258         int err;
3259         struct f2fs_sb_info *sbi = fio->sbi;
3260         unsigned int segno;
3261
3262         fio->new_blkaddr = fio->old_blkaddr;
3263         /* i/o temperature is needed for passing down write hints */
3264         __get_segment_type(fio);
3265
3266         segno = GET_SEGNO(sbi, fio->new_blkaddr);
3267
3268         if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3269                 set_sbi_flag(sbi, SBI_NEED_FSCK);
3270                 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3271                           __func__, segno);
3272                 return -EFSCORRUPTED;
3273         }
3274
3275         stat_inc_inplace_blocks(fio->sbi);
3276
3277         if (fio->bio)
3278                 err = f2fs_merge_page_bio(fio);
3279         else
3280                 err = f2fs_submit_page_bio(fio);
3281         if (!err) {
3282                 update_device_state(fio);
3283                 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3284         }
3285
3286         return err;
3287 }
3288
3289 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3290                                                 unsigned int segno)
3291 {
3292         int i;
3293
3294         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3295                 if (CURSEG_I(sbi, i)->segno == segno)
3296                         break;
3297         }
3298         return i;
3299 }
3300
3301 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3302                                 block_t old_blkaddr, block_t new_blkaddr,
3303                                 bool recover_curseg, bool recover_newaddr)
3304 {
3305         struct sit_info *sit_i = SIT_I(sbi);
3306         struct curseg_info *curseg;
3307         unsigned int segno, old_cursegno;
3308         struct seg_entry *se;
3309         int type;
3310         unsigned short old_blkoff;
3311
3312         segno = GET_SEGNO(sbi, new_blkaddr);
3313         se = get_seg_entry(sbi, segno);
3314         type = se->type;
3315
3316         down_write(&SM_I(sbi)->curseg_lock);
3317
3318         if (!recover_curseg) {
3319                 /* for recovery flow */
3320                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3321                         if (old_blkaddr == NULL_ADDR)
3322                                 type = CURSEG_COLD_DATA;
3323                         else
3324                                 type = CURSEG_WARM_DATA;
3325                 }
3326         } else {
3327                 if (IS_CURSEG(sbi, segno)) {
3328                         /* se->type is volatile as SSR allocation */
3329                         type = __f2fs_get_curseg(sbi, segno);
3330                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3331                 } else {
3332                         type = CURSEG_WARM_DATA;
3333                 }
3334         }
3335
3336         f2fs_bug_on(sbi, !IS_DATASEG(type));
3337         curseg = CURSEG_I(sbi, type);
3338
3339         mutex_lock(&curseg->curseg_mutex);
3340         down_write(&sit_i->sentry_lock);
3341
3342         old_cursegno = curseg->segno;
3343         old_blkoff = curseg->next_blkoff;
3344
3345         /* change the current segment */
3346         if (segno != curseg->segno) {
3347                 curseg->next_segno = segno;
3348                 change_curseg(sbi, type);
3349         }
3350
3351         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3352         __add_sum_entry(sbi, type, sum);
3353
3354         if (!recover_curseg || recover_newaddr)
3355                 update_sit_entry(sbi, new_blkaddr, 1);
3356         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3357                 invalidate_mapping_pages(META_MAPPING(sbi),
3358                                         old_blkaddr, old_blkaddr);
3359                 update_sit_entry(sbi, old_blkaddr, -1);
3360         }
3361
3362         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3363         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3364
3365         locate_dirty_segment(sbi, old_cursegno);
3366
3367         if (recover_curseg) {
3368                 if (old_cursegno != curseg->segno) {
3369                         curseg->next_segno = old_cursegno;
3370                         change_curseg(sbi, type);
3371                 }
3372                 curseg->next_blkoff = old_blkoff;
3373         }
3374
3375         up_write(&sit_i->sentry_lock);
3376         mutex_unlock(&curseg->curseg_mutex);
3377         up_write(&SM_I(sbi)->curseg_lock);
3378 }
3379
3380 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3381                                 block_t old_addr, block_t new_addr,
3382                                 unsigned char version, bool recover_curseg,
3383                                 bool recover_newaddr)
3384 {
3385         struct f2fs_summary sum;
3386
3387         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3388
3389         f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3390                                         recover_curseg, recover_newaddr);
3391
3392         f2fs_update_data_blkaddr(dn, new_addr);
3393 }
3394
3395 void f2fs_wait_on_page_writeback(struct page *page,
3396                                 enum page_type type, bool ordered, bool locked)
3397 {
3398         if (PageWriteback(page)) {
3399                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3400
3401                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3402                 if (ordered) {
3403                         wait_on_page_writeback(page);
3404                         f2fs_bug_on(sbi, locked && PageWriteback(page));
3405                 } else {
3406                         wait_for_stable_page(page);
3407                 }
3408         }
3409 }
3410
3411 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3412 {
3413         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3414         struct page *cpage;
3415
3416         if (!f2fs_post_read_required(inode))
3417                 return;
3418
3419         if (!__is_valid_data_blkaddr(blkaddr))
3420                 return;
3421
3422         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3423         if (cpage) {
3424                 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3425                 f2fs_put_page(cpage, 1);
3426         }
3427 }
3428
3429 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3430                                                                 block_t len)
3431 {
3432         block_t i;
3433
3434         for (i = 0; i < len; i++)
3435                 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3436 }
3437
3438 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3439 {
3440         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3441         struct curseg_info *seg_i;
3442         unsigned char *kaddr;
3443         struct page *page;
3444         block_t start;
3445         int i, j, offset;
3446
3447         start = start_sum_block(sbi);
3448
3449         page = f2fs_get_meta_page(sbi, start++);
3450         if (IS_ERR(page))
3451                 return PTR_ERR(page);
3452         kaddr = (unsigned char *)page_address(page);
3453
3454         /* Step 1: restore nat cache */
3455         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3456         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3457
3458         /* Step 2: restore sit cache */
3459         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3460         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3461         offset = 2 * SUM_JOURNAL_SIZE;
3462
3463         /* Step 3: restore summary entries */
3464         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3465                 unsigned short blk_off;
3466                 unsigned int segno;
3467
3468                 seg_i = CURSEG_I(sbi, i);
3469                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3470                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3471                 seg_i->next_segno = segno;
3472                 reset_curseg(sbi, i, 0);
3473                 seg_i->alloc_type = ckpt->alloc_type[i];
3474                 seg_i->next_blkoff = blk_off;
3475
3476                 if (seg_i->alloc_type == SSR)
3477                         blk_off = sbi->blocks_per_seg;
3478
3479                 for (j = 0; j < blk_off; j++) {
3480                         struct f2fs_summary *s;
3481                         s = (struct f2fs_summary *)(kaddr + offset);
3482                         seg_i->sum_blk->entries[j] = *s;
3483                         offset += SUMMARY_SIZE;
3484                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3485                                                 SUM_FOOTER_SIZE)
3486                                 continue;
3487
3488                         f2fs_put_page(page, 1);
3489                         page = NULL;
3490
3491                         page = f2fs_get_meta_page(sbi, start++);
3492                         if (IS_ERR(page))
3493                                 return PTR_ERR(page);
3494                         kaddr = (unsigned char *)page_address(page);
3495                         offset = 0;
3496                 }
3497         }
3498         f2fs_put_page(page, 1);
3499         return 0;
3500 }
3501
3502 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3503 {
3504         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3505         struct f2fs_summary_block *sum;
3506         struct curseg_info *curseg;
3507         struct page *new;
3508         unsigned short blk_off;
3509         unsigned int segno = 0;
3510         block_t blk_addr = 0;
3511         int err = 0;
3512
3513         /* get segment number and block addr */
3514         if (IS_DATASEG(type)) {
3515                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3516                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3517                                                         CURSEG_HOT_DATA]);
3518                 if (__exist_node_summaries(sbi))
3519                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3520                 else
3521                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3522         } else {
3523                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3524                                                         CURSEG_HOT_NODE]);
3525                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3526                                                         CURSEG_HOT_NODE]);
3527                 if (__exist_node_summaries(sbi))
3528                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3529                                                         type - CURSEG_HOT_NODE);
3530                 else
3531                         blk_addr = GET_SUM_BLOCK(sbi, segno);
3532         }
3533
3534         new = f2fs_get_meta_page(sbi, blk_addr);
3535         if (IS_ERR(new))
3536                 return PTR_ERR(new);
3537         sum = (struct f2fs_summary_block *)page_address(new);
3538
3539         if (IS_NODESEG(type)) {
3540                 if (__exist_node_summaries(sbi)) {
3541                         struct f2fs_summary *ns = &sum->entries[0];
3542                         int i;
3543                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3544                                 ns->version = 0;
3545                                 ns->ofs_in_node = 0;
3546                         }
3547                 } else {
3548                         err = f2fs_restore_node_summary(sbi, segno, sum);
3549                         if (err)
3550                                 goto out;
3551                 }
3552         }
3553
3554         /* set uncompleted segment to curseg */
3555         curseg = CURSEG_I(sbi, type);
3556         mutex_lock(&curseg->curseg_mutex);
3557
3558         /* update journal info */
3559         down_write(&curseg->journal_rwsem);
3560         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3561         up_write(&curseg->journal_rwsem);
3562
3563         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3564         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3565         curseg->next_segno = segno;
3566         reset_curseg(sbi, type, 0);
3567         curseg->alloc_type = ckpt->alloc_type[type];
3568         curseg->next_blkoff = blk_off;
3569         mutex_unlock(&curseg->curseg_mutex);
3570 out:
3571         f2fs_put_page(new, 1);
3572         return err;
3573 }
3574
3575 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3576 {
3577         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3578         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3579         int type = CURSEG_HOT_DATA;
3580         int err;
3581
3582         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3583                 int npages = f2fs_npages_for_summary_flush(sbi, true);
3584
3585                 if (npages >= 2)
3586                         f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3587                                                         META_CP, true);
3588
3589                 /* restore for compacted data summary */
3590                 err = read_compacted_summaries(sbi);
3591                 if (err)
3592                         return err;
3593                 type = CURSEG_HOT_NODE;
3594         }
3595
3596         if (__exist_node_summaries(sbi))
3597                 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3598                                         NR_CURSEG_TYPE - type, META_CP, true);
3599
3600         for (; type <= CURSEG_COLD_NODE; type++) {
3601                 err = read_normal_summaries(sbi, type);
3602                 if (err)
3603                         return err;
3604         }
3605
3606         /* sanity check for summary blocks */
3607         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3608                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3609                 f2fs_err(sbi, "invalid journal entries nats %u sits %u\n",
3610                          nats_in_cursum(nat_j), sits_in_cursum(sit_j));
3611                 return -EINVAL;
3612         }
3613
3614         return 0;
3615 }
3616
3617 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3618 {
3619         struct page *page;
3620         unsigned char *kaddr;
3621         struct f2fs_summary *summary;
3622         struct curseg_info *seg_i;
3623         int written_size = 0;
3624         int i, j;
3625
3626         page = f2fs_grab_meta_page(sbi, blkaddr++);
3627         kaddr = (unsigned char *)page_address(page);
3628         memset(kaddr, 0, PAGE_SIZE);
3629
3630         /* Step 1: write nat cache */
3631         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3632         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3633         written_size += SUM_JOURNAL_SIZE;
3634
3635         /* Step 2: write sit cache */
3636         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3637         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3638         written_size += SUM_JOURNAL_SIZE;
3639
3640         /* Step 3: write summary entries */
3641         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3642                 unsigned short blkoff;
3643                 seg_i = CURSEG_I(sbi, i);
3644                 if (sbi->ckpt->alloc_type[i] == SSR)
3645                         blkoff = sbi->blocks_per_seg;
3646                 else
3647                         blkoff = curseg_blkoff(sbi, i);
3648
3649                 for (j = 0; j < blkoff; j++) {
3650                         if (!page) {
3651                                 page = f2fs_grab_meta_page(sbi, blkaddr++);
3652                                 kaddr = (unsigned char *)page_address(page);
3653                                 memset(kaddr, 0, PAGE_SIZE);
3654                                 written_size = 0;
3655                         }
3656                         summary = (struct f2fs_summary *)(kaddr + written_size);
3657                         *summary = seg_i->sum_blk->entries[j];
3658                         written_size += SUMMARY_SIZE;
3659
3660                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3661                                                         SUM_FOOTER_SIZE)
3662                                 continue;
3663
3664                         set_page_dirty(page);
3665                         f2fs_put_page(page, 1);
3666                         page = NULL;
3667                 }
3668         }
3669         if (page) {
3670                 set_page_dirty(page);
3671                 f2fs_put_page(page, 1);
3672         }
3673 }
3674
3675 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3676                                         block_t blkaddr, int type)
3677 {
3678         int i, end;
3679         if (IS_DATASEG(type))
3680                 end = type + NR_CURSEG_DATA_TYPE;
3681         else
3682                 end = type + NR_CURSEG_NODE_TYPE;
3683
3684         for (i = type; i < end; i++)
3685                 write_current_sum_page(sbi, i, blkaddr + (i - type));
3686 }
3687
3688 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3689 {
3690         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3691                 write_compacted_summaries(sbi, start_blk);
3692         else
3693                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3694 }
3695
3696 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3697 {
3698         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3699 }
3700
3701 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3702                                         unsigned int val, int alloc)
3703 {
3704         int i;
3705
3706         if (type == NAT_JOURNAL) {
3707                 for (i = 0; i < nats_in_cursum(journal); i++) {
3708                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3709                                 return i;
3710                 }
3711                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3712                         return update_nats_in_cursum(journal, 1);
3713         } else if (type == SIT_JOURNAL) {
3714                 for (i = 0; i < sits_in_cursum(journal); i++)
3715                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3716                                 return i;
3717                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3718                         return update_sits_in_cursum(journal, 1);
3719         }
3720         return -1;
3721 }
3722
3723 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3724                                         unsigned int segno)
3725 {
3726         return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
3727 }
3728
3729 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3730                                         unsigned int start)
3731 {
3732         struct sit_info *sit_i = SIT_I(sbi);
3733         struct page *page;
3734         pgoff_t src_off, dst_off;
3735
3736         src_off = current_sit_addr(sbi, start);
3737         dst_off = next_sit_addr(sbi, src_off);
3738
3739         page = f2fs_grab_meta_page(sbi, dst_off);
3740         seg_info_to_sit_page(sbi, page, start);
3741
3742         set_page_dirty(page);
3743         set_to_next_sit(sit_i, start);
3744
3745         return page;
3746 }
3747
3748 static struct sit_entry_set *grab_sit_entry_set(void)
3749 {
3750         struct sit_entry_set *ses =
3751                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3752
3753         ses->entry_cnt = 0;
3754         INIT_LIST_HEAD(&ses->set_list);
3755         return ses;
3756 }
3757
3758 static void release_sit_entry_set(struct sit_entry_set *ses)
3759 {
3760         list_del(&ses->set_list);
3761         kmem_cache_free(sit_entry_set_slab, ses);
3762 }
3763
3764 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3765                                                 struct list_head *head)
3766 {
3767         struct sit_entry_set *next = ses;
3768
3769         if (list_is_last(&ses->set_list, head))
3770                 return;
3771
3772         list_for_each_entry_continue(next, head, set_list)
3773                 if (ses->entry_cnt <= next->entry_cnt)
3774                         break;
3775
3776         list_move_tail(&ses->set_list, &next->set_list);
3777 }
3778
3779 static void add_sit_entry(unsigned int segno, struct list_head *head)
3780 {
3781         struct sit_entry_set *ses;
3782         unsigned int start_segno = START_SEGNO(segno);
3783
3784         list_for_each_entry(ses, head, set_list) {
3785                 if (ses->start_segno == start_segno) {
3786                         ses->entry_cnt++;
3787                         adjust_sit_entry_set(ses, head);
3788                         return;
3789                 }
3790         }
3791
3792         ses = grab_sit_entry_set();
3793
3794         ses->start_segno = start_segno;
3795         ses->entry_cnt++;
3796         list_add(&ses->set_list, head);
3797 }
3798
3799 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3800 {
3801         struct f2fs_sm_info *sm_info = SM_I(sbi);
3802         struct list_head *set_list = &sm_info->sit_entry_set;
3803         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3804         unsigned int segno;
3805
3806         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3807                 add_sit_entry(segno, set_list);
3808 }
3809
3810 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3811 {
3812         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3813         struct f2fs_journal *journal = curseg->journal;
3814         int i;
3815
3816         down_write(&curseg->journal_rwsem);
3817         for (i = 0; i < sits_in_cursum(journal); i++) {
3818                 unsigned int segno;
3819                 bool dirtied;
3820
3821                 segno = le32_to_cpu(segno_in_journal(journal, i));
3822                 dirtied = __mark_sit_entry_dirty(sbi, segno);
3823
3824                 if (!dirtied)
3825                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3826         }
3827         update_sits_in_cursum(journal, -i);
3828         up_write(&curseg->journal_rwsem);
3829 }
3830
3831 /*
3832  * CP calls this function, which flushes SIT entries including sit_journal,
3833  * and moves prefree segs to free segs.
3834  */
3835 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3836 {
3837         struct sit_info *sit_i = SIT_I(sbi);
3838         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3839         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3840         struct f2fs_journal *journal = curseg->journal;
3841         struct sit_entry_set *ses, *tmp;
3842         struct list_head *head = &SM_I(sbi)->sit_entry_set;
3843         bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
3844         struct seg_entry *se;
3845
3846         down_write(&sit_i->sentry_lock);
3847
3848         if (!sit_i->dirty_sentries)
3849                 goto out;
3850
3851         /*
3852          * add and account sit entries of dirty bitmap in sit entry
3853          * set temporarily
3854          */
3855         add_sits_in_set(sbi);
3856
3857         /*
3858          * if there are no enough space in journal to store dirty sit
3859          * entries, remove all entries from journal and add and account
3860          * them in sit entry set.
3861          */
3862         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
3863                                                                 !to_journal)
3864                 remove_sits_in_journal(sbi);
3865
3866         /*
3867          * there are two steps to flush sit entries:
3868          * #1, flush sit entries to journal in current cold data summary block.
3869          * #2, flush sit entries to sit page.
3870          */
3871         list_for_each_entry_safe(ses, tmp, head, set_list) {
3872                 struct page *page = NULL;
3873                 struct f2fs_sit_block *raw_sit = NULL;
3874                 unsigned int start_segno = ses->start_segno;
3875                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3876                                                 (unsigned long)MAIN_SEGS(sbi));
3877                 unsigned int segno = start_segno;
3878
3879                 if (to_journal &&
3880                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3881                         to_journal = false;
3882
3883                 if (to_journal) {
3884                         down_write(&curseg->journal_rwsem);
3885                 } else {
3886                         page = get_next_sit_page(sbi, start_segno);
3887                         raw_sit = page_address(page);
3888                 }
3889
3890                 /* flush dirty sit entries in region of current sit set */
3891                 for_each_set_bit_from(segno, bitmap, end) {
3892                         int offset, sit_offset;
3893
3894                         se = get_seg_entry(sbi, segno);
3895 #ifdef CONFIG_F2FS_CHECK_FS
3896                         if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3897                                                 SIT_VBLOCK_MAP_SIZE))
3898                                 f2fs_bug_on(sbi, 1);
3899 #endif
3900
3901                         /* add discard candidates */
3902                         if (!(cpc->reason & CP_DISCARD)) {
3903                                 cpc->trim_start = segno;
3904                                 add_discard_addrs(sbi, cpc, false);
3905                         }
3906
3907                         if (to_journal) {
3908                                 offset = f2fs_lookup_journal_in_cursum(journal,
3909                                                         SIT_JOURNAL, segno, 1);
3910                                 f2fs_bug_on(sbi, offset < 0);
3911                                 segno_in_journal(journal, offset) =
3912                                                         cpu_to_le32(segno);
3913                                 seg_info_to_raw_sit(se,
3914                                         &sit_in_journal(journal, offset));
3915                                 check_block_count(sbi, segno,
3916                                         &sit_in_journal(journal, offset));
3917                         } else {
3918                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3919                                 seg_info_to_raw_sit(se,
3920                                                 &raw_sit->entries[sit_offset]);
3921                                 check_block_count(sbi, segno,
3922                                                 &raw_sit->entries[sit_offset]);
3923                         }
3924
3925                         __clear_bit(segno, bitmap);
3926                         sit_i->dirty_sentries--;
3927                         ses->entry_cnt--;
3928                 }
3929
3930                 if (to_journal)
3931                         up_write(&curseg->journal_rwsem);
3932                 else
3933                         f2fs_put_page(page, 1);
3934
3935                 f2fs_bug_on(sbi, ses->entry_cnt);
3936                 release_sit_entry_set(ses);
3937         }
3938
3939         f2fs_bug_on(sbi, !list_empty(head));
3940         f2fs_bug_on(sbi, sit_i->dirty_sentries);
3941 out:
3942         if (cpc->reason & CP_DISCARD) {
3943                 __u64 trim_start = cpc->trim_start;
3944
3945                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3946                         add_discard_addrs(sbi, cpc, false);
3947
3948                 cpc->trim_start = trim_start;
3949         }
3950         up_write(&sit_i->sentry_lock);
3951
3952         set_prefree_as_free_segments(sbi);
3953 }
3954
3955 static int build_sit_info(struct f2fs_sb_info *sbi)
3956 {
3957         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3958         struct sit_info *sit_i;
3959         unsigned int sit_segs, start;
3960         char *src_bitmap, *bitmap;
3961         unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
3962
3963         /* allocate memory for SIT information */
3964         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3965         if (!sit_i)
3966                 return -ENOMEM;
3967
3968         SM_I(sbi)->sit_info = sit_i;
3969
3970         sit_i->sentries =
3971                 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3972                                               MAIN_SEGS(sbi)),
3973                               GFP_KERNEL);
3974         if (!sit_i->sentries)
3975                 return -ENOMEM;
3976
3977         main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3978         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
3979                                                                 GFP_KERNEL);
3980         if (!sit_i->dirty_sentries_bitmap)
3981                 return -ENOMEM;
3982
3983 #ifdef CONFIG_F2FS_CHECK_FS
3984         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 4;
3985 #else
3986         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 3;
3987 #endif
3988         sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
3989         if (!sit_i->bitmap)
3990                 return -ENOMEM;
3991
3992         bitmap = sit_i->bitmap;
3993
3994         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3995                 sit_i->sentries[start].cur_valid_map = bitmap;
3996                 bitmap += SIT_VBLOCK_MAP_SIZE;
3997
3998                 sit_i->sentries[start].ckpt_valid_map = bitmap;
3999                 bitmap += SIT_VBLOCK_MAP_SIZE;
4000
4001 #ifdef CONFIG_F2FS_CHECK_FS
4002                 sit_i->sentries[start].cur_valid_map_mir = bitmap;
4003                 bitmap += SIT_VBLOCK_MAP_SIZE;
4004 #endif
4005
4006                 sit_i->sentries[start].discard_map = bitmap;
4007                 bitmap += SIT_VBLOCK_MAP_SIZE;
4008         }
4009
4010         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4011         if (!sit_i->tmp_map)
4012                 return -ENOMEM;
4013
4014         if (__is_large_section(sbi)) {
4015                 sit_i->sec_entries =
4016                         f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4017                                                       MAIN_SECS(sbi)),
4018                                       GFP_KERNEL);
4019                 if (!sit_i->sec_entries)
4020                         return -ENOMEM;
4021         }
4022
4023         /* get information related with SIT */
4024         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4025
4026         /* setup SIT bitmap from ckeckpoint pack */
4027         sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4028         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4029
4030         sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4031         if (!sit_i->sit_bitmap)
4032                 return -ENOMEM;
4033
4034 #ifdef CONFIG_F2FS_CHECK_FS
4035         sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4036                                         sit_bitmap_size, GFP_KERNEL);
4037         if (!sit_i->sit_bitmap_mir)
4038                 return -ENOMEM;
4039
4040         sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4041                                         main_bitmap_size, GFP_KERNEL);
4042         if (!sit_i->invalid_segmap)
4043                 return -ENOMEM;
4044 #endif
4045
4046         /* init SIT information */
4047         sit_i->s_ops = &default_salloc_ops;
4048
4049         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4050         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4051         sit_i->written_valid_blocks = 0;
4052         sit_i->bitmap_size = sit_bitmap_size;
4053         sit_i->dirty_sentries = 0;
4054         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4055         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4056         sit_i->mounted_time = ktime_get_real_seconds();
4057         init_rwsem(&sit_i->sentry_lock);
4058         return 0;
4059 }
4060
4061 static int build_free_segmap(struct f2fs_sb_info *sbi)
4062 {
4063         struct free_segmap_info *free_i;
4064         unsigned int bitmap_size, sec_bitmap_size;
4065
4066         /* allocate memory for free segmap information */
4067         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4068         if (!free_i)
4069                 return -ENOMEM;
4070
4071         SM_I(sbi)->free_info = free_i;
4072
4073         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4074         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4075         if (!free_i->free_segmap)
4076                 return -ENOMEM;
4077
4078         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4079         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4080         if (!free_i->free_secmap)
4081                 return -ENOMEM;
4082
4083         /* set all segments as dirty temporarily */
4084         memset(free_i->free_segmap, 0xff, bitmap_size);
4085         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4086
4087         /* init free segmap information */
4088         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4089         free_i->free_segments = 0;
4090         free_i->free_sections = 0;
4091         spin_lock_init(&free_i->segmap_lock);
4092         return 0;
4093 }
4094
4095 static int build_curseg(struct f2fs_sb_info *sbi)
4096 {
4097         struct curseg_info *array;
4098         int i;
4099
4100         array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
4101                              GFP_KERNEL);
4102         if (!array)
4103                 return -ENOMEM;
4104
4105         SM_I(sbi)->curseg_array = array;
4106
4107         for (i = 0; i < NR_CURSEG_TYPE; i++) {
4108                 mutex_init(&array[i].curseg_mutex);
4109                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4110                 if (!array[i].sum_blk)
4111                         return -ENOMEM;
4112                 init_rwsem(&array[i].journal_rwsem);
4113                 array[i].journal = f2fs_kzalloc(sbi,
4114                                 sizeof(struct f2fs_journal), GFP_KERNEL);
4115                 if (!array[i].journal)
4116                         return -ENOMEM;
4117                 array[i].segno = NULL_SEGNO;
4118                 array[i].next_blkoff = 0;
4119         }
4120         return restore_curseg_summaries(sbi);
4121 }
4122
4123 static int build_sit_entries(struct f2fs_sb_info *sbi)
4124 {
4125         struct sit_info *sit_i = SIT_I(sbi);
4126         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4127         struct f2fs_journal *journal = curseg->journal;
4128         struct seg_entry *se;
4129         struct f2fs_sit_entry sit;
4130         int sit_blk_cnt = SIT_BLK_CNT(sbi);
4131         unsigned int i, start, end;
4132         unsigned int readed, start_blk = 0;
4133         int err = 0;
4134         block_t total_node_blocks = 0;
4135
4136         do {
4137                 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
4138                                                         META_SIT, true);
4139
4140                 start = start_blk * sit_i->sents_per_block;
4141                 end = (start_blk + readed) * sit_i->sents_per_block;
4142
4143                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4144                         struct f2fs_sit_block *sit_blk;
4145                         struct page *page;
4146
4147                         se = &sit_i->sentries[start];
4148                         page = get_current_sit_page(sbi, start);
4149                         if (IS_ERR(page))
4150                                 return PTR_ERR(page);
4151                         sit_blk = (struct f2fs_sit_block *)page_address(page);
4152                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4153                         f2fs_put_page(page, 1);
4154
4155                         err = check_block_count(sbi, start, &sit);
4156                         if (err)
4157                                 return err;
4158                         seg_info_from_raw_sit(se, &sit);
4159                         if (IS_NODESEG(se->type))
4160                                 total_node_blocks += se->valid_blocks;
4161
4162                         /* build discard map only one time */
4163                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4164                                 memset(se->discard_map, 0xff,
4165                                         SIT_VBLOCK_MAP_SIZE);
4166                         } else {
4167                                 memcpy(se->discard_map,
4168                                         se->cur_valid_map,
4169                                         SIT_VBLOCK_MAP_SIZE);
4170                                 sbi->discard_blks +=
4171                                         sbi->blocks_per_seg -
4172                                         se->valid_blocks;
4173                         }
4174
4175                         if (__is_large_section(sbi))
4176                                 get_sec_entry(sbi, start)->valid_blocks +=
4177                                                         se->valid_blocks;
4178                 }
4179                 start_blk += readed;
4180         } while (start_blk < sit_blk_cnt);
4181
4182         down_read(&curseg->journal_rwsem);
4183         for (i = 0; i < sits_in_cursum(journal); i++) {
4184                 unsigned int old_valid_blocks;
4185
4186                 start = le32_to_cpu(segno_in_journal(journal, i));
4187                 if (start >= MAIN_SEGS(sbi)) {
4188                         f2fs_err(sbi, "Wrong journal entry on segno %u",
4189                                  start);
4190                         err = -EFSCORRUPTED;
4191                         break;
4192                 }
4193
4194                 se = &sit_i->sentries[start];
4195                 sit = sit_in_journal(journal, i);
4196
4197                 old_valid_blocks = se->valid_blocks;
4198                 if (IS_NODESEG(se->type))
4199                         total_node_blocks -= old_valid_blocks;
4200
4201                 err = check_block_count(sbi, start, &sit);
4202                 if (err)
4203                         break;
4204                 seg_info_from_raw_sit(se, &sit);
4205                 if (IS_NODESEG(se->type))
4206                         total_node_blocks += se->valid_blocks;
4207
4208                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4209                         memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4210                 } else {
4211                         memcpy(se->discard_map, se->cur_valid_map,
4212                                                 SIT_VBLOCK_MAP_SIZE);
4213                         sbi->discard_blks += old_valid_blocks;
4214                         sbi->discard_blks -= se->valid_blocks;
4215                 }
4216
4217                 if (__is_large_section(sbi)) {
4218                         get_sec_entry(sbi, start)->valid_blocks +=
4219                                                         se->valid_blocks;
4220                         get_sec_entry(sbi, start)->valid_blocks -=
4221                                                         old_valid_blocks;
4222                 }
4223         }
4224         up_read(&curseg->journal_rwsem);
4225
4226         if (!err && total_node_blocks != valid_node_count(sbi)) {
4227                 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4228                          total_node_blocks, valid_node_count(sbi));
4229                 err = -EFSCORRUPTED;
4230         }
4231
4232         return err;
4233 }
4234
4235 static void init_free_segmap(struct f2fs_sb_info *sbi)
4236 {
4237         unsigned int start;
4238         int type;
4239
4240         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4241                 struct seg_entry *sentry = get_seg_entry(sbi, start);
4242                 if (!sentry->valid_blocks)
4243                         __set_free(sbi, start);
4244                 else
4245                         SIT_I(sbi)->written_valid_blocks +=
4246                                                 sentry->valid_blocks;
4247         }
4248
4249         /* set use the current segments */
4250         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4251                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4252                 __set_test_and_inuse(sbi, curseg_t->segno);
4253         }
4254 }
4255
4256 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4257 {
4258         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4259         struct free_segmap_info *free_i = FREE_I(sbi);
4260         unsigned int segno = 0, offset = 0;
4261         unsigned short valid_blocks;
4262
4263         while (1) {
4264                 /* find dirty segment based on free segmap */
4265                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4266                 if (segno >= MAIN_SEGS(sbi))
4267                         break;
4268                 offset = segno + 1;
4269                 valid_blocks = get_valid_blocks(sbi, segno, false);
4270                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
4271                         continue;
4272                 if (valid_blocks > sbi->blocks_per_seg) {
4273                         f2fs_bug_on(sbi, 1);
4274                         continue;
4275                 }
4276                 mutex_lock(&dirty_i->seglist_lock);
4277                 __locate_dirty_segment(sbi, segno, DIRTY);
4278                 mutex_unlock(&dirty_i->seglist_lock);
4279         }
4280 }
4281
4282 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4283 {
4284         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4285         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4286
4287         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4288         if (!dirty_i->victim_secmap)
4289                 return -ENOMEM;
4290         return 0;
4291 }
4292
4293 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4294 {
4295         struct dirty_seglist_info *dirty_i;
4296         unsigned int bitmap_size, i;
4297
4298         /* allocate memory for dirty segments list information */
4299         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4300                                                                 GFP_KERNEL);
4301         if (!dirty_i)
4302                 return -ENOMEM;
4303
4304         SM_I(sbi)->dirty_info = dirty_i;
4305         mutex_init(&dirty_i->seglist_lock);
4306
4307         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4308
4309         for (i = 0; i < NR_DIRTY_TYPE; i++) {
4310                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4311                                                                 GFP_KERNEL);
4312                 if (!dirty_i->dirty_segmap[i])
4313                         return -ENOMEM;
4314         }
4315
4316         init_dirty_segmap(sbi);
4317         return init_victim_secmap(sbi);
4318 }
4319
4320 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4321 {
4322         int i;
4323
4324         /*
4325          * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4326          * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4327          */
4328         for (i = 0; i < NO_CHECK_TYPE; i++) {
4329                 struct curseg_info *curseg = CURSEG_I(sbi, i);
4330                 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4331                 unsigned int blkofs = curseg->next_blkoff;
4332
4333                 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4334                         goto out;
4335
4336                 if (curseg->alloc_type == SSR)
4337                         continue;
4338
4339                 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4340                         if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4341                                 continue;
4342 out:
4343                         f2fs_err(sbi,
4344                                  "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4345                                  i, curseg->segno, curseg->alloc_type,
4346                                  curseg->next_blkoff, blkofs);
4347                         return -EFSCORRUPTED;
4348                 }
4349         }
4350         return 0;
4351 }
4352
4353 /*
4354  * Update min, max modified time for cost-benefit GC algorithm
4355  */
4356 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4357 {
4358         struct sit_info *sit_i = SIT_I(sbi);
4359         unsigned int segno;
4360
4361         down_write(&sit_i->sentry_lock);
4362
4363         sit_i->min_mtime = ULLONG_MAX;
4364
4365         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4366                 unsigned int i;
4367                 unsigned long long mtime = 0;
4368
4369                 for (i = 0; i < sbi->segs_per_sec; i++)
4370                         mtime += get_seg_entry(sbi, segno + i)->mtime;
4371
4372                 mtime = div_u64(mtime, sbi->segs_per_sec);
4373
4374                 if (sit_i->min_mtime > mtime)
4375                         sit_i->min_mtime = mtime;
4376         }
4377         sit_i->max_mtime = get_mtime(sbi, false);
4378         up_write(&sit_i->sentry_lock);
4379 }
4380
4381 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
4382 {
4383         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4384         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4385         struct f2fs_sm_info *sm_info;
4386         int err;
4387
4388         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
4389         if (!sm_info)
4390                 return -ENOMEM;
4391
4392         /* init sm info */
4393         sbi->sm_info = sm_info;
4394         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4395         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4396         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4397         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4398         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4399         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4400         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
4401         sm_info->rec_prefree_segments = sm_info->main_segments *
4402                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
4403         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4404                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4405
4406         if (!test_opt(sbi, LFS))
4407                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
4408         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
4409         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
4410         sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
4411         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
4412         sm_info->min_ssr_sections = reserved_sections(sbi);
4413
4414         INIT_LIST_HEAD(&sm_info->sit_entry_set);
4415
4416         init_rwsem(&sm_info->curseg_lock);
4417
4418         if (!f2fs_readonly(sbi->sb)) {
4419                 err = f2fs_create_flush_cmd_control(sbi);
4420                 if (err)
4421                         return err;
4422         }
4423
4424         err = create_discard_cmd_control(sbi);
4425         if (err)
4426                 return err;
4427
4428         err = build_sit_info(sbi);
4429         if (err)
4430                 return err;
4431         err = build_free_segmap(sbi);
4432         if (err)
4433                 return err;
4434         err = build_curseg(sbi);
4435         if (err)
4436                 return err;
4437
4438         /* reinit free segmap based on SIT */
4439         err = build_sit_entries(sbi);
4440         if (err)
4441                 return err;
4442
4443         init_free_segmap(sbi);
4444         err = build_dirty_segmap(sbi);
4445         if (err)
4446                 return err;
4447
4448         err = sanity_check_curseg(sbi);
4449         if (err)
4450                 return err;
4451
4452         init_min_max_mtime(sbi);
4453         return 0;
4454 }
4455
4456 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4457                 enum dirty_type dirty_type)
4458 {
4459         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4460
4461         mutex_lock(&dirty_i->seglist_lock);
4462         kvfree(dirty_i->dirty_segmap[dirty_type]);
4463         dirty_i->nr_dirty[dirty_type] = 0;
4464         mutex_unlock(&dirty_i->seglist_lock);
4465 }
4466
4467 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
4468 {
4469         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4470         kvfree(dirty_i->victim_secmap);
4471 }
4472
4473 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4474 {
4475         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4476         int i;
4477
4478         if (!dirty_i)
4479                 return;
4480
4481         /* discard pre-free/dirty segments list */
4482         for (i = 0; i < NR_DIRTY_TYPE; i++)
4483                 discard_dirty_segmap(sbi, i);
4484
4485         destroy_victim_secmap(sbi);
4486         SM_I(sbi)->dirty_info = NULL;
4487         kvfree(dirty_i);
4488 }
4489
4490 static void destroy_curseg(struct f2fs_sb_info *sbi)
4491 {
4492         struct curseg_info *array = SM_I(sbi)->curseg_array;
4493         int i;
4494
4495         if (!array)
4496                 return;
4497         SM_I(sbi)->curseg_array = NULL;
4498         for (i = 0; i < NR_CURSEG_TYPE; i++) {
4499                 kvfree(array[i].sum_blk);
4500                 kvfree(array[i].journal);
4501         }
4502         kvfree(array);
4503 }
4504
4505 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4506 {
4507         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4508         if (!free_i)
4509                 return;
4510         SM_I(sbi)->free_info = NULL;
4511         kvfree(free_i->free_segmap);
4512         kvfree(free_i->free_secmap);
4513         kvfree(free_i);
4514 }
4515
4516 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4517 {
4518         struct sit_info *sit_i = SIT_I(sbi);
4519
4520         if (!sit_i)
4521                 return;
4522
4523         if (sit_i->sentries)
4524                 kvfree(sit_i->bitmap);
4525         kvfree(sit_i->tmp_map);
4526
4527         kvfree(sit_i->sentries);
4528         kvfree(sit_i->sec_entries);
4529         kvfree(sit_i->dirty_sentries_bitmap);
4530
4531         SM_I(sbi)->sit_info = NULL;
4532         kvfree(sit_i->sit_bitmap);
4533 #ifdef CONFIG_F2FS_CHECK_FS
4534         kvfree(sit_i->sit_bitmap_mir);
4535         kvfree(sit_i->invalid_segmap);
4536 #endif
4537         kvfree(sit_i);
4538 }
4539
4540 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
4541 {
4542         struct f2fs_sm_info *sm_info = SM_I(sbi);
4543
4544         if (!sm_info)
4545                 return;
4546         f2fs_destroy_flush_cmd_control(sbi, true);
4547         destroy_discard_cmd_control(sbi);
4548         destroy_dirty_segmap(sbi);
4549         destroy_curseg(sbi);
4550         destroy_free_segmap(sbi);
4551         destroy_sit_info(sbi);
4552         sbi->sm_info = NULL;
4553         kvfree(sm_info);
4554 }
4555
4556 int __init f2fs_create_segment_manager_caches(void)
4557 {
4558         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4559                         sizeof(struct discard_entry));
4560         if (!discard_entry_slab)
4561                 goto fail;
4562
4563         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4564                         sizeof(struct discard_cmd));
4565         if (!discard_cmd_slab)
4566                 goto destroy_discard_entry;
4567
4568         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4569                         sizeof(struct sit_entry_set));
4570         if (!sit_entry_set_slab)
4571                 goto destroy_discard_cmd;
4572
4573         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4574                         sizeof(struct inmem_pages));
4575         if (!inmem_entry_slab)
4576                 goto destroy_sit_entry_set;
4577         return 0;
4578
4579 destroy_sit_entry_set:
4580         kmem_cache_destroy(sit_entry_set_slab);
4581 destroy_discard_cmd:
4582         kmem_cache_destroy(discard_cmd_slab);
4583 destroy_discard_entry:
4584         kmem_cache_destroy(discard_entry_slab);
4585 fail:
4586         return -ENOMEM;
4587 }
4588
4589 void f2fs_destroy_segment_manager_caches(void)
4590 {
4591         kmem_cache_destroy(sit_entry_set_slab);
4592         kmem_cache_destroy(discard_cmd_slab);
4593         kmem_cache_destroy(discard_entry_slab);
4594         kmem_cache_destroy(inmem_entry_slab);
4595 }