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
 356         f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
 357
 358         mutex_lock(&fi->inmem_lock);
 359         list_for_each_entry(cur, head, list) {
 360                 if (cur->page == page)
 361                         break;
 362         }
 363
 364         f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
 365         list_del(&cur->list);
 366         mutex_unlock(&fi->inmem_lock);
 367
 368         dec_page_count(sbi, F2FS_INMEM_PAGES);
 369         kmem_cache_free(inmem_entry_slab, cur);
 370
 371         ClearPageUptodate(page);
 372         f2fs_clear_page_private(page);
 373         f2fs_put_page(page, 0);
 374
 375         trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 376 }
 377
 378 static int __f2fs_commit_inmem_pages(struct inode *inode)
 379 {
 380         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 381         struct f2fs_inode_info *fi = F2FS_I(inode);
 382         struct inmem_pages *cur, *tmp;
 383         struct f2fs_io_info fio = {
 384                 .sbi = sbi,
 385                 .ino = inode->i_ino,
 386                 .type = DATA,
 387                 .op = REQ_OP_WRITE,
 388                 .op_flags = REQ_SYNC | REQ_PRIO,
 389                 .io_type = FS_DATA_IO,
 390         };
 391         struct list_head revoke_list;
 392         bool submit_bio = false;
 393         int err = 0;
 394
 395         INIT_LIST_HEAD(&revoke_list);
 396
 397         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 398                 struct page *page = cur->page;
 399
 400                 lock_page(page);
 401                 if (page->mapping == inode->i_mapping) {
 402                         trace_f2fs_commit_inmem_page(page, INMEM);
 403
 404                         f2fs_wait_on_page_writeback(page, DATA, true, true);
 405
 406                         set_page_dirty(page);
 407                         if (clear_page_dirty_for_io(page)) {
 408                                 inode_dec_dirty_pages(inode);
 409                                 f2fs_remove_dirty_inode(inode);
 410                         }
 411 retry:
 412                         fio.page = page;
 413                         fio.old_blkaddr = NULL_ADDR;
 414                         fio.encrypted_page = NULL;
 415                         fio.need_lock = LOCK_DONE;
 416                         err = f2fs_do_write_data_page(&fio);
 417                         if (err) {
 418                                 if (err == -ENOMEM) {
 419                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 420                                         cond_resched();
 421                                         goto retry;
 422                                 }
 423                                 unlock_page(page);
 424                                 break;
 425                         }
 426                         /* record old blkaddr for revoking */
 427                         cur->old_addr = fio.old_blkaddr;
 428                         submit_bio = true;
 429                 }
 430                 unlock_page(page);
 431                 list_move_tail(&cur->list, &revoke_list);
 432         }
 433
 434         if (submit_bio)
 435                 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
 436
 437         if (err) {
 438                 /*
 439                  * try to revoke all committed pages, but still we could fail
 440                  * due to no memory or other reason, if that happened, EAGAIN
 441                  * will be returned, which means in such case, transaction is
 442                  * already not integrity, caller should use journal to do the
 443                  * recovery or rewrite & commit last transaction. For other
 444                  * error number, revoking was done by filesystem itself.
 445                  */
 446                 err = __revoke_inmem_pages(inode, &revoke_list,
 447                                                 false, true, false);
 448
 449                 /* drop all uncommitted pages */
 450                 __revoke_inmem_pages(inode, &fi->inmem_pages,
 451                                                 true, false, false);
 452         } else {
 453                 __revoke_inmem_pages(inode, &revoke_list,
 454                                                 false, false, false);
 455         }
 456
 457         return err;
 458 }
 459
 460 int f2fs_commit_inmem_pages(struct inode *inode)
 461 {
 462         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 463         struct f2fs_inode_info *fi = F2FS_I(inode);
 464         int err;
 465
 466         f2fs_balance_fs(sbi, true);
 467
 468         down_write(&fi->i_gc_rwsem[WRITE]);
 469
 470         f2fs_lock_op(sbi);
 471         set_inode_flag(inode, FI_ATOMIC_COMMIT);
 472
 473         mutex_lock(&fi->inmem_lock);
 474         err = __f2fs_commit_inmem_pages(inode);
 475         mutex_unlock(&fi->inmem_lock);
 476
 477         clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 478
 479         f2fs_unlock_op(sbi);
 480         up_write(&fi->i_gc_rwsem[WRITE]);
 481
 482         return err;
 483 }
 484
 485 /*
 486  * This function balances dirty node and dentry pages.
 487  * In addition, it controls garbage collection.
 488  */
 489 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 490 {
 491         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 492                 f2fs_show_injection_info(FAULT_CHECKPOINT);
 493                 f2fs_stop_checkpoint(sbi, false);
 494         }
 495
 496         /* balance_fs_bg is able to be pending */
 497         if (need && excess_cached_nats(sbi))
 498                 f2fs_balance_fs_bg(sbi);
 499
 500         if (!f2fs_is_checkpoint_ready(sbi))
 501                 return;
 502
 503         /*
 504          * We should do GC or end up with checkpoint, if there are so many dirty
 505          * dir/node pages without enough free segments.
 506          */
 507         if (has_not_enough_free_secs(sbi, 0, 0)) {
 508                 mutex_lock(&sbi->gc_mutex);
 509                 f2fs_gc(sbi, false, false, NULL_SEGNO);
 510         }
 511 }
 512
 513 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 514 {
 515         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 516                 return;
 517
 518         /* try to shrink extent cache when there is no enough memory */
 519         if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
 520                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 521
 522         /* check the # of cached NAT entries */
 523         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
 524                 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 525
 526         if (!f2fs_available_free_memory(sbi, FREE_NIDS))
 527                 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
 528         else
 529                 f2fs_build_free_nids(sbi, false, false);
 530
 531         if (!is_idle(sbi, REQ_TIME) &&
 532                 (!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
 533                 return;
 534
 535         /* checkpoint is the only way to shrink partial cached entries */
 536         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
 537                         !f2fs_available_free_memory(sbi, INO_ENTRIES) ||
 538                         excess_prefree_segs(sbi) ||
 539                         excess_dirty_nats(sbi) ||
 540                         excess_dirty_nodes(sbi) ||
 541                         f2fs_time_over(sbi, CP_TIME)) {
 542                 if (test_opt(sbi, DATA_FLUSH)) {
 543                         struct blk_plug plug;
 544
 545                         mutex_lock(&sbi->flush_lock);
 546
 547                         blk_start_plug(&plug);
 548                         f2fs_sync_dirty_inodes(sbi, FILE_INODE);
 549                         blk_finish_plug(&plug);
 550
 551                         mutex_unlock(&sbi->flush_lock);
 552                 }
 553                 f2fs_sync_fs(sbi->sb, true);
 554                 stat_inc_bg_cp_count(sbi->stat_info);
 555         }
 556 }
 557
 558 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 559                                 struct block_device *bdev)
 560 {
 561         struct bio *bio;
 562         int ret;
 563
 564         bio = f2fs_bio_alloc(sbi, 0, false);
 565         if (!bio)
 566                 return -ENOMEM;
 567
 568         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
 569         bio_set_dev(bio, bdev);
 570         ret = submit_bio_wait(bio);
 571         bio_put(bio);
 572
 573         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 574                                 test_opt(sbi, FLUSH_MERGE), ret);
 575         return ret;
 576 }
 577
 578 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 579 {
 580         int ret = 0;
 581         int i;
 582
 583         if (!f2fs_is_multi_device(sbi))
 584                 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 585
 586         for (i = 0; i < sbi->s_ndevs; i++) {
 587                 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
 588                         continue;
 589                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 590                 if (ret)
 591                         break;
 592         }
 593         return ret;
 594 }
 595
 596 static int issue_flush_thread(void *data)
 597 {
 598         struct f2fs_sb_info *sbi = data;
 599         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 600         wait_queue_head_t *q = &fcc->flush_wait_queue;
 601 repeat:
 602         if (kthread_should_stop())
 603                 return 0;
 604
 605         sb_start_intwrite(sbi->sb);
 606
 607         if (!llist_empty(&fcc->issue_list)) {
 608                 struct flush_cmd *cmd, *next;
 609                 int ret;
 610
 611                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 612                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 613
 614                 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 615
 616                 ret = submit_flush_wait(sbi, cmd->ino);
 617                 atomic_inc(&fcc->issued_flush);
 618
 619                 llist_for_each_entry_safe(cmd, next,
 620                                           fcc->dispatch_list, llnode) {
 621                         cmd->ret = ret;
 622                         complete(&cmd->wait);
 623                 }
 624                 fcc->dispatch_list = NULL;
 625         }
 626
 627         sb_end_intwrite(sbi->sb);
 628
 629         wait_event_interruptible(*q,
 630                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 631         goto repeat;
 632 }
 633
 634 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 635 {
 636         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 637         struct flush_cmd cmd;
 638         int ret;
 639
 640         if (test_opt(sbi, NOBARRIER))
 641                 return 0;
 642
 643         if (!test_opt(sbi, FLUSH_MERGE)) {
 644                 atomic_inc(&fcc->queued_flush);
 645                 ret = submit_flush_wait(sbi, ino);
 646                 atomic_dec(&fcc->queued_flush);
 647                 atomic_inc(&fcc->issued_flush);
 648                 return ret;
 649         }
 650
 651         if (atomic_inc_return(&fcc->queued_flush) == 1 ||
 652             f2fs_is_multi_device(sbi)) {
 653                 ret = submit_flush_wait(sbi, ino);
 654                 atomic_dec(&fcc->queued_flush);
 655
 656                 atomic_inc(&fcc->issued_flush);
 657                 return ret;
 658         }
 659
 660         cmd.ino = ino;
 661         init_completion(&cmd.wait);
 662
 663         llist_add(&cmd.llnode, &fcc->issue_list);
 664
 665         /* update issue_list before we wake up issue_flush thread */
 666         smp_mb();
 667
 668         if (waitqueue_active(&fcc->flush_wait_queue))
 669                 wake_up(&fcc->flush_wait_queue);
 670
 671         if (fcc->f2fs_issue_flush) {
 672                 wait_for_completion(&cmd.wait);
 673                 atomic_dec(&fcc->queued_flush);
 674         } else {
 675                 struct llist_node *list;
 676
 677                 list = llist_del_all(&fcc->issue_list);
 678                 if (!list) {
 679                         wait_for_completion(&cmd.wait);
 680                         atomic_dec(&fcc->queued_flush);
 681                 } else {
 682                         struct flush_cmd *tmp, *next;
 683
 684                         ret = submit_flush_wait(sbi, ino);
 685
 686                         llist_for_each_entry_safe(tmp, next, list, llnode) {
 687                                 if (tmp == &cmd) {
 688                                         cmd.ret = ret;
 689                                         atomic_dec(&fcc->queued_flush);
 690                                         continue;
 691                                 }
 692                                 tmp->ret = ret;
 693                                 complete(&tmp->wait);
 694                         }
 695                 }
 696         }
 697
 698         return cmd.ret;
 699 }
 700
 701 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
 702 {
 703         dev_t dev = sbi->sb->s_bdev->bd_dev;
 704         struct flush_cmd_control *fcc;
 705         int err = 0;
 706
 707         if (SM_I(sbi)->fcc_info) {
 708                 fcc = SM_I(sbi)->fcc_info;
 709                 if (fcc->f2fs_issue_flush)
 710                         return err;
 711                 goto init_thread;
 712         }
 713
 714         fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
 715         if (!fcc)
 716                 return -ENOMEM;
 717         atomic_set(&fcc->issued_flush, 0);
 718         atomic_set(&fcc->queued_flush, 0);
 719         init_waitqueue_head(&fcc->flush_wait_queue);
 720         init_llist_head(&fcc->issue_list);
 721         SM_I(sbi)->fcc_info = fcc;
 722         if (!test_opt(sbi, FLUSH_MERGE))
 723                 return err;
 724
 725 init_thread:
 726         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 727                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 728         if (IS_ERR(fcc->f2fs_issue_flush)) {
 729                 err = PTR_ERR(fcc->f2fs_issue_flush);
 730                 kvfree(fcc);
 731                 SM_I(sbi)->fcc_info = NULL;
 732                 return err;
 733         }
 734
 735         return err;
 736 }
 737
 738 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 739 {
 740         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 741
 742         if (fcc && fcc->f2fs_issue_flush) {
 743                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 744
 745                 fcc->f2fs_issue_flush = NULL;
 746                 kthread_stop(flush_thread);
 747         }
 748         if (free) {
 749                 kvfree(fcc);
 750                 SM_I(sbi)->fcc_info = NULL;
 751         }
 752 }
 753
 754 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
 755 {
 756         int ret = 0, i;
 757
 758         if (!f2fs_is_multi_device(sbi))
 759                 return 0;
 760
 761         for (i = 1; i < sbi->s_ndevs; i++) {
 762                 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
 763                         continue;
 764                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 765                 if (ret)
 766                         break;
 767
 768                 spin_lock(&sbi->dev_lock);
 769                 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
 770                 spin_unlock(&sbi->dev_lock);
 771         }
 772
 773         return ret;
 774 }
 775
 776 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 777                 enum dirty_type dirty_type)
 778 {
 779         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 780
 781         /* need not be added */
 782         if (IS_CURSEG(sbi, segno))
 783                 return;
 784
 785         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 786                 dirty_i->nr_dirty[dirty_type]++;
 787
 788         if (dirty_type == DIRTY) {
 789                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 790                 enum dirty_type t = sentry->type;
 791
 792                 if (unlikely(t >= DIRTY)) {
 793                         f2fs_bug_on(sbi, 1);
 794                         return;
 795                 }
 796                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 797                         dirty_i->nr_dirty[t]++;
 798         }
 799 }
 800
 801 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 802                 enum dirty_type dirty_type)
 803 {
 804         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 805
 806         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 807                 dirty_i->nr_dirty[dirty_type]--;
 808
 809         if (dirty_type == DIRTY) {
 810                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 811                 enum dirty_type t = sentry->type;
 812
 813                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 814                         dirty_i->nr_dirty[t]--;
 815
 816                 if (get_valid_blocks(sbi, segno, true) == 0) {
 817                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 818                                                 dirty_i->victim_secmap);
 819 #ifdef CONFIG_F2FS_CHECK_FS
 820                         clear_bit(segno, SIT_I(sbi)->invalid_segmap);
 821 #endif
 822                 }
 823         }
 824 }
 825
 826 /*
 827  * Should not occur error such as -ENOMEM.
 828  * Adding dirty entry into seglist is not critical operation.
 829  * If a given segment is one of current working segments, it won't be added.
 830  */
 831 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 832 {
 833         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 834         unsigned short valid_blocks, ckpt_valid_blocks;
 835
 836         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 837                 return;
 838
 839         mutex_lock(&dirty_i->seglist_lock);
 840
 841         valid_blocks = get_valid_blocks(sbi, segno, false);
 842         ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno);
 843
 844         if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
 845                                 ckpt_valid_blocks == sbi->blocks_per_seg)) {
 846                 __locate_dirty_segment(sbi, segno, PRE);
 847                 __remove_dirty_segment(sbi, segno, DIRTY);
 848         } else if (valid_blocks < sbi->blocks_per_seg) {
 849                 __locate_dirty_segment(sbi, segno, DIRTY);
 850         } else {
 851                 /* Recovery routine with SSR needs this */
 852                 __remove_dirty_segment(sbi, segno, DIRTY);
 853         }
 854
 855         mutex_unlock(&dirty_i->seglist_lock);
 856 }
 857
 858 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
 859 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
 860 {
 861         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 862         unsigned int segno;
 863
 864         mutex_lock(&dirty_i->seglist_lock);
 865         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 866                 if (get_valid_blocks(sbi, segno, false))
 867                         continue;
 868                 if (IS_CURSEG(sbi, segno))
 869                         continue;
 870                 __locate_dirty_segment(sbi, segno, PRE);
 871                 __remove_dirty_segment(sbi, segno, DIRTY);
 872         }
 873         mutex_unlock(&dirty_i->seglist_lock);
 874 }
 875
 876 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
 877 {
 878         int ovp_hole_segs =
 879                 (overprovision_segments(sbi) - reserved_segments(sbi));
 880         block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
 881         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 882         block_t holes[2] = {0, 0};      /* DATA and NODE */
 883         block_t unusable;
 884         struct seg_entry *se;
 885         unsigned int segno;
 886
 887         mutex_lock(&dirty_i->seglist_lock);
 888         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 889                 se = get_seg_entry(sbi, segno);
 890                 if (IS_NODESEG(se->type))
 891                         holes[NODE] += sbi->blocks_per_seg - se->valid_blocks;
 892                 else
 893                         holes[DATA] += sbi->blocks_per_seg - se->valid_blocks;
 894         }
 895         mutex_unlock(&dirty_i->seglist_lock);
 896
 897         unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
 898         if (unusable > ovp_holes)
 899                 return unusable - ovp_holes;
 900         return 0;
 901 }
 902
 903 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
 904 {
 905         int ovp_hole_segs =
 906                 (overprovision_segments(sbi) - reserved_segments(sbi));
 907         if (unusable > F2FS_OPTION(sbi).unusable_cap)
 908                 return -EAGAIN;
 909         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
 910                 dirty_segments(sbi) > ovp_hole_segs)
 911                 return -EAGAIN;
 912         return 0;
 913 }
 914
 915 /* This is only used by SBI_CP_DISABLED */
 916 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
 917 {
 918         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 919         unsigned int segno = 0;
 920
 921         mutex_lock(&dirty_i->seglist_lock);
 922         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 923                 if (get_valid_blocks(sbi, segno, false))
 924                         continue;
 925                 if (get_ckpt_valid_blocks(sbi, segno))
 926                         continue;
 927                 mutex_unlock(&dirty_i->seglist_lock);
 928                 return segno;
 929         }
 930         mutex_unlock(&dirty_i->seglist_lock);
 931         return NULL_SEGNO;
 932 }
 933
 934 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 935                 struct block_device *bdev, block_t lstart,
 936                 block_t start, block_t len)
 937 {
 938         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 939         struct list_head *pend_list;
 940         struct discard_cmd *dc;
 941
 942         f2fs_bug_on(sbi, !len);
 943
 944         pend_list = &dcc->pend_list[plist_idx(len)];
 945
 946         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 947         INIT_LIST_HEAD(&dc->list);
 948         dc->bdev = bdev;
 949         dc->lstart = lstart;
 950         dc->start = start;
 951         dc->len = len;
 952         dc->ref = 0;
 953         dc->state = D_PREP;
 954         dc->queued = 0;
 955         dc->error = 0;
 956         init_completion(&dc->wait);
 957         list_add_tail(&dc->list, pend_list);
 958         spin_lock_init(&dc->lock);
 959         dc->bio_ref = 0;
 960         atomic_inc(&dcc->discard_cmd_cnt);
 961         dcc->undiscard_blks += len;
 962
 963         return dc;
 964 }
 965
 966 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 967                                 struct block_device *bdev, block_t lstart,
 968                                 block_t start, block_t len,
 969                                 struct rb_node *parent, struct rb_node **p,
 970                                 bool leftmost)
 971 {
 972         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 973         struct discard_cmd *dc;
 974
 975         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
 976
 977         rb_link_node(&dc->rb_node, parent, p);
 978         rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
 979
 980         return dc;
 981 }
 982
 983 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
 984                                                         struct discard_cmd *dc)
 985 {
 986         if (dc->state == D_DONE)
 987                 atomic_sub(dc->queued, &dcc->queued_discard);
 988
 989         list_del(&dc->list);
 990         rb_erase_cached(&dc->rb_node, &dcc->root);
 991         dcc->undiscard_blks -= dc->len;
 992
 993         kmem_cache_free(discard_cmd_slab, dc);
 994
 995         atomic_dec(&dcc->discard_cmd_cnt);
 996 }
 997
 998 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
 999                                                         struct discard_cmd *dc)
1000 {
1001         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1002         unsigned long flags;
1003
1004         trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
1005
1006         spin_lock_irqsave(&dc->lock, flags);
1007         if (dc->bio_ref) {
1008                 spin_unlock_irqrestore(&dc->lock, flags);
1009                 return;
1010         }
1011         spin_unlock_irqrestore(&dc->lock, flags);
1012
1013         f2fs_bug_on(sbi, dc->ref);
1014
1015         if (dc->error == -EOPNOTSUPP)
1016                 dc->error = 0;
1017
1018         if (dc->error)
1019                 printk_ratelimited(
1020                         "%sF2FS-fs: Issue discard(%u, %u, %u) failed, ret: %d",
1021                         KERN_INFO, dc->lstart, dc->start, dc->len, dc->error);
1022         __detach_discard_cmd(dcc, dc);
1023 }
1024
1025 static void f2fs_submit_discard_endio(struct bio *bio)
1026 {
1027         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1028         unsigned long flags;
1029
1030         dc->error = blk_status_to_errno(bio->bi_status);
1031
1032         spin_lock_irqsave(&dc->lock, flags);
1033         dc->bio_ref--;
1034         if (!dc->bio_ref && dc->state == D_SUBMIT) {
1035                 dc->state = D_DONE;
1036                 complete_all(&dc->wait);
1037         }
1038         spin_unlock_irqrestore(&dc->lock, flags);
1039         bio_put(bio);
1040 }
1041
1042 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1043                                 block_t start, block_t end)
1044 {
1045 #ifdef CONFIG_F2FS_CHECK_FS
1046         struct seg_entry *sentry;
1047         unsigned int segno;
1048         block_t blk = start;
1049         unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1050         unsigned long *map;
1051
1052         while (blk < end) {
1053                 segno = GET_SEGNO(sbi, blk);
1054                 sentry = get_seg_entry(sbi, segno);
1055                 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1056
1057                 if (end < START_BLOCK(sbi, segno + 1))
1058                         size = GET_BLKOFF_FROM_SEG0(sbi, end);
1059                 else
1060                         size = max_blocks;
1061                 map = (unsigned long *)(sentry->cur_valid_map);
1062                 offset = __find_rev_next_bit(map, size, offset);
1063                 f2fs_bug_on(sbi, offset != size);
1064                 blk = START_BLOCK(sbi, segno + 1);
1065         }
1066 #endif
1067 }
1068
1069 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1070                                 struct discard_policy *dpolicy,
1071                                 int discard_type, unsigned int granularity)
1072 {
1073         /* common policy */
1074         dpolicy->type = discard_type;
1075         dpolicy->sync = true;
1076         dpolicy->ordered = false;
1077         dpolicy->granularity = granularity;
1078
1079         dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1080         dpolicy->io_aware_gran = MAX_PLIST_NUM;
1081         dpolicy->timeout = 0;
1082
1083         if (discard_type == DPOLICY_BG) {
1084                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1085                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1086                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1087                 dpolicy->io_aware = true;
1088                 dpolicy->sync = false;
1089                 dpolicy->ordered = true;
1090                 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1091                         dpolicy->granularity = 1;
1092                         dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1093                 }
1094         } else if (discard_type == DPOLICY_FORCE) {
1095                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1096                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1097                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1098                 dpolicy->io_aware = false;
1099         } else if (discard_type == DPOLICY_FSTRIM) {
1100                 dpolicy->io_aware = false;
1101         } else if (discard_type == DPOLICY_UMOUNT) {
1102                 dpolicy->max_requests = UINT_MAX;
1103                 dpolicy->io_aware = false;
1104                 /* we need to issue all to keep CP_TRIMMED_FLAG */
1105                 dpolicy->granularity = 1;
1106         }
1107 }
1108
1109 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1110                                 struct block_device *bdev, block_t lstart,
1111                                 block_t start, block_t len);
1112 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1113 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1114                                                 struct discard_policy *dpolicy,
1115                                                 struct discard_cmd *dc,
1116                                                 unsigned int *issued)
1117 {
1118         struct block_device *bdev = dc->bdev;
1119         struct request_queue *q = bdev_get_queue(bdev);
1120         unsigned int max_discard_blocks =
1121                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1122         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1123         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1124                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1125         int flag = dpolicy->sync ? REQ_SYNC : 0;
1126         block_t lstart, start, len, total_len;
1127         int err = 0;
1128
1129         if (dc->state != D_PREP)
1130                 return 0;
1131
1132         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1133                 return 0;
1134
1135         trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1136
1137         lstart = dc->lstart;
1138         start = dc->start;
1139         len = dc->len;
1140         total_len = len;
1141
1142         dc->len = 0;
1143
1144         while (total_len && *issued < dpolicy->max_requests && !err) {
1145                 struct bio *bio = NULL;
1146                 unsigned long flags;
1147                 bool last = true;
1148
1149                 if (len > max_discard_blocks) {
1150                         len = max_discard_blocks;
1151                         last = false;
1152                 }
1153
1154                 (*issued)++;
1155                 if (*issued == dpolicy->max_requests)
1156                         last = true;
1157
1158                 dc->len += len;
1159
1160                 if (time_to_inject(sbi, FAULT_DISCARD)) {
1161                         f2fs_show_injection_info(FAULT_DISCARD);
1162                         err = -EIO;
1163                         goto submit;
1164                 }
1165                 err = __blkdev_issue_discard(bdev,
1166                                         SECTOR_FROM_BLOCK(start),
1167                                         SECTOR_FROM_BLOCK(len),
1168                                         GFP_NOFS, 0, &bio);
1169 submit:
1170                 if (err) {
1171                         spin_lock_irqsave(&dc->lock, flags);
1172                         if (dc->state == D_PARTIAL)
1173                                 dc->state = D_SUBMIT;
1174                         spin_unlock_irqrestore(&dc->lock, flags);
1175
1176                         break;
1177                 }
1178
1179                 f2fs_bug_on(sbi, !bio);
1180
1181                 /*
1182                  * should keep before submission to avoid D_DONE
1183                  * right away
1184                  */
1185                 spin_lock_irqsave(&dc->lock, flags);
1186                 if (last)
1187                         dc->state = D_SUBMIT;
1188                 else
1189                         dc->state = D_PARTIAL;
1190                 dc->bio_ref++;
1191                 spin_unlock_irqrestore(&dc->lock, flags);
1192
1193                 atomic_inc(&dcc->queued_discard);
1194                 dc->queued++;
1195                 list_move_tail(&dc->list, wait_list);
1196
1197                 /* sanity check on discard range */
1198                 __check_sit_bitmap(sbi, lstart, lstart + len);
1199
1200                 bio->bi_private = dc;
1201                 bio->bi_end_io = f2fs_submit_discard_endio;
1202                 bio->bi_opf |= flag;
1203                 submit_bio(bio);
1204
1205                 atomic_inc(&dcc->issued_discard);
1206
1207                 f2fs_update_iostat(sbi, FS_DISCARD, 1);
1208
1209                 lstart += len;
1210                 start += len;
1211                 total_len -= len;
1212                 len = total_len;
1213         }
1214
1215         if (!err && len)
1216                 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1217         return err;
1218 }
1219
1220 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1221                                 struct block_device *bdev, block_t lstart,
1222                                 block_t start, block_t len,
1223                                 struct rb_node **insert_p,
1224                                 struct rb_node *insert_parent)
1225 {
1226         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1227         struct rb_node **p;
1228         struct rb_node *parent = NULL;
1229         struct discard_cmd *dc = NULL;
1230         bool leftmost = true;
1231
1232         if (insert_p && insert_parent) {
1233                 parent = insert_parent;
1234                 p = insert_p;
1235                 goto do_insert;
1236         }
1237
1238         p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1239                                                         lstart, &leftmost);
1240 do_insert:
1241         dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1242                                                                 p, leftmost);
1243         if (!dc)
1244                 return NULL;
1245
1246         return dc;
1247 }
1248
1249 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1250                                                 struct discard_cmd *dc)
1251 {
1252         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1253 }
1254
1255 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1256                                 struct discard_cmd *dc, block_t blkaddr)
1257 {
1258         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1259         struct discard_info di = dc->di;
1260         bool modified = false;
1261
1262         if (dc->state == D_DONE || dc->len == 1) {
1263                 __remove_discard_cmd(sbi, dc);
1264                 return;
1265         }
1266
1267         dcc->undiscard_blks -= di.len;
1268
1269         if (blkaddr > di.lstart) {
1270                 dc->len = blkaddr - dc->lstart;
1271                 dcc->undiscard_blks += dc->len;
1272                 __relocate_discard_cmd(dcc, dc);
1273                 modified = true;
1274         }
1275
1276         if (blkaddr < di.lstart + di.len - 1) {
1277                 if (modified) {
1278                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1279                                         di.start + blkaddr + 1 - di.lstart,
1280                                         di.lstart + di.len - 1 - blkaddr,
1281                                         NULL, NULL);
1282                 } else {
1283                         dc->lstart++;
1284                         dc->len--;
1285                         dc->start++;
1286                         dcc->undiscard_blks += dc->len;
1287                         __relocate_discard_cmd(dcc, dc);
1288                 }
1289         }
1290 }
1291
1292 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1293                                 struct block_device *bdev, block_t lstart,
1294                                 block_t start, block_t len)
1295 {
1296         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1297         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1298         struct discard_cmd *dc;
1299         struct discard_info di = {0};
1300         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1301         struct request_queue *q = bdev_get_queue(bdev);
1302         unsigned int max_discard_blocks =
1303                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1304         block_t end = lstart + len;
1305
1306         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1307                                         NULL, lstart,
1308                                         (struct rb_entry **)&prev_dc,
1309                                         (struct rb_entry **)&next_dc,
1310                                         &insert_p, &insert_parent, true, NULL);
1311         if (dc)
1312                 prev_dc = dc;
1313
1314         if (!prev_dc) {
1315                 di.lstart = lstart;
1316                 di.len = next_dc ? next_dc->lstart - lstart : len;
1317                 di.len = min(di.len, len);
1318                 di.start = start;
1319         }
1320
1321         while (1) {
1322                 struct rb_node *node;
1323                 bool merged = false;
1324                 struct discard_cmd *tdc = NULL;
1325
1326                 if (prev_dc) {
1327                         di.lstart = prev_dc->lstart + prev_dc->len;
1328                         if (di.lstart < lstart)
1329                                 di.lstart = lstart;
1330                         if (di.lstart >= end)
1331                                 break;
1332
1333                         if (!next_dc || next_dc->lstart > end)
1334                                 di.len = end - di.lstart;
1335                         else
1336                                 di.len = next_dc->lstart - di.lstart;
1337                         di.start = start + di.lstart - lstart;
1338                 }
1339
1340                 if (!di.len)
1341                         goto next;
1342
1343                 if (prev_dc && prev_dc->state == D_PREP &&
1344                         prev_dc->bdev == bdev &&
1345                         __is_discard_back_mergeable(&di, &prev_dc->di,
1346                                                         max_discard_blocks)) {
1347                         prev_dc->di.len += di.len;
1348                         dcc->undiscard_blks += di.len;
1349                         __relocate_discard_cmd(dcc, prev_dc);
1350                         di = prev_dc->di;
1351                         tdc = prev_dc;
1352                         merged = true;
1353                 }
1354
1355                 if (next_dc && next_dc->state == D_PREP &&
1356                         next_dc->bdev == bdev &&
1357                         __is_discard_front_mergeable(&di, &next_dc->di,
1358                                                         max_discard_blocks)) {
1359                         next_dc->di.lstart = di.lstart;
1360                         next_dc->di.len += di.len;
1361                         next_dc->di.start = di.start;
1362                         dcc->undiscard_blks += di.len;
1363                         __relocate_discard_cmd(dcc, next_dc);
1364                         if (tdc)
1365                                 __remove_discard_cmd(sbi, tdc);
1366                         merged = true;
1367                 }
1368
1369                 if (!merged) {
1370                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1371                                                         di.len, NULL, NULL);
1372                 }
1373  next:
1374                 prev_dc = next_dc;
1375                 if (!prev_dc)
1376                         break;
1377
1378                 node = rb_next(&prev_dc->rb_node);
1379                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1380         }
1381 }
1382
1383 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1384                 struct block_device *bdev, block_t blkstart, block_t blklen)
1385 {
1386         block_t lblkstart = blkstart;
1387
1388         if (!f2fs_bdev_support_discard(bdev))
1389                 return 0;
1390
1391         trace_f2fs_queue_discard(bdev, blkstart, blklen);
1392
1393         if (f2fs_is_multi_device(sbi)) {
1394                 int devi = f2fs_target_device_index(sbi, blkstart);
1395
1396                 blkstart -= FDEV(devi).start_blk;
1397         }
1398         mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1399         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1400         mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1401         return 0;
1402 }
1403
1404 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1405                                         struct discard_policy *dpolicy)
1406 {
1407         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1408         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1409         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1410         struct discard_cmd *dc;
1411         struct blk_plug plug;
1412         unsigned int pos = dcc->next_pos;
1413         unsigned int issued = 0;
1414         bool io_interrupted = false;
1415
1416         mutex_lock(&dcc->cmd_lock);
1417         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1418                                         NULL, pos,
1419                                         (struct rb_entry **)&prev_dc,
1420                                         (struct rb_entry **)&next_dc,
1421                                         &insert_p, &insert_parent, true, NULL);
1422         if (!dc)
1423                 dc = next_dc;
1424
1425         blk_start_plug(&plug);
1426
1427         while (dc) {
1428                 struct rb_node *node;
1429                 int err = 0;
1430
1431                 if (dc->state != D_PREP)
1432                         goto next;
1433
1434                 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1435                         io_interrupted = true;
1436                         break;
1437                 }
1438
1439                 dcc->next_pos = dc->lstart + dc->len;
1440                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1441
1442                 if (issued >= dpolicy->max_requests)
1443                         break;
1444 next:
1445                 node = rb_next(&dc->rb_node);
1446                 if (err)
1447                         __remove_discard_cmd(sbi, dc);
1448                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1449         }
1450
1451         blk_finish_plug(&plug);
1452
1453         if (!dc)
1454                 dcc->next_pos = 0;
1455
1456         mutex_unlock(&dcc->cmd_lock);
1457
1458         if (!issued && io_interrupted)
1459                 issued = -1;
1460
1461         return issued;
1462 }
1463
1464 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1465                                         struct discard_policy *dpolicy)
1466 {
1467         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1468         struct list_head *pend_list;
1469         struct discard_cmd *dc, *tmp;
1470         struct blk_plug plug;
1471         int i, issued = 0;
1472         bool io_interrupted = false;
1473
1474         if (dpolicy->timeout != 0)
1475                 f2fs_update_time(sbi, dpolicy->timeout);
1476
1477         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1478                 if (dpolicy->timeout != 0 &&
1479                                 f2fs_time_over(sbi, dpolicy->timeout))
1480                         break;
1481
1482                 if (i + 1 < dpolicy->granularity)
1483                         break;
1484
1485                 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1486                         return __issue_discard_cmd_orderly(sbi, dpolicy);
1487
1488                 pend_list = &dcc->pend_list[i];
1489
1490                 mutex_lock(&dcc->cmd_lock);
1491                 if (list_empty(pend_list))
1492                         goto next;
1493                 if (unlikely(dcc->rbtree_check))
1494                         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1495                                                                 &dcc->root));
1496                 blk_start_plug(&plug);
1497                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1498                         f2fs_bug_on(sbi, dc->state != D_PREP);
1499
1500                         if (dpolicy->timeout != 0 &&
1501                                 f2fs_time_over(sbi, dpolicy->timeout))
1502                                 break;
1503
1504                         if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1505                                                 !is_idle(sbi, DISCARD_TIME)) {
1506                                 io_interrupted = true;
1507                                 break;
1508                         }
1509
1510                         __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1511
1512                         if (issued >= dpolicy->max_requests)
1513                                 break;
1514                 }
1515                 blk_finish_plug(&plug);
1516 next:
1517                 mutex_unlock(&dcc->cmd_lock);
1518
1519                 if (issued >= dpolicy->max_requests || io_interrupted)
1520                         break;
1521         }
1522
1523         if (!issued && io_interrupted)
1524                 issued = -1;
1525
1526         return issued;
1527 }
1528
1529 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1530 {
1531         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1532         struct list_head *pend_list;
1533         struct discard_cmd *dc, *tmp;
1534         int i;
1535         bool dropped = false;
1536
1537         mutex_lock(&dcc->cmd_lock);
1538         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1539                 pend_list = &dcc->pend_list[i];
1540                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1541                         f2fs_bug_on(sbi, dc->state != D_PREP);
1542                         __remove_discard_cmd(sbi, dc);
1543                         dropped = true;
1544                 }
1545         }
1546         mutex_unlock(&dcc->cmd_lock);
1547
1548         return dropped;
1549 }
1550
1551 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1552 {
1553         __drop_discard_cmd(sbi);
1554 }
1555
1556 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1557                                                         struct discard_cmd *dc)
1558 {
1559         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1560         unsigned int len = 0;
1561
1562         wait_for_completion_io(&dc->wait);
1563         mutex_lock(&dcc->cmd_lock);
1564         f2fs_bug_on(sbi, dc->state != D_DONE);
1565         dc->ref--;
1566         if (!dc->ref) {
1567                 if (!dc->error)
1568                         len = dc->len;
1569                 __remove_discard_cmd(sbi, dc);
1570         }
1571         mutex_unlock(&dcc->cmd_lock);
1572
1573         return len;
1574 }
1575
1576 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1577                                                 struct discard_policy *dpolicy,
1578                                                 block_t start, block_t end)
1579 {
1580         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1581         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1582                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1583         struct discard_cmd *dc, *tmp;
1584         bool need_wait;
1585         unsigned int trimmed = 0;
1586
1587 next:
1588         need_wait = false;
1589
1590         mutex_lock(&dcc->cmd_lock);
1591         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1592                 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1593                         continue;
1594                 if (dc->len < dpolicy->granularity)
1595                         continue;
1596                 if (dc->state == D_DONE && !dc->ref) {
1597                         wait_for_completion_io(&dc->wait);
1598                         if (!dc->error)
1599                                 trimmed += dc->len;
1600                         __remove_discard_cmd(sbi, dc);
1601                 } else {
1602                         dc->ref++;
1603                         need_wait = true;
1604                         break;
1605                 }
1606         }
1607         mutex_unlock(&dcc->cmd_lock);
1608
1609         if (need_wait) {
1610                 trimmed += __wait_one_discard_bio(sbi, dc);
1611                 goto next;
1612         }
1613
1614         return trimmed;
1615 }
1616
1617 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1618                                                 struct discard_policy *dpolicy)
1619 {
1620         struct discard_policy dp;
1621         unsigned int discard_blks;
1622
1623         if (dpolicy)
1624                 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1625
1626         /* wait all */
1627         __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1628         discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1629         __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1630         discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1631
1632         return discard_blks;
1633 }
1634
1635 /* This should be covered by global mutex, &sit_i->sentry_lock */
1636 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1637 {
1638         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1639         struct discard_cmd *dc;
1640         bool need_wait = false;
1641
1642         mutex_lock(&dcc->cmd_lock);
1643         dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1644                                                         NULL, blkaddr);
1645         if (dc) {
1646                 if (dc->state == D_PREP) {
1647                         __punch_discard_cmd(sbi, dc, blkaddr);
1648                 } else {
1649                         dc->ref++;
1650                         need_wait = true;
1651                 }
1652         }
1653         mutex_unlock(&dcc->cmd_lock);
1654
1655         if (need_wait)
1656                 __wait_one_discard_bio(sbi, dc);
1657 }
1658
1659 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1660 {
1661         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1662
1663         if (dcc && dcc->f2fs_issue_discard) {
1664                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1665
1666                 dcc->f2fs_issue_discard = NULL;
1667                 kthread_stop(discard_thread);
1668         }
1669 }
1670
1671 /* This comes from f2fs_put_super */
1672 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1673 {
1674         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1675         struct discard_policy dpolicy;
1676         bool dropped;
1677
1678         __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1679                                         dcc->discard_granularity);
1680         dpolicy.timeout = UMOUNT_DISCARD_TIMEOUT;
1681         __issue_discard_cmd(sbi, &dpolicy);
1682         dropped = __drop_discard_cmd(sbi);
1683
1684         /* just to make sure there is no pending discard commands */
1685         __wait_all_discard_cmd(sbi, NULL);
1686
1687         f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1688         return dropped;
1689 }
1690
1691 static int issue_discard_thread(void *data)
1692 {
1693         struct f2fs_sb_info *sbi = data;
1694         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1695         wait_queue_head_t *q = &dcc->discard_wait_queue;
1696         struct discard_policy dpolicy;
1697         unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1698         int issued;
1699
1700         set_freezable();
1701
1702         do {
1703                 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1704                                         dcc->discard_granularity);
1705
1706                 wait_event_interruptible_timeout(*q,
1707                                 kthread_should_stop() || freezing(current) ||
1708                                 dcc->discard_wake,
1709                                 msecs_to_jiffies(wait_ms));
1710
1711                 if (dcc->discard_wake)
1712                         dcc->discard_wake = 0;
1713
1714                 /* clean up pending candidates before going to sleep */
1715                 if (atomic_read(&dcc->queued_discard))
1716                         __wait_all_discard_cmd(sbi, NULL);
1717
1718                 if (try_to_freeze())
1719                         continue;
1720                 if (f2fs_readonly(sbi->sb))
1721                         continue;
1722                 if (kthread_should_stop())
1723                         return 0;
1724                 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1725                         wait_ms = dpolicy.max_interval;
1726                         continue;
1727                 }
1728
1729                 if (sbi->gc_mode == GC_URGENT)
1730                         __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1731
1732                 sb_start_intwrite(sbi->sb);
1733
1734                 issued = __issue_discard_cmd(sbi, &dpolicy);
1735                 if (issued > 0) {
1736                         __wait_all_discard_cmd(sbi, &dpolicy);
1737                         wait_ms = dpolicy.min_interval;
1738                 } else if (issued == -1){
1739                         wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1740                         if (!wait_ms)
1741                                 wait_ms = dpolicy.mid_interval;
1742                 } else {
1743                         wait_ms = dpolicy.max_interval;
1744                 }
1745
1746                 sb_end_intwrite(sbi->sb);
1747
1748         } while (!kthread_should_stop());
1749         return 0;
1750 }
1751
1752 #ifdef CONFIG_BLK_DEV_ZONED
1753 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1754                 struct block_device *bdev, block_t blkstart, block_t blklen)
1755 {
1756         sector_t sector, nr_sects;
1757         block_t lblkstart = blkstart;
1758         int devi = 0;
1759
1760         if (f2fs_is_multi_device(sbi)) {
1761                 devi = f2fs_target_device_index(sbi, blkstart);
1762                 if (blkstart < FDEV(devi).start_blk ||
1763                     blkstart > FDEV(devi).end_blk) {
1764                         f2fs_err(sbi, "Invalid block %x", blkstart);
1765                         return -EIO;
1766                 }
1767                 blkstart -= FDEV(devi).start_blk;
1768         }
1769
1770         /* For sequential zones, reset the zone write pointer */
1771         if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1772                 sector = SECTOR_FROM_BLOCK(blkstart);
1773                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1774
1775                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1776                                 nr_sects != bdev_zone_sectors(bdev)) {
1777                         f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1778                                  devi, sbi->s_ndevs ? FDEV(devi).path : "",
1779                                  blkstart, blklen);
1780                         return -EIO;
1781                 }
1782                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1783                 return blkdev_reset_zones(bdev, sector, nr_sects, GFP_NOFS);
1784         }
1785
1786         /* For conventional zones, use regular discard if supported */
1787         return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1788 }
1789 #endif
1790
1791 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1792                 struct block_device *bdev, block_t blkstart, block_t blklen)
1793 {
1794 #ifdef CONFIG_BLK_DEV_ZONED
1795         if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1796                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1797 #endif
1798         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1799 }
1800
1801 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1802                                 block_t blkstart, block_t blklen)
1803 {
1804         sector_t start = blkstart, len = 0;
1805         struct block_device *bdev;
1806         struct seg_entry *se;
1807         unsigned int offset;
1808         block_t i;
1809         int err = 0;
1810
1811         bdev = f2fs_target_device(sbi, blkstart, NULL);
1812
1813         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1814                 if (i != start) {
1815                         struct block_device *bdev2 =
1816                                 f2fs_target_device(sbi, i, NULL);
1817
1818                         if (bdev2 != bdev) {
1819                                 err = __issue_discard_async(sbi, bdev,
1820                                                 start, len);
1821                                 if (err)
1822                                         return err;
1823                                 bdev = bdev2;
1824                                 start = i;
1825                                 len = 0;
1826                         }
1827                 }
1828
1829                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1830                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1831
1832                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1833                         sbi->discard_blks--;
1834         }
1835
1836         if (len)
1837                 err = __issue_discard_async(sbi, bdev, start, len);
1838         return err;
1839 }
1840
1841 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1842                                                         bool check_only)
1843 {
1844         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1845         int max_blocks = sbi->blocks_per_seg;
1846         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1847         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1848         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1849         unsigned long *discard_map = (unsigned long *)se->discard_map;
1850         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1851         unsigned int start = 0, end = -1;
1852         bool force = (cpc->reason & CP_DISCARD);
1853         struct discard_entry *de = NULL;
1854         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1855         int i;
1856
1857         if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
1858                 return false;
1859
1860         if (!force) {
1861                 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1862                         SM_I(sbi)->dcc_info->nr_discards >=
1863                                 SM_I(sbi)->dcc_info->max_discards)
1864                         return false;
1865         }
1866
1867         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1868         for (i = 0; i < entries; i++)
1869                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1870                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1871
1872         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1873                                 SM_I(sbi)->dcc_info->max_discards) {
1874                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1875                 if (start >= max_blocks)
1876                         break;
1877
1878                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1879                 if (force && start && end != max_blocks
1880                                         && (end - start) < cpc->trim_minlen)
1881                         continue;
1882
1883                 if (check_only)
1884                         return true;
1885
1886                 if (!de) {
1887                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1888                                                                 GFP_F2FS_ZERO);
1889                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1890                         list_add_tail(&de->list, head);
1891                 }
1892
1893                 for (i = start; i < end; i++)
1894                         __set_bit_le(i, (void *)de->discard_map);
1895
1896                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1897         }
1898         return false;
1899 }
1900
1901 static void release_discard_addr(struct discard_entry *entry)
1902 {
1903         list_del(&entry->list);
1904         kmem_cache_free(discard_entry_slab, entry);
1905 }
1906
1907 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1908 {
1909         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1910         struct discard_entry *entry, *this;
1911
1912         /* drop caches */
1913         list_for_each_entry_safe(entry, this, head, list)
1914                 release_discard_addr(entry);
1915 }
1916
1917 /*
1918  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1919  */
1920 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1921 {
1922         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1923         unsigned int segno;
1924
1925         mutex_lock(&dirty_i->seglist_lock);
1926         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1927                 __set_test_and_free(sbi, segno);
1928         mutex_unlock(&dirty_i->seglist_lock);
1929 }
1930
1931 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1932                                                 struct cp_control *cpc)
1933 {
1934         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1935         struct list_head *head = &dcc->entry_list;
1936         struct discard_entry *entry, *this;
1937         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1938         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1939         unsigned int start = 0, end = -1;
1940         unsigned int secno, start_segno;
1941         bool force = (cpc->reason & CP_DISCARD);
1942         bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
1943
1944         mutex_lock(&dirty_i->seglist_lock);
1945
1946         while (1) {
1947                 int i;
1948
1949                 if (need_align && end != -1)
1950                         end--;
1951                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1952                 if (start >= MAIN_SEGS(sbi))
1953                         break;
1954                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1955                                                                 start + 1);
1956
1957                 if (need_align) {
1958                         start = rounddown(start, sbi->segs_per_sec);
1959                         end = roundup(end, sbi->segs_per_sec);
1960                 }
1961
1962                 for (i = start; i < end; i++) {
1963                         if (test_and_clear_bit(i, prefree_map))
1964                                 dirty_i->nr_dirty[PRE]--;
1965                 }
1966
1967                 if (!f2fs_realtime_discard_enable(sbi))
1968                         continue;
1969
1970                 if (force && start >= cpc->trim_start &&
1971                                         (end - 1) <= cpc->trim_end)
1972                                 continue;
1973
1974                 if (!test_opt(sbi, LFS) || !__is_large_section(sbi)) {
1975                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1976                                 (end - start) << sbi->log_blocks_per_seg);
1977                         continue;
1978                 }
1979 next:
1980                 secno = GET_SEC_FROM_SEG(sbi, start);
1981                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1982                 if (!IS_CURSEC(sbi, secno) &&
1983                         !get_valid_blocks(sbi, start, true))
1984                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1985                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1986
1987                 start = start_segno + sbi->segs_per_sec;
1988                 if (start < end)
1989                         goto next;
1990                 else
1991                         end = start - 1;
1992         }
1993         mutex_unlock(&dirty_i->seglist_lock);
1994
1995         /* send small discards */
1996         list_for_each_entry_safe(entry, this, head, list) {
1997                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1998                 bool is_valid = test_bit_le(0, entry->discard_map);
1999
2000 find_next:
2001                 if (is_valid) {
2002                         next_pos = find_next_zero_bit_le(entry->discard_map,
2003                                         sbi->blocks_per_seg, cur_pos);
2004                         len = next_pos - cur_pos;
2005
2006                         if (f2fs_sb_has_blkzoned(sbi) ||
2007                             (force && len < cpc->trim_minlen))
2008                                 goto skip;
2009
2010                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2011                                                                         len);
2012                         total_len += len;
2013                 } else {
2014                         next_pos = find_next_bit_le(entry->discard_map,
2015                                         sbi->blocks_per_seg, cur_pos);
2016                 }
2017 skip:
2018                 cur_pos = next_pos;
2019                 is_valid = !is_valid;
2020
2021                 if (cur_pos < sbi->blocks_per_seg)
2022                         goto find_next;
2023
2024                 release_discard_addr(entry);
2025                 dcc->nr_discards -= total_len;
2026         }
2027
2028         wake_up_discard_thread(sbi, false);
2029 }
2030
2031 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2032 {
2033         dev_t dev = sbi->sb->s_bdev->bd_dev;
2034         struct discard_cmd_control *dcc;
2035         int err = 0, i;
2036
2037         if (SM_I(sbi)->dcc_info) {
2038                 dcc = SM_I(sbi)->dcc_info;
2039                 goto init_thread;
2040         }
2041
2042         dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2043         if (!dcc)
2044                 return -ENOMEM;
2045
2046         dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2047         INIT_LIST_HEAD(&dcc->entry_list);
2048         for (i = 0; i < MAX_PLIST_NUM; i++)
2049                 INIT_LIST_HEAD(&dcc->pend_list[i]);
2050         INIT_LIST_HEAD(&dcc->wait_list);
2051         INIT_LIST_HEAD(&dcc->fstrim_list);
2052         mutex_init(&dcc->cmd_lock);
2053         atomic_set(&dcc->issued_discard, 0);
2054         atomic_set(&dcc->queued_discard, 0);
2055         atomic_set(&dcc->discard_cmd_cnt, 0);
2056         dcc->nr_discards = 0;
2057         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2058         dcc->undiscard_blks = 0;
2059         dcc->next_pos = 0;
2060         dcc->root = RB_ROOT_CACHED;
2061         dcc->rbtree_check = false;
2062
2063         init_waitqueue_head(&dcc->discard_wait_queue);
2064         SM_I(sbi)->dcc_info = dcc;
2065 init_thread:
2066         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2067                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2068         if (IS_ERR(dcc->f2fs_issue_discard)) {
2069                 err = PTR_ERR(dcc->f2fs_issue_discard);
2070                 kvfree(dcc);
2071                 SM_I(sbi)->dcc_info = NULL;
2072                 return err;
2073         }
2074
2075         return err;
2076 }
2077
2078 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2079 {
2080         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2081
2082         if (!dcc)
2083                 return;
2084
2085         f2fs_stop_discard_thread(sbi);
2086
2087         /*
2088          * Recovery can cache discard commands, so in error path of
2089          * fill_super(), it needs to give a chance to handle them.
2090          */
2091         if (unlikely(atomic_read(&dcc->discard_cmd_cnt)))
2092                 f2fs_issue_discard_timeout(sbi);
2093
2094         kvfree(dcc);
2095         SM_I(sbi)->dcc_info = NULL;
2096 }
2097
2098 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2099 {
2100         struct sit_info *sit_i = SIT_I(sbi);
2101
2102         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2103                 sit_i->dirty_sentries++;
2104                 return false;
2105         }
2106
2107         return true;
2108 }
2109
2110 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2111                                         unsigned int segno, int modified)
2112 {
2113         struct seg_entry *se = get_seg_entry(sbi, segno);
2114         se->type = type;
2115         if (modified)
2116                 __mark_sit_entry_dirty(sbi, segno);
2117 }
2118
2119 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2120 {
2121         struct seg_entry *se;
2122         unsigned int segno, offset;
2123         long int new_vblocks;
2124         bool exist;
2125 #ifdef CONFIG_F2FS_CHECK_FS
2126         bool mir_exist;
2127 #endif
2128
2129         segno = GET_SEGNO(sbi, blkaddr);
2130
2131         se = get_seg_entry(sbi, segno);
2132         new_vblocks = se->valid_blocks + del;
2133         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2134
2135         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
2136                                 (new_vblocks > sbi->blocks_per_seg)));
2137
2138         se->valid_blocks = new_vblocks;
2139         se->mtime = get_mtime(sbi, false);
2140         if (se->mtime > SIT_I(sbi)->max_mtime)
2141                 SIT_I(sbi)->max_mtime = se->mtime;
2142
2143         /* Update valid block bitmap */
2144         if (del > 0) {
2145                 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2146 #ifdef CONFIG_F2FS_CHECK_FS
2147                 mir_exist = f2fs_test_and_set_bit(offset,
2148                                                 se->cur_valid_map_mir);
2149                 if (unlikely(exist != mir_exist)) {
2150                         f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2151                                  blkaddr, exist);
2152                         f2fs_bug_on(sbi, 1);
2153                 }
2154 #endif
2155                 if (unlikely(exist)) {
2156                         f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2157                                  blkaddr);
2158                         f2fs_bug_on(sbi, 1);
2159                         se->valid_blocks--;
2160                         del = 0;
2161                 }
2162
2163                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
2164                         sbi->discard_blks--;
2165
2166                 /*
2167                  * SSR should never reuse block which is checkpointed
2168                  * or newly invalidated.
2169                  */
2170                 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2171                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2172                                 se->ckpt_valid_blocks++;
2173                 }
2174         } else {
2175                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2176 #ifdef CONFIG_F2FS_CHECK_FS
2177                 mir_exist = f2fs_test_and_clear_bit(offset,
2178                                                 se->cur_valid_map_mir);
2179                 if (unlikely(exist != mir_exist)) {
2180                         f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2181                                  blkaddr, exist);
2182                         f2fs_bug_on(sbi, 1);
2183                 }
2184 #endif
2185                 if (unlikely(!exist)) {
2186                         f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2187                                  blkaddr);
2188                         f2fs_bug_on(sbi, 1);
2189                         se->valid_blocks++;
2190                         del = 0;
2191                 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2192                         /*
2193                          * If checkpoints are off, we must not reuse data that
2194                          * was used in the previous checkpoint. If it was used
2195                          * before, we must track that to know how much space we
2196                          * really have.
2197                          */
2198                         if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2199                                 spin_lock(&sbi->stat_lock);
2200                                 sbi->unusable_block_count++;
2201                                 spin_unlock(&sbi->stat_lock);
2202                         }
2203                 }
2204
2205                 if (f2fs_test_and_clear_bit(offset, se->discard_map))
2206                         sbi->discard_blks++;
2207         }
2208         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2209                 se->ckpt_valid_blocks += del;
2210
2211         __mark_sit_entry_dirty(sbi, segno);
2212
2213         /* update total number of valid blocks to be written in ckpt area */
2214         SIT_I(sbi)->written_valid_blocks += del;
2215
2216         if (__is_large_section(sbi))
2217                 get_sec_entry(sbi, segno)->valid_blocks += del;
2218 }
2219
2220 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2221 {
2222         unsigned int segno = GET_SEGNO(sbi, addr);
2223         struct sit_info *sit_i = SIT_I(sbi);
2224
2225         f2fs_bug_on(sbi, addr == NULL_ADDR);
2226         if (addr == NEW_ADDR)
2227                 return;
2228
2229         invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2230
2231         /* add it into sit main buffer */
2232         down_write(&sit_i->sentry_lock);
2233
2234         update_sit_entry(sbi, addr, -1);
2235
2236         /* add it into dirty seglist */
2237         locate_dirty_segment(sbi, segno);
2238
2239         up_write(&sit_i->sentry_lock);
2240 }
2241
2242 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2243 {
2244         struct sit_info *sit_i = SIT_I(sbi);
2245         unsigned int segno, offset;
2246         struct seg_entry *se;
2247         bool is_cp = false;
2248
2249         if (!__is_valid_data_blkaddr(blkaddr))
2250                 return true;
2251
2252         down_read(&sit_i->sentry_lock);
2253
2254         segno = GET_SEGNO(sbi, blkaddr);
2255         se = get_seg_entry(sbi, segno);
2256         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2257
2258         if (f2fs_test_bit(offset, se->ckpt_valid_map))
2259                 is_cp = true;
2260
2261         up_read(&sit_i->sentry_lock);
2262
2263         return is_cp;
2264 }
2265
2266 /*
2267  * This function should be resided under the curseg_mutex lock
2268  */
2269 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2270                                         struct f2fs_summary *sum)
2271 {
2272         struct curseg_info *curseg = CURSEG_I(sbi, type);
2273         void *addr = curseg->sum_blk;
2274         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2275         memcpy(addr, sum, sizeof(struct f2fs_summary));
2276 }
2277
2278 /*
2279  * Calculate the number of current summary pages for writing
2280  */
2281 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2282 {
2283         int valid_sum_count = 0;
2284         int i, sum_in_page;
2285
2286         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2287                 if (sbi->ckpt->alloc_type[i] == SSR)
2288                         valid_sum_count += sbi->blocks_per_seg;
2289                 else {
2290                         if (for_ra)
2291                                 valid_sum_count += le16_to_cpu(
2292                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2293                         else
2294                                 valid_sum_count += curseg_blkoff(sbi, i);
2295                 }
2296         }
2297
2298         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2299                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2300         if (valid_sum_count <= sum_in_page)
2301                 return 1;
2302         else if ((valid_sum_count - sum_in_page) <=
2303                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2304                 return 2;
2305         return 3;
2306 }
2307
2308 /*
2309  * Caller should put this summary page
2310  */
2311 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2312 {
2313         return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
2314 }
2315
2316 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2317                                         void *src, block_t blk_addr)
2318 {
2319         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2320
2321         memcpy(page_address(page), src, PAGE_SIZE);
2322         set_page_dirty(page);
2323         f2fs_put_page(page, 1);
2324 }
2325
2326 static void write_sum_page(struct f2fs_sb_info *sbi,
2327                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
2328 {
2329         f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2330 }
2331
2332 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2333                                                 int type, block_t blk_addr)
2334 {
2335         struct curseg_info *curseg = CURSEG_I(sbi, type);
2336         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2337         struct f2fs_summary_block *src = curseg->sum_blk;
2338         struct f2fs_summary_block *dst;
2339
2340         dst = (struct f2fs_summary_block *)page_address(page);
2341         memset(dst, 0, PAGE_SIZE);
2342
2343         mutex_lock(&curseg->curseg_mutex);
2344
2345         down_read(&curseg->journal_rwsem);
2346         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2347         up_read(&curseg->journal_rwsem);
2348
2349         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2350         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2351
2352         mutex_unlock(&curseg->curseg_mutex);
2353
2354         set_page_dirty(page);
2355         f2fs_put_page(page, 1);
2356 }
2357
2358 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2359 {
2360         struct curseg_info *curseg = CURSEG_I(sbi, type);
2361         unsigned int segno = curseg->segno + 1;
2362         struct free_segmap_info *free_i = FREE_I(sbi);
2363
2364         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2365                 return !test_bit(segno, free_i->free_segmap);
2366         return 0;
2367 }
2368
2369 /*
2370  * Find a new segment from the free segments bitmap to right order
2371  * This function should be returned with success, otherwise BUG
2372  */
2373 static void get_new_segment(struct f2fs_sb_info *sbi,
2374                         unsigned int *newseg, bool new_sec, int dir)
2375 {
2376         struct free_segmap_info *free_i = FREE_I(sbi);
2377         unsigned int segno, secno, zoneno;
2378         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2379         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2380         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2381         unsigned int left_start = hint;
2382         bool init = true;
2383         int go_left = 0;
2384         int i;
2385
2386         spin_lock(&free_i->segmap_lock);
2387
2388         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2389                 segno = find_next_zero_bit(free_i->free_segmap,
2390                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2391                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2392                         goto got_it;
2393         }
2394 find_other_zone:
2395         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2396         if (secno >= MAIN_SECS(sbi)) {
2397                 if (dir == ALLOC_RIGHT) {
2398                         secno = find_next_zero_bit(free_i->free_secmap,
2399                                                         MAIN_SECS(sbi), 0);
2400                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2401                 } else {
2402                         go_left = 1;
2403                         left_start = hint - 1;
2404                 }
2405         }
2406         if (go_left == 0)
2407                 goto skip_left;
2408
2409         while (test_bit(left_start, free_i->free_secmap)) {
2410                 if (left_start > 0) {
2411                         left_start--;
2412                         continue;
2413                 }
2414                 left_start = find_next_zero_bit(free_i->free_secmap,
2415                                                         MAIN_SECS(sbi), 0);
2416                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2417                 break;
2418         }
2419         secno = left_start;
2420 skip_left:
2421         segno = GET_SEG_FROM_SEC(sbi, secno);
2422         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2423
2424         /* give up on finding another zone */
2425         if (!init)
2426                 goto got_it;
2427         if (sbi->secs_per_zone == 1)
2428                 goto got_it;
2429         if (zoneno == old_zoneno)
2430                 goto got_it;
2431         if (dir == ALLOC_LEFT) {
2432                 if (!go_left && zoneno + 1 >= total_zones)
2433                         goto got_it;
2434                 if (go_left && zoneno == 0)
2435                         goto got_it;
2436         }
2437         for (i = 0; i < NR_CURSEG_TYPE; i++)
2438                 if (CURSEG_I(sbi, i)->zone == zoneno)
2439                         break;
2440
2441         if (i < NR_CURSEG_TYPE) {
2442                 /* zone is in user, try another */
2443                 if (go_left)
2444                         hint = zoneno * sbi->secs_per_zone - 1;
2445                 else if (zoneno + 1 >= total_zones)
2446                         hint = 0;
2447                 else
2448                         hint = (zoneno + 1) * sbi->secs_per_zone;
2449                 init = false;
2450                 goto find_other_zone;
2451         }
2452 got_it:
2453         /* set it as dirty segment in free segmap */
2454         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2455         __set_inuse(sbi, segno);
2456         *newseg = segno;
2457         spin_unlock(&free_i->segmap_lock);
2458 }
2459
2460 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2461 {
2462         struct curseg_info *curseg = CURSEG_I(sbi, type);
2463         struct summary_footer *sum_footer;
2464
2465         curseg->segno = curseg->next_segno;
2466         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2467         curseg->next_blkoff = 0;
2468         curseg->next_segno = NULL_SEGNO;
2469
2470         sum_footer = &(curseg->sum_blk->footer);
2471         memset(sum_footer, 0, sizeof(struct summary_footer));
2472         if (IS_DATASEG(type))
2473                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2474         if (IS_NODESEG(type))
2475                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2476         __set_sit_entry_type(sbi, type, curseg->segno, modified);
2477 }
2478
2479 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2480 {
2481         /* if segs_per_sec is large than 1, we need to keep original policy. */
2482         if (__is_large_section(sbi))
2483                 return CURSEG_I(sbi, type)->segno;
2484
2485         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2486                 return 0;
2487
2488         if (test_opt(sbi, NOHEAP) &&
2489                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2490                 return 0;
2491
2492         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2493                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2494
2495         /* find segments from 0 to reuse freed segments */
2496         if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2497                 return 0;
2498
2499         return CURSEG_I(sbi, type)->segno;
2500 }
2501
2502 /*
2503  * Allocate a current working segment.
2504  * This function always allocates a free segment in LFS manner.
2505  */
2506 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2507 {
2508         struct curseg_info *curseg = CURSEG_I(sbi, type);
2509         unsigned int segno = curseg->segno;
2510         int dir = ALLOC_LEFT;
2511
2512         write_sum_page(sbi, curseg->sum_blk,
2513                                 GET_SUM_BLOCK(sbi, segno));
2514         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2515                 dir = ALLOC_RIGHT;
2516
2517         if (test_opt(sbi, NOHEAP))
2518                 dir = ALLOC_RIGHT;
2519
2520         segno = __get_next_segno(sbi, type);
2521         get_new_segment(sbi, &segno, new_sec, dir);
2522         curseg->next_segno = segno;
2523         reset_curseg(sbi, type, 1);
2524         curseg->alloc_type = LFS;
2525 }
2526
2527 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2528                         struct curseg_info *seg, block_t start)
2529 {
2530         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2531         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2532         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2533         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2534         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2535         int i, pos;
2536
2537         for (i = 0; i < entries; i++)
2538                 target_map[i] = ckpt_map[i] | cur_map[i];
2539
2540         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2541
2542         seg->next_blkoff = pos;
2543 }
2544
2545 /*
2546  * If a segment is written by LFS manner, next block offset is just obtained
2547  * by increasing the current block offset. However, if a segment is written by
2548  * SSR manner, next block offset obtained by calling __next_free_blkoff
2549  */
2550 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2551                                 struct curseg_info *seg)
2552 {
2553         if (seg->alloc_type == SSR)
2554                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2555         else
2556                 seg->next_blkoff++;
2557 }
2558
2559 /*
2560  * This function always allocates a used segment(from dirty seglist) by SSR
2561  * manner, so it should recover the existing segment information of valid blocks
2562  */
2563 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2564 {
2565         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2566         struct curseg_info *curseg = CURSEG_I(sbi, type);
2567         unsigned int new_segno = curseg->next_segno;
2568         struct f2fs_summary_block *sum_node;
2569         struct page *sum_page;
2570
2571         write_sum_page(sbi, curseg->sum_blk,
2572                                 GET_SUM_BLOCK(sbi, curseg->segno));
2573         __set_test_and_inuse(sbi, new_segno);
2574
2575         mutex_lock(&dirty_i->seglist_lock);
2576         __remove_dirty_segment(sbi, new_segno, PRE);
2577         __remove_dirty_segment(sbi, new_segno, DIRTY);
2578         mutex_unlock(&dirty_i->seglist_lock);
2579
2580         reset_curseg(sbi, type, 1);
2581         curseg->alloc_type = SSR;
2582         __next_free_blkoff(sbi, curseg, 0);
2583
2584         sum_page = f2fs_get_sum_page(sbi, new_segno);
2585         f2fs_bug_on(sbi, IS_ERR(sum_page));
2586         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2587         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2588         f2fs_put_page(sum_page, 1);
2589 }
2590
2591 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2592 {
2593         struct curseg_info *curseg = CURSEG_I(sbi, type);
2594         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2595         unsigned segno = NULL_SEGNO;
2596         int i, cnt;
2597         bool reversed = false;
2598
2599         /* f2fs_need_SSR() already forces to do this */
2600         if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2601                 curseg->next_segno = segno;
2602                 return 1;
2603         }
2604
2605         /* For node segments, let's do SSR more intensively */
2606         if (IS_NODESEG(type)) {
2607                 if (type >= CURSEG_WARM_NODE) {
2608                         reversed = true;
2609                         i = CURSEG_COLD_NODE;
2610                 } else {
2611                         i = CURSEG_HOT_NODE;
2612                 }
2613                 cnt = NR_CURSEG_NODE_TYPE;
2614         } else {
2615                 if (type >= CURSEG_WARM_DATA) {
2616                         reversed = true;
2617                         i = CURSEG_COLD_DATA;
2618                 } else {
2619                         i = CURSEG_HOT_DATA;
2620                 }
2621                 cnt = NR_CURSEG_DATA_TYPE;
2622         }
2623
2624         for (; cnt-- > 0; reversed ? i-- : i++) {
2625                 if (i == type)
2626                         continue;
2627                 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2628                         curseg->next_segno = segno;
2629                         return 1;
2630                 }
2631         }
2632
2633         /* find valid_blocks=0 in dirty list */
2634         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2635                 segno = get_free_segment(sbi);
2636                 if (segno != NULL_SEGNO) {
2637                         curseg->next_segno = segno;
2638                         return 1;
2639                 }
2640         }
2641         return 0;
2642 }
2643
2644 /*
2645  * flush out current segment and replace it with new segment
2646  * This function should be returned with success, otherwise BUG
2647  */
2648 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2649                                                 int type, bool force)
2650 {
2651         struct curseg_info *curseg = CURSEG_I(sbi, type);
2652
2653         if (force)
2654                 new_curseg(sbi, type, true);
2655         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2656                                         type == CURSEG_WARM_NODE)
2657                 new_curseg(sbi, type, false);
2658         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2659                         likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2660                 new_curseg(sbi, type, false);
2661         else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2662                 change_curseg(sbi, type);
2663         else
2664                 new_curseg(sbi, type, false);
2665
2666         stat_inc_seg_type(sbi, curseg);
2667 }
2668
2669 void allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2670                                         unsigned int start, unsigned int end)
2671 {
2672         struct curseg_info *curseg = CURSEG_I(sbi, type);
2673         unsigned int segno;
2674
2675         down_read(&SM_I(sbi)->curseg_lock);
2676         mutex_lock(&curseg->curseg_mutex);
2677         down_write(&SIT_I(sbi)->sentry_lock);
2678
2679         segno = CURSEG_I(sbi, type)->segno;
2680         if (segno < start || segno > end)
2681                 goto unlock;
2682
2683         if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2684                 change_curseg(sbi, type);
2685         else
2686                 new_curseg(sbi, type, true);
2687
2688         stat_inc_seg_type(sbi, curseg);
2689
2690         locate_dirty_segment(sbi, segno);
2691 unlock:
2692         up_write(&SIT_I(sbi)->sentry_lock);
2693
2694         if (segno != curseg->segno)
2695                 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2696                             type, segno, curseg->segno);
2697
2698         mutex_unlock(&curseg->curseg_mutex);
2699         up_read(&SM_I(sbi)->curseg_lock);
2700 }
2701
2702 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2703 {
2704         struct curseg_info *curseg;
2705         unsigned int old_segno;
2706         int i;
2707
2708         down_write(&SIT_I(sbi)->sentry_lock);
2709
2710         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2711                 curseg = CURSEG_I(sbi, i);
2712                 old_segno = curseg->segno;
2713                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2714                 locate_dirty_segment(sbi, old_segno);
2715         }
2716
2717         up_write(&SIT_I(sbi)->sentry_lock);
2718 }
2719
2720 static const struct segment_allocation default_salloc_ops = {
2721         .allocate_segment = allocate_segment_by_default,
2722 };
2723
2724 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2725                                                 struct cp_control *cpc)
2726 {
2727         __u64 trim_start = cpc->trim_start;
2728         bool has_candidate = false;
2729
2730         down_write(&SIT_I(sbi)->sentry_lock);
2731         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2732                 if (add_discard_addrs(sbi, cpc, true)) {
2733                         has_candidate = true;
2734                         break;
2735                 }
2736         }
2737         up_write(&SIT_I(sbi)->sentry_lock);
2738
2739         cpc->trim_start = trim_start;
2740         return has_candidate;
2741 }
2742
2743 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2744                                         struct discard_policy *dpolicy,
2745                                         unsigned int start, unsigned int end)
2746 {
2747         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2748         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2749         struct rb_node **insert_p = NULL, *insert_parent = NULL;
2750         struct discard_cmd *dc;
2751         struct blk_plug plug;
2752         int issued;
2753         unsigned int trimmed = 0;
2754
2755 next:
2756         issued = 0;
2757
2758         mutex_lock(&dcc->cmd_lock);
2759         if (unlikely(dcc->rbtree_check))
2760                 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2761                                                                 &dcc->root));
2762
2763         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2764                                         NULL, start,
2765                                         (struct rb_entry **)&prev_dc,
2766                                         (struct rb_entry **)&next_dc,
2767                                         &insert_p, &insert_parent, true, NULL);
2768         if (!dc)
2769                 dc = next_dc;
2770
2771         blk_start_plug(&plug);
2772
2773         while (dc && dc->lstart <= end) {
2774                 struct rb_node *node;
2775                 int err = 0;
2776
2777                 if (dc->len < dpolicy->granularity)
2778                         goto skip;
2779
2780                 if (dc->state != D_PREP) {
2781                         list_move_tail(&dc->list, &dcc->fstrim_list);
2782                         goto skip;
2783                 }
2784
2785                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2786
2787                 if (issued >= dpolicy->max_requests) {
2788                         start = dc->lstart + dc->len;
2789
2790                         if (err)
2791                                 __remove_discard_cmd(sbi, dc);
2792
2793                         blk_finish_plug(&plug);
2794                         mutex_unlock(&dcc->cmd_lock);
2795                         trimmed += __wait_all_discard_cmd(sbi, NULL);
2796                         congestion_wait(BLK_RW_ASYNC, HZ/50);
2797                         goto next;
2798                 }
2799 skip:
2800                 node = rb_next(&dc->rb_node);
2801                 if (err)
2802                         __remove_discard_cmd(sbi, dc);
2803                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2804
2805                 if (fatal_signal_pending(current))
2806                         break;
2807         }
2808
2809         blk_finish_plug(&plug);
2810         mutex_unlock(&dcc->cmd_lock);
2811
2812         return trimmed;
2813 }
2814
2815 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2816 {
2817         __u64 start = F2FS_BYTES_TO_BLK(range->start);
2818         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2819         unsigned int start_segno, end_segno;
2820         block_t start_block, end_block;
2821         struct cp_control cpc;
2822         struct discard_policy dpolicy;
2823         unsigned long long trimmed = 0;
2824         int err = 0;
2825         bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
2826
2827         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2828                 return -EINVAL;
2829
2830         if (end < MAIN_BLKADDR(sbi))
2831                 goto out;
2832
2833         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2834                 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
2835                 return -EFSCORRUPTED;
2836         }
2837
2838         /* start/end segment number in main_area */
2839         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2840         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2841                                                 GET_SEGNO(sbi, end);
2842         if (need_align) {
2843                 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2844                 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2845         }
2846
2847         cpc.reason = CP_DISCARD;
2848         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2849         cpc.trim_start = start_segno;
2850         cpc.trim_end = end_segno;
2851
2852         if (sbi->discard_blks == 0)
2853                 goto out;
2854
2855         mutex_lock(&sbi->gc_mutex);
2856         err = f2fs_write_checkpoint(sbi, &cpc);
2857         mutex_unlock(&sbi->gc_mutex);
2858         if (err)
2859                 goto out;
2860
2861         /*
2862          * We filed discard candidates, but actually we don't need to wait for
2863          * all of them, since they'll be issued in idle time along with runtime
2864          * discard option. User configuration looks like using runtime discard
2865          * or periodic fstrim instead of it.
2866          */
2867         if (f2fs_realtime_discard_enable(sbi))
2868                 goto out;
2869
2870         start_block = START_BLOCK(sbi, start_segno);
2871         end_block = START_BLOCK(sbi, end_segno + 1);
2872
2873         __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2874         trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2875                                         start_block, end_block);
2876
2877         trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
2878                                         start_block, end_block);
2879 out:
2880         if (!err)
2881                 range->len = F2FS_BLK_TO_BYTES(trimmed);
2882         return err;
2883 }
2884
2885 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2886 {
2887         struct curseg_info *curseg = CURSEG_I(sbi, type);
2888         if (curseg->next_blkoff < sbi->blocks_per_seg)
2889                 return true;
2890         return false;
2891 }
2892
2893 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
2894 {
2895         switch (hint) {
2896         case WRITE_LIFE_SHORT:
2897                 return CURSEG_HOT_DATA;
2898         case WRITE_LIFE_EXTREME:
2899                 return CURSEG_COLD_DATA;
2900         default:
2901                 return CURSEG_WARM_DATA;
2902         }
2903 }
2904
2905 /* This returns write hints for each segment type. This hints will be
2906  * passed down to block layer. There are mapping tables which depend on
2907  * the mount option 'whint_mode'.
2908  *
2909  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2910  *
2911  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2912  *
2913  * User                  F2FS                     Block
2914  * ----                  ----                     -----
2915  *                       META                     WRITE_LIFE_NOT_SET
2916  *                       HOT_NODE                 "
2917  *                       WARM_NODE                "
2918  *                       COLD_NODE                "
2919  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2920  * extension list        "                        "
2921  *
2922  * -- buffered io
2923  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2924  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2925  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2926  * WRITE_LIFE_NONE       "                        "
2927  * WRITE_LIFE_MEDIUM     "                        "
2928  * WRITE_LIFE_LONG       "                        "
2929  *
2930  * -- direct io
2931  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2932  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2933  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2934  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2935  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2936  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2937  *
2938  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2939  *
2940  * User                  F2FS                     Block
2941  * ----                  ----                     -----
2942  *                       META                     WRITE_LIFE_MEDIUM;
2943  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
2944  *                       WARM_NODE                "
2945  *                       COLD_NODE                WRITE_LIFE_NONE
2946  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2947  * extension list        "                        "
2948  *
2949  * -- buffered io
2950  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2951  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2952  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
2953  * WRITE_LIFE_NONE       "                        "
2954  * WRITE_LIFE_MEDIUM     "                        "
2955  * WRITE_LIFE_LONG       "                        "
2956  *
2957  * -- direct io
2958  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2959  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2960  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2961  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2962  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2963  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2964  */
2965
2966 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2967                                 enum page_type type, enum temp_type temp)
2968 {
2969         if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2970                 if (type == DATA) {
2971                         if (temp == WARM)
2972                                 return WRITE_LIFE_NOT_SET;
2973                         else if (temp == HOT)
2974                                 return WRITE_LIFE_SHORT;
2975                         else if (temp == COLD)
2976                                 return WRITE_LIFE_EXTREME;
2977                 } else {
2978                         return WRITE_LIFE_NOT_SET;
2979                 }
2980         } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2981                 if (type == DATA) {
2982                         if (temp == WARM)
2983                                 return WRITE_LIFE_LONG;
2984                         else if (temp == HOT)
2985                                 return WRITE_LIFE_SHORT;
2986                         else if (temp == COLD)
2987                                 return WRITE_LIFE_EXTREME;
2988                 } else if (type == NODE) {
2989                         if (temp == WARM || temp == HOT)
2990                                 return WRITE_LIFE_NOT_SET;
2991                         else if (temp == COLD)
2992                                 return WRITE_LIFE_NONE;
2993                 } else if (type == META) {
2994                         return WRITE_LIFE_MEDIUM;
2995                 }
2996         }
2997         return WRITE_LIFE_NOT_SET;
2998 }
2999
3000 static int __get_segment_type_2(struct f2fs_io_info *fio)
3001 {
3002         if (fio->type == DATA)
3003                 return CURSEG_HOT_DATA;
3004         else
3005                 return CURSEG_HOT_NODE;
3006 }
3007
3008 static int __get_segment_type_4(struct f2fs_io_info *fio)
3009 {
3010         if (fio->type == DATA) {
3011                 struct inode *inode = fio->page->mapping->host;
3012
3013                 if (S_ISDIR(inode->i_mode))
3014                         return CURSEG_HOT_DATA;
3015                 else
3016                         return CURSEG_COLD_DATA;
3017         } else {
3018                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3019                         return CURSEG_WARM_NODE;
3020                 else
3021                         return CURSEG_COLD_NODE;
3022         }
3023 }
3024
3025 static int __get_segment_type_6(struct f2fs_io_info *fio)
3026 {
3027         if (fio->type == DATA) {
3028                 struct inode *inode = fio->page->mapping->host;
3029
3030                 if (is_cold_data(fio->page) || file_is_cold(inode))
3031                         return CURSEG_COLD_DATA;
3032                 if (file_is_hot(inode) ||
3033                                 is_inode_flag_set(inode, FI_HOT_DATA) ||
3034                                 f2fs_is_atomic_file(inode) ||
3035                                 f2fs_is_volatile_file(inode))
3036                         return CURSEG_HOT_DATA;
3037                 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3038         } else {
3039                 if (IS_DNODE(fio->page))
3040                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3041                                                 CURSEG_HOT_NODE;
3042                 return CURSEG_COLD_NODE;
3043         }
3044 }
3045
3046 static int __get_segment_type(struct f2fs_io_info *fio)
3047 {
3048         int type = 0;
3049
3050         switch (F2FS_OPTION(fio->sbi).active_logs) {
3051         case 2:
3052                 type = __get_segment_type_2(fio);
3053                 break;
3054         case 4:
3055                 type = __get_segment_type_4(fio);
3056                 break;
3057         case 6:
3058                 type = __get_segment_type_6(fio);
3059                 break;
3060         default:
3061                 f2fs_bug_on(fio->sbi, true);
3062         }
3063
3064         if (IS_HOT(type))
3065                 fio->temp = HOT;
3066         else if (IS_WARM(type))
3067                 fio->temp = WARM;
3068         else
3069                 fio->temp = COLD;
3070         return type;
3071 }
3072
3073 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3074                 block_t old_blkaddr, block_t *new_blkaddr,
3075                 struct f2fs_summary *sum, int type,
3076                 struct f2fs_io_info *fio, bool add_list)
3077 {
3078         struct sit_info *sit_i = SIT_I(sbi);
3079         struct curseg_info *curseg = CURSEG_I(sbi, type);
3080
3081         down_read(&SM_I(sbi)->curseg_lock);
3082
3083         mutex_lock(&curseg->curseg_mutex);
3084         down_write(&sit_i->sentry_lock);
3085
3086         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3087
3088         f2fs_wait_discard_bio(sbi, *new_blkaddr);
3089
3090         /*
3091          * __add_sum_entry should be resided under the curseg_mutex
3092          * because, this function updates a summary entry in the
3093          * current summary block.
3094          */
3095         __add_sum_entry(sbi, type, sum);
3096
3097         __refresh_next_blkoff(sbi, curseg);
3098
3099         stat_inc_block_count(sbi, curseg);
3100
3101         /*
3102          * SIT information should be updated before segment allocation,
3103          * since SSR needs latest valid block information.
3104          */
3105         update_sit_entry(sbi, *new_blkaddr, 1);
3106         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3107                 update_sit_entry(sbi, old_blkaddr, -1);
3108
3109         if (!__has_curseg_space(sbi, type))
3110                 sit_i->s_ops->allocate_segment(sbi, type, false);
3111
3112         /*
3113          * segment dirty status should be updated after segment allocation,
3114          * so we just need to update status only one time after previous
3115          * segment being closed.
3116          */
3117         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3118         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3119
3120         up_write(&sit_i->sentry_lock);
3121
3122         if (page && IS_NODESEG(type)) {
3123                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3124
3125                 f2fs_inode_chksum_set(sbi, page);
3126         }
3127
3128         if (F2FS_IO_ALIGNED(sbi))
3129                 fio->retry = false;
3130
3131         if (add_list) {
3132                 struct f2fs_bio_info *io;
3133
3134                 INIT_LIST_HEAD(&fio->list);
3135                 fio->in_list = true;
3136                 io = sbi->write_io[fio->type] + fio->temp;
3137                 spin_lock(&io->io_lock);
3138                 list_add_tail(&fio->list, &io->io_list);
3139                 spin_unlock(&io->io_lock);
3140         }
3141
3142         mutex_unlock(&curseg->curseg_mutex);
3143
3144         up_read(&SM_I(sbi)->curseg_lock);
3145 }
3146
3147 static void update_device_state(struct f2fs_io_info *fio)
3148 {
3149         struct f2fs_sb_info *sbi = fio->sbi;
3150         unsigned int devidx;
3151
3152         if (!f2fs_is_multi_device(sbi))
3153                 return;
3154
3155         devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3156
3157         /* update device state for fsync */
3158         f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
3159
3160         /* update device state for checkpoint */
3161         if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3162                 spin_lock(&sbi->dev_lock);
3163                 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3164                 spin_unlock(&sbi->dev_lock);
3165         }
3166 }
3167
3168 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3169 {
3170         int type = __get_segment_type(fio);
3171         bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
3172
3173         if (keep_order)
3174                 down_read(&fio->sbi->io_order_lock);
3175 reallocate:
3176         f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3177                         &fio->new_blkaddr, sum, type, fio, true);
3178         if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3179                 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3180                                         fio->old_blkaddr, fio->old_blkaddr);
3181
3182         /* writeout dirty page into bdev */
3183         f2fs_submit_page_write(fio);
3184         if (fio->retry) {
3185                 fio->old_blkaddr = fio->new_blkaddr;
3186                 goto reallocate;
3187         }
3188
3189         update_device_state(fio);
3190
3191         if (keep_order)
3192                 up_read(&fio->sbi->io_order_lock);
3193 }
3194
3195 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3196                                         enum iostat_type io_type)
3197 {
3198         struct f2fs_io_info fio = {
3199                 .sbi = sbi,
3200                 .type = META,
3201                 .temp = HOT,
3202                 .op = REQ_OP_WRITE,
3203                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3204                 .old_blkaddr = page->index,
3205                 .new_blkaddr = page->index,
3206                 .page = page,
3207                 .encrypted_page = NULL,
3208                 .in_list = false,
3209         };
3210
3211         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3212                 fio.op_flags &= ~REQ_META;
3213
3214         set_page_writeback(page);
3215         ClearPageError(page);
3216         f2fs_submit_page_write(&fio);
3217
3218         stat_inc_meta_count(sbi, page->index);
3219         f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3220 }
3221
3222 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3223 {
3224         struct f2fs_summary sum;
3225
3226         set_summary(&sum, nid, 0, 0);
3227         do_write_page(&sum, fio);
3228
3229         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3230 }
3231
3232 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3233                                         struct f2fs_io_info *fio)
3234 {
3235         struct f2fs_sb_info *sbi = fio->sbi;
3236         struct f2fs_summary sum;
3237
3238         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3239         set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3240         do_write_page(&sum, fio);
3241         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3242
3243         f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3244 }
3245
3246 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3247 {
3248         int err;
3249         struct f2fs_sb_info *sbi = fio->sbi;
3250         unsigned int segno;
3251
3252         fio->new_blkaddr = fio->old_blkaddr;
3253         /* i/o temperature is needed for passing down write hints */
3254         __get_segment_type(fio);
3255
3256         segno = GET_SEGNO(sbi, fio->new_blkaddr);
3257
3258         if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3259                 set_sbi_flag(sbi, SBI_NEED_FSCK);
3260                 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3261                           __func__, segno);
3262                 return -EFSCORRUPTED;
3263         }
3264
3265         stat_inc_inplace_blocks(fio->sbi);
3266
3267         if (fio->bio)
3268                 err = f2fs_merge_page_bio(fio);
3269         else
3270                 err = f2fs_submit_page_bio(fio);
3271         if (!err) {
3272                 update_device_state(fio);
3273                 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3274         }
3275
3276         return err;
3277 }
3278
3279 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3280                                                 unsigned int segno)
3281 {
3282         int i;
3283
3284         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3285                 if (CURSEG_I(sbi, i)->segno == segno)
3286                         break;
3287         }
3288         return i;
3289 }
3290
3291 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3292                                 block_t old_blkaddr, block_t new_blkaddr,
3293                                 bool recover_curseg, bool recover_newaddr)
3294 {
3295         struct sit_info *sit_i = SIT_I(sbi);
3296         struct curseg_info *curseg;
3297         unsigned int segno, old_cursegno;
3298         struct seg_entry *se;
3299         int type;
3300         unsigned short old_blkoff;
3301
3302         segno = GET_SEGNO(sbi, new_blkaddr);
3303         se = get_seg_entry(sbi, segno);
3304         type = se->type;
3305
3306         down_write(&SM_I(sbi)->curseg_lock);
3307
3308         if (!recover_curseg) {
3309                 /* for recovery flow */
3310                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3311                         if (old_blkaddr == NULL_ADDR)
3312                                 type = CURSEG_COLD_DATA;
3313                         else
3314                                 type = CURSEG_WARM_DATA;
3315                 }
3316         } else {
3317                 if (IS_CURSEG(sbi, segno)) {
3318                         /* se->type is volatile as SSR allocation */
3319                         type = __f2fs_get_curseg(sbi, segno);
3320                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3321                 } else {
3322                         type = CURSEG_WARM_DATA;
3323                 }
3324         }
3325
3326         f2fs_bug_on(sbi, !IS_DATASEG(type));
3327         curseg = CURSEG_I(sbi, type);
3328
3329         mutex_lock(&curseg->curseg_mutex);
3330         down_write(&sit_i->sentry_lock);
3331
3332         old_cursegno = curseg->segno;
3333         old_blkoff = curseg->next_blkoff;
3334
3335         /* change the current segment */
3336         if (segno != curseg->segno) {
3337                 curseg->next_segno = segno;
3338                 change_curseg(sbi, type);
3339         }
3340
3341         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3342         __add_sum_entry(sbi, type, sum);
3343
3344         if (!recover_curseg || recover_newaddr)
3345                 update_sit_entry(sbi, new_blkaddr, 1);
3346         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3347                 invalidate_mapping_pages(META_MAPPING(sbi),
3348                                         old_blkaddr, old_blkaddr);
3349                 update_sit_entry(sbi, old_blkaddr, -1);
3350         }
3351
3352         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3353         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3354
3355         locate_dirty_segment(sbi, old_cursegno);
3356
3357         if (recover_curseg) {
3358                 if (old_cursegno != curseg->segno) {
3359                         curseg->next_segno = old_cursegno;
3360                         change_curseg(sbi, type);
3361                 }
3362                 curseg->next_blkoff = old_blkoff;
3363         }
3364
3365         up_write(&sit_i->sentry_lock);
3366         mutex_unlock(&curseg->curseg_mutex);
3367         up_write(&SM_I(sbi)->curseg_lock);
3368 }
3369
3370 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3371                                 block_t old_addr, block_t new_addr,
3372                                 unsigned char version, bool recover_curseg,
3373                                 bool recover_newaddr)
3374 {
3375         struct f2fs_summary sum;
3376
3377         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3378
3379         f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3380                                         recover_curseg, recover_newaddr);
3381
3382         f2fs_update_data_blkaddr(dn, new_addr);
3383 }
3384
3385 void f2fs_wait_on_page_writeback(struct page *page,
3386                                 enum page_type type, bool ordered, bool locked)
3387 {
3388         if (PageWriteback(page)) {
3389                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3390
3391                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3392                 if (ordered) {
3393                         wait_on_page_writeback(page);
3394                         f2fs_bug_on(sbi, locked && PageWriteback(page));
3395                 } else {
3396                         wait_for_stable_page(page);
3397                 }
3398         }
3399 }
3400
3401 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3402 {
3403         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3404         struct page *cpage;
3405
3406         if (!f2fs_post_read_required(inode))
3407                 return;
3408
3409         if (!__is_valid_data_blkaddr(blkaddr))
3410                 return;
3411
3412         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3413         if (cpage) {
3414                 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3415                 f2fs_put_page(cpage, 1);
3416         }
3417 }
3418
3419 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3420                                                                 block_t len)
3421 {
3422         block_t i;
3423
3424         for (i = 0; i < len; i++)
3425                 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3426 }
3427
3428 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3429 {
3430         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3431         struct curseg_info *seg_i;
3432         unsigned char *kaddr;
3433         struct page *page;
3434         block_t start;
3435         int i, j, offset;
3436
3437         start = start_sum_block(sbi);
3438
3439         page = f2fs_get_meta_page(sbi, start++);
3440         if (IS_ERR(page))
3441                 return PTR_ERR(page);
3442         kaddr = (unsigned char *)page_address(page);
3443
3444         /* Step 1: restore nat cache */
3445         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3446         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3447
3448         /* Step 2: restore sit cache */
3449         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3450         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3451         offset = 2 * SUM_JOURNAL_SIZE;
3452
3453         /* Step 3: restore summary entries */
3454         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3455                 unsigned short blk_off;
3456                 unsigned int segno;
3457
3458                 seg_i = CURSEG_I(sbi, i);
3459                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3460                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3461                 seg_i->next_segno = segno;
3462                 reset_curseg(sbi, i, 0);
3463                 seg_i->alloc_type = ckpt->alloc_type[i];
3464                 seg_i->next_blkoff = blk_off;
3465
3466                 if (seg_i->alloc_type == SSR)
3467                         blk_off = sbi->blocks_per_seg;
3468
3469                 for (j = 0; j < blk_off; j++) {
3470                         struct f2fs_summary *s;
3471                         s = (struct f2fs_summary *)(kaddr + offset);
3472                         seg_i->sum_blk->entries[j] = *s;
3473                         offset += SUMMARY_SIZE;
3474                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3475                                                 SUM_FOOTER_SIZE)
3476                                 continue;
3477
3478                         f2fs_put_page(page, 1);
3479                         page = NULL;
3480
3481                         page = f2fs_get_meta_page(sbi, start++);
3482                         if (IS_ERR(page))
3483                                 return PTR_ERR(page);
3484                         kaddr = (unsigned char *)page_address(page);
3485                         offset = 0;
3486                 }
3487         }
3488         f2fs_put_page(page, 1);
3489         return 0;
3490 }
3491
3492 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3493 {
3494         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3495         struct f2fs_summary_block *sum;
3496         struct curseg_info *curseg;
3497         struct page *new;
3498         unsigned short blk_off;
3499         unsigned int segno = 0;
3500         block_t blk_addr = 0;
3501         int err = 0;
3502
3503         /* get segment number and block addr */
3504         if (IS_DATASEG(type)) {
3505                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3506                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3507                                                         CURSEG_HOT_DATA]);
3508                 if (__exist_node_summaries(sbi))
3509                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3510                 else
3511                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3512         } else {
3513                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3514                                                         CURSEG_HOT_NODE]);
3515                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3516                                                         CURSEG_HOT_NODE]);
3517                 if (__exist_node_summaries(sbi))
3518                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3519                                                         type - CURSEG_HOT_NODE);
3520                 else
3521                         blk_addr = GET_SUM_BLOCK(sbi, segno);
3522         }
3523
3524         new = f2fs_get_meta_page(sbi, blk_addr);
3525         if (IS_ERR(new))
3526                 return PTR_ERR(new);
3527         sum = (struct f2fs_summary_block *)page_address(new);
3528
3529         if (IS_NODESEG(type)) {
3530                 if (__exist_node_summaries(sbi)) {
3531                         struct f2fs_summary *ns = &sum->entries[0];
3532                         int i;
3533                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3534                                 ns->version = 0;
3535                                 ns->ofs_in_node = 0;
3536                         }
3537                 } else {
3538                         err = f2fs_restore_node_summary(sbi, segno, sum);
3539                         if (err)
3540                                 goto out;
3541                 }
3542         }
3543
3544         /* set uncompleted segment to curseg */
3545         curseg = CURSEG_I(sbi, type);
3546         mutex_lock(&curseg->curseg_mutex);
3547
3548         /* update journal info */
3549         down_write(&curseg->journal_rwsem);
3550         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3551         up_write(&curseg->journal_rwsem);
3552
3553         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3554         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3555         curseg->next_segno = segno;
3556         reset_curseg(sbi, type, 0);
3557         curseg->alloc_type = ckpt->alloc_type[type];
3558         curseg->next_blkoff = blk_off;
3559         mutex_unlock(&curseg->curseg_mutex);
3560 out:
3561         f2fs_put_page(new, 1);
3562         return err;
3563 }
3564
3565 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3566 {
3567         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3568         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3569         int type = CURSEG_HOT_DATA;
3570         int err;
3571
3572         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3573                 int npages = f2fs_npages_for_summary_flush(sbi, true);
3574
3575                 if (npages >= 2)
3576                         f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3577                                                         META_CP, true);
3578
3579                 /* restore for compacted data summary */
3580                 err = read_compacted_summaries(sbi);
3581                 if (err)
3582                         return err;
3583                 type = CURSEG_HOT_NODE;
3584         }
3585
3586         if (__exist_node_summaries(sbi))
3587                 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3588                                         NR_CURSEG_TYPE - type, META_CP, true);
3589
3590         for (; type <= CURSEG_COLD_NODE; type++) {
3591                 err = read_normal_summaries(sbi, type);
3592                 if (err)
3593                         return err;
3594         }
3595
3596         /* sanity check for summary blocks */
3597         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3598                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3599                 f2fs_err(sbi, "invalid journal entries nats %u sits %u\n",
3600                          nats_in_cursum(nat_j), sits_in_cursum(sit_j));
3601                 return -EINVAL;
3602         }
3603
3604         return 0;
3605 }
3606
3607 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3608 {
3609         struct page *page;
3610         unsigned char *kaddr;
3611         struct f2fs_summary *summary;
3612         struct curseg_info *seg_i;
3613         int written_size = 0;
3614         int i, j;
3615
3616         page = f2fs_grab_meta_page(sbi, blkaddr++);
3617         kaddr = (unsigned char *)page_address(page);
3618         memset(kaddr, 0, PAGE_SIZE);
3619
3620         /* Step 1: write nat cache */
3621         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3622         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3623         written_size += SUM_JOURNAL_SIZE;
3624
3625         /* Step 2: write sit cache */
3626         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3627         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3628         written_size += SUM_JOURNAL_SIZE;
3629
3630         /* Step 3: write summary entries */
3631         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3632                 unsigned short blkoff;
3633                 seg_i = CURSEG_I(sbi, i);
3634                 if (sbi->ckpt->alloc_type[i] == SSR)
3635                         blkoff = sbi->blocks_per_seg;
3636                 else
3637                         blkoff = curseg_blkoff(sbi, i);
3638
3639                 for (j = 0; j < blkoff; j++) {
3640                         if (!page) {
3641                                 page = f2fs_grab_meta_page(sbi, blkaddr++);
3642                                 kaddr = (unsigned char *)page_address(page);
3643                                 memset(kaddr, 0, PAGE_SIZE);
3644                                 written_size = 0;
3645                         }
3646                         summary = (struct f2fs_summary *)(kaddr + written_size);
3647                         *summary = seg_i->sum_blk->entries[j];
3648                         written_size += SUMMARY_SIZE;
3649
3650                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3651                                                         SUM_FOOTER_SIZE)
3652                                 continue;
3653
3654                         set_page_dirty(page);
3655                         f2fs_put_page(page, 1);
3656                         page = NULL;
3657                 }
3658         }
3659         if (page) {
3660                 set_page_dirty(page);
3661                 f2fs_put_page(page, 1);
3662         }
3663 }
3664
3665 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3666                                         block_t blkaddr, int type)
3667 {
3668         int i, end;
3669         if (IS_DATASEG(type))
3670                 end = type + NR_CURSEG_DATA_TYPE;
3671         else
3672                 end = type + NR_CURSEG_NODE_TYPE;
3673
3674         for (i = type; i < end; i++)
3675                 write_current_sum_page(sbi, i, blkaddr + (i - type));
3676 }
3677
3678 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3679 {
3680         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3681                 write_compacted_summaries(sbi, start_blk);
3682         else
3683                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3684 }
3685
3686 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3687 {
3688         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3689 }
3690
3691 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3692                                         unsigned int val, int alloc)
3693 {
3694         int i;
3695
3696         if (type == NAT_JOURNAL) {
3697                 for (i = 0; i < nats_in_cursum(journal); i++) {
3698                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3699                                 return i;
3700                 }
3701                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3702                         return update_nats_in_cursum(journal, 1);
3703         } else if (type == SIT_JOURNAL) {
3704                 for (i = 0; i < sits_in_cursum(journal); i++)
3705                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3706                                 return i;
3707                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3708                         return update_sits_in_cursum(journal, 1);
3709         }
3710         return -1;
3711 }
3712
3713 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3714                                         unsigned int segno)
3715 {
3716         return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
3717 }
3718
3719 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3720                                         unsigned int start)
3721 {
3722         struct sit_info *sit_i = SIT_I(sbi);
3723         struct page *page;
3724         pgoff_t src_off, dst_off;
3725
3726         src_off = current_sit_addr(sbi, start);
3727         dst_off = next_sit_addr(sbi, src_off);
3728
3729         page = f2fs_grab_meta_page(sbi, dst_off);
3730         seg_info_to_sit_page(sbi, page, start);
3731
3732         set_page_dirty(page);
3733         set_to_next_sit(sit_i, start);
3734
3735         return page;
3736 }
3737
3738 static struct sit_entry_set *grab_sit_entry_set(void)
3739 {
3740         struct sit_entry_set *ses =
3741                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3742
3743         ses->entry_cnt = 0;
3744         INIT_LIST_HEAD(&ses->set_list);
3745         return ses;
3746 }
3747
3748 static void release_sit_entry_set(struct sit_entry_set *ses)
3749 {
3750         list_del(&ses->set_list);
3751         kmem_cache_free(sit_entry_set_slab, ses);
3752 }
3753
3754 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3755                                                 struct list_head *head)
3756 {
3757         struct sit_entry_set *next = ses;
3758
3759         if (list_is_last(&ses->set_list, head))
3760                 return;
3761
3762         list_for_each_entry_continue(next, head, set_list)
3763                 if (ses->entry_cnt <= next->entry_cnt)
3764                         break;
3765
3766         list_move_tail(&ses->set_list, &next->set_list);
3767 }
3768
3769 static void add_sit_entry(unsigned int segno, struct list_head *head)
3770 {
3771         struct sit_entry_set *ses;
3772         unsigned int start_segno = START_SEGNO(segno);
3773
3774         list_for_each_entry(ses, head, set_list) {
3775                 if (ses->start_segno == start_segno) {
3776                         ses->entry_cnt++;
3777                         adjust_sit_entry_set(ses, head);
3778                         return;
3779                 }
3780         }
3781
3782         ses = grab_sit_entry_set();
3783
3784         ses->start_segno = start_segno;
3785         ses->entry_cnt++;
3786         list_add(&ses->set_list, head);
3787 }
3788
3789 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3790 {
3791         struct f2fs_sm_info *sm_info = SM_I(sbi);
3792         struct list_head *set_list = &sm_info->sit_entry_set;
3793         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3794         unsigned int segno;
3795
3796         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3797                 add_sit_entry(segno, set_list);
3798 }
3799
3800 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3801 {
3802         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3803         struct f2fs_journal *journal = curseg->journal;
3804         int i;
3805
3806         down_write(&curseg->journal_rwsem);
3807         for (i = 0; i < sits_in_cursum(journal); i++) {
3808                 unsigned int segno;
3809                 bool dirtied;
3810
3811                 segno = le32_to_cpu(segno_in_journal(journal, i));
3812                 dirtied = __mark_sit_entry_dirty(sbi, segno);
3813
3814                 if (!dirtied)
3815                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3816         }
3817         update_sits_in_cursum(journal, -i);
3818         up_write(&curseg->journal_rwsem);
3819 }
3820
3821 /*
3822  * CP calls this function, which flushes SIT entries including sit_journal,
3823  * and moves prefree segs to free segs.
3824  */
3825 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3826 {
3827         struct sit_info *sit_i = SIT_I(sbi);
3828         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3829         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3830         struct f2fs_journal *journal = curseg->journal;
3831         struct sit_entry_set *ses, *tmp;
3832         struct list_head *head = &SM_I(sbi)->sit_entry_set;
3833         bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
3834         struct seg_entry *se;
3835
3836         down_write(&sit_i->sentry_lock);
3837
3838         if (!sit_i->dirty_sentries)
3839                 goto out;
3840
3841         /*
3842          * add and account sit entries of dirty bitmap in sit entry
3843          * set temporarily
3844          */
3845         add_sits_in_set(sbi);
3846
3847         /*
3848          * if there are no enough space in journal to store dirty sit
3849          * entries, remove all entries from journal and add and account
3850          * them in sit entry set.
3851          */
3852         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
3853                                                                 !to_journal)
3854                 remove_sits_in_journal(sbi);
3855
3856         /*
3857          * there are two steps to flush sit entries:
3858          * #1, flush sit entries to journal in current cold data summary block.
3859          * #2, flush sit entries to sit page.
3860          */
3861         list_for_each_entry_safe(ses, tmp, head, set_list) {
3862                 struct page *page = NULL;
3863                 struct f2fs_sit_block *raw_sit = NULL;
3864                 unsigned int start_segno = ses->start_segno;
3865                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3866                                                 (unsigned long)MAIN_SEGS(sbi));
3867                 unsigned int segno = start_segno;
3868
3869                 if (to_journal &&
3870                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3871                         to_journal = false;
3872
3873                 if (to_journal) {
3874                         down_write(&curseg->journal_rwsem);
3875                 } else {
3876                         page = get_next_sit_page(sbi, start_segno);
3877                         raw_sit = page_address(page);
3878                 }
3879
3880                 /* flush dirty sit entries in region of current sit set */
3881                 for_each_set_bit_from(segno, bitmap, end) {
3882                         int offset, sit_offset;
3883
3884                         se = get_seg_entry(sbi, segno);
3885 #ifdef CONFIG_F2FS_CHECK_FS
3886                         if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3887                                                 SIT_VBLOCK_MAP_SIZE))
3888                                 f2fs_bug_on(sbi, 1);
3889 #endif
3890
3891                         /* add discard candidates */
3892                         if (!(cpc->reason & CP_DISCARD)) {
3893                                 cpc->trim_start = segno;
3894                                 add_discard_addrs(sbi, cpc, false);
3895                         }
3896
3897                         if (to_journal) {
3898                                 offset = f2fs_lookup_journal_in_cursum(journal,
3899                                                         SIT_JOURNAL, segno, 1);
3900                                 f2fs_bug_on(sbi, offset < 0);
3901                                 segno_in_journal(journal, offset) =
3902                                                         cpu_to_le32(segno);
3903                                 seg_info_to_raw_sit(se,
3904                                         &sit_in_journal(journal, offset));
3905                                 check_block_count(sbi, segno,
3906                                         &sit_in_journal(journal, offset));
3907                         } else {
3908                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3909                                 seg_info_to_raw_sit(se,
3910                                                 &raw_sit->entries[sit_offset]);
3911                                 check_block_count(sbi, segno,
3912                                                 &raw_sit->entries[sit_offset]);
3913                         }
3914
3915                         __clear_bit(segno, bitmap);
3916                         sit_i->dirty_sentries--;
3917                         ses->entry_cnt--;
3918                 }
3919
3920                 if (to_journal)
3921                         up_write(&curseg->journal_rwsem);
3922                 else
3923                         f2fs_put_page(page, 1);
3924
3925                 f2fs_bug_on(sbi, ses->entry_cnt);
3926                 release_sit_entry_set(ses);
3927         }
3928
3929         f2fs_bug_on(sbi, !list_empty(head));
3930         f2fs_bug_on(sbi, sit_i->dirty_sentries);
3931 out:
3932         if (cpc->reason & CP_DISCARD) {
3933                 __u64 trim_start = cpc->trim_start;
3934
3935                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3936                         add_discard_addrs(sbi, cpc, false);
3937
3938                 cpc->trim_start = trim_start;
3939         }
3940         up_write(&sit_i->sentry_lock);
3941
3942         set_prefree_as_free_segments(sbi);
3943 }
3944
3945 static int build_sit_info(struct f2fs_sb_info *sbi)
3946 {
3947         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3948         struct sit_info *sit_i;
3949         unsigned int sit_segs, start;
3950         char *src_bitmap, *bitmap;
3951         unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
3952
3953         /* allocate memory for SIT information */
3954         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3955         if (!sit_i)
3956                 return -ENOMEM;
3957
3958         SM_I(sbi)->sit_info = sit_i;
3959
3960         sit_i->sentries =
3961                 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3962                                               MAIN_SEGS(sbi)),
3963                               GFP_KERNEL);
3964         if (!sit_i->sentries)
3965                 return -ENOMEM;
3966
3967         main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3968         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
3969                                                                 GFP_KERNEL);
3970         if (!sit_i->dirty_sentries_bitmap)
3971                 return -ENOMEM;
3972
3973 #ifdef CONFIG_F2FS_CHECK_FS
3974         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 4;
3975 #else
3976         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 3;
3977 #endif
3978         sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
3979         if (!sit_i->bitmap)
3980                 return -ENOMEM;
3981
3982         bitmap = sit_i->bitmap;
3983
3984         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3985                 sit_i->sentries[start].cur_valid_map = bitmap;
3986                 bitmap += SIT_VBLOCK_MAP_SIZE;
3987
3988                 sit_i->sentries[start].ckpt_valid_map = bitmap;
3989                 bitmap += SIT_VBLOCK_MAP_SIZE;
3990
3991 #ifdef CONFIG_F2FS_CHECK_FS
3992                 sit_i->sentries[start].cur_valid_map_mir = bitmap;
3993                 bitmap += SIT_VBLOCK_MAP_SIZE;
3994 #endif
3995
3996                 sit_i->sentries[start].discard_map = bitmap;
3997                 bitmap += SIT_VBLOCK_MAP_SIZE;
3998         }
3999
4000         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4001         if (!sit_i->tmp_map)
4002                 return -ENOMEM;
4003
4004         if (__is_large_section(sbi)) {
4005                 sit_i->sec_entries =
4006                         f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4007                                                       MAIN_SECS(sbi)),
4008                                       GFP_KERNEL);
4009                 if (!sit_i->sec_entries)
4010                         return -ENOMEM;
4011         }
4012
4013         /* get information related with SIT */
4014         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4015
4016         /* setup SIT bitmap from ckeckpoint pack */
4017         sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4018         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4019
4020         sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4021         if (!sit_i->sit_bitmap)
4022                 return -ENOMEM;
4023
4024 #ifdef CONFIG_F2FS_CHECK_FS
4025         sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4026                                         sit_bitmap_size, GFP_KERNEL);
4027         if (!sit_i->sit_bitmap_mir)
4028                 return -ENOMEM;
4029
4030         sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4031                                         main_bitmap_size, GFP_KERNEL);
4032         if (!sit_i->invalid_segmap)
4033                 return -ENOMEM;
4034 #endif
4035
4036         /* init SIT information */
4037         sit_i->s_ops = &default_salloc_ops;
4038
4039         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4040         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4041         sit_i->written_valid_blocks = 0;
4042         sit_i->bitmap_size = sit_bitmap_size;
4043         sit_i->dirty_sentries = 0;
4044         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4045         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4046         sit_i->mounted_time = ktime_get_real_seconds();
4047         init_rwsem(&sit_i->sentry_lock);
4048         return 0;
4049 }
4050
4051 static int build_free_segmap(struct f2fs_sb_info *sbi)
4052 {
4053         struct free_segmap_info *free_i;
4054         unsigned int bitmap_size, sec_bitmap_size;
4055
4056         /* allocate memory for free segmap information */
4057         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4058         if (!free_i)
4059                 return -ENOMEM;
4060
4061         SM_I(sbi)->free_info = free_i;
4062
4063         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4064         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4065         if (!free_i->free_segmap)
4066                 return -ENOMEM;
4067
4068         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4069         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4070         if (!free_i->free_secmap)
4071                 return -ENOMEM;
4072
4073         /* set all segments as dirty temporarily */
4074         memset(free_i->free_segmap, 0xff, bitmap_size);
4075         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4076
4077         /* init free segmap information */
4078         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4079         free_i->free_segments = 0;
4080         free_i->free_sections = 0;
4081         spin_lock_init(&free_i->segmap_lock);
4082         return 0;
4083 }
4084
4085 static int build_curseg(struct f2fs_sb_info *sbi)
4086 {
4087         struct curseg_info *array;
4088         int i;
4089
4090         array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
4091                              GFP_KERNEL);
4092         if (!array)
4093                 return -ENOMEM;
4094
4095         SM_I(sbi)->curseg_array = array;
4096
4097         for (i = 0; i < NR_CURSEG_TYPE; i++) {
4098                 mutex_init(&array[i].curseg_mutex);
4099                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4100                 if (!array[i].sum_blk)
4101                         return -ENOMEM;
4102                 init_rwsem(&array[i].journal_rwsem);
4103                 array[i].journal = f2fs_kzalloc(sbi,
4104                                 sizeof(struct f2fs_journal), GFP_KERNEL);
4105                 if (!array[i].journal)
4106                         return -ENOMEM;
4107                 array[i].segno = NULL_SEGNO;
4108                 array[i].next_blkoff = 0;
4109         }
4110         return restore_curseg_summaries(sbi);
4111 }
4112
4113 static int build_sit_entries(struct f2fs_sb_info *sbi)
4114 {
4115         struct sit_info *sit_i = SIT_I(sbi);
4116         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4117         struct f2fs_journal *journal = curseg->journal;
4118         struct seg_entry *se;
4119         struct f2fs_sit_entry sit;
4120         int sit_blk_cnt = SIT_BLK_CNT(sbi);
4121         unsigned int i, start, end;
4122         unsigned int readed, start_blk = 0;
4123         int err = 0;
4124         block_t total_node_blocks = 0;
4125
4126         do {
4127                 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
4128                                                         META_SIT, true);
4129
4130                 start = start_blk * sit_i->sents_per_block;
4131                 end = (start_blk + readed) * sit_i->sents_per_block;
4132
4133                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4134                         struct f2fs_sit_block *sit_blk;
4135                         struct page *page;
4136
4137                         se = &sit_i->sentries[start];
4138                         page = get_current_sit_page(sbi, start);
4139                         if (IS_ERR(page))
4140                                 return PTR_ERR(page);
4141                         sit_blk = (struct f2fs_sit_block *)page_address(page);
4142                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4143                         f2fs_put_page(page, 1);
4144
4145                         err = check_block_count(sbi, start, &sit);
4146                         if (err)
4147                                 return err;
4148                         seg_info_from_raw_sit(se, &sit);
4149                         if (IS_NODESEG(se->type))
4150                                 total_node_blocks += se->valid_blocks;
4151
4152                         /* build discard map only one time */
4153                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4154                                 memset(se->discard_map, 0xff,
4155                                         SIT_VBLOCK_MAP_SIZE);
4156                         } else {
4157                                 memcpy(se->discard_map,
4158                                         se->cur_valid_map,
4159                                         SIT_VBLOCK_MAP_SIZE);
4160                                 sbi->discard_blks +=
4161                                         sbi->blocks_per_seg -
4162                                         se->valid_blocks;
4163                         }
4164
4165                         if (__is_large_section(sbi))
4166                                 get_sec_entry(sbi, start)->valid_blocks +=
4167                                                         se->valid_blocks;
4168                 }
4169                 start_blk += readed;
4170         } while (start_blk < sit_blk_cnt);
4171
4172         down_read(&curseg->journal_rwsem);
4173         for (i = 0; i < sits_in_cursum(journal); i++) {
4174                 unsigned int old_valid_blocks;
4175
4176                 start = le32_to_cpu(segno_in_journal(journal, i));
4177                 if (start >= MAIN_SEGS(sbi)) {
4178                         f2fs_err(sbi, "Wrong journal entry on segno %u",
4179                                  start);
4180                         err = -EFSCORRUPTED;
4181                         break;
4182                 }
4183
4184                 se = &sit_i->sentries[start];
4185                 sit = sit_in_journal(journal, i);
4186
4187                 old_valid_blocks = se->valid_blocks;
4188                 if (IS_NODESEG(se->type))
4189                         total_node_blocks -= old_valid_blocks;
4190
4191                 err = check_block_count(sbi, start, &sit);
4192                 if (err)
4193                         break;
4194                 seg_info_from_raw_sit(se, &sit);
4195                 if (IS_NODESEG(se->type))
4196                         total_node_blocks += se->valid_blocks;
4197
4198                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4199                         memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4200                 } else {
4201                         memcpy(se->discard_map, se->cur_valid_map,
4202                                                 SIT_VBLOCK_MAP_SIZE);
4203                         sbi->discard_blks += old_valid_blocks;
4204                         sbi->discard_blks -= se->valid_blocks;
4205                 }
4206
4207                 if (__is_large_section(sbi)) {
4208                         get_sec_entry(sbi, start)->valid_blocks +=
4209                                                         se->valid_blocks;
4210                         get_sec_entry(sbi, start)->valid_blocks -=
4211                                                         old_valid_blocks;
4212                 }
4213         }
4214         up_read(&curseg->journal_rwsem);
4215
4216         if (!err && total_node_blocks != valid_node_count(sbi)) {
4217                 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4218                          total_node_blocks, valid_node_count(sbi));
4219                 err = -EFSCORRUPTED;
4220         }
4221
4222         return err;
4223 }
4224
4225 static void init_free_segmap(struct f2fs_sb_info *sbi)
4226 {
4227         unsigned int start;
4228         int type;
4229
4230         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4231                 struct seg_entry *sentry = get_seg_entry(sbi, start);
4232                 if (!sentry->valid_blocks)
4233                         __set_free(sbi, start);
4234                 else
4235                         SIT_I(sbi)->written_valid_blocks +=
4236                                                 sentry->valid_blocks;
4237         }
4238
4239         /* set use the current segments */
4240         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4241                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4242                 __set_test_and_inuse(sbi, curseg_t->segno);
4243         }
4244 }
4245
4246 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4247 {
4248         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4249         struct free_segmap_info *free_i = FREE_I(sbi);
4250         unsigned int segno = 0, offset = 0;
4251         unsigned short valid_blocks;
4252
4253         while (1) {
4254                 /* find dirty segment based on free segmap */
4255                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4256                 if (segno >= MAIN_SEGS(sbi))
4257                         break;
4258                 offset = segno + 1;
4259                 valid_blocks = get_valid_blocks(sbi, segno, false);
4260                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
4261                         continue;
4262                 if (valid_blocks > sbi->blocks_per_seg) {
4263                         f2fs_bug_on(sbi, 1);
4264                         continue;
4265                 }
4266                 mutex_lock(&dirty_i->seglist_lock);
4267                 __locate_dirty_segment(sbi, segno, DIRTY);
4268                 mutex_unlock(&dirty_i->seglist_lock);
4269         }
4270 }
4271
4272 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4273 {
4274         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4275         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4276
4277         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4278         if (!dirty_i->victim_secmap)
4279                 return -ENOMEM;
4280         return 0;
4281 }
4282
4283 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4284 {
4285         struct dirty_seglist_info *dirty_i;
4286         unsigned int bitmap_size, i;
4287
4288         /* allocate memory for dirty segments list information */
4289         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4290                                                                 GFP_KERNEL);
4291         if (!dirty_i)
4292                 return -ENOMEM;
4293
4294         SM_I(sbi)->dirty_info = dirty_i;
4295         mutex_init(&dirty_i->seglist_lock);
4296
4297         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4298
4299         for (i = 0; i < NR_DIRTY_TYPE; i++) {
4300                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4301                                                                 GFP_KERNEL);
4302                 if (!dirty_i->dirty_segmap[i])
4303                         return -ENOMEM;
4304         }
4305
4306         init_dirty_segmap(sbi);
4307         return init_victim_secmap(sbi);
4308 }
4309
4310 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4311 {
4312         int i;
4313
4314         /*
4315          * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4316          * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4317          */
4318         for (i = 0; i < NO_CHECK_TYPE; i++) {
4319                 struct curseg_info *curseg = CURSEG_I(sbi, i);
4320                 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4321                 unsigned int blkofs = curseg->next_blkoff;
4322
4323                 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4324                         goto out;
4325
4326                 if (curseg->alloc_type == SSR)
4327                         continue;
4328
4329                 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4330                         if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4331                                 continue;
4332 out:
4333                         f2fs_err(sbi,
4334                                  "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4335                                  i, curseg->segno, curseg->alloc_type,
4336                                  curseg->next_blkoff, blkofs);
4337                         return -EFSCORRUPTED;
4338                 }
4339         }
4340         return 0;
4341 }
4342
4343 /*
4344  * Update min, max modified time for cost-benefit GC algorithm
4345  */
4346 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4347 {
4348         struct sit_info *sit_i = SIT_I(sbi);
4349         unsigned int segno;
4350
4351         down_write(&sit_i->sentry_lock);
4352
4353         sit_i->min_mtime = ULLONG_MAX;
4354
4355         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4356                 unsigned int i;
4357                 unsigned long long mtime = 0;
4358
4359                 for (i = 0; i < sbi->segs_per_sec; i++)
4360                         mtime += get_seg_entry(sbi, segno + i)->mtime;
4361
4362                 mtime = div_u64(mtime, sbi->segs_per_sec);
4363
4364                 if (sit_i->min_mtime > mtime)
4365                         sit_i->min_mtime = mtime;
4366         }
4367         sit_i->max_mtime = get_mtime(sbi, false);
4368         up_write(&sit_i->sentry_lock);
4369 }
4370
4371 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
4372 {
4373         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4374         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4375         struct f2fs_sm_info *sm_info;
4376         int err;
4377
4378         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
4379         if (!sm_info)
4380                 return -ENOMEM;
4381
4382         /* init sm info */
4383         sbi->sm_info = sm_info;
4384         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4385         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4386         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4387         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4388         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4389         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4390         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
4391         sm_info->rec_prefree_segments = sm_info->main_segments *
4392                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
4393         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4394                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4395
4396         if (!test_opt(sbi, LFS))
4397                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
4398         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
4399         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
4400         sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
4401         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
4402         sm_info->min_ssr_sections = reserved_sections(sbi);
4403
4404         INIT_LIST_HEAD(&sm_info->sit_entry_set);
4405
4406         init_rwsem(&sm_info->curseg_lock);
4407
4408         if (!f2fs_readonly(sbi->sb)) {
4409                 err = f2fs_create_flush_cmd_control(sbi);
4410                 if (err)
4411                         return err;
4412         }
4413
4414         err = create_discard_cmd_control(sbi);
4415         if (err)
4416                 return err;
4417
4418         err = build_sit_info(sbi);
4419         if (err)
4420                 return err;
4421         err = build_free_segmap(sbi);
4422         if (err)
4423                 return err;
4424         err = build_curseg(sbi);
4425         if (err)
4426                 return err;
4427
4428         /* reinit free segmap based on SIT */
4429         err = build_sit_entries(sbi);
4430         if (err)
4431                 return err;
4432
4433         init_free_segmap(sbi);
4434         err = build_dirty_segmap(sbi);
4435         if (err)
4436                 return err;
4437
4438         err = sanity_check_curseg(sbi);
4439         if (err)
4440                 return err;
4441
4442         init_min_max_mtime(sbi);
4443         return 0;
4444 }
4445
4446 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4447                 enum dirty_type dirty_type)
4448 {
4449         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4450
4451         mutex_lock(&dirty_i->seglist_lock);
4452         kvfree(dirty_i->dirty_segmap[dirty_type]);
4453         dirty_i->nr_dirty[dirty_type] = 0;
4454         mutex_unlock(&dirty_i->seglist_lock);
4455 }
4456
4457 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
4458 {
4459         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4460         kvfree(dirty_i->victim_secmap);
4461 }
4462
4463 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4464 {
4465         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4466         int i;
4467
4468         if (!dirty_i)
4469                 return;
4470
4471         /* discard pre-free/dirty segments list */
4472         for (i = 0; i < NR_DIRTY_TYPE; i++)
4473                 discard_dirty_segmap(sbi, i);
4474
4475         destroy_victim_secmap(sbi);
4476         SM_I(sbi)->dirty_info = NULL;
4477         kvfree(dirty_i);
4478 }
4479
4480 static void destroy_curseg(struct f2fs_sb_info *sbi)
4481 {
4482         struct curseg_info *array = SM_I(sbi)->curseg_array;
4483         int i;
4484
4485         if (!array)
4486                 return;
4487         SM_I(sbi)->curseg_array = NULL;
4488         for (i = 0; i < NR_CURSEG_TYPE; i++) {
4489                 kvfree(array[i].sum_blk);
4490                 kvfree(array[i].journal);
4491         }
4492         kvfree(array);
4493 }
4494
4495 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4496 {
4497         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4498         if (!free_i)
4499                 return;
4500         SM_I(sbi)->free_info = NULL;
4501         kvfree(free_i->free_segmap);
4502         kvfree(free_i->free_secmap);
4503         kvfree(free_i);
4504 }
4505
4506 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4507 {
4508         struct sit_info *sit_i = SIT_I(sbi);
4509
4510         if (!sit_i)
4511                 return;
4512
4513         if (sit_i->sentries)
4514                 kvfree(sit_i->bitmap);
4515         kvfree(sit_i->tmp_map);
4516
4517         kvfree(sit_i->sentries);
4518         kvfree(sit_i->sec_entries);
4519         kvfree(sit_i->dirty_sentries_bitmap);
4520
4521         SM_I(sbi)->sit_info = NULL;
4522         kvfree(sit_i->sit_bitmap);
4523 #ifdef CONFIG_F2FS_CHECK_FS
4524         kvfree(sit_i->sit_bitmap_mir);
4525         kvfree(sit_i->invalid_segmap);
4526 #endif
4527         kvfree(sit_i);
4528 }
4529
4530 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
4531 {
4532         struct f2fs_sm_info *sm_info = SM_I(sbi);
4533
4534         if (!sm_info)
4535                 return;
4536         f2fs_destroy_flush_cmd_control(sbi, true);
4537         destroy_discard_cmd_control(sbi);
4538         destroy_dirty_segmap(sbi);
4539         destroy_curseg(sbi);
4540         destroy_free_segmap(sbi);
4541         destroy_sit_info(sbi);
4542         sbi->sm_info = NULL;
4543         kvfree(sm_info);
4544 }
4545
4546 int __init f2fs_create_segment_manager_caches(void)
4547 {
4548         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4549                         sizeof(struct discard_entry));
4550         if (!discard_entry_slab)
4551                 goto fail;
4552
4553         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4554                         sizeof(struct discard_cmd));
4555         if (!discard_cmd_slab)
4556                 goto destroy_discard_entry;
4557
4558         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4559                         sizeof(struct sit_entry_set));
4560         if (!sit_entry_set_slab)
4561                 goto destroy_discard_cmd;
4562
4563         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4564                         sizeof(struct inmem_pages));
4565         if (!inmem_entry_slab)
4566                 goto destroy_sit_entry_set;
4567         return 0;
4568
4569 destroy_sit_entry_set:
4570         kmem_cache_destroy(sit_entry_set_slab);
4571 destroy_discard_cmd:
4572         kmem_cache_destroy(discard_cmd_slab);
4573 destroy_discard_entry:
4574         kmem_cache_destroy(discard_entry_slab);
4575 fail:
4576         return -ENOMEM;
4577 }
4578
4579 void f2fs_destroy_segment_manager_caches(void)
4580 {
4581         kmem_cache_destroy(sit_entry_set_slab);
4582         kmem_cache_destroy(discard_cmd_slab);
4583         kmem_cache_destroy(discard_entry_slab);
4584         kmem_cache_destroy(inmem_entry_slab);
4585 }