fs/f2fs/segment.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/f2fs/segment.c
   4  *
   5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6  *             http://www.samsung.com/
   7  */
   8 #include <linux/fs.h>
   9 #include <linux/f2fs_fs.h>
  10 #include <linux/bio.h>
  11 #include <linux/blkdev.h>
  12 #include <linux/prefetch.h>
  13 #include <linux/kthread.h>
  14 #include <linux/swap.h>
  15 #include <linux/timer.h>
  16 #include <linux/freezer.h>
  17 #include <linux/sched/signal.h>
  18
  19 #include "f2fs.h"
  20 #include "segment.h"
  21 #include "node.h"
  22 #include "gc.h"
  23 #include "trace.h"
  24 #include <trace/events/f2fs.h>
  25
  26 #define __reverse_ffz(x) __reverse_ffs(~(x))
  27
  28 static struct kmem_cache *discard_entry_slab;
  29 static struct kmem_cache *discard_cmd_slab;
  30 static struct kmem_cache *sit_entry_set_slab;
  31 static struct kmem_cache *inmem_entry_slab;
  32
  33 static unsigned long __reverse_ulong(unsigned char *str)
  34 {
  35         unsigned long tmp = 0;
  36         int shift = 24, idx = 0;
  37
  38 #if BITS_PER_LONG == 64
  39         shift = 56;
  40 #endif
  41         while (shift >= 0) {
  42                 tmp |= (unsigned long)str[idx++] << shift;
  43                 shift -= BITS_PER_BYTE;
  44         }
  45         return tmp;
  46 }
  47
  48 /*
  49  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  50  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  51  */
  52 static inline unsigned long __reverse_ffs(unsigned long word)
  53 {
  54         int num = 0;
  55
  56 #if BITS_PER_LONG == 64
  57         if ((word & 0xffffffff00000000UL) == 0)
  58                 num += 32;
  59         else
  60                 word >>= 32;
  61 #endif
  62         if ((word & 0xffff0000) == 0)
  63                 num += 16;
  64         else
  65                 word >>= 16;
  66
  67         if ((word & 0xff00) == 0)
  68                 num += 8;
  69         else
  70                 word >>= 8;
  71
  72         if ((word & 0xf0) == 0)
  73                 num += 4;
  74         else
  75                 word >>= 4;
  76
  77         if ((word & 0xc) == 0)
  78                 num += 2;
  79         else
  80                 word >>= 2;
  81
  82         if ((word & 0x2) == 0)
  83                 num += 1;
  84         return num;
  85 }
  86
  87 /*
  88  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  89  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  90  * @size must be integral times of unsigned long.
  91  * Example:
  92  *                             MSB <--> LSB
  93  *   f2fs_set_bit(0, bitmap) => 1000 0000
  94  *   f2fs_set_bit(7, bitmap) => 0000 0001
  95  */
  96 static unsigned long __find_rev_next_bit(const unsigned long *addr,
  97                         unsigned long size, unsigned long offset)
  98 {
  99         const unsigned long *p = addr + BIT_WORD(offset);
 100         unsigned long result = size;
 101         unsigned long tmp;
 102
 103         if (offset >= size)
 104                 return size;
 105
 106         size -= (offset & ~(BITS_PER_LONG - 1));
 107         offset %= BITS_PER_LONG;
 108
 109         while (1) {
 110                 if (*p == 0)
 111                         goto pass;
 112
 113                 tmp = __reverse_ulong((unsigned char *)p);
 114
 115                 tmp &= ~0UL >> offset;
 116                 if (size < BITS_PER_LONG)
 117                         tmp &= (~0UL << (BITS_PER_LONG - size));
 118                 if (tmp)
 119                         goto found;
 120 pass:
 121                 if (size <= BITS_PER_LONG)
 122                         break;
 123                 size -= BITS_PER_LONG;
 124                 offset = 0;
 125                 p++;
 126         }
 127         return result;
 128 found:
 129         return result - size + __reverse_ffs(tmp);
 130 }
 131
 132 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 133                         unsigned long size, unsigned long offset)
 134 {
 135         const unsigned long *p = addr + BIT_WORD(offset);
 136         unsigned long result = size;
 137         unsigned long tmp;
 138
 139         if (offset >= size)
 140                 return size;
 141
 142         size -= (offset & ~(BITS_PER_LONG - 1));
 143         offset %= BITS_PER_LONG;
 144
 145         while (1) {
 146                 if (*p == ~0UL)
 147                         goto pass;
 148
 149                 tmp = __reverse_ulong((unsigned char *)p);
 150
 151                 if (offset)
 152                         tmp |= ~0UL << (BITS_PER_LONG - offset);
 153                 if (size < BITS_PER_LONG)
 154                         tmp |= ~0UL >> size;
 155                 if (tmp != ~0UL)
 156                         goto found;
 157 pass:
 158                 if (size <= BITS_PER_LONG)
 159                         break;
 160                 size -= BITS_PER_LONG;
 161                 offset = 0;
 162                 p++;
 163         }
 164         return result;
 165 found:
 166         return result - size + __reverse_ffz(tmp);
 167 }
 168
 169 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
 170 {
 171         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 172         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 173         int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
 174
 175         if (test_opt(sbi, LFS))
 176                 return false;
 177         if (sbi->gc_mode == GC_URGENT)
 178                 return true;
 179         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
 180                 return true;
 181
 182         return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
 183                         SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
 184 }
 185
 186 void f2fs_register_inmem_page(struct inode *inode, struct page *page)
 187 {
 188         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 189         struct f2fs_inode_info *fi = F2FS_I(inode);
 190         struct inmem_pages *new;
 191
 192         f2fs_trace_pid(page);
 193
 194         f2fs_set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
 195
 196         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 197
 198         /* add atomic page indices to the list */
 199         new->page = page;
 200         INIT_LIST_HEAD(&new->list);
 201
 202         /* increase reference count with clean state */
 203         mutex_lock(&fi->inmem_lock);
 204         get_page(page);
 205         list_add_tail(&new->list, &fi->inmem_pages);
 206         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 207         if (list_empty(&fi->inmem_ilist))
 208                 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
 209         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 210         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 211         mutex_unlock(&fi->inmem_lock);
 212
 213         trace_f2fs_register_inmem_page(page, INMEM);
 214 }
 215
 216 static int __revoke_inmem_pages(struct inode *inode,
 217                                 struct list_head *head, bool drop, bool recover,
 218                                 bool trylock)
 219 {
 220         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 221         struct inmem_pages *cur, *tmp;
 222         int err = 0;
 223
 224         list_for_each_entry_safe(cur, tmp, head, list) {
 225                 struct page *page = cur->page;
 226
 227                 if (drop)
 228                         trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 229
 230                 if (trylock) {
 231                         /*
 232                          * to avoid deadlock in between page lock and
 233                          * inmem_lock.
 234                          */
 235                         if (!trylock_page(page))
 236                                 continue;
 237                 } else {
 238                         lock_page(page);
 239                 }
 240
 241                 f2fs_wait_on_page_writeback(page, DATA, true, true);
 242
 243                 if (recover) {
 244                         struct dnode_of_data dn;
 245                         struct node_info ni;
 246
 247                         trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 248 retry:
 249                         set_new_dnode(&dn, inode, NULL, NULL, 0);
 250                         err = f2fs_get_dnode_of_data(&dn, page->index,
 251                                                                 LOOKUP_NODE);
 252                         if (err) {
 253                                 if (err == -ENOMEM) {
 254                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 255                                         cond_resched();
 256                                         goto retry;
 257                                 }
 258                                 err = -EAGAIN;
 259                                 goto next;
 260                         }
 261
 262                         err = f2fs_get_node_info(sbi, dn.nid, &ni);
 263                         if (err) {
 264                                 f2fs_put_dnode(&dn);
 265                                 return err;
 266                         }
 267
 268                         if (cur->old_addr == NEW_ADDR) {
 269                                 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
 270                                 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
 271                         } else
 272                                 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 273                                         cur->old_addr, ni.version, true, true);
 274                         f2fs_put_dnode(&dn);
 275                 }
 276 next:
 277                 /* we don't need to invalidate this in the sccessful status */
 278                 if (drop || recover) {
 279                         ClearPageUptodate(page);
 280                         clear_cold_data(page);
 281                 }
 282                 f2fs_clear_page_private(page);
 283                 f2fs_put_page(page, 1);
 284
 285                 list_del(&cur->list);
 286                 kmem_cache_free(inmem_entry_slab, cur);
 287                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 288         }
 289         return err;
 290 }
 291
 292 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
 293 {
 294         struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
 295         struct inode *inode;
 296         struct f2fs_inode_info *fi;
 297 next:
 298         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 299         if (list_empty(head)) {
 300                 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 301                 return;
 302         }
 303         fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
 304         inode = igrab(&fi->vfs_inode);
 305         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 306
 307         if (inode) {
 308                 if (gc_failure) {
 309                         if (fi->i_gc_failures[GC_FAILURE_ATOMIC])
 310                                 goto drop;
 311                         goto skip;
 312                 }
 313 drop:
 314                 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
 315                 f2fs_drop_inmem_pages(inode);
 316                 iput(inode);
 317         }
 318 skip:
 319         congestion_wait(BLK_RW_ASYNC, HZ/50);
 320         cond_resched();
 321         goto next;
 322 }
 323
 324 void f2fs_drop_inmem_pages(struct inode *inode)
 325 {
 326         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 327         struct f2fs_inode_info *fi = F2FS_I(inode);
 328
 329         while (!list_empty(&fi->inmem_pages)) {
 330                 mutex_lock(&fi->inmem_lock);
 331                 __revoke_inmem_pages(inode, &fi->inmem_pages,
 332                                                 true, false, true);
 333
 334                 if (list_empty(&fi->inmem_pages)) {
 335                         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 336                         if (!list_empty(&fi->inmem_ilist))
 337                                 list_del_init(&fi->inmem_ilist);
 338                         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 339                 }
 340                 mutex_unlock(&fi->inmem_lock);
 341         }
 342
 343         clear_inode_flag(inode, FI_ATOMIC_FILE);
 344         fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
 345         stat_dec_atomic_write(inode);
 346 }
 347
 348 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
 349 {
 350         struct f2fs_inode_info *fi = F2FS_I(inode);
 351         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 352         struct list_head *head = &fi->inmem_pages;
 353         struct inmem_pages *cur = NULL;
 354
 355         f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
 356
 357         mutex_lock(&fi->inmem_lock);
 358         list_for_each_entry(cur, head, list) {
 359                 if (cur->page == page)
 360                         break;
 361         }
 362
 363         f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
 364         list_del(&cur->list);
 365         mutex_unlock(&fi->inmem_lock);
 366
 367         dec_page_count(sbi, F2FS_INMEM_PAGES);
 368         kmem_cache_free(inmem_entry_slab, cur);
 369
 370         ClearPageUptodate(page);
 371         f2fs_clear_page_private(page);
 372         f2fs_put_page(page, 0);
 373
 374         trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 375 }
 376
 377 static int __f2fs_commit_inmem_pages(struct inode *inode)
 378 {
 379         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 380         struct f2fs_inode_info *fi = F2FS_I(inode);
 381         struct inmem_pages *cur, *tmp;
 382         struct f2fs_io_info fio = {
 383                 .sbi = sbi,
 384                 .ino = inode->i_ino,
 385                 .type = DATA,
 386                 .op = REQ_OP_WRITE,
 387                 .op_flags = REQ_SYNC | REQ_PRIO,
 388                 .io_type = FS_DATA_IO,
 389         };
 390         struct list_head revoke_list;
 391         bool submit_bio = false;
 392         int err = 0;
 393
 394         INIT_LIST_HEAD(&revoke_list);
 395
 396         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 397                 struct page *page = cur->page;
 398
 399                 lock_page(page);
 400                 if (page->mapping == inode->i_mapping) {
 401                         trace_f2fs_commit_inmem_page(page, INMEM);
 402
 403                         f2fs_wait_on_page_writeback(page, DATA, true, true);
 404
 405                         set_page_dirty(page);
 406                         if (clear_page_dirty_for_io(page)) {
 407                                 inode_dec_dirty_pages(inode);
 408                                 f2fs_remove_dirty_inode(inode);
 409                         }
 410 retry:
 411                         fio.page = page;
 412                         fio.old_blkaddr = NULL_ADDR;
 413                         fio.encrypted_page = NULL;
 414                         fio.need_lock = LOCK_DONE;
 415                         err = f2fs_do_write_data_page(&fio);
 416                         if (err) {
 417                                 if (err == -ENOMEM) {
 418                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 419                                         cond_resched();
 420                                         goto retry;
 421                                 }
 422                                 unlock_page(page);
 423                                 break;
 424                         }
 425                         /* record old blkaddr for revoking */
 426                         cur->old_addr = fio.old_blkaddr;
 427                         submit_bio = true;
 428                 }
 429                 unlock_page(page);
 430                 list_move_tail(&cur->list, &revoke_list);
 431         }
 432
 433         if (submit_bio)
 434                 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
 435
 436         if (err) {
 437                 /*
 438                  * try to revoke all committed pages, but still we could fail
 439                  * due to no memory or other reason, if that happened, EAGAIN
 440                  * will be returned, which means in such case, transaction is
 441                  * already not integrity, caller should use journal to do the
 442                  * recovery or rewrite & commit last transaction. For other
 443                  * error number, revoking was done by filesystem itself.
 444                  */
 445                 err = __revoke_inmem_pages(inode, &revoke_list,
 446                                                 false, true, false);
 447
 448                 /* drop all uncommitted pages */
 449                 __revoke_inmem_pages(inode, &fi->inmem_pages,
 450                                                 true, false, false);
 451         } else {
 452                 __revoke_inmem_pages(inode, &revoke_list,
 453                                                 false, false, false);
 454         }
 455
 456         return err;
 457 }
 458
 459 int f2fs_commit_inmem_pages(struct inode *inode)
 460 {
 461         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 462         struct f2fs_inode_info *fi = F2FS_I(inode);
 463         int err;
 464
 465         f2fs_balance_fs(sbi, true);
 466
 467         down_write(&fi->i_gc_rwsem[WRITE]);
 468
 469         f2fs_lock_op(sbi);
 470         set_inode_flag(inode, FI_ATOMIC_COMMIT);
 471
 472         mutex_lock(&fi->inmem_lock);
 473         err = __f2fs_commit_inmem_pages(inode);
 474
 475         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 476         if (!list_empty(&fi->inmem_ilist))
 477                 list_del_init(&fi->inmem_ilist);
 478         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 479         mutex_unlock(&fi->inmem_lock);
 480
 481         clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 482
 483         f2fs_unlock_op(sbi);
 484         up_write(&fi->i_gc_rwsem[WRITE]);
 485
 486         return err;
 487 }
 488
 489 /*
 490  * This function balances dirty node and dentry pages.
 491  * In addition, it controls garbage collection.
 492  */
 493 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 494 {
 495         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 496                 f2fs_show_injection_info(FAULT_CHECKPOINT);
 497                 f2fs_stop_checkpoint(sbi, false);
 498         }
 499
 500         /* balance_fs_bg is able to be pending */
 501         if (need && excess_cached_nats(sbi))
 502                 f2fs_balance_fs_bg(sbi);
 503
 504         if (f2fs_is_checkpoint_ready(sbi))
 505                 return;
 506
 507         /*
 508          * We should do GC or end up with checkpoint, if there are so many dirty
 509          * dir/node pages without enough free segments.
 510          */
 511         if (has_not_enough_free_secs(sbi, 0, 0)) {
 512                 mutex_lock(&sbi->gc_mutex);
 513                 f2fs_gc(sbi, false, false, NULL_SEGNO);
 514         }
 515 }
 516
 517 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 518 {
 519         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 520                 return;
 521
 522         /* try to shrink extent cache when there is no enough memory */
 523         if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
 524                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 525
 526         /* check the # of cached NAT entries */
 527         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
 528                 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 529
 530         if (!f2fs_available_free_memory(sbi, FREE_NIDS))
 531                 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
 532         else
 533                 f2fs_build_free_nids(sbi, false, false);
 534
 535         if (!is_idle(sbi, REQ_TIME) &&
 536                 (!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
 537                 return;
 538
 539         /* checkpoint is the only way to shrink partial cached entries */
 540         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
 541                         !f2fs_available_free_memory(sbi, INO_ENTRIES) ||
 542                         excess_prefree_segs(sbi) ||
 543                         excess_dirty_nats(sbi) ||
 544                         excess_dirty_nodes(sbi) ||
 545                         f2fs_time_over(sbi, CP_TIME)) {
 546                 if (test_opt(sbi, DATA_FLUSH)) {
 547                         struct blk_plug plug;
 548
 549                         mutex_lock(&sbi->flush_lock);
 550
 551                         blk_start_plug(&plug);
 552                         f2fs_sync_dirty_inodes(sbi, FILE_INODE);
 553                         blk_finish_plug(&plug);
 554
 555                         mutex_unlock(&sbi->flush_lock);
 556                 }
 557                 f2fs_sync_fs(sbi->sb, true);
 558                 stat_inc_bg_cp_count(sbi->stat_info);
 559         }
 560 }
 561
 562 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 563                                 struct block_device *bdev)
 564 {
 565         struct bio *bio;
 566         int ret;
 567
 568         bio = f2fs_bio_alloc(sbi, 0, false);
 569         if (!bio)
 570                 return -ENOMEM;
 571
 572         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
 573         bio_set_dev(bio, bdev);
 574         ret = submit_bio_wait(bio);
 575         bio_put(bio);
 576
 577         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 578                                 test_opt(sbi, FLUSH_MERGE), ret);
 579         return ret;
 580 }
 581
 582 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 583 {
 584         int ret = 0;
 585         int i;
 586
 587         if (!f2fs_is_multi_device(sbi))
 588                 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 589
 590         for (i = 0; i < sbi->s_ndevs; i++) {
 591                 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
 592                         continue;
 593                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 594                 if (ret)
 595                         break;
 596         }
 597         return ret;
 598 }
 599
 600 static int issue_flush_thread(void *data)
 601 {
 602         struct f2fs_sb_info *sbi = data;
 603         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 604         wait_queue_head_t *q = &fcc->flush_wait_queue;
 605 repeat:
 606         if (kthread_should_stop())
 607                 return 0;
 608
 609         sb_start_intwrite(sbi->sb);
 610
 611         if (!llist_empty(&fcc->issue_list)) {
 612                 struct flush_cmd *cmd, *next;
 613                 int ret;
 614
 615                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 616                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 617
 618                 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 619
 620                 ret = submit_flush_wait(sbi, cmd->ino);
 621                 atomic_inc(&fcc->issued_flush);
 622
 623                 llist_for_each_entry_safe(cmd, next,
 624                                           fcc->dispatch_list, llnode) {
 625                         cmd->ret = ret;
 626                         complete(&cmd->wait);
 627                 }
 628                 fcc->dispatch_list = NULL;
 629         }
 630
 631         sb_end_intwrite(sbi->sb);
 632
 633         wait_event_interruptible(*q,
 634                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 635         goto repeat;
 636 }
 637
 638 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 639 {
 640         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 641         struct flush_cmd cmd;
 642         int ret;
 643
 644         if (test_opt(sbi, NOBARRIER))
 645                 return 0;
 646
 647         if (!test_opt(sbi, FLUSH_MERGE)) {
 648                 atomic_inc(&fcc->queued_flush);
 649                 ret = submit_flush_wait(sbi, ino);
 650                 atomic_dec(&fcc->queued_flush);
 651                 atomic_inc(&fcc->issued_flush);
 652                 return ret;
 653         }
 654
 655         if (atomic_inc_return(&fcc->queued_flush) == 1 ||
 656             f2fs_is_multi_device(sbi)) {
 657                 ret = submit_flush_wait(sbi, ino);
 658                 atomic_dec(&fcc->queued_flush);
 659
 660                 atomic_inc(&fcc->issued_flush);
 661                 return ret;
 662         }
 663
 664         cmd.ino = ino;
 665         init_completion(&cmd.wait);
 666
 667         llist_add(&cmd.llnode, &fcc->issue_list);
 668
 669         /* update issue_list before we wake up issue_flush thread */
 670         smp_mb();
 671
 672         if (waitqueue_active(&fcc->flush_wait_queue))
 673                 wake_up(&fcc->flush_wait_queue);
 674
 675         if (fcc->f2fs_issue_flush) {
 676                 wait_for_completion(&cmd.wait);
 677                 atomic_dec(&fcc->queued_flush);
 678         } else {
 679                 struct llist_node *list;
 680
 681                 list = llist_del_all(&fcc->issue_list);
 682                 if (!list) {
 683                         wait_for_completion(&cmd.wait);
 684                         atomic_dec(&fcc->queued_flush);
 685                 } else {
 686                         struct flush_cmd *tmp, *next;
 687
 688                         ret = submit_flush_wait(sbi, ino);
 689
 690                         llist_for_each_entry_safe(tmp, next, list, llnode) {
 691                                 if (tmp == &cmd) {
 692                                         cmd.ret = ret;
 693                                         atomic_dec(&fcc->queued_flush);
 694                                         continue;
 695                                 }
 696                                 tmp->ret = ret;
 697                                 complete(&tmp->wait);
 698                         }
 699                 }
 700         }
 701
 702         return cmd.ret;
 703 }
 704
 705 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
 706 {
 707         dev_t dev = sbi->sb->s_bdev->bd_dev;
 708         struct flush_cmd_control *fcc;
 709         int err = 0;
 710
 711         if (SM_I(sbi)->fcc_info) {
 712                 fcc = SM_I(sbi)->fcc_info;
 713                 if (fcc->f2fs_issue_flush)
 714                         return err;
 715                 goto init_thread;
 716         }
 717
 718         fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
 719         if (!fcc)
 720                 return -ENOMEM;
 721         atomic_set(&fcc->issued_flush, 0);
 722         atomic_set(&fcc->queued_flush, 0);
 723         init_waitqueue_head(&fcc->flush_wait_queue);
 724         init_llist_head(&fcc->issue_list);
 725         SM_I(sbi)->fcc_info = fcc;
 726         if (!test_opt(sbi, FLUSH_MERGE))
 727                 return err;
 728
 729 init_thread:
 730         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 731                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 732         if (IS_ERR(fcc->f2fs_issue_flush)) {
 733                 err = PTR_ERR(fcc->f2fs_issue_flush);
 734                 kvfree(fcc);
 735                 SM_I(sbi)->fcc_info = NULL;
 736                 return err;
 737         }
 738
 739         return err;
 740 }
 741
 742 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 743 {
 744         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 745
 746         if (fcc && fcc->f2fs_issue_flush) {
 747                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 748
 749                 fcc->f2fs_issue_flush = NULL;
 750                 kthread_stop(flush_thread);
 751         }
 752         if (free) {
 753                 kvfree(fcc);
 754                 SM_I(sbi)->fcc_info = NULL;
 755         }
 756 }
 757
 758 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
 759 {
 760         int ret = 0, i;
 761
 762         if (!f2fs_is_multi_device(sbi))
 763                 return 0;
 764
 765         for (i = 1; i < sbi->s_ndevs; i++) {
 766                 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
 767                         continue;
 768                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 769                 if (ret)
 770                         break;
 771
 772                 spin_lock(&sbi->dev_lock);
 773                 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
 774                 spin_unlock(&sbi->dev_lock);
 775         }
 776
 777         return ret;
 778 }
 779
 780 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 781                 enum dirty_type dirty_type)
 782 {
 783         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 784
 785         /* need not be added */
 786         if (IS_CURSEG(sbi, segno))
 787                 return;
 788
 789         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 790                 dirty_i->nr_dirty[dirty_type]++;
 791
 792         if (dirty_type == DIRTY) {
 793                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 794                 enum dirty_type t = sentry->type;
 795
 796                 if (unlikely(t >= DIRTY)) {
 797                         f2fs_bug_on(sbi, 1);
 798                         return;
 799                 }
 800                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 801                         dirty_i->nr_dirty[t]++;
 802         }
 803 }
 804
 805 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 806                 enum dirty_type dirty_type)
 807 {
 808         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 809
 810         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 811                 dirty_i->nr_dirty[dirty_type]--;
 812
 813         if (dirty_type == DIRTY) {
 814                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 815                 enum dirty_type t = sentry->type;
 816
 817                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 818                         dirty_i->nr_dirty[t]--;
 819
 820                 if (get_valid_blocks(sbi, segno, true) == 0)
 821                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 822                                                 dirty_i->victim_secmap);
 823         }
 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                 /* don't overwrite by SSR to keep node chain */
2167                 if (IS_NODESEG(se->type) &&
2168                                 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2169                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2170                                 se->ckpt_valid_blocks++;
2171                 }
2172         } else {
2173                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2174 #ifdef CONFIG_F2FS_CHECK_FS
2175                 mir_exist = f2fs_test_and_clear_bit(offset,
2176                                                 se->cur_valid_map_mir);
2177                 if (unlikely(exist != mir_exist)) {
2178                         f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2179                                  blkaddr, exist);
2180                         f2fs_bug_on(sbi, 1);
2181                 }
2182 #endif
2183                 if (unlikely(!exist)) {
2184                         f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2185                                  blkaddr);
2186                         f2fs_bug_on(sbi, 1);
2187                         se->valid_blocks++;
2188                         del = 0;
2189                 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2190                         /*
2191                          * If checkpoints are off, we must not reuse data that
2192                          * was used in the previous checkpoint. If it was used
2193                          * before, we must track that to know how much space we
2194                          * really have.
2195                          */
2196                         if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2197                                 spin_lock(&sbi->stat_lock);
2198                                 sbi->unusable_block_count++;
2199                                 spin_unlock(&sbi->stat_lock);
2200                         }
2201                 }
2202
2203                 if (f2fs_test_and_clear_bit(offset, se->discard_map))
2204                         sbi->discard_blks++;
2205         }
2206         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2207                 se->ckpt_valid_blocks += del;
2208
2209         __mark_sit_entry_dirty(sbi, segno);
2210
2211         /* update total number of valid blocks to be written in ckpt area */
2212         SIT_I(sbi)->written_valid_blocks += del;
2213
2214         if (__is_large_section(sbi))
2215                 get_sec_entry(sbi, segno)->valid_blocks += del;
2216 }
2217
2218 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2219 {
2220         unsigned int segno = GET_SEGNO(sbi, addr);
2221         struct sit_info *sit_i = SIT_I(sbi);
2222
2223         f2fs_bug_on(sbi, addr == NULL_ADDR);
2224         if (addr == NEW_ADDR)
2225                 return;
2226
2227         invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2228
2229         /* add it into sit main buffer */
2230         down_write(&sit_i->sentry_lock);
2231
2232         update_sit_entry(sbi, addr, -1);
2233
2234         /* add it into dirty seglist */
2235         locate_dirty_segment(sbi, segno);
2236
2237         up_write(&sit_i->sentry_lock);
2238 }
2239
2240 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2241 {
2242         struct sit_info *sit_i = SIT_I(sbi);
2243         unsigned int segno, offset;
2244         struct seg_entry *se;
2245         bool is_cp = false;
2246
2247         if (!__is_valid_data_blkaddr(blkaddr))
2248                 return true;
2249
2250         down_read(&sit_i->sentry_lock);
2251
2252         segno = GET_SEGNO(sbi, blkaddr);
2253         se = get_seg_entry(sbi, segno);
2254         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2255
2256         if (f2fs_test_bit(offset, se->ckpt_valid_map))
2257                 is_cp = true;
2258
2259         up_read(&sit_i->sentry_lock);
2260
2261         return is_cp;
2262 }
2263
2264 /*
2265  * This function should be resided under the curseg_mutex lock
2266  */
2267 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2268                                         struct f2fs_summary *sum)
2269 {
2270         struct curseg_info *curseg = CURSEG_I(sbi, type);
2271         void *addr = curseg->sum_blk;
2272         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2273         memcpy(addr, sum, sizeof(struct f2fs_summary));
2274 }
2275
2276 /*
2277  * Calculate the number of current summary pages for writing
2278  */
2279 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2280 {
2281         int valid_sum_count = 0;
2282         int i, sum_in_page;
2283
2284         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2285                 if (sbi->ckpt->alloc_type[i] == SSR)
2286                         valid_sum_count += sbi->blocks_per_seg;
2287                 else {
2288                         if (for_ra)
2289                                 valid_sum_count += le16_to_cpu(
2290                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2291                         else
2292                                 valid_sum_count += curseg_blkoff(sbi, i);
2293                 }
2294         }
2295
2296         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2297                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2298         if (valid_sum_count <= sum_in_page)
2299                 return 1;
2300         else if ((valid_sum_count - sum_in_page) <=
2301                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2302                 return 2;
2303         return 3;
2304 }
2305
2306 /*
2307  * Caller should put this summary page
2308  */
2309 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2310 {
2311         return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
2312 }
2313
2314 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2315                                         void *src, block_t blk_addr)
2316 {
2317         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2318
2319         memcpy(page_address(page), src, PAGE_SIZE);
2320         set_page_dirty(page);
2321         f2fs_put_page(page, 1);
2322 }
2323
2324 static void write_sum_page(struct f2fs_sb_info *sbi,
2325                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
2326 {
2327         f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2328 }
2329
2330 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2331                                                 int type, block_t blk_addr)
2332 {
2333         struct curseg_info *curseg = CURSEG_I(sbi, type);
2334         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2335         struct f2fs_summary_block *src = curseg->sum_blk;
2336         struct f2fs_summary_block *dst;
2337
2338         dst = (struct f2fs_summary_block *)page_address(page);
2339         memset(dst, 0, PAGE_SIZE);
2340
2341         mutex_lock(&curseg->curseg_mutex);
2342
2343         down_read(&curseg->journal_rwsem);
2344         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2345         up_read(&curseg->journal_rwsem);
2346
2347         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2348         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2349
2350         mutex_unlock(&curseg->curseg_mutex);
2351
2352         set_page_dirty(page);
2353         f2fs_put_page(page, 1);
2354 }
2355
2356 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2357 {
2358         struct curseg_info *curseg = CURSEG_I(sbi, type);
2359         unsigned int segno = curseg->segno + 1;
2360         struct free_segmap_info *free_i = FREE_I(sbi);
2361
2362         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2363                 return !test_bit(segno, free_i->free_segmap);
2364         return 0;
2365 }
2366
2367 /*
2368  * Find a new segment from the free segments bitmap to right order
2369  * This function should be returned with success, otherwise BUG
2370  */
2371 static void get_new_segment(struct f2fs_sb_info *sbi,
2372                         unsigned int *newseg, bool new_sec, int dir)
2373 {
2374         struct free_segmap_info *free_i = FREE_I(sbi);
2375         unsigned int segno, secno, zoneno;
2376         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2377         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2378         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2379         unsigned int left_start = hint;
2380         bool init = true;
2381         int go_left = 0;
2382         int i;
2383
2384         spin_lock(&free_i->segmap_lock);
2385
2386         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2387                 segno = find_next_zero_bit(free_i->free_segmap,
2388                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2389                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2390                         goto got_it;
2391         }
2392 find_other_zone:
2393         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2394         if (secno >= MAIN_SECS(sbi)) {
2395                 if (dir == ALLOC_RIGHT) {
2396                         secno = find_next_zero_bit(free_i->free_secmap,
2397                                                         MAIN_SECS(sbi), 0);
2398                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2399                 } else {
2400                         go_left = 1;
2401                         left_start = hint - 1;
2402                 }
2403         }
2404         if (go_left == 0)
2405                 goto skip_left;
2406
2407         while (test_bit(left_start, free_i->free_secmap)) {
2408                 if (left_start > 0) {
2409                         left_start--;
2410                         continue;
2411                 }
2412                 left_start = find_next_zero_bit(free_i->free_secmap,
2413                                                         MAIN_SECS(sbi), 0);
2414                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2415                 break;
2416         }
2417         secno = left_start;
2418 skip_left:
2419         segno = GET_SEG_FROM_SEC(sbi, secno);
2420         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2421
2422         /* give up on finding another zone */
2423         if (!init)
2424                 goto got_it;
2425         if (sbi->secs_per_zone == 1)
2426                 goto got_it;
2427         if (zoneno == old_zoneno)
2428                 goto got_it;
2429         if (dir == ALLOC_LEFT) {
2430                 if (!go_left && zoneno + 1 >= total_zones)
2431                         goto got_it;
2432                 if (go_left && zoneno == 0)
2433                         goto got_it;
2434         }
2435         for (i = 0; i < NR_CURSEG_TYPE; i++)
2436                 if (CURSEG_I(sbi, i)->zone == zoneno)
2437                         break;
2438
2439         if (i < NR_CURSEG_TYPE) {
2440                 /* zone is in user, try another */
2441                 if (go_left)
2442                         hint = zoneno * sbi->secs_per_zone - 1;
2443                 else if (zoneno + 1 >= total_zones)
2444                         hint = 0;
2445                 else
2446                         hint = (zoneno + 1) * sbi->secs_per_zone;
2447                 init = false;
2448                 goto find_other_zone;
2449         }
2450 got_it:
2451         /* set it as dirty segment in free segmap */
2452         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2453         __set_inuse(sbi, segno);
2454         *newseg = segno;
2455         spin_unlock(&free_i->segmap_lock);
2456 }
2457
2458 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2459 {
2460         struct curseg_info *curseg = CURSEG_I(sbi, type);
2461         struct summary_footer *sum_footer;
2462
2463         curseg->segno = curseg->next_segno;
2464         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2465         curseg->next_blkoff = 0;
2466         curseg->next_segno = NULL_SEGNO;
2467
2468         sum_footer = &(curseg->sum_blk->footer);
2469         memset(sum_footer, 0, sizeof(struct summary_footer));
2470         if (IS_DATASEG(type))
2471                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2472         if (IS_NODESEG(type))
2473                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2474         __set_sit_entry_type(sbi, type, curseg->segno, modified);
2475 }
2476
2477 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2478 {
2479         /* if segs_per_sec is large than 1, we need to keep original policy. */
2480         if (__is_large_section(sbi))
2481                 return CURSEG_I(sbi, type)->segno;
2482
2483         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2484                 return 0;
2485
2486         if (test_opt(sbi, NOHEAP) &&
2487                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2488                 return 0;
2489
2490         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2491                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2492
2493         /* find segments from 0 to reuse freed segments */
2494         if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2495                 return 0;
2496
2497         return CURSEG_I(sbi, type)->segno;
2498 }
2499
2500 /*
2501  * Allocate a current working segment.
2502  * This function always allocates a free segment in LFS manner.
2503  */
2504 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2505 {
2506         struct curseg_info *curseg = CURSEG_I(sbi, type);
2507         unsigned int segno = curseg->segno;
2508         int dir = ALLOC_LEFT;
2509
2510         write_sum_page(sbi, curseg->sum_blk,
2511                                 GET_SUM_BLOCK(sbi, segno));
2512         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2513                 dir = ALLOC_RIGHT;
2514
2515         if (test_opt(sbi, NOHEAP))
2516                 dir = ALLOC_RIGHT;
2517
2518         segno = __get_next_segno(sbi, type);
2519         get_new_segment(sbi, &segno, new_sec, dir);
2520         curseg->next_segno = segno;
2521         reset_curseg(sbi, type, 1);
2522         curseg->alloc_type = LFS;
2523 }
2524
2525 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2526                         struct curseg_info *seg, block_t start)
2527 {
2528         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2529         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2530         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2531         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2532         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2533         int i, pos;
2534
2535         for (i = 0; i < entries; i++)
2536                 target_map[i] = ckpt_map[i] | cur_map[i];
2537
2538         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2539
2540         seg->next_blkoff = pos;
2541 }
2542
2543 /*
2544  * If a segment is written by LFS manner, next block offset is just obtained
2545  * by increasing the current block offset. However, if a segment is written by
2546  * SSR manner, next block offset obtained by calling __next_free_blkoff
2547  */
2548 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2549                                 struct curseg_info *seg)
2550 {
2551         if (seg->alloc_type == SSR)
2552                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2553         else
2554                 seg->next_blkoff++;
2555 }
2556
2557 /*
2558  * This function always allocates a used segment(from dirty seglist) by SSR
2559  * manner, so it should recover the existing segment information of valid blocks
2560  */
2561 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2562 {
2563         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2564         struct curseg_info *curseg = CURSEG_I(sbi, type);
2565         unsigned int new_segno = curseg->next_segno;
2566         struct f2fs_summary_block *sum_node;
2567         struct page *sum_page;
2568
2569         write_sum_page(sbi, curseg->sum_blk,
2570                                 GET_SUM_BLOCK(sbi, curseg->segno));
2571         __set_test_and_inuse(sbi, new_segno);
2572
2573         mutex_lock(&dirty_i->seglist_lock);
2574         __remove_dirty_segment(sbi, new_segno, PRE);
2575         __remove_dirty_segment(sbi, new_segno, DIRTY);
2576         mutex_unlock(&dirty_i->seglist_lock);
2577
2578         reset_curseg(sbi, type, 1);
2579         curseg->alloc_type = SSR;
2580         __next_free_blkoff(sbi, curseg, 0);
2581
2582         sum_page = f2fs_get_sum_page(sbi, new_segno);
2583         f2fs_bug_on(sbi, IS_ERR(sum_page));
2584         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2585         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2586         f2fs_put_page(sum_page, 1);
2587 }
2588
2589 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2590 {
2591         struct curseg_info *curseg = CURSEG_I(sbi, type);
2592         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2593         unsigned segno = NULL_SEGNO;
2594         int i, cnt;
2595         bool reversed = false;
2596
2597         /* f2fs_need_SSR() already forces to do this */
2598         if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2599                 curseg->next_segno = segno;
2600                 return 1;
2601         }
2602
2603         /* For node segments, let's do SSR more intensively */
2604         if (IS_NODESEG(type)) {
2605                 if (type >= CURSEG_WARM_NODE) {
2606                         reversed = true;
2607                         i = CURSEG_COLD_NODE;
2608                 } else {
2609                         i = CURSEG_HOT_NODE;
2610                 }
2611                 cnt = NR_CURSEG_NODE_TYPE;
2612         } else {
2613                 if (type >= CURSEG_WARM_DATA) {
2614                         reversed = true;
2615                         i = CURSEG_COLD_DATA;
2616                 } else {
2617                         i = CURSEG_HOT_DATA;
2618                 }
2619                 cnt = NR_CURSEG_DATA_TYPE;
2620         }
2621
2622         for (; cnt-- > 0; reversed ? i-- : i++) {
2623                 if (i == type)
2624                         continue;
2625                 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2626                         curseg->next_segno = segno;
2627                         return 1;
2628                 }
2629         }
2630
2631         /* find valid_blocks=0 in dirty list */
2632         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2633                 segno = get_free_segment(sbi);
2634                 if (segno != NULL_SEGNO) {
2635                         curseg->next_segno = segno;
2636                         return 1;
2637                 }
2638         }
2639         return 0;
2640 }
2641
2642 /*
2643  * flush out current segment and replace it with new segment
2644  * This function should be returned with success, otherwise BUG
2645  */
2646 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2647                                                 int type, bool force)
2648 {
2649         struct curseg_info *curseg = CURSEG_I(sbi, type);
2650
2651         if (force)
2652                 new_curseg(sbi, type, true);
2653         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2654                                         type == CURSEG_WARM_NODE)
2655                 new_curseg(sbi, type, false);
2656         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2657                         likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2658                 new_curseg(sbi, type, false);
2659         else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2660                 change_curseg(sbi, type);
2661         else
2662                 new_curseg(sbi, type, false);
2663
2664         stat_inc_seg_type(sbi, curseg);
2665 }
2666
2667 void allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2668                                         unsigned int start, unsigned int end)
2669 {
2670         struct curseg_info *curseg = CURSEG_I(sbi, type);
2671         unsigned int segno;
2672
2673         down_read(&SM_I(sbi)->curseg_lock);
2674         mutex_lock(&curseg->curseg_mutex);
2675         down_write(&SIT_I(sbi)->sentry_lock);
2676
2677         segno = CURSEG_I(sbi, type)->segno;
2678         if (segno < start || segno > end)
2679                 goto unlock;
2680
2681         if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2682                 change_curseg(sbi, type);
2683         else
2684                 new_curseg(sbi, type, true);
2685
2686         stat_inc_seg_type(sbi, curseg);
2687
2688         locate_dirty_segment(sbi, segno);
2689 unlock:
2690         up_write(&SIT_I(sbi)->sentry_lock);
2691
2692         if (segno != curseg->segno)
2693                 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2694                             type, segno, curseg->segno);
2695
2696         mutex_unlock(&curseg->curseg_mutex);
2697         up_read(&SM_I(sbi)->curseg_lock);
2698 }
2699
2700 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2701 {
2702         struct curseg_info *curseg;
2703         unsigned int old_segno;
2704         int i;
2705
2706         down_write(&SIT_I(sbi)->sentry_lock);
2707
2708         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2709                 curseg = CURSEG_I(sbi, i);
2710                 old_segno = curseg->segno;
2711                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2712                 locate_dirty_segment(sbi, old_segno);
2713         }
2714
2715         up_write(&SIT_I(sbi)->sentry_lock);
2716 }
2717
2718 static const struct segment_allocation default_salloc_ops = {
2719         .allocate_segment = allocate_segment_by_default,
2720 };
2721
2722 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2723                                                 struct cp_control *cpc)
2724 {
2725         __u64 trim_start = cpc->trim_start;
2726         bool has_candidate = false;
2727
2728         down_write(&SIT_I(sbi)->sentry_lock);
2729         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2730                 if (add_discard_addrs(sbi, cpc, true)) {
2731                         has_candidate = true;
2732                         break;
2733                 }
2734         }
2735         up_write(&SIT_I(sbi)->sentry_lock);
2736
2737         cpc->trim_start = trim_start;
2738         return has_candidate;
2739 }
2740
2741 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2742                                         struct discard_policy *dpolicy,
2743                                         unsigned int start, unsigned int end)
2744 {
2745         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2746         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2747         struct rb_node **insert_p = NULL, *insert_parent = NULL;
2748         struct discard_cmd *dc;
2749         struct blk_plug plug;
2750         int issued;
2751         unsigned int trimmed = 0;
2752
2753 next:
2754         issued = 0;
2755
2756         mutex_lock(&dcc->cmd_lock);
2757         if (unlikely(dcc->rbtree_check))
2758                 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2759                                                                 &dcc->root));
2760
2761         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2762                                         NULL, start,
2763                                         (struct rb_entry **)&prev_dc,
2764                                         (struct rb_entry **)&next_dc,
2765                                         &insert_p, &insert_parent, true, NULL);
2766         if (!dc)
2767                 dc = next_dc;
2768
2769         blk_start_plug(&plug);
2770
2771         while (dc && dc->lstart <= end) {
2772                 struct rb_node *node;
2773                 int err = 0;
2774
2775                 if (dc->len < dpolicy->granularity)
2776                         goto skip;
2777
2778                 if (dc->state != D_PREP) {
2779                         list_move_tail(&dc->list, &dcc->fstrim_list);
2780                         goto skip;
2781                 }
2782
2783                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2784
2785                 if (issued >= dpolicy->max_requests) {
2786                         start = dc->lstart + dc->len;
2787
2788                         if (err)
2789                                 __remove_discard_cmd(sbi, dc);
2790
2791                         blk_finish_plug(&plug);
2792                         mutex_unlock(&dcc->cmd_lock);
2793                         trimmed += __wait_all_discard_cmd(sbi, NULL);
2794                         congestion_wait(BLK_RW_ASYNC, HZ/50);
2795                         goto next;
2796                 }
2797 skip:
2798                 node = rb_next(&dc->rb_node);
2799                 if (err)
2800                         __remove_discard_cmd(sbi, dc);
2801                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2802
2803                 if (fatal_signal_pending(current))
2804                         break;
2805         }
2806
2807         blk_finish_plug(&plug);
2808         mutex_unlock(&dcc->cmd_lock);
2809
2810         return trimmed;
2811 }
2812
2813 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2814 {
2815         __u64 start = F2FS_BYTES_TO_BLK(range->start);
2816         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2817         unsigned int start_segno, end_segno;
2818         block_t start_block, end_block;
2819         struct cp_control cpc;
2820         struct discard_policy dpolicy;
2821         unsigned long long trimmed = 0;
2822         int err = 0;
2823         bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
2824
2825         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2826                 return -EINVAL;
2827
2828         if (end < MAIN_BLKADDR(sbi))
2829                 goto out;
2830
2831         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2832                 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
2833                 return -EFSCORRUPTED;
2834         }
2835
2836         /* start/end segment number in main_area */
2837         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2838         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2839                                                 GET_SEGNO(sbi, end);
2840         if (need_align) {
2841                 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2842                 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2843         }
2844
2845         cpc.reason = CP_DISCARD;
2846         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2847         cpc.trim_start = start_segno;
2848         cpc.trim_end = end_segno;
2849
2850         if (sbi->discard_blks == 0)
2851                 goto out;
2852
2853         mutex_lock(&sbi->gc_mutex);
2854         err = f2fs_write_checkpoint(sbi, &cpc);
2855         mutex_unlock(&sbi->gc_mutex);
2856         if (err)
2857                 goto out;
2858
2859         /*
2860          * We filed discard candidates, but actually we don't need to wait for
2861          * all of them, since they'll be issued in idle time along with runtime
2862          * discard option. User configuration looks like using runtime discard
2863          * or periodic fstrim instead of it.
2864          */
2865         if (f2fs_realtime_discard_enable(sbi))
2866                 goto out;
2867
2868         start_block = START_BLOCK(sbi, start_segno);
2869         end_block = START_BLOCK(sbi, end_segno + 1);
2870
2871         __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2872         trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2873                                         start_block, end_block);
2874
2875         trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
2876                                         start_block, end_block);
2877 out:
2878         if (!err)
2879                 range->len = F2FS_BLK_TO_BYTES(trimmed);
2880         return err;
2881 }
2882
2883 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2884 {
2885         struct curseg_info *curseg = CURSEG_I(sbi, type);
2886         if (curseg->next_blkoff < sbi->blocks_per_seg)
2887                 return true;
2888         return false;
2889 }
2890
2891 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
2892 {
2893         switch (hint) {
2894         case WRITE_LIFE_SHORT:
2895                 return CURSEG_HOT_DATA;
2896         case WRITE_LIFE_EXTREME:
2897                 return CURSEG_COLD_DATA;
2898         default:
2899                 return CURSEG_WARM_DATA;
2900         }
2901 }
2902
2903 /* This returns write hints for each segment type. This hints will be
2904  * passed down to block layer. There are mapping tables which depend on
2905  * the mount option 'whint_mode'.
2906  *
2907  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2908  *
2909  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2910  *
2911  * User                  F2FS                     Block
2912  * ----                  ----                     -----
2913  *                       META                     WRITE_LIFE_NOT_SET
2914  *                       HOT_NODE                 "
2915  *                       WARM_NODE                "
2916  *                       COLD_NODE                "
2917  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2918  * extension list        "                        "
2919  *
2920  * -- buffered io
2921  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2922  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2923  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2924  * WRITE_LIFE_NONE       "                        "
2925  * WRITE_LIFE_MEDIUM     "                        "
2926  * WRITE_LIFE_LONG       "                        "
2927  *
2928  * -- direct io
2929  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2930  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2931  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2932  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2933  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2934  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2935  *
2936  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2937  *
2938  * User                  F2FS                     Block
2939  * ----                  ----                     -----
2940  *                       META                     WRITE_LIFE_MEDIUM;
2941  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
2942  *                       WARM_NODE                "
2943  *                       COLD_NODE                WRITE_LIFE_NONE
2944  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2945  * extension list        "                        "
2946  *
2947  * -- buffered io
2948  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2949  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2950  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
2951  * WRITE_LIFE_NONE       "                        "
2952  * WRITE_LIFE_MEDIUM     "                        "
2953  * WRITE_LIFE_LONG       "                        "
2954  *
2955  * -- direct io
2956  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2957  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2958  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2959  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2960  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2961  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2962  */
2963
2964 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2965                                 enum page_type type, enum temp_type temp)
2966 {
2967         if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2968                 if (type == DATA) {
2969                         if (temp == WARM)
2970                                 return WRITE_LIFE_NOT_SET;
2971                         else if (temp == HOT)
2972                                 return WRITE_LIFE_SHORT;
2973                         else if (temp == COLD)
2974                                 return WRITE_LIFE_EXTREME;
2975                 } else {
2976                         return WRITE_LIFE_NOT_SET;
2977                 }
2978         } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2979                 if (type == DATA) {
2980                         if (temp == WARM)
2981                                 return WRITE_LIFE_LONG;
2982                         else if (temp == HOT)
2983                                 return WRITE_LIFE_SHORT;
2984                         else if (temp == COLD)
2985                                 return WRITE_LIFE_EXTREME;
2986                 } else if (type == NODE) {
2987                         if (temp == WARM || temp == HOT)
2988                                 return WRITE_LIFE_NOT_SET;
2989                         else if (temp == COLD)
2990                                 return WRITE_LIFE_NONE;
2991                 } else if (type == META) {
2992                         return WRITE_LIFE_MEDIUM;
2993                 }
2994         }
2995         return WRITE_LIFE_NOT_SET;
2996 }
2997
2998 static int __get_segment_type_2(struct f2fs_io_info *fio)
2999 {
3000         if (fio->type == DATA)
3001                 return CURSEG_HOT_DATA;
3002         else
3003                 return CURSEG_HOT_NODE;
3004 }
3005
3006 static int __get_segment_type_4(struct f2fs_io_info *fio)
3007 {
3008         if (fio->type == DATA) {
3009                 struct inode *inode = fio->page->mapping->host;
3010
3011                 if (S_ISDIR(inode->i_mode))
3012                         return CURSEG_HOT_DATA;
3013                 else
3014                         return CURSEG_COLD_DATA;
3015         } else {
3016                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3017                         return CURSEG_WARM_NODE;
3018                 else
3019                         return CURSEG_COLD_NODE;
3020         }
3021 }
3022
3023 static int __get_segment_type_6(struct f2fs_io_info *fio)
3024 {
3025         if (fio->type == DATA) {
3026                 struct inode *inode = fio->page->mapping->host;
3027
3028                 if (is_cold_data(fio->page) || file_is_cold(inode))
3029                         return CURSEG_COLD_DATA;
3030                 if (file_is_hot(inode) ||
3031                                 is_inode_flag_set(inode, FI_HOT_DATA) ||
3032                                 f2fs_is_atomic_file(inode) ||
3033                                 f2fs_is_volatile_file(inode))
3034                         return CURSEG_HOT_DATA;
3035                 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3036         } else {
3037                 if (IS_DNODE(fio->page))
3038                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3039                                                 CURSEG_HOT_NODE;
3040                 return CURSEG_COLD_NODE;
3041         }
3042 }
3043
3044 static int __get_segment_type(struct f2fs_io_info *fio)
3045 {
3046         int type = 0;
3047
3048         switch (F2FS_OPTION(fio->sbi).active_logs) {
3049         case 2:
3050                 type = __get_segment_type_2(fio);
3051                 break;
3052         case 4:
3053                 type = __get_segment_type_4(fio);
3054                 break;
3055         case 6:
3056                 type = __get_segment_type_6(fio);
3057                 break;
3058         default:
3059                 f2fs_bug_on(fio->sbi, true);
3060         }
3061
3062         if (IS_HOT(type))
3063                 fio->temp = HOT;
3064         else if (IS_WARM(type))
3065                 fio->temp = WARM;
3066         else
3067                 fio->temp = COLD;
3068         return type;
3069 }
3070
3071 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3072                 block_t old_blkaddr, block_t *new_blkaddr,
3073                 struct f2fs_summary *sum, int type,
3074                 struct f2fs_io_info *fio, bool add_list)
3075 {
3076         struct sit_info *sit_i = SIT_I(sbi);
3077         struct curseg_info *curseg = CURSEG_I(sbi, type);
3078
3079         down_read(&SM_I(sbi)->curseg_lock);
3080
3081         mutex_lock(&curseg->curseg_mutex);
3082         down_write(&sit_i->sentry_lock);
3083
3084         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3085
3086         f2fs_wait_discard_bio(sbi, *new_blkaddr);
3087
3088         /*
3089          * __add_sum_entry should be resided under the curseg_mutex
3090          * because, this function updates a summary entry in the
3091          * current summary block.
3092          */
3093         __add_sum_entry(sbi, type, sum);
3094
3095         __refresh_next_blkoff(sbi, curseg);
3096
3097         stat_inc_block_count(sbi, curseg);
3098
3099         /*
3100          * SIT information should be updated before segment allocation,
3101          * since SSR needs latest valid block information.
3102          */
3103         update_sit_entry(sbi, *new_blkaddr, 1);
3104         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3105                 update_sit_entry(sbi, old_blkaddr, -1);
3106
3107         if (!__has_curseg_space(sbi, type))
3108                 sit_i->s_ops->allocate_segment(sbi, type, false);
3109
3110         /*
3111          * segment dirty status should be updated after segment allocation,
3112          * so we just need to update status only one time after previous
3113          * segment being closed.
3114          */
3115         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3116         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3117
3118         up_write(&sit_i->sentry_lock);
3119
3120         if (page && IS_NODESEG(type)) {
3121                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3122
3123                 f2fs_inode_chksum_set(sbi, page);
3124         }
3125
3126         if (add_list) {
3127                 struct f2fs_bio_info *io;
3128
3129                 INIT_LIST_HEAD(&fio->list);
3130                 fio->in_list = true;
3131                 fio->retry = false;
3132                 io = sbi->write_io[fio->type] + fio->temp;
3133                 spin_lock(&io->io_lock);
3134                 list_add_tail(&fio->list, &io->io_list);
3135                 spin_unlock(&io->io_lock);
3136         }
3137
3138         mutex_unlock(&curseg->curseg_mutex);
3139
3140         up_read(&SM_I(sbi)->curseg_lock);
3141 }
3142
3143 static void update_device_state(struct f2fs_io_info *fio)
3144 {
3145         struct f2fs_sb_info *sbi = fio->sbi;
3146         unsigned int devidx;
3147
3148         if (!f2fs_is_multi_device(sbi))
3149                 return;
3150
3151         devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3152
3153         /* update device state for fsync */
3154         f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
3155
3156         /* update device state for checkpoint */
3157         if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3158                 spin_lock(&sbi->dev_lock);
3159                 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3160                 spin_unlock(&sbi->dev_lock);
3161         }
3162 }
3163
3164 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3165 {
3166         int type = __get_segment_type(fio);
3167         bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
3168
3169         if (keep_order)
3170                 down_read(&fio->sbi->io_order_lock);
3171 reallocate:
3172         f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3173                         &fio->new_blkaddr, sum, type, fio, true);
3174         if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3175                 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3176                                         fio->old_blkaddr, fio->old_blkaddr);
3177
3178         /* writeout dirty page into bdev */
3179         f2fs_submit_page_write(fio);
3180         if (fio->retry) {
3181                 fio->old_blkaddr = fio->new_blkaddr;
3182                 goto reallocate;
3183         }
3184
3185         update_device_state(fio);
3186
3187         if (keep_order)
3188                 up_read(&fio->sbi->io_order_lock);
3189 }
3190
3191 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3192                                         enum iostat_type io_type)
3193 {
3194         struct f2fs_io_info fio = {
3195                 .sbi = sbi,
3196                 .type = META,
3197                 .temp = HOT,
3198                 .op = REQ_OP_WRITE,
3199                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3200                 .old_blkaddr = page->index,
3201                 .new_blkaddr = page->index,
3202                 .page = page,
3203                 .encrypted_page = NULL,
3204                 .in_list = false,
3205         };
3206
3207         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3208                 fio.op_flags &= ~REQ_META;
3209
3210         set_page_writeback(page);
3211         ClearPageError(page);
3212         f2fs_submit_page_write(&fio);
3213
3214         stat_inc_meta_count(sbi, page->index);
3215         f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3216 }
3217
3218 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3219 {
3220         struct f2fs_summary sum;
3221
3222         set_summary(&sum, nid, 0, 0);
3223         do_write_page(&sum, fio);
3224
3225         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3226 }
3227
3228 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3229                                         struct f2fs_io_info *fio)
3230 {
3231         struct f2fs_sb_info *sbi = fio->sbi;
3232         struct f2fs_summary sum;
3233
3234         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3235         set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3236         do_write_page(&sum, fio);
3237         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3238
3239         f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3240 }
3241
3242 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3243 {
3244         int err;
3245         struct f2fs_sb_info *sbi = fio->sbi;
3246         unsigned int segno;
3247
3248         fio->new_blkaddr = fio->old_blkaddr;
3249         /* i/o temperature is needed for passing down write hints */
3250         __get_segment_type(fio);
3251
3252         segno = GET_SEGNO(sbi, fio->new_blkaddr);
3253
3254         if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3255                 set_sbi_flag(sbi, SBI_NEED_FSCK);
3256                 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3257                           __func__, segno);
3258                 return -EFSCORRUPTED;
3259         }
3260
3261         stat_inc_inplace_blocks(fio->sbi);
3262
3263         if (fio->bio)
3264                 err = f2fs_merge_page_bio(fio);
3265         else
3266                 err = f2fs_submit_page_bio(fio);
3267         if (!err) {
3268                 update_device_state(fio);
3269                 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3270         }
3271
3272         return err;
3273 }
3274
3275 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3276                                                 unsigned int segno)
3277 {
3278         int i;
3279
3280         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3281                 if (CURSEG_I(sbi, i)->segno == segno)
3282                         break;
3283         }
3284         return i;
3285 }
3286
3287 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3288                                 block_t old_blkaddr, block_t new_blkaddr,
3289                                 bool recover_curseg, bool recover_newaddr)
3290 {
3291         struct sit_info *sit_i = SIT_I(sbi);
3292         struct curseg_info *curseg;
3293         unsigned int segno, old_cursegno;
3294         struct seg_entry *se;
3295         int type;
3296         unsigned short old_blkoff;
3297
3298         segno = GET_SEGNO(sbi, new_blkaddr);
3299         se = get_seg_entry(sbi, segno);
3300         type = se->type;
3301
3302         down_write(&SM_I(sbi)->curseg_lock);
3303
3304         if (!recover_curseg) {
3305                 /* for recovery flow */
3306                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3307                         if (old_blkaddr == NULL_ADDR)
3308                                 type = CURSEG_COLD_DATA;
3309                         else
3310                                 type = CURSEG_WARM_DATA;
3311                 }
3312         } else {
3313                 if (IS_CURSEG(sbi, segno)) {
3314                         /* se->type is volatile as SSR allocation */
3315                         type = __f2fs_get_curseg(sbi, segno);
3316                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3317                 } else {
3318                         type = CURSEG_WARM_DATA;
3319                 }
3320         }
3321
3322         f2fs_bug_on(sbi, !IS_DATASEG(type));
3323         curseg = CURSEG_I(sbi, type);
3324
3325         mutex_lock(&curseg->curseg_mutex);
3326         down_write(&sit_i->sentry_lock);
3327
3328         old_cursegno = curseg->segno;
3329         old_blkoff = curseg->next_blkoff;
3330
3331         /* change the current segment */
3332         if (segno != curseg->segno) {
3333                 curseg->next_segno = segno;
3334                 change_curseg(sbi, type);
3335         }
3336
3337         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3338         __add_sum_entry(sbi, type, sum);
3339
3340         if (!recover_curseg || recover_newaddr)
3341                 update_sit_entry(sbi, new_blkaddr, 1);
3342         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3343                 invalidate_mapping_pages(META_MAPPING(sbi),
3344                                         old_blkaddr, old_blkaddr);
3345                 update_sit_entry(sbi, old_blkaddr, -1);
3346         }
3347
3348         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3349         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3350
3351         locate_dirty_segment(sbi, old_cursegno);
3352
3353         if (recover_curseg) {
3354                 if (old_cursegno != curseg->segno) {
3355                         curseg->next_segno = old_cursegno;
3356                         change_curseg(sbi, type);
3357                 }
3358                 curseg->next_blkoff = old_blkoff;
3359         }
3360
3361         up_write(&sit_i->sentry_lock);
3362         mutex_unlock(&curseg->curseg_mutex);
3363         up_write(&SM_I(sbi)->curseg_lock);
3364 }
3365
3366 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3367                                 block_t old_addr, block_t new_addr,
3368                                 unsigned char version, bool recover_curseg,
3369                                 bool recover_newaddr)
3370 {
3371         struct f2fs_summary sum;
3372
3373         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3374
3375         f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3376                                         recover_curseg, recover_newaddr);
3377
3378         f2fs_update_data_blkaddr(dn, new_addr);
3379 }
3380
3381 void f2fs_wait_on_page_writeback(struct page *page,
3382                                 enum page_type type, bool ordered, bool locked)
3383 {
3384         if (PageWriteback(page)) {
3385                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3386
3387                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3388                 if (ordered) {
3389                         wait_on_page_writeback(page);
3390                         f2fs_bug_on(sbi, locked && PageWriteback(page));
3391                 } else {
3392                         wait_for_stable_page(page);
3393                 }
3394         }
3395 }
3396
3397 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3398 {
3399         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3400         struct page *cpage;
3401
3402         if (!f2fs_post_read_required(inode))
3403                 return;
3404
3405         if (!__is_valid_data_blkaddr(blkaddr))
3406                 return;
3407
3408         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3409         if (cpage) {
3410                 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3411                 f2fs_put_page(cpage, 1);
3412         }
3413 }
3414
3415 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3416                                                                 block_t len)
3417 {
3418         block_t i;
3419
3420         for (i = 0; i < len; i++)
3421                 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3422 }
3423
3424 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3425 {
3426         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3427         struct curseg_info *seg_i;
3428         unsigned char *kaddr;
3429         struct page *page;
3430         block_t start;
3431         int i, j, offset;
3432
3433         start = start_sum_block(sbi);
3434
3435         page = f2fs_get_meta_page(sbi, start++);
3436         if (IS_ERR(page))
3437                 return PTR_ERR(page);
3438         kaddr = (unsigned char *)page_address(page);
3439
3440         /* Step 1: restore nat cache */
3441         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3442         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3443
3444         /* Step 2: restore sit cache */
3445         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3446         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3447         offset = 2 * SUM_JOURNAL_SIZE;
3448
3449         /* Step 3: restore summary entries */
3450         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3451                 unsigned short blk_off;
3452                 unsigned int segno;
3453
3454                 seg_i = CURSEG_I(sbi, i);
3455                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3456                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3457                 if (blk_off > ENTRIES_IN_SUM) {
3458                         f2fs_bug_on(sbi, 1);
3459                         f2fs_put_page(page, 1);
3460                         return -EFAULT;
3461                 }
3462                 seg_i->next_segno = segno;
3463                 reset_curseg(sbi, i, 0);
3464                 seg_i->alloc_type = ckpt->alloc_type[i];
3465                 seg_i->next_blkoff = blk_off;
3466
3467                 if (seg_i->alloc_type == SSR)
3468                         blk_off = sbi->blocks_per_seg;
3469
3470                 for (j = 0; j < blk_off; j++) {
3471                         struct f2fs_summary *s;
3472                         s = (struct f2fs_summary *)(kaddr + offset);
3473                         seg_i->sum_blk->entries[j] = *s;
3474                         offset += SUMMARY_SIZE;
3475                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3476                                                 SUM_FOOTER_SIZE)
3477                                 continue;
3478
3479                         f2fs_put_page(page, 1);
3480                         page = NULL;
3481
3482                         page = f2fs_get_meta_page(sbi, start++);
3483                         if (IS_ERR(page))
3484                                 return PTR_ERR(page);
3485                         kaddr = (unsigned char *)page_address(page);
3486                         offset = 0;
3487                 }
3488         }
3489         f2fs_put_page(page, 1);
3490         return 0;
3491 }
3492
3493 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3494 {
3495         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3496         struct f2fs_summary_block *sum;
3497         struct curseg_info *curseg;
3498         struct page *new;
3499         unsigned short blk_off;
3500         unsigned int segno = 0;
3501         block_t blk_addr = 0;
3502         int err = 0;
3503
3504         /* get segment number and block addr */
3505         if (IS_DATASEG(type)) {
3506                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3507                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3508                                                         CURSEG_HOT_DATA]);
3509                 if (__exist_node_summaries(sbi))
3510                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3511                 else
3512                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3513         } else {
3514                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3515                                                         CURSEG_HOT_NODE]);
3516                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3517                                                         CURSEG_HOT_NODE]);
3518                 if (__exist_node_summaries(sbi))
3519                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3520                                                         type - CURSEG_HOT_NODE);
3521                 else
3522                         blk_addr = GET_SUM_BLOCK(sbi, segno);
3523         }
3524
3525         new = f2fs_get_meta_page(sbi, blk_addr);
3526         if (IS_ERR(new))
3527                 return PTR_ERR(new);
3528         sum = (struct f2fs_summary_block *)page_address(new);
3529
3530         if (IS_NODESEG(type)) {
3531                 if (__exist_node_summaries(sbi)) {
3532                         struct f2fs_summary *ns = &sum->entries[0];
3533                         int i;
3534                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3535                                 ns->version = 0;
3536                                 ns->ofs_in_node = 0;
3537                         }
3538                 } else {
3539                         err = f2fs_restore_node_summary(sbi, segno, sum);
3540                         if (err)
3541                                 goto out;
3542                 }
3543         }
3544
3545         /* set uncompleted segment to curseg */
3546         curseg = CURSEG_I(sbi, type);
3547         mutex_lock(&curseg->curseg_mutex);
3548
3549         /* update journal info */
3550         down_write(&curseg->journal_rwsem);
3551         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3552         up_write(&curseg->journal_rwsem);
3553
3554         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3555         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3556         curseg->next_segno = segno;
3557         reset_curseg(sbi, type, 0);
3558         curseg->alloc_type = ckpt->alloc_type[type];
3559         curseg->next_blkoff = blk_off;
3560         mutex_unlock(&curseg->curseg_mutex);
3561 out:
3562         f2fs_put_page(new, 1);
3563         return err;
3564 }
3565
3566 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3567 {
3568         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3569         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3570         int type = CURSEG_HOT_DATA;
3571         int err;
3572
3573         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3574                 int npages = f2fs_npages_for_summary_flush(sbi, true);
3575
3576                 if (npages >= 2)
3577                         f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3578                                                         META_CP, true);
3579
3580                 /* restore for compacted data summary */
3581                 err = read_compacted_summaries(sbi);
3582                 if (err)
3583                         return err;
3584                 type = CURSEG_HOT_NODE;
3585         }
3586
3587         if (__exist_node_summaries(sbi))
3588                 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3589                                         NR_CURSEG_TYPE - type, META_CP, true);
3590
3591         for (; type <= CURSEG_COLD_NODE; type++) {
3592                 err = read_normal_summaries(sbi, type);
3593                 if (err)
3594                         return err;
3595         }
3596
3597         /* sanity check for summary blocks */
3598         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3599                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3600                 f2fs_err(sbi, "invalid journal entries nats %u sits %u\n",
3601                          nats_in_cursum(nat_j), sits_in_cursum(sit_j));
3602                 return -EINVAL;
3603         }
3604
3605         return 0;
3606 }
3607
3608 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3609 {
3610         struct page *page;
3611         unsigned char *kaddr;
3612         struct f2fs_summary *summary;
3613         struct curseg_info *seg_i;
3614         int written_size = 0;
3615         int i, j;
3616
3617         page = f2fs_grab_meta_page(sbi, blkaddr++);
3618         kaddr = (unsigned char *)page_address(page);
3619         memset(kaddr, 0, PAGE_SIZE);
3620
3621         /* Step 1: write nat cache */
3622         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3623         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3624         written_size += SUM_JOURNAL_SIZE;
3625
3626         /* Step 2: write sit cache */
3627         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3628         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3629         written_size += SUM_JOURNAL_SIZE;
3630
3631         /* Step 3: write summary entries */
3632         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3633                 unsigned short blkoff;
3634                 seg_i = CURSEG_I(sbi, i);
3635                 if (sbi->ckpt->alloc_type[i] == SSR)
3636                         blkoff = sbi->blocks_per_seg;
3637                 else
3638                         blkoff = curseg_blkoff(sbi, i);
3639
3640                 for (j = 0; j < blkoff; j++) {
3641                         if (!page) {
3642                                 page = f2fs_grab_meta_page(sbi, blkaddr++);
3643                                 kaddr = (unsigned char *)page_address(page);
3644                                 memset(kaddr, 0, PAGE_SIZE);
3645                                 written_size = 0;
3646                         }
3647                         summary = (struct f2fs_summary *)(kaddr + written_size);
3648                         *summary = seg_i->sum_blk->entries[j];
3649                         written_size += SUMMARY_SIZE;
3650
3651                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3652                                                         SUM_FOOTER_SIZE)
3653                                 continue;
3654
3655                         set_page_dirty(page);
3656                         f2fs_put_page(page, 1);
3657                         page = NULL;
3658                 }
3659         }
3660         if (page) {
3661                 set_page_dirty(page);
3662                 f2fs_put_page(page, 1);
3663         }
3664 }
3665
3666 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3667                                         block_t blkaddr, int type)
3668 {
3669         int i, end;
3670         if (IS_DATASEG(type))
3671                 end = type + NR_CURSEG_DATA_TYPE;
3672         else
3673                 end = type + NR_CURSEG_NODE_TYPE;
3674
3675         for (i = type; i < end; i++)
3676                 write_current_sum_page(sbi, i, blkaddr + (i - type));
3677 }
3678
3679 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3680 {
3681         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3682                 write_compacted_summaries(sbi, start_blk);
3683         else
3684                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3685 }
3686
3687 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3688 {
3689         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3690 }
3691
3692 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3693                                         unsigned int val, int alloc)
3694 {
3695         int i;
3696
3697         if (type == NAT_JOURNAL) {
3698                 for (i = 0; i < nats_in_cursum(journal); i++) {
3699                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3700                                 return i;
3701                 }
3702                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3703                         return update_nats_in_cursum(journal, 1);
3704         } else if (type == SIT_JOURNAL) {
3705                 for (i = 0; i < sits_in_cursum(journal); i++)
3706                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3707                                 return i;
3708                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3709                         return update_sits_in_cursum(journal, 1);
3710         }
3711         return -1;
3712 }
3713
3714 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3715                                         unsigned int segno)
3716 {
3717         return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
3718 }
3719
3720 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3721                                         unsigned int start)
3722 {
3723         struct sit_info *sit_i = SIT_I(sbi);
3724         struct page *page;
3725         pgoff_t src_off, dst_off;
3726
3727         src_off = current_sit_addr(sbi, start);
3728         dst_off = next_sit_addr(sbi, src_off);
3729
3730         page = f2fs_grab_meta_page(sbi, dst_off);
3731         seg_info_to_sit_page(sbi, page, start);
3732
3733         set_page_dirty(page);
3734         set_to_next_sit(sit_i, start);
3735
3736         return page;
3737 }
3738
3739 static struct sit_entry_set *grab_sit_entry_set(void)
3740 {
3741         struct sit_entry_set *ses =
3742                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3743
3744         ses->entry_cnt = 0;
3745         INIT_LIST_HEAD(&ses->set_list);
3746         return ses;
3747 }
3748
3749 static void release_sit_entry_set(struct sit_entry_set *ses)
3750 {
3751         list_del(&ses->set_list);
3752         kmem_cache_free(sit_entry_set_slab, ses);
3753 }
3754
3755 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3756                                                 struct list_head *head)
3757 {
3758         struct sit_entry_set *next = ses;
3759
3760         if (list_is_last(&ses->set_list, head))
3761                 return;
3762
3763         list_for_each_entry_continue(next, head, set_list)
3764                 if (ses->entry_cnt <= next->entry_cnt)
3765                         break;
3766
3767         list_move_tail(&ses->set_list, &next->set_list);
3768 }
3769
3770 static void add_sit_entry(unsigned int segno, struct list_head *head)
3771 {
3772         struct sit_entry_set *ses;
3773         unsigned int start_segno = START_SEGNO(segno);
3774
3775         list_for_each_entry(ses, head, set_list) {
3776                 if (ses->start_segno == start_segno) {
3777                         ses->entry_cnt++;
3778                         adjust_sit_entry_set(ses, head);
3779                         return;
3780                 }
3781         }
3782
3783         ses = grab_sit_entry_set();
3784
3785         ses->start_segno = start_segno;
3786         ses->entry_cnt++;
3787         list_add(&ses->set_list, head);
3788 }
3789
3790 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3791 {
3792         struct f2fs_sm_info *sm_info = SM_I(sbi);
3793         struct list_head *set_list = &sm_info->sit_entry_set;
3794         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3795         unsigned int segno;
3796
3797         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3798                 add_sit_entry(segno, set_list);
3799 }
3800
3801 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3802 {
3803         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3804         struct f2fs_journal *journal = curseg->journal;
3805         int i;
3806
3807         down_write(&curseg->journal_rwsem);
3808         for (i = 0; i < sits_in_cursum(journal); i++) {
3809                 unsigned int segno;
3810                 bool dirtied;
3811
3812                 segno = le32_to_cpu(segno_in_journal(journal, i));
3813                 dirtied = __mark_sit_entry_dirty(sbi, segno);
3814
3815                 if (!dirtied)
3816                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3817         }
3818         update_sits_in_cursum(journal, -i);
3819         up_write(&curseg->journal_rwsem);
3820 }
3821
3822 /*
3823  * CP calls this function, which flushes SIT entries including sit_journal,
3824  * and moves prefree segs to free segs.
3825  */
3826 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3827 {
3828         struct sit_info *sit_i = SIT_I(sbi);
3829         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3830         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3831         struct f2fs_journal *journal = curseg->journal;
3832         struct sit_entry_set *ses, *tmp;
3833         struct list_head *head = &SM_I(sbi)->sit_entry_set;
3834         bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
3835         struct seg_entry *se;
3836
3837         down_write(&sit_i->sentry_lock);
3838
3839         if (!sit_i->dirty_sentries)
3840                 goto out;
3841
3842         /*
3843          * add and account sit entries of dirty bitmap in sit entry
3844          * set temporarily
3845          */
3846         add_sits_in_set(sbi);
3847
3848         /*
3849          * if there are no enough space in journal to store dirty sit
3850          * entries, remove all entries from journal and add and account
3851          * them in sit entry set.
3852          */
3853         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
3854                                                                 !to_journal)
3855                 remove_sits_in_journal(sbi);
3856
3857         /*
3858          * there are two steps to flush sit entries:
3859          * #1, flush sit entries to journal in current cold data summary block.
3860          * #2, flush sit entries to sit page.
3861          */
3862         list_for_each_entry_safe(ses, tmp, head, set_list) {
3863                 struct page *page = NULL;
3864                 struct f2fs_sit_block *raw_sit = NULL;
3865                 unsigned int start_segno = ses->start_segno;
3866                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3867                                                 (unsigned long)MAIN_SEGS(sbi));
3868                 unsigned int segno = start_segno;
3869
3870                 if (to_journal &&
3871                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3872                         to_journal = false;
3873
3874                 if (to_journal) {
3875                         down_write(&curseg->journal_rwsem);
3876                 } else {
3877                         page = get_next_sit_page(sbi, start_segno);
3878                         raw_sit = page_address(page);
3879                 }
3880
3881                 /* flush dirty sit entries in region of current sit set */
3882                 for_each_set_bit_from(segno, bitmap, end) {
3883                         int offset, sit_offset;
3884
3885                         se = get_seg_entry(sbi, segno);
3886 #ifdef CONFIG_F2FS_CHECK_FS
3887                         if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3888                                                 SIT_VBLOCK_MAP_SIZE))
3889                                 f2fs_bug_on(sbi, 1);
3890 #endif
3891
3892                         /* add discard candidates */
3893                         if (!(cpc->reason & CP_DISCARD)) {
3894                                 cpc->trim_start = segno;
3895                                 add_discard_addrs(sbi, cpc, false);
3896                         }
3897
3898                         if (to_journal) {
3899                                 offset = f2fs_lookup_journal_in_cursum(journal,
3900                                                         SIT_JOURNAL, segno, 1);
3901                                 f2fs_bug_on(sbi, offset < 0);
3902                                 segno_in_journal(journal, offset) =
3903                                                         cpu_to_le32(segno);
3904                                 seg_info_to_raw_sit(se,
3905                                         &sit_in_journal(journal, offset));
3906                                 check_block_count(sbi, segno,
3907                                         &sit_in_journal(journal, offset));
3908                         } else {
3909                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3910                                 seg_info_to_raw_sit(se,
3911                                                 &raw_sit->entries[sit_offset]);
3912                                 check_block_count(sbi, segno,
3913                                                 &raw_sit->entries[sit_offset]);
3914                         }
3915
3916                         __clear_bit(segno, bitmap);
3917                         sit_i->dirty_sentries--;
3918                         ses->entry_cnt--;
3919                 }
3920
3921                 if (to_journal)
3922                         up_write(&curseg->journal_rwsem);
3923                 else
3924                         f2fs_put_page(page, 1);
3925
3926                 f2fs_bug_on(sbi, ses->entry_cnt);
3927                 release_sit_entry_set(ses);
3928         }
3929
3930         f2fs_bug_on(sbi, !list_empty(head));
3931         f2fs_bug_on(sbi, sit_i->dirty_sentries);
3932 out:
3933         if (cpc->reason & CP_DISCARD) {
3934                 __u64 trim_start = cpc->trim_start;
3935
3936                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3937                         add_discard_addrs(sbi, cpc, false);
3938
3939                 cpc->trim_start = trim_start;
3940         }
3941         up_write(&sit_i->sentry_lock);
3942
3943         set_prefree_as_free_segments(sbi);
3944 }
3945
3946 static int build_sit_info(struct f2fs_sb_info *sbi)
3947 {
3948         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3949         struct sit_info *sit_i;
3950         unsigned int sit_segs, start;
3951         char *src_bitmap;
3952         unsigned int bitmap_size;
3953
3954         /* allocate memory for SIT information */
3955         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3956         if (!sit_i)
3957                 return -ENOMEM;
3958
3959         SM_I(sbi)->sit_info = sit_i;
3960
3961         sit_i->sentries =
3962                 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3963                                               MAIN_SEGS(sbi)),
3964                               GFP_KERNEL);
3965         if (!sit_i->sentries)
3966                 return -ENOMEM;
3967
3968         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3969         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3970                                                                 GFP_KERNEL);
3971         if (!sit_i->dirty_sentries_bitmap)
3972                 return -ENOMEM;
3973
3974         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3975                 sit_i->sentries[start].cur_valid_map
3976                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3977                 sit_i->sentries[start].ckpt_valid_map
3978                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3979                 if (!sit_i->sentries[start].cur_valid_map ||
3980                                 !sit_i->sentries[start].ckpt_valid_map)
3981                         return -ENOMEM;
3982
3983 #ifdef CONFIG_F2FS_CHECK_FS
3984                 sit_i->sentries[start].cur_valid_map_mir
3985                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3986                 if (!sit_i->sentries[start].cur_valid_map_mir)
3987                         return -ENOMEM;
3988 #endif
3989
3990                 sit_i->sentries[start].discard_map
3991                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3992                                                         GFP_KERNEL);
3993                 if (!sit_i->sentries[start].discard_map)
3994                         return -ENOMEM;
3995         }
3996
3997         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3998         if (!sit_i->tmp_map)
3999                 return -ENOMEM;
4000
4001         if (__is_large_section(sbi)) {
4002                 sit_i->sec_entries =
4003                         f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4004                                                       MAIN_SECS(sbi)),
4005                                       GFP_KERNEL);
4006                 if (!sit_i->sec_entries)
4007                         return -ENOMEM;
4008         }
4009
4010         /* get information related with SIT */
4011         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4012
4013         /* setup SIT bitmap from ckeckpoint pack */
4014         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4015         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4016
4017         sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
4018         if (!sit_i->sit_bitmap)
4019                 return -ENOMEM;
4020
4021 #ifdef CONFIG_F2FS_CHECK_FS
4022         sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
4023         if (!sit_i->sit_bitmap_mir)
4024                 return -ENOMEM;
4025 #endif
4026
4027         /* init SIT information */
4028         sit_i->s_ops = &default_salloc_ops;
4029
4030         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4031         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4032         sit_i->written_valid_blocks = 0;
4033         sit_i->bitmap_size = bitmap_size;
4034         sit_i->dirty_sentries = 0;
4035         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4036         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4037         sit_i->mounted_time = ktime_get_real_seconds();
4038         init_rwsem(&sit_i->sentry_lock);
4039         return 0;
4040 }
4041
4042 static int build_free_segmap(struct f2fs_sb_info *sbi)
4043 {
4044         struct free_segmap_info *free_i;
4045         unsigned int bitmap_size, sec_bitmap_size;
4046
4047         /* allocate memory for free segmap information */
4048         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4049         if (!free_i)
4050                 return -ENOMEM;
4051
4052         SM_I(sbi)->free_info = free_i;
4053
4054         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4055         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4056         if (!free_i->free_segmap)
4057                 return -ENOMEM;
4058
4059         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4060         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4061         if (!free_i->free_secmap)
4062                 return -ENOMEM;
4063
4064         /* set all segments as dirty temporarily */
4065         memset(free_i->free_segmap, 0xff, bitmap_size);
4066         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4067
4068         /* init free segmap information */
4069         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4070         free_i->free_segments = 0;
4071         free_i->free_sections = 0;
4072         spin_lock_init(&free_i->segmap_lock);
4073         return 0;
4074 }
4075
4076 static int build_curseg(struct f2fs_sb_info *sbi)
4077 {
4078         struct curseg_info *array;
4079         int i;
4080
4081         array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
4082                              GFP_KERNEL);
4083         if (!array)
4084                 return -ENOMEM;
4085
4086         SM_I(sbi)->curseg_array = array;
4087
4088         for (i = 0; i < NR_CURSEG_TYPE; i++) {
4089                 mutex_init(&array[i].curseg_mutex);
4090                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4091                 if (!array[i].sum_blk)
4092                         return -ENOMEM;
4093                 init_rwsem(&array[i].journal_rwsem);
4094                 array[i].journal = f2fs_kzalloc(sbi,
4095                                 sizeof(struct f2fs_journal), GFP_KERNEL);
4096                 if (!array[i].journal)
4097                         return -ENOMEM;
4098                 array[i].segno = NULL_SEGNO;
4099                 array[i].next_blkoff = 0;
4100         }
4101         return restore_curseg_summaries(sbi);
4102 }
4103
4104 static int build_sit_entries(struct f2fs_sb_info *sbi)
4105 {
4106         struct sit_info *sit_i = SIT_I(sbi);
4107         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4108         struct f2fs_journal *journal = curseg->journal;
4109         struct seg_entry *se;
4110         struct f2fs_sit_entry sit;
4111         int sit_blk_cnt = SIT_BLK_CNT(sbi);
4112         unsigned int i, start, end;
4113         unsigned int readed, start_blk = 0;
4114         int err = 0;
4115         block_t total_node_blocks = 0;
4116
4117         do {
4118                 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
4119                                                         META_SIT, true);
4120
4121                 start = start_blk * sit_i->sents_per_block;
4122                 end = (start_blk + readed) * sit_i->sents_per_block;
4123
4124                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4125                         struct f2fs_sit_block *sit_blk;
4126                         struct page *page;
4127
4128                         se = &sit_i->sentries[start];
4129                         page = get_current_sit_page(sbi, start);
4130                         if (IS_ERR(page))
4131                                 return PTR_ERR(page);
4132                         sit_blk = (struct f2fs_sit_block *)page_address(page);
4133                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4134                         f2fs_put_page(page, 1);
4135
4136                         err = check_block_count(sbi, start, &sit);
4137                         if (err)
4138                                 return err;
4139                         seg_info_from_raw_sit(se, &sit);
4140                         if (IS_NODESEG(se->type))
4141                                 total_node_blocks += se->valid_blocks;
4142
4143                         /* build discard map only one time */
4144                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4145                                 memset(se->discard_map, 0xff,
4146                                         SIT_VBLOCK_MAP_SIZE);
4147                         } else {
4148                                 memcpy(se->discard_map,
4149                                         se->cur_valid_map,
4150                                         SIT_VBLOCK_MAP_SIZE);
4151                                 sbi->discard_blks +=
4152                                         sbi->blocks_per_seg -
4153                                         se->valid_blocks;
4154                         }
4155
4156                         if (__is_large_section(sbi))
4157                                 get_sec_entry(sbi, start)->valid_blocks +=
4158                                                         se->valid_blocks;
4159                 }
4160                 start_blk += readed;
4161         } while (start_blk < sit_blk_cnt);
4162
4163         down_read(&curseg->journal_rwsem);
4164         for (i = 0; i < sits_in_cursum(journal); i++) {
4165                 unsigned int old_valid_blocks;
4166
4167                 start = le32_to_cpu(segno_in_journal(journal, i));
4168                 if (start >= MAIN_SEGS(sbi)) {
4169                         f2fs_err(sbi, "Wrong journal entry on segno %u",
4170                                  start);
4171                         err = -EFSCORRUPTED;
4172                         break;
4173                 }
4174
4175                 se = &sit_i->sentries[start];
4176                 sit = sit_in_journal(journal, i);
4177
4178                 old_valid_blocks = se->valid_blocks;
4179                 if (IS_NODESEG(se->type))
4180                         total_node_blocks -= old_valid_blocks;
4181
4182                 err = check_block_count(sbi, start, &sit);
4183                 if (err)
4184                         break;
4185                 seg_info_from_raw_sit(se, &sit);
4186                 if (IS_NODESEG(se->type))
4187                         total_node_blocks += se->valid_blocks;
4188
4189                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4190                         memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4191                 } else {
4192                         memcpy(se->discard_map, se->cur_valid_map,
4193                                                 SIT_VBLOCK_MAP_SIZE);
4194                         sbi->discard_blks += old_valid_blocks;
4195                         sbi->discard_blks -= se->valid_blocks;
4196                 }
4197
4198                 if (__is_large_section(sbi)) {
4199                         get_sec_entry(sbi, start)->valid_blocks +=
4200                                                         se->valid_blocks;
4201                         get_sec_entry(sbi, start)->valid_blocks -=
4202                                                         old_valid_blocks;
4203                 }
4204         }
4205         up_read(&curseg->journal_rwsem);
4206
4207         if (!err && total_node_blocks != valid_node_count(sbi)) {
4208                 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4209                          total_node_blocks, valid_node_count(sbi));
4210                 err = -EFSCORRUPTED;
4211         }
4212
4213         return err;
4214 }
4215
4216 static void init_free_segmap(struct f2fs_sb_info *sbi)
4217 {
4218         unsigned int start;
4219         int type;
4220
4221         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4222                 struct seg_entry *sentry = get_seg_entry(sbi, start);
4223                 if (!sentry->valid_blocks)
4224                         __set_free(sbi, start);
4225                 else
4226                         SIT_I(sbi)->written_valid_blocks +=
4227                                                 sentry->valid_blocks;
4228         }
4229
4230         /* set use the current segments */
4231         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4232                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4233                 __set_test_and_inuse(sbi, curseg_t->segno);
4234         }
4235 }
4236
4237 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4238 {
4239         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4240         struct free_segmap_info *free_i = FREE_I(sbi);
4241         unsigned int segno = 0, offset = 0;
4242         unsigned short valid_blocks;
4243
4244         while (1) {
4245                 /* find dirty segment based on free segmap */
4246                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4247                 if (segno >= MAIN_SEGS(sbi))
4248                         break;
4249                 offset = segno + 1;
4250                 valid_blocks = get_valid_blocks(sbi, segno, false);
4251                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
4252                         continue;
4253                 if (valid_blocks > sbi->blocks_per_seg) {
4254                         f2fs_bug_on(sbi, 1);
4255                         continue;
4256                 }
4257                 mutex_lock(&dirty_i->seglist_lock);
4258                 __locate_dirty_segment(sbi, segno, DIRTY);
4259                 mutex_unlock(&dirty_i->seglist_lock);
4260         }
4261 }
4262
4263 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4264 {
4265         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4266         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4267
4268         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4269         if (!dirty_i->victim_secmap)
4270                 return -ENOMEM;
4271         return 0;
4272 }
4273
4274 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4275 {
4276         struct dirty_seglist_info *dirty_i;
4277         unsigned int bitmap_size, i;
4278
4279         /* allocate memory for dirty segments list information */
4280         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4281                                                                 GFP_KERNEL);
4282         if (!dirty_i)
4283                 return -ENOMEM;
4284
4285         SM_I(sbi)->dirty_info = dirty_i;
4286         mutex_init(&dirty_i->seglist_lock);
4287
4288         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4289
4290         for (i = 0; i < NR_DIRTY_TYPE; i++) {
4291                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4292                                                                 GFP_KERNEL);
4293                 if (!dirty_i->dirty_segmap[i])
4294                         return -ENOMEM;
4295         }
4296
4297         init_dirty_segmap(sbi);
4298         return init_victim_secmap(sbi);
4299 }
4300
4301 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4302 {
4303         int i;
4304
4305         /*
4306          * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4307          * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4308          */
4309         for (i = 0; i < NO_CHECK_TYPE; i++) {
4310                 struct curseg_info *curseg = CURSEG_I(sbi, i);
4311                 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4312                 unsigned int blkofs = curseg->next_blkoff;
4313
4314                 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4315                         goto out;
4316
4317                 if (curseg->alloc_type == SSR)
4318                         continue;
4319
4320                 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4321                         if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4322                                 continue;
4323 out:
4324                         f2fs_err(sbi,
4325                                  "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4326                                  i, curseg->segno, curseg->alloc_type,
4327                                  curseg->next_blkoff, blkofs);
4328                         return -EFSCORRUPTED;
4329                 }
4330         }
4331         return 0;
4332 }
4333
4334 /*
4335  * Update min, max modified time for cost-benefit GC algorithm
4336  */
4337 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4338 {
4339         struct sit_info *sit_i = SIT_I(sbi);
4340         unsigned int segno;
4341
4342         down_write(&sit_i->sentry_lock);
4343
4344         sit_i->min_mtime = ULLONG_MAX;
4345
4346         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4347                 unsigned int i;
4348                 unsigned long long mtime = 0;
4349
4350                 for (i = 0; i < sbi->segs_per_sec; i++)
4351                         mtime += get_seg_entry(sbi, segno + i)->mtime;
4352
4353                 mtime = div_u64(mtime, sbi->segs_per_sec);
4354
4355                 if (sit_i->min_mtime > mtime)
4356                         sit_i->min_mtime = mtime;
4357         }
4358         sit_i->max_mtime = get_mtime(sbi, false);
4359         up_write(&sit_i->sentry_lock);
4360 }
4361
4362 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
4363 {
4364         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4365         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4366         struct f2fs_sm_info *sm_info;
4367         int err;
4368
4369         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
4370         if (!sm_info)
4371                 return -ENOMEM;
4372
4373         /* init sm info */
4374         sbi->sm_info = sm_info;
4375         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4376         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4377         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4378         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4379         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4380         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4381         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
4382         sm_info->rec_prefree_segments = sm_info->main_segments *
4383                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
4384         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4385                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4386
4387         if (!test_opt(sbi, LFS))
4388                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
4389         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
4390         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
4391         sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
4392         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
4393         sm_info->min_ssr_sections = reserved_sections(sbi);
4394
4395         INIT_LIST_HEAD(&sm_info->sit_entry_set);
4396
4397         init_rwsem(&sm_info->curseg_lock);
4398
4399         if (!f2fs_readonly(sbi->sb)) {
4400                 err = f2fs_create_flush_cmd_control(sbi);
4401                 if (err)
4402                         return err;
4403         }
4404
4405         err = create_discard_cmd_control(sbi);
4406         if (err)
4407                 return err;
4408
4409         err = build_sit_info(sbi);
4410         if (err)
4411                 return err;
4412         err = build_free_segmap(sbi);
4413         if (err)
4414                 return err;
4415         err = build_curseg(sbi);
4416         if (err)
4417                 return err;
4418
4419         /* reinit free segmap based on SIT */
4420         err = build_sit_entries(sbi);
4421         if (err)
4422                 return err;
4423
4424         init_free_segmap(sbi);
4425         err = build_dirty_segmap(sbi);
4426         if (err)
4427                 return err;
4428
4429         err = sanity_check_curseg(sbi);
4430         if (err)
4431                 return err;
4432
4433         init_min_max_mtime(sbi);
4434         return 0;
4435 }
4436
4437 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4438                 enum dirty_type dirty_type)
4439 {
4440         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4441
4442         mutex_lock(&dirty_i->seglist_lock);
4443         kvfree(dirty_i->dirty_segmap[dirty_type]);
4444         dirty_i->nr_dirty[dirty_type] = 0;
4445         mutex_unlock(&dirty_i->seglist_lock);
4446 }
4447
4448 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
4449 {
4450         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4451         kvfree(dirty_i->victim_secmap);
4452 }
4453
4454 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4455 {
4456         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4457         int i;
4458
4459         if (!dirty_i)
4460                 return;
4461
4462         /* discard pre-free/dirty segments list */
4463         for (i = 0; i < NR_DIRTY_TYPE; i++)
4464                 discard_dirty_segmap(sbi, i);
4465
4466         destroy_victim_secmap(sbi);
4467         SM_I(sbi)->dirty_info = NULL;
4468         kvfree(dirty_i);
4469 }
4470
4471 static void destroy_curseg(struct f2fs_sb_info *sbi)
4472 {
4473         struct curseg_info *array = SM_I(sbi)->curseg_array;
4474         int i;
4475
4476         if (!array)
4477                 return;
4478         SM_I(sbi)->curseg_array = NULL;
4479         for (i = 0; i < NR_CURSEG_TYPE; i++) {
4480                 kvfree(array[i].sum_blk);
4481                 kvfree(array[i].journal);
4482         }
4483         kvfree(array);
4484 }
4485
4486 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4487 {
4488         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4489         if (!free_i)
4490                 return;
4491         SM_I(sbi)->free_info = NULL;
4492         kvfree(free_i->free_segmap);
4493         kvfree(free_i->free_secmap);
4494         kvfree(free_i);
4495 }
4496
4497 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4498 {
4499         struct sit_info *sit_i = SIT_I(sbi);
4500         unsigned int start;
4501
4502         if (!sit_i)
4503                 return;
4504
4505         if (sit_i->sentries) {
4506                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4507                         kvfree(sit_i->sentries[start].cur_valid_map);
4508 #ifdef CONFIG_F2FS_CHECK_FS
4509                         kvfree(sit_i->sentries[start].cur_valid_map_mir);
4510 #endif
4511                         kvfree(sit_i->sentries[start].ckpt_valid_map);
4512                         kvfree(sit_i->sentries[start].discard_map);
4513                 }
4514         }
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 #endif
4526         kvfree(sit_i);
4527 }
4528
4529 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
4530 {
4531         struct f2fs_sm_info *sm_info = SM_I(sbi);
4532
4533         if (!sm_info)
4534                 return;
4535         f2fs_destroy_flush_cmd_control(sbi, true);
4536         destroy_discard_cmd_control(sbi);
4537         destroy_dirty_segmap(sbi);
4538         destroy_curseg(sbi);
4539         destroy_free_segmap(sbi);
4540         destroy_sit_info(sbi);
4541         sbi->sm_info = NULL;
4542         kvfree(sm_info);
4543 }
4544
4545 int __init f2fs_create_segment_manager_caches(void)
4546 {
4547         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4548                         sizeof(struct discard_entry));
4549         if (!discard_entry_slab)
4550                 goto fail;
4551
4552         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4553                         sizeof(struct discard_cmd));
4554         if (!discard_cmd_slab)
4555                 goto destroy_discard_entry;
4556
4557         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4558                         sizeof(struct sit_entry_set));
4559         if (!sit_entry_set_slab)
4560                 goto destroy_discard_cmd;
4561
4562         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4563                         sizeof(struct inmem_pages));
4564         if (!inmem_entry_slab)
4565                 goto destroy_sit_entry_set;
4566         return 0;
4567
4568 destroy_sit_entry_set:
4569         kmem_cache_destroy(sit_entry_set_slab);
4570 destroy_discard_cmd:
4571         kmem_cache_destroy(discard_cmd_slab);
4572 destroy_discard_entry:
4573         kmem_cache_destroy(discard_entry_slab);
4574 fail:
4575         return -ENOMEM;
4576 }
4577
4578 void f2fs_destroy_segment_manager_caches(void)
4579 {
4580         kmem_cache_destroy(sit_entry_set_slab);
4581         kmem_cache_destroy(discard_cmd_slab);
4582         kmem_cache_destroy(discard_entry_slab);
4583         kmem_cache_destroy(inmem_entry_slab);
4584 }