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 (f2fs_lfs_mode(sbi))
 176                 return false;
 177         if (sbi->gc_mode == GC_URGENT_HIGH)
 178                 return true;
 179         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
 180                 return true;
 181
 182         return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
 183                         SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
 184 }
 185
 186 void f2fs_register_inmem_page(struct inode *inode, struct page *page)
 187 {
 188         struct inmem_pages *new;
 189
 190         f2fs_trace_pid(page);
 191
 192         f2fs_set_page_private(page, ATOMIC_WRITTEN_PAGE);
 193
 194         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 195
 196         /* add atomic page indices to the list */
 197         new->page = page;
 198         INIT_LIST_HEAD(&new->list);
 199
 200         /* increase reference count with clean state */
 201         get_page(page);
 202         mutex_lock(&F2FS_I(inode)->inmem_lock);
 203         list_add_tail(&new->list, &F2FS_I(inode)->inmem_pages);
 204         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 205         mutex_unlock(&F2FS_I(inode)->inmem_lock);
 206
 207         trace_f2fs_register_inmem_page(page, INMEM);
 208 }
 209
 210 static int __revoke_inmem_pages(struct inode *inode,
 211                                 struct list_head *head, bool drop, bool recover,
 212                                 bool trylock)
 213 {
 214         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 215         struct inmem_pages *cur, *tmp;
 216         int err = 0;
 217
 218         list_for_each_entry_safe(cur, tmp, head, list) {
 219                 struct page *page = cur->page;
 220
 221                 if (drop)
 222                         trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 223
 224                 if (trylock) {
 225                         /*
 226                          * to avoid deadlock in between page lock and
 227                          * inmem_lock.
 228                          */
 229                         if (!trylock_page(page))
 230                                 continue;
 231                 } else {
 232                         lock_page(page);
 233                 }
 234
 235                 f2fs_wait_on_page_writeback(page, DATA, true, true);
 236
 237                 if (recover) {
 238                         struct dnode_of_data dn;
 239                         struct node_info ni;
 240
 241                         trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 242 retry:
 243                         set_new_dnode(&dn, inode, NULL, NULL, 0);
 244                         err = f2fs_get_dnode_of_data(&dn, page->index,
 245                                                                 LOOKUP_NODE);
 246                         if (err) {
 247                                 if (err == -ENOMEM) {
 248                                         congestion_wait(BLK_RW_ASYNC,
 249                                                         DEFAULT_IO_TIMEOUT);
 250                                         cond_resched();
 251                                         goto retry;
 252                                 }
 253                                 err = -EAGAIN;
 254                                 goto next;
 255                         }
 256
 257                         err = f2fs_get_node_info(sbi, dn.nid, &ni);
 258                         if (err) {
 259                                 f2fs_put_dnode(&dn);
 260                                 return err;
 261                         }
 262
 263                         if (cur->old_addr == NEW_ADDR) {
 264                                 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
 265                                 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
 266                         } else
 267                                 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 268                                         cur->old_addr, ni.version, true, true);
 269                         f2fs_put_dnode(&dn);
 270                 }
 271 next:
 272                 /* we don't need to invalidate this in the sccessful status */
 273                 if (drop || recover) {
 274                         ClearPageUptodate(page);
 275                         clear_cold_data(page);
 276                 }
 277                 f2fs_clear_page_private(page);
 278                 f2fs_put_page(page, 1);
 279
 280                 list_del(&cur->list);
 281                 kmem_cache_free(inmem_entry_slab, cur);
 282                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 283         }
 284         return err;
 285 }
 286
 287 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
 288 {
 289         struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
 290         struct inode *inode;
 291         struct f2fs_inode_info *fi;
 292         unsigned int count = sbi->atomic_files;
 293         unsigned int looped = 0;
 294 next:
 295         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 296         if (list_empty(head)) {
 297                 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 298                 return;
 299         }
 300         fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
 301         inode = igrab(&fi->vfs_inode);
 302         if (inode)
 303                 list_move_tail(&fi->inmem_ilist, head);
 304         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 305
 306         if (inode) {
 307                 if (gc_failure) {
 308                         if (!fi->i_gc_failures[GC_FAILURE_ATOMIC])
 309                                 goto skip;
 310                 }
 311                 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
 312                 f2fs_drop_inmem_pages(inode);
 313 skip:
 314                 iput(inode);
 315         }
 316         congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
 317         cond_resched();
 318         if (gc_failure) {
 319                 if (++looped >= count)
 320                         return;
 321         }
 322         goto next;
 323 }
 324
 325 void f2fs_drop_inmem_pages(struct inode *inode)
 326 {
 327         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 328         struct f2fs_inode_info *fi = F2FS_I(inode);
 329
 330         while (!list_empty(&fi->inmem_pages)) {
 331                 mutex_lock(&fi->inmem_lock);
 332                 __revoke_inmem_pages(inode, &fi->inmem_pages,
 333                                                 true, false, true);
 334                 mutex_unlock(&fi->inmem_lock);
 335         }
 336
 337         fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
 338
 339         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 340         if (!list_empty(&fi->inmem_ilist))
 341                 list_del_init(&fi->inmem_ilist);
 342         if (f2fs_is_atomic_file(inode)) {
 343                 clear_inode_flag(inode, FI_ATOMIC_FILE);
 344                 sbi->atomic_files--;
 345         }
 346         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 347 }
 348
 349 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
 350 {
 351         struct f2fs_inode_info *fi = F2FS_I(inode);
 352         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 353         struct list_head *head = &fi->inmem_pages;
 354         struct inmem_pages *cur = NULL;
 355
 356         f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
 357
 358         mutex_lock(&fi->inmem_lock);
 359         list_for_each_entry(cur, head, list) {
 360                 if (cur->page == page)
 361                         break;
 362         }
 363
 364         f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
 365         list_del(&cur->list);
 366         mutex_unlock(&fi->inmem_lock);
 367
 368         dec_page_count(sbi, F2FS_INMEM_PAGES);
 369         kmem_cache_free(inmem_entry_slab, cur);
 370
 371         ClearPageUptodate(page);
 372         f2fs_clear_page_private(page);
 373         f2fs_put_page(page, 0);
 374
 375         trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 376 }
 377
 378 static int __f2fs_commit_inmem_pages(struct inode *inode)
 379 {
 380         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 381         struct f2fs_inode_info *fi = F2FS_I(inode);
 382         struct inmem_pages *cur, *tmp;
 383         struct f2fs_io_info fio = {
 384                 .sbi = sbi,
 385                 .ino = inode->i_ino,
 386                 .type = DATA,
 387                 .op = REQ_OP_WRITE,
 388                 .op_flags = REQ_SYNC | REQ_PRIO,
 389                 .io_type = FS_DATA_IO,
 390         };
 391         struct list_head revoke_list;
 392         bool submit_bio = false;
 393         int err = 0;
 394
 395         INIT_LIST_HEAD(&revoke_list);
 396
 397         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 398                 struct page *page = cur->page;
 399
 400                 lock_page(page);
 401                 if (page->mapping == inode->i_mapping) {
 402                         trace_f2fs_commit_inmem_page(page, INMEM);
 403
 404                         f2fs_wait_on_page_writeback(page, DATA, true, true);
 405
 406                         set_page_dirty(page);
 407                         if (clear_page_dirty_for_io(page)) {
 408                                 inode_dec_dirty_pages(inode);
 409                                 f2fs_remove_dirty_inode(inode);
 410                         }
 411 retry:
 412                         fio.page = page;
 413                         fio.old_blkaddr = NULL_ADDR;
 414                         fio.encrypted_page = NULL;
 415                         fio.need_lock = LOCK_DONE;
 416                         err = f2fs_do_write_data_page(&fio);
 417                         if (err) {
 418                                 if (err == -ENOMEM) {
 419                                         congestion_wait(BLK_RW_ASYNC,
 420                                                         DEFAULT_IO_TIMEOUT);
 421                                         cond_resched();
 422                                         goto retry;
 423                                 }
 424                                 unlock_page(page);
 425                                 break;
 426                         }
 427                         /* record old blkaddr for revoking */
 428                         cur->old_addr = fio.old_blkaddr;
 429                         submit_bio = true;
 430                 }
 431                 unlock_page(page);
 432                 list_move_tail(&cur->list, &revoke_list);
 433         }
 434
 435         if (submit_bio)
 436                 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
 437
 438         if (err) {
 439                 /*
 440                  * try to revoke all committed pages, but still we could fail
 441                  * due to no memory or other reason, if that happened, EAGAIN
 442                  * will be returned, which means in such case, transaction is
 443                  * already not integrity, caller should use journal to do the
 444                  * recovery or rewrite & commit last transaction. For other
 445                  * error number, revoking was done by filesystem itself.
 446                  */
 447                 err = __revoke_inmem_pages(inode, &revoke_list,
 448                                                 false, true, false);
 449
 450                 /* drop all uncommitted pages */
 451                 __revoke_inmem_pages(inode, &fi->inmem_pages,
 452                                                 true, false, false);
 453         } else {
 454                 __revoke_inmem_pages(inode, &revoke_list,
 455                                                 false, false, false);
 456         }
 457
 458         return err;
 459 }
 460
 461 int f2fs_commit_inmem_pages(struct inode *inode)
 462 {
 463         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 464         struct f2fs_inode_info *fi = F2FS_I(inode);
 465         int err;
 466
 467         f2fs_balance_fs(sbi, true);
 468
 469         down_write(&fi->i_gc_rwsem[WRITE]);
 470
 471         f2fs_lock_op(sbi);
 472         set_inode_flag(inode, FI_ATOMIC_COMMIT);
 473
 474         mutex_lock(&fi->inmem_lock);
 475         err = __f2fs_commit_inmem_pages(inode);
 476         mutex_unlock(&fi->inmem_lock);
 477
 478         clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 479
 480         f2fs_unlock_op(sbi);
 481         up_write(&fi->i_gc_rwsem[WRITE]);
 482
 483         return err;
 484 }
 485
 486 /*
 487  * This function balances dirty node and dentry pages.
 488  * In addition, it controls garbage collection.
 489  */
 490 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 491 {
 492         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 493                 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
 494                 f2fs_stop_checkpoint(sbi, false);
 495         }
 496
 497         /* balance_fs_bg is able to be pending */
 498         if (need && excess_cached_nats(sbi))
 499                 f2fs_balance_fs_bg(sbi, false);
 500
 501         if (!f2fs_is_checkpoint_ready(sbi))
 502                 return;
 503
 504         /*
 505          * We should do GC or end up with checkpoint, if there are so many dirty
 506          * dir/node pages without enough free segments.
 507          */
 508         if (has_not_enough_free_secs(sbi, 0, 0)) {
 509                 down_write(&sbi->gc_lock);
 510                 f2fs_gc(sbi, false, false, NULL_SEGNO);
 511         }
 512 }
 513
 514 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
 515 {
 516         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 517                 return;
 518
 519         /* try to shrink extent cache when there is no enough memory */
 520         if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
 521                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 522
 523         /* check the # of cached NAT entries */
 524         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
 525                 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 526
 527         if (!f2fs_available_free_memory(sbi, FREE_NIDS))
 528                 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
 529         else
 530                 f2fs_build_free_nids(sbi, false, false);
 531
 532         if (!is_idle(sbi, REQ_TIME) &&
 533                 (!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
 534                 return;
 535
 536         /* checkpoint is the only way to shrink partial cached entries */
 537         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
 538                         !f2fs_available_free_memory(sbi, INO_ENTRIES) ||
 539                         excess_prefree_segs(sbi) ||
 540                         excess_dirty_nats(sbi) ||
 541                         excess_dirty_nodes(sbi) ||
 542                         f2fs_time_over(sbi, CP_TIME)) {
 543                 if (test_opt(sbi, DATA_FLUSH) && from_bg) {
 544                         struct blk_plug plug;
 545
 546                         mutex_lock(&sbi->flush_lock);
 547
 548                         blk_start_plug(&plug);
 549                         f2fs_sync_dirty_inodes(sbi, FILE_INODE);
 550                         blk_finish_plug(&plug);
 551
 552                         mutex_unlock(&sbi->flush_lock);
 553                 }
 554                 f2fs_sync_fs(sbi->sb, true);
 555                 stat_inc_bg_cp_count(sbi->stat_info);
 556         }
 557 }
 558
 559 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 560                                 struct block_device *bdev)
 561 {
 562         struct bio *bio;
 563         int ret;
 564
 565         bio = f2fs_bio_alloc(sbi, 0, false);
 566         if (!bio)
 567                 return -ENOMEM;
 568
 569         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
 570         bio_set_dev(bio, bdev);
 571         ret = submit_bio_wait(bio);
 572         bio_put(bio);
 573
 574         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 575                                 test_opt(sbi, FLUSH_MERGE), ret);
 576         return ret;
 577 }
 578
 579 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 580 {
 581         int ret = 0;
 582         int i;
 583
 584         if (!f2fs_is_multi_device(sbi))
 585                 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 586
 587         for (i = 0; i < sbi->s_ndevs; i++) {
 588                 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
 589                         continue;
 590                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 591                 if (ret)
 592                         break;
 593         }
 594         return ret;
 595 }
 596
 597 static int issue_flush_thread(void *data)
 598 {
 599         struct f2fs_sb_info *sbi = data;
 600         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 601         wait_queue_head_t *q = &fcc->flush_wait_queue;
 602 repeat:
 603         if (kthread_should_stop())
 604                 return 0;
 605
 606         sb_start_intwrite(sbi->sb);
 607
 608         if (!llist_empty(&fcc->issue_list)) {
 609                 struct flush_cmd *cmd, *next;
 610                 int ret;
 611
 612                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 613                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 614
 615                 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 616
 617                 ret = submit_flush_wait(sbi, cmd->ino);
 618                 atomic_inc(&fcc->issued_flush);
 619
 620                 llist_for_each_entry_safe(cmd, next,
 621                                           fcc->dispatch_list, llnode) {
 622                         cmd->ret = ret;
 623                         complete(&cmd->wait);
 624                 }
 625                 fcc->dispatch_list = NULL;
 626         }
 627
 628         sb_end_intwrite(sbi->sb);
 629
 630         wait_event_interruptible(*q,
 631                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 632         goto repeat;
 633 }
 634
 635 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 636 {
 637         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 638         struct flush_cmd cmd;
 639         int ret;
 640
 641         if (test_opt(sbi, NOBARRIER))
 642                 return 0;
 643
 644         if (!test_opt(sbi, FLUSH_MERGE)) {
 645                 atomic_inc(&fcc->queued_flush);
 646                 ret = submit_flush_wait(sbi, ino);
 647                 atomic_dec(&fcc->queued_flush);
 648                 atomic_inc(&fcc->issued_flush);
 649                 return ret;
 650         }
 651
 652         if (atomic_inc_return(&fcc->queued_flush) == 1 ||
 653             f2fs_is_multi_device(sbi)) {
 654                 ret = submit_flush_wait(sbi, ino);
 655                 atomic_dec(&fcc->queued_flush);
 656
 657                 atomic_inc(&fcc->issued_flush);
 658                 return ret;
 659         }
 660
 661         cmd.ino = ino;
 662         init_completion(&cmd.wait);
 663
 664         llist_add(&cmd.llnode, &fcc->issue_list);
 665
 666         /* update issue_list before we wake up issue_flush thread */
 667         smp_mb();
 668
 669         if (waitqueue_active(&fcc->flush_wait_queue))
 670                 wake_up(&fcc->flush_wait_queue);
 671
 672         if (fcc->f2fs_issue_flush) {
 673                 wait_for_completion(&cmd.wait);
 674                 atomic_dec(&fcc->queued_flush);
 675         } else {
 676                 struct llist_node *list;
 677
 678                 list = llist_del_all(&fcc->issue_list);
 679                 if (!list) {
 680                         wait_for_completion(&cmd.wait);
 681                         atomic_dec(&fcc->queued_flush);
 682                 } else {
 683                         struct flush_cmd *tmp, *next;
 684
 685                         ret = submit_flush_wait(sbi, ino);
 686
 687                         llist_for_each_entry_safe(tmp, next, list, llnode) {
 688                                 if (tmp == &cmd) {
 689                                         cmd.ret = ret;
 690                                         atomic_dec(&fcc->queued_flush);
 691                                         continue;
 692                                 }
 693                                 tmp->ret = ret;
 694                                 complete(&tmp->wait);
 695                         }
 696                 }
 697         }
 698
 699         return cmd.ret;
 700 }
 701
 702 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
 703 {
 704         dev_t dev = sbi->sb->s_bdev->bd_dev;
 705         struct flush_cmd_control *fcc;
 706         int err = 0;
 707
 708         if (SM_I(sbi)->fcc_info) {
 709                 fcc = SM_I(sbi)->fcc_info;
 710                 if (fcc->f2fs_issue_flush)
 711                         return err;
 712                 goto init_thread;
 713         }
 714
 715         fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
 716         if (!fcc)
 717                 return -ENOMEM;
 718         atomic_set(&fcc->issued_flush, 0);
 719         atomic_set(&fcc->queued_flush, 0);
 720         init_waitqueue_head(&fcc->flush_wait_queue);
 721         init_llist_head(&fcc->issue_list);
 722         SM_I(sbi)->fcc_info = fcc;
 723         if (!test_opt(sbi, FLUSH_MERGE))
 724                 return err;
 725
 726 init_thread:
 727         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 728                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 729         if (IS_ERR(fcc->f2fs_issue_flush)) {
 730                 err = PTR_ERR(fcc->f2fs_issue_flush);
 731                 kvfree(fcc);
 732                 SM_I(sbi)->fcc_info = NULL;
 733                 return err;
 734         }
 735
 736         return err;
 737 }
 738
 739 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 740 {
 741         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 742
 743         if (fcc && fcc->f2fs_issue_flush) {
 744                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 745
 746                 fcc->f2fs_issue_flush = NULL;
 747                 kthread_stop(flush_thread);
 748         }
 749         if (free) {
 750                 kvfree(fcc);
 751                 SM_I(sbi)->fcc_info = NULL;
 752         }
 753 }
 754
 755 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
 756 {
 757         int ret = 0, i;
 758
 759         if (!f2fs_is_multi_device(sbi))
 760                 return 0;
 761
 762         for (i = 1; i < sbi->s_ndevs; i++) {
 763                 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
 764                         continue;
 765                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 766                 if (ret)
 767                         break;
 768
 769                 spin_lock(&sbi->dev_lock);
 770                 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
 771                 spin_unlock(&sbi->dev_lock);
 772         }
 773
 774         return ret;
 775 }
 776
 777 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 778                 enum dirty_type dirty_type)
 779 {
 780         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 781
 782         /* need not be added */
 783         if (IS_CURSEG(sbi, segno))
 784                 return;
 785
 786         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 787                 dirty_i->nr_dirty[dirty_type]++;
 788
 789         if (dirty_type == DIRTY) {
 790                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 791                 enum dirty_type t = sentry->type;
 792
 793                 if (unlikely(t >= DIRTY)) {
 794                         f2fs_bug_on(sbi, 1);
 795                         return;
 796                 }
 797                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 798                         dirty_i->nr_dirty[t]++;
 799
 800                 if (__is_large_section(sbi)) {
 801                         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 802                         block_t valid_blocks =
 803                                 get_valid_blocks(sbi, segno, true);
 804
 805                         f2fs_bug_on(sbi, unlikely(!valid_blocks ||
 806                                         valid_blocks == BLKS_PER_SEC(sbi)));
 807
 808                         if (!IS_CURSEC(sbi, secno))
 809                                 set_bit(secno, dirty_i->dirty_secmap);
 810                 }
 811         }
 812 }
 813
 814 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 815                 enum dirty_type dirty_type)
 816 {
 817         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 818         block_t valid_blocks;
 819
 820         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 821                 dirty_i->nr_dirty[dirty_type]--;
 822
 823         if (dirty_type == DIRTY) {
 824                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 825                 enum dirty_type t = sentry->type;
 826
 827                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 828                         dirty_i->nr_dirty[t]--;
 829
 830                 valid_blocks = get_valid_blocks(sbi, segno, true);
 831                 if (valid_blocks == 0) {
 832                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 833                                                 dirty_i->victim_secmap);
 834 #ifdef CONFIG_F2FS_CHECK_FS
 835                         clear_bit(segno, SIT_I(sbi)->invalid_segmap);
 836 #endif
 837                 }
 838                 if (__is_large_section(sbi)) {
 839                         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 840
 841                         if (!valid_blocks ||
 842                                         valid_blocks == BLKS_PER_SEC(sbi)) {
 843                                 clear_bit(secno, dirty_i->dirty_secmap);
 844                                 return;
 845                         }
 846
 847                         if (!IS_CURSEC(sbi, secno))
 848                                 set_bit(secno, dirty_i->dirty_secmap);
 849                 }
 850         }
 851 }
 852
 853 /*
 854  * Should not occur error such as -ENOMEM.
 855  * Adding dirty entry into seglist is not critical operation.
 856  * If a given segment is one of current working segments, it won't be added.
 857  */
 858 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 859 {
 860         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 861         unsigned short valid_blocks, ckpt_valid_blocks;
 862         unsigned int usable_blocks;
 863
 864         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 865                 return;
 866
 867         usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
 868         mutex_lock(&dirty_i->seglist_lock);
 869
 870         valid_blocks = get_valid_blocks(sbi, segno, false);
 871         ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno);
 872
 873         if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
 874                 ckpt_valid_blocks == usable_blocks)) {
 875                 __locate_dirty_segment(sbi, segno, PRE);
 876                 __remove_dirty_segment(sbi, segno, DIRTY);
 877         } else if (valid_blocks < usable_blocks) {
 878                 __locate_dirty_segment(sbi, segno, DIRTY);
 879         } else {
 880                 /* Recovery routine with SSR needs this */
 881                 __remove_dirty_segment(sbi, segno, DIRTY);
 882         }
 883
 884         mutex_unlock(&dirty_i->seglist_lock);
 885 }
 886
 887 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
 888 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
 889 {
 890         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 891         unsigned int segno;
 892
 893         mutex_lock(&dirty_i->seglist_lock);
 894         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 895                 if (get_valid_blocks(sbi, segno, false))
 896                         continue;
 897                 if (IS_CURSEG(sbi, segno))
 898                         continue;
 899                 __locate_dirty_segment(sbi, segno, PRE);
 900                 __remove_dirty_segment(sbi, segno, DIRTY);
 901         }
 902         mutex_unlock(&dirty_i->seglist_lock);
 903 }
 904
 905 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
 906 {
 907         int ovp_hole_segs =
 908                 (overprovision_segments(sbi) - reserved_segments(sbi));
 909         block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
 910         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 911         block_t holes[2] = {0, 0};      /* DATA and NODE */
 912         block_t unusable;
 913         struct seg_entry *se;
 914         unsigned int segno;
 915
 916         mutex_lock(&dirty_i->seglist_lock);
 917         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 918                 se = get_seg_entry(sbi, segno);
 919                 if (IS_NODESEG(se->type))
 920                         holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
 921                                                         se->valid_blocks;
 922                 else
 923                         holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
 924                                                         se->valid_blocks;
 925         }
 926         mutex_unlock(&dirty_i->seglist_lock);
 927
 928         unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
 929         if (unusable > ovp_holes)
 930                 return unusable - ovp_holes;
 931         return 0;
 932 }
 933
 934 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
 935 {
 936         int ovp_hole_segs =
 937                 (overprovision_segments(sbi) - reserved_segments(sbi));
 938         if (unusable > F2FS_OPTION(sbi).unusable_cap)
 939                 return -EAGAIN;
 940         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
 941                 dirty_segments(sbi) > ovp_hole_segs)
 942                 return -EAGAIN;
 943         return 0;
 944 }
 945
 946 /* This is only used by SBI_CP_DISABLED */
 947 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
 948 {
 949         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 950         unsigned int segno = 0;
 951
 952         mutex_lock(&dirty_i->seglist_lock);
 953         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 954                 if (get_valid_blocks(sbi, segno, false))
 955                         continue;
 956                 if (get_ckpt_valid_blocks(sbi, segno))
 957                         continue;
 958                 mutex_unlock(&dirty_i->seglist_lock);
 959                 return segno;
 960         }
 961         mutex_unlock(&dirty_i->seglist_lock);
 962         return NULL_SEGNO;
 963 }
 964
 965 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 966                 struct block_device *bdev, block_t lstart,
 967                 block_t start, block_t len)
 968 {
 969         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 970         struct list_head *pend_list;
 971         struct discard_cmd *dc;
 972
 973         f2fs_bug_on(sbi, !len);
 974
 975         pend_list = &dcc->pend_list[plist_idx(len)];
 976
 977         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 978         INIT_LIST_HEAD(&dc->list);
 979         dc->bdev = bdev;
 980         dc->lstart = lstart;
 981         dc->start = start;
 982         dc->len = len;
 983         dc->ref = 0;
 984         dc->state = D_PREP;
 985         dc->queued = 0;
 986         dc->error = 0;
 987         init_completion(&dc->wait);
 988         list_add_tail(&dc->list, pend_list);
 989         spin_lock_init(&dc->lock);
 990         dc->bio_ref = 0;
 991         atomic_inc(&dcc->discard_cmd_cnt);
 992         dcc->undiscard_blks += len;
 993
 994         return dc;
 995 }
 996
 997 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 998                                 struct block_device *bdev, block_t lstart,
 999                                 block_t start, block_t len,
1000                                 struct rb_node *parent, struct rb_node **p,
1001                                 bool leftmost)
1002 {
1003         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1004         struct discard_cmd *dc;
1005
1006         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1007
1008         rb_link_node(&dc->rb_node, parent, p);
1009         rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
1010
1011         return dc;
1012 }
1013
1014 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1015                                                         struct discard_cmd *dc)
1016 {
1017         if (dc->state == D_DONE)
1018                 atomic_sub(dc->queued, &dcc->queued_discard);
1019
1020         list_del(&dc->list);
1021         rb_erase_cached(&dc->rb_node, &dcc->root);
1022         dcc->undiscard_blks -= dc->len;
1023
1024         kmem_cache_free(discard_cmd_slab, dc);
1025
1026         atomic_dec(&dcc->discard_cmd_cnt);
1027 }
1028
1029 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1030                                                         struct discard_cmd *dc)
1031 {
1032         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1033         unsigned long flags;
1034
1035         trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
1036
1037         spin_lock_irqsave(&dc->lock, flags);
1038         if (dc->bio_ref) {
1039                 spin_unlock_irqrestore(&dc->lock, flags);
1040                 return;
1041         }
1042         spin_unlock_irqrestore(&dc->lock, flags);
1043
1044         f2fs_bug_on(sbi, dc->ref);
1045
1046         if (dc->error == -EOPNOTSUPP)
1047                 dc->error = 0;
1048
1049         if (dc->error)
1050                 printk_ratelimited(
1051                         "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
1052                         KERN_INFO, sbi->sb->s_id,
1053                         dc->lstart, dc->start, dc->len, dc->error);
1054         __detach_discard_cmd(dcc, dc);
1055 }
1056
1057 static void f2fs_submit_discard_endio(struct bio *bio)
1058 {
1059         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1060         unsigned long flags;
1061
1062         spin_lock_irqsave(&dc->lock, flags);
1063         if (!dc->error)
1064                 dc->error = blk_status_to_errno(bio->bi_status);
1065         dc->bio_ref--;
1066         if (!dc->bio_ref && dc->state == D_SUBMIT) {
1067                 dc->state = D_DONE;
1068                 complete_all(&dc->wait);
1069         }
1070         spin_unlock_irqrestore(&dc->lock, flags);
1071         bio_put(bio);
1072 }
1073
1074 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1075                                 block_t start, block_t end)
1076 {
1077 #ifdef CONFIG_F2FS_CHECK_FS
1078         struct seg_entry *sentry;
1079         unsigned int segno;
1080         block_t blk = start;
1081         unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1082         unsigned long *map;
1083
1084         while (blk < end) {
1085                 segno = GET_SEGNO(sbi, blk);
1086                 sentry = get_seg_entry(sbi, segno);
1087                 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1088
1089                 if (end < START_BLOCK(sbi, segno + 1))
1090                         size = GET_BLKOFF_FROM_SEG0(sbi, end);
1091                 else
1092                         size = max_blocks;
1093                 map = (unsigned long *)(sentry->cur_valid_map);
1094                 offset = __find_rev_next_bit(map, size, offset);
1095                 f2fs_bug_on(sbi, offset != size);
1096                 blk = START_BLOCK(sbi, segno + 1);
1097         }
1098 #endif
1099 }
1100
1101 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1102                                 struct discard_policy *dpolicy,
1103                                 int discard_type, unsigned int granularity)
1104 {
1105         /* common policy */
1106         dpolicy->type = discard_type;
1107         dpolicy->sync = true;
1108         dpolicy->ordered = false;
1109         dpolicy->granularity = granularity;
1110
1111         dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1112         dpolicy->io_aware_gran = MAX_PLIST_NUM;
1113         dpolicy->timeout = false;
1114
1115         if (discard_type == DPOLICY_BG) {
1116                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1117                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1118                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1119                 dpolicy->io_aware = true;
1120                 dpolicy->sync = false;
1121                 dpolicy->ordered = true;
1122                 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1123                         dpolicy->granularity = 1;
1124                         dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1125                 }
1126         } else if (discard_type == DPOLICY_FORCE) {
1127                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1128                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1129                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1130                 dpolicy->io_aware = false;
1131         } else if (discard_type == DPOLICY_FSTRIM) {
1132                 dpolicy->io_aware = false;
1133         } else if (discard_type == DPOLICY_UMOUNT) {
1134                 dpolicy->io_aware = false;
1135                 /* we need to issue all to keep CP_TRIMMED_FLAG */
1136                 dpolicy->granularity = 1;
1137                 dpolicy->timeout = true;
1138         }
1139 }
1140
1141 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1142                                 struct block_device *bdev, block_t lstart,
1143                                 block_t start, block_t len);
1144 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1145 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1146                                                 struct discard_policy *dpolicy,
1147                                                 struct discard_cmd *dc,
1148                                                 unsigned int *issued)
1149 {
1150         struct block_device *bdev = dc->bdev;
1151         struct request_queue *q = bdev_get_queue(bdev);
1152         unsigned int max_discard_blocks =
1153                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1154         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1155         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1156                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1157         int flag = dpolicy->sync ? REQ_SYNC : 0;
1158         block_t lstart, start, len, total_len;
1159         int err = 0;
1160
1161         if (dc->state != D_PREP)
1162                 return 0;
1163
1164         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1165                 return 0;
1166
1167         trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1168
1169         lstart = dc->lstart;
1170         start = dc->start;
1171         len = dc->len;
1172         total_len = len;
1173
1174         dc->len = 0;
1175
1176         while (total_len && *issued < dpolicy->max_requests && !err) {
1177                 struct bio *bio = NULL;
1178                 unsigned long flags;
1179                 bool last = true;
1180
1181                 if (len > max_discard_blocks) {
1182                         len = max_discard_blocks;
1183                         last = false;
1184                 }
1185
1186                 (*issued)++;
1187                 if (*issued == dpolicy->max_requests)
1188                         last = true;
1189
1190                 dc->len += len;
1191
1192                 if (time_to_inject(sbi, FAULT_DISCARD)) {
1193                         f2fs_show_injection_info(sbi, FAULT_DISCARD);
1194                         err = -EIO;
1195                         goto submit;
1196                 }
1197                 err = __blkdev_issue_discard(bdev,
1198                                         SECTOR_FROM_BLOCK(start),
1199                                         SECTOR_FROM_BLOCK(len),
1200                                         GFP_NOFS, 0, &bio);
1201 submit:
1202                 if (err) {
1203                         spin_lock_irqsave(&dc->lock, flags);
1204                         if (dc->state == D_PARTIAL)
1205                                 dc->state = D_SUBMIT;
1206                         spin_unlock_irqrestore(&dc->lock, flags);
1207
1208                         break;
1209                 }
1210
1211                 f2fs_bug_on(sbi, !bio);
1212
1213                 /*
1214                  * should keep before submission to avoid D_DONE
1215                  * right away
1216                  */
1217                 spin_lock_irqsave(&dc->lock, flags);
1218                 if (last)
1219                         dc->state = D_SUBMIT;
1220                 else
1221                         dc->state = D_PARTIAL;
1222                 dc->bio_ref++;
1223                 spin_unlock_irqrestore(&dc->lock, flags);
1224
1225                 atomic_inc(&dcc->queued_discard);
1226                 dc->queued++;
1227                 list_move_tail(&dc->list, wait_list);
1228
1229                 /* sanity check on discard range */
1230                 __check_sit_bitmap(sbi, lstart, lstart + len);
1231
1232                 bio->bi_private = dc;
1233                 bio->bi_end_io = f2fs_submit_discard_endio;
1234                 bio->bi_opf |= flag;
1235                 submit_bio(bio);
1236
1237                 atomic_inc(&dcc->issued_discard);
1238
1239                 f2fs_update_iostat(sbi, FS_DISCARD, 1);
1240
1241                 lstart += len;
1242                 start += len;
1243                 total_len -= len;
1244                 len = total_len;
1245         }
1246
1247         if (!err && len) {
1248                 dcc->undiscard_blks -= len;
1249                 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1250         }
1251         return err;
1252 }
1253
1254 static void __insert_discard_tree(struct f2fs_sb_info *sbi,
1255                                 struct block_device *bdev, block_t lstart,
1256                                 block_t start, block_t len,
1257                                 struct rb_node **insert_p,
1258                                 struct rb_node *insert_parent)
1259 {
1260         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1261         struct rb_node **p;
1262         struct rb_node *parent = NULL;
1263         bool leftmost = true;
1264
1265         if (insert_p && insert_parent) {
1266                 parent = insert_parent;
1267                 p = insert_p;
1268                 goto do_insert;
1269         }
1270
1271         p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1272                                                         lstart, &leftmost);
1273 do_insert:
1274         __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1275                                                                 p, leftmost);
1276 }
1277
1278 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1279                                                 struct discard_cmd *dc)
1280 {
1281         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1282 }
1283
1284 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1285                                 struct discard_cmd *dc, block_t blkaddr)
1286 {
1287         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1288         struct discard_info di = dc->di;
1289         bool modified = false;
1290
1291         if (dc->state == D_DONE || dc->len == 1) {
1292                 __remove_discard_cmd(sbi, dc);
1293                 return;
1294         }
1295
1296         dcc->undiscard_blks -= di.len;
1297
1298         if (blkaddr > di.lstart) {
1299                 dc->len = blkaddr - dc->lstart;
1300                 dcc->undiscard_blks += dc->len;
1301                 __relocate_discard_cmd(dcc, dc);
1302                 modified = true;
1303         }
1304
1305         if (blkaddr < di.lstart + di.len - 1) {
1306                 if (modified) {
1307                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1308                                         di.start + blkaddr + 1 - di.lstart,
1309                                         di.lstart + di.len - 1 - blkaddr,
1310                                         NULL, NULL);
1311                 } else {
1312                         dc->lstart++;
1313                         dc->len--;
1314                         dc->start++;
1315                         dcc->undiscard_blks += dc->len;
1316                         __relocate_discard_cmd(dcc, dc);
1317                 }
1318         }
1319 }
1320
1321 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1322                                 struct block_device *bdev, block_t lstart,
1323                                 block_t start, block_t len)
1324 {
1325         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1326         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1327         struct discard_cmd *dc;
1328         struct discard_info di = {0};
1329         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1330         struct request_queue *q = bdev_get_queue(bdev);
1331         unsigned int max_discard_blocks =
1332                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1333         block_t end = lstart + len;
1334
1335         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1336                                         NULL, lstart,
1337                                         (struct rb_entry **)&prev_dc,
1338                                         (struct rb_entry **)&next_dc,
1339                                         &insert_p, &insert_parent, true, NULL);
1340         if (dc)
1341                 prev_dc = dc;
1342
1343         if (!prev_dc) {
1344                 di.lstart = lstart;
1345                 di.len = next_dc ? next_dc->lstart - lstart : len;
1346                 di.len = min(di.len, len);
1347                 di.start = start;
1348         }
1349
1350         while (1) {
1351                 struct rb_node *node;
1352                 bool merged = false;
1353                 struct discard_cmd *tdc = NULL;
1354
1355                 if (prev_dc) {
1356                         di.lstart = prev_dc->lstart + prev_dc->len;
1357                         if (di.lstart < lstart)
1358                                 di.lstart = lstart;
1359                         if (di.lstart >= end)
1360                                 break;
1361
1362                         if (!next_dc || next_dc->lstart > end)
1363                                 di.len = end - di.lstart;
1364                         else
1365                                 di.len = next_dc->lstart - di.lstart;
1366                         di.start = start + di.lstart - lstart;
1367                 }
1368
1369                 if (!di.len)
1370                         goto next;
1371
1372                 if (prev_dc && prev_dc->state == D_PREP &&
1373                         prev_dc->bdev == bdev &&
1374                         __is_discard_back_mergeable(&di, &prev_dc->di,
1375                                                         max_discard_blocks)) {
1376                         prev_dc->di.len += di.len;
1377                         dcc->undiscard_blks += di.len;
1378                         __relocate_discard_cmd(dcc, prev_dc);
1379                         di = prev_dc->di;
1380                         tdc = prev_dc;
1381                         merged = true;
1382                 }
1383
1384                 if (next_dc && next_dc->state == D_PREP &&
1385                         next_dc->bdev == bdev &&
1386                         __is_discard_front_mergeable(&di, &next_dc->di,
1387                                                         max_discard_blocks)) {
1388                         next_dc->di.lstart = di.lstart;
1389                         next_dc->di.len += di.len;
1390                         next_dc->di.start = di.start;
1391                         dcc->undiscard_blks += di.len;
1392                         __relocate_discard_cmd(dcc, next_dc);
1393                         if (tdc)
1394                                 __remove_discard_cmd(sbi, tdc);
1395                         merged = true;
1396                 }
1397
1398                 if (!merged) {
1399                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1400                                                         di.len, NULL, NULL);
1401                 }
1402  next:
1403                 prev_dc = next_dc;
1404                 if (!prev_dc)
1405                         break;
1406
1407                 node = rb_next(&prev_dc->rb_node);
1408                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1409         }
1410 }
1411
1412 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1413                 struct block_device *bdev, block_t blkstart, block_t blklen)
1414 {
1415         block_t lblkstart = blkstart;
1416
1417         if (!f2fs_bdev_support_discard(bdev))
1418                 return 0;
1419
1420         trace_f2fs_queue_discard(bdev, blkstart, blklen);
1421
1422         if (f2fs_is_multi_device(sbi)) {
1423                 int devi = f2fs_target_device_index(sbi, blkstart);
1424
1425                 blkstart -= FDEV(devi).start_blk;
1426         }
1427         mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1428         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1429         mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1430         return 0;
1431 }
1432
1433 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1434                                         struct discard_policy *dpolicy)
1435 {
1436         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1437         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1438         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1439         struct discard_cmd *dc;
1440         struct blk_plug plug;
1441         unsigned int pos = dcc->next_pos;
1442         unsigned int issued = 0;
1443         bool io_interrupted = false;
1444
1445         mutex_lock(&dcc->cmd_lock);
1446         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1447                                         NULL, pos,
1448                                         (struct rb_entry **)&prev_dc,
1449                                         (struct rb_entry **)&next_dc,
1450                                         &insert_p, &insert_parent, true, NULL);
1451         if (!dc)
1452                 dc = next_dc;
1453
1454         blk_start_plug(&plug);
1455
1456         while (dc) {
1457                 struct rb_node *node;
1458                 int err = 0;
1459
1460                 if (dc->state != D_PREP)
1461                         goto next;
1462
1463                 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1464                         io_interrupted = true;
1465                         break;
1466                 }
1467
1468                 dcc->next_pos = dc->lstart + dc->len;
1469                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1470
1471                 if (issued >= dpolicy->max_requests)
1472                         break;
1473 next:
1474                 node = rb_next(&dc->rb_node);
1475                 if (err)
1476                         __remove_discard_cmd(sbi, dc);
1477                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1478         }
1479
1480         blk_finish_plug(&plug);
1481
1482         if (!dc)
1483                 dcc->next_pos = 0;
1484
1485         mutex_unlock(&dcc->cmd_lock);
1486
1487         if (!issued && io_interrupted)
1488                 issued = -1;
1489
1490         return issued;
1491 }
1492 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1493                                         struct discard_policy *dpolicy);
1494
1495 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1496                                         struct discard_policy *dpolicy)
1497 {
1498         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1499         struct list_head *pend_list;
1500         struct discard_cmd *dc, *tmp;
1501         struct blk_plug plug;
1502         int i, issued;
1503         bool io_interrupted = false;
1504
1505         if (dpolicy->timeout)
1506                 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1507
1508 retry:
1509         issued = 0;
1510         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1511                 if (dpolicy->timeout &&
1512                                 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1513                         break;
1514
1515                 if (i + 1 < dpolicy->granularity)
1516                         break;
1517
1518                 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1519                         return __issue_discard_cmd_orderly(sbi, dpolicy);
1520
1521                 pend_list = &dcc->pend_list[i];
1522
1523                 mutex_lock(&dcc->cmd_lock);
1524                 if (list_empty(pend_list))
1525                         goto next;
1526                 if (unlikely(dcc->rbtree_check))
1527                         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1528                                                                 &dcc->root));
1529                 blk_start_plug(&plug);
1530                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1531                         f2fs_bug_on(sbi, dc->state != D_PREP);
1532
1533                         if (dpolicy->timeout &&
1534                                 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1535                                 break;
1536
1537                         if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1538                                                 !is_idle(sbi, DISCARD_TIME)) {
1539                                 io_interrupted = true;
1540                                 break;
1541                         }
1542
1543                         __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1544
1545                         if (issued >= dpolicy->max_requests)
1546                                 break;
1547                 }
1548                 blk_finish_plug(&plug);
1549 next:
1550                 mutex_unlock(&dcc->cmd_lock);
1551
1552                 if (issued >= dpolicy->max_requests || io_interrupted)
1553                         break;
1554         }
1555
1556         if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1557                 __wait_all_discard_cmd(sbi, dpolicy);
1558                 goto retry;
1559         }
1560
1561         if (!issued && io_interrupted)
1562                 issued = -1;
1563
1564         return issued;
1565 }
1566
1567 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1568 {
1569         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1570         struct list_head *pend_list;
1571         struct discard_cmd *dc, *tmp;
1572         int i;
1573         bool dropped = false;
1574
1575         mutex_lock(&dcc->cmd_lock);
1576         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1577                 pend_list = &dcc->pend_list[i];
1578                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1579                         f2fs_bug_on(sbi, dc->state != D_PREP);
1580                         __remove_discard_cmd(sbi, dc);
1581                         dropped = true;
1582                 }
1583         }
1584         mutex_unlock(&dcc->cmd_lock);
1585
1586         return dropped;
1587 }
1588
1589 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1590 {
1591         __drop_discard_cmd(sbi);
1592 }
1593
1594 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1595                                                         struct discard_cmd *dc)
1596 {
1597         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1598         unsigned int len = 0;
1599
1600         wait_for_completion_io(&dc->wait);
1601         mutex_lock(&dcc->cmd_lock);
1602         f2fs_bug_on(sbi, dc->state != D_DONE);
1603         dc->ref--;
1604         if (!dc->ref) {
1605                 if (!dc->error)
1606                         len = dc->len;
1607                 __remove_discard_cmd(sbi, dc);
1608         }
1609         mutex_unlock(&dcc->cmd_lock);
1610
1611         return len;
1612 }
1613
1614 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1615                                                 struct discard_policy *dpolicy,
1616                                                 block_t start, block_t end)
1617 {
1618         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1619         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1620                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1621         struct discard_cmd *dc, *tmp;
1622         bool need_wait;
1623         unsigned int trimmed = 0;
1624
1625 next:
1626         need_wait = false;
1627
1628         mutex_lock(&dcc->cmd_lock);
1629         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1630                 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1631                         continue;
1632                 if (dc->len < dpolicy->granularity)
1633                         continue;
1634                 if (dc->state == D_DONE && !dc->ref) {
1635                         wait_for_completion_io(&dc->wait);
1636                         if (!dc->error)
1637                                 trimmed += dc->len;
1638                         __remove_discard_cmd(sbi, dc);
1639                 } else {
1640                         dc->ref++;
1641                         need_wait = true;
1642                         break;
1643                 }
1644         }
1645         mutex_unlock(&dcc->cmd_lock);
1646
1647         if (need_wait) {
1648                 trimmed += __wait_one_discard_bio(sbi, dc);
1649                 goto next;
1650         }
1651
1652         return trimmed;
1653 }
1654
1655 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1656                                                 struct discard_policy *dpolicy)
1657 {
1658         struct discard_policy dp;
1659         unsigned int discard_blks;
1660
1661         if (dpolicy)
1662                 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1663
1664         /* wait all */
1665         __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1666         discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1667         __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1668         discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1669
1670         return discard_blks;
1671 }
1672
1673 /* This should be covered by global mutex, &sit_i->sentry_lock */
1674 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1675 {
1676         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1677         struct discard_cmd *dc;
1678         bool need_wait = false;
1679
1680         mutex_lock(&dcc->cmd_lock);
1681         dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1682                                                         NULL, blkaddr);
1683         if (dc) {
1684                 if (dc->state == D_PREP) {
1685                         __punch_discard_cmd(sbi, dc, blkaddr);
1686                 } else {
1687                         dc->ref++;
1688                         need_wait = true;
1689                 }
1690         }
1691         mutex_unlock(&dcc->cmd_lock);
1692
1693         if (need_wait)
1694                 __wait_one_discard_bio(sbi, dc);
1695 }
1696
1697 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1698 {
1699         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1700
1701         if (dcc && dcc->f2fs_issue_discard) {
1702                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1703
1704                 dcc->f2fs_issue_discard = NULL;
1705                 kthread_stop(discard_thread);
1706         }
1707 }
1708
1709 /* This comes from f2fs_put_super */
1710 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1711 {
1712         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1713         struct discard_policy dpolicy;
1714         bool dropped;
1715
1716         __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1717                                         dcc->discard_granularity);
1718         __issue_discard_cmd(sbi, &dpolicy);
1719         dropped = __drop_discard_cmd(sbi);
1720
1721         /* just to make sure there is no pending discard commands */
1722         __wait_all_discard_cmd(sbi, NULL);
1723
1724         f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1725         return dropped;
1726 }
1727
1728 static int issue_discard_thread(void *data)
1729 {
1730         struct f2fs_sb_info *sbi = data;
1731         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1732         wait_queue_head_t *q = &dcc->discard_wait_queue;
1733         struct discard_policy dpolicy;
1734         unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1735         int issued;
1736
1737         set_freezable();
1738
1739         do {
1740                 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1741                                         dcc->discard_granularity);
1742
1743                 wait_event_interruptible_timeout(*q,
1744                                 kthread_should_stop() || freezing(current) ||
1745                                 dcc->discard_wake,
1746                                 msecs_to_jiffies(wait_ms));
1747
1748                 if (dcc->discard_wake)
1749                         dcc->discard_wake = 0;
1750
1751                 /* clean up pending candidates before going to sleep */
1752                 if (atomic_read(&dcc->queued_discard))
1753                         __wait_all_discard_cmd(sbi, NULL);
1754
1755                 if (try_to_freeze())
1756                         continue;
1757                 if (f2fs_readonly(sbi->sb))
1758                         continue;
1759                 if (kthread_should_stop())
1760                         return 0;
1761                 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1762                         wait_ms = dpolicy.max_interval;
1763                         continue;
1764                 }
1765
1766                 if (sbi->gc_mode == GC_URGENT_HIGH)
1767                         __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1768
1769                 sb_start_intwrite(sbi->sb);
1770
1771                 issued = __issue_discard_cmd(sbi, &dpolicy);
1772                 if (issued > 0) {
1773                         __wait_all_discard_cmd(sbi, &dpolicy);
1774                         wait_ms = dpolicy.min_interval;
1775                 } else if (issued == -1){
1776                         wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1777                         if (!wait_ms)
1778                                 wait_ms = dpolicy.mid_interval;
1779                 } else {
1780                         wait_ms = dpolicy.max_interval;
1781                 }
1782
1783                 sb_end_intwrite(sbi->sb);
1784
1785         } while (!kthread_should_stop());
1786         return 0;
1787 }
1788
1789 #ifdef CONFIG_BLK_DEV_ZONED
1790 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1791                 struct block_device *bdev, block_t blkstart, block_t blklen)
1792 {
1793         sector_t sector, nr_sects;
1794         block_t lblkstart = blkstart;
1795         int devi = 0;
1796
1797         if (f2fs_is_multi_device(sbi)) {
1798                 devi = f2fs_target_device_index(sbi, blkstart);
1799                 if (blkstart < FDEV(devi).start_blk ||
1800                     blkstart > FDEV(devi).end_blk) {
1801                         f2fs_err(sbi, "Invalid block %x", blkstart);
1802                         return -EIO;
1803                 }
1804                 blkstart -= FDEV(devi).start_blk;
1805         }
1806
1807         /* For sequential zones, reset the zone write pointer */
1808         if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1809                 sector = SECTOR_FROM_BLOCK(blkstart);
1810                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1811
1812                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1813                                 nr_sects != bdev_zone_sectors(bdev)) {
1814                         f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1815                                  devi, sbi->s_ndevs ? FDEV(devi).path : "",
1816                                  blkstart, blklen);
1817                         return -EIO;
1818                 }
1819                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1820                 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1821                                         sector, nr_sects, GFP_NOFS);
1822         }
1823
1824         /* For conventional zones, use regular discard if supported */
1825         return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1826 }
1827 #endif
1828
1829 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1830                 struct block_device *bdev, block_t blkstart, block_t blklen)
1831 {
1832 #ifdef CONFIG_BLK_DEV_ZONED
1833         if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1834                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1835 #endif
1836         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1837 }
1838
1839 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1840                                 block_t blkstart, block_t blklen)
1841 {
1842         sector_t start = blkstart, len = 0;
1843         struct block_device *bdev;
1844         struct seg_entry *se;
1845         unsigned int offset;
1846         block_t i;
1847         int err = 0;
1848
1849         bdev = f2fs_target_device(sbi, blkstart, NULL);
1850
1851         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1852                 if (i != start) {
1853                         struct block_device *bdev2 =
1854                                 f2fs_target_device(sbi, i, NULL);
1855
1856                         if (bdev2 != bdev) {
1857                                 err = __issue_discard_async(sbi, bdev,
1858                                                 start, len);
1859                                 if (err)
1860                                         return err;
1861                                 bdev = bdev2;
1862                                 start = i;
1863                                 len = 0;
1864                         }
1865                 }
1866
1867                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1868                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1869
1870                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1871                         sbi->discard_blks--;
1872         }
1873
1874         if (len)
1875                 err = __issue_discard_async(sbi, bdev, start, len);
1876         return err;
1877 }
1878
1879 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1880                                                         bool check_only)
1881 {
1882         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1883         int max_blocks = sbi->blocks_per_seg;
1884         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1885         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1886         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1887         unsigned long *discard_map = (unsigned long *)se->discard_map;
1888         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1889         unsigned int start = 0, end = -1;
1890         bool force = (cpc->reason & CP_DISCARD);
1891         struct discard_entry *de = NULL;
1892         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1893         int i;
1894
1895         if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
1896                 return false;
1897
1898         if (!force) {
1899                 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1900                         SM_I(sbi)->dcc_info->nr_discards >=
1901                                 SM_I(sbi)->dcc_info->max_discards)
1902                         return false;
1903         }
1904
1905         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1906         for (i = 0; i < entries; i++)
1907                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1908                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1909
1910         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1911                                 SM_I(sbi)->dcc_info->max_discards) {
1912                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1913                 if (start >= max_blocks)
1914                         break;
1915
1916                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1917                 if (force && start && end != max_blocks
1918                                         && (end - start) < cpc->trim_minlen)
1919                         continue;
1920
1921                 if (check_only)
1922                         return true;
1923
1924                 if (!de) {
1925                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1926                                                                 GFP_F2FS_ZERO);
1927                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1928                         list_add_tail(&de->list, head);
1929                 }
1930
1931                 for (i = start; i < end; i++)
1932                         __set_bit_le(i, (void *)de->discard_map);
1933
1934                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1935         }
1936         return false;
1937 }
1938
1939 static void release_discard_addr(struct discard_entry *entry)
1940 {
1941         list_del(&entry->list);
1942         kmem_cache_free(discard_entry_slab, entry);
1943 }
1944
1945 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1946 {
1947         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1948         struct discard_entry *entry, *this;
1949
1950         /* drop caches */
1951         list_for_each_entry_safe(entry, this, head, list)
1952                 release_discard_addr(entry);
1953 }
1954
1955 /*
1956  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1957  */
1958 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1959 {
1960         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1961         unsigned int segno;
1962
1963         mutex_lock(&dirty_i->seglist_lock);
1964         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1965                 __set_test_and_free(sbi, segno, false);
1966         mutex_unlock(&dirty_i->seglist_lock);
1967 }
1968
1969 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1970                                                 struct cp_control *cpc)
1971 {
1972         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1973         struct list_head *head = &dcc->entry_list;
1974         struct discard_entry *entry, *this;
1975         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1976         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1977         unsigned int start = 0, end = -1;
1978         unsigned int secno, start_segno;
1979         bool force = (cpc->reason & CP_DISCARD);
1980         bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
1981
1982         mutex_lock(&dirty_i->seglist_lock);
1983
1984         while (1) {
1985                 int i;
1986
1987                 if (need_align && end != -1)
1988                         end--;
1989                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1990                 if (start >= MAIN_SEGS(sbi))
1991                         break;
1992                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1993                                                                 start + 1);
1994
1995                 if (need_align) {
1996                         start = rounddown(start, sbi->segs_per_sec);
1997                         end = roundup(end, sbi->segs_per_sec);
1998                 }
1999
2000                 for (i = start; i < end; i++) {
2001                         if (test_and_clear_bit(i, prefree_map))
2002                                 dirty_i->nr_dirty[PRE]--;
2003                 }
2004
2005                 if (!f2fs_realtime_discard_enable(sbi))
2006                         continue;
2007
2008                 if (force && start >= cpc->trim_start &&
2009                                         (end - 1) <= cpc->trim_end)
2010                                 continue;
2011
2012                 if (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi)) {
2013                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2014                                 (end - start) << sbi->log_blocks_per_seg);
2015                         continue;
2016                 }
2017 next:
2018                 secno = GET_SEC_FROM_SEG(sbi, start);
2019                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
2020                 if (!IS_CURSEC(sbi, secno) &&
2021                         !get_valid_blocks(sbi, start, true))
2022                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2023                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
2024
2025                 start = start_segno + sbi->segs_per_sec;
2026                 if (start < end)
2027                         goto next;
2028                 else
2029                         end = start - 1;
2030         }
2031         mutex_unlock(&dirty_i->seglist_lock);
2032
2033         /* send small discards */
2034         list_for_each_entry_safe(entry, this, head, list) {
2035                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2036                 bool is_valid = test_bit_le(0, entry->discard_map);
2037
2038 find_next:
2039                 if (is_valid) {
2040                         next_pos = find_next_zero_bit_le(entry->discard_map,
2041                                         sbi->blocks_per_seg, cur_pos);
2042                         len = next_pos - cur_pos;
2043
2044                         if (f2fs_sb_has_blkzoned(sbi) ||
2045                             (force && len < cpc->trim_minlen))
2046                                 goto skip;
2047
2048                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2049                                                                         len);
2050                         total_len += len;
2051                 } else {
2052                         next_pos = find_next_bit_le(entry->discard_map,
2053                                         sbi->blocks_per_seg, cur_pos);
2054                 }
2055 skip:
2056                 cur_pos = next_pos;
2057                 is_valid = !is_valid;
2058
2059                 if (cur_pos < sbi->blocks_per_seg)
2060                         goto find_next;
2061
2062                 release_discard_addr(entry);
2063                 dcc->nr_discards -= total_len;
2064         }
2065
2066         wake_up_discard_thread(sbi, false);
2067 }
2068
2069 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2070 {
2071         dev_t dev = sbi->sb->s_bdev->bd_dev;
2072         struct discard_cmd_control *dcc;
2073         int err = 0, i;
2074
2075         if (SM_I(sbi)->dcc_info) {
2076                 dcc = SM_I(sbi)->dcc_info;
2077                 goto init_thread;
2078         }
2079
2080         dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2081         if (!dcc)
2082                 return -ENOMEM;
2083
2084         dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2085         INIT_LIST_HEAD(&dcc->entry_list);
2086         for (i = 0; i < MAX_PLIST_NUM; i++)
2087                 INIT_LIST_HEAD(&dcc->pend_list[i]);
2088         INIT_LIST_HEAD(&dcc->wait_list);
2089         INIT_LIST_HEAD(&dcc->fstrim_list);
2090         mutex_init(&dcc->cmd_lock);
2091         atomic_set(&dcc->issued_discard, 0);
2092         atomic_set(&dcc->queued_discard, 0);
2093         atomic_set(&dcc->discard_cmd_cnt, 0);
2094         dcc->nr_discards = 0;
2095         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2096         dcc->undiscard_blks = 0;
2097         dcc->next_pos = 0;
2098         dcc->root = RB_ROOT_CACHED;
2099         dcc->rbtree_check = false;
2100
2101         init_waitqueue_head(&dcc->discard_wait_queue);
2102         SM_I(sbi)->dcc_info = dcc;
2103 init_thread:
2104         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2105                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2106         if (IS_ERR(dcc->f2fs_issue_discard)) {
2107                 err = PTR_ERR(dcc->f2fs_issue_discard);
2108                 kvfree(dcc);
2109                 SM_I(sbi)->dcc_info = NULL;
2110                 return err;
2111         }
2112
2113         return err;
2114 }
2115
2116 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2117 {
2118         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2119
2120         if (!dcc)
2121                 return;
2122
2123         f2fs_stop_discard_thread(sbi);
2124
2125         /*
2126          * Recovery can cache discard commands, so in error path of
2127          * fill_super(), it needs to give a chance to handle them.
2128          */
2129         if (unlikely(atomic_read(&dcc->discard_cmd_cnt)))
2130                 f2fs_issue_discard_timeout(sbi);
2131
2132         kvfree(dcc);
2133         SM_I(sbi)->dcc_info = NULL;
2134 }
2135
2136 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2137 {
2138         struct sit_info *sit_i = SIT_I(sbi);
2139
2140         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2141                 sit_i->dirty_sentries++;
2142                 return false;
2143         }
2144
2145         return true;
2146 }
2147
2148 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2149                                         unsigned int segno, int modified)
2150 {
2151         struct seg_entry *se = get_seg_entry(sbi, segno);
2152         se->type = type;
2153         if (modified)
2154                 __mark_sit_entry_dirty(sbi, segno);
2155 }
2156
2157 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr)
2158 {
2159         unsigned int segno = GET_SEGNO(sbi, blkaddr);
2160         struct seg_entry *se = get_seg_entry(sbi, segno);
2161         unsigned long long mtime = get_mtime(sbi, false);
2162
2163         if (!se->mtime)
2164                 se->mtime = mtime;
2165         else
2166                 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2167                                                 se->valid_blocks + 1);
2168
2169         if (mtime > SIT_I(sbi)->max_mtime)
2170                 SIT_I(sbi)->max_mtime = mtime;
2171 }
2172
2173 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2174 {
2175         struct seg_entry *se;
2176         unsigned int segno, offset;
2177         long int new_vblocks;
2178         bool exist;
2179 #ifdef CONFIG_F2FS_CHECK_FS
2180         bool mir_exist;
2181 #endif
2182
2183         segno = GET_SEGNO(sbi, blkaddr);
2184
2185         se = get_seg_entry(sbi, segno);
2186         new_vblocks = se->valid_blocks + del;
2187         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2188
2189         f2fs_bug_on(sbi, (new_vblocks < 0 ||
2190                         (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2191
2192         update_segment_mtime(sbi, blkaddr);
2193
2194         se->valid_blocks = new_vblocks;
2195
2196         /* Update valid block bitmap */
2197         if (del > 0) {
2198                 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2199 #ifdef CONFIG_F2FS_CHECK_FS
2200                 mir_exist = f2fs_test_and_set_bit(offset,
2201                                                 se->cur_valid_map_mir);
2202                 if (unlikely(exist != mir_exist)) {
2203                         f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2204                                  blkaddr, exist);
2205                         f2fs_bug_on(sbi, 1);
2206                 }
2207 #endif
2208                 if (unlikely(exist)) {
2209                         f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2210                                  blkaddr);
2211                         f2fs_bug_on(sbi, 1);
2212                         se->valid_blocks--;
2213                         del = 0;
2214                 }
2215
2216                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
2217                         sbi->discard_blks--;
2218
2219                 /*
2220                  * SSR should never reuse block which is checkpointed
2221                  * or newly invalidated.
2222                  */
2223                 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2224                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2225                                 se->ckpt_valid_blocks++;
2226                 }
2227         } else {
2228                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2229 #ifdef CONFIG_F2FS_CHECK_FS
2230                 mir_exist = f2fs_test_and_clear_bit(offset,
2231                                                 se->cur_valid_map_mir);
2232                 if (unlikely(exist != mir_exist)) {
2233                         f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2234                                  blkaddr, exist);
2235                         f2fs_bug_on(sbi, 1);
2236                 }
2237 #endif
2238                 if (unlikely(!exist)) {
2239                         f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2240                                  blkaddr);
2241                         f2fs_bug_on(sbi, 1);
2242                         se->valid_blocks++;
2243                         del = 0;
2244                 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2245                         /*
2246                          * If checkpoints are off, we must not reuse data that
2247                          * was used in the previous checkpoint. If it was used
2248                          * before, we must track that to know how much space we
2249                          * really have.
2250                          */
2251                         if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2252                                 spin_lock(&sbi->stat_lock);
2253                                 sbi->unusable_block_count++;
2254                                 spin_unlock(&sbi->stat_lock);
2255                         }
2256                 }
2257
2258                 if (f2fs_test_and_clear_bit(offset, se->discard_map))
2259                         sbi->discard_blks++;
2260         }
2261         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2262                 se->ckpt_valid_blocks += del;
2263
2264         __mark_sit_entry_dirty(sbi, segno);
2265
2266         /* update total number of valid blocks to be written in ckpt area */
2267         SIT_I(sbi)->written_valid_blocks += del;
2268
2269         if (__is_large_section(sbi))
2270                 get_sec_entry(sbi, segno)->valid_blocks += del;
2271 }
2272
2273 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2274 {
2275         unsigned int segno = GET_SEGNO(sbi, addr);
2276         struct sit_info *sit_i = SIT_I(sbi);
2277
2278         f2fs_bug_on(sbi, addr == NULL_ADDR);
2279         if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2280                 return;
2281
2282         invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2283
2284         /* add it into sit main buffer */
2285         down_write(&sit_i->sentry_lock);
2286
2287         update_sit_entry(sbi, addr, -1);
2288
2289         /* add it into dirty seglist */
2290         locate_dirty_segment(sbi, segno);
2291
2292         up_write(&sit_i->sentry_lock);
2293 }
2294
2295 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2296 {
2297         struct sit_info *sit_i = SIT_I(sbi);
2298         unsigned int segno, offset;
2299         struct seg_entry *se;
2300         bool is_cp = false;
2301
2302         if (!__is_valid_data_blkaddr(blkaddr))
2303                 return true;
2304
2305         down_read(&sit_i->sentry_lock);
2306
2307         segno = GET_SEGNO(sbi, blkaddr);
2308         se = get_seg_entry(sbi, segno);
2309         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2310
2311         if (f2fs_test_bit(offset, se->ckpt_valid_map))
2312                 is_cp = true;
2313
2314         up_read(&sit_i->sentry_lock);
2315
2316         return is_cp;
2317 }
2318
2319 /*
2320  * This function should be resided under the curseg_mutex lock
2321  */
2322 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2323                                         struct f2fs_summary *sum)
2324 {
2325         struct curseg_info *curseg = CURSEG_I(sbi, type);
2326         void *addr = curseg->sum_blk;
2327         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2328         memcpy(addr, sum, sizeof(struct f2fs_summary));
2329 }
2330
2331 /*
2332  * Calculate the number of current summary pages for writing
2333  */
2334 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2335 {
2336         int valid_sum_count = 0;
2337         int i, sum_in_page;
2338
2339         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2340                 if (sbi->ckpt->alloc_type[i] == SSR)
2341                         valid_sum_count += sbi->blocks_per_seg;
2342                 else {
2343                         if (for_ra)
2344                                 valid_sum_count += le16_to_cpu(
2345                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2346                         else
2347                                 valid_sum_count += curseg_blkoff(sbi, i);
2348                 }
2349         }
2350
2351         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2352                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2353         if (valid_sum_count <= sum_in_page)
2354                 return 1;
2355         else if ((valid_sum_count - sum_in_page) <=
2356                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2357                 return 2;
2358         return 3;
2359 }
2360
2361 /*
2362  * Caller should put this summary page
2363  */
2364 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2365 {
2366         return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
2367 }
2368
2369 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2370                                         void *src, block_t blk_addr)
2371 {
2372         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2373
2374         memcpy(page_address(page), src, PAGE_SIZE);
2375         set_page_dirty(page);
2376         f2fs_put_page(page, 1);
2377 }
2378
2379 static void write_sum_page(struct f2fs_sb_info *sbi,
2380                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
2381 {
2382         f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2383 }
2384
2385 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2386                                                 int type, block_t blk_addr)
2387 {
2388         struct curseg_info *curseg = CURSEG_I(sbi, type);
2389         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2390         struct f2fs_summary_block *src = curseg->sum_blk;
2391         struct f2fs_summary_block *dst;
2392
2393         dst = (struct f2fs_summary_block *)page_address(page);
2394         memset(dst, 0, PAGE_SIZE);
2395
2396         mutex_lock(&curseg->curseg_mutex);
2397
2398         down_read(&curseg->journal_rwsem);
2399         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2400         up_read(&curseg->journal_rwsem);
2401
2402         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2403         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2404
2405         mutex_unlock(&curseg->curseg_mutex);
2406
2407         set_page_dirty(page);
2408         f2fs_put_page(page, 1);
2409 }
2410
2411 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2412 {
2413         struct curseg_info *curseg = CURSEG_I(sbi, type);
2414         unsigned int segno = curseg->segno + 1;
2415         struct free_segmap_info *free_i = FREE_I(sbi);
2416
2417         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2418                 return !test_bit(segno, free_i->free_segmap);
2419         return 0;
2420 }
2421
2422 /*
2423  * Find a new segment from the free segments bitmap to right order
2424  * This function should be returned with success, otherwise BUG
2425  */
2426 static void get_new_segment(struct f2fs_sb_info *sbi,
2427                         unsigned int *newseg, bool new_sec, int dir)
2428 {
2429         struct free_segmap_info *free_i = FREE_I(sbi);
2430         unsigned int segno, secno, zoneno;
2431         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2432         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2433         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2434         unsigned int left_start = hint;
2435         bool init = true;
2436         int go_left = 0;
2437         int i;
2438
2439         spin_lock(&free_i->segmap_lock);
2440
2441         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2442                 segno = find_next_zero_bit(free_i->free_segmap,
2443                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2444                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2445                         goto got_it;
2446         }
2447 find_other_zone:
2448         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2449         if (secno >= MAIN_SECS(sbi)) {
2450                 if (dir == ALLOC_RIGHT) {
2451                         secno = find_next_zero_bit(free_i->free_secmap,
2452                                                         MAIN_SECS(sbi), 0);
2453                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2454                 } else {
2455                         go_left = 1;
2456                         left_start = hint - 1;
2457                 }
2458         }
2459         if (go_left == 0)
2460                 goto skip_left;
2461
2462         while (test_bit(left_start, free_i->free_secmap)) {
2463                 if (left_start > 0) {
2464                         left_start--;
2465                         continue;
2466                 }
2467                 left_start = find_next_zero_bit(free_i->free_secmap,
2468                                                         MAIN_SECS(sbi), 0);
2469                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2470                 break;
2471         }
2472         secno = left_start;
2473 skip_left:
2474         segno = GET_SEG_FROM_SEC(sbi, secno);
2475         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2476
2477         /* give up on finding another zone */
2478         if (!init)
2479                 goto got_it;
2480         if (sbi->secs_per_zone == 1)
2481                 goto got_it;
2482         if (zoneno == old_zoneno)
2483                 goto got_it;
2484         if (dir == ALLOC_LEFT) {
2485                 if (!go_left && zoneno + 1 >= total_zones)
2486                         goto got_it;
2487                 if (go_left && zoneno == 0)
2488                         goto got_it;
2489         }
2490         for (i = 0; i < NR_CURSEG_TYPE; i++)
2491                 if (CURSEG_I(sbi, i)->zone == zoneno)
2492                         break;
2493
2494         if (i < NR_CURSEG_TYPE) {
2495                 /* zone is in user, try another */
2496                 if (go_left)
2497                         hint = zoneno * sbi->secs_per_zone - 1;
2498                 else if (zoneno + 1 >= total_zones)
2499                         hint = 0;
2500                 else
2501                         hint = (zoneno + 1) * sbi->secs_per_zone;
2502                 init = false;
2503                 goto find_other_zone;
2504         }
2505 got_it:
2506         /* set it as dirty segment in free segmap */
2507         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2508         __set_inuse(sbi, segno);
2509         *newseg = segno;
2510         spin_unlock(&free_i->segmap_lock);
2511 }
2512
2513 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2514 {
2515         struct curseg_info *curseg = CURSEG_I(sbi, type);
2516         struct summary_footer *sum_footer;
2517
2518         curseg->inited = true;
2519         curseg->segno = curseg->next_segno;
2520         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2521         curseg->next_blkoff = 0;
2522         curseg->next_segno = NULL_SEGNO;
2523
2524         sum_footer = &(curseg->sum_blk->footer);
2525         memset(sum_footer, 0, sizeof(struct summary_footer));
2526         if (IS_DATASEG(curseg->seg_type))
2527                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2528         if (IS_NODESEG(curseg->seg_type))
2529                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2530         __set_sit_entry_type(sbi, curseg->seg_type, curseg->segno, modified);
2531 }
2532
2533 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2534 {
2535         struct curseg_info *curseg = CURSEG_I(sbi, type);
2536
2537         /* if segs_per_sec is large than 1, we need to keep original policy. */
2538         if (__is_large_section(sbi))
2539                 return curseg->segno;
2540
2541         /* inmem log may not locate on any segment after mount */
2542         if (!curseg->inited)
2543                 return 0;
2544
2545         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2546                 return 0;
2547
2548         if (test_opt(sbi, NOHEAP) &&
2549                 (curseg->seg_type == CURSEG_HOT_DATA ||
2550                 IS_NODESEG(curseg->seg_type)))
2551                 return 0;
2552
2553         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2554                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2555
2556         /* find segments from 0 to reuse freed segments */
2557         if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2558                 return 0;
2559
2560         return curseg->segno;
2561 }
2562
2563 /*
2564  * Allocate a current working segment.
2565  * This function always allocates a free segment in LFS manner.
2566  */
2567 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2568 {
2569         struct curseg_info *curseg = CURSEG_I(sbi, type);
2570         unsigned short seg_type = curseg->seg_type;
2571         unsigned int segno = curseg->segno;
2572         int dir = ALLOC_LEFT;
2573
2574         if (curseg->inited)
2575                 write_sum_page(sbi, curseg->sum_blk,
2576                                 GET_SUM_BLOCK(sbi, segno));
2577         if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA)
2578                 dir = ALLOC_RIGHT;
2579
2580         if (test_opt(sbi, NOHEAP))
2581                 dir = ALLOC_RIGHT;
2582
2583         segno = __get_next_segno(sbi, type);
2584         get_new_segment(sbi, &segno, new_sec, dir);
2585         curseg->next_segno = segno;
2586         reset_curseg(sbi, type, 1);
2587         curseg->alloc_type = LFS;
2588 }
2589
2590 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2591                         struct curseg_info *seg, block_t start)
2592 {
2593         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2594         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2595         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2596         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2597         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2598         int i, pos;
2599
2600         for (i = 0; i < entries; i++)
2601                 target_map[i] = ckpt_map[i] | cur_map[i];
2602
2603         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2604
2605         seg->next_blkoff = pos;
2606 }
2607
2608 /*
2609  * If a segment is written by LFS manner, next block offset is just obtained
2610  * by increasing the current block offset. However, if a segment is written by
2611  * SSR manner, next block offset obtained by calling __next_free_blkoff
2612  */
2613 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2614                                 struct curseg_info *seg)
2615 {
2616         if (seg->alloc_type == SSR)
2617                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2618         else
2619                 seg->next_blkoff++;
2620 }
2621
2622 /*
2623  * This function always allocates a used segment(from dirty seglist) by SSR
2624  * manner, so it should recover the existing segment information of valid blocks
2625  */
2626 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2627 {
2628         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2629         struct curseg_info *curseg = CURSEG_I(sbi, type);
2630         unsigned int new_segno = curseg->next_segno;
2631         struct f2fs_summary_block *sum_node;
2632         struct page *sum_page;
2633
2634         write_sum_page(sbi, curseg->sum_blk,
2635                                 GET_SUM_BLOCK(sbi, curseg->segno));
2636         __set_test_and_inuse(sbi, new_segno);
2637
2638         mutex_lock(&dirty_i->seglist_lock);
2639         __remove_dirty_segment(sbi, new_segno, PRE);
2640         __remove_dirty_segment(sbi, new_segno, DIRTY);
2641         mutex_unlock(&dirty_i->seglist_lock);
2642
2643         reset_curseg(sbi, type, 1);
2644         curseg->alloc_type = SSR;
2645         __next_free_blkoff(sbi, curseg, 0);
2646
2647         sum_page = f2fs_get_sum_page(sbi, new_segno);
2648         f2fs_bug_on(sbi, IS_ERR(sum_page));
2649         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2650         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2651         f2fs_put_page(sum_page, 1);
2652 }
2653
2654 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2655 {
2656         struct curseg_info *curseg = CURSEG_I(sbi, type);
2657
2658         mutex_lock(&curseg->curseg_mutex);
2659         if (!curseg->inited)
2660                 goto out;
2661
2662         if (get_valid_blocks(sbi, curseg->segno, false)) {
2663                 write_sum_page(sbi, curseg->sum_blk,
2664                                 GET_SUM_BLOCK(sbi, curseg->segno));
2665         } else {
2666                 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2667                 __set_test_and_free(sbi, curseg->segno, true);
2668                 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2669         }
2670 out:
2671         mutex_unlock(&curseg->curseg_mutex);
2672 }
2673
2674 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2675 {
2676         struct curseg_info *curseg = CURSEG_I(sbi, type);
2677
2678         mutex_lock(&curseg->curseg_mutex);
2679         if (!curseg->inited)
2680                 goto out;
2681         if (get_valid_blocks(sbi, curseg->segno, false))
2682                 goto out;
2683
2684         mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2685         __set_test_and_inuse(sbi, curseg->segno);
2686         mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2687 out:
2688         mutex_unlock(&curseg->curseg_mutex);
2689 }
2690
2691 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2692 {
2693         struct curseg_info *curseg = CURSEG_I(sbi, type);
2694         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2695         unsigned segno = NULL_SEGNO;
2696         int i, cnt;
2697         bool reversed = false;
2698
2699         /* f2fs_need_SSR() already forces to do this */
2700         if (!v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2701                 curseg->next_segno = segno;
2702                 return 1;
2703         }
2704
2705         /* For node segments, let's do SSR more intensively */
2706         if (IS_NODESEG(type)) {
2707                 if (type >= CURSEG_WARM_NODE) {
2708                         reversed = true;
2709                         i = CURSEG_COLD_NODE;
2710                 } else {
2711                         i = CURSEG_HOT_NODE;
2712                 }
2713                 cnt = NR_CURSEG_NODE_TYPE;
2714         } else {
2715                 if (type >= CURSEG_WARM_DATA) {
2716                         reversed = true;
2717                         i = CURSEG_COLD_DATA;
2718                 } else {
2719                         i = CURSEG_HOT_DATA;
2720                 }
2721                 cnt = NR_CURSEG_DATA_TYPE;
2722         }
2723
2724         for (; cnt-- > 0; reversed ? i-- : i++) {
2725                 if (i == type)
2726                         continue;
2727                 if (!v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2728                         curseg->next_segno = segno;
2729                         return 1;
2730                 }
2731         }
2732
2733         /* find valid_blocks=0 in dirty list */
2734         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2735                 segno = get_free_segment(sbi);
2736                 if (segno != NULL_SEGNO) {
2737                         curseg->next_segno = segno;
2738                         return 1;
2739                 }
2740         }
2741         return 0;
2742 }
2743
2744 /*
2745  * flush out current segment and replace it with new segment
2746  * This function should be returned with success, otherwise BUG
2747  */
2748 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2749                                                 int type, bool force)
2750 {
2751         struct curseg_info *curseg = CURSEG_I(sbi, type);
2752
2753         if (force)
2754                 new_curseg(sbi, type, true);
2755         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2756                                         type == CURSEG_WARM_NODE)
2757                 new_curseg(sbi, type, false);
2758         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2759                         likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2760                 new_curseg(sbi, type, false);
2761         else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2762                 change_curseg(sbi, type);
2763         else
2764                 new_curseg(sbi, type, false);
2765
2766         stat_inc_seg_type(sbi, curseg);
2767 }
2768
2769 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2770                                         unsigned int start, unsigned int end)
2771 {
2772         struct curseg_info *curseg = CURSEG_I(sbi, type);
2773         unsigned int segno;
2774
2775         down_read(&SM_I(sbi)->curseg_lock);
2776         mutex_lock(&curseg->curseg_mutex);
2777         down_write(&SIT_I(sbi)->sentry_lock);
2778
2779         segno = CURSEG_I(sbi, type)->segno;
2780         if (segno < start || segno > end)
2781                 goto unlock;
2782
2783         if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2784                 change_curseg(sbi, type);
2785         else
2786                 new_curseg(sbi, type, true);
2787
2788         stat_inc_seg_type(sbi, curseg);
2789
2790         locate_dirty_segment(sbi, segno);
2791 unlock:
2792         up_write(&SIT_I(sbi)->sentry_lock);
2793
2794         if (segno != curseg->segno)
2795                 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2796                             type, segno, curseg->segno);
2797
2798         mutex_unlock(&curseg->curseg_mutex);
2799         up_read(&SM_I(sbi)->curseg_lock);
2800 }
2801
2802 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type)
2803 {
2804         struct curseg_info *curseg = CURSEG_I(sbi, type);
2805         unsigned int old_segno;
2806
2807         if (!curseg->inited)
2808                 goto alloc;
2809
2810         if (!curseg->next_blkoff &&
2811                 !get_valid_blocks(sbi, curseg->segno, false) &&
2812                 !get_ckpt_valid_blocks(sbi, curseg->segno))
2813                 return;
2814
2815 alloc:
2816         old_segno = curseg->segno;
2817         SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
2818         locate_dirty_segment(sbi, old_segno);
2819 }
2820
2821 void f2fs_allocate_new_segment(struct f2fs_sb_info *sbi, int type)
2822 {
2823         down_write(&SIT_I(sbi)->sentry_lock);
2824         __allocate_new_segment(sbi, type);
2825         up_write(&SIT_I(sbi)->sentry_lock);
2826 }
2827
2828 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2829 {
2830         int i;
2831
2832         down_write(&SIT_I(sbi)->sentry_lock);
2833         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
2834                 __allocate_new_segment(sbi, i);
2835         up_write(&SIT_I(sbi)->sentry_lock);
2836 }
2837
2838 static const struct segment_allocation default_salloc_ops = {
2839         .allocate_segment = allocate_segment_by_default,
2840 };
2841
2842 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2843                                                 struct cp_control *cpc)
2844 {
2845         __u64 trim_start = cpc->trim_start;
2846         bool has_candidate = false;
2847
2848         down_write(&SIT_I(sbi)->sentry_lock);
2849         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2850                 if (add_discard_addrs(sbi, cpc, true)) {
2851                         has_candidate = true;
2852                         break;
2853                 }
2854         }
2855         up_write(&SIT_I(sbi)->sentry_lock);
2856
2857         cpc->trim_start = trim_start;
2858         return has_candidate;
2859 }
2860
2861 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2862                                         struct discard_policy *dpolicy,
2863                                         unsigned int start, unsigned int end)
2864 {
2865         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2866         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2867         struct rb_node **insert_p = NULL, *insert_parent = NULL;
2868         struct discard_cmd *dc;
2869         struct blk_plug plug;
2870         int issued;
2871         unsigned int trimmed = 0;
2872
2873 next:
2874         issued = 0;
2875
2876         mutex_lock(&dcc->cmd_lock);
2877         if (unlikely(dcc->rbtree_check))
2878                 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2879                                                                 &dcc->root));
2880
2881         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2882                                         NULL, start,
2883                                         (struct rb_entry **)&prev_dc,
2884                                         (struct rb_entry **)&next_dc,
2885                                         &insert_p, &insert_parent, true, NULL);
2886         if (!dc)
2887                 dc = next_dc;
2888
2889         blk_start_plug(&plug);
2890
2891         while (dc && dc->lstart <= end) {
2892                 struct rb_node *node;
2893                 int err = 0;
2894
2895                 if (dc->len < dpolicy->granularity)
2896                         goto skip;
2897
2898                 if (dc->state != D_PREP) {
2899                         list_move_tail(&dc->list, &dcc->fstrim_list);
2900                         goto skip;
2901                 }
2902
2903                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2904
2905                 if (issued >= dpolicy->max_requests) {
2906                         start = dc->lstart + dc->len;
2907
2908                         if (err)
2909                                 __remove_discard_cmd(sbi, dc);
2910
2911                         blk_finish_plug(&plug);
2912                         mutex_unlock(&dcc->cmd_lock);
2913                         trimmed += __wait_all_discard_cmd(sbi, NULL);
2914                         congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2915                         goto next;
2916                 }
2917 skip:
2918                 node = rb_next(&dc->rb_node);
2919                 if (err)
2920                         __remove_discard_cmd(sbi, dc);
2921                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2922
2923                 if (fatal_signal_pending(current))
2924                         break;
2925         }
2926
2927         blk_finish_plug(&plug);
2928         mutex_unlock(&dcc->cmd_lock);
2929
2930         return trimmed;
2931 }
2932
2933 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2934 {
2935         __u64 start = F2FS_BYTES_TO_BLK(range->start);
2936         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2937         unsigned int start_segno, end_segno;
2938         block_t start_block, end_block;
2939         struct cp_control cpc;
2940         struct discard_policy dpolicy;
2941         unsigned long long trimmed = 0;
2942         int err = 0;
2943         bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
2944
2945         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2946                 return -EINVAL;
2947
2948         if (end < MAIN_BLKADDR(sbi))
2949                 goto out;
2950
2951         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2952                 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
2953                 return -EFSCORRUPTED;
2954         }
2955
2956         /* start/end segment number in main_area */
2957         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2958         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2959                                                 GET_SEGNO(sbi, end);
2960         if (need_align) {
2961                 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2962                 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2963         }
2964
2965         cpc.reason = CP_DISCARD;
2966         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2967         cpc.trim_start = start_segno;
2968         cpc.trim_end = end_segno;
2969
2970         if (sbi->discard_blks == 0)
2971                 goto out;
2972
2973         down_write(&sbi->gc_lock);
2974         err = f2fs_write_checkpoint(sbi, &cpc);
2975         up_write(&sbi->gc_lock);
2976         if (err)
2977                 goto out;
2978
2979         /*
2980          * We filed discard candidates, but actually we don't need to wait for
2981          * all of them, since they'll be issued in idle time along with runtime
2982          * discard option. User configuration looks like using runtime discard
2983          * or periodic fstrim instead of it.
2984          */
2985         if (f2fs_realtime_discard_enable(sbi))
2986                 goto out;
2987
2988         start_block = START_BLOCK(sbi, start_segno);
2989         end_block = START_BLOCK(sbi, end_segno + 1);
2990
2991         __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2992         trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2993                                         start_block, end_block);
2994
2995         trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
2996                                         start_block, end_block);
2997 out:
2998         if (!err)
2999                 range->len = F2FS_BLK_TO_BYTES(trimmed);
3000         return err;
3001 }
3002
3003 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
3004 {
3005         struct curseg_info *curseg = CURSEG_I(sbi, type);
3006
3007         return curseg->next_blkoff < f2fs_usable_blks_in_seg(sbi,
3008                                                         curseg->segno);
3009 }
3010
3011 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
3012 {
3013         switch (hint) {
3014         case WRITE_LIFE_SHORT:
3015                 return CURSEG_HOT_DATA;
3016         case WRITE_LIFE_EXTREME:
3017                 return CURSEG_COLD_DATA;
3018         default:
3019                 return CURSEG_WARM_DATA;
3020         }
3021 }
3022
3023 /* This returns write hints for each segment type. This hints will be
3024  * passed down to block layer. There are mapping tables which depend on
3025  * the mount option 'whint_mode'.
3026  *
3027  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
3028  *
3029  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
3030  *
3031  * User                  F2FS                     Block
3032  * ----                  ----                     -----
3033  *                       META                     WRITE_LIFE_NOT_SET
3034  *                       HOT_NODE                 "
3035  *                       WARM_NODE                "
3036  *                       COLD_NODE                "
3037  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
3038  * extension list        "                        "
3039  *
3040  * -- buffered io
3041  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3042  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3043  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
3044  * WRITE_LIFE_NONE       "                        "
3045  * WRITE_LIFE_MEDIUM     "                        "
3046  * WRITE_LIFE_LONG       "                        "
3047  *
3048  * -- direct io
3049  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3050  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3051  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
3052  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
3053  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
3054  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
3055  *
3056  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
3057  *
3058  * User                  F2FS                     Block
3059  * ----                  ----                     -----
3060  *                       META                     WRITE_LIFE_MEDIUM;
3061  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
3062  *                       WARM_NODE                "
3063  *                       COLD_NODE                WRITE_LIFE_NONE
3064  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
3065  * extension list        "                        "
3066  *
3067  * -- buffered io
3068  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3069  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3070  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
3071  * WRITE_LIFE_NONE       "                        "
3072  * WRITE_LIFE_MEDIUM     "                        "
3073  * WRITE_LIFE_LONG       "                        "
3074  *
3075  * -- direct io
3076  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3077  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3078  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
3079  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
3080  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
3081  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
3082  */
3083
3084 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
3085                                 enum page_type type, enum temp_type temp)
3086 {
3087         if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
3088                 if (type == DATA) {
3089                         if (temp == WARM)
3090                                 return WRITE_LIFE_NOT_SET;
3091                         else if (temp == HOT)
3092                                 return WRITE_LIFE_SHORT;
3093                         else if (temp == COLD)
3094                                 return WRITE_LIFE_EXTREME;
3095                 } else {
3096                         return WRITE_LIFE_NOT_SET;
3097                 }
3098         } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
3099                 if (type == DATA) {
3100                         if (temp == WARM)
3101                                 return WRITE_LIFE_LONG;
3102                         else if (temp == HOT)
3103                                 return WRITE_LIFE_SHORT;
3104                         else if (temp == COLD)
3105                                 return WRITE_LIFE_EXTREME;
3106                 } else if (type == NODE) {
3107                         if (temp == WARM || temp == HOT)
3108                                 return WRITE_LIFE_NOT_SET;
3109                         else if (temp == COLD)
3110                                 return WRITE_LIFE_NONE;
3111                 } else if (type == META) {
3112                         return WRITE_LIFE_MEDIUM;
3113                 }
3114         }
3115         return WRITE_LIFE_NOT_SET;
3116 }
3117
3118 static int __get_segment_type_2(struct f2fs_io_info *fio)
3119 {
3120         if (fio->type == DATA)
3121                 return CURSEG_HOT_DATA;
3122         else
3123                 return CURSEG_HOT_NODE;
3124 }
3125
3126 static int __get_segment_type_4(struct f2fs_io_info *fio)
3127 {
3128         if (fio->type == DATA) {
3129                 struct inode *inode = fio->page->mapping->host;
3130
3131                 if (S_ISDIR(inode->i_mode))
3132                         return CURSEG_HOT_DATA;
3133                 else
3134                         return CURSEG_COLD_DATA;
3135         } else {
3136                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3137                         return CURSEG_WARM_NODE;
3138                 else
3139                         return CURSEG_COLD_NODE;
3140         }
3141 }
3142
3143 static int __get_segment_type_6(struct f2fs_io_info *fio)
3144 {
3145         if (fio->type == DATA) {
3146                 struct inode *inode = fio->page->mapping->host;
3147
3148                 if (is_cold_data(fio->page) || file_is_cold(inode) ||
3149                                 f2fs_compressed_file(inode))
3150                         return CURSEG_COLD_DATA;
3151                 if (file_is_hot(inode) ||
3152                                 is_inode_flag_set(inode, FI_HOT_DATA) ||
3153                                 f2fs_is_atomic_file(inode) ||
3154                                 f2fs_is_volatile_file(inode))
3155                         return CURSEG_HOT_DATA;
3156                 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3157         } else {
3158                 if (IS_DNODE(fio->page))
3159                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3160                                                 CURSEG_HOT_NODE;
3161                 return CURSEG_COLD_NODE;
3162         }
3163 }
3164
3165 static int __get_segment_type(struct f2fs_io_info *fio)
3166 {
3167         int type = 0;
3168
3169         switch (F2FS_OPTION(fio->sbi).active_logs) {
3170         case 2:
3171                 type = __get_segment_type_2(fio);
3172                 break;
3173         case 4:
3174                 type = __get_segment_type_4(fio);
3175                 break;
3176         case 6:
3177                 type = __get_segment_type_6(fio);
3178                 break;
3179         default:
3180                 f2fs_bug_on(fio->sbi, true);
3181         }
3182
3183         if (IS_HOT(type))
3184                 fio->temp = HOT;
3185         else if (IS_WARM(type))
3186                 fio->temp = WARM;
3187         else
3188                 fio->temp = COLD;
3189         return type;
3190 }
3191
3192 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3193                 block_t old_blkaddr, block_t *new_blkaddr,
3194                 struct f2fs_summary *sum, int type,
3195                 struct f2fs_io_info *fio)
3196 {
3197         struct sit_info *sit_i = SIT_I(sbi);
3198         struct curseg_info *curseg = CURSEG_I(sbi, type);
3199
3200         down_read(&SM_I(sbi)->curseg_lock);
3201
3202         mutex_lock(&curseg->curseg_mutex);
3203         down_write(&sit_i->sentry_lock);
3204
3205         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3206
3207         f2fs_wait_discard_bio(sbi, *new_blkaddr);
3208
3209         /*
3210          * __add_sum_entry should be resided under the curseg_mutex
3211          * because, this function updates a summary entry in the
3212          * current summary block.
3213          */
3214         __add_sum_entry(sbi, type, sum);
3215
3216         __refresh_next_blkoff(sbi, curseg);
3217
3218         stat_inc_block_count(sbi, curseg);
3219
3220         /*
3221          * SIT information should be updated before segment allocation,
3222          * since SSR needs latest valid block information.
3223          */
3224         update_sit_entry(sbi, *new_blkaddr, 1);
3225         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3226                 update_sit_entry(sbi, old_blkaddr, -1);
3227
3228         if (!__has_curseg_space(sbi, type))
3229                 sit_i->s_ops->allocate_segment(sbi, type, false);
3230
3231         /*
3232          * segment dirty status should be updated after segment allocation,
3233          * so we just need to update status only one time after previous
3234          * segment being closed.
3235          */
3236         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3237         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3238
3239         up_write(&sit_i->sentry_lock);
3240
3241         if (page && IS_NODESEG(type)) {
3242                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3243
3244                 f2fs_inode_chksum_set(sbi, page);
3245         }
3246
3247         if (F2FS_IO_ALIGNED(sbi))
3248                 fio->retry = false;
3249
3250         if (fio) {
3251                 struct f2fs_bio_info *io;
3252
3253                 INIT_LIST_HEAD(&fio->list);
3254                 fio->in_list = true;
3255                 io = sbi->write_io[fio->type] + fio->temp;
3256                 spin_lock(&io->io_lock);
3257                 list_add_tail(&fio->list, &io->io_list);
3258                 spin_unlock(&io->io_lock);
3259         }
3260
3261         mutex_unlock(&curseg->curseg_mutex);
3262
3263         up_read(&SM_I(sbi)->curseg_lock);
3264 }
3265
3266 static void update_device_state(struct f2fs_io_info *fio)
3267 {
3268         struct f2fs_sb_info *sbi = fio->sbi;
3269         unsigned int devidx;
3270
3271         if (!f2fs_is_multi_device(sbi))
3272                 return;
3273
3274         devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3275
3276         /* update device state for fsync */
3277         f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
3278
3279         /* update device state for checkpoint */
3280         if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3281                 spin_lock(&sbi->dev_lock);
3282                 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3283                 spin_unlock(&sbi->dev_lock);
3284         }
3285 }
3286
3287 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3288 {
3289         int type = __get_segment_type(fio);
3290         bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3291
3292         if (keep_order)
3293                 down_read(&fio->sbi->io_order_lock);
3294 reallocate:
3295         f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3296                         &fio->new_blkaddr, sum, type, fio);
3297         if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3298                 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3299                                         fio->old_blkaddr, fio->old_blkaddr);
3300
3301         /* writeout dirty page into bdev */
3302         f2fs_submit_page_write(fio);
3303         if (fio->retry) {
3304                 fio->old_blkaddr = fio->new_blkaddr;
3305                 goto reallocate;
3306         }
3307
3308         update_device_state(fio);
3309
3310         if (keep_order)
3311                 up_read(&fio->sbi->io_order_lock);
3312 }
3313
3314 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3315                                         enum iostat_type io_type)
3316 {
3317         struct f2fs_io_info fio = {
3318                 .sbi = sbi,
3319                 .type = META,
3320                 .temp = HOT,
3321                 .op = REQ_OP_WRITE,
3322                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3323                 .old_blkaddr = page->index,
3324                 .new_blkaddr = page->index,
3325                 .page = page,
3326                 .encrypted_page = NULL,
3327                 .in_list = false,
3328         };
3329
3330         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3331                 fio.op_flags &= ~REQ_META;
3332
3333         set_page_writeback(page);
3334         ClearPageError(page);
3335         f2fs_submit_page_write(&fio);
3336
3337         stat_inc_meta_count(sbi, page->index);
3338         f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3339 }
3340
3341 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3342 {
3343         struct f2fs_summary sum;
3344
3345         set_summary(&sum, nid, 0, 0);
3346         do_write_page(&sum, fio);
3347
3348         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3349 }
3350
3351 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3352                                         struct f2fs_io_info *fio)
3353 {
3354         struct f2fs_sb_info *sbi = fio->sbi;
3355         struct f2fs_summary sum;
3356
3357         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3358         set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3359         do_write_page(&sum, fio);
3360         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3361
3362         f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3363 }
3364
3365 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3366 {
3367         int err;
3368         struct f2fs_sb_info *sbi = fio->sbi;
3369         unsigned int segno;
3370
3371         fio->new_blkaddr = fio->old_blkaddr;
3372         /* i/o temperature is needed for passing down write hints */
3373         __get_segment_type(fio);
3374
3375         segno = GET_SEGNO(sbi, fio->new_blkaddr);
3376
3377         if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3378                 set_sbi_flag(sbi, SBI_NEED_FSCK);
3379                 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3380                           __func__, segno);
3381                 return -EFSCORRUPTED;
3382         }
3383
3384         stat_inc_inplace_blocks(fio->sbi);
3385
3386         if (fio->bio && !(SM_I(sbi)->ipu_policy & (1 << F2FS_IPU_NOCACHE)))
3387                 err = f2fs_merge_page_bio(fio);
3388         else
3389                 err = f2fs_submit_page_bio(fio);
3390         if (!err) {
3391                 update_device_state(fio);
3392                 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3393         }
3394
3395         return err;
3396 }
3397
3398 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3399                                                 unsigned int segno)
3400 {
3401         int i;
3402
3403         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3404                 if (CURSEG_I(sbi, i)->segno == segno)
3405                         break;
3406         }
3407         return i;
3408 }
3409
3410 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3411                                 block_t old_blkaddr, block_t new_blkaddr,
3412                                 bool recover_curseg, bool recover_newaddr)
3413 {
3414         struct sit_info *sit_i = SIT_I(sbi);
3415         struct curseg_info *curseg;
3416         unsigned int segno, old_cursegno;
3417         struct seg_entry *se;
3418         int type;
3419         unsigned short old_blkoff;
3420
3421         segno = GET_SEGNO(sbi, new_blkaddr);
3422         se = get_seg_entry(sbi, segno);
3423         type = se->type;
3424
3425         down_write(&SM_I(sbi)->curseg_lock);
3426
3427         if (!recover_curseg) {
3428                 /* for recovery flow */
3429                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3430                         if (old_blkaddr == NULL_ADDR)
3431                                 type = CURSEG_COLD_DATA;
3432                         else
3433                                 type = CURSEG_WARM_DATA;
3434                 }
3435         } else {
3436                 if (IS_CURSEG(sbi, segno)) {
3437                         /* se->type is volatile as SSR allocation */
3438                         type = __f2fs_get_curseg(sbi, segno);
3439                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3440                 } else {
3441                         type = CURSEG_WARM_DATA;
3442                 }
3443         }
3444
3445         f2fs_bug_on(sbi, !IS_DATASEG(type));
3446         curseg = CURSEG_I(sbi, type);
3447
3448         mutex_lock(&curseg->curseg_mutex);
3449         down_write(&sit_i->sentry_lock);
3450
3451         old_cursegno = curseg->segno;
3452         old_blkoff = curseg->next_blkoff;
3453
3454         /* change the current segment */
3455         if (segno != curseg->segno) {
3456                 curseg->next_segno = segno;
3457                 change_curseg(sbi, type);
3458         }
3459
3460         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3461         __add_sum_entry(sbi, type, sum);
3462
3463         if (!recover_curseg || recover_newaddr)
3464                 update_sit_entry(sbi, new_blkaddr, 1);
3465         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3466                 invalidate_mapping_pages(META_MAPPING(sbi),
3467                                         old_blkaddr, old_blkaddr);
3468                 update_sit_entry(sbi, old_blkaddr, -1);
3469         }
3470
3471         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3472         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3473
3474         locate_dirty_segment(sbi, old_cursegno);
3475
3476         if (recover_curseg) {
3477                 if (old_cursegno != curseg->segno) {
3478                         curseg->next_segno = old_cursegno;
3479                         change_curseg(sbi, type);
3480                 }
3481                 curseg->next_blkoff = old_blkoff;
3482         }
3483
3484         up_write(&sit_i->sentry_lock);
3485         mutex_unlock(&curseg->curseg_mutex);
3486         up_write(&SM_I(sbi)->curseg_lock);
3487 }
3488
3489 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3490                                 block_t old_addr, block_t new_addr,
3491                                 unsigned char version, bool recover_curseg,
3492                                 bool recover_newaddr)
3493 {
3494         struct f2fs_summary sum;
3495
3496         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3497
3498         f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3499                                         recover_curseg, recover_newaddr);
3500
3501         f2fs_update_data_blkaddr(dn, new_addr);
3502 }
3503
3504 void f2fs_wait_on_page_writeback(struct page *page,
3505                                 enum page_type type, bool ordered, bool locked)
3506 {
3507         if (PageWriteback(page)) {
3508                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3509
3510                 /* submit cached LFS IO */
3511                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3512                 /* sbumit cached IPU IO */
3513                 f2fs_submit_merged_ipu_write(sbi, NULL, page);
3514                 if (ordered) {
3515                         wait_on_page_writeback(page);
3516                         f2fs_bug_on(sbi, locked && PageWriteback(page));
3517                 } else {
3518                         wait_for_stable_page(page);
3519                 }
3520         }
3521 }
3522
3523 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3524 {
3525         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3526         struct page *cpage;
3527
3528         if (!f2fs_post_read_required(inode))
3529                 return;
3530
3531         if (!__is_valid_data_blkaddr(blkaddr))
3532                 return;
3533
3534         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3535         if (cpage) {
3536                 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3537                 f2fs_put_page(cpage, 1);
3538         }
3539 }
3540
3541 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3542                                                                 block_t len)
3543 {
3544         block_t i;
3545
3546         for (i = 0; i < len; i++)
3547                 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3548 }
3549
3550 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3551 {
3552         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3553         struct curseg_info *seg_i;
3554         unsigned char *kaddr;
3555         struct page *page;
3556         block_t start;
3557         int i, j, offset;
3558
3559         start = start_sum_block(sbi);
3560
3561         page = f2fs_get_meta_page(sbi, start++);
3562         if (IS_ERR(page))
3563                 return PTR_ERR(page);
3564         kaddr = (unsigned char *)page_address(page);
3565
3566         /* Step 1: restore nat cache */
3567         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3568         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3569
3570         /* Step 2: restore sit cache */
3571         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3572         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3573         offset = 2 * SUM_JOURNAL_SIZE;
3574
3575         /* Step 3: restore summary entries */
3576         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3577                 unsigned short blk_off;
3578                 unsigned int segno;
3579
3580                 seg_i = CURSEG_I(sbi, i);
3581                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3582                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3583                 seg_i->next_segno = segno;
3584                 reset_curseg(sbi, i, 0);
3585                 seg_i->alloc_type = ckpt->alloc_type[i];
3586                 seg_i->next_blkoff = blk_off;
3587
3588                 if (seg_i->alloc_type == SSR)
3589                         blk_off = sbi->blocks_per_seg;
3590
3591                 for (j = 0; j < blk_off; j++) {
3592                         struct f2fs_summary *s;
3593                         s = (struct f2fs_summary *)(kaddr + offset);
3594                         seg_i->sum_blk->entries[j] = *s;
3595                         offset += SUMMARY_SIZE;
3596                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3597                                                 SUM_FOOTER_SIZE)
3598                                 continue;
3599
3600                         f2fs_put_page(page, 1);
3601                         page = NULL;
3602
3603                         page = f2fs_get_meta_page(sbi, start++);
3604                         if (IS_ERR(page))
3605                                 return PTR_ERR(page);
3606                         kaddr = (unsigned char *)page_address(page);
3607                         offset = 0;
3608                 }
3609         }
3610         f2fs_put_page(page, 1);
3611         return 0;
3612 }
3613
3614 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3615 {
3616         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3617         struct f2fs_summary_block *sum;
3618         struct curseg_info *curseg;
3619         struct page *new;
3620         unsigned short blk_off;
3621         unsigned int segno = 0;
3622         block_t blk_addr = 0;
3623         int err = 0;
3624
3625         /* get segment number and block addr */
3626         if (IS_DATASEG(type)) {
3627                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3628                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3629                                                         CURSEG_HOT_DATA]);
3630                 if (__exist_node_summaries(sbi))
3631                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
3632                 else
3633                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3634         } else {
3635                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3636                                                         CURSEG_HOT_NODE]);
3637                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3638                                                         CURSEG_HOT_NODE]);
3639                 if (__exist_node_summaries(sbi))
3640                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3641                                                         type - CURSEG_HOT_NODE);
3642                 else
3643                         blk_addr = GET_SUM_BLOCK(sbi, segno);
3644         }
3645
3646         new = f2fs_get_meta_page(sbi, blk_addr);
3647         if (IS_ERR(new))
3648                 return PTR_ERR(new);
3649         sum = (struct f2fs_summary_block *)page_address(new);
3650
3651         if (IS_NODESEG(type)) {
3652                 if (__exist_node_summaries(sbi)) {
3653                         struct f2fs_summary *ns = &sum->entries[0];
3654                         int i;
3655                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3656                                 ns->version = 0;
3657                                 ns->ofs_in_node = 0;
3658                         }
3659                 } else {
3660                         err = f2fs_restore_node_summary(sbi, segno, sum);
3661                         if (err)
3662                                 goto out;
3663                 }
3664         }
3665
3666         /* set uncompleted segment to curseg */
3667         curseg = CURSEG_I(sbi, type);
3668         mutex_lock(&curseg->curseg_mutex);
3669
3670         /* update journal info */
3671         down_write(&curseg->journal_rwsem);
3672         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3673         up_write(&curseg->journal_rwsem);
3674
3675         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3676         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3677         curseg->next_segno = segno;
3678         reset_curseg(sbi, type, 0);
3679         curseg->alloc_type = ckpt->alloc_type[type];
3680         curseg->next_blkoff = blk_off;
3681         mutex_unlock(&curseg->curseg_mutex);
3682 out:
3683         f2fs_put_page(new, 1);
3684         return err;
3685 }
3686
3687 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3688 {
3689         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3690         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3691         int type = CURSEG_HOT_DATA;
3692         int err;
3693
3694         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3695                 int npages = f2fs_npages_for_summary_flush(sbi, true);
3696
3697                 if (npages >= 2)
3698                         f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3699                                                         META_CP, true);
3700
3701                 /* restore for compacted data summary */
3702                 err = read_compacted_summaries(sbi);
3703                 if (err)
3704                         return err;
3705                 type = CURSEG_HOT_NODE;
3706         }
3707
3708         if (__exist_node_summaries(sbi))
3709                 f2fs_ra_meta_pages(sbi,
3710                                 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
3711                                 NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
3712
3713         for (; type <= CURSEG_COLD_NODE; type++) {
3714                 err = read_normal_summaries(sbi, type);
3715                 if (err)
3716                         return err;
3717         }
3718
3719         /* sanity check for summary blocks */
3720         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3721                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3722                 f2fs_err(sbi, "invalid journal entries nats %u sits %u\n",
3723                          nats_in_cursum(nat_j), sits_in_cursum(sit_j));
3724                 return -EINVAL;
3725         }
3726
3727         return 0;
3728 }
3729
3730 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3731 {
3732         struct page *page;
3733         unsigned char *kaddr;
3734         struct f2fs_summary *summary;
3735         struct curseg_info *seg_i;
3736         int written_size = 0;
3737         int i, j;
3738
3739         page = f2fs_grab_meta_page(sbi, blkaddr++);
3740         kaddr = (unsigned char *)page_address(page);
3741         memset(kaddr, 0, PAGE_SIZE);
3742
3743         /* Step 1: write nat cache */
3744         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3745         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3746         written_size += SUM_JOURNAL_SIZE;
3747
3748         /* Step 2: write sit cache */
3749         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3750         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3751         written_size += SUM_JOURNAL_SIZE;
3752
3753         /* Step 3: write summary entries */
3754         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3755                 unsigned short blkoff;
3756                 seg_i = CURSEG_I(sbi, i);
3757                 if (sbi->ckpt->alloc_type[i] == SSR)
3758                         blkoff = sbi->blocks_per_seg;
3759                 else
3760                         blkoff = curseg_blkoff(sbi, i);
3761
3762                 for (j = 0; j < blkoff; j++) {
3763                         if (!page) {
3764                                 page = f2fs_grab_meta_page(sbi, blkaddr++);
3765                                 kaddr = (unsigned char *)page_address(page);
3766                                 memset(kaddr, 0, PAGE_SIZE);
3767                                 written_size = 0;
3768                         }
3769                         summary = (struct f2fs_summary *)(kaddr + written_size);
3770                         *summary = seg_i->sum_blk->entries[j];
3771                         written_size += SUMMARY_SIZE;
3772
3773                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3774                                                         SUM_FOOTER_SIZE)
3775                                 continue;
3776
3777                         set_page_dirty(page);
3778                         f2fs_put_page(page, 1);
3779                         page = NULL;
3780                 }
3781         }
3782         if (page) {
3783                 set_page_dirty(page);
3784                 f2fs_put_page(page, 1);
3785         }
3786 }
3787
3788 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3789                                         block_t blkaddr, int type)
3790 {
3791         int i, end;
3792         if (IS_DATASEG(type))
3793                 end = type + NR_CURSEG_DATA_TYPE;
3794         else
3795                 end = type + NR_CURSEG_NODE_TYPE;
3796
3797         for (i = type; i < end; i++)
3798                 write_current_sum_page(sbi, i, blkaddr + (i - type));
3799 }
3800
3801 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3802 {
3803         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3804                 write_compacted_summaries(sbi, start_blk);
3805         else
3806                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3807 }
3808
3809 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3810 {
3811         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3812 }
3813
3814 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3815                                         unsigned int val, int alloc)
3816 {
3817         int i;
3818
3819         if (type == NAT_JOURNAL) {
3820                 for (i = 0; i < nats_in_cursum(journal); i++) {
3821                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3822                                 return i;
3823                 }
3824                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3825                         return update_nats_in_cursum(journal, 1);
3826         } else if (type == SIT_JOURNAL) {
3827                 for (i = 0; i < sits_in_cursum(journal); i++)
3828                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3829                                 return i;
3830                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3831                         return update_sits_in_cursum(journal, 1);
3832         }
3833         return -1;
3834 }
3835
3836 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3837                                         unsigned int segno)
3838 {
3839         return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
3840 }
3841
3842 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3843                                         unsigned int start)
3844 {
3845         struct sit_info *sit_i = SIT_I(sbi);
3846         struct page *page;
3847         pgoff_t src_off, dst_off;
3848
3849         src_off = current_sit_addr(sbi, start);
3850         dst_off = next_sit_addr(sbi, src_off);
3851
3852         page = f2fs_grab_meta_page(sbi, dst_off);
3853         seg_info_to_sit_page(sbi, page, start);
3854
3855         set_page_dirty(page);
3856         set_to_next_sit(sit_i, start);
3857
3858         return page;
3859 }
3860
3861 static struct sit_entry_set *grab_sit_entry_set(void)
3862 {
3863         struct sit_entry_set *ses =
3864                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3865
3866         ses->entry_cnt = 0;
3867         INIT_LIST_HEAD(&ses->set_list);
3868         return ses;
3869 }
3870
3871 static void release_sit_entry_set(struct sit_entry_set *ses)
3872 {
3873         list_del(&ses->set_list);
3874         kmem_cache_free(sit_entry_set_slab, ses);
3875 }
3876
3877 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3878                                                 struct list_head *head)
3879 {
3880         struct sit_entry_set *next = ses;
3881
3882         if (list_is_last(&ses->set_list, head))
3883                 return;
3884
3885         list_for_each_entry_continue(next, head, set_list)
3886                 if (ses->entry_cnt <= next->entry_cnt)
3887                         break;
3888
3889         list_move_tail(&ses->set_list, &next->set_list);
3890 }
3891
3892 static void add_sit_entry(unsigned int segno, struct list_head *head)
3893 {
3894         struct sit_entry_set *ses;
3895         unsigned int start_segno = START_SEGNO(segno);
3896
3897         list_for_each_entry(ses, head, set_list) {
3898                 if (ses->start_segno == start_segno) {
3899                         ses->entry_cnt++;
3900                         adjust_sit_entry_set(ses, head);
3901                         return;
3902                 }
3903         }
3904
3905         ses = grab_sit_entry_set();
3906
3907         ses->start_segno = start_segno;
3908         ses->entry_cnt++;
3909         list_add(&ses->set_list, head);
3910 }
3911
3912 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3913 {
3914         struct f2fs_sm_info *sm_info = SM_I(sbi);
3915         struct list_head *set_list = &sm_info->sit_entry_set;
3916         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3917         unsigned int segno;
3918
3919         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3920                 add_sit_entry(segno, set_list);
3921 }
3922
3923 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3924 {
3925         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3926         struct f2fs_journal *journal = curseg->journal;
3927         int i;
3928
3929         down_write(&curseg->journal_rwsem);
3930         for (i = 0; i < sits_in_cursum(journal); i++) {
3931                 unsigned int segno;
3932                 bool dirtied;
3933
3934                 segno = le32_to_cpu(segno_in_journal(journal, i));
3935                 dirtied = __mark_sit_entry_dirty(sbi, segno);
3936
3937                 if (!dirtied)
3938                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3939         }
3940         update_sits_in_cursum(journal, -i);
3941         up_write(&curseg->journal_rwsem);
3942 }
3943
3944 /*
3945  * CP calls this function, which flushes SIT entries including sit_journal,
3946  * and moves prefree segs to free segs.
3947  */
3948 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3949 {
3950         struct sit_info *sit_i = SIT_I(sbi);
3951         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3952         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3953         struct f2fs_journal *journal = curseg->journal;
3954         struct sit_entry_set *ses, *tmp;
3955         struct list_head *head = &SM_I(sbi)->sit_entry_set;
3956         bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
3957         struct seg_entry *se;
3958
3959         down_write(&sit_i->sentry_lock);
3960
3961         if (!sit_i->dirty_sentries)
3962                 goto out;
3963
3964         /*
3965          * add and account sit entries of dirty bitmap in sit entry
3966          * set temporarily
3967          */
3968         add_sits_in_set(sbi);
3969
3970         /*
3971          * if there are no enough space in journal to store dirty sit
3972          * entries, remove all entries from journal and add and account
3973          * them in sit entry set.
3974          */
3975         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
3976                                                                 !to_journal)
3977                 remove_sits_in_journal(sbi);
3978
3979         /*
3980          * there are two steps to flush sit entries:
3981          * #1, flush sit entries to journal in current cold data summary block.
3982          * #2, flush sit entries to sit page.
3983          */
3984         list_for_each_entry_safe(ses, tmp, head, set_list) {
3985                 struct page *page = NULL;
3986                 struct f2fs_sit_block *raw_sit = NULL;
3987                 unsigned int start_segno = ses->start_segno;
3988                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3989                                                 (unsigned long)MAIN_SEGS(sbi));
3990                 unsigned int segno = start_segno;
3991
3992                 if (to_journal &&
3993                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3994                         to_journal = false;
3995
3996                 if (to_journal) {
3997                         down_write(&curseg->journal_rwsem);
3998                 } else {
3999                         page = get_next_sit_page(sbi, start_segno);
4000                         raw_sit = page_address(page);
4001                 }
4002
4003                 /* flush dirty sit entries in region of current sit set */
4004                 for_each_set_bit_from(segno, bitmap, end) {
4005                         int offset, sit_offset;
4006
4007                         se = get_seg_entry(sbi, segno);
4008 #ifdef CONFIG_F2FS_CHECK_FS
4009                         if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4010                                                 SIT_VBLOCK_MAP_SIZE))
4011                                 f2fs_bug_on(sbi, 1);
4012 #endif
4013
4014                         /* add discard candidates */
4015                         if (!(cpc->reason & CP_DISCARD)) {
4016                                 cpc->trim_start = segno;
4017                                 add_discard_addrs(sbi, cpc, false);
4018                         }
4019
4020                         if (to_journal) {
4021                                 offset = f2fs_lookup_journal_in_cursum(journal,
4022                                                         SIT_JOURNAL, segno, 1);
4023                                 f2fs_bug_on(sbi, offset < 0);
4024                                 segno_in_journal(journal, offset) =
4025                                                         cpu_to_le32(segno);
4026                                 seg_info_to_raw_sit(se,
4027                                         &sit_in_journal(journal, offset));
4028                                 check_block_count(sbi, segno,
4029                                         &sit_in_journal(journal, offset));
4030                         } else {
4031                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4032                                 seg_info_to_raw_sit(se,
4033                                                 &raw_sit->entries[sit_offset]);
4034                                 check_block_count(sbi, segno,
4035                                                 &raw_sit->entries[sit_offset]);
4036                         }
4037
4038                         __clear_bit(segno, bitmap);
4039                         sit_i->dirty_sentries--;
4040                         ses->entry_cnt--;
4041                 }
4042
4043                 if (to_journal)
4044                         up_write(&curseg->journal_rwsem);
4045                 else
4046                         f2fs_put_page(page, 1);
4047
4048                 f2fs_bug_on(sbi, ses->entry_cnt);
4049                 release_sit_entry_set(ses);
4050         }
4051
4052         f2fs_bug_on(sbi, !list_empty(head));
4053         f2fs_bug_on(sbi, sit_i->dirty_sentries);
4054 out:
4055         if (cpc->reason & CP_DISCARD) {
4056                 __u64 trim_start = cpc->trim_start;
4057
4058                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4059                         add_discard_addrs(sbi, cpc, false);
4060
4061                 cpc->trim_start = trim_start;
4062         }
4063         up_write(&sit_i->sentry_lock);
4064
4065         set_prefree_as_free_segments(sbi);
4066 }
4067
4068 static int build_sit_info(struct f2fs_sb_info *sbi)
4069 {
4070         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4071         struct sit_info *sit_i;
4072         unsigned int sit_segs, start;
4073         char *src_bitmap, *bitmap;
4074         unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4075
4076         /* allocate memory for SIT information */
4077         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4078         if (!sit_i)
4079                 return -ENOMEM;
4080
4081         SM_I(sbi)->sit_info = sit_i;
4082
4083         sit_i->sentries =
4084                 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4085                                               MAIN_SEGS(sbi)),
4086                               GFP_KERNEL);
4087         if (!sit_i->sentries)
4088                 return -ENOMEM;
4089
4090         main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4091         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4092                                                                 GFP_KERNEL);
4093         if (!sit_i->dirty_sentries_bitmap)
4094                 return -ENOMEM;
4095
4096 #ifdef CONFIG_F2FS_CHECK_FS
4097         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 4;
4098 #else
4099         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 3;
4100 #endif
4101         sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4102         if (!sit_i->bitmap)
4103                 return -ENOMEM;
4104
4105         bitmap = sit_i->bitmap;
4106
4107         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4108                 sit_i->sentries[start].cur_valid_map = bitmap;
4109                 bitmap += SIT_VBLOCK_MAP_SIZE;
4110
4111                 sit_i->sentries[start].ckpt_valid_map = bitmap;
4112                 bitmap += SIT_VBLOCK_MAP_SIZE;
4113
4114 #ifdef CONFIG_F2FS_CHECK_FS
4115                 sit_i->sentries[start].cur_valid_map_mir = bitmap;
4116                 bitmap += SIT_VBLOCK_MAP_SIZE;
4117 #endif
4118
4119                 sit_i->sentries[start].discard_map = bitmap;
4120                 bitmap += SIT_VBLOCK_MAP_SIZE;
4121         }
4122
4123         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4124         if (!sit_i->tmp_map)
4125                 return -ENOMEM;
4126
4127         if (__is_large_section(sbi)) {
4128                 sit_i->sec_entries =
4129                         f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4130                                                       MAIN_SECS(sbi)),
4131                                       GFP_KERNEL);
4132                 if (!sit_i->sec_entries)
4133                         return -ENOMEM;
4134         }
4135
4136         /* get information related with SIT */
4137         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4138
4139         /* setup SIT bitmap from ckeckpoint pack */
4140         sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4141         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4142
4143         sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4144         if (!sit_i->sit_bitmap)
4145                 return -ENOMEM;
4146
4147 #ifdef CONFIG_F2FS_CHECK_FS
4148         sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4149                                         sit_bitmap_size, GFP_KERNEL);
4150         if (!sit_i->sit_bitmap_mir)
4151                 return -ENOMEM;
4152
4153         sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4154                                         main_bitmap_size, GFP_KERNEL);
4155         if (!sit_i->invalid_segmap)
4156                 return -ENOMEM;
4157 #endif
4158
4159         /* init SIT information */
4160         sit_i->s_ops = &default_salloc_ops;
4161
4162         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4163         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4164         sit_i->written_valid_blocks = 0;
4165         sit_i->bitmap_size = sit_bitmap_size;
4166         sit_i->dirty_sentries = 0;
4167         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4168         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4169         sit_i->mounted_time = ktime_get_boottime_seconds();
4170         init_rwsem(&sit_i->sentry_lock);
4171         return 0;
4172 }
4173
4174 static int build_free_segmap(struct f2fs_sb_info *sbi)
4175 {
4176         struct free_segmap_info *free_i;
4177         unsigned int bitmap_size, sec_bitmap_size;
4178
4179         /* allocate memory for free segmap information */
4180         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4181         if (!free_i)
4182                 return -ENOMEM;
4183
4184         SM_I(sbi)->free_info = free_i;
4185
4186         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4187         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4188         if (!free_i->free_segmap)
4189                 return -ENOMEM;
4190
4191         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4192         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4193         if (!free_i->free_secmap)
4194                 return -ENOMEM;
4195
4196         /* set all segments as dirty temporarily */
4197         memset(free_i->free_segmap, 0xff, bitmap_size);
4198         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4199
4200         /* init free segmap information */
4201         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4202         free_i->free_segments = 0;
4203         free_i->free_sections = 0;
4204         spin_lock_init(&free_i->segmap_lock);
4205         return 0;
4206 }
4207
4208 static int build_curseg(struct f2fs_sb_info *sbi)
4209 {
4210         struct curseg_info *array;
4211         int i;
4212
4213         array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4214                                         sizeof(*array)), GFP_KERNEL);
4215         if (!array)
4216                 return -ENOMEM;
4217
4218         SM_I(sbi)->curseg_array = array;
4219
4220         for (i = 0; i < NO_CHECK_TYPE; i++) {
4221                 mutex_init(&array[i].curseg_mutex);
4222                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4223                 if (!array[i].sum_blk)
4224                         return -ENOMEM;
4225                 init_rwsem(&array[i].journal_rwsem);
4226                 array[i].journal = f2fs_kzalloc(sbi,
4227                                 sizeof(struct f2fs_journal), GFP_KERNEL);
4228                 if (!array[i].journal)
4229                         return -ENOMEM;
4230                 if (i < NR_PERSISTENT_LOG)
4231                         array[i].seg_type = CURSEG_HOT_DATA + i;
4232                 else if (i == CURSEG_COLD_DATA_PINNED)
4233                         array[i].seg_type = CURSEG_COLD_DATA;
4234                 array[i].segno = NULL_SEGNO;
4235                 array[i].next_blkoff = 0;
4236                 array[i].inited = false;
4237         }
4238         return restore_curseg_summaries(sbi);
4239 }
4240
4241 static int build_sit_entries(struct f2fs_sb_info *sbi)
4242 {
4243         struct sit_info *sit_i = SIT_I(sbi);
4244         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4245         struct f2fs_journal *journal = curseg->journal;
4246         struct seg_entry *se;
4247         struct f2fs_sit_entry sit;
4248         int sit_blk_cnt = SIT_BLK_CNT(sbi);
4249         unsigned int i, start, end;
4250         unsigned int readed, start_blk = 0;
4251         int err = 0;
4252         block_t total_node_blocks = 0;
4253
4254         do {
4255                 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
4256                                                         META_SIT, true);
4257
4258                 start = start_blk * sit_i->sents_per_block;
4259                 end = (start_blk + readed) * sit_i->sents_per_block;
4260
4261                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4262                         struct f2fs_sit_block *sit_blk;
4263                         struct page *page;
4264
4265                         se = &sit_i->sentries[start];
4266                         page = get_current_sit_page(sbi, start);
4267                         if (IS_ERR(page))
4268                                 return PTR_ERR(page);
4269                         sit_blk = (struct f2fs_sit_block *)page_address(page);
4270                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4271                         f2fs_put_page(page, 1);
4272
4273                         err = check_block_count(sbi, start, &sit);
4274                         if (err)
4275                                 return err;
4276                         seg_info_from_raw_sit(se, &sit);
4277                         if (IS_NODESEG(se->type))
4278                                 total_node_blocks += se->valid_blocks;
4279
4280                         /* build discard map only one time */
4281                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4282                                 memset(se->discard_map, 0xff,
4283                                         SIT_VBLOCK_MAP_SIZE);
4284                         } else {
4285                                 memcpy(se->discard_map,
4286                                         se->cur_valid_map,
4287                                         SIT_VBLOCK_MAP_SIZE);
4288                                 sbi->discard_blks +=
4289                                         sbi->blocks_per_seg -
4290                                         se->valid_blocks;
4291                         }
4292
4293                         if (__is_large_section(sbi))
4294                                 get_sec_entry(sbi, start)->valid_blocks +=
4295                                                         se->valid_blocks;
4296                 }
4297                 start_blk += readed;
4298         } while (start_blk < sit_blk_cnt);
4299
4300         down_read(&curseg->journal_rwsem);
4301         for (i = 0; i < sits_in_cursum(journal); i++) {
4302                 unsigned int old_valid_blocks;
4303
4304                 start = le32_to_cpu(segno_in_journal(journal, i));
4305                 if (start >= MAIN_SEGS(sbi)) {
4306                         f2fs_err(sbi, "Wrong journal entry on segno %u",
4307                                  start);
4308                         err = -EFSCORRUPTED;
4309                         break;
4310                 }
4311
4312                 se = &sit_i->sentries[start];
4313                 sit = sit_in_journal(journal, i);
4314
4315                 old_valid_blocks = se->valid_blocks;
4316                 if (IS_NODESEG(se->type))
4317                         total_node_blocks -= old_valid_blocks;
4318
4319                 err = check_block_count(sbi, start, &sit);
4320                 if (err)
4321                         break;
4322                 seg_info_from_raw_sit(se, &sit);
4323                 if (IS_NODESEG(se->type))
4324                         total_node_blocks += se->valid_blocks;
4325
4326                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4327                         memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4328                 } else {
4329                         memcpy(se->discard_map, se->cur_valid_map,
4330                                                 SIT_VBLOCK_MAP_SIZE);
4331                         sbi->discard_blks += old_valid_blocks;
4332                         sbi->discard_blks -= se->valid_blocks;
4333                 }
4334
4335                 if (__is_large_section(sbi)) {
4336                         get_sec_entry(sbi, start)->valid_blocks +=
4337                                                         se->valid_blocks;
4338                         get_sec_entry(sbi, start)->valid_blocks -=
4339                                                         old_valid_blocks;
4340                 }
4341         }
4342         up_read(&curseg->journal_rwsem);
4343
4344         if (!err && total_node_blocks != valid_node_count(sbi)) {
4345                 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4346                          total_node_blocks, valid_node_count(sbi));
4347                 err = -EFSCORRUPTED;
4348         }
4349
4350         return err;
4351 }
4352
4353 static void init_free_segmap(struct f2fs_sb_info *sbi)
4354 {
4355         unsigned int start;
4356         int type;
4357         struct seg_entry *sentry;
4358
4359         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4360                 if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4361                         continue;
4362                 sentry = get_seg_entry(sbi, start);
4363                 if (!sentry->valid_blocks)
4364                         __set_free(sbi, start);
4365                 else
4366                         SIT_I(sbi)->written_valid_blocks +=
4367                                                 sentry->valid_blocks;
4368         }
4369
4370         /* set use the current segments */
4371         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4372                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4373                 __set_test_and_inuse(sbi, curseg_t->segno);
4374         }
4375 }
4376
4377 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4378 {
4379         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4380         struct free_segmap_info *free_i = FREE_I(sbi);
4381         unsigned int segno = 0, offset = 0, secno;
4382         block_t valid_blocks, usable_blks_in_seg;
4383         block_t blks_per_sec = BLKS_PER_SEC(sbi);
4384
4385         while (1) {
4386                 /* find dirty segment based on free segmap */
4387                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4388                 if (segno >= MAIN_SEGS(sbi))
4389                         break;
4390                 offset = segno + 1;
4391                 valid_blocks = get_valid_blocks(sbi, segno, false);
4392                 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4393                 if (valid_blocks == usable_blks_in_seg || !valid_blocks)
4394                         continue;
4395                 if (valid_blocks > usable_blks_in_seg) {
4396                         f2fs_bug_on(sbi, 1);
4397                         continue;
4398                 }
4399                 mutex_lock(&dirty_i->seglist_lock);
4400                 __locate_dirty_segment(sbi, segno, DIRTY);
4401                 mutex_unlock(&dirty_i->seglist_lock);
4402         }
4403
4404         if (!__is_large_section(sbi))
4405                 return;
4406
4407         mutex_lock(&dirty_i->seglist_lock);
4408         for (segno = 0; segno < MAIN_SECS(sbi); segno += blks_per_sec) {
4409                 valid_blocks = get_valid_blocks(sbi, segno, true);
4410                 secno = GET_SEC_FROM_SEG(sbi, segno);
4411
4412                 if (!valid_blocks || valid_blocks == blks_per_sec)
4413                         continue;
4414                 if (IS_CURSEC(sbi, secno))
4415                         continue;
4416                 set_bit(secno, dirty_i->dirty_secmap);
4417         }
4418         mutex_unlock(&dirty_i->seglist_lock);
4419 }
4420
4421 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4422 {
4423         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4424         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4425
4426         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4427         if (!dirty_i->victim_secmap)
4428                 return -ENOMEM;
4429         return 0;
4430 }
4431
4432 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4433 {
4434         struct dirty_seglist_info *dirty_i;
4435         unsigned int bitmap_size, i;
4436
4437         /* allocate memory for dirty segments list information */
4438         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4439                                                                 GFP_KERNEL);
4440         if (!dirty_i)
4441                 return -ENOMEM;
4442
4443         SM_I(sbi)->dirty_info = dirty_i;
4444         mutex_init(&dirty_i->seglist_lock);
4445
4446         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4447
4448         for (i = 0; i < NR_DIRTY_TYPE; i++) {
4449                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4450                                                                 GFP_KERNEL);
4451                 if (!dirty_i->dirty_segmap[i])
4452                         return -ENOMEM;
4453         }
4454
4455         if (__is_large_section(sbi)) {
4456                 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4457                 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
4458                                                 bitmap_size, GFP_KERNEL);
4459                 if (!dirty_i->dirty_secmap)
4460                         return -ENOMEM;
4461         }
4462
4463         init_dirty_segmap(sbi);
4464         return init_victim_secmap(sbi);
4465 }
4466
4467 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4468 {
4469         int i;
4470
4471         /*
4472          * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4473          * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4474          */
4475         for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4476                 struct curseg_info *curseg = CURSEG_I(sbi, i);
4477                 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4478                 unsigned int blkofs = curseg->next_blkoff;
4479
4480                 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4481                         goto out;
4482
4483                 if (curseg->alloc_type == SSR)
4484                         continue;
4485
4486                 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4487                         if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4488                                 continue;
4489 out:
4490                         f2fs_err(sbi,
4491                                  "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4492                                  i, curseg->segno, curseg->alloc_type,
4493                                  curseg->next_blkoff, blkofs);
4494                         return -EFSCORRUPTED;
4495                 }
4496         }
4497         return 0;
4498 }
4499
4500 #ifdef CONFIG_BLK_DEV_ZONED
4501
4502 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4503                                     struct f2fs_dev_info *fdev,
4504                                     struct blk_zone *zone)
4505 {
4506         unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno;
4507         block_t zone_block, wp_block, last_valid_block;
4508         unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4509         int i, s, b, ret;
4510         struct seg_entry *se;
4511
4512         if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4513                 return 0;
4514
4515         wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block);
4516         wp_segno = GET_SEGNO(sbi, wp_block);
4517         wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4518         zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4519         zone_segno = GET_SEGNO(sbi, zone_block);
4520         zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno);
4521
4522         if (zone_segno >= MAIN_SEGS(sbi))
4523                 return 0;
4524
4525         /*
4526          * Skip check of zones cursegs point to, since
4527          * fix_curseg_write_pointer() checks them.
4528          */
4529         for (i = 0; i < NO_CHECK_TYPE; i++)
4530                 if (zone_secno == GET_SEC_FROM_SEG(sbi,
4531                                                    CURSEG_I(sbi, i)->segno))
4532                         return 0;
4533
4534         /*
4535          * Get last valid block of the zone.
4536          */
4537         last_valid_block = zone_block - 1;
4538         for (s = sbi->segs_per_sec - 1; s >= 0; s--) {
4539                 segno = zone_segno + s;
4540                 se = get_seg_entry(sbi, segno);
4541                 for (b = sbi->blocks_per_seg - 1; b >= 0; b--)
4542                         if (f2fs_test_bit(b, se->cur_valid_map)) {
4543                                 last_valid_block = START_BLOCK(sbi, segno) + b;
4544                                 break;
4545                         }
4546                 if (last_valid_block >= zone_block)
4547                         break;
4548         }
4549
4550         /*
4551          * If last valid block is beyond the write pointer, report the
4552          * inconsistency. This inconsistency does not cause write error
4553          * because the zone will not be selected for write operation until
4554          * it get discarded. Just report it.
4555          */
4556         if (last_valid_block >= wp_block) {
4557                 f2fs_notice(sbi, "Valid block beyond write pointer: "
4558                             "valid block[0x%x,0x%x] wp[0x%x,0x%x]",
4559                             GET_SEGNO(sbi, last_valid_block),
4560                             GET_BLKOFF_FROM_SEG0(sbi, last_valid_block),
4561                             wp_segno, wp_blkoff);
4562                 return 0;
4563         }
4564
4565         /*
4566          * If there is no valid block in the zone and if write pointer is
4567          * not at zone start, reset the write pointer.
4568          */
4569         if (last_valid_block + 1 == zone_block && zone->wp != zone->start) {
4570                 f2fs_notice(sbi,
4571                             "Zone without valid block has non-zero write "
4572                             "pointer. Reset the write pointer: wp[0x%x,0x%x]",
4573                             wp_segno, wp_blkoff);
4574                 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
4575                                         zone->len >> log_sectors_per_block);
4576                 if (ret) {
4577                         f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4578                                  fdev->path, ret);
4579                         return ret;
4580                 }
4581         }
4582
4583         return 0;
4584 }
4585
4586 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
4587                                                   block_t zone_blkaddr)
4588 {
4589         int i;
4590
4591         for (i = 0; i < sbi->s_ndevs; i++) {
4592                 if (!bdev_is_zoned(FDEV(i).bdev))
4593                         continue;
4594                 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
4595                                 zone_blkaddr <= FDEV(i).end_blk))
4596                         return &FDEV(i);
4597         }
4598
4599         return NULL;
4600 }
4601
4602 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
4603                               void *data) {
4604         memcpy(data, zone, sizeof(struct blk_zone));
4605         return 0;
4606 }
4607
4608 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
4609 {
4610         struct curseg_info *cs = CURSEG_I(sbi, type);
4611         struct f2fs_dev_info *zbd;
4612         struct blk_zone zone;
4613         unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
4614         block_t cs_zone_block, wp_block;
4615         unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4616         sector_t zone_sector;
4617         int err;
4618
4619         cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4620         cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4621
4622         zbd = get_target_zoned_dev(sbi, cs_zone_block);
4623         if (!zbd)
4624                 return 0;
4625
4626         /* report zone for the sector the curseg points to */
4627         zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4628                 << log_sectors_per_block;
4629         err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4630                                   report_one_zone_cb, &zone);
4631         if (err != 1) {
4632                 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4633                          zbd->path, err);
4634                 return err;
4635         }
4636
4637         if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4638                 return 0;
4639
4640         wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
4641         wp_segno = GET_SEGNO(sbi, wp_block);
4642         wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4643         wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
4644
4645         if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
4646                 wp_sector_off == 0)
4647                 return 0;
4648
4649         f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
4650                     "curseg[0x%x,0x%x] wp[0x%x,0x%x]",
4651                     type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff);
4652
4653         f2fs_notice(sbi, "Assign new section to curseg[%d]: "
4654                     "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff);
4655         allocate_segment_by_default(sbi, type, true);
4656
4657         /* check consistency of the zone curseg pointed to */
4658         if (check_zone_write_pointer(sbi, zbd, &zone))
4659                 return -EIO;
4660
4661         /* check newly assigned zone */
4662         cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4663         cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4664
4665         zbd = get_target_zoned_dev(sbi, cs_zone_block);
4666         if (!zbd)
4667                 return 0;
4668
4669         zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4670                 << log_sectors_per_block;
4671         err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4672                                   report_one_zone_cb, &zone);
4673         if (err != 1) {
4674                 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4675                          zbd->path, err);
4676                 return err;
4677         }
4678
4679         if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4680                 return 0;
4681
4682         if (zone.wp != zone.start) {
4683                 f2fs_notice(sbi,
4684                             "New zone for curseg[%d] is not yet discarded. "
4685                             "Reset the zone: curseg[0x%x,0x%x]",
4686                             type, cs->segno, cs->next_blkoff);
4687                 err = __f2fs_issue_discard_zone(sbi, zbd->bdev,
4688                                 zone_sector >> log_sectors_per_block,
4689                                 zone.len >> log_sectors_per_block);
4690                 if (err) {
4691                         f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4692                                  zbd->path, err);
4693                         return err;
4694                 }
4695         }
4696
4697         return 0;
4698 }
4699
4700 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4701 {
4702         int i, ret;
4703
4704         for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4705                 ret = fix_curseg_write_pointer(sbi, i);
4706                 if (ret)
4707                         return ret;
4708         }
4709
4710         return 0;
4711 }
4712
4713 struct check_zone_write_pointer_args {
4714         struct f2fs_sb_info *sbi;
4715         struct f2fs_dev_info *fdev;
4716 };
4717
4718 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
4719                                       void *data) {
4720         struct check_zone_write_pointer_args *args;
4721         args = (struct check_zone_write_pointer_args *)data;
4722
4723         return check_zone_write_pointer(args->sbi, args->fdev, zone);
4724 }
4725
4726 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4727 {
4728         int i, ret;
4729         struct check_zone_write_pointer_args args;
4730
4731         for (i = 0; i < sbi->s_ndevs; i++) {
4732                 if (!bdev_is_zoned(FDEV(i).bdev))
4733                         continue;
4734
4735                 args.sbi = sbi;
4736                 args.fdev = &FDEV(i);
4737                 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
4738                                           check_zone_write_pointer_cb, &args);
4739                 if (ret < 0)
4740                         return ret;
4741         }
4742
4743         return 0;
4744 }
4745
4746 static bool is_conv_zone(struct f2fs_sb_info *sbi, unsigned int zone_idx,
4747                                                 unsigned int dev_idx)
4748 {
4749         if (!bdev_is_zoned(FDEV(dev_idx).bdev))
4750                 return true;
4751         return !test_bit(zone_idx, FDEV(dev_idx).blkz_seq);
4752 }
4753
4754 /* Return the zone index in the given device */
4755 static unsigned int get_zone_idx(struct f2fs_sb_info *sbi, unsigned int secno,
4756                                         int dev_idx)
4757 {
4758         block_t sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
4759
4760         return (sec_start_blkaddr - FDEV(dev_idx).start_blk) >>
4761                                                 sbi->log_blocks_per_blkz;
4762 }
4763
4764 /*
4765  * Return the usable segments in a section based on the zone's
4766  * corresponding zone capacity. Zone is equal to a section.
4767  */
4768 static inline unsigned int f2fs_usable_zone_segs_in_sec(
4769                 struct f2fs_sb_info *sbi, unsigned int segno)
4770 {
4771         unsigned int dev_idx, zone_idx, unusable_segs_in_sec;
4772
4773         dev_idx = f2fs_target_device_index(sbi, START_BLOCK(sbi, segno));
4774         zone_idx = get_zone_idx(sbi, GET_SEC_FROM_SEG(sbi, segno), dev_idx);
4775
4776         /* Conventional zone's capacity is always equal to zone size */
4777         if (is_conv_zone(sbi, zone_idx, dev_idx))
4778                 return sbi->segs_per_sec;
4779
4780         /*
4781          * If the zone_capacity_blocks array is NULL, then zone capacity
4782          * is equal to the zone size for all zones
4783          */
4784         if (!FDEV(dev_idx).zone_capacity_blocks)
4785                 return sbi->segs_per_sec;
4786
4787         /* Get the segment count beyond zone capacity block */
4788         unusable_segs_in_sec = (sbi->blocks_per_blkz -
4789                                 FDEV(dev_idx).zone_capacity_blocks[zone_idx]) >>
4790                                 sbi->log_blocks_per_seg;
4791         return sbi->segs_per_sec - unusable_segs_in_sec;
4792 }
4793
4794 /*
4795  * Return the number of usable blocks in a segment. The number of blocks
4796  * returned is always equal to the number of blocks in a segment for
4797  * segments fully contained within a sequential zone capacity or a
4798  * conventional zone. For segments partially contained in a sequential
4799  * zone capacity, the number of usable blocks up to the zone capacity
4800  * is returned. 0 is returned in all other cases.
4801  */
4802 static inline unsigned int f2fs_usable_zone_blks_in_seg(
4803                         struct f2fs_sb_info *sbi, unsigned int segno)
4804 {
4805         block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
4806         unsigned int zone_idx, dev_idx, secno;
4807
4808         secno = GET_SEC_FROM_SEG(sbi, segno);
4809         seg_start = START_BLOCK(sbi, segno);
4810         dev_idx = f2fs_target_device_index(sbi, seg_start);
4811         zone_idx = get_zone_idx(sbi, secno, dev_idx);
4812
4813         /*
4814          * Conventional zone's capacity is always equal to zone size,
4815          * so, blocks per segment is unchanged.
4816          */
4817         if (is_conv_zone(sbi, zone_idx, dev_idx))
4818                 return sbi->blocks_per_seg;
4819
4820         if (!FDEV(dev_idx).zone_capacity_blocks)
4821                 return sbi->blocks_per_seg;
4822
4823         sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
4824         sec_cap_blkaddr = sec_start_blkaddr +
4825                                 FDEV(dev_idx).zone_capacity_blocks[zone_idx];
4826
4827         /*
4828          * If segment starts before zone capacity and spans beyond
4829          * zone capacity, then usable blocks are from seg start to
4830          * zone capacity. If the segment starts after the zone capacity,
4831          * then there are no usable blocks.
4832          */
4833         if (seg_start >= sec_cap_blkaddr)
4834                 return 0;
4835         if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr)
4836                 return sec_cap_blkaddr - seg_start;
4837
4838         return sbi->blocks_per_seg;
4839 }
4840 #else
4841 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4842 {
4843         return 0;
4844 }
4845
4846 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4847 {
4848         return 0;
4849 }
4850
4851 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
4852                                                         unsigned int segno)
4853 {
4854         return 0;
4855 }
4856
4857 static inline unsigned int f2fs_usable_zone_segs_in_sec(struct f2fs_sb_info *sbi,
4858                                                         unsigned int segno)
4859 {
4860         return 0;
4861 }
4862 #endif
4863 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
4864                                         unsigned int segno)
4865 {
4866         if (f2fs_sb_has_blkzoned(sbi))
4867                 return f2fs_usable_zone_blks_in_seg(sbi, segno);
4868
4869         return sbi->blocks_per_seg;
4870 }
4871
4872 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi,
4873                                         unsigned int segno)
4874 {
4875         if (f2fs_sb_has_blkzoned(sbi))
4876                 return f2fs_usable_zone_segs_in_sec(sbi, segno);
4877
4878         return sbi->segs_per_sec;
4879 }
4880
4881 /*
4882  * Update min, max modified time for cost-benefit GC algorithm
4883  */
4884 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4885 {
4886         struct sit_info *sit_i = SIT_I(sbi);
4887         unsigned int segno;
4888
4889         down_write(&sit_i->sentry_lock);
4890
4891         sit_i->min_mtime = ULLONG_MAX;
4892
4893         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4894                 unsigned int i;
4895                 unsigned long long mtime = 0;
4896
4897                 for (i = 0; i < sbi->segs_per_sec; i++)
4898                         mtime += get_seg_entry(sbi, segno + i)->mtime;
4899
4900                 mtime = div_u64(mtime, sbi->segs_per_sec);
4901
4902                 if (sit_i->min_mtime > mtime)
4903                         sit_i->min_mtime = mtime;
4904         }
4905         sit_i->max_mtime = get_mtime(sbi, false);
4906         up_write(&sit_i->sentry_lock);
4907 }
4908
4909 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
4910 {
4911         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4912         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4913         struct f2fs_sm_info *sm_info;
4914         int err;
4915
4916         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
4917         if (!sm_info)
4918                 return -ENOMEM;
4919
4920         /* init sm info */
4921         sbi->sm_info = sm_info;
4922         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4923         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4924         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4925         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4926         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4927         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4928         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
4929         sm_info->rec_prefree_segments = sm_info->main_segments *
4930                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
4931         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4932                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4933
4934         if (!f2fs_lfs_mode(sbi))
4935                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
4936         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
4937         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
4938         sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
4939         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
4940         sm_info->min_ssr_sections = reserved_sections(sbi);
4941
4942         INIT_LIST_HEAD(&sm_info->sit_entry_set);
4943
4944         init_rwsem(&sm_info->curseg_lock);
4945
4946         if (!f2fs_readonly(sbi->sb)) {
4947                 err = f2fs_create_flush_cmd_control(sbi);
4948                 if (err)
4949                         return err;
4950         }
4951
4952         err = create_discard_cmd_control(sbi);
4953         if (err)
4954                 return err;
4955
4956         err = build_sit_info(sbi);
4957         if (err)
4958                 return err;
4959         err = build_free_segmap(sbi);
4960         if (err)
4961                 return err;
4962         err = build_curseg(sbi);
4963         if (err)
4964                 return err;
4965
4966         /* reinit free segmap based on SIT */
4967         err = build_sit_entries(sbi);
4968         if (err)
4969                 return err;
4970
4971         init_free_segmap(sbi);
4972         err = build_dirty_segmap(sbi);
4973         if (err)
4974                 return err;
4975
4976         err = sanity_check_curseg(sbi);
4977         if (err)
4978                 return err;
4979
4980         init_min_max_mtime(sbi);
4981         return 0;
4982 }
4983
4984 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4985                 enum dirty_type dirty_type)
4986 {
4987         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4988
4989         mutex_lock(&dirty_i->seglist_lock);
4990         kvfree(dirty_i->dirty_segmap[dirty_type]);
4991         dirty_i->nr_dirty[dirty_type] = 0;
4992         mutex_unlock(&dirty_i->seglist_lock);
4993 }
4994
4995 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
4996 {
4997         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4998         kvfree(dirty_i->victim_secmap);
4999 }
5000
5001 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5002 {
5003         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5004         int i;
5005
5006         if (!dirty_i)
5007                 return;
5008
5009         /* discard pre-free/dirty segments list */
5010         for (i = 0; i < NR_DIRTY_TYPE; i++)
5011                 discard_dirty_segmap(sbi, i);
5012
5013         if (__is_large_section(sbi)) {
5014                 mutex_lock(&dirty_i->seglist_lock);
5015                 kvfree(dirty_i->dirty_secmap);
5016                 mutex_unlock(&dirty_i->seglist_lock);
5017         }
5018
5019         destroy_victim_secmap(sbi);
5020         SM_I(sbi)->dirty_info = NULL;
5021         kvfree(dirty_i);
5022 }
5023
5024 static void destroy_curseg(struct f2fs_sb_info *sbi)
5025 {
5026         struct curseg_info *array = SM_I(sbi)->curseg_array;
5027         int i;
5028
5029         if (!array)
5030                 return;
5031         SM_I(sbi)->curseg_array = NULL;
5032         for (i = 0; i < NR_CURSEG_TYPE; i++) {
5033                 kvfree(array[i].sum_blk);
5034                 kvfree(array[i].journal);
5035         }
5036         kvfree(array);
5037 }
5038
5039 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5040 {
5041         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5042         if (!free_i)
5043                 return;
5044         SM_I(sbi)->free_info = NULL;
5045         kvfree(free_i->free_segmap);
5046         kvfree(free_i->free_secmap);
5047         kvfree(free_i);
5048 }
5049
5050 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5051 {
5052         struct sit_info *sit_i = SIT_I(sbi);
5053
5054         if (!sit_i)
5055                 return;
5056
5057         if (sit_i->sentries)
5058                 kvfree(sit_i->bitmap);
5059         kvfree(sit_i->tmp_map);
5060
5061         kvfree(sit_i->sentries);
5062         kvfree(sit_i->sec_entries);
5063         kvfree(sit_i->dirty_sentries_bitmap);
5064
5065         SM_I(sbi)->sit_info = NULL;
5066         kvfree(sit_i->sit_bitmap);
5067 #ifdef CONFIG_F2FS_CHECK_FS
5068         kvfree(sit_i->sit_bitmap_mir);
5069         kvfree(sit_i->invalid_segmap);
5070 #endif
5071         kvfree(sit_i);
5072 }
5073
5074 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5075 {
5076         struct f2fs_sm_info *sm_info = SM_I(sbi);
5077
5078         if (!sm_info)
5079                 return;
5080         f2fs_destroy_flush_cmd_control(sbi, true);
5081         destroy_discard_cmd_control(sbi);
5082         destroy_dirty_segmap(sbi);
5083         destroy_curseg(sbi);
5084         destroy_free_segmap(sbi);
5085         destroy_sit_info(sbi);
5086         sbi->sm_info = NULL;
5087         kvfree(sm_info);
5088 }
5089
5090 int __init f2fs_create_segment_manager_caches(void)
5091 {
5092         discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5093                         sizeof(struct discard_entry));
5094         if (!discard_entry_slab)
5095                 goto fail;
5096
5097         discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5098                         sizeof(struct discard_cmd));
5099         if (!discard_cmd_slab)
5100                 goto destroy_discard_entry;
5101
5102         sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5103                         sizeof(struct sit_entry_set));
5104         if (!sit_entry_set_slab)
5105                 goto destroy_discard_cmd;
5106
5107         inmem_entry_slab = f2fs_kmem_cache_create("f2fs_inmem_page_entry",
5108                         sizeof(struct inmem_pages));
5109         if (!inmem_entry_slab)
5110                 goto destroy_sit_entry_set;
5111         return 0;
5112
5113 destroy_sit_entry_set:
5114         kmem_cache_destroy(sit_entry_set_slab);
5115 destroy_discard_cmd:
5116         kmem_cache_destroy(discard_cmd_slab);
5117 destroy_discard_entry:
5118         kmem_cache_destroy(discard_entry_slab);
5119 fail:
5120         return -ENOMEM;
5121 }
5122
5123 void f2fs_destroy_segment_manager_caches(void)
5124 {
5125         kmem_cache_destroy(sit_entry_set_slab);
5126         kmem_cache_destroy(discard_cmd_slab);
5127         kmem_cache_destroy(discard_entry_slab);
5128         kmem_cache_destroy(inmem_entry_slab);
5129 }