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/events/f2fs.h>
  24
  25 #define __reverse_ffz(x) __reverse_ffs(~(x))
  26
  27 static struct kmem_cache *discard_entry_slab;
  28 static struct kmem_cache *discard_cmd_slab;
  29 static struct kmem_cache *sit_entry_set_slab;
  30 static struct kmem_cache *inmem_entry_slab;
  31
  32 static unsigned long __reverse_ulong(unsigned char *str)
  33 {
  34         unsigned long tmp = 0;
  35         int shift = 24, idx = 0;
  36
  37 #if BITS_PER_LONG == 64
  38         shift = 56;
  39 #endif
  40         while (shift >= 0) {
  41                 tmp |= (unsigned long)str[idx++] << shift;
  42                 shift -= BITS_PER_BYTE;
  43         }
  44         return tmp;
  45 }
  46
  47 /*
  48  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  49  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  50  */
  51 static inline unsigned long __reverse_ffs(unsigned long word)
  52 {
  53         int num = 0;
  54
  55 #if BITS_PER_LONG == 64
  56         if ((word & 0xffffffff00000000UL) == 0)
  57                 num += 32;
  58         else
  59                 word >>= 32;
  60 #endif
  61         if ((word & 0xffff0000) == 0)
  62                 num += 16;
  63         else
  64                 word >>= 16;
  65
  66         if ((word & 0xff00) == 0)
  67                 num += 8;
  68         else
  69                 word >>= 8;
  70
  71         if ((word & 0xf0) == 0)
  72                 num += 4;
  73         else
  74                 word >>= 4;
  75
  76         if ((word & 0xc) == 0)
  77                 num += 2;
  78         else
  79                 word >>= 2;
  80
  81         if ((word & 0x2) == 0)
  82                 num += 1;
  83         return num;
  84 }
  85
  86 /*
  87  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  88  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  89  * @size must be integral times of unsigned long.
  90  * Example:
  91  *                             MSB <--> LSB
  92  *   f2fs_set_bit(0, bitmap) => 1000 0000
  93  *   f2fs_set_bit(7, bitmap) => 0000 0001
  94  */
  95 static unsigned long __find_rev_next_bit(const unsigned long *addr,
  96                         unsigned long size, unsigned long offset)
  97 {
  98         const unsigned long *p = addr + BIT_WORD(offset);
  99         unsigned long result = size;
 100         unsigned long tmp;
 101
 102         if (offset >= size)
 103                 return size;
 104
 105         size -= (offset & ~(BITS_PER_LONG - 1));
 106         offset %= BITS_PER_LONG;
 107
 108         while (1) {
 109                 if (*p == 0)
 110                         goto pass;
 111
 112                 tmp = __reverse_ulong((unsigned char *)p);
 113
 114                 tmp &= ~0UL >> offset;
 115                 if (size < BITS_PER_LONG)
 116                         tmp &= (~0UL << (BITS_PER_LONG - size));
 117                 if (tmp)
 118                         goto found;
 119 pass:
 120                 if (size <= BITS_PER_LONG)
 121                         break;
 122                 size -= BITS_PER_LONG;
 123                 offset = 0;
 124                 p++;
 125         }
 126         return result;
 127 found:
 128         return result - size + __reverse_ffs(tmp);
 129 }
 130
 131 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 132                         unsigned long size, unsigned long offset)
 133 {
 134         const unsigned long *p = addr + BIT_WORD(offset);
 135         unsigned long result = size;
 136         unsigned long tmp;
 137
 138         if (offset >= size)
 139                 return size;
 140
 141         size -= (offset & ~(BITS_PER_LONG - 1));
 142         offset %= BITS_PER_LONG;
 143
 144         while (1) {
 145                 if (*p == ~0UL)
 146                         goto pass;
 147
 148                 tmp = __reverse_ulong((unsigned char *)p);
 149
 150                 if (offset)
 151                         tmp |= ~0UL << (BITS_PER_LONG - offset);
 152                 if (size < BITS_PER_LONG)
 153                         tmp |= ~0UL >> size;
 154                 if (tmp != ~0UL)
 155                         goto found;
 156 pass:
 157                 if (size <= BITS_PER_LONG)
 158                         break;
 159                 size -= BITS_PER_LONG;
 160                 offset = 0;
 161                 p++;
 162         }
 163         return result;
 164 found:
 165         return result - size + __reverse_ffz(tmp);
 166 }
 167
 168 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
 169 {
 170         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 171         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 172         int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
 173
 174         if (f2fs_lfs_mode(sbi))
 175                 return false;
 176         if (sbi->gc_mode == GC_URGENT_HIGH)
 177                 return true;
 178         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
 179                 return true;
 180
 181         return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
 182                         SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
 183 }
 184
 185 void f2fs_register_inmem_page(struct inode *inode, struct page *page)
 186 {
 187         struct inmem_pages *new;
 188
 189         set_page_private_atomic(page);
 190
 191         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 192
 193         /* add atomic page indices to the list */
 194         new->page = page;
 195         INIT_LIST_HEAD(&new->list);
 196
 197         /* increase reference count with clean state */
 198         get_page(page);
 199         mutex_lock(&F2FS_I(inode)->inmem_lock);
 200         list_add_tail(&new->list, &F2FS_I(inode)->inmem_pages);
 201         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 202         mutex_unlock(&F2FS_I(inode)->inmem_lock);
 203
 204         trace_f2fs_register_inmem_page(page, INMEM);
 205 }
 206
 207 static int __revoke_inmem_pages(struct inode *inode,
 208                                 struct list_head *head, bool drop, bool recover,
 209                                 bool trylock)
 210 {
 211         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 212         struct inmem_pages *cur, *tmp;
 213         int err = 0;
 214
 215         list_for_each_entry_safe(cur, tmp, head, list) {
 216                 struct page *page = cur->page;
 217
 218                 if (drop)
 219                         trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 220
 221                 if (trylock) {
 222                         /*
 223                          * to avoid deadlock in between page lock and
 224                          * inmem_lock.
 225                          */
 226                         if (!trylock_page(page))
 227                                 continue;
 228                 } else {
 229                         lock_page(page);
 230                 }
 231
 232                 f2fs_wait_on_page_writeback(page, DATA, true, true);
 233
 234                 if (recover) {
 235                         struct dnode_of_data dn;
 236                         struct node_info ni;
 237
 238                         trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 239 retry:
 240                         set_new_dnode(&dn, inode, NULL, NULL, 0);
 241                         err = f2fs_get_dnode_of_data(&dn, page->index,
 242                                                                 LOOKUP_NODE);
 243                         if (err) {
 244                                 if (err == -ENOMEM) {
 245                                         congestion_wait(BLK_RW_ASYNC,
 246                                                         DEFAULT_IO_TIMEOUT);
 247                                         cond_resched();
 248                                         goto retry;
 249                                 }
 250                                 err = -EAGAIN;
 251                                 goto next;
 252                         }
 253
 254                         err = f2fs_get_node_info(sbi, dn.nid, &ni);
 255                         if (err) {
 256                                 f2fs_put_dnode(&dn);
 257                                 return err;
 258                         }
 259
 260                         if (cur->old_addr == NEW_ADDR) {
 261                                 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
 262                                 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
 263                         } else
 264                                 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 265                                         cur->old_addr, ni.version, true, true);
 266                         f2fs_put_dnode(&dn);
 267                 }
 268 next:
 269                 /* we don't need to invalidate this in the sccessful status */
 270                 if (drop || recover) {
 271                         ClearPageUptodate(page);
 272                         clear_page_private_gcing(page);
 273                 }
 274                 detach_page_private(page);
 275                 set_page_private(page, 0);
 276                 f2fs_put_page(page, 1);
 277
 278                 list_del(&cur->list);
 279                 kmem_cache_free(inmem_entry_slab, cur);
 280                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 281         }
 282         return err;
 283 }
 284
 285 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
 286 {
 287         struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
 288         struct inode *inode;
 289         struct f2fs_inode_info *fi;
 290         unsigned int count = sbi->atomic_files;
 291         unsigned int looped = 0;
 292 next:
 293         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 294         if (list_empty(head)) {
 295                 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 296                 return;
 297         }
 298         fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
 299         inode = igrab(&fi->vfs_inode);
 300         if (inode)
 301                 list_move_tail(&fi->inmem_ilist, head);
 302         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 303
 304         if (inode) {
 305                 if (gc_failure) {
 306                         if (!fi->i_gc_failures[GC_FAILURE_ATOMIC])
 307                                 goto skip;
 308                 }
 309                 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
 310                 f2fs_drop_inmem_pages(inode);
 311 skip:
 312                 iput(inode);
 313         }
 314         congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
 315         cond_resched();
 316         if (gc_failure) {
 317                 if (++looped >= count)
 318                         return;
 319         }
 320         goto next;
 321 }
 322
 323 void f2fs_drop_inmem_pages(struct inode *inode)
 324 {
 325         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 326         struct f2fs_inode_info *fi = F2FS_I(inode);
 327
 328         do {
 329                 mutex_lock(&fi->inmem_lock);
 330                 if (list_empty(&fi->inmem_pages)) {
 331                         fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
 332
 333                         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 334                         if (!list_empty(&fi->inmem_ilist))
 335                                 list_del_init(&fi->inmem_ilist);
 336                         if (f2fs_is_atomic_file(inode)) {
 337                                 clear_inode_flag(inode, FI_ATOMIC_FILE);
 338                                 sbi->atomic_files--;
 339                         }
 340                         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 341
 342                         mutex_unlock(&fi->inmem_lock);
 343                         break;
 344                 }
 345                 __revoke_inmem_pages(inode, &fi->inmem_pages,
 346                                                 true, false, true);
 347                 mutex_unlock(&fi->inmem_lock);
 348         } while (1);
 349 }
 350
 351 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
 352 {
 353         struct f2fs_inode_info *fi = F2FS_I(inode);
 354         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 355         struct list_head *head = &fi->inmem_pages;
 356         struct inmem_pages *cur = NULL;
 357
 358         f2fs_bug_on(sbi, !page_private_atomic(page));
 359
 360         mutex_lock(&fi->inmem_lock);
 361         list_for_each_entry(cur, head, list) {
 362                 if (cur->page == page)
 363                         break;
 364         }
 365
 366         f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
 367         list_del(&cur->list);
 368         mutex_unlock(&fi->inmem_lock);
 369
 370         dec_page_count(sbi, F2FS_INMEM_PAGES);
 371         kmem_cache_free(inmem_entry_slab, cur);
 372
 373         ClearPageUptodate(page);
 374         clear_page_private_atomic(page);
 375         f2fs_put_page(page, 0);
 376
 377         detach_page_private(page);
 378         set_page_private(page, 0);
 379
 380         trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 381 }
 382
 383 static int __f2fs_commit_inmem_pages(struct inode *inode)
 384 {
 385         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 386         struct f2fs_inode_info *fi = F2FS_I(inode);
 387         struct inmem_pages *cur, *tmp;
 388         struct f2fs_io_info fio = {
 389                 .sbi = sbi,
 390                 .ino = inode->i_ino,
 391                 .type = DATA,
 392                 .op = REQ_OP_WRITE,
 393                 .op_flags = REQ_SYNC | REQ_PRIO,
 394                 .io_type = FS_DATA_IO,
 395         };
 396         struct list_head revoke_list;
 397         bool submit_bio = false;
 398         int err = 0;
 399
 400         INIT_LIST_HEAD(&revoke_list);
 401
 402         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 403                 struct page *page = cur->page;
 404
 405                 lock_page(page);
 406                 if (page->mapping == inode->i_mapping) {
 407                         trace_f2fs_commit_inmem_page(page, INMEM);
 408
 409                         f2fs_wait_on_page_writeback(page, DATA, true, true);
 410
 411                         set_page_dirty(page);
 412                         if (clear_page_dirty_for_io(page)) {
 413                                 inode_dec_dirty_pages(inode);
 414                                 f2fs_remove_dirty_inode(inode);
 415                         }
 416 retry:
 417                         fio.page = page;
 418                         fio.old_blkaddr = NULL_ADDR;
 419                         fio.encrypted_page = NULL;
 420                         fio.need_lock = LOCK_DONE;
 421                         err = f2fs_do_write_data_page(&fio);
 422                         if (err) {
 423                                 if (err == -ENOMEM) {
 424                                         congestion_wait(BLK_RW_ASYNC,
 425                                                         DEFAULT_IO_TIMEOUT);
 426                                         cond_resched();
 427                                         goto retry;
 428                                 }
 429                                 unlock_page(page);
 430                                 break;
 431                         }
 432                         /* record old blkaddr for revoking */
 433                         cur->old_addr = fio.old_blkaddr;
 434                         submit_bio = true;
 435                 }
 436                 unlock_page(page);
 437                 list_move_tail(&cur->list, &revoke_list);
 438         }
 439
 440         if (submit_bio)
 441                 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
 442
 443         if (err) {
 444                 /*
 445                  * try to revoke all committed pages, but still we could fail
 446                  * due to no memory or other reason, if that happened, EAGAIN
 447                  * will be returned, which means in such case, transaction is
 448                  * already not integrity, caller should use journal to do the
 449                  * recovery or rewrite & commit last transaction. For other
 450                  * error number, revoking was done by filesystem itself.
 451                  */
 452                 err = __revoke_inmem_pages(inode, &revoke_list,
 453                                                 false, true, false);
 454
 455                 /* drop all uncommitted pages */
 456                 __revoke_inmem_pages(inode, &fi->inmem_pages,
 457                                                 true, false, false);
 458         } else {
 459                 __revoke_inmem_pages(inode, &revoke_list,
 460                                                 false, false, false);
 461         }
 462
 463         return err;
 464 }
 465
 466 int f2fs_commit_inmem_pages(struct inode *inode)
 467 {
 468         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 469         struct f2fs_inode_info *fi = F2FS_I(inode);
 470         int err;
 471
 472         f2fs_balance_fs(sbi, true);
 473
 474         down_write(&fi->i_gc_rwsem[WRITE]);
 475
 476         f2fs_lock_op(sbi);
 477         set_inode_flag(inode, FI_ATOMIC_COMMIT);
 478
 479         mutex_lock(&fi->inmem_lock);
 480         err = __f2fs_commit_inmem_pages(inode);
 481         mutex_unlock(&fi->inmem_lock);
 482
 483         clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 484
 485         f2fs_unlock_op(sbi);
 486         up_write(&fi->i_gc_rwsem[WRITE]);
 487
 488         return err;
 489 }
 490
 491 /*
 492  * This function balances dirty node and dentry pages.
 493  * In addition, it controls garbage collection.
 494  */
 495 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 496 {
 497         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 498                 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
 499                 f2fs_stop_checkpoint(sbi, false);
 500         }
 501
 502         /* balance_fs_bg is able to be pending */
 503         if (need && excess_cached_nats(sbi))
 504                 f2fs_balance_fs_bg(sbi, false);
 505
 506         if (!f2fs_is_checkpoint_ready(sbi))
 507                 return;
 508
 509         /*
 510          * We should do GC or end up with checkpoint, if there are so many dirty
 511          * dir/node pages without enough free segments.
 512          */
 513         if (has_not_enough_free_secs(sbi, 0, 0)) {
 514                 if (test_opt(sbi, GC_MERGE) && sbi->gc_thread &&
 515                                         sbi->gc_thread->f2fs_gc_task) {
 516                         DEFINE_WAIT(wait);
 517
 518                         prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait,
 519                                                 TASK_UNINTERRUPTIBLE);
 520                         wake_up(&sbi->gc_thread->gc_wait_queue_head);
 521                         io_schedule();
 522                         finish_wait(&sbi->gc_thread->fggc_wq, &wait);
 523                 } else {
 524                         down_write(&sbi->gc_lock);
 525                         f2fs_gc(sbi, false, false, false, NULL_SEGNO);
 526                 }
 527         }
 528 }
 529
 530 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
 531 {
 532         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 533                 return;
 534
 535         /* try to shrink extent cache when there is no enough memory */
 536         if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
 537                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 538
 539         /* check the # of cached NAT entries */
 540         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
 541                 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 542
 543         if (!f2fs_available_free_memory(sbi, FREE_NIDS))
 544                 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
 545         else
 546                 f2fs_build_free_nids(sbi, false, false);
 547
 548         if (excess_dirty_nats(sbi) || excess_dirty_nodes(sbi) ||
 549                 excess_prefree_segs(sbi))
 550                 goto do_sync;
 551
 552         /* there is background inflight IO or foreground operation recently */
 553         if (is_inflight_io(sbi, REQ_TIME) ||
 554                 (!f2fs_time_over(sbi, REQ_TIME) && rwsem_is_locked(&sbi->cp_rwsem)))
 555                 return;
 556
 557         /* exceed periodical checkpoint timeout threshold */
 558         if (f2fs_time_over(sbi, CP_TIME))
 559                 goto do_sync;
 560
 561         /* checkpoint is the only way to shrink partial cached entries */
 562         if (f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
 563                 f2fs_available_free_memory(sbi, INO_ENTRIES))
 564                 return;
 565
 566 do_sync:
 567         if (test_opt(sbi, DATA_FLUSH) && from_bg) {
 568                 struct blk_plug plug;
 569
 570                 mutex_lock(&sbi->flush_lock);
 571
 572                 blk_start_plug(&plug);
 573                 f2fs_sync_dirty_inodes(sbi, FILE_INODE);
 574                 blk_finish_plug(&plug);
 575
 576                 mutex_unlock(&sbi->flush_lock);
 577         }
 578         f2fs_sync_fs(sbi->sb, true);
 579         stat_inc_bg_cp_count(sbi->stat_info);
 580 }
 581
 582 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 583                                 struct block_device *bdev)
 584 {
 585         int ret = blkdev_issue_flush(bdev);
 586
 587         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 588                                 test_opt(sbi, FLUSH_MERGE), ret);
 589         return ret;
 590 }
 591
 592 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 593 {
 594         int ret = 0;
 595         int i;
 596
 597         if (!f2fs_is_multi_device(sbi))
 598                 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 599
 600         for (i = 0; i < sbi->s_ndevs; i++) {
 601                 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
 602                         continue;
 603                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 604                 if (ret)
 605                         break;
 606         }
 607         return ret;
 608 }
 609
 610 static int issue_flush_thread(void *data)
 611 {
 612         struct f2fs_sb_info *sbi = data;
 613         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 614         wait_queue_head_t *q = &fcc->flush_wait_queue;
 615 repeat:
 616         if (kthread_should_stop())
 617                 return 0;
 618
 619         if (!llist_empty(&fcc->issue_list)) {
 620                 struct flush_cmd *cmd, *next;
 621                 int ret;
 622
 623                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 624                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 625
 626                 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 627
 628                 ret = submit_flush_wait(sbi, cmd->ino);
 629                 atomic_inc(&fcc->issued_flush);
 630
 631                 llist_for_each_entry_safe(cmd, next,
 632                                           fcc->dispatch_list, llnode) {
 633                         cmd->ret = ret;
 634                         complete(&cmd->wait);
 635                 }
 636                 fcc->dispatch_list = NULL;
 637         }
 638
 639         wait_event_interruptible(*q,
 640                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 641         goto repeat;
 642 }
 643
 644 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 645 {
 646         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 647         struct flush_cmd cmd;
 648         int ret;
 649
 650         if (test_opt(sbi, NOBARRIER))
 651                 return 0;
 652
 653         if (!test_opt(sbi, FLUSH_MERGE)) {
 654                 atomic_inc(&fcc->queued_flush);
 655                 ret = submit_flush_wait(sbi, ino);
 656                 atomic_dec(&fcc->queued_flush);
 657                 atomic_inc(&fcc->issued_flush);
 658                 return ret;
 659         }
 660
 661         if (atomic_inc_return(&fcc->queued_flush) == 1 ||
 662             f2fs_is_multi_device(sbi)) {
 663                 ret = submit_flush_wait(sbi, ino);
 664                 atomic_dec(&fcc->queued_flush);
 665
 666                 atomic_inc(&fcc->issued_flush);
 667                 return ret;
 668         }
 669
 670         cmd.ino = ino;
 671         init_completion(&cmd.wait);
 672
 673         llist_add(&cmd.llnode, &fcc->issue_list);
 674
 675         /*
 676          * update issue_list before we wake up issue_flush thread, this
 677          * smp_mb() pairs with another barrier in ___wait_event(), see
 678          * more details in comments of waitqueue_active().
 679          */
 680         smp_mb();
 681
 682         if (waitqueue_active(&fcc->flush_wait_queue))
 683                 wake_up(&fcc->flush_wait_queue);
 684
 685         if (fcc->f2fs_issue_flush) {
 686                 wait_for_completion(&cmd.wait);
 687                 atomic_dec(&fcc->queued_flush);
 688         } else {
 689                 struct llist_node *list;
 690
 691                 list = llist_del_all(&fcc->issue_list);
 692                 if (!list) {
 693                         wait_for_completion(&cmd.wait);
 694                         atomic_dec(&fcc->queued_flush);
 695                 } else {
 696                         struct flush_cmd *tmp, *next;
 697
 698                         ret = submit_flush_wait(sbi, ino);
 699
 700                         llist_for_each_entry_safe(tmp, next, list, llnode) {
 701                                 if (tmp == &cmd) {
 702                                         cmd.ret = ret;
 703                                         atomic_dec(&fcc->queued_flush);
 704                                         continue;
 705                                 }
 706                                 tmp->ret = ret;
 707                                 complete(&tmp->wait);
 708                         }
 709                 }
 710         }
 711
 712         return cmd.ret;
 713 }
 714
 715 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
 716 {
 717         dev_t dev = sbi->sb->s_bdev->bd_dev;
 718         struct flush_cmd_control *fcc;
 719         int err = 0;
 720
 721         if (SM_I(sbi)->fcc_info) {
 722                 fcc = SM_I(sbi)->fcc_info;
 723                 if (fcc->f2fs_issue_flush)
 724                         return err;
 725                 goto init_thread;
 726         }
 727
 728         fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
 729         if (!fcc)
 730                 return -ENOMEM;
 731         atomic_set(&fcc->issued_flush, 0);
 732         atomic_set(&fcc->queued_flush, 0);
 733         init_waitqueue_head(&fcc->flush_wait_queue);
 734         init_llist_head(&fcc->issue_list);
 735         SM_I(sbi)->fcc_info = fcc;
 736         if (!test_opt(sbi, FLUSH_MERGE))
 737                 return err;
 738
 739 init_thread:
 740         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 741                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 742         if (IS_ERR(fcc->f2fs_issue_flush)) {
 743                 err = PTR_ERR(fcc->f2fs_issue_flush);
 744                 kfree(fcc);
 745                 SM_I(sbi)->fcc_info = NULL;
 746                 return err;
 747         }
 748
 749         return err;
 750 }
 751
 752 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 753 {
 754         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 755
 756         if (fcc && fcc->f2fs_issue_flush) {
 757                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 758
 759                 fcc->f2fs_issue_flush = NULL;
 760                 kthread_stop(flush_thread);
 761         }
 762         if (free) {
 763                 kfree(fcc);
 764                 SM_I(sbi)->fcc_info = NULL;
 765         }
 766 }
 767
 768 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
 769 {
 770         int ret = 0, i;
 771
 772         if (!f2fs_is_multi_device(sbi))
 773                 return 0;
 774
 775         if (test_opt(sbi, NOBARRIER))
 776                 return 0;
 777
 778         for (i = 1; i < sbi->s_ndevs; i++) {
 779                 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
 780                         continue;
 781                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 782                 if (ret)
 783                         break;
 784
 785                 spin_lock(&sbi->dev_lock);
 786                 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
 787                 spin_unlock(&sbi->dev_lock);
 788         }
 789
 790         return ret;
 791 }
 792
 793 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 794                 enum dirty_type dirty_type)
 795 {
 796         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 797
 798         /* need not be added */
 799         if (IS_CURSEG(sbi, segno))
 800                 return;
 801
 802         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 803                 dirty_i->nr_dirty[dirty_type]++;
 804
 805         if (dirty_type == DIRTY) {
 806                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 807                 enum dirty_type t = sentry->type;
 808
 809                 if (unlikely(t >= DIRTY)) {
 810                         f2fs_bug_on(sbi, 1);
 811                         return;
 812                 }
 813                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 814                         dirty_i->nr_dirty[t]++;
 815
 816                 if (__is_large_section(sbi)) {
 817                         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 818                         block_t valid_blocks =
 819                                 get_valid_blocks(sbi, segno, true);
 820
 821                         f2fs_bug_on(sbi, unlikely(!valid_blocks ||
 822                                         valid_blocks == BLKS_PER_SEC(sbi)));
 823
 824                         if (!IS_CURSEC(sbi, secno))
 825                                 set_bit(secno, dirty_i->dirty_secmap);
 826                 }
 827         }
 828 }
 829
 830 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 831                 enum dirty_type dirty_type)
 832 {
 833         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 834         block_t valid_blocks;
 835
 836         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 837                 dirty_i->nr_dirty[dirty_type]--;
 838
 839         if (dirty_type == DIRTY) {
 840                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 841                 enum dirty_type t = sentry->type;
 842
 843                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 844                         dirty_i->nr_dirty[t]--;
 845
 846                 valid_blocks = get_valid_blocks(sbi, segno, true);
 847                 if (valid_blocks == 0) {
 848                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 849                                                 dirty_i->victim_secmap);
 850 #ifdef CONFIG_F2FS_CHECK_FS
 851                         clear_bit(segno, SIT_I(sbi)->invalid_segmap);
 852 #endif
 853                 }
 854                 if (__is_large_section(sbi)) {
 855                         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 856
 857                         if (!valid_blocks ||
 858                                         valid_blocks == BLKS_PER_SEC(sbi)) {
 859                                 clear_bit(secno, dirty_i->dirty_secmap);
 860                                 return;
 861                         }
 862
 863                         if (!IS_CURSEC(sbi, secno))
 864                                 set_bit(secno, dirty_i->dirty_secmap);
 865                 }
 866         }
 867 }
 868
 869 /*
 870  * Should not occur error such as -ENOMEM.
 871  * Adding dirty entry into seglist is not critical operation.
 872  * If a given segment is one of current working segments, it won't be added.
 873  */
 874 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 875 {
 876         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 877         unsigned short valid_blocks, ckpt_valid_blocks;
 878         unsigned int usable_blocks;
 879
 880         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 881                 return;
 882
 883         usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
 884         mutex_lock(&dirty_i->seglist_lock);
 885
 886         valid_blocks = get_valid_blocks(sbi, segno, false);
 887         ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false);
 888
 889         if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
 890                 ckpt_valid_blocks == usable_blocks)) {
 891                 __locate_dirty_segment(sbi, segno, PRE);
 892                 __remove_dirty_segment(sbi, segno, DIRTY);
 893         } else if (valid_blocks < usable_blocks) {
 894                 __locate_dirty_segment(sbi, segno, DIRTY);
 895         } else {
 896                 /* Recovery routine with SSR needs this */
 897                 __remove_dirty_segment(sbi, segno, DIRTY);
 898         }
 899
 900         mutex_unlock(&dirty_i->seglist_lock);
 901 }
 902
 903 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
 904 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
 905 {
 906         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 907         unsigned int segno;
 908
 909         mutex_lock(&dirty_i->seglist_lock);
 910         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 911                 if (get_valid_blocks(sbi, segno, false))
 912                         continue;
 913                 if (IS_CURSEG(sbi, segno))
 914                         continue;
 915                 __locate_dirty_segment(sbi, segno, PRE);
 916                 __remove_dirty_segment(sbi, segno, DIRTY);
 917         }
 918         mutex_unlock(&dirty_i->seglist_lock);
 919 }
 920
 921 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
 922 {
 923         int ovp_hole_segs =
 924                 (overprovision_segments(sbi) - reserved_segments(sbi));
 925         block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
 926         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 927         block_t holes[2] = {0, 0};      /* DATA and NODE */
 928         block_t unusable;
 929         struct seg_entry *se;
 930         unsigned int segno;
 931
 932         mutex_lock(&dirty_i->seglist_lock);
 933         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 934                 se = get_seg_entry(sbi, segno);
 935                 if (IS_NODESEG(se->type))
 936                         holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
 937                                                         se->valid_blocks;
 938                 else
 939                         holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
 940                                                         se->valid_blocks;
 941         }
 942         mutex_unlock(&dirty_i->seglist_lock);
 943
 944         unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
 945         if (unusable > ovp_holes)
 946                 return unusable - ovp_holes;
 947         return 0;
 948 }
 949
 950 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
 951 {
 952         int ovp_hole_segs =
 953                 (overprovision_segments(sbi) - reserved_segments(sbi));
 954         if (unusable > F2FS_OPTION(sbi).unusable_cap)
 955                 return -EAGAIN;
 956         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
 957                 dirty_segments(sbi) > ovp_hole_segs)
 958                 return -EAGAIN;
 959         return 0;
 960 }
 961
 962 /* This is only used by SBI_CP_DISABLED */
 963 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
 964 {
 965         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 966         unsigned int segno = 0;
 967
 968         mutex_lock(&dirty_i->seglist_lock);
 969         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 970                 if (get_valid_blocks(sbi, segno, false))
 971                         continue;
 972                 if (get_ckpt_valid_blocks(sbi, segno, false))
 973                         continue;
 974                 mutex_unlock(&dirty_i->seglist_lock);
 975                 return segno;
 976         }
 977         mutex_unlock(&dirty_i->seglist_lock);
 978         return NULL_SEGNO;
 979 }
 980
 981 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 982                 struct block_device *bdev, block_t lstart,
 983                 block_t start, block_t len)
 984 {
 985         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 986         struct list_head *pend_list;
 987         struct discard_cmd *dc;
 988
 989         f2fs_bug_on(sbi, !len);
 990
 991         pend_list = &dcc->pend_list[plist_idx(len)];
 992
 993         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 994         INIT_LIST_HEAD(&dc->list);
 995         dc->bdev = bdev;
 996         dc->lstart = lstart;
 997         dc->start = start;
 998         dc->len = len;
 999         dc->ref = 0;
1000         dc->state = D_PREP;
1001         dc->queued = 0;
1002         dc->error = 0;
1003         init_completion(&dc->wait);
1004         list_add_tail(&dc->list, pend_list);
1005         spin_lock_init(&dc->lock);
1006         dc->bio_ref = 0;
1007         atomic_inc(&dcc->discard_cmd_cnt);
1008         dcc->undiscard_blks += len;
1009
1010         return dc;
1011 }
1012
1013 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
1014                                 struct block_device *bdev, block_t lstart,
1015                                 block_t start, block_t len,
1016                                 struct rb_node *parent, struct rb_node **p,
1017                                 bool leftmost)
1018 {
1019         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1020         struct discard_cmd *dc;
1021
1022         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1023
1024         rb_link_node(&dc->rb_node, parent, p);
1025         rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
1026
1027         return dc;
1028 }
1029
1030 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1031                                                         struct discard_cmd *dc)
1032 {
1033         if (dc->state == D_DONE)
1034                 atomic_sub(dc->queued, &dcc->queued_discard);
1035
1036         list_del(&dc->list);
1037         rb_erase_cached(&dc->rb_node, &dcc->root);
1038         dcc->undiscard_blks -= dc->len;
1039
1040         kmem_cache_free(discard_cmd_slab, dc);
1041
1042         atomic_dec(&dcc->discard_cmd_cnt);
1043 }
1044
1045 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1046                                                         struct discard_cmd *dc)
1047 {
1048         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1049         unsigned long flags;
1050
1051         trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
1052
1053         spin_lock_irqsave(&dc->lock, flags);
1054         if (dc->bio_ref) {
1055                 spin_unlock_irqrestore(&dc->lock, flags);
1056                 return;
1057         }
1058         spin_unlock_irqrestore(&dc->lock, flags);
1059
1060         f2fs_bug_on(sbi, dc->ref);
1061
1062         if (dc->error == -EOPNOTSUPP)
1063                 dc->error = 0;
1064
1065         if (dc->error)
1066                 printk_ratelimited(
1067                         "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
1068                         KERN_INFO, sbi->sb->s_id,
1069                         dc->lstart, dc->start, dc->len, dc->error);
1070         __detach_discard_cmd(dcc, dc);
1071 }
1072
1073 static void f2fs_submit_discard_endio(struct bio *bio)
1074 {
1075         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1076         unsigned long flags;
1077
1078         spin_lock_irqsave(&dc->lock, flags);
1079         if (!dc->error)
1080                 dc->error = blk_status_to_errno(bio->bi_status);
1081         dc->bio_ref--;
1082         if (!dc->bio_ref && dc->state == D_SUBMIT) {
1083                 dc->state = D_DONE;
1084                 complete_all(&dc->wait);
1085         }
1086         spin_unlock_irqrestore(&dc->lock, flags);
1087         bio_put(bio);
1088 }
1089
1090 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1091                                 block_t start, block_t end)
1092 {
1093 #ifdef CONFIG_F2FS_CHECK_FS
1094         struct seg_entry *sentry;
1095         unsigned int segno;
1096         block_t blk = start;
1097         unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1098         unsigned long *map;
1099
1100         while (blk < end) {
1101                 segno = GET_SEGNO(sbi, blk);
1102                 sentry = get_seg_entry(sbi, segno);
1103                 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1104
1105                 if (end < START_BLOCK(sbi, segno + 1))
1106                         size = GET_BLKOFF_FROM_SEG0(sbi, end);
1107                 else
1108                         size = max_blocks;
1109                 map = (unsigned long *)(sentry->cur_valid_map);
1110                 offset = __find_rev_next_bit(map, size, offset);
1111                 f2fs_bug_on(sbi, offset != size);
1112                 blk = START_BLOCK(sbi, segno + 1);
1113         }
1114 #endif
1115 }
1116
1117 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1118                                 struct discard_policy *dpolicy,
1119                                 int discard_type, unsigned int granularity)
1120 {
1121         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1122
1123         /* common policy */
1124         dpolicy->type = discard_type;
1125         dpolicy->sync = true;
1126         dpolicy->ordered = false;
1127         dpolicy->granularity = granularity;
1128
1129         dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1130         dpolicy->io_aware_gran = MAX_PLIST_NUM;
1131         dpolicy->timeout = false;
1132
1133         if (discard_type == DPOLICY_BG) {
1134                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1135                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1136                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1137                 dpolicy->io_aware = true;
1138                 dpolicy->sync = false;
1139                 dpolicy->ordered = true;
1140                 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1141                         dpolicy->granularity = 1;
1142                         if (atomic_read(&dcc->discard_cmd_cnt))
1143                                 dpolicy->max_interval =
1144                                         DEF_MIN_DISCARD_ISSUE_TIME;
1145                 }
1146         } else if (discard_type == DPOLICY_FORCE) {
1147                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1148                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1149                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1150                 dpolicy->io_aware = false;
1151         } else if (discard_type == DPOLICY_FSTRIM) {
1152                 dpolicy->io_aware = false;
1153         } else if (discard_type == DPOLICY_UMOUNT) {
1154                 dpolicy->io_aware = false;
1155                 /* we need to issue all to keep CP_TRIMMED_FLAG */
1156                 dpolicy->granularity = 1;
1157                 dpolicy->timeout = true;
1158         }
1159 }
1160
1161 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1162                                 struct block_device *bdev, block_t lstart,
1163                                 block_t start, block_t len);
1164 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1165 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1166                                                 struct discard_policy *dpolicy,
1167                                                 struct discard_cmd *dc,
1168                                                 unsigned int *issued)
1169 {
1170         struct block_device *bdev = dc->bdev;
1171         struct request_queue *q = bdev_get_queue(bdev);
1172         unsigned int max_discard_blocks =
1173                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1174         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1175         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1176                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1177         int flag = dpolicy->sync ? REQ_SYNC : 0;
1178         block_t lstart, start, len, total_len;
1179         int err = 0;
1180
1181         if (dc->state != D_PREP)
1182                 return 0;
1183
1184         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1185                 return 0;
1186
1187         trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1188
1189         lstart = dc->lstart;
1190         start = dc->start;
1191         len = dc->len;
1192         total_len = len;
1193
1194         dc->len = 0;
1195
1196         while (total_len && *issued < dpolicy->max_requests && !err) {
1197                 struct bio *bio = NULL;
1198                 unsigned long flags;
1199                 bool last = true;
1200
1201                 if (len > max_discard_blocks) {
1202                         len = max_discard_blocks;
1203                         last = false;
1204                 }
1205
1206                 (*issued)++;
1207                 if (*issued == dpolicy->max_requests)
1208                         last = true;
1209
1210                 dc->len += len;
1211
1212                 if (time_to_inject(sbi, FAULT_DISCARD)) {
1213                         f2fs_show_injection_info(sbi, FAULT_DISCARD);
1214                         err = -EIO;
1215                         goto submit;
1216                 }
1217                 err = __blkdev_issue_discard(bdev,
1218                                         SECTOR_FROM_BLOCK(start),
1219                                         SECTOR_FROM_BLOCK(len),
1220                                         GFP_NOFS, 0, &bio);
1221 submit:
1222                 if (err) {
1223                         spin_lock_irqsave(&dc->lock, flags);
1224                         if (dc->state == D_PARTIAL)
1225                                 dc->state = D_SUBMIT;
1226                         spin_unlock_irqrestore(&dc->lock, flags);
1227
1228                         break;
1229                 }
1230
1231                 f2fs_bug_on(sbi, !bio);
1232
1233                 /*
1234                  * should keep before submission to avoid D_DONE
1235                  * right away
1236                  */
1237                 spin_lock_irqsave(&dc->lock, flags);
1238                 if (last)
1239                         dc->state = D_SUBMIT;
1240                 else
1241                         dc->state = D_PARTIAL;
1242                 dc->bio_ref++;
1243                 spin_unlock_irqrestore(&dc->lock, flags);
1244
1245                 atomic_inc(&dcc->queued_discard);
1246                 dc->queued++;
1247                 list_move_tail(&dc->list, wait_list);
1248
1249                 /* sanity check on discard range */
1250                 __check_sit_bitmap(sbi, lstart, lstart + len);
1251
1252                 bio->bi_private = dc;
1253                 bio->bi_end_io = f2fs_submit_discard_endio;
1254                 bio->bi_opf |= flag;
1255                 submit_bio(bio);
1256
1257                 atomic_inc(&dcc->issued_discard);
1258
1259                 f2fs_update_iostat(sbi, FS_DISCARD, 1);
1260
1261                 lstart += len;
1262                 start += len;
1263                 total_len -= len;
1264                 len = total_len;
1265         }
1266
1267         if (!err && len) {
1268                 dcc->undiscard_blks -= len;
1269                 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1270         }
1271         return err;
1272 }
1273
1274 static void __insert_discard_tree(struct f2fs_sb_info *sbi,
1275                                 struct block_device *bdev, block_t lstart,
1276                                 block_t start, block_t len,
1277                                 struct rb_node **insert_p,
1278                                 struct rb_node *insert_parent)
1279 {
1280         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1281         struct rb_node **p;
1282         struct rb_node *parent = NULL;
1283         bool leftmost = true;
1284
1285         if (insert_p && insert_parent) {
1286                 parent = insert_parent;
1287                 p = insert_p;
1288                 goto do_insert;
1289         }
1290
1291         p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1292                                                         lstart, &leftmost);
1293 do_insert:
1294         __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1295                                                                 p, leftmost);
1296 }
1297
1298 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1299                                                 struct discard_cmd *dc)
1300 {
1301         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1302 }
1303
1304 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1305                                 struct discard_cmd *dc, block_t blkaddr)
1306 {
1307         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1308         struct discard_info di = dc->di;
1309         bool modified = false;
1310
1311         if (dc->state == D_DONE || dc->len == 1) {
1312                 __remove_discard_cmd(sbi, dc);
1313                 return;
1314         }
1315
1316         dcc->undiscard_blks -= di.len;
1317
1318         if (blkaddr > di.lstart) {
1319                 dc->len = blkaddr - dc->lstart;
1320                 dcc->undiscard_blks += dc->len;
1321                 __relocate_discard_cmd(dcc, dc);
1322                 modified = true;
1323         }
1324
1325         if (blkaddr < di.lstart + di.len - 1) {
1326                 if (modified) {
1327                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1328                                         di.start + blkaddr + 1 - di.lstart,
1329                                         di.lstart + di.len - 1 - blkaddr,
1330                                         NULL, NULL);
1331                 } else {
1332                         dc->lstart++;
1333                         dc->len--;
1334                         dc->start++;
1335                         dcc->undiscard_blks += dc->len;
1336                         __relocate_discard_cmd(dcc, dc);
1337                 }
1338         }
1339 }
1340
1341 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1342                                 struct block_device *bdev, block_t lstart,
1343                                 block_t start, block_t len)
1344 {
1345         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1346         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1347         struct discard_cmd *dc;
1348         struct discard_info di = {0};
1349         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1350         struct request_queue *q = bdev_get_queue(bdev);
1351         unsigned int max_discard_blocks =
1352                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1353         block_t end = lstart + len;
1354
1355         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1356                                         NULL, lstart,
1357                                         (struct rb_entry **)&prev_dc,
1358                                         (struct rb_entry **)&next_dc,
1359                                         &insert_p, &insert_parent, true, NULL);
1360         if (dc)
1361                 prev_dc = dc;
1362
1363         if (!prev_dc) {
1364                 di.lstart = lstart;
1365                 di.len = next_dc ? next_dc->lstart - lstart : len;
1366                 di.len = min(di.len, len);
1367                 di.start = start;
1368         }
1369
1370         while (1) {
1371                 struct rb_node *node;
1372                 bool merged = false;
1373                 struct discard_cmd *tdc = NULL;
1374
1375                 if (prev_dc) {
1376                         di.lstart = prev_dc->lstart + prev_dc->len;
1377                         if (di.lstart < lstart)
1378                                 di.lstart = lstart;
1379                         if (di.lstart >= end)
1380                                 break;
1381
1382                         if (!next_dc || next_dc->lstart > end)
1383                                 di.len = end - di.lstart;
1384                         else
1385                                 di.len = next_dc->lstart - di.lstart;
1386                         di.start = start + di.lstart - lstart;
1387                 }
1388
1389                 if (!di.len)
1390                         goto next;
1391
1392                 if (prev_dc && prev_dc->state == D_PREP &&
1393                         prev_dc->bdev == bdev &&
1394                         __is_discard_back_mergeable(&di, &prev_dc->di,
1395                                                         max_discard_blocks)) {
1396                         prev_dc->di.len += di.len;
1397                         dcc->undiscard_blks += di.len;
1398                         __relocate_discard_cmd(dcc, prev_dc);
1399                         di = prev_dc->di;
1400                         tdc = prev_dc;
1401                         merged = true;
1402                 }
1403
1404                 if (next_dc && next_dc->state == D_PREP &&
1405                         next_dc->bdev == bdev &&
1406                         __is_discard_front_mergeable(&di, &next_dc->di,
1407                                                         max_discard_blocks)) {
1408                         next_dc->di.lstart = di.lstart;
1409                         next_dc->di.len += di.len;
1410                         next_dc->di.start = di.start;
1411                         dcc->undiscard_blks += di.len;
1412                         __relocate_discard_cmd(dcc, next_dc);
1413                         if (tdc)
1414                                 __remove_discard_cmd(sbi, tdc);
1415                         merged = true;
1416                 }
1417
1418                 if (!merged) {
1419                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1420                                                         di.len, NULL, NULL);
1421                 }
1422  next:
1423                 prev_dc = next_dc;
1424                 if (!prev_dc)
1425                         break;
1426
1427                 node = rb_next(&prev_dc->rb_node);
1428                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1429         }
1430 }
1431
1432 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1433                 struct block_device *bdev, block_t blkstart, block_t blklen)
1434 {
1435         block_t lblkstart = blkstart;
1436
1437         if (!f2fs_bdev_support_discard(bdev))
1438                 return 0;
1439
1440         trace_f2fs_queue_discard(bdev, blkstart, blklen);
1441
1442         if (f2fs_is_multi_device(sbi)) {
1443                 int devi = f2fs_target_device_index(sbi, blkstart);
1444
1445                 blkstart -= FDEV(devi).start_blk;
1446         }
1447         mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1448         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1449         mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1450         return 0;
1451 }
1452
1453 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1454                                         struct discard_policy *dpolicy)
1455 {
1456         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1457         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1458         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1459         struct discard_cmd *dc;
1460         struct blk_plug plug;
1461         unsigned int pos = dcc->next_pos;
1462         unsigned int issued = 0;
1463         bool io_interrupted = false;
1464
1465         mutex_lock(&dcc->cmd_lock);
1466         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1467                                         NULL, pos,
1468                                         (struct rb_entry **)&prev_dc,
1469                                         (struct rb_entry **)&next_dc,
1470                                         &insert_p, &insert_parent, true, NULL);
1471         if (!dc)
1472                 dc = next_dc;
1473
1474         blk_start_plug(&plug);
1475
1476         while (dc) {
1477                 struct rb_node *node;
1478                 int err = 0;
1479
1480                 if (dc->state != D_PREP)
1481                         goto next;
1482
1483                 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1484                         io_interrupted = true;
1485                         break;
1486                 }
1487
1488                 dcc->next_pos = dc->lstart + dc->len;
1489                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1490
1491                 if (issued >= dpolicy->max_requests)
1492                         break;
1493 next:
1494                 node = rb_next(&dc->rb_node);
1495                 if (err)
1496                         __remove_discard_cmd(sbi, dc);
1497                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1498         }
1499
1500         blk_finish_plug(&plug);
1501
1502         if (!dc)
1503                 dcc->next_pos = 0;
1504
1505         mutex_unlock(&dcc->cmd_lock);
1506
1507         if (!issued && io_interrupted)
1508                 issued = -1;
1509
1510         return issued;
1511 }
1512 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1513                                         struct discard_policy *dpolicy);
1514
1515 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1516                                         struct discard_policy *dpolicy)
1517 {
1518         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1519         struct list_head *pend_list;
1520         struct discard_cmd *dc, *tmp;
1521         struct blk_plug plug;
1522         int i, issued;
1523         bool io_interrupted = false;
1524
1525         if (dpolicy->timeout)
1526                 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1527
1528 retry:
1529         issued = 0;
1530         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1531                 if (dpolicy->timeout &&
1532                                 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1533                         break;
1534
1535                 if (i + 1 < dpolicy->granularity)
1536                         break;
1537
1538                 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1539                         return __issue_discard_cmd_orderly(sbi, dpolicy);
1540
1541                 pend_list = &dcc->pend_list[i];
1542
1543                 mutex_lock(&dcc->cmd_lock);
1544                 if (list_empty(pend_list))
1545                         goto next;
1546                 if (unlikely(dcc->rbtree_check))
1547                         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1548                                                         &dcc->root, false));
1549                 blk_start_plug(&plug);
1550                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1551                         f2fs_bug_on(sbi, dc->state != D_PREP);
1552
1553                         if (dpolicy->timeout &&
1554                                 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1555                                 break;
1556
1557                         if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1558                                                 !is_idle(sbi, DISCARD_TIME)) {
1559                                 io_interrupted = true;
1560                                 break;
1561                         }
1562
1563                         __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1564
1565                         if (issued >= dpolicy->max_requests)
1566                                 break;
1567                 }
1568                 blk_finish_plug(&plug);
1569 next:
1570                 mutex_unlock(&dcc->cmd_lock);
1571
1572                 if (issued >= dpolicy->max_requests || io_interrupted)
1573                         break;
1574         }
1575
1576         if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1577                 __wait_all_discard_cmd(sbi, dpolicy);
1578                 goto retry;
1579         }
1580
1581         if (!issued && io_interrupted)
1582                 issued = -1;
1583
1584         return issued;
1585 }
1586
1587 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1588 {
1589         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1590         struct list_head *pend_list;
1591         struct discard_cmd *dc, *tmp;
1592         int i;
1593         bool dropped = false;
1594
1595         mutex_lock(&dcc->cmd_lock);
1596         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1597                 pend_list = &dcc->pend_list[i];
1598                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1599                         f2fs_bug_on(sbi, dc->state != D_PREP);
1600                         __remove_discard_cmd(sbi, dc);
1601                         dropped = true;
1602                 }
1603         }
1604         mutex_unlock(&dcc->cmd_lock);
1605
1606         return dropped;
1607 }
1608
1609 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1610 {
1611         __drop_discard_cmd(sbi);
1612 }
1613
1614 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1615                                                         struct discard_cmd *dc)
1616 {
1617         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1618         unsigned int len = 0;
1619
1620         wait_for_completion_io(&dc->wait);
1621         mutex_lock(&dcc->cmd_lock);
1622         f2fs_bug_on(sbi, dc->state != D_DONE);
1623         dc->ref--;
1624         if (!dc->ref) {
1625                 if (!dc->error)
1626                         len = dc->len;
1627                 __remove_discard_cmd(sbi, dc);
1628         }
1629         mutex_unlock(&dcc->cmd_lock);
1630
1631         return len;
1632 }
1633
1634 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1635                                                 struct discard_policy *dpolicy,
1636                                                 block_t start, block_t end)
1637 {
1638         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1639         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1640                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1641         struct discard_cmd *dc, *tmp;
1642         bool need_wait;
1643         unsigned int trimmed = 0;
1644
1645 next:
1646         need_wait = false;
1647
1648         mutex_lock(&dcc->cmd_lock);
1649         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1650                 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1651                         continue;
1652                 if (dc->len < dpolicy->granularity)
1653                         continue;
1654                 if (dc->state == D_DONE && !dc->ref) {
1655                         wait_for_completion_io(&dc->wait);
1656                         if (!dc->error)
1657                                 trimmed += dc->len;
1658                         __remove_discard_cmd(sbi, dc);
1659                 } else {
1660                         dc->ref++;
1661                         need_wait = true;
1662                         break;
1663                 }
1664         }
1665         mutex_unlock(&dcc->cmd_lock);
1666
1667         if (need_wait) {
1668                 trimmed += __wait_one_discard_bio(sbi, dc);
1669                 goto next;
1670         }
1671
1672         return trimmed;
1673 }
1674
1675 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1676                                                 struct discard_policy *dpolicy)
1677 {
1678         struct discard_policy dp;
1679         unsigned int discard_blks;
1680
1681         if (dpolicy)
1682                 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1683
1684         /* wait all */
1685         __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1686         discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1687         __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1688         discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1689
1690         return discard_blks;
1691 }
1692
1693 /* This should be covered by global mutex, &sit_i->sentry_lock */
1694 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1695 {
1696         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1697         struct discard_cmd *dc;
1698         bool need_wait = false;
1699
1700         mutex_lock(&dcc->cmd_lock);
1701         dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1702                                                         NULL, blkaddr);
1703         if (dc) {
1704                 if (dc->state == D_PREP) {
1705                         __punch_discard_cmd(sbi, dc, blkaddr);
1706                 } else {
1707                         dc->ref++;
1708                         need_wait = true;
1709                 }
1710         }
1711         mutex_unlock(&dcc->cmd_lock);
1712
1713         if (need_wait)
1714                 __wait_one_discard_bio(sbi, dc);
1715 }
1716
1717 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1718 {
1719         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1720
1721         if (dcc && dcc->f2fs_issue_discard) {
1722                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1723
1724                 dcc->f2fs_issue_discard = NULL;
1725                 kthread_stop(discard_thread);
1726         }
1727 }
1728
1729 /* This comes from f2fs_put_super */
1730 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1731 {
1732         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1733         struct discard_policy dpolicy;
1734         bool dropped;
1735
1736         __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1737                                         dcc->discard_granularity);
1738         __issue_discard_cmd(sbi, &dpolicy);
1739         dropped = __drop_discard_cmd(sbi);
1740
1741         /* just to make sure there is no pending discard commands */
1742         __wait_all_discard_cmd(sbi, NULL);
1743
1744         f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1745         return dropped;
1746 }
1747
1748 static int issue_discard_thread(void *data)
1749 {
1750         struct f2fs_sb_info *sbi = data;
1751         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1752         wait_queue_head_t *q = &dcc->discard_wait_queue;
1753         struct discard_policy dpolicy;
1754         unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1755         int issued;
1756
1757         set_freezable();
1758
1759         do {
1760                 if (sbi->gc_mode == GC_URGENT_HIGH ||
1761                         !f2fs_available_free_memory(sbi, DISCARD_CACHE))
1762                         __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1763                 else
1764                         __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1765                                                 dcc->discard_granularity);
1766
1767                 if (!atomic_read(&dcc->discard_cmd_cnt))
1768                        wait_ms = dpolicy.max_interval;
1769
1770                 wait_event_interruptible_timeout(*q,
1771                                 kthread_should_stop() || freezing(current) ||
1772                                 dcc->discard_wake,
1773                                 msecs_to_jiffies(wait_ms));
1774
1775                 if (dcc->discard_wake)
1776                         dcc->discard_wake = 0;
1777
1778                 /* clean up pending candidates before going to sleep */
1779                 if (atomic_read(&dcc->queued_discard))
1780                         __wait_all_discard_cmd(sbi, NULL);
1781
1782                 if (try_to_freeze())
1783                         continue;
1784                 if (f2fs_readonly(sbi->sb))
1785                         continue;
1786                 if (kthread_should_stop())
1787                         return 0;
1788                 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1789                         wait_ms = dpolicy.max_interval;
1790                         continue;
1791                 }
1792                 if (!atomic_read(&dcc->discard_cmd_cnt))
1793                         continue;
1794
1795                 sb_start_intwrite(sbi->sb);
1796
1797                 issued = __issue_discard_cmd(sbi, &dpolicy);
1798                 if (issued > 0) {
1799                         __wait_all_discard_cmd(sbi, &dpolicy);
1800                         wait_ms = dpolicy.min_interval;
1801                 } else if (issued == -1) {
1802                         wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1803                         if (!wait_ms)
1804                                 wait_ms = dpolicy.mid_interval;
1805                 } else {
1806                         wait_ms = dpolicy.max_interval;
1807                 }
1808
1809                 sb_end_intwrite(sbi->sb);
1810
1811         } while (!kthread_should_stop());
1812         return 0;
1813 }
1814
1815 #ifdef CONFIG_BLK_DEV_ZONED
1816 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1817                 struct block_device *bdev, block_t blkstart, block_t blklen)
1818 {
1819         sector_t sector, nr_sects;
1820         block_t lblkstart = blkstart;
1821         int devi = 0;
1822
1823         if (f2fs_is_multi_device(sbi)) {
1824                 devi = f2fs_target_device_index(sbi, blkstart);
1825                 if (blkstart < FDEV(devi).start_blk ||
1826                     blkstart > FDEV(devi).end_blk) {
1827                         f2fs_err(sbi, "Invalid block %x", blkstart);
1828                         return -EIO;
1829                 }
1830                 blkstart -= FDEV(devi).start_blk;
1831         }
1832
1833         /* For sequential zones, reset the zone write pointer */
1834         if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1835                 sector = SECTOR_FROM_BLOCK(blkstart);
1836                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1837
1838                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1839                                 nr_sects != bdev_zone_sectors(bdev)) {
1840                         f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1841                                  devi, sbi->s_ndevs ? FDEV(devi).path : "",
1842                                  blkstart, blklen);
1843                         return -EIO;
1844                 }
1845                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1846                 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1847                                         sector, nr_sects, GFP_NOFS);
1848         }
1849
1850         /* For conventional zones, use regular discard if supported */
1851         return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1852 }
1853 #endif
1854
1855 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1856                 struct block_device *bdev, block_t blkstart, block_t blklen)
1857 {
1858 #ifdef CONFIG_BLK_DEV_ZONED
1859         if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1860                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1861 #endif
1862         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1863 }
1864
1865 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1866                                 block_t blkstart, block_t blklen)
1867 {
1868         sector_t start = blkstart, len = 0;
1869         struct block_device *bdev;
1870         struct seg_entry *se;
1871         unsigned int offset;
1872         block_t i;
1873         int err = 0;
1874
1875         bdev = f2fs_target_device(sbi, blkstart, NULL);
1876
1877         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1878                 if (i != start) {
1879                         struct block_device *bdev2 =
1880                                 f2fs_target_device(sbi, i, NULL);
1881
1882                         if (bdev2 != bdev) {
1883                                 err = __issue_discard_async(sbi, bdev,
1884                                                 start, len);
1885                                 if (err)
1886                                         return err;
1887                                 bdev = bdev2;
1888                                 start = i;
1889                                 len = 0;
1890                         }
1891                 }
1892
1893                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1894                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1895
1896                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1897                         sbi->discard_blks--;
1898         }
1899
1900         if (len)
1901                 err = __issue_discard_async(sbi, bdev, start, len);
1902         return err;
1903 }
1904
1905 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1906                                                         bool check_only)
1907 {
1908         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1909         int max_blocks = sbi->blocks_per_seg;
1910         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1911         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1912         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1913         unsigned long *discard_map = (unsigned long *)se->discard_map;
1914         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1915         unsigned int start = 0, end = -1;
1916         bool force = (cpc->reason & CP_DISCARD);
1917         struct discard_entry *de = NULL;
1918         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1919         int i;
1920
1921         if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
1922                 return false;
1923
1924         if (!force) {
1925                 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1926                         SM_I(sbi)->dcc_info->nr_discards >=
1927                                 SM_I(sbi)->dcc_info->max_discards)
1928                         return false;
1929         }
1930
1931         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1932         for (i = 0; i < entries; i++)
1933                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1934                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1935
1936         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1937                                 SM_I(sbi)->dcc_info->max_discards) {
1938                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1939                 if (start >= max_blocks)
1940                         break;
1941
1942                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1943                 if (force && start && end != max_blocks
1944                                         && (end - start) < cpc->trim_minlen)
1945                         continue;
1946
1947                 if (check_only)
1948                         return true;
1949
1950                 if (!de) {
1951                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1952                                                                 GFP_F2FS_ZERO);
1953                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1954                         list_add_tail(&de->list, head);
1955                 }
1956
1957                 for (i = start; i < end; i++)
1958                         __set_bit_le(i, (void *)de->discard_map);
1959
1960                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1961         }
1962         return false;
1963 }
1964
1965 static void release_discard_addr(struct discard_entry *entry)
1966 {
1967         list_del(&entry->list);
1968         kmem_cache_free(discard_entry_slab, entry);
1969 }
1970
1971 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1972 {
1973         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1974         struct discard_entry *entry, *this;
1975
1976         /* drop caches */
1977         list_for_each_entry_safe(entry, this, head, list)
1978                 release_discard_addr(entry);
1979 }
1980
1981 /*
1982  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1983  */
1984 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1985 {
1986         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1987         unsigned int segno;
1988
1989         mutex_lock(&dirty_i->seglist_lock);
1990         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1991                 __set_test_and_free(sbi, segno, false);
1992         mutex_unlock(&dirty_i->seglist_lock);
1993 }
1994
1995 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1996                                                 struct cp_control *cpc)
1997 {
1998         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1999         struct list_head *head = &dcc->entry_list;
2000         struct discard_entry *entry, *this;
2001         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2002         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
2003         unsigned int start = 0, end = -1;
2004         unsigned int secno, start_segno;
2005         bool force = (cpc->reason & CP_DISCARD);
2006         bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
2007
2008         mutex_lock(&dirty_i->seglist_lock);
2009
2010         while (1) {
2011                 int i;
2012
2013                 if (need_align && end != -1)
2014                         end--;
2015                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
2016                 if (start >= MAIN_SEGS(sbi))
2017                         break;
2018                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
2019                                                                 start + 1);
2020
2021                 if (need_align) {
2022                         start = rounddown(start, sbi->segs_per_sec);
2023                         end = roundup(end, sbi->segs_per_sec);
2024                 }
2025
2026                 for (i = start; i < end; i++) {
2027                         if (test_and_clear_bit(i, prefree_map))
2028                                 dirty_i->nr_dirty[PRE]--;
2029                 }
2030
2031                 if (!f2fs_realtime_discard_enable(sbi))
2032                         continue;
2033
2034                 if (force && start >= cpc->trim_start &&
2035                                         (end - 1) <= cpc->trim_end)
2036                                 continue;
2037
2038                 if (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi)) {
2039                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2040                                 (end - start) << sbi->log_blocks_per_seg);
2041                         continue;
2042                 }
2043 next:
2044                 secno = GET_SEC_FROM_SEG(sbi, start);
2045                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
2046                 if (!IS_CURSEC(sbi, secno) &&
2047                         !get_valid_blocks(sbi, start, true))
2048                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2049                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
2050
2051                 start = start_segno + sbi->segs_per_sec;
2052                 if (start < end)
2053                         goto next;
2054                 else
2055                         end = start - 1;
2056         }
2057         mutex_unlock(&dirty_i->seglist_lock);
2058
2059         /* send small discards */
2060         list_for_each_entry_safe(entry, this, head, list) {
2061                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2062                 bool is_valid = test_bit_le(0, entry->discard_map);
2063
2064 find_next:
2065                 if (is_valid) {
2066                         next_pos = find_next_zero_bit_le(entry->discard_map,
2067                                         sbi->blocks_per_seg, cur_pos);
2068                         len = next_pos - cur_pos;
2069
2070                         if (f2fs_sb_has_blkzoned(sbi) ||
2071                             (force && len < cpc->trim_minlen))
2072                                 goto skip;
2073
2074                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2075                                                                         len);
2076                         total_len += len;
2077                 } else {
2078                         next_pos = find_next_bit_le(entry->discard_map,
2079                                         sbi->blocks_per_seg, cur_pos);
2080                 }
2081 skip:
2082                 cur_pos = next_pos;
2083                 is_valid = !is_valid;
2084
2085                 if (cur_pos < sbi->blocks_per_seg)
2086                         goto find_next;
2087
2088                 release_discard_addr(entry);
2089                 dcc->nr_discards -= total_len;
2090         }
2091
2092         wake_up_discard_thread(sbi, false);
2093 }
2094
2095 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2096 {
2097         dev_t dev = sbi->sb->s_bdev->bd_dev;
2098         struct discard_cmd_control *dcc;
2099         int err = 0, i;
2100
2101         if (SM_I(sbi)->dcc_info) {
2102                 dcc = SM_I(sbi)->dcc_info;
2103                 goto init_thread;
2104         }
2105
2106         dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2107         if (!dcc)
2108                 return -ENOMEM;
2109
2110         dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2111         INIT_LIST_HEAD(&dcc->entry_list);
2112         for (i = 0; i < MAX_PLIST_NUM; i++)
2113                 INIT_LIST_HEAD(&dcc->pend_list[i]);
2114         INIT_LIST_HEAD(&dcc->wait_list);
2115         INIT_LIST_HEAD(&dcc->fstrim_list);
2116         mutex_init(&dcc->cmd_lock);
2117         atomic_set(&dcc->issued_discard, 0);
2118         atomic_set(&dcc->queued_discard, 0);
2119         atomic_set(&dcc->discard_cmd_cnt, 0);
2120         dcc->nr_discards = 0;
2121         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2122         dcc->undiscard_blks = 0;
2123         dcc->next_pos = 0;
2124         dcc->root = RB_ROOT_CACHED;
2125         dcc->rbtree_check = false;
2126
2127         init_waitqueue_head(&dcc->discard_wait_queue);
2128         SM_I(sbi)->dcc_info = dcc;
2129 init_thread:
2130         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2131                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2132         if (IS_ERR(dcc->f2fs_issue_discard)) {
2133                 err = PTR_ERR(dcc->f2fs_issue_discard);
2134                 kfree(dcc);
2135                 SM_I(sbi)->dcc_info = NULL;
2136                 return err;
2137         }
2138
2139         return err;
2140 }
2141
2142 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2143 {
2144         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2145
2146         if (!dcc)
2147                 return;
2148
2149         f2fs_stop_discard_thread(sbi);
2150
2151         /*
2152          * Recovery can cache discard commands, so in error path of
2153          * fill_super(), it needs to give a chance to handle them.
2154          */
2155         if (unlikely(atomic_read(&dcc->discard_cmd_cnt)))
2156                 f2fs_issue_discard_timeout(sbi);
2157
2158         kfree(dcc);
2159         SM_I(sbi)->dcc_info = NULL;
2160 }
2161
2162 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2163 {
2164         struct sit_info *sit_i = SIT_I(sbi);
2165
2166         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2167                 sit_i->dirty_sentries++;
2168                 return false;
2169         }
2170
2171         return true;
2172 }
2173
2174 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2175                                         unsigned int segno, int modified)
2176 {
2177         struct seg_entry *se = get_seg_entry(sbi, segno);
2178
2179         se->type = type;
2180         if (modified)
2181                 __mark_sit_entry_dirty(sbi, segno);
2182 }
2183
2184 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2185                                                                 block_t blkaddr)
2186 {
2187         unsigned int segno = GET_SEGNO(sbi, blkaddr);
2188
2189         if (segno == NULL_SEGNO)
2190                 return 0;
2191         return get_seg_entry(sbi, segno)->mtime;
2192 }
2193
2194 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2195                                                 unsigned long long old_mtime)
2196 {
2197         struct seg_entry *se;
2198         unsigned int segno = GET_SEGNO(sbi, blkaddr);
2199         unsigned long long ctime = get_mtime(sbi, false);
2200         unsigned long long mtime = old_mtime ? old_mtime : ctime;
2201
2202         if (segno == NULL_SEGNO)
2203                 return;
2204
2205         se = get_seg_entry(sbi, segno);
2206
2207         if (!se->mtime)
2208                 se->mtime = mtime;
2209         else
2210                 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2211                                                 se->valid_blocks + 1);
2212
2213         if (ctime > SIT_I(sbi)->max_mtime)
2214                 SIT_I(sbi)->max_mtime = ctime;
2215 }
2216
2217 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2218 {
2219         struct seg_entry *se;
2220         unsigned int segno, offset;
2221         long int new_vblocks;
2222         bool exist;
2223 #ifdef CONFIG_F2FS_CHECK_FS
2224         bool mir_exist;
2225 #endif
2226
2227         segno = GET_SEGNO(sbi, blkaddr);
2228
2229         se = get_seg_entry(sbi, segno);
2230         new_vblocks = se->valid_blocks + del;
2231         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2232
2233         f2fs_bug_on(sbi, (new_vblocks < 0 ||
2234                         (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2235
2236         se->valid_blocks = new_vblocks;
2237
2238         /* Update valid block bitmap */
2239         if (del > 0) {
2240                 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2241 #ifdef CONFIG_F2FS_CHECK_FS
2242                 mir_exist = f2fs_test_and_set_bit(offset,
2243                                                 se->cur_valid_map_mir);
2244                 if (unlikely(exist != mir_exist)) {
2245                         f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2246                                  blkaddr, exist);
2247                         f2fs_bug_on(sbi, 1);
2248                 }
2249 #endif
2250                 if (unlikely(exist)) {
2251                         f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2252                                  blkaddr);
2253                         f2fs_bug_on(sbi, 1);
2254                         se->valid_blocks--;
2255                         del = 0;
2256                 }
2257
2258                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
2259                         sbi->discard_blks--;
2260
2261                 /*
2262                  * SSR should never reuse block which is checkpointed
2263                  * or newly invalidated.
2264                  */
2265                 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2266                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2267                                 se->ckpt_valid_blocks++;
2268                 }
2269         } else {
2270                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2271 #ifdef CONFIG_F2FS_CHECK_FS
2272                 mir_exist = f2fs_test_and_clear_bit(offset,
2273                                                 se->cur_valid_map_mir);
2274                 if (unlikely(exist != mir_exist)) {
2275                         f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2276                                  blkaddr, exist);
2277                         f2fs_bug_on(sbi, 1);
2278                 }
2279 #endif
2280                 if (unlikely(!exist)) {
2281                         f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2282                                  blkaddr);
2283                         f2fs_bug_on(sbi, 1);
2284                         se->valid_blocks++;
2285                         del = 0;
2286                 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2287                         /*
2288                          * If checkpoints are off, we must not reuse data that
2289                          * was used in the previous checkpoint. If it was used
2290                          * before, we must track that to know how much space we
2291                          * really have.
2292                          */
2293                         if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2294                                 spin_lock(&sbi->stat_lock);
2295                                 sbi->unusable_block_count++;
2296                                 spin_unlock(&sbi->stat_lock);
2297                         }
2298                 }
2299
2300                 if (f2fs_test_and_clear_bit(offset, se->discard_map))
2301                         sbi->discard_blks++;
2302         }
2303         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2304                 se->ckpt_valid_blocks += del;
2305
2306         __mark_sit_entry_dirty(sbi, segno);
2307
2308         /* update total number of valid blocks to be written in ckpt area */
2309         SIT_I(sbi)->written_valid_blocks += del;
2310
2311         if (__is_large_section(sbi))
2312                 get_sec_entry(sbi, segno)->valid_blocks += del;
2313 }
2314
2315 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2316 {
2317         unsigned int segno = GET_SEGNO(sbi, addr);
2318         struct sit_info *sit_i = SIT_I(sbi);
2319
2320         f2fs_bug_on(sbi, addr == NULL_ADDR);
2321         if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2322                 return;
2323
2324         invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2325
2326         /* add it into sit main buffer */
2327         down_write(&sit_i->sentry_lock);
2328
2329         update_segment_mtime(sbi, addr, 0);
2330         update_sit_entry(sbi, addr, -1);
2331
2332         /* add it into dirty seglist */
2333         locate_dirty_segment(sbi, segno);
2334
2335         up_write(&sit_i->sentry_lock);
2336 }
2337
2338 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2339 {
2340         struct sit_info *sit_i = SIT_I(sbi);
2341         unsigned int segno, offset;
2342         struct seg_entry *se;
2343         bool is_cp = false;
2344
2345         if (!__is_valid_data_blkaddr(blkaddr))
2346                 return true;
2347
2348         down_read(&sit_i->sentry_lock);
2349
2350         segno = GET_SEGNO(sbi, blkaddr);
2351         se = get_seg_entry(sbi, segno);
2352         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2353
2354         if (f2fs_test_bit(offset, se->ckpt_valid_map))
2355                 is_cp = true;
2356
2357         up_read(&sit_i->sentry_lock);
2358
2359         return is_cp;
2360 }
2361
2362 /*
2363  * This function should be resided under the curseg_mutex lock
2364  */
2365 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2366                                         struct f2fs_summary *sum)
2367 {
2368         struct curseg_info *curseg = CURSEG_I(sbi, type);
2369         void *addr = curseg->sum_blk;
2370
2371         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2372         memcpy(addr, sum, sizeof(struct f2fs_summary));
2373 }
2374
2375 /*
2376  * Calculate the number of current summary pages for writing
2377  */
2378 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2379 {
2380         int valid_sum_count = 0;
2381         int i, sum_in_page;
2382
2383         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2384                 if (sbi->ckpt->alloc_type[i] == SSR)
2385                         valid_sum_count += sbi->blocks_per_seg;
2386                 else {
2387                         if (for_ra)
2388                                 valid_sum_count += le16_to_cpu(
2389                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2390                         else
2391                                 valid_sum_count += curseg_blkoff(sbi, i);
2392                 }
2393         }
2394
2395         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2396                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2397         if (valid_sum_count <= sum_in_page)
2398                 return 1;
2399         else if ((valid_sum_count - sum_in_page) <=
2400                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2401                 return 2;
2402         return 3;
2403 }
2404
2405 /*
2406  * Caller should put this summary page
2407  */
2408 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2409 {
2410         if (unlikely(f2fs_cp_error(sbi)))
2411                 return ERR_PTR(-EIO);
2412         return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno));
2413 }
2414
2415 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2416                                         void *src, block_t blk_addr)
2417 {
2418         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2419
2420         memcpy(page_address(page), src, PAGE_SIZE);
2421         set_page_dirty(page);
2422         f2fs_put_page(page, 1);
2423 }
2424
2425 static void write_sum_page(struct f2fs_sb_info *sbi,
2426                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
2427 {
2428         f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2429 }
2430
2431 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2432                                                 int type, block_t blk_addr)
2433 {
2434         struct curseg_info *curseg = CURSEG_I(sbi, type);
2435         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2436         struct f2fs_summary_block *src = curseg->sum_blk;
2437         struct f2fs_summary_block *dst;
2438
2439         dst = (struct f2fs_summary_block *)page_address(page);
2440         memset(dst, 0, PAGE_SIZE);
2441
2442         mutex_lock(&curseg->curseg_mutex);
2443
2444         down_read(&curseg->journal_rwsem);
2445         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2446         up_read(&curseg->journal_rwsem);
2447
2448         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2449         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2450
2451         mutex_unlock(&curseg->curseg_mutex);
2452
2453         set_page_dirty(page);
2454         f2fs_put_page(page, 1);
2455 }
2456
2457 static int is_next_segment_free(struct f2fs_sb_info *sbi,
2458                                 struct curseg_info *curseg, int type)
2459 {
2460         unsigned int segno = curseg->segno + 1;
2461         struct free_segmap_info *free_i = FREE_I(sbi);
2462
2463         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2464                 return !test_bit(segno, free_i->free_segmap);
2465         return 0;
2466 }
2467
2468 /*
2469  * Find a new segment from the free segments bitmap to right order
2470  * This function should be returned with success, otherwise BUG
2471  */
2472 static void get_new_segment(struct f2fs_sb_info *sbi,
2473                         unsigned int *newseg, bool new_sec, int dir)
2474 {
2475         struct free_segmap_info *free_i = FREE_I(sbi);
2476         unsigned int segno, secno, zoneno;
2477         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2478         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2479         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2480         unsigned int left_start = hint;
2481         bool init = true;
2482         int go_left = 0;
2483         int i;
2484
2485         spin_lock(&free_i->segmap_lock);
2486
2487         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2488                 segno = find_next_zero_bit(free_i->free_segmap,
2489                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2490                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2491                         goto got_it;
2492         }
2493 find_other_zone:
2494         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2495         if (secno >= MAIN_SECS(sbi)) {
2496                 if (dir == ALLOC_RIGHT) {
2497                         secno = find_next_zero_bit(free_i->free_secmap,
2498                                                         MAIN_SECS(sbi), 0);
2499                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2500                 } else {
2501                         go_left = 1;
2502                         left_start = hint - 1;
2503                 }
2504         }
2505         if (go_left == 0)
2506                 goto skip_left;
2507
2508         while (test_bit(left_start, free_i->free_secmap)) {
2509                 if (left_start > 0) {
2510                         left_start--;
2511                         continue;
2512                 }
2513                 left_start = find_next_zero_bit(free_i->free_secmap,
2514                                                         MAIN_SECS(sbi), 0);
2515                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2516                 break;
2517         }
2518         secno = left_start;
2519 skip_left:
2520         segno = GET_SEG_FROM_SEC(sbi, secno);
2521         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2522
2523         /* give up on finding another zone */
2524         if (!init)
2525                 goto got_it;
2526         if (sbi->secs_per_zone == 1)
2527                 goto got_it;
2528         if (zoneno == old_zoneno)
2529                 goto got_it;
2530         if (dir == ALLOC_LEFT) {
2531                 if (!go_left && zoneno + 1 >= total_zones)
2532                         goto got_it;
2533                 if (go_left && zoneno == 0)
2534                         goto got_it;
2535         }
2536         for (i = 0; i < NR_CURSEG_TYPE; i++)
2537                 if (CURSEG_I(sbi, i)->zone == zoneno)
2538                         break;
2539
2540         if (i < NR_CURSEG_TYPE) {
2541                 /* zone is in user, try another */
2542                 if (go_left)
2543                         hint = zoneno * sbi->secs_per_zone - 1;
2544                 else if (zoneno + 1 >= total_zones)
2545                         hint = 0;
2546                 else
2547                         hint = (zoneno + 1) * sbi->secs_per_zone;
2548                 init = false;
2549                 goto find_other_zone;
2550         }
2551 got_it:
2552         /* set it as dirty segment in free segmap */
2553         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2554         __set_inuse(sbi, segno);
2555         *newseg = segno;
2556         spin_unlock(&free_i->segmap_lock);
2557 }
2558
2559 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2560 {
2561         struct curseg_info *curseg = CURSEG_I(sbi, type);
2562         struct summary_footer *sum_footer;
2563         unsigned short seg_type = curseg->seg_type;
2564
2565         curseg->inited = true;
2566         curseg->segno = curseg->next_segno;
2567         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2568         curseg->next_blkoff = 0;
2569         curseg->next_segno = NULL_SEGNO;
2570
2571         sum_footer = &(curseg->sum_blk->footer);
2572         memset(sum_footer, 0, sizeof(struct summary_footer));
2573
2574         sanity_check_seg_type(sbi, seg_type);
2575
2576         if (IS_DATASEG(seg_type))
2577                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2578         if (IS_NODESEG(seg_type))
2579                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2580         __set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
2581 }
2582
2583 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2584 {
2585         struct curseg_info *curseg = CURSEG_I(sbi, type);
2586         unsigned short seg_type = curseg->seg_type;
2587
2588         sanity_check_seg_type(sbi, seg_type);
2589
2590         /* if segs_per_sec is large than 1, we need to keep original policy. */
2591         if (__is_large_section(sbi))
2592                 return curseg->segno;
2593
2594         /* inmem log may not locate on any segment after mount */
2595         if (!curseg->inited)
2596                 return 0;
2597
2598         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2599                 return 0;
2600
2601         if (test_opt(sbi, NOHEAP) &&
2602                 (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type)))
2603                 return 0;
2604
2605         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2606                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2607
2608         /* find segments from 0 to reuse freed segments */
2609         if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2610                 return 0;
2611
2612         return curseg->segno;
2613 }
2614
2615 /*
2616  * Allocate a current working segment.
2617  * This function always allocates a free segment in LFS manner.
2618  */
2619 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2620 {
2621         struct curseg_info *curseg = CURSEG_I(sbi, type);
2622         unsigned short seg_type = curseg->seg_type;
2623         unsigned int segno = curseg->segno;
2624         int dir = ALLOC_LEFT;
2625
2626         if (curseg->inited)
2627                 write_sum_page(sbi, curseg->sum_blk,
2628                                 GET_SUM_BLOCK(sbi, segno));
2629         if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA)
2630                 dir = ALLOC_RIGHT;
2631
2632         if (test_opt(sbi, NOHEAP))
2633                 dir = ALLOC_RIGHT;
2634
2635         segno = __get_next_segno(sbi, type);
2636         get_new_segment(sbi, &segno, new_sec, dir);
2637         curseg->next_segno = segno;
2638         reset_curseg(sbi, type, 1);
2639         curseg->alloc_type = LFS;
2640 }
2641
2642 static int __next_free_blkoff(struct f2fs_sb_info *sbi,
2643                                         int segno, block_t start)
2644 {
2645         struct seg_entry *se = get_seg_entry(sbi, segno);
2646         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2647         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2648         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2649         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2650         int i;
2651
2652         for (i = 0; i < entries; i++)
2653                 target_map[i] = ckpt_map[i] | cur_map[i];
2654
2655         return __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2656 }
2657
2658 /*
2659  * If a segment is written by LFS manner, next block offset is just obtained
2660  * by increasing the current block offset. However, if a segment is written by
2661  * SSR manner, next block offset obtained by calling __next_free_blkoff
2662  */
2663 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2664                                 struct curseg_info *seg)
2665 {
2666         if (seg->alloc_type == SSR)
2667                 seg->next_blkoff =
2668                         __next_free_blkoff(sbi, seg->segno,
2669                                                 seg->next_blkoff + 1);
2670         else
2671                 seg->next_blkoff++;
2672 }
2673
2674 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno)
2675 {
2676         return __next_free_blkoff(sbi, segno, 0) < sbi->blocks_per_seg;
2677 }
2678
2679 /*
2680  * This function always allocates a used segment(from dirty seglist) by SSR
2681  * manner, so it should recover the existing segment information of valid blocks
2682  */
2683 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool flush)
2684 {
2685         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2686         struct curseg_info *curseg = CURSEG_I(sbi, type);
2687         unsigned int new_segno = curseg->next_segno;
2688         struct f2fs_summary_block *sum_node;
2689         struct page *sum_page;
2690
2691         if (flush)
2692                 write_sum_page(sbi, curseg->sum_blk,
2693                                         GET_SUM_BLOCK(sbi, curseg->segno));
2694
2695         __set_test_and_inuse(sbi, new_segno);
2696
2697         mutex_lock(&dirty_i->seglist_lock);
2698         __remove_dirty_segment(sbi, new_segno, PRE);
2699         __remove_dirty_segment(sbi, new_segno, DIRTY);
2700         mutex_unlock(&dirty_i->seglist_lock);
2701
2702         reset_curseg(sbi, type, 1);
2703         curseg->alloc_type = SSR;
2704         curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0);
2705
2706         sum_page = f2fs_get_sum_page(sbi, new_segno);
2707         if (IS_ERR(sum_page)) {
2708                 /* GC won't be able to use stale summary pages by cp_error */
2709                 memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
2710                 return;
2711         }
2712         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2713         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2714         f2fs_put_page(sum_page, 1);
2715 }
2716
2717 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2718                                 int alloc_mode, unsigned long long age);
2719
2720 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type,
2721                                         int target_type, int alloc_mode,
2722                                         unsigned long long age)
2723 {
2724         struct curseg_info *curseg = CURSEG_I(sbi, type);
2725
2726         curseg->seg_type = target_type;
2727
2728         if (get_ssr_segment(sbi, type, alloc_mode, age)) {
2729                 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
2730
2731                 curseg->seg_type = se->type;
2732                 change_curseg(sbi, type, true);
2733         } else {
2734                 /* allocate cold segment by default */
2735                 curseg->seg_type = CURSEG_COLD_DATA;
2736                 new_curseg(sbi, type, true);
2737         }
2738         stat_inc_seg_type(sbi, curseg);
2739 }
2740
2741 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
2742 {
2743         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
2744
2745         if (!sbi->am.atgc_enabled)
2746                 return;
2747
2748         down_read(&SM_I(sbi)->curseg_lock);
2749
2750         mutex_lock(&curseg->curseg_mutex);
2751         down_write(&SIT_I(sbi)->sentry_lock);
2752
2753         get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0);
2754
2755         up_write(&SIT_I(sbi)->sentry_lock);
2756         mutex_unlock(&curseg->curseg_mutex);
2757
2758         up_read(&SM_I(sbi)->curseg_lock);
2759
2760 }
2761 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
2762 {
2763         __f2fs_init_atgc_curseg(sbi);
2764 }
2765
2766 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2767 {
2768         struct curseg_info *curseg = CURSEG_I(sbi, type);
2769
2770         mutex_lock(&curseg->curseg_mutex);
2771         if (!curseg->inited)
2772                 goto out;
2773
2774         if (get_valid_blocks(sbi, curseg->segno, false)) {
2775                 write_sum_page(sbi, curseg->sum_blk,
2776                                 GET_SUM_BLOCK(sbi, curseg->segno));
2777         } else {
2778                 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2779                 __set_test_and_free(sbi, curseg->segno, true);
2780                 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2781         }
2782 out:
2783         mutex_unlock(&curseg->curseg_mutex);
2784 }
2785
2786 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
2787 {
2788         __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2789
2790         if (sbi->am.atgc_enabled)
2791                 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2792 }
2793
2794 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2795 {
2796         struct curseg_info *curseg = CURSEG_I(sbi, type);
2797
2798         mutex_lock(&curseg->curseg_mutex);
2799         if (!curseg->inited)
2800                 goto out;
2801         if (get_valid_blocks(sbi, curseg->segno, false))
2802                 goto out;
2803
2804         mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2805         __set_test_and_inuse(sbi, curseg->segno);
2806         mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2807 out:
2808         mutex_unlock(&curseg->curseg_mutex);
2809 }
2810
2811 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
2812 {
2813         __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2814
2815         if (sbi->am.atgc_enabled)
2816                 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2817 }
2818
2819 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2820                                 int alloc_mode, unsigned long long age)
2821 {
2822         struct curseg_info *curseg = CURSEG_I(sbi, type);
2823         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2824         unsigned segno = NULL_SEGNO;
2825         unsigned short seg_type = curseg->seg_type;
2826         int i, cnt;
2827         bool reversed = false;
2828
2829         sanity_check_seg_type(sbi, seg_type);
2830
2831         /* f2fs_need_SSR() already forces to do this */
2832         if (!v_ops->get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) {
2833                 curseg->next_segno = segno;
2834                 return 1;
2835         }
2836
2837         /* For node segments, let's do SSR more intensively */
2838         if (IS_NODESEG(seg_type)) {
2839                 if (seg_type >= CURSEG_WARM_NODE) {
2840                         reversed = true;
2841                         i = CURSEG_COLD_NODE;
2842                 } else {
2843                         i = CURSEG_HOT_NODE;
2844                 }
2845                 cnt = NR_CURSEG_NODE_TYPE;
2846         } else {
2847                 if (seg_type >= CURSEG_WARM_DATA) {
2848                         reversed = true;
2849                         i = CURSEG_COLD_DATA;
2850                 } else {
2851                         i = CURSEG_HOT_DATA;
2852                 }
2853                 cnt = NR_CURSEG_DATA_TYPE;
2854         }
2855
2856         for (; cnt-- > 0; reversed ? i-- : i++) {
2857                 if (i == seg_type)
2858                         continue;
2859                 if (!v_ops->get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) {
2860                         curseg->next_segno = segno;
2861                         return 1;
2862                 }
2863         }
2864
2865         /* find valid_blocks=0 in dirty list */
2866         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2867                 segno = get_free_segment(sbi);
2868                 if (segno != NULL_SEGNO) {
2869                         curseg->next_segno = segno;
2870                         return 1;
2871                 }
2872         }
2873         return 0;
2874 }
2875
2876 /*
2877  * flush out current segment and replace it with new segment
2878  * This function should be returned with success, otherwise BUG
2879  */
2880 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2881                                                 int type, bool force)
2882 {
2883         struct curseg_info *curseg = CURSEG_I(sbi, type);
2884
2885         if (force)
2886                 new_curseg(sbi, type, true);
2887         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2888                                         curseg->seg_type == CURSEG_WARM_NODE)
2889                 new_curseg(sbi, type, false);
2890         else if (curseg->alloc_type == LFS &&
2891                         is_next_segment_free(sbi, curseg, type) &&
2892                         likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2893                 new_curseg(sbi, type, false);
2894         else if (f2fs_need_SSR(sbi) &&
2895                         get_ssr_segment(sbi, type, SSR, 0))
2896                 change_curseg(sbi, type, true);
2897         else
2898                 new_curseg(sbi, type, false);
2899
2900         stat_inc_seg_type(sbi, curseg);
2901 }
2902
2903 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2904                                         unsigned int start, unsigned int end)
2905 {
2906         struct curseg_info *curseg = CURSEG_I(sbi, type);
2907         unsigned int segno;
2908
2909         down_read(&SM_I(sbi)->curseg_lock);
2910         mutex_lock(&curseg->curseg_mutex);
2911         down_write(&SIT_I(sbi)->sentry_lock);
2912
2913         segno = CURSEG_I(sbi, type)->segno;
2914         if (segno < start || segno > end)
2915                 goto unlock;
2916
2917         if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
2918                 change_curseg(sbi, type, true);
2919         else
2920                 new_curseg(sbi, type, true);
2921
2922         stat_inc_seg_type(sbi, curseg);
2923
2924         locate_dirty_segment(sbi, segno);
2925 unlock:
2926         up_write(&SIT_I(sbi)->sentry_lock);
2927
2928         if (segno != curseg->segno)
2929                 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2930                             type, segno, curseg->segno);
2931
2932         mutex_unlock(&curseg->curseg_mutex);
2933         up_read(&SM_I(sbi)->curseg_lock);
2934 }
2935
2936 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type,
2937                                                 bool new_sec, bool force)
2938 {
2939         struct curseg_info *curseg = CURSEG_I(sbi, type);
2940         unsigned int old_segno;
2941
2942         if (!curseg->inited)
2943                 goto alloc;
2944
2945         if (force || curseg->next_blkoff ||
2946                 get_valid_blocks(sbi, curseg->segno, new_sec))
2947                 goto alloc;
2948
2949         if (!get_ckpt_valid_blocks(sbi, curseg->segno, new_sec))
2950                 return;
2951 alloc:
2952         old_segno = curseg->segno;
2953         SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
2954         locate_dirty_segment(sbi, old_segno);
2955 }
2956
2957 static void __allocate_new_section(struct f2fs_sb_info *sbi,
2958                                                 int type, bool force)
2959 {
2960         __allocate_new_segment(sbi, type, true, force);
2961 }
2962
2963 void f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force)
2964 {
2965         down_read(&SM_I(sbi)->curseg_lock);
2966         down_write(&SIT_I(sbi)->sentry_lock);
2967         __allocate_new_section(sbi, type, force);
2968         up_write(&SIT_I(sbi)->sentry_lock);
2969         up_read(&SM_I(sbi)->curseg_lock);
2970 }
2971
2972 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2973 {
2974         int i;
2975
2976         down_read(&SM_I(sbi)->curseg_lock);
2977         down_write(&SIT_I(sbi)->sentry_lock);
2978         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
2979                 __allocate_new_segment(sbi, i, false, false);
2980         up_write(&SIT_I(sbi)->sentry_lock);
2981         up_read(&SM_I(sbi)->curseg_lock);
2982 }
2983
2984 static const struct segment_allocation default_salloc_ops = {
2985         .allocate_segment = allocate_segment_by_default,
2986 };
2987
2988 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2989                                                 struct cp_control *cpc)
2990 {
2991         __u64 trim_start = cpc->trim_start;
2992         bool has_candidate = false;
2993
2994         down_write(&SIT_I(sbi)->sentry_lock);
2995         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2996                 if (add_discard_addrs(sbi, cpc, true)) {
2997                         has_candidate = true;
2998                         break;
2999                 }
3000         }
3001         up_write(&SIT_I(sbi)->sentry_lock);
3002
3003         cpc->trim_start = trim_start;
3004         return has_candidate;
3005 }
3006
3007 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
3008                                         struct discard_policy *dpolicy,
3009                                         unsigned int start, unsigned int end)
3010 {
3011         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
3012         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
3013         struct rb_node **insert_p = NULL, *insert_parent = NULL;
3014         struct discard_cmd *dc;
3015         struct blk_plug plug;
3016         int issued;
3017         unsigned int trimmed = 0;
3018
3019 next:
3020         issued = 0;
3021
3022         mutex_lock(&dcc->cmd_lock);
3023         if (unlikely(dcc->rbtree_check))
3024                 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
3025                                                         &dcc->root, false));
3026
3027         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
3028                                         NULL, start,
3029                                         (struct rb_entry **)&prev_dc,
3030                                         (struct rb_entry **)&next_dc,
3031                                         &insert_p, &insert_parent, true, NULL);
3032         if (!dc)
3033                 dc = next_dc;
3034
3035         blk_start_plug(&plug);
3036
3037         while (dc && dc->lstart <= end) {
3038                 struct rb_node *node;
3039                 int err = 0;
3040
3041                 if (dc->len < dpolicy->granularity)
3042                         goto skip;
3043
3044                 if (dc->state != D_PREP) {
3045                         list_move_tail(&dc->list, &dcc->fstrim_list);
3046                         goto skip;
3047                 }
3048
3049                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
3050
3051                 if (issued >= dpolicy->max_requests) {
3052                         start = dc->lstart + dc->len;
3053
3054                         if (err)
3055                                 __remove_discard_cmd(sbi, dc);
3056
3057                         blk_finish_plug(&plug);
3058                         mutex_unlock(&dcc->cmd_lock);
3059                         trimmed += __wait_all_discard_cmd(sbi, NULL);
3060                         congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
3061                         goto next;
3062                 }
3063 skip:
3064                 node = rb_next(&dc->rb_node);
3065                 if (err)
3066                         __remove_discard_cmd(sbi, dc);
3067                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3068
3069                 if (fatal_signal_pending(current))
3070                         break;
3071         }
3072
3073         blk_finish_plug(&plug);
3074         mutex_unlock(&dcc->cmd_lock);
3075
3076         return trimmed;
3077 }
3078
3079 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3080 {
3081         __u64 start = F2FS_BYTES_TO_BLK(range->start);
3082         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
3083         unsigned int start_segno, end_segno;
3084         block_t start_block, end_block;
3085         struct cp_control cpc;
3086         struct discard_policy dpolicy;
3087         unsigned long long trimmed = 0;
3088         int err = 0;
3089         bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
3090
3091         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
3092                 return -EINVAL;
3093
3094         if (end < MAIN_BLKADDR(sbi))
3095                 goto out;
3096
3097         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
3098                 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
3099                 return -EFSCORRUPTED;
3100         }
3101
3102         /* start/end segment number in main_area */
3103         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3104         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3105                                                 GET_SEGNO(sbi, end);
3106         if (need_align) {
3107                 start_segno = rounddown(start_segno, sbi->segs_per_sec);
3108                 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
3109         }
3110
3111         cpc.reason = CP_DISCARD;
3112         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
3113         cpc.trim_start = start_segno;
3114         cpc.trim_end = end_segno;
3115
3116         if (sbi->discard_blks == 0)
3117                 goto out;
3118
3119         down_write(&sbi->gc_lock);
3120         err = f2fs_write_checkpoint(sbi, &cpc);
3121         up_write(&sbi->gc_lock);
3122         if (err)
3123                 goto out;
3124
3125         /*
3126          * We filed discard candidates, but actually we don't need to wait for
3127          * all of them, since they'll be issued in idle time along with runtime
3128          * discard option. User configuration looks like using runtime discard
3129          * or periodic fstrim instead of it.
3130          */
3131         if (f2fs_realtime_discard_enable(sbi))
3132                 goto out;
3133
3134         start_block = START_BLOCK(sbi, start_segno);
3135         end_block = START_BLOCK(sbi, end_segno + 1);
3136
3137         __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3138         trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3139                                         start_block, end_block);
3140
3141         trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3142                                         start_block, end_block);
3143 out:
3144         if (!err)
3145                 range->len = F2FS_BLK_TO_BYTES(trimmed);
3146         return err;
3147 }
3148
3149 static bool __has_curseg_space(struct f2fs_sb_info *sbi,
3150                                         struct curseg_info *curseg)
3151 {
3152         return curseg->next_blkoff < f2fs_usable_blks_in_seg(sbi,
3153                                                         curseg->segno);
3154 }
3155
3156 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
3157 {
3158         switch (hint) {
3159         case WRITE_LIFE_SHORT:
3160                 return CURSEG_HOT_DATA;
3161         case WRITE_LIFE_EXTREME:
3162                 return CURSEG_COLD_DATA;
3163         default:
3164                 return CURSEG_WARM_DATA;
3165         }
3166 }
3167
3168 /* This returns write hints for each segment type. This hints will be
3169  * passed down to block layer. There are mapping tables which depend on
3170  * the mount option 'whint_mode'.
3171  *
3172  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
3173  *
3174  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
3175  *
3176  * User                  F2FS                     Block
3177  * ----                  ----                     -----
3178  *                       META                     WRITE_LIFE_NOT_SET
3179  *                       HOT_NODE                 "
3180  *                       WARM_NODE                "
3181  *                       COLD_NODE                "
3182  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
3183  * extension list        "                        "
3184  *
3185  * -- buffered io
3186  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3187  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3188  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
3189  * WRITE_LIFE_NONE       "                        "
3190  * WRITE_LIFE_MEDIUM     "                        "
3191  * WRITE_LIFE_LONG       "                        "
3192  *
3193  * -- direct io
3194  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3195  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3196  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
3197  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
3198  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
3199  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
3200  *
3201  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
3202  *
3203  * User                  F2FS                     Block
3204  * ----                  ----                     -----
3205  *                       META                     WRITE_LIFE_MEDIUM;
3206  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
3207  *                       WARM_NODE                "
3208  *                       COLD_NODE                WRITE_LIFE_NONE
3209  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
3210  * extension list        "                        "
3211  *
3212  * -- buffered io
3213  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3214  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3215  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
3216  * WRITE_LIFE_NONE       "                        "
3217  * WRITE_LIFE_MEDIUM     "                        "
3218  * WRITE_LIFE_LONG       "                        "
3219  *
3220  * -- direct io
3221  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3222  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3223  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
3224  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
3225  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
3226  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
3227  */
3228
3229 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
3230                                 enum page_type type, enum temp_type temp)
3231 {
3232         if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
3233                 if (type == DATA) {
3234                         if (temp == WARM)
3235                                 return WRITE_LIFE_NOT_SET;
3236                         else if (temp == HOT)
3237                                 return WRITE_LIFE_SHORT;
3238                         else if (temp == COLD)
3239                                 return WRITE_LIFE_EXTREME;
3240                 } else {
3241                         return WRITE_LIFE_NOT_SET;
3242                 }
3243         } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
3244                 if (type == DATA) {
3245                         if (temp == WARM)
3246                                 return WRITE_LIFE_LONG;
3247                         else if (temp == HOT)
3248                                 return WRITE_LIFE_SHORT;
3249                         else if (temp == COLD)
3250                                 return WRITE_LIFE_EXTREME;
3251                 } else if (type == NODE) {
3252                         if (temp == WARM || temp == HOT)
3253                                 return WRITE_LIFE_NOT_SET;
3254                         else if (temp == COLD)
3255                                 return WRITE_LIFE_NONE;
3256                 } else if (type == META) {
3257                         return WRITE_LIFE_MEDIUM;
3258                 }
3259         }
3260         return WRITE_LIFE_NOT_SET;
3261 }
3262
3263 static int __get_segment_type_2(struct f2fs_io_info *fio)
3264 {
3265         if (fio->type == DATA)
3266                 return CURSEG_HOT_DATA;
3267         else
3268                 return CURSEG_HOT_NODE;
3269 }
3270
3271 static int __get_segment_type_4(struct f2fs_io_info *fio)
3272 {
3273         if (fio->type == DATA) {
3274                 struct inode *inode = fio->page->mapping->host;
3275
3276                 if (S_ISDIR(inode->i_mode))
3277                         return CURSEG_HOT_DATA;
3278                 else
3279                         return CURSEG_COLD_DATA;
3280         } else {
3281                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3282                         return CURSEG_WARM_NODE;
3283                 else
3284                         return CURSEG_COLD_NODE;
3285         }
3286 }
3287
3288 static int __get_segment_type_6(struct f2fs_io_info *fio)
3289 {
3290         if (fio->type == DATA) {
3291                 struct inode *inode = fio->page->mapping->host;
3292
3293                 if (page_private_gcing(fio->page)) {
3294                         if (fio->sbi->am.atgc_enabled &&
3295                                 (fio->io_type == FS_DATA_IO) &&
3296                                 (fio->sbi->gc_mode != GC_URGENT_HIGH))
3297                                 return CURSEG_ALL_DATA_ATGC;
3298                         else
3299                                 return CURSEG_COLD_DATA;
3300                 }
3301                 if (file_is_cold(inode) || f2fs_need_compress_data(inode))
3302                         return CURSEG_COLD_DATA;
3303                 if (file_is_hot(inode) ||
3304                                 is_inode_flag_set(inode, FI_HOT_DATA) ||
3305                                 f2fs_is_atomic_file(inode) ||
3306                                 f2fs_is_volatile_file(inode))
3307                         return CURSEG_HOT_DATA;
3308                 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3309         } else {
3310                 if (IS_DNODE(fio->page))
3311                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3312                                                 CURSEG_HOT_NODE;
3313                 return CURSEG_COLD_NODE;
3314         }
3315 }
3316
3317 static int __get_segment_type(struct f2fs_io_info *fio)
3318 {
3319         int type = 0;
3320
3321         switch (F2FS_OPTION(fio->sbi).active_logs) {
3322         case 2:
3323                 type = __get_segment_type_2(fio);
3324                 break;
3325         case 4:
3326                 type = __get_segment_type_4(fio);
3327                 break;
3328         case 6:
3329                 type = __get_segment_type_6(fio);
3330                 break;
3331         default:
3332                 f2fs_bug_on(fio->sbi, true);
3333         }
3334
3335         if (IS_HOT(type))
3336                 fio->temp = HOT;
3337         else if (IS_WARM(type))
3338                 fio->temp = WARM;
3339         else
3340                 fio->temp = COLD;
3341         return type;
3342 }
3343
3344 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3345                 block_t old_blkaddr, block_t *new_blkaddr,
3346                 struct f2fs_summary *sum, int type,
3347                 struct f2fs_io_info *fio)
3348 {
3349         struct sit_info *sit_i = SIT_I(sbi);
3350         struct curseg_info *curseg = CURSEG_I(sbi, type);
3351         unsigned long long old_mtime;
3352         bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3353         struct seg_entry *se = NULL;
3354
3355         down_read(&SM_I(sbi)->curseg_lock);
3356
3357         mutex_lock(&curseg->curseg_mutex);
3358         down_write(&sit_i->sentry_lock);
3359
3360         if (from_gc) {
3361                 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3362                 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3363                 sanity_check_seg_type(sbi, se->type);
3364                 f2fs_bug_on(sbi, IS_NODESEG(se->type));
3365         }
3366         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3367
3368         f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg);
3369
3370         f2fs_wait_discard_bio(sbi, *new_blkaddr);
3371
3372         /*
3373          * __add_sum_entry should be resided under the curseg_mutex
3374          * because, this function updates a summary entry in the
3375          * current summary block.
3376          */
3377         __add_sum_entry(sbi, type, sum);
3378
3379         __refresh_next_blkoff(sbi, curseg);
3380
3381         stat_inc_block_count(sbi, curseg);
3382
3383         if (from_gc) {
3384                 old_mtime = get_segment_mtime(sbi, old_blkaddr);
3385         } else {
3386                 update_segment_mtime(sbi, old_blkaddr, 0);
3387                 old_mtime = 0;
3388         }
3389         update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3390
3391         /*
3392          * SIT information should be updated before segment allocation,
3393          * since SSR needs latest valid block information.
3394          */
3395         update_sit_entry(sbi, *new_blkaddr, 1);
3396         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3397                 update_sit_entry(sbi, old_blkaddr, -1);
3398
3399         if (!__has_curseg_space(sbi, curseg)) {
3400                 if (from_gc)
3401                         get_atssr_segment(sbi, type, se->type,
3402                                                 AT_SSR, se->mtime);
3403                 else
3404                         sit_i->s_ops->allocate_segment(sbi, type, false);
3405         }
3406         /*
3407          * segment dirty status should be updated after segment allocation,
3408          * so we just need to update status only one time after previous
3409          * segment being closed.
3410          */
3411         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3412         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3413
3414         up_write(&sit_i->sentry_lock);
3415
3416         if (page && IS_NODESEG(type)) {
3417                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3418
3419                 f2fs_inode_chksum_set(sbi, page);
3420         }
3421
3422         if (fio) {
3423                 struct f2fs_bio_info *io;
3424
3425                 if (F2FS_IO_ALIGNED(sbi))
3426                         fio->retry = false;
3427
3428                 INIT_LIST_HEAD(&fio->list);
3429                 fio->in_list = true;
3430                 io = sbi->write_io[fio->type] + fio->temp;
3431                 spin_lock(&io->io_lock);
3432                 list_add_tail(&fio->list, &io->io_list);
3433                 spin_unlock(&io->io_lock);
3434         }
3435
3436         mutex_unlock(&curseg->curseg_mutex);
3437
3438         up_read(&SM_I(sbi)->curseg_lock);
3439 }
3440
3441 static void update_device_state(struct f2fs_io_info *fio)
3442 {
3443         struct f2fs_sb_info *sbi = fio->sbi;
3444         unsigned int devidx;
3445
3446         if (!f2fs_is_multi_device(sbi))
3447                 return;
3448
3449         devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3450
3451         /* update device state for fsync */
3452         f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
3453
3454         /* update device state for checkpoint */
3455         if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3456                 spin_lock(&sbi->dev_lock);
3457                 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3458                 spin_unlock(&sbi->dev_lock);
3459         }
3460 }
3461
3462 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3463 {
3464         int type = __get_segment_type(fio);
3465         bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3466
3467         if (keep_order)
3468                 down_read(&fio->sbi->io_order_lock);
3469 reallocate:
3470         f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3471                         &fio->new_blkaddr, sum, type, fio);
3472         if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3473                 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3474                                         fio->old_blkaddr, fio->old_blkaddr);
3475
3476         /* writeout dirty page into bdev */
3477         f2fs_submit_page_write(fio);
3478         if (fio->retry) {
3479                 fio->old_blkaddr = fio->new_blkaddr;
3480                 goto reallocate;
3481         }
3482
3483         update_device_state(fio);
3484
3485         if (keep_order)
3486                 up_read(&fio->sbi->io_order_lock);
3487 }
3488
3489 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3490                                         enum iostat_type io_type)
3491 {
3492         struct f2fs_io_info fio = {
3493                 .sbi = sbi,
3494                 .type = META,
3495                 .temp = HOT,
3496                 .op = REQ_OP_WRITE,
3497                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3498                 .old_blkaddr = page->index,
3499                 .new_blkaddr = page->index,
3500                 .page = page,
3501                 .encrypted_page = NULL,
3502                 .in_list = false,
3503         };
3504
3505         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3506                 fio.op_flags &= ~REQ_META;
3507
3508         set_page_writeback(page);
3509         ClearPageError(page);
3510         f2fs_submit_page_write(&fio);
3511
3512         stat_inc_meta_count(sbi, page->index);
3513         f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3514 }
3515
3516 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3517 {
3518         struct f2fs_summary sum;
3519
3520         set_summary(&sum, nid, 0, 0);
3521         do_write_page(&sum, fio);
3522
3523         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3524 }
3525
3526 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3527                                         struct f2fs_io_info *fio)
3528 {
3529         struct f2fs_sb_info *sbi = fio->sbi;
3530         struct f2fs_summary sum;
3531
3532         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3533         set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3534         do_write_page(&sum, fio);
3535         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3536
3537         f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3538 }
3539
3540 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3541 {
3542         int err;
3543         struct f2fs_sb_info *sbi = fio->sbi;
3544         unsigned int segno;
3545
3546         fio->new_blkaddr = fio->old_blkaddr;
3547         /* i/o temperature is needed for passing down write hints */
3548         __get_segment_type(fio);
3549
3550         segno = GET_SEGNO(sbi, fio->new_blkaddr);
3551
3552         if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3553                 set_sbi_flag(sbi, SBI_NEED_FSCK);
3554                 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3555                           __func__, segno);
3556                 err = -EFSCORRUPTED;
3557                 goto drop_bio;
3558         }
3559
3560         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) || f2fs_cp_error(sbi)) {
3561                 err = -EIO;
3562                 goto drop_bio;
3563         }
3564
3565         stat_inc_inplace_blocks(fio->sbi);
3566
3567         if (fio->bio && !(SM_I(sbi)->ipu_policy & (1 << F2FS_IPU_NOCACHE)))
3568                 err = f2fs_merge_page_bio(fio);
3569         else
3570                 err = f2fs_submit_page_bio(fio);
3571         if (!err) {
3572                 update_device_state(fio);
3573                 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3574         }
3575
3576         return err;
3577 drop_bio:
3578         if (fio->bio && *(fio->bio)) {
3579                 struct bio *bio = *(fio->bio);
3580
3581                 bio->bi_status = BLK_STS_IOERR;
3582                 bio_endio(bio);
3583                 *(fio->bio) = NULL;
3584         }
3585         return err;
3586 }
3587
3588 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3589                                                 unsigned int segno)
3590 {
3591         int i;
3592
3593         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3594                 if (CURSEG_I(sbi, i)->segno == segno)
3595                         break;
3596         }
3597         return i;
3598 }
3599
3600 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3601                                 block_t old_blkaddr, block_t new_blkaddr,
3602                                 bool recover_curseg, bool recover_newaddr,
3603                                 bool from_gc)
3604 {
3605         struct sit_info *sit_i = SIT_I(sbi);
3606         struct curseg_info *curseg;
3607         unsigned int segno, old_cursegno;
3608         struct seg_entry *se;
3609         int type;
3610         unsigned short old_blkoff;
3611         unsigned char old_alloc_type;
3612
3613         segno = GET_SEGNO(sbi, new_blkaddr);
3614         se = get_seg_entry(sbi, segno);
3615         type = se->type;
3616
3617         down_write(&SM_I(sbi)->curseg_lock);
3618
3619         if (!recover_curseg) {
3620                 /* for recovery flow */
3621                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3622                         if (old_blkaddr == NULL_ADDR)
3623                                 type = CURSEG_COLD_DATA;
3624                         else
3625                                 type = CURSEG_WARM_DATA;
3626                 }
3627         } else {
3628                 if (IS_CURSEG(sbi, segno)) {
3629                         /* se->type is volatile as SSR allocation */
3630                         type = __f2fs_get_curseg(sbi, segno);
3631                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3632                 } else {
3633                         type = CURSEG_WARM_DATA;
3634                 }
3635         }
3636
3637         f2fs_bug_on(sbi, !IS_DATASEG(type));
3638         curseg = CURSEG_I(sbi, type);
3639
3640         mutex_lock(&curseg->curseg_mutex);
3641         down_write(&sit_i->sentry_lock);
3642
3643         old_cursegno = curseg->segno;
3644         old_blkoff = curseg->next_blkoff;
3645         old_alloc_type = curseg->alloc_type;
3646
3647         /* change the current segment */
3648         if (segno != curseg->segno) {
3649                 curseg->next_segno = segno;
3650                 change_curseg(sbi, type, true);
3651         }
3652
3653         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3654         __add_sum_entry(sbi, type, sum);
3655
3656         if (!recover_curseg || recover_newaddr) {
3657                 if (!from_gc)
3658                         update_segment_mtime(sbi, new_blkaddr, 0);
3659                 update_sit_entry(sbi, new_blkaddr, 1);
3660         }
3661         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3662                 invalidate_mapping_pages(META_MAPPING(sbi),
3663                                         old_blkaddr, old_blkaddr);
3664                 if (!from_gc)
3665                         update_segment_mtime(sbi, old_blkaddr, 0);
3666                 update_sit_entry(sbi, old_blkaddr, -1);
3667         }
3668
3669         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3670         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3671
3672         locate_dirty_segment(sbi, old_cursegno);
3673
3674         if (recover_curseg) {
3675                 if (old_cursegno != curseg->segno) {
3676                         curseg->next_segno = old_cursegno;
3677                         change_curseg(sbi, type, true);
3678                 }
3679                 curseg->next_blkoff = old_blkoff;
3680                 curseg->alloc_type = old_alloc_type;
3681         }
3682
3683         up_write(&sit_i->sentry_lock);
3684         mutex_unlock(&curseg->curseg_mutex);
3685         up_write(&SM_I(sbi)->curseg_lock);
3686 }
3687
3688 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3689                                 block_t old_addr, block_t new_addr,
3690                                 unsigned char version, bool recover_curseg,
3691                                 bool recover_newaddr)
3692 {
3693         struct f2fs_summary sum;
3694
3695         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3696
3697         f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3698                                         recover_curseg, recover_newaddr, false);
3699
3700         f2fs_update_data_blkaddr(dn, new_addr);
3701 }
3702
3703 void f2fs_wait_on_page_writeback(struct page *page,
3704                                 enum page_type type, bool ordered, bool locked)
3705 {
3706         if (PageWriteback(page)) {
3707                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3708
3709                 /* submit cached LFS IO */
3710                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3711                 /* sbumit cached IPU IO */
3712                 f2fs_submit_merged_ipu_write(sbi, NULL, page);
3713                 if (ordered) {
3714                         wait_on_page_writeback(page);
3715                         f2fs_bug_on(sbi, locked && PageWriteback(page));
3716                 } else {
3717                         wait_for_stable_page(page);
3718                 }
3719         }
3720 }
3721
3722 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3723 {
3724         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3725         struct page *cpage;
3726
3727         if (!f2fs_post_read_required(inode))
3728                 return;
3729
3730         if (!__is_valid_data_blkaddr(blkaddr))
3731                 return;
3732
3733         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3734         if (cpage) {
3735                 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3736                 f2fs_put_page(cpage, 1);
3737         }
3738 }
3739
3740 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3741                                                                 block_t len)
3742 {
3743         block_t i;
3744
3745         for (i = 0; i < len; i++)
3746                 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3747 }
3748
3749 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3750 {
3751         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3752         struct curseg_info *seg_i;
3753         unsigned char *kaddr;
3754         struct page *page;
3755         block_t start;
3756         int i, j, offset;
3757
3758         start = start_sum_block(sbi);
3759
3760         page = f2fs_get_meta_page(sbi, start++);
3761         if (IS_ERR(page))
3762                 return PTR_ERR(page);
3763         kaddr = (unsigned char *)page_address(page);
3764
3765         /* Step 1: restore nat cache */
3766         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3767         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3768
3769         /* Step 2: restore sit cache */
3770         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3771         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3772         offset = 2 * SUM_JOURNAL_SIZE;
3773
3774         /* Step 3: restore summary entries */
3775         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3776                 unsigned short blk_off;
3777                 unsigned int segno;
3778
3779                 seg_i = CURSEG_I(sbi, i);
3780                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3781                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3782                 seg_i->next_segno = segno;
3783                 reset_curseg(sbi, i, 0);
3784                 seg_i->alloc_type = ckpt->alloc_type[i];
3785                 seg_i->next_blkoff = blk_off;
3786
3787                 if (seg_i->alloc_type == SSR)
3788                         blk_off = sbi->blocks_per_seg;
3789
3790                 for (j = 0; j < blk_off; j++) {
3791                         struct f2fs_summary *s;
3792
3793                         s = (struct f2fs_summary *)(kaddr + offset);
3794                         seg_i->sum_blk->entries[j] = *s;
3795                         offset += SUMMARY_SIZE;
3796                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3797                                                 SUM_FOOTER_SIZE)
3798                                 continue;
3799
3800                         f2fs_put_page(page, 1);
3801                         page = NULL;
3802
3803                         page = f2fs_get_meta_page(sbi, start++);
3804                         if (IS_ERR(page))
3805                                 return PTR_ERR(page);
3806                         kaddr = (unsigned char *)page_address(page);
3807                         offset = 0;
3808                 }
3809         }
3810         f2fs_put_page(page, 1);
3811         return 0;
3812 }
3813
3814 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3815 {
3816         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3817         struct f2fs_summary_block *sum;
3818         struct curseg_info *curseg;
3819         struct page *new;
3820         unsigned short blk_off;
3821         unsigned int segno = 0;
3822         block_t blk_addr = 0;
3823         int err = 0;
3824
3825         /* get segment number and block addr */
3826         if (IS_DATASEG(type)) {
3827                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3828                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3829                                                         CURSEG_HOT_DATA]);
3830                 if (__exist_node_summaries(sbi))
3831                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
3832                 else
3833                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3834         } else {
3835                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3836                                                         CURSEG_HOT_NODE]);
3837                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3838                                                         CURSEG_HOT_NODE]);
3839                 if (__exist_node_summaries(sbi))
3840                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3841                                                         type - CURSEG_HOT_NODE);
3842                 else
3843                         blk_addr = GET_SUM_BLOCK(sbi, segno);
3844         }
3845
3846         new = f2fs_get_meta_page(sbi, blk_addr);
3847         if (IS_ERR(new))
3848                 return PTR_ERR(new);
3849         sum = (struct f2fs_summary_block *)page_address(new);
3850
3851         if (IS_NODESEG(type)) {
3852                 if (__exist_node_summaries(sbi)) {
3853                         struct f2fs_summary *ns = &sum->entries[0];
3854                         int i;
3855
3856                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3857                                 ns->version = 0;
3858                                 ns->ofs_in_node = 0;
3859                         }
3860                 } else {
3861                         err = f2fs_restore_node_summary(sbi, segno, sum);
3862                         if (err)
3863                                 goto out;
3864                 }
3865         }
3866
3867         /* set uncompleted segment to curseg */
3868         curseg = CURSEG_I(sbi, type);
3869         mutex_lock(&curseg->curseg_mutex);
3870
3871         /* update journal info */
3872         down_write(&curseg->journal_rwsem);
3873         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3874         up_write(&curseg->journal_rwsem);
3875
3876         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3877         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3878         curseg->next_segno = segno;
3879         reset_curseg(sbi, type, 0);
3880         curseg->alloc_type = ckpt->alloc_type[type];
3881         curseg->next_blkoff = blk_off;
3882         mutex_unlock(&curseg->curseg_mutex);
3883 out:
3884         f2fs_put_page(new, 1);
3885         return err;
3886 }
3887
3888 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3889 {
3890         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3891         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3892         int type = CURSEG_HOT_DATA;
3893         int err;
3894
3895         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3896                 int npages = f2fs_npages_for_summary_flush(sbi, true);
3897
3898                 if (npages >= 2)
3899                         f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3900                                                         META_CP, true);
3901
3902                 /* restore for compacted data summary */
3903                 err = read_compacted_summaries(sbi);
3904                 if (err)
3905                         return err;
3906                 type = CURSEG_HOT_NODE;
3907         }
3908
3909         if (__exist_node_summaries(sbi))
3910                 f2fs_ra_meta_pages(sbi,
3911                                 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
3912                                 NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
3913
3914         for (; type <= CURSEG_COLD_NODE; type++) {
3915                 err = read_normal_summaries(sbi, type);
3916                 if (err)
3917                         return err;
3918         }
3919
3920         /* sanity check for summary blocks */
3921         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3922                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3923                 f2fs_err(sbi, "invalid journal entries nats %u sits %u\n",
3924                          nats_in_cursum(nat_j), sits_in_cursum(sit_j));
3925                 return -EINVAL;
3926         }
3927
3928         return 0;
3929 }
3930
3931 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3932 {
3933         struct page *page;
3934         unsigned char *kaddr;
3935         struct f2fs_summary *summary;
3936         struct curseg_info *seg_i;
3937         int written_size = 0;
3938         int i, j;
3939
3940         page = f2fs_grab_meta_page(sbi, blkaddr++);
3941         kaddr = (unsigned char *)page_address(page);
3942         memset(kaddr, 0, PAGE_SIZE);
3943
3944         /* Step 1: write nat cache */
3945         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3946         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3947         written_size += SUM_JOURNAL_SIZE;
3948
3949         /* Step 2: write sit cache */
3950         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3951         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3952         written_size += SUM_JOURNAL_SIZE;
3953
3954         /* Step 3: write summary entries */
3955         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3956                 unsigned short blkoff;
3957
3958                 seg_i = CURSEG_I(sbi, i);
3959                 if (sbi->ckpt->alloc_type[i] == SSR)
3960                         blkoff = sbi->blocks_per_seg;
3961                 else
3962                         blkoff = curseg_blkoff(sbi, i);
3963
3964                 for (j = 0; j < blkoff; j++) {
3965                         if (!page) {
3966                                 page = f2fs_grab_meta_page(sbi, blkaddr++);
3967                                 kaddr = (unsigned char *)page_address(page);
3968                                 memset(kaddr, 0, PAGE_SIZE);
3969                                 written_size = 0;
3970                         }
3971                         summary = (struct f2fs_summary *)(kaddr + written_size);
3972                         *summary = seg_i->sum_blk->entries[j];
3973                         written_size += SUMMARY_SIZE;
3974
3975                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3976                                                         SUM_FOOTER_SIZE)
3977                                 continue;
3978
3979                         set_page_dirty(page);
3980                         f2fs_put_page(page, 1);
3981                         page = NULL;
3982                 }
3983         }
3984         if (page) {
3985                 set_page_dirty(page);
3986                 f2fs_put_page(page, 1);
3987         }
3988 }
3989
3990 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3991                                         block_t blkaddr, int type)
3992 {
3993         int i, end;
3994
3995         if (IS_DATASEG(type))
3996                 end = type + NR_CURSEG_DATA_TYPE;
3997         else
3998                 end = type + NR_CURSEG_NODE_TYPE;
3999
4000         for (i = type; i < end; i++)
4001                 write_current_sum_page(sbi, i, blkaddr + (i - type));
4002 }
4003
4004 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4005 {
4006         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
4007                 write_compacted_summaries(sbi, start_blk);
4008         else
4009                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
4010 }
4011
4012 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4013 {
4014         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
4015 }
4016
4017 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
4018                                         unsigned int val, int alloc)
4019 {
4020         int i;
4021
4022         if (type == NAT_JOURNAL) {
4023                 for (i = 0; i < nats_in_cursum(journal); i++) {
4024                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
4025                                 return i;
4026                 }
4027                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
4028                         return update_nats_in_cursum(journal, 1);
4029         } else if (type == SIT_JOURNAL) {
4030                 for (i = 0; i < sits_in_cursum(journal); i++)
4031                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
4032                                 return i;
4033                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
4034                         return update_sits_in_cursum(journal, 1);
4035         }
4036         return -1;
4037 }
4038
4039 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
4040                                         unsigned int segno)
4041 {
4042         return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
4043 }
4044
4045 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
4046                                         unsigned int start)
4047 {
4048         struct sit_info *sit_i = SIT_I(sbi);
4049         struct page *page;
4050         pgoff_t src_off, dst_off;
4051
4052         src_off = current_sit_addr(sbi, start);
4053         dst_off = next_sit_addr(sbi, src_off);
4054
4055         page = f2fs_grab_meta_page(sbi, dst_off);
4056         seg_info_to_sit_page(sbi, page, start);
4057
4058         set_page_dirty(page);
4059         set_to_next_sit(sit_i, start);
4060
4061         return page;
4062 }
4063
4064 static struct sit_entry_set *grab_sit_entry_set(void)
4065 {
4066         struct sit_entry_set *ses =
4067                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
4068
4069         ses->entry_cnt = 0;
4070         INIT_LIST_HEAD(&ses->set_list);
4071         return ses;
4072 }
4073
4074 static void release_sit_entry_set(struct sit_entry_set *ses)
4075 {
4076         list_del(&ses->set_list);
4077         kmem_cache_free(sit_entry_set_slab, ses);
4078 }
4079
4080 static void adjust_sit_entry_set(struct sit_entry_set *ses,
4081                                                 struct list_head *head)
4082 {
4083         struct sit_entry_set *next = ses;
4084
4085         if (list_is_last(&ses->set_list, head))
4086                 return;
4087
4088         list_for_each_entry_continue(next, head, set_list)
4089                 if (ses->entry_cnt <= next->entry_cnt)
4090                         break;
4091
4092         list_move_tail(&ses->set_list, &next->set_list);
4093 }
4094
4095 static void add_sit_entry(unsigned int segno, struct list_head *head)
4096 {
4097         struct sit_entry_set *ses;
4098         unsigned int start_segno = START_SEGNO(segno);
4099
4100         list_for_each_entry(ses, head, set_list) {
4101                 if (ses->start_segno == start_segno) {
4102                         ses->entry_cnt++;
4103                         adjust_sit_entry_set(ses, head);
4104                         return;
4105                 }
4106         }
4107
4108         ses = grab_sit_entry_set();
4109
4110         ses->start_segno = start_segno;
4111         ses->entry_cnt++;
4112         list_add(&ses->set_list, head);
4113 }
4114
4115 static void add_sits_in_set(struct f2fs_sb_info *sbi)
4116 {
4117         struct f2fs_sm_info *sm_info = SM_I(sbi);
4118         struct list_head *set_list = &sm_info->sit_entry_set;
4119         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
4120         unsigned int segno;
4121
4122         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
4123                 add_sit_entry(segno, set_list);
4124 }
4125
4126 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
4127 {
4128         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4129         struct f2fs_journal *journal = curseg->journal;
4130         int i;
4131
4132         down_write(&curseg->journal_rwsem);
4133         for (i = 0; i < sits_in_cursum(journal); i++) {
4134                 unsigned int segno;
4135                 bool dirtied;
4136
4137                 segno = le32_to_cpu(segno_in_journal(journal, i));
4138                 dirtied = __mark_sit_entry_dirty(sbi, segno);
4139
4140                 if (!dirtied)
4141                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
4142         }
4143         update_sits_in_cursum(journal, -i);
4144         up_write(&curseg->journal_rwsem);
4145 }
4146
4147 /*
4148  * CP calls this function, which flushes SIT entries including sit_journal,
4149  * and moves prefree segs to free segs.
4150  */
4151 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
4152 {
4153         struct sit_info *sit_i = SIT_I(sbi);
4154         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4155         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4156         struct f2fs_journal *journal = curseg->journal;
4157         struct sit_entry_set *ses, *tmp;
4158         struct list_head *head = &SM_I(sbi)->sit_entry_set;
4159         bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4160         struct seg_entry *se;
4161
4162         down_write(&sit_i->sentry_lock);
4163
4164         if (!sit_i->dirty_sentries)
4165                 goto out;
4166
4167         /*
4168          * add and account sit entries of dirty bitmap in sit entry
4169          * set temporarily
4170          */
4171         add_sits_in_set(sbi);
4172
4173         /*
4174          * if there are no enough space in journal to store dirty sit
4175          * entries, remove all entries from journal and add and account
4176          * them in sit entry set.
4177          */
4178         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4179                                                                 !to_journal)
4180                 remove_sits_in_journal(sbi);
4181
4182         /*
4183          * there are two steps to flush sit entries:
4184          * #1, flush sit entries to journal in current cold data summary block.
4185          * #2, flush sit entries to sit page.
4186          */
4187         list_for_each_entry_safe(ses, tmp, head, set_list) {
4188                 struct page *page = NULL;
4189                 struct f2fs_sit_block *raw_sit = NULL;
4190                 unsigned int start_segno = ses->start_segno;
4191                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4192                                                 (unsigned long)MAIN_SEGS(sbi));
4193                 unsigned int segno = start_segno;
4194
4195                 if (to_journal &&
4196                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4197                         to_journal = false;
4198
4199                 if (to_journal) {
4200                         down_write(&curseg->journal_rwsem);
4201                 } else {
4202                         page = get_next_sit_page(sbi, start_segno);
4203                         raw_sit = page_address(page);
4204                 }
4205
4206                 /* flush dirty sit entries in region of current sit set */
4207                 for_each_set_bit_from(segno, bitmap, end) {
4208                         int offset, sit_offset;
4209
4210                         se = get_seg_entry(sbi, segno);
4211 #ifdef CONFIG_F2FS_CHECK_FS
4212                         if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4213                                                 SIT_VBLOCK_MAP_SIZE))
4214                                 f2fs_bug_on(sbi, 1);
4215 #endif
4216
4217                         /* add discard candidates */
4218                         if (!(cpc->reason & CP_DISCARD)) {
4219                                 cpc->trim_start = segno;
4220                                 add_discard_addrs(sbi, cpc, false);
4221                         }
4222
4223                         if (to_journal) {
4224                                 offset = f2fs_lookup_journal_in_cursum(journal,
4225                                                         SIT_JOURNAL, segno, 1);
4226                                 f2fs_bug_on(sbi, offset < 0);
4227                                 segno_in_journal(journal, offset) =
4228                                                         cpu_to_le32(segno);
4229                                 seg_info_to_raw_sit(se,
4230                                         &sit_in_journal(journal, offset));
4231                                 check_block_count(sbi, segno,
4232                                         &sit_in_journal(journal, offset));
4233                         } else {
4234                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4235                                 seg_info_to_raw_sit(se,
4236                                                 &raw_sit->entries[sit_offset]);
4237                                 check_block_count(sbi, segno,
4238                                                 &raw_sit->entries[sit_offset]);
4239                         }
4240
4241                         __clear_bit(segno, bitmap);
4242                         sit_i->dirty_sentries--;
4243                         ses->entry_cnt--;
4244                 }
4245
4246                 if (to_journal)
4247                         up_write(&curseg->journal_rwsem);
4248                 else
4249                         f2fs_put_page(page, 1);
4250
4251                 f2fs_bug_on(sbi, ses->entry_cnt);
4252                 release_sit_entry_set(ses);
4253         }
4254
4255         f2fs_bug_on(sbi, !list_empty(head));
4256         f2fs_bug_on(sbi, sit_i->dirty_sentries);
4257 out:
4258         if (cpc->reason & CP_DISCARD) {
4259                 __u64 trim_start = cpc->trim_start;
4260
4261                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4262                         add_discard_addrs(sbi, cpc, false);
4263
4264                 cpc->trim_start = trim_start;
4265         }
4266         up_write(&sit_i->sentry_lock);
4267
4268         set_prefree_as_free_segments(sbi);
4269 }
4270
4271 static int build_sit_info(struct f2fs_sb_info *sbi)
4272 {
4273         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4274         struct sit_info *sit_i;
4275         unsigned int sit_segs, start;
4276         char *src_bitmap, *bitmap;
4277         unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4278
4279         /* allocate memory for SIT information */
4280         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4281         if (!sit_i)
4282                 return -ENOMEM;
4283
4284         SM_I(sbi)->sit_info = sit_i;
4285
4286         sit_i->sentries =
4287                 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4288                                               MAIN_SEGS(sbi)),
4289                               GFP_KERNEL);
4290         if (!sit_i->sentries)
4291                 return -ENOMEM;
4292
4293         main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4294         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4295                                                                 GFP_KERNEL);
4296         if (!sit_i->dirty_sentries_bitmap)
4297                 return -ENOMEM;
4298
4299 #ifdef CONFIG_F2FS_CHECK_FS
4300         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 4;
4301 #else
4302         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 3;
4303 #endif
4304         sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4305         if (!sit_i->bitmap)
4306                 return -ENOMEM;
4307
4308         bitmap = sit_i->bitmap;
4309
4310         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4311                 sit_i->sentries[start].cur_valid_map = bitmap;
4312                 bitmap += SIT_VBLOCK_MAP_SIZE;
4313
4314                 sit_i->sentries[start].ckpt_valid_map = bitmap;
4315                 bitmap += SIT_VBLOCK_MAP_SIZE;
4316
4317 #ifdef CONFIG_F2FS_CHECK_FS
4318                 sit_i->sentries[start].cur_valid_map_mir = bitmap;
4319                 bitmap += SIT_VBLOCK_MAP_SIZE;
4320 #endif
4321
4322                 sit_i->sentries[start].discard_map = bitmap;
4323                 bitmap += SIT_VBLOCK_MAP_SIZE;
4324         }
4325
4326         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4327         if (!sit_i->tmp_map)
4328                 return -ENOMEM;
4329
4330         if (__is_large_section(sbi)) {
4331                 sit_i->sec_entries =
4332                         f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4333                                                       MAIN_SECS(sbi)),
4334                                       GFP_KERNEL);
4335                 if (!sit_i->sec_entries)
4336                         return -ENOMEM;
4337         }
4338
4339         /* get information related with SIT */
4340         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4341
4342         /* setup SIT bitmap from ckeckpoint pack */
4343         sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4344         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4345
4346         sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4347         if (!sit_i->sit_bitmap)
4348                 return -ENOMEM;
4349
4350 #ifdef CONFIG_F2FS_CHECK_FS
4351         sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4352                                         sit_bitmap_size, GFP_KERNEL);
4353         if (!sit_i->sit_bitmap_mir)
4354                 return -ENOMEM;
4355
4356         sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4357                                         main_bitmap_size, GFP_KERNEL);
4358         if (!sit_i->invalid_segmap)
4359                 return -ENOMEM;
4360 #endif
4361
4362         /* init SIT information */
4363         sit_i->s_ops = &default_salloc_ops;
4364
4365         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4366         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4367         sit_i->written_valid_blocks = 0;
4368         sit_i->bitmap_size = sit_bitmap_size;
4369         sit_i->dirty_sentries = 0;
4370         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4371         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4372         sit_i->mounted_time = ktime_get_boottime_seconds();
4373         init_rwsem(&sit_i->sentry_lock);
4374         return 0;
4375 }
4376
4377 static int build_free_segmap(struct f2fs_sb_info *sbi)
4378 {
4379         struct free_segmap_info *free_i;
4380         unsigned int bitmap_size, sec_bitmap_size;
4381
4382         /* allocate memory for free segmap information */
4383         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4384         if (!free_i)
4385                 return -ENOMEM;
4386
4387         SM_I(sbi)->free_info = free_i;
4388
4389         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4390         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4391         if (!free_i->free_segmap)
4392                 return -ENOMEM;
4393
4394         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4395         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4396         if (!free_i->free_secmap)
4397                 return -ENOMEM;
4398
4399         /* set all segments as dirty temporarily */
4400         memset(free_i->free_segmap, 0xff, bitmap_size);
4401         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4402
4403         /* init free segmap information */
4404         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4405         free_i->free_segments = 0;
4406         free_i->free_sections = 0;
4407         spin_lock_init(&free_i->segmap_lock);
4408         return 0;
4409 }
4410
4411 static int build_curseg(struct f2fs_sb_info *sbi)
4412 {
4413         struct curseg_info *array;
4414         int i;
4415
4416         array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4417                                         sizeof(*array)), GFP_KERNEL);
4418         if (!array)
4419                 return -ENOMEM;
4420
4421         SM_I(sbi)->curseg_array = array;
4422
4423         for (i = 0; i < NO_CHECK_TYPE; i++) {
4424                 mutex_init(&array[i].curseg_mutex);
4425                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4426                 if (!array[i].sum_blk)
4427                         return -ENOMEM;
4428                 init_rwsem(&array[i].journal_rwsem);
4429                 array[i].journal = f2fs_kzalloc(sbi,
4430                                 sizeof(struct f2fs_journal), GFP_KERNEL);
4431                 if (!array[i].journal)
4432                         return -ENOMEM;
4433                 if (i < NR_PERSISTENT_LOG)
4434                         array[i].seg_type = CURSEG_HOT_DATA + i;
4435                 else if (i == CURSEG_COLD_DATA_PINNED)
4436                         array[i].seg_type = CURSEG_COLD_DATA;
4437                 else if (i == CURSEG_ALL_DATA_ATGC)
4438                         array[i].seg_type = CURSEG_COLD_DATA;
4439                 array[i].segno = NULL_SEGNO;
4440                 array[i].next_blkoff = 0;
4441                 array[i].inited = false;
4442         }
4443         return restore_curseg_summaries(sbi);
4444 }
4445
4446 static int build_sit_entries(struct f2fs_sb_info *sbi)
4447 {
4448         struct sit_info *sit_i = SIT_I(sbi);
4449         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4450         struct f2fs_journal *journal = curseg->journal;
4451         struct seg_entry *se;
4452         struct f2fs_sit_entry sit;
4453         int sit_blk_cnt = SIT_BLK_CNT(sbi);
4454         unsigned int i, start, end;
4455         unsigned int readed, start_blk = 0;
4456         int err = 0;
4457         block_t total_node_blocks = 0;
4458
4459         do {
4460                 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS,
4461                                                         META_SIT, true);
4462
4463                 start = start_blk * sit_i->sents_per_block;
4464                 end = (start_blk + readed) * sit_i->sents_per_block;
4465
4466                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4467                         struct f2fs_sit_block *sit_blk;
4468                         struct page *page;
4469
4470                         se = &sit_i->sentries[start];
4471                         page = get_current_sit_page(sbi, start);
4472                         if (IS_ERR(page))
4473                                 return PTR_ERR(page);
4474                         sit_blk = (struct f2fs_sit_block *)page_address(page);
4475                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4476                         f2fs_put_page(page, 1);
4477
4478                         err = check_block_count(sbi, start, &sit);
4479                         if (err)
4480                                 return err;
4481                         seg_info_from_raw_sit(se, &sit);
4482                         if (IS_NODESEG(se->type))
4483                                 total_node_blocks += se->valid_blocks;
4484
4485                         /* build discard map only one time */
4486                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4487                                 memset(se->discard_map, 0xff,
4488                                         SIT_VBLOCK_MAP_SIZE);
4489                         } else {
4490                                 memcpy(se->discard_map,
4491                                         se->cur_valid_map,
4492                                         SIT_VBLOCK_MAP_SIZE);
4493                                 sbi->discard_blks +=
4494                                         sbi->blocks_per_seg -
4495                                         se->valid_blocks;
4496                         }
4497
4498                         if (__is_large_section(sbi))
4499                                 get_sec_entry(sbi, start)->valid_blocks +=
4500                                                         se->valid_blocks;
4501                 }
4502                 start_blk += readed;
4503         } while (start_blk < sit_blk_cnt);
4504
4505         down_read(&curseg->journal_rwsem);
4506         for (i = 0; i < sits_in_cursum(journal); i++) {
4507                 unsigned int old_valid_blocks;
4508
4509                 start = le32_to_cpu(segno_in_journal(journal, i));
4510                 if (start >= MAIN_SEGS(sbi)) {
4511                         f2fs_err(sbi, "Wrong journal entry on segno %u",
4512                                  start);
4513                         err = -EFSCORRUPTED;
4514                         break;
4515                 }
4516
4517                 se = &sit_i->sentries[start];
4518                 sit = sit_in_journal(journal, i);
4519
4520                 old_valid_blocks = se->valid_blocks;
4521                 if (IS_NODESEG(se->type))
4522                         total_node_blocks -= old_valid_blocks;
4523
4524                 err = check_block_count(sbi, start, &sit);
4525                 if (err)
4526                         break;
4527                 seg_info_from_raw_sit(se, &sit);
4528                 if (IS_NODESEG(se->type))
4529                         total_node_blocks += se->valid_blocks;
4530
4531                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4532                         memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4533                 } else {
4534                         memcpy(se->discard_map, se->cur_valid_map,
4535                                                 SIT_VBLOCK_MAP_SIZE);
4536                         sbi->discard_blks += old_valid_blocks;
4537                         sbi->discard_blks -= se->valid_blocks;
4538                 }
4539
4540                 if (__is_large_section(sbi)) {
4541                         get_sec_entry(sbi, start)->valid_blocks +=
4542                                                         se->valid_blocks;
4543                         get_sec_entry(sbi, start)->valid_blocks -=
4544                                                         old_valid_blocks;
4545                 }
4546         }
4547         up_read(&curseg->journal_rwsem);
4548
4549         if (!err && total_node_blocks != valid_node_count(sbi)) {
4550                 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4551                          total_node_blocks, valid_node_count(sbi));
4552                 err = -EFSCORRUPTED;
4553         }
4554
4555         return err;
4556 }
4557
4558 static void init_free_segmap(struct f2fs_sb_info *sbi)
4559 {
4560         unsigned int start;
4561         int type;
4562         struct seg_entry *sentry;
4563
4564         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4565                 if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4566                         continue;
4567                 sentry = get_seg_entry(sbi, start);
4568                 if (!sentry->valid_blocks)
4569                         __set_free(sbi, start);
4570                 else
4571                         SIT_I(sbi)->written_valid_blocks +=
4572                                                 sentry->valid_blocks;
4573         }
4574
4575         /* set use the current segments */
4576         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4577                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4578
4579                 __set_test_and_inuse(sbi, curseg_t->segno);
4580         }
4581 }
4582
4583 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4584 {
4585         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4586         struct free_segmap_info *free_i = FREE_I(sbi);
4587         unsigned int segno = 0, offset = 0, secno;
4588         block_t valid_blocks, usable_blks_in_seg;
4589         block_t blks_per_sec = BLKS_PER_SEC(sbi);
4590
4591         while (1) {
4592                 /* find dirty segment based on free segmap */
4593                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4594                 if (segno >= MAIN_SEGS(sbi))
4595                         break;
4596                 offset = segno + 1;
4597                 valid_blocks = get_valid_blocks(sbi, segno, false);
4598                 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4599                 if (valid_blocks == usable_blks_in_seg || !valid_blocks)
4600                         continue;
4601                 if (valid_blocks > usable_blks_in_seg) {
4602                         f2fs_bug_on(sbi, 1);
4603                         continue;
4604                 }
4605                 mutex_lock(&dirty_i->seglist_lock);
4606                 __locate_dirty_segment(sbi, segno, DIRTY);
4607                 mutex_unlock(&dirty_i->seglist_lock);
4608         }
4609
4610         if (!__is_large_section(sbi))
4611                 return;
4612
4613         mutex_lock(&dirty_i->seglist_lock);
4614         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4615                 valid_blocks = get_valid_blocks(sbi, segno, true);
4616                 secno = GET_SEC_FROM_SEG(sbi, segno);
4617
4618                 if (!valid_blocks || valid_blocks == blks_per_sec)
4619                         continue;
4620                 if (IS_CURSEC(sbi, secno))
4621                         continue;
4622                 set_bit(secno, dirty_i->dirty_secmap);
4623         }
4624         mutex_unlock(&dirty_i->seglist_lock);
4625 }
4626
4627 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4628 {
4629         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4630         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4631
4632         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4633         if (!dirty_i->victim_secmap)
4634                 return -ENOMEM;
4635         return 0;
4636 }
4637
4638 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4639 {
4640         struct dirty_seglist_info *dirty_i;
4641         unsigned int bitmap_size, i;
4642
4643         /* allocate memory for dirty segments list information */
4644         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4645                                                                 GFP_KERNEL);
4646         if (!dirty_i)
4647                 return -ENOMEM;
4648
4649         SM_I(sbi)->dirty_info = dirty_i;
4650         mutex_init(&dirty_i->seglist_lock);
4651
4652         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4653
4654         for (i = 0; i < NR_DIRTY_TYPE; i++) {
4655                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4656                                                                 GFP_KERNEL);
4657                 if (!dirty_i->dirty_segmap[i])
4658                         return -ENOMEM;
4659         }
4660
4661         if (__is_large_section(sbi)) {
4662                 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4663                 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
4664                                                 bitmap_size, GFP_KERNEL);
4665                 if (!dirty_i->dirty_secmap)
4666                         return -ENOMEM;
4667         }
4668
4669         init_dirty_segmap(sbi);
4670         return init_victim_secmap(sbi);
4671 }
4672
4673 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4674 {
4675         int i;
4676
4677         /*
4678          * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4679          * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4680          */
4681         for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4682                 struct curseg_info *curseg = CURSEG_I(sbi, i);
4683                 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4684                 unsigned int blkofs = curseg->next_blkoff;
4685
4686                 sanity_check_seg_type(sbi, curseg->seg_type);
4687
4688                 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4689                         goto out;
4690
4691                 if (curseg->alloc_type == SSR)
4692                         continue;
4693
4694                 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4695                         if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4696                                 continue;
4697 out:
4698                         f2fs_err(sbi,
4699                                  "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4700                                  i, curseg->segno, curseg->alloc_type,
4701                                  curseg->next_blkoff, blkofs);
4702                         return -EFSCORRUPTED;
4703                 }
4704         }
4705         return 0;
4706 }
4707
4708 #ifdef CONFIG_BLK_DEV_ZONED
4709
4710 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4711                                     struct f2fs_dev_info *fdev,
4712                                     struct blk_zone *zone)
4713 {
4714         unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno;
4715         block_t zone_block, wp_block, last_valid_block;
4716         unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4717         int i, s, b, ret;
4718         struct seg_entry *se;
4719
4720         if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4721                 return 0;
4722
4723         wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block);
4724         wp_segno = GET_SEGNO(sbi, wp_block);
4725         wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4726         zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4727         zone_segno = GET_SEGNO(sbi, zone_block);
4728         zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno);
4729
4730         if (zone_segno >= MAIN_SEGS(sbi))
4731                 return 0;
4732
4733         /*
4734          * Skip check of zones cursegs point to, since
4735          * fix_curseg_write_pointer() checks them.
4736          */
4737         for (i = 0; i < NO_CHECK_TYPE; i++)
4738                 if (zone_secno == GET_SEC_FROM_SEG(sbi,
4739                                                    CURSEG_I(sbi, i)->segno))
4740                         return 0;
4741
4742         /*
4743          * Get last valid block of the zone.
4744          */
4745         last_valid_block = zone_block - 1;
4746         for (s = sbi->segs_per_sec - 1; s >= 0; s--) {
4747                 segno = zone_segno + s;
4748                 se = get_seg_entry(sbi, segno);
4749                 for (b = sbi->blocks_per_seg - 1; b >= 0; b--)
4750                         if (f2fs_test_bit(b, se->cur_valid_map)) {
4751                                 last_valid_block = START_BLOCK(sbi, segno) + b;
4752                                 break;
4753                         }
4754                 if (last_valid_block >= zone_block)
4755                         break;
4756         }
4757
4758         /*
4759          * If last valid block is beyond the write pointer, report the
4760          * inconsistency. This inconsistency does not cause write error
4761          * because the zone will not be selected for write operation until
4762          * it get discarded. Just report it.
4763          */
4764         if (last_valid_block >= wp_block) {
4765                 f2fs_notice(sbi, "Valid block beyond write pointer: "
4766                             "valid block[0x%x,0x%x] wp[0x%x,0x%x]",
4767                             GET_SEGNO(sbi, last_valid_block),
4768                             GET_BLKOFF_FROM_SEG0(sbi, last_valid_block),
4769                             wp_segno, wp_blkoff);
4770                 return 0;
4771         }
4772
4773         /*
4774          * If there is no valid block in the zone and if write pointer is
4775          * not at zone start, reset the write pointer.
4776          */
4777         if (last_valid_block + 1 == zone_block && zone->wp != zone->start) {
4778                 f2fs_notice(sbi,
4779                             "Zone without valid block has non-zero write "
4780                             "pointer. Reset the write pointer: wp[0x%x,0x%x]",
4781                             wp_segno, wp_blkoff);
4782                 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
4783                                         zone->len >> log_sectors_per_block);
4784                 if (ret) {
4785                         f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4786                                  fdev->path, ret);
4787                         return ret;
4788                 }
4789         }
4790
4791         return 0;
4792 }
4793
4794 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
4795                                                   block_t zone_blkaddr)
4796 {
4797         int i;
4798
4799         for (i = 0; i < sbi->s_ndevs; i++) {
4800                 if (!bdev_is_zoned(FDEV(i).bdev))
4801                         continue;
4802                 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
4803                                 zone_blkaddr <= FDEV(i).end_blk))
4804                         return &FDEV(i);
4805         }
4806
4807         return NULL;
4808 }
4809
4810 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
4811                               void *data)
4812 {
4813         memcpy(data, zone, sizeof(struct blk_zone));
4814         return 0;
4815 }
4816
4817 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
4818 {
4819         struct curseg_info *cs = CURSEG_I(sbi, type);
4820         struct f2fs_dev_info *zbd;
4821         struct blk_zone zone;
4822         unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
4823         block_t cs_zone_block, wp_block;
4824         unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4825         sector_t zone_sector;
4826         int err;
4827
4828         cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4829         cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4830
4831         zbd = get_target_zoned_dev(sbi, cs_zone_block);
4832         if (!zbd)
4833                 return 0;
4834
4835         /* report zone for the sector the curseg points to */
4836         zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4837                 << log_sectors_per_block;
4838         err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4839                                   report_one_zone_cb, &zone);
4840         if (err != 1) {
4841                 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4842                          zbd->path, err);
4843                 return err;
4844         }
4845
4846         if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4847                 return 0;
4848
4849         wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
4850         wp_segno = GET_SEGNO(sbi, wp_block);
4851         wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4852         wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
4853
4854         if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
4855                 wp_sector_off == 0)
4856                 return 0;
4857
4858         f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
4859                     "curseg[0x%x,0x%x] wp[0x%x,0x%x]",
4860                     type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff);
4861
4862         f2fs_notice(sbi, "Assign new section to curseg[%d]: "
4863                     "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff);
4864
4865         f2fs_allocate_new_section(sbi, type, true);
4866
4867         /* check consistency of the zone curseg pointed to */
4868         if (check_zone_write_pointer(sbi, zbd, &zone))
4869                 return -EIO;
4870
4871         /* check newly assigned zone */
4872         cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4873         cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4874
4875         zbd = get_target_zoned_dev(sbi, cs_zone_block);
4876         if (!zbd)
4877                 return 0;
4878
4879         zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4880                 << log_sectors_per_block;
4881         err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4882                                   report_one_zone_cb, &zone);
4883         if (err != 1) {
4884                 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4885                          zbd->path, err);
4886                 return err;
4887         }
4888
4889         if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4890                 return 0;
4891
4892         if (zone.wp != zone.start) {
4893                 f2fs_notice(sbi,
4894                             "New zone for curseg[%d] is not yet discarded. "
4895                             "Reset the zone: curseg[0x%x,0x%x]",
4896                             type, cs->segno, cs->next_blkoff);
4897                 err = __f2fs_issue_discard_zone(sbi, zbd->bdev,
4898                                 zone_sector >> log_sectors_per_block,
4899                                 zone.len >> log_sectors_per_block);
4900                 if (err) {
4901                         f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4902                                  zbd->path, err);
4903                         return err;
4904                 }
4905         }
4906
4907         return 0;
4908 }
4909
4910 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4911 {
4912         int i, ret;
4913
4914         for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4915                 ret = fix_curseg_write_pointer(sbi, i);
4916                 if (ret)
4917                         return ret;
4918         }
4919
4920         return 0;
4921 }
4922
4923 struct check_zone_write_pointer_args {
4924         struct f2fs_sb_info *sbi;
4925         struct f2fs_dev_info *fdev;
4926 };
4927
4928 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
4929                                       void *data)
4930 {
4931         struct check_zone_write_pointer_args *args;
4932
4933         args = (struct check_zone_write_pointer_args *)data;
4934
4935         return check_zone_write_pointer(args->sbi, args->fdev, zone);
4936 }
4937
4938 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4939 {
4940         int i, ret;
4941         struct check_zone_write_pointer_args args;
4942
4943         for (i = 0; i < sbi->s_ndevs; i++) {
4944                 if (!bdev_is_zoned(FDEV(i).bdev))
4945                         continue;
4946
4947                 args.sbi = sbi;
4948                 args.fdev = &FDEV(i);
4949                 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
4950                                           check_zone_write_pointer_cb, &args);
4951                 if (ret < 0)
4952                         return ret;
4953         }
4954
4955         return 0;
4956 }
4957
4958 static bool is_conv_zone(struct f2fs_sb_info *sbi, unsigned int zone_idx,
4959                                                 unsigned int dev_idx)
4960 {
4961         if (!bdev_is_zoned(FDEV(dev_idx).bdev))
4962                 return true;
4963         return !test_bit(zone_idx, FDEV(dev_idx).blkz_seq);
4964 }
4965
4966 /* Return the zone index in the given device */
4967 static unsigned int get_zone_idx(struct f2fs_sb_info *sbi, unsigned int secno,
4968                                         int dev_idx)
4969 {
4970         block_t sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
4971
4972         return (sec_start_blkaddr - FDEV(dev_idx).start_blk) >>
4973                                                 sbi->log_blocks_per_blkz;
4974 }
4975
4976 /*
4977  * Return the usable segments in a section based on the zone's
4978  * corresponding zone capacity. Zone is equal to a section.
4979  */
4980 static inline unsigned int f2fs_usable_zone_segs_in_sec(
4981                 struct f2fs_sb_info *sbi, unsigned int segno)
4982 {
4983         unsigned int dev_idx, zone_idx, unusable_segs_in_sec;
4984
4985         dev_idx = f2fs_target_device_index(sbi, START_BLOCK(sbi, segno));
4986         zone_idx = get_zone_idx(sbi, GET_SEC_FROM_SEG(sbi, segno), dev_idx);
4987
4988         /* Conventional zone's capacity is always equal to zone size */
4989         if (is_conv_zone(sbi, zone_idx, dev_idx))
4990                 return sbi->segs_per_sec;
4991
4992         /*
4993          * If the zone_capacity_blocks array is NULL, then zone capacity
4994          * is equal to the zone size for all zones
4995          */
4996         if (!FDEV(dev_idx).zone_capacity_blocks)
4997                 return sbi->segs_per_sec;
4998
4999         /* Get the segment count beyond zone capacity block */
5000         unusable_segs_in_sec = (sbi->blocks_per_blkz -
5001                                 FDEV(dev_idx).zone_capacity_blocks[zone_idx]) >>
5002                                 sbi->log_blocks_per_seg;
5003         return sbi->segs_per_sec - unusable_segs_in_sec;
5004 }
5005
5006 /*
5007  * Return the number of usable blocks in a segment. The number of blocks
5008  * returned is always equal to the number of blocks in a segment for
5009  * segments fully contained within a sequential zone capacity or a
5010  * conventional zone. For segments partially contained in a sequential
5011  * zone capacity, the number of usable blocks up to the zone capacity
5012  * is returned. 0 is returned in all other cases.
5013  */
5014 static inline unsigned int f2fs_usable_zone_blks_in_seg(
5015                         struct f2fs_sb_info *sbi, unsigned int segno)
5016 {
5017         block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
5018         unsigned int zone_idx, dev_idx, secno;
5019
5020         secno = GET_SEC_FROM_SEG(sbi, segno);
5021         seg_start = START_BLOCK(sbi, segno);
5022         dev_idx = f2fs_target_device_index(sbi, seg_start);
5023         zone_idx = get_zone_idx(sbi, secno, dev_idx);
5024
5025         /*
5026          * Conventional zone's capacity is always equal to zone size,
5027          * so, blocks per segment is unchanged.
5028          */
5029         if (is_conv_zone(sbi, zone_idx, dev_idx))
5030                 return sbi->blocks_per_seg;
5031
5032         if (!FDEV(dev_idx).zone_capacity_blocks)
5033                 return sbi->blocks_per_seg;
5034
5035         sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
5036         sec_cap_blkaddr = sec_start_blkaddr +
5037                                 FDEV(dev_idx).zone_capacity_blocks[zone_idx];
5038
5039         /*
5040          * If segment starts before zone capacity and spans beyond
5041          * zone capacity, then usable blocks are from seg start to
5042          * zone capacity. If the segment starts after the zone capacity,
5043          * then there are no usable blocks.
5044          */
5045         if (seg_start >= sec_cap_blkaddr)
5046                 return 0;
5047         if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr)
5048                 return sec_cap_blkaddr - seg_start;
5049
5050         return sbi->blocks_per_seg;
5051 }
5052 #else
5053 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5054 {
5055         return 0;
5056 }
5057
5058 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5059 {
5060         return 0;
5061 }
5062
5063 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
5064                                                         unsigned int segno)
5065 {
5066         return 0;
5067 }
5068
5069 static inline unsigned int f2fs_usable_zone_segs_in_sec(struct f2fs_sb_info *sbi,
5070                                                         unsigned int segno)
5071 {
5072         return 0;
5073 }
5074 #endif
5075 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
5076                                         unsigned int segno)
5077 {
5078         if (f2fs_sb_has_blkzoned(sbi))
5079                 return f2fs_usable_zone_blks_in_seg(sbi, segno);
5080
5081         return sbi->blocks_per_seg;
5082 }
5083
5084 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi,
5085                                         unsigned int segno)
5086 {
5087         if (f2fs_sb_has_blkzoned(sbi))
5088                 return f2fs_usable_zone_segs_in_sec(sbi, segno);
5089
5090         return sbi->segs_per_sec;
5091 }
5092
5093 /*
5094  * Update min, max modified time for cost-benefit GC algorithm
5095  */
5096 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5097 {
5098         struct sit_info *sit_i = SIT_I(sbi);
5099         unsigned int segno;
5100
5101         down_write(&sit_i->sentry_lock);
5102
5103         sit_i->min_mtime = ULLONG_MAX;
5104
5105         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
5106                 unsigned int i;
5107                 unsigned long long mtime = 0;
5108
5109                 for (i = 0; i < sbi->segs_per_sec; i++)
5110                         mtime += get_seg_entry(sbi, segno + i)->mtime;
5111
5112                 mtime = div_u64(mtime, sbi->segs_per_sec);
5113
5114                 if (sit_i->min_mtime > mtime)
5115                         sit_i->min_mtime = mtime;
5116         }
5117         sit_i->max_mtime = get_mtime(sbi, false);
5118         sit_i->dirty_max_mtime = 0;
5119         up_write(&sit_i->sentry_lock);
5120 }
5121
5122 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
5123 {
5124         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5125         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
5126         struct f2fs_sm_info *sm_info;
5127         int err;
5128
5129         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
5130         if (!sm_info)
5131                 return -ENOMEM;
5132
5133         /* init sm info */
5134         sbi->sm_info = sm_info;
5135         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5136         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5137         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5138         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5139         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5140         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5141         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
5142         sm_info->rec_prefree_segments = sm_info->main_segments *
5143                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
5144         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5145                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5146
5147         if (!f2fs_lfs_mode(sbi))
5148                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
5149         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
5150         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
5151         sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
5152         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
5153         sm_info->min_ssr_sections = reserved_sections(sbi);
5154
5155         INIT_LIST_HEAD(&sm_info->sit_entry_set);
5156
5157         init_rwsem(&sm_info->curseg_lock);
5158
5159         if (!f2fs_readonly(sbi->sb)) {
5160                 err = f2fs_create_flush_cmd_control(sbi);
5161                 if (err)
5162                         return err;
5163         }
5164
5165         err = create_discard_cmd_control(sbi);
5166         if (err)
5167                 return err;
5168
5169         err = build_sit_info(sbi);
5170         if (err)
5171                 return err;
5172         err = build_free_segmap(sbi);
5173         if (err)
5174                 return err;
5175         err = build_curseg(sbi);
5176         if (err)
5177                 return err;
5178
5179         /* reinit free segmap based on SIT */
5180         err = build_sit_entries(sbi);
5181         if (err)
5182                 return err;
5183
5184         init_free_segmap(sbi);
5185         err = build_dirty_segmap(sbi);
5186         if (err)
5187                 return err;
5188
5189         err = sanity_check_curseg(sbi);
5190         if (err)
5191                 return err;
5192
5193         init_min_max_mtime(sbi);
5194         return 0;
5195 }
5196
5197 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5198                 enum dirty_type dirty_type)
5199 {
5200         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5201
5202         mutex_lock(&dirty_i->seglist_lock);
5203         kvfree(dirty_i->dirty_segmap[dirty_type]);
5204         dirty_i->nr_dirty[dirty_type] = 0;
5205         mutex_unlock(&dirty_i->seglist_lock);
5206 }
5207
5208 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5209 {
5210         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5211
5212         kvfree(dirty_i->victim_secmap);
5213 }
5214
5215 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5216 {
5217         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5218         int i;
5219
5220         if (!dirty_i)
5221                 return;
5222
5223         /* discard pre-free/dirty segments list */
5224         for (i = 0; i < NR_DIRTY_TYPE; i++)
5225                 discard_dirty_segmap(sbi, i);
5226
5227         if (__is_large_section(sbi)) {
5228                 mutex_lock(&dirty_i->seglist_lock);
5229                 kvfree(dirty_i->dirty_secmap);
5230                 mutex_unlock(&dirty_i->seglist_lock);
5231         }
5232
5233         destroy_victim_secmap(sbi);
5234         SM_I(sbi)->dirty_info = NULL;
5235         kfree(dirty_i);
5236 }
5237
5238 static void destroy_curseg(struct f2fs_sb_info *sbi)
5239 {
5240         struct curseg_info *array = SM_I(sbi)->curseg_array;
5241         int i;
5242
5243         if (!array)
5244                 return;
5245         SM_I(sbi)->curseg_array = NULL;
5246         for (i = 0; i < NR_CURSEG_TYPE; i++) {
5247                 kfree(array[i].sum_blk);
5248                 kfree(array[i].journal);
5249         }
5250         kfree(array);
5251 }
5252
5253 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5254 {
5255         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5256
5257         if (!free_i)
5258                 return;
5259         SM_I(sbi)->free_info = NULL;
5260         kvfree(free_i->free_segmap);
5261         kvfree(free_i->free_secmap);
5262         kfree(free_i);
5263 }
5264
5265 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5266 {
5267         struct sit_info *sit_i = SIT_I(sbi);
5268
5269         if (!sit_i)
5270                 return;
5271
5272         if (sit_i->sentries)
5273                 kvfree(sit_i->bitmap);
5274         kfree(sit_i->tmp_map);
5275
5276         kvfree(sit_i->sentries);
5277         kvfree(sit_i->sec_entries);
5278         kvfree(sit_i->dirty_sentries_bitmap);
5279
5280         SM_I(sbi)->sit_info = NULL;
5281         kvfree(sit_i->sit_bitmap);
5282 #ifdef CONFIG_F2FS_CHECK_FS
5283         kvfree(sit_i->sit_bitmap_mir);
5284         kvfree(sit_i->invalid_segmap);
5285 #endif
5286         kfree(sit_i);
5287 }
5288
5289 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5290 {
5291         struct f2fs_sm_info *sm_info = SM_I(sbi);
5292
5293         if (!sm_info)
5294                 return;
5295         f2fs_destroy_flush_cmd_control(sbi, true);
5296         destroy_discard_cmd_control(sbi);
5297         destroy_dirty_segmap(sbi);
5298         destroy_curseg(sbi);
5299         destroy_free_segmap(sbi);
5300         destroy_sit_info(sbi);
5301         sbi->sm_info = NULL;
5302         kfree(sm_info);
5303 }
5304
5305 int __init f2fs_create_segment_manager_caches(void)
5306 {
5307         discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5308                         sizeof(struct discard_entry));
5309         if (!discard_entry_slab)
5310                 goto fail;
5311
5312         discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5313                         sizeof(struct discard_cmd));
5314         if (!discard_cmd_slab)
5315                 goto destroy_discard_entry;
5316
5317         sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5318                         sizeof(struct sit_entry_set));
5319         if (!sit_entry_set_slab)
5320                 goto destroy_discard_cmd;
5321
5322         inmem_entry_slab = f2fs_kmem_cache_create("f2fs_inmem_page_entry",
5323                         sizeof(struct inmem_pages));
5324         if (!inmem_entry_slab)
5325                 goto destroy_sit_entry_set;
5326         return 0;
5327
5328 destroy_sit_entry_set:
5329         kmem_cache_destroy(sit_entry_set_slab);
5330 destroy_discard_cmd:
5331         kmem_cache_destroy(discard_cmd_slab);
5332 destroy_discard_entry:
5333         kmem_cache_destroy(discard_entry_slab);
5334 fail:
5335         return -ENOMEM;
5336 }
5337
5338 void f2fs_destroy_segment_manager_caches(void)
5339 {
5340         kmem_cache_destroy(sit_entry_set_slab);
5341         kmem_cache_destroy(discard_cmd_slab);
5342         kmem_cache_destroy(discard_entry_slab);
5343         kmem_cache_destroy(inmem_entry_slab);
5344 }