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/sched/mm.h>
  13 #include <linux/prefetch.h>
  14 #include <linux/kthread.h>
  15 #include <linux/swap.h>
  16 #include <linux/timer.h>
  17 #include <linux/freezer.h>
  18 #include <linux/sched/signal.h>
  19 #include <linux/random.h>
  20
  21 #include "f2fs.h"
  22 #include "segment.h"
  23 #include "node.h"
  24 #include "gc.h"
  25 #include "iostat.h"
  26 #include <trace/events/f2fs.h>
  27
  28 #define __reverse_ffz(x) __reverse_ffs(~(x))
  29
  30 static struct kmem_cache *discard_entry_slab;
  31 static struct kmem_cache *discard_cmd_slab;
  32 static struct kmem_cache *sit_entry_set_slab;
  33 static struct kmem_cache *revoke_entry_slab;
  34
  35 static unsigned long __reverse_ulong(unsigned char *str)
  36 {
  37         unsigned long tmp = 0;
  38         int shift = 24, idx = 0;
  39
  40 #if BITS_PER_LONG == 64
  41         shift = 56;
  42 #endif
  43         while (shift >= 0) {
  44                 tmp |= (unsigned long)str[idx++] << shift;
  45                 shift -= BITS_PER_BYTE;
  46         }
  47         return tmp;
  48 }
  49
  50 /*
  51  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  52  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  53  */
  54 static inline unsigned long __reverse_ffs(unsigned long word)
  55 {
  56         int num = 0;
  57
  58 #if BITS_PER_LONG == 64
  59         if ((word & 0xffffffff00000000UL) == 0)
  60                 num += 32;
  61         else
  62                 word >>= 32;
  63 #endif
  64         if ((word & 0xffff0000) == 0)
  65                 num += 16;
  66         else
  67                 word >>= 16;
  68
  69         if ((word & 0xff00) == 0)
  70                 num += 8;
  71         else
  72                 word >>= 8;
  73
  74         if ((word & 0xf0) == 0)
  75                 num += 4;
  76         else
  77                 word >>= 4;
  78
  79         if ((word & 0xc) == 0)
  80                 num += 2;
  81         else
  82                 word >>= 2;
  83
  84         if ((word & 0x2) == 0)
  85                 num += 1;
  86         return num;
  87 }
  88
  89 /*
  90  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  91  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  92  * @size must be integral times of unsigned long.
  93  * Example:
  94  *                             MSB <--> LSB
  95  *   f2fs_set_bit(0, bitmap) => 1000 0000
  96  *   f2fs_set_bit(7, bitmap) => 0000 0001
  97  */
  98 static unsigned long __find_rev_next_bit(const unsigned long *addr,
  99                         unsigned long size, unsigned long offset)
 100 {
 101         const unsigned long *p = addr + BIT_WORD(offset);
 102         unsigned long result = size;
 103         unsigned long tmp;
 104
 105         if (offset >= size)
 106                 return size;
 107
 108         size -= (offset & ~(BITS_PER_LONG - 1));
 109         offset %= BITS_PER_LONG;
 110
 111         while (1) {
 112                 if (*p == 0)
 113                         goto pass;
 114
 115                 tmp = __reverse_ulong((unsigned char *)p);
 116
 117                 tmp &= ~0UL >> offset;
 118                 if (size < BITS_PER_LONG)
 119                         tmp &= (~0UL << (BITS_PER_LONG - size));
 120                 if (tmp)
 121                         goto found;
 122 pass:
 123                 if (size <= BITS_PER_LONG)
 124                         break;
 125                 size -= BITS_PER_LONG;
 126                 offset = 0;
 127                 p++;
 128         }
 129         return result;
 130 found:
 131         return result - size + __reverse_ffs(tmp);
 132 }
 133
 134 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 135                         unsigned long size, unsigned long offset)
 136 {
 137         const unsigned long *p = addr + BIT_WORD(offset);
 138         unsigned long result = size;
 139         unsigned long tmp;
 140
 141         if (offset >= size)
 142                 return size;
 143
 144         size -= (offset & ~(BITS_PER_LONG - 1));
 145         offset %= BITS_PER_LONG;
 146
 147         while (1) {
 148                 if (*p == ~0UL)
 149                         goto pass;
 150
 151                 tmp = __reverse_ulong((unsigned char *)p);
 152
 153                 if (offset)
 154                         tmp |= ~0UL << (BITS_PER_LONG - offset);
 155                 if (size < BITS_PER_LONG)
 156                         tmp |= ~0UL >> size;
 157                 if (tmp != ~0UL)
 158                         goto found;
 159 pass:
 160                 if (size <= BITS_PER_LONG)
 161                         break;
 162                 size -= BITS_PER_LONG;
 163                 offset = 0;
 164                 p++;
 165         }
 166         return result;
 167 found:
 168         return result - size + __reverse_ffz(tmp);
 169 }
 170
 171 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
 172 {
 173         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 174         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 175         int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
 176
 177         if (f2fs_lfs_mode(sbi))
 178                 return false;
 179         if (sbi->gc_mode == GC_URGENT_HIGH)
 180                 return true;
 181         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
 182                 return true;
 183
 184         return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
 185                         SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
 186 }
 187
 188 void f2fs_abort_atomic_write(struct inode *inode, bool clean)
 189 {
 190         struct f2fs_inode_info *fi = F2FS_I(inode);
 191
 192         if (!f2fs_is_atomic_file(inode))
 193                 return;
 194
 195         clear_inode_flag(fi->cow_inode, FI_COW_FILE);
 196         iput(fi->cow_inode);
 197         fi->cow_inode = NULL;
 198         release_atomic_write_cnt(inode);
 199         clear_inode_flag(inode, FI_ATOMIC_COMMITTED);
 200         clear_inode_flag(inode, FI_ATOMIC_REPLACE);
 201         clear_inode_flag(inode, FI_ATOMIC_FILE);
 202         stat_dec_atomic_inode(inode);
 203
 204         if (clean) {
 205                 truncate_inode_pages_final(inode->i_mapping);
 206                 f2fs_i_size_write(inode, fi->original_i_size);
 207         }
 208 }
 209
 210 static int __replace_atomic_write_block(struct inode *inode, pgoff_t index,
 211                         block_t new_addr, block_t *old_addr, bool recover)
 212 {
 213         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 214         struct dnode_of_data dn;
 215         struct node_info ni;
 216         int err;
 217
 218 retry:
 219         set_new_dnode(&dn, inode, NULL, NULL, 0);
 220         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE_RA);
 221         if (err) {
 222                 if (err == -ENOMEM) {
 223                         f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
 224                         goto retry;
 225                 }
 226                 return err;
 227         }
 228
 229         err = f2fs_get_node_info(sbi, dn.nid, &ni, false);
 230         if (err) {
 231                 f2fs_put_dnode(&dn);
 232                 return err;
 233         }
 234
 235         if (recover) {
 236                 /* dn.data_blkaddr is always valid */
 237                 if (!__is_valid_data_blkaddr(new_addr)) {
 238                         if (new_addr == NULL_ADDR)
 239                                 dec_valid_block_count(sbi, inode, 1);
 240                         f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
 241                         f2fs_update_data_blkaddr(&dn, new_addr);
 242                 } else {
 243                         f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 244                                 new_addr, ni.version, true, true);
 245                 }
 246         } else {
 247                 blkcnt_t count = 1;
 248
 249                 *old_addr = dn.data_blkaddr;
 250                 f2fs_truncate_data_blocks_range(&dn, 1);
 251                 dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count);
 252                 inc_valid_block_count(sbi, inode, &count);
 253                 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr,
 254                                         ni.version, true, false);
 255         }
 256
 257         f2fs_put_dnode(&dn);
 258         return 0;
 259 }
 260
 261 static void __complete_revoke_list(struct inode *inode, struct list_head *head,
 262                                         bool revoke)
 263 {
 264         struct revoke_entry *cur, *tmp;
 265         bool truncate = is_inode_flag_set(inode, FI_ATOMIC_REPLACE);
 266
 267         list_for_each_entry_safe(cur, tmp, head, list) {
 268                 if (revoke)
 269                         __replace_atomic_write_block(inode, cur->index,
 270                                                 cur->old_addr, NULL, true);
 271
 272                 list_del(&cur->list);
 273                 kmem_cache_free(revoke_entry_slab, cur);
 274         }
 275
 276         if (!revoke && truncate)
 277                 f2fs_do_truncate_blocks(inode, 0, false);
 278 }
 279
 280 static int __f2fs_commit_atomic_write(struct inode *inode)
 281 {
 282         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 283         struct f2fs_inode_info *fi = F2FS_I(inode);
 284         struct inode *cow_inode = fi->cow_inode;
 285         struct revoke_entry *new;
 286         struct list_head revoke_list;
 287         block_t blkaddr;
 288         struct dnode_of_data dn;
 289         pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
 290         pgoff_t off = 0, blen, index;
 291         int ret = 0, i;
 292
 293         INIT_LIST_HEAD(&revoke_list);
 294
 295         while (len) {
 296                 blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len);
 297
 298                 set_new_dnode(&dn, cow_inode, NULL, NULL, 0);
 299                 ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA);
 300                 if (ret && ret != -ENOENT) {
 301                         goto out;
 302                 } else if (ret == -ENOENT) {
 303                         ret = 0;
 304                         if (dn.max_level == 0)
 305                                 goto out;
 306                         goto next;
 307                 }
 308
 309                 blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_page, cow_inode),
 310                                 len);
 311                 index = off;
 312                 for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) {
 313                         blkaddr = f2fs_data_blkaddr(&dn);
 314
 315                         if (!__is_valid_data_blkaddr(blkaddr)) {
 316                                 continue;
 317                         } else if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
 318                                         DATA_GENERIC_ENHANCE)) {
 319                                 f2fs_put_dnode(&dn);
 320                                 ret = -EFSCORRUPTED;
 321                                 f2fs_handle_error(sbi,
 322                                                 ERROR_INVALID_BLKADDR);
 323                                 goto out;
 324                         }
 325
 326                         new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS,
 327                                                         true, NULL);
 328
 329                         ret = __replace_atomic_write_block(inode, index, blkaddr,
 330                                                         &new->old_addr, false);
 331                         if (ret) {
 332                                 f2fs_put_dnode(&dn);
 333                                 kmem_cache_free(revoke_entry_slab, new);
 334                                 goto out;
 335                         }
 336
 337                         f2fs_update_data_blkaddr(&dn, NULL_ADDR);
 338                         new->index = index;
 339                         list_add_tail(&new->list, &revoke_list);
 340                 }
 341                 f2fs_put_dnode(&dn);
 342 next:
 343                 off += blen;
 344                 len -= blen;
 345         }
 346
 347 out:
 348         if (ret) {
 349                 sbi->revoked_atomic_block += fi->atomic_write_cnt;
 350         } else {
 351                 sbi->committed_atomic_block += fi->atomic_write_cnt;
 352                 set_inode_flag(inode, FI_ATOMIC_COMMITTED);
 353         }
 354
 355         __complete_revoke_list(inode, &revoke_list, ret ? true : false);
 356
 357         return ret;
 358 }
 359
 360 int f2fs_commit_atomic_write(struct inode *inode)
 361 {
 362         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 363         struct f2fs_inode_info *fi = F2FS_I(inode);
 364         int err;
 365
 366         err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX);
 367         if (err)
 368                 return err;
 369
 370         f2fs_down_write(&fi->i_gc_rwsem[WRITE]);
 371         f2fs_lock_op(sbi);
 372
 373         err = __f2fs_commit_atomic_write(inode);
 374
 375         f2fs_unlock_op(sbi);
 376         f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
 377
 378         return err;
 379 }
 380
 381 /*
 382  * This function balances dirty node and dentry pages.
 383  * In addition, it controls garbage collection.
 384  */
 385 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 386 {
 387         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 388                 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
 389                 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT);
 390         }
 391
 392         /* balance_fs_bg is able to be pending */
 393         if (need && excess_cached_nats(sbi))
 394                 f2fs_balance_fs_bg(sbi, false);
 395
 396         if (!f2fs_is_checkpoint_ready(sbi))
 397                 return;
 398
 399         /*
 400          * We should do GC or end up with checkpoint, if there are so many dirty
 401          * dir/node pages without enough free segments.
 402          */
 403         if (has_not_enough_free_secs(sbi, 0, 0)) {
 404                 if (test_opt(sbi, GC_MERGE) && sbi->gc_thread &&
 405                                         sbi->gc_thread->f2fs_gc_task) {
 406                         DEFINE_WAIT(wait);
 407
 408                         prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait,
 409                                                 TASK_UNINTERRUPTIBLE);
 410                         wake_up(&sbi->gc_thread->gc_wait_queue_head);
 411                         io_schedule();
 412                         finish_wait(&sbi->gc_thread->fggc_wq, &wait);
 413                 } else {
 414                         struct f2fs_gc_control gc_control = {
 415                                 .victim_segno = NULL_SEGNO,
 416                                 .init_gc_type = BG_GC,
 417                                 .no_bg_gc = true,
 418                                 .should_migrate_blocks = false,
 419                                 .err_gc_skipped = false,
 420                                 .nr_free_secs = 1 };
 421                         f2fs_down_write(&sbi->gc_lock);
 422                         f2fs_gc(sbi, &gc_control);
 423                 }
 424         }
 425 }
 426
 427 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi)
 428 {
 429         int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2;
 430         unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS);
 431         unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA);
 432         unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES);
 433         unsigned int meta = get_pages(sbi, F2FS_DIRTY_META);
 434         unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA);
 435         unsigned int threshold = sbi->blocks_per_seg * factor *
 436                                         DEFAULT_DIRTY_THRESHOLD;
 437         unsigned int global_threshold = threshold * 3 / 2;
 438
 439         if (dents >= threshold || qdata >= threshold ||
 440                 nodes >= threshold || meta >= threshold ||
 441                 imeta >= threshold)
 442                 return true;
 443         return dents + qdata + nodes + meta + imeta >  global_threshold;
 444 }
 445
 446 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
 447 {
 448         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 449                 return;
 450
 451         /* try to shrink extent cache when there is no enough memory */
 452         if (!f2fs_available_free_memory(sbi, READ_EXTENT_CACHE))
 453                 f2fs_shrink_read_extent_tree(sbi,
 454                                 READ_EXTENT_CACHE_SHRINK_NUMBER);
 455
 456         /* try to shrink age extent cache when there is no enough memory */
 457         if (!f2fs_available_free_memory(sbi, AGE_EXTENT_CACHE))
 458                 f2fs_shrink_age_extent_tree(sbi,
 459                                 AGE_EXTENT_CACHE_SHRINK_NUMBER);
 460
 461         /* check the # of cached NAT entries */
 462         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
 463                 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 464
 465         if (!f2fs_available_free_memory(sbi, FREE_NIDS))
 466                 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
 467         else
 468                 f2fs_build_free_nids(sbi, false, false);
 469
 470         if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) ||
 471                 excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi))
 472                 goto do_sync;
 473
 474         /* there is background inflight IO or foreground operation recently */
 475         if (is_inflight_io(sbi, REQ_TIME) ||
 476                 (!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem)))
 477                 return;
 478
 479         /* exceed periodical checkpoint timeout threshold */
 480         if (f2fs_time_over(sbi, CP_TIME))
 481                 goto do_sync;
 482
 483         /* checkpoint is the only way to shrink partial cached entries */
 484         if (f2fs_available_free_memory(sbi, NAT_ENTRIES) &&
 485                 f2fs_available_free_memory(sbi, INO_ENTRIES))
 486                 return;
 487
 488 do_sync:
 489         if (test_opt(sbi, DATA_FLUSH) && from_bg) {
 490                 struct blk_plug plug;
 491
 492                 mutex_lock(&sbi->flush_lock);
 493
 494                 blk_start_plug(&plug);
 495                 f2fs_sync_dirty_inodes(sbi, FILE_INODE, false);
 496                 blk_finish_plug(&plug);
 497
 498                 mutex_unlock(&sbi->flush_lock);
 499         }
 500         f2fs_sync_fs(sbi->sb, 1);
 501         stat_inc_bg_cp_count(sbi->stat_info);
 502 }
 503
 504 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 505                                 struct block_device *bdev)
 506 {
 507         int ret = blkdev_issue_flush(bdev);
 508
 509         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 510                                 test_opt(sbi, FLUSH_MERGE), ret);
 511         return ret;
 512 }
 513
 514 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 515 {
 516         int ret = 0;
 517         int i;
 518
 519         if (!f2fs_is_multi_device(sbi))
 520                 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 521
 522         for (i = 0; i < sbi->s_ndevs; i++) {
 523                 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
 524                         continue;
 525                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 526                 if (ret)
 527                         break;
 528         }
 529         return ret;
 530 }
 531
 532 static int issue_flush_thread(void *data)
 533 {
 534         struct f2fs_sb_info *sbi = data;
 535         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 536         wait_queue_head_t *q = &fcc->flush_wait_queue;
 537 repeat:
 538         if (kthread_should_stop())
 539                 return 0;
 540
 541         if (!llist_empty(&fcc->issue_list)) {
 542                 struct flush_cmd *cmd, *next;
 543                 int ret;
 544
 545                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 546                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 547
 548                 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 549
 550                 ret = submit_flush_wait(sbi, cmd->ino);
 551                 atomic_inc(&fcc->issued_flush);
 552
 553                 llist_for_each_entry_safe(cmd, next,
 554                                           fcc->dispatch_list, llnode) {
 555                         cmd->ret = ret;
 556                         complete(&cmd->wait);
 557                 }
 558                 fcc->dispatch_list = NULL;
 559         }
 560
 561         wait_event_interruptible(*q,
 562                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 563         goto repeat;
 564 }
 565
 566 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 567 {
 568         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 569         struct flush_cmd cmd;
 570         int ret;
 571
 572         if (test_opt(sbi, NOBARRIER))
 573                 return 0;
 574
 575         if (!test_opt(sbi, FLUSH_MERGE)) {
 576                 atomic_inc(&fcc->queued_flush);
 577                 ret = submit_flush_wait(sbi, ino);
 578                 atomic_dec(&fcc->queued_flush);
 579                 atomic_inc(&fcc->issued_flush);
 580                 return ret;
 581         }
 582
 583         if (atomic_inc_return(&fcc->queued_flush) == 1 ||
 584             f2fs_is_multi_device(sbi)) {
 585                 ret = submit_flush_wait(sbi, ino);
 586                 atomic_dec(&fcc->queued_flush);
 587
 588                 atomic_inc(&fcc->issued_flush);
 589                 return ret;
 590         }
 591
 592         cmd.ino = ino;
 593         init_completion(&cmd.wait);
 594
 595         llist_add(&cmd.llnode, &fcc->issue_list);
 596
 597         /*
 598          * update issue_list before we wake up issue_flush thread, this
 599          * smp_mb() pairs with another barrier in ___wait_event(), see
 600          * more details in comments of waitqueue_active().
 601          */
 602         smp_mb();
 603
 604         if (waitqueue_active(&fcc->flush_wait_queue))
 605                 wake_up(&fcc->flush_wait_queue);
 606
 607         if (fcc->f2fs_issue_flush) {
 608                 wait_for_completion(&cmd.wait);
 609                 atomic_dec(&fcc->queued_flush);
 610         } else {
 611                 struct llist_node *list;
 612
 613                 list = llist_del_all(&fcc->issue_list);
 614                 if (!list) {
 615                         wait_for_completion(&cmd.wait);
 616                         atomic_dec(&fcc->queued_flush);
 617                 } else {
 618                         struct flush_cmd *tmp, *next;
 619
 620                         ret = submit_flush_wait(sbi, ino);
 621
 622                         llist_for_each_entry_safe(tmp, next, list, llnode) {
 623                                 if (tmp == &cmd) {
 624                                         cmd.ret = ret;
 625                                         atomic_dec(&fcc->queued_flush);
 626                                         continue;
 627                                 }
 628                                 tmp->ret = ret;
 629                                 complete(&tmp->wait);
 630                         }
 631                 }
 632         }
 633
 634         return cmd.ret;
 635 }
 636
 637 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
 638 {
 639         dev_t dev = sbi->sb->s_bdev->bd_dev;
 640         struct flush_cmd_control *fcc;
 641
 642         if (SM_I(sbi)->fcc_info) {
 643                 fcc = SM_I(sbi)->fcc_info;
 644                 if (fcc->f2fs_issue_flush)
 645                         return 0;
 646                 goto init_thread;
 647         }
 648
 649         fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
 650         if (!fcc)
 651                 return -ENOMEM;
 652         atomic_set(&fcc->issued_flush, 0);
 653         atomic_set(&fcc->queued_flush, 0);
 654         init_waitqueue_head(&fcc->flush_wait_queue);
 655         init_llist_head(&fcc->issue_list);
 656         SM_I(sbi)->fcc_info = fcc;
 657         if (!test_opt(sbi, FLUSH_MERGE))
 658                 return 0;
 659
 660 init_thread:
 661         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 662                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 663         if (IS_ERR(fcc->f2fs_issue_flush)) {
 664                 int err = PTR_ERR(fcc->f2fs_issue_flush);
 665
 666                 fcc->f2fs_issue_flush = NULL;
 667                 return err;
 668         }
 669
 670         return 0;
 671 }
 672
 673 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 674 {
 675         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 676
 677         if (fcc && fcc->f2fs_issue_flush) {
 678                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 679
 680                 fcc->f2fs_issue_flush = NULL;
 681                 kthread_stop(flush_thread);
 682         }
 683         if (free) {
 684                 kfree(fcc);
 685                 SM_I(sbi)->fcc_info = NULL;
 686         }
 687 }
 688
 689 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
 690 {
 691         int ret = 0, i;
 692
 693         if (!f2fs_is_multi_device(sbi))
 694                 return 0;
 695
 696         if (test_opt(sbi, NOBARRIER))
 697                 return 0;
 698
 699         for (i = 1; i < sbi->s_ndevs; i++) {
 700                 int count = DEFAULT_RETRY_IO_COUNT;
 701
 702                 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
 703                         continue;
 704
 705                 do {
 706                         ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 707                         if (ret)
 708                                 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
 709                 } while (ret && --count);
 710
 711                 if (ret) {
 712                         f2fs_stop_checkpoint(sbi, false,
 713                                         STOP_CP_REASON_FLUSH_FAIL);
 714                         break;
 715                 }
 716
 717                 spin_lock(&sbi->dev_lock);
 718                 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
 719                 spin_unlock(&sbi->dev_lock);
 720         }
 721
 722         return ret;
 723 }
 724
 725 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 726                 enum dirty_type dirty_type)
 727 {
 728         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 729
 730         /* need not be added */
 731         if (IS_CURSEG(sbi, segno))
 732                 return;
 733
 734         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 735                 dirty_i->nr_dirty[dirty_type]++;
 736
 737         if (dirty_type == DIRTY) {
 738                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 739                 enum dirty_type t = sentry->type;
 740
 741                 if (unlikely(t >= DIRTY)) {
 742                         f2fs_bug_on(sbi, 1);
 743                         return;
 744                 }
 745                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 746                         dirty_i->nr_dirty[t]++;
 747
 748                 if (__is_large_section(sbi)) {
 749                         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 750                         block_t valid_blocks =
 751                                 get_valid_blocks(sbi, segno, true);
 752
 753                         f2fs_bug_on(sbi, unlikely(!valid_blocks ||
 754                                         valid_blocks == CAP_BLKS_PER_SEC(sbi)));
 755
 756                         if (!IS_CURSEC(sbi, secno))
 757                                 set_bit(secno, dirty_i->dirty_secmap);
 758                 }
 759         }
 760 }
 761
 762 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 763                 enum dirty_type dirty_type)
 764 {
 765         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 766         block_t valid_blocks;
 767
 768         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 769                 dirty_i->nr_dirty[dirty_type]--;
 770
 771         if (dirty_type == DIRTY) {
 772                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 773                 enum dirty_type t = sentry->type;
 774
 775                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 776                         dirty_i->nr_dirty[t]--;
 777
 778                 valid_blocks = get_valid_blocks(sbi, segno, true);
 779                 if (valid_blocks == 0) {
 780                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 781                                                 dirty_i->victim_secmap);
 782 #ifdef CONFIG_F2FS_CHECK_FS
 783                         clear_bit(segno, SIT_I(sbi)->invalid_segmap);
 784 #endif
 785                 }
 786                 if (__is_large_section(sbi)) {
 787                         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 788
 789                         if (!valid_blocks ||
 790                                         valid_blocks == CAP_BLKS_PER_SEC(sbi)) {
 791                                 clear_bit(secno, dirty_i->dirty_secmap);
 792                                 return;
 793                         }
 794
 795                         if (!IS_CURSEC(sbi, secno))
 796                                 set_bit(secno, dirty_i->dirty_secmap);
 797                 }
 798         }
 799 }
 800
 801 /*
 802  * Should not occur error such as -ENOMEM.
 803  * Adding dirty entry into seglist is not critical operation.
 804  * If a given segment is one of current working segments, it won't be added.
 805  */
 806 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 807 {
 808         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 809         unsigned short valid_blocks, ckpt_valid_blocks;
 810         unsigned int usable_blocks;
 811
 812         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 813                 return;
 814
 815         usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
 816         mutex_lock(&dirty_i->seglist_lock);
 817
 818         valid_blocks = get_valid_blocks(sbi, segno, false);
 819         ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false);
 820
 821         if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
 822                 ckpt_valid_blocks == usable_blocks)) {
 823                 __locate_dirty_segment(sbi, segno, PRE);
 824                 __remove_dirty_segment(sbi, segno, DIRTY);
 825         } else if (valid_blocks < usable_blocks) {
 826                 __locate_dirty_segment(sbi, segno, DIRTY);
 827         } else {
 828                 /* Recovery routine with SSR needs this */
 829                 __remove_dirty_segment(sbi, segno, DIRTY);
 830         }
 831
 832         mutex_unlock(&dirty_i->seglist_lock);
 833 }
 834
 835 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
 836 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
 837 {
 838         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 839         unsigned int segno;
 840
 841         mutex_lock(&dirty_i->seglist_lock);
 842         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 843                 if (get_valid_blocks(sbi, segno, false))
 844                         continue;
 845                 if (IS_CURSEG(sbi, segno))
 846                         continue;
 847                 __locate_dirty_segment(sbi, segno, PRE);
 848                 __remove_dirty_segment(sbi, segno, DIRTY);
 849         }
 850         mutex_unlock(&dirty_i->seglist_lock);
 851 }
 852
 853 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
 854 {
 855         int ovp_hole_segs =
 856                 (overprovision_segments(sbi) - reserved_segments(sbi));
 857         block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
 858         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 859         block_t holes[2] = {0, 0};      /* DATA and NODE */
 860         block_t unusable;
 861         struct seg_entry *se;
 862         unsigned int segno;
 863
 864         mutex_lock(&dirty_i->seglist_lock);
 865         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 866                 se = get_seg_entry(sbi, segno);
 867                 if (IS_NODESEG(se->type))
 868                         holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
 869                                                         se->valid_blocks;
 870                 else
 871                         holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
 872                                                         se->valid_blocks;
 873         }
 874         mutex_unlock(&dirty_i->seglist_lock);
 875
 876         unusable = max(holes[DATA], holes[NODE]);
 877         if (unusable > ovp_holes)
 878                 return unusable - ovp_holes;
 879         return 0;
 880 }
 881
 882 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
 883 {
 884         int ovp_hole_segs =
 885                 (overprovision_segments(sbi) - reserved_segments(sbi));
 886         if (unusable > F2FS_OPTION(sbi).unusable_cap)
 887                 return -EAGAIN;
 888         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
 889                 dirty_segments(sbi) > ovp_hole_segs)
 890                 return -EAGAIN;
 891         return 0;
 892 }
 893
 894 /* This is only used by SBI_CP_DISABLED */
 895 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
 896 {
 897         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 898         unsigned int segno = 0;
 899
 900         mutex_lock(&dirty_i->seglist_lock);
 901         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 902                 if (get_valid_blocks(sbi, segno, false))
 903                         continue;
 904                 if (get_ckpt_valid_blocks(sbi, segno, false))
 905                         continue;
 906                 mutex_unlock(&dirty_i->seglist_lock);
 907                 return segno;
 908         }
 909         mutex_unlock(&dirty_i->seglist_lock);
 910         return NULL_SEGNO;
 911 }
 912
 913 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 914                 struct block_device *bdev, block_t lstart,
 915                 block_t start, block_t len)
 916 {
 917         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 918         struct list_head *pend_list;
 919         struct discard_cmd *dc;
 920
 921         f2fs_bug_on(sbi, !len);
 922
 923         pend_list = &dcc->pend_list[plist_idx(len)];
 924
 925         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL);
 926         INIT_LIST_HEAD(&dc->list);
 927         dc->bdev = bdev;
 928         dc->lstart = lstart;
 929         dc->start = start;
 930         dc->len = len;
 931         dc->ref = 0;
 932         dc->state = D_PREP;
 933         dc->queued = 0;
 934         dc->error = 0;
 935         init_completion(&dc->wait);
 936         list_add_tail(&dc->list, pend_list);
 937         spin_lock_init(&dc->lock);
 938         dc->bio_ref = 0;
 939         atomic_inc(&dcc->discard_cmd_cnt);
 940         dcc->undiscard_blks += len;
 941
 942         return dc;
 943 }
 944
 945 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 946                                 struct block_device *bdev, block_t lstart,
 947                                 block_t start, block_t len,
 948                                 struct rb_node *parent, struct rb_node **p,
 949                                 bool leftmost)
 950 {
 951         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 952         struct discard_cmd *dc;
 953
 954         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
 955
 956         rb_link_node(&dc->rb_node, parent, p);
 957         rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
 958
 959         return dc;
 960 }
 961
 962 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
 963                                                         struct discard_cmd *dc)
 964 {
 965         if (dc->state == D_DONE)
 966                 atomic_sub(dc->queued, &dcc->queued_discard);
 967
 968         list_del(&dc->list);
 969         rb_erase_cached(&dc->rb_node, &dcc->root);
 970         dcc->undiscard_blks -= dc->len;
 971
 972         kmem_cache_free(discard_cmd_slab, dc);
 973
 974         atomic_dec(&dcc->discard_cmd_cnt);
 975 }
 976
 977 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
 978                                                         struct discard_cmd *dc)
 979 {
 980         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 981         unsigned long flags;
 982
 983         trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
 984
 985         spin_lock_irqsave(&dc->lock, flags);
 986         if (dc->bio_ref) {
 987                 spin_unlock_irqrestore(&dc->lock, flags);
 988                 return;
 989         }
 990         spin_unlock_irqrestore(&dc->lock, flags);
 991
 992         f2fs_bug_on(sbi, dc->ref);
 993
 994         if (dc->error == -EOPNOTSUPP)
 995                 dc->error = 0;
 996
 997         if (dc->error)
 998                 printk_ratelimited(
 999                         "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
1000                         KERN_INFO, sbi->sb->s_id,
1001                         dc->lstart, dc->start, dc->len, dc->error);
1002         __detach_discard_cmd(dcc, dc);
1003 }
1004
1005 static void f2fs_submit_discard_endio(struct bio *bio)
1006 {
1007         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1008         unsigned long flags;
1009
1010         spin_lock_irqsave(&dc->lock, flags);
1011         if (!dc->error)
1012                 dc->error = blk_status_to_errno(bio->bi_status);
1013         dc->bio_ref--;
1014         if (!dc->bio_ref && dc->state == D_SUBMIT) {
1015                 dc->state = D_DONE;
1016                 complete_all(&dc->wait);
1017         }
1018         spin_unlock_irqrestore(&dc->lock, flags);
1019         bio_put(bio);
1020 }
1021
1022 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1023                                 block_t start, block_t end)
1024 {
1025 #ifdef CONFIG_F2FS_CHECK_FS
1026         struct seg_entry *sentry;
1027         unsigned int segno;
1028         block_t blk = start;
1029         unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1030         unsigned long *map;
1031
1032         while (blk < end) {
1033                 segno = GET_SEGNO(sbi, blk);
1034                 sentry = get_seg_entry(sbi, segno);
1035                 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1036
1037                 if (end < START_BLOCK(sbi, segno + 1))
1038                         size = GET_BLKOFF_FROM_SEG0(sbi, end);
1039                 else
1040                         size = max_blocks;
1041                 map = (unsigned long *)(sentry->cur_valid_map);
1042                 offset = __find_rev_next_bit(map, size, offset);
1043                 f2fs_bug_on(sbi, offset != size);
1044                 blk = START_BLOCK(sbi, segno + 1);
1045         }
1046 #endif
1047 }
1048
1049 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1050                                 struct discard_policy *dpolicy,
1051                                 int discard_type, unsigned int granularity)
1052 {
1053         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1054
1055         /* common policy */
1056         dpolicy->type = discard_type;
1057         dpolicy->sync = true;
1058         dpolicy->ordered = false;
1059         dpolicy->granularity = granularity;
1060
1061         dpolicy->max_requests = dcc->max_discard_request;
1062         dpolicy->io_aware_gran = MAX_PLIST_NUM;
1063         dpolicy->timeout = false;
1064
1065         if (discard_type == DPOLICY_BG) {
1066                 dpolicy->min_interval = dcc->min_discard_issue_time;
1067                 dpolicy->mid_interval = dcc->mid_discard_issue_time;
1068                 dpolicy->max_interval = dcc->max_discard_issue_time;
1069                 dpolicy->io_aware = true;
1070                 dpolicy->sync = false;
1071                 dpolicy->ordered = true;
1072                 if (utilization(sbi) > dcc->discard_urgent_util) {
1073                         dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1074                         if (atomic_read(&dcc->discard_cmd_cnt))
1075                                 dpolicy->max_interval =
1076                                         dcc->min_discard_issue_time;
1077                 }
1078         } else if (discard_type == DPOLICY_FORCE) {
1079                 dpolicy->min_interval = dcc->min_discard_issue_time;
1080                 dpolicy->mid_interval = dcc->mid_discard_issue_time;
1081                 dpolicy->max_interval = dcc->max_discard_issue_time;
1082                 dpolicy->io_aware = false;
1083         } else if (discard_type == DPOLICY_FSTRIM) {
1084                 dpolicy->io_aware = false;
1085         } else if (discard_type == DPOLICY_UMOUNT) {
1086                 dpolicy->io_aware = false;
1087                 /* we need to issue all to keep CP_TRIMMED_FLAG */
1088                 dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1089                 dpolicy->timeout = true;
1090         }
1091 }
1092
1093 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1094                                 struct block_device *bdev, block_t lstart,
1095                                 block_t start, block_t len);
1096 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1097 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1098                                                 struct discard_policy *dpolicy,
1099                                                 struct discard_cmd *dc,
1100                                                 unsigned int *issued)
1101 {
1102         struct block_device *bdev = dc->bdev;
1103         unsigned int max_discard_blocks =
1104                         SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1105         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1106         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1107                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1108         blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0;
1109         block_t lstart, start, len, total_len;
1110         int err = 0;
1111
1112         if (dc->state != D_PREP)
1113                 return 0;
1114
1115         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1116                 return 0;
1117
1118         trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1119
1120         lstart = dc->lstart;
1121         start = dc->start;
1122         len = dc->len;
1123         total_len = len;
1124
1125         dc->len = 0;
1126
1127         while (total_len && *issued < dpolicy->max_requests && !err) {
1128                 struct bio *bio = NULL;
1129                 unsigned long flags;
1130                 bool last = true;
1131
1132                 if (len > max_discard_blocks) {
1133                         len = max_discard_blocks;
1134                         last = false;
1135                 }
1136
1137                 (*issued)++;
1138                 if (*issued == dpolicy->max_requests)
1139                         last = true;
1140
1141                 dc->len += len;
1142
1143                 if (time_to_inject(sbi, FAULT_DISCARD)) {
1144                         f2fs_show_injection_info(sbi, FAULT_DISCARD);
1145                         err = -EIO;
1146                 } else {
1147                         err = __blkdev_issue_discard(bdev,
1148                                         SECTOR_FROM_BLOCK(start),
1149                                         SECTOR_FROM_BLOCK(len),
1150                                         GFP_NOFS, &bio);
1151                 }
1152                 if (err) {
1153                         spin_lock_irqsave(&dc->lock, flags);
1154                         if (dc->state == D_PARTIAL)
1155                                 dc->state = D_SUBMIT;
1156                         spin_unlock_irqrestore(&dc->lock, flags);
1157
1158                         break;
1159                 }
1160
1161                 f2fs_bug_on(sbi, !bio);
1162
1163                 /*
1164                  * should keep before submission to avoid D_DONE
1165                  * right away
1166                  */
1167                 spin_lock_irqsave(&dc->lock, flags);
1168                 if (last)
1169                         dc->state = D_SUBMIT;
1170                 else
1171                         dc->state = D_PARTIAL;
1172                 dc->bio_ref++;
1173                 spin_unlock_irqrestore(&dc->lock, flags);
1174
1175                 atomic_inc(&dcc->queued_discard);
1176                 dc->queued++;
1177                 list_move_tail(&dc->list, wait_list);
1178
1179                 /* sanity check on discard range */
1180                 __check_sit_bitmap(sbi, lstart, lstart + len);
1181
1182                 bio->bi_private = dc;
1183                 bio->bi_end_io = f2fs_submit_discard_endio;
1184                 bio->bi_opf |= flag;
1185                 submit_bio(bio);
1186
1187                 atomic_inc(&dcc->issued_discard);
1188
1189                 f2fs_update_iostat(sbi, NULL, FS_DISCARD, len * F2FS_BLKSIZE);
1190
1191                 lstart += len;
1192                 start += len;
1193                 total_len -= len;
1194                 len = total_len;
1195         }
1196
1197         if (!err && len) {
1198                 dcc->undiscard_blks -= len;
1199                 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1200         }
1201         return err;
1202 }
1203
1204 static void __insert_discard_tree(struct f2fs_sb_info *sbi,
1205                                 struct block_device *bdev, block_t lstart,
1206                                 block_t start, block_t len,
1207                                 struct rb_node **insert_p,
1208                                 struct rb_node *insert_parent)
1209 {
1210         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1211         struct rb_node **p;
1212         struct rb_node *parent = NULL;
1213         bool leftmost = true;
1214
1215         if (insert_p && insert_parent) {
1216                 parent = insert_parent;
1217                 p = insert_p;
1218                 goto do_insert;
1219         }
1220
1221         p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1222                                                         lstart, &leftmost);
1223 do_insert:
1224         __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1225                                                                 p, leftmost);
1226 }
1227
1228 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1229                                                 struct discard_cmd *dc)
1230 {
1231         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1232 }
1233
1234 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1235                                 struct discard_cmd *dc, block_t blkaddr)
1236 {
1237         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1238         struct discard_info di = dc->di;
1239         bool modified = false;
1240
1241         if (dc->state == D_DONE || dc->len == 1) {
1242                 __remove_discard_cmd(sbi, dc);
1243                 return;
1244         }
1245
1246         dcc->undiscard_blks -= di.len;
1247
1248         if (blkaddr > di.lstart) {
1249                 dc->len = blkaddr - dc->lstart;
1250                 dcc->undiscard_blks += dc->len;
1251                 __relocate_discard_cmd(dcc, dc);
1252                 modified = true;
1253         }
1254
1255         if (blkaddr < di.lstart + di.len - 1) {
1256                 if (modified) {
1257                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1258                                         di.start + blkaddr + 1 - di.lstart,
1259                                         di.lstart + di.len - 1 - blkaddr,
1260                                         NULL, NULL);
1261                 } else {
1262                         dc->lstart++;
1263                         dc->len--;
1264                         dc->start++;
1265                         dcc->undiscard_blks += dc->len;
1266                         __relocate_discard_cmd(dcc, dc);
1267                 }
1268         }
1269 }
1270
1271 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1272                                 struct block_device *bdev, block_t lstart,
1273                                 block_t start, block_t len)
1274 {
1275         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1276         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1277         struct discard_cmd *dc;
1278         struct discard_info di = {0};
1279         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1280         unsigned int max_discard_blocks =
1281                         SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1282         block_t end = lstart + len;
1283
1284         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1285                                         NULL, lstart,
1286                                         (struct rb_entry **)&prev_dc,
1287                                         (struct rb_entry **)&next_dc,
1288                                         &insert_p, &insert_parent, true, NULL);
1289         if (dc)
1290                 prev_dc = dc;
1291
1292         if (!prev_dc) {
1293                 di.lstart = lstart;
1294                 di.len = next_dc ? next_dc->lstart - lstart : len;
1295                 di.len = min(di.len, len);
1296                 di.start = start;
1297         }
1298
1299         while (1) {
1300                 struct rb_node *node;
1301                 bool merged = false;
1302                 struct discard_cmd *tdc = NULL;
1303
1304                 if (prev_dc) {
1305                         di.lstart = prev_dc->lstart + prev_dc->len;
1306                         if (di.lstart < lstart)
1307                                 di.lstart = lstart;
1308                         if (di.lstart >= end)
1309                                 break;
1310
1311                         if (!next_dc || next_dc->lstart > end)
1312                                 di.len = end - di.lstart;
1313                         else
1314                                 di.len = next_dc->lstart - di.lstart;
1315                         di.start = start + di.lstart - lstart;
1316                 }
1317
1318                 if (!di.len)
1319                         goto next;
1320
1321                 if (prev_dc && prev_dc->state == D_PREP &&
1322                         prev_dc->bdev == bdev &&
1323                         __is_discard_back_mergeable(&di, &prev_dc->di,
1324                                                         max_discard_blocks)) {
1325                         prev_dc->di.len += di.len;
1326                         dcc->undiscard_blks += di.len;
1327                         __relocate_discard_cmd(dcc, prev_dc);
1328                         di = prev_dc->di;
1329                         tdc = prev_dc;
1330                         merged = true;
1331                 }
1332
1333                 if (next_dc && next_dc->state == D_PREP &&
1334                         next_dc->bdev == bdev &&
1335                         __is_discard_front_mergeable(&di, &next_dc->di,
1336                                                         max_discard_blocks)) {
1337                         next_dc->di.lstart = di.lstart;
1338                         next_dc->di.len += di.len;
1339                         next_dc->di.start = di.start;
1340                         dcc->undiscard_blks += di.len;
1341                         __relocate_discard_cmd(dcc, next_dc);
1342                         if (tdc)
1343                                 __remove_discard_cmd(sbi, tdc);
1344                         merged = true;
1345                 }
1346
1347                 if (!merged) {
1348                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1349                                                         di.len, NULL, NULL);
1350                 }
1351  next:
1352                 prev_dc = next_dc;
1353                 if (!prev_dc)
1354                         break;
1355
1356                 node = rb_next(&prev_dc->rb_node);
1357                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1358         }
1359 }
1360
1361 static void __queue_discard_cmd(struct f2fs_sb_info *sbi,
1362                 struct block_device *bdev, block_t blkstart, block_t blklen)
1363 {
1364         block_t lblkstart = blkstart;
1365
1366         if (!f2fs_bdev_support_discard(bdev))
1367                 return;
1368
1369         trace_f2fs_queue_discard(bdev, blkstart, blklen);
1370
1371         if (f2fs_is_multi_device(sbi)) {
1372                 int devi = f2fs_target_device_index(sbi, blkstart);
1373
1374                 blkstart -= FDEV(devi).start_blk;
1375         }
1376         mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1377         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1378         mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1379 }
1380
1381 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1382                                         struct discard_policy *dpolicy)
1383 {
1384         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1385         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1386         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1387         struct discard_cmd *dc;
1388         struct blk_plug plug;
1389         unsigned int pos = dcc->next_pos;
1390         unsigned int issued = 0;
1391         bool io_interrupted = false;
1392
1393         mutex_lock(&dcc->cmd_lock);
1394         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1395                                         NULL, pos,
1396                                         (struct rb_entry **)&prev_dc,
1397                                         (struct rb_entry **)&next_dc,
1398                                         &insert_p, &insert_parent, true, NULL);
1399         if (!dc)
1400                 dc = next_dc;
1401
1402         blk_start_plug(&plug);
1403
1404         while (dc) {
1405                 struct rb_node *node;
1406                 int err = 0;
1407
1408                 if (dc->state != D_PREP)
1409                         goto next;
1410
1411                 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1412                         io_interrupted = true;
1413                         break;
1414                 }
1415
1416                 dcc->next_pos = dc->lstart + dc->len;
1417                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1418
1419                 if (issued >= dpolicy->max_requests)
1420                         break;
1421 next:
1422                 node = rb_next(&dc->rb_node);
1423                 if (err)
1424                         __remove_discard_cmd(sbi, dc);
1425                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1426         }
1427
1428         blk_finish_plug(&plug);
1429
1430         if (!dc)
1431                 dcc->next_pos = 0;
1432
1433         mutex_unlock(&dcc->cmd_lock);
1434
1435         if (!issued && io_interrupted)
1436                 issued = -1;
1437
1438         return issued;
1439 }
1440 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1441                                         struct discard_policy *dpolicy);
1442
1443 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1444                                         struct discard_policy *dpolicy)
1445 {
1446         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1447         struct list_head *pend_list;
1448         struct discard_cmd *dc, *tmp;
1449         struct blk_plug plug;
1450         int i, issued;
1451         bool io_interrupted = false;
1452
1453         if (dpolicy->timeout)
1454                 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1455
1456 retry:
1457         issued = 0;
1458         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1459                 if (dpolicy->timeout &&
1460                                 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1461                         break;
1462
1463                 if (i + 1 < dpolicy->granularity)
1464                         break;
1465
1466                 if (i + 1 < dcc->max_ordered_discard && dpolicy->ordered)
1467                         return __issue_discard_cmd_orderly(sbi, dpolicy);
1468
1469                 pend_list = &dcc->pend_list[i];
1470
1471                 mutex_lock(&dcc->cmd_lock);
1472                 if (list_empty(pend_list))
1473                         goto next;
1474                 if (unlikely(dcc->rbtree_check))
1475                         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1476                                                         &dcc->root, false));
1477                 blk_start_plug(&plug);
1478                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1479                         f2fs_bug_on(sbi, dc->state != D_PREP);
1480
1481                         if (dpolicy->timeout &&
1482                                 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1483                                 break;
1484
1485                         if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1486                                                 !is_idle(sbi, DISCARD_TIME)) {
1487                                 io_interrupted = true;
1488                                 break;
1489                         }
1490
1491                         __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1492
1493                         if (issued >= dpolicy->max_requests)
1494                                 break;
1495                 }
1496                 blk_finish_plug(&plug);
1497 next:
1498                 mutex_unlock(&dcc->cmd_lock);
1499
1500                 if (issued >= dpolicy->max_requests || io_interrupted)
1501                         break;
1502         }
1503
1504         if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1505                 __wait_all_discard_cmd(sbi, dpolicy);
1506                 goto retry;
1507         }
1508
1509         if (!issued && io_interrupted)
1510                 issued = -1;
1511
1512         return issued;
1513 }
1514
1515 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1516 {
1517         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1518         struct list_head *pend_list;
1519         struct discard_cmd *dc, *tmp;
1520         int i;
1521         bool dropped = false;
1522
1523         mutex_lock(&dcc->cmd_lock);
1524         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1525                 pend_list = &dcc->pend_list[i];
1526                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1527                         f2fs_bug_on(sbi, dc->state != D_PREP);
1528                         __remove_discard_cmd(sbi, dc);
1529                         dropped = true;
1530                 }
1531         }
1532         mutex_unlock(&dcc->cmd_lock);
1533
1534         return dropped;
1535 }
1536
1537 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1538 {
1539         __drop_discard_cmd(sbi);
1540 }
1541
1542 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1543                                                         struct discard_cmd *dc)
1544 {
1545         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1546         unsigned int len = 0;
1547
1548         wait_for_completion_io(&dc->wait);
1549         mutex_lock(&dcc->cmd_lock);
1550         f2fs_bug_on(sbi, dc->state != D_DONE);
1551         dc->ref--;
1552         if (!dc->ref) {
1553                 if (!dc->error)
1554                         len = dc->len;
1555                 __remove_discard_cmd(sbi, dc);
1556         }
1557         mutex_unlock(&dcc->cmd_lock);
1558
1559         return len;
1560 }
1561
1562 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1563                                                 struct discard_policy *dpolicy,
1564                                                 block_t start, block_t end)
1565 {
1566         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1567         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1568                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1569         struct discard_cmd *dc = NULL, *iter, *tmp;
1570         unsigned int trimmed = 0;
1571
1572 next:
1573         dc = NULL;
1574
1575         mutex_lock(&dcc->cmd_lock);
1576         list_for_each_entry_safe(iter, tmp, wait_list, list) {
1577                 if (iter->lstart + iter->len <= start || end <= iter->lstart)
1578                         continue;
1579                 if (iter->len < dpolicy->granularity)
1580                         continue;
1581                 if (iter->state == D_DONE && !iter->ref) {
1582                         wait_for_completion_io(&iter->wait);
1583                         if (!iter->error)
1584                                 trimmed += iter->len;
1585                         __remove_discard_cmd(sbi, iter);
1586                 } else {
1587                         iter->ref++;
1588                         dc = iter;
1589                         break;
1590                 }
1591         }
1592         mutex_unlock(&dcc->cmd_lock);
1593
1594         if (dc) {
1595                 trimmed += __wait_one_discard_bio(sbi, dc);
1596                 goto next;
1597         }
1598
1599         return trimmed;
1600 }
1601
1602 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1603                                                 struct discard_policy *dpolicy)
1604 {
1605         struct discard_policy dp;
1606         unsigned int discard_blks;
1607
1608         if (dpolicy)
1609                 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1610
1611         /* wait all */
1612         __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1613         discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1614         __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1615         discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1616
1617         return discard_blks;
1618 }
1619
1620 /* This should be covered by global mutex, &sit_i->sentry_lock */
1621 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1622 {
1623         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1624         struct discard_cmd *dc;
1625         bool need_wait = false;
1626
1627         mutex_lock(&dcc->cmd_lock);
1628         dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1629                                                         NULL, blkaddr);
1630         if (dc) {
1631                 if (dc->state == D_PREP) {
1632                         __punch_discard_cmd(sbi, dc, blkaddr);
1633                 } else {
1634                         dc->ref++;
1635                         need_wait = true;
1636                 }
1637         }
1638         mutex_unlock(&dcc->cmd_lock);
1639
1640         if (need_wait)
1641                 __wait_one_discard_bio(sbi, dc);
1642 }
1643
1644 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1645 {
1646         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1647
1648         if (dcc && dcc->f2fs_issue_discard) {
1649                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1650
1651                 dcc->f2fs_issue_discard = NULL;
1652                 kthread_stop(discard_thread);
1653         }
1654 }
1655
1656 /* This comes from f2fs_put_super */
1657 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1658 {
1659         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1660         struct discard_policy dpolicy;
1661         bool dropped;
1662
1663         if (!atomic_read(&dcc->discard_cmd_cnt))
1664                 return false;
1665
1666         __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1667                                         dcc->discard_granularity);
1668         __issue_discard_cmd(sbi, &dpolicy);
1669         dropped = __drop_discard_cmd(sbi);
1670
1671         /* just to make sure there is no pending discard commands */
1672         __wait_all_discard_cmd(sbi, NULL);
1673
1674         f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1675         return dropped;
1676 }
1677
1678 static int issue_discard_thread(void *data)
1679 {
1680         struct f2fs_sb_info *sbi = data;
1681         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1682         wait_queue_head_t *q = &dcc->discard_wait_queue;
1683         struct discard_policy dpolicy;
1684         unsigned int wait_ms = dcc->min_discard_issue_time;
1685         int issued;
1686
1687         set_freezable();
1688
1689         do {
1690                 wait_event_interruptible_timeout(*q,
1691                                 kthread_should_stop() || freezing(current) ||
1692                                 dcc->discard_wake,
1693                                 msecs_to_jiffies(wait_ms));
1694
1695                 if (sbi->gc_mode == GC_URGENT_HIGH ||
1696                         !f2fs_available_free_memory(sbi, DISCARD_CACHE))
1697                         __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1698                 else
1699                         __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1700                                                 dcc->discard_granularity);
1701
1702                 if (dcc->discard_wake)
1703                         dcc->discard_wake = 0;
1704
1705                 /* clean up pending candidates before going to sleep */
1706                 if (atomic_read(&dcc->queued_discard))
1707                         __wait_all_discard_cmd(sbi, NULL);
1708
1709                 if (try_to_freeze())
1710                         continue;
1711                 if (f2fs_readonly(sbi->sb))
1712                         continue;
1713                 if (kthread_should_stop())
1714                         return 0;
1715                 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) ||
1716                         !atomic_read(&dcc->discard_cmd_cnt)) {
1717                         wait_ms = dpolicy.max_interval;
1718                         continue;
1719                 }
1720
1721                 sb_start_intwrite(sbi->sb);
1722
1723                 issued = __issue_discard_cmd(sbi, &dpolicy);
1724                 if (issued > 0) {
1725                         __wait_all_discard_cmd(sbi, &dpolicy);
1726                         wait_ms = dpolicy.min_interval;
1727                 } else if (issued == -1) {
1728                         wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1729                         if (!wait_ms)
1730                                 wait_ms = dpolicy.mid_interval;
1731                 } else {
1732                         wait_ms = dpolicy.max_interval;
1733                 }
1734                 if (!atomic_read(&dcc->discard_cmd_cnt))
1735                         wait_ms = dpolicy.max_interval;
1736
1737                 sb_end_intwrite(sbi->sb);
1738
1739         } while (!kthread_should_stop());
1740         return 0;
1741 }
1742
1743 #ifdef CONFIG_BLK_DEV_ZONED
1744 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1745                 struct block_device *bdev, block_t blkstart, block_t blklen)
1746 {
1747         sector_t sector, nr_sects;
1748         block_t lblkstart = blkstart;
1749         int devi = 0;
1750
1751         if (f2fs_is_multi_device(sbi)) {
1752                 devi = f2fs_target_device_index(sbi, blkstart);
1753                 if (blkstart < FDEV(devi).start_blk ||
1754                     blkstart > FDEV(devi).end_blk) {
1755                         f2fs_err(sbi, "Invalid block %x", blkstart);
1756                         return -EIO;
1757                 }
1758                 blkstart -= FDEV(devi).start_blk;
1759         }
1760
1761         /* For sequential zones, reset the zone write pointer */
1762         if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1763                 sector = SECTOR_FROM_BLOCK(blkstart);
1764                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1765
1766                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1767                                 nr_sects != bdev_zone_sectors(bdev)) {
1768                         f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1769                                  devi, sbi->s_ndevs ? FDEV(devi).path : "",
1770                                  blkstart, blklen);
1771                         return -EIO;
1772                 }
1773                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1774                 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1775                                         sector, nr_sects, GFP_NOFS);
1776         }
1777
1778         /* For conventional zones, use regular discard if supported */
1779         __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1780         return 0;
1781 }
1782 #endif
1783
1784 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1785                 struct block_device *bdev, block_t blkstart, block_t blklen)
1786 {
1787 #ifdef CONFIG_BLK_DEV_ZONED
1788         if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1789                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1790 #endif
1791         __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1792         return 0;
1793 }
1794
1795 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1796                                 block_t blkstart, block_t blklen)
1797 {
1798         sector_t start = blkstart, len = 0;
1799         struct block_device *bdev;
1800         struct seg_entry *se;
1801         unsigned int offset;
1802         block_t i;
1803         int err = 0;
1804
1805         bdev = f2fs_target_device(sbi, blkstart, NULL);
1806
1807         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1808                 if (i != start) {
1809                         struct block_device *bdev2 =
1810                                 f2fs_target_device(sbi, i, NULL);
1811
1812                         if (bdev2 != bdev) {
1813                                 err = __issue_discard_async(sbi, bdev,
1814                                                 start, len);
1815                                 if (err)
1816                                         return err;
1817                                 bdev = bdev2;
1818                                 start = i;
1819                                 len = 0;
1820                         }
1821                 }
1822
1823                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1824                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1825
1826                 if (f2fs_block_unit_discard(sbi) &&
1827                                 !f2fs_test_and_set_bit(offset, se->discard_map))
1828                         sbi->discard_blks--;
1829         }
1830
1831         if (len)
1832                 err = __issue_discard_async(sbi, bdev, start, len);
1833         return err;
1834 }
1835
1836 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1837                                                         bool check_only)
1838 {
1839         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1840         int max_blocks = sbi->blocks_per_seg;
1841         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1842         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1843         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1844         unsigned long *discard_map = (unsigned long *)se->discard_map;
1845         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1846         unsigned int start = 0, end = -1;
1847         bool force = (cpc->reason & CP_DISCARD);
1848         struct discard_entry *de = NULL;
1849         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1850         int i;
1851
1852         if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi) ||
1853                         !f2fs_block_unit_discard(sbi))
1854                 return false;
1855
1856         if (!force) {
1857                 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1858                         SM_I(sbi)->dcc_info->nr_discards >=
1859                                 SM_I(sbi)->dcc_info->max_discards)
1860                         return false;
1861         }
1862
1863         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1864         for (i = 0; i < entries; i++)
1865                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1866                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1867
1868         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1869                                 SM_I(sbi)->dcc_info->max_discards) {
1870                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1871                 if (start >= max_blocks)
1872                         break;
1873
1874                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1875                 if (force && start && end != max_blocks
1876                                         && (end - start) < cpc->trim_minlen)
1877                         continue;
1878
1879                 if (check_only)
1880                         return true;
1881
1882                 if (!de) {
1883                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1884                                                 GFP_F2FS_ZERO, true, NULL);
1885                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1886                         list_add_tail(&de->list, head);
1887                 }
1888
1889                 for (i = start; i < end; i++)
1890                         __set_bit_le(i, (void *)de->discard_map);
1891
1892                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1893         }
1894         return false;
1895 }
1896
1897 static void release_discard_addr(struct discard_entry *entry)
1898 {
1899         list_del(&entry->list);
1900         kmem_cache_free(discard_entry_slab, entry);
1901 }
1902
1903 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1904 {
1905         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1906         struct discard_entry *entry, *this;
1907
1908         /* drop caches */
1909         list_for_each_entry_safe(entry, this, head, list)
1910                 release_discard_addr(entry);
1911 }
1912
1913 /*
1914  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1915  */
1916 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1917 {
1918         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1919         unsigned int segno;
1920
1921         mutex_lock(&dirty_i->seglist_lock);
1922         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1923                 __set_test_and_free(sbi, segno, false);
1924         mutex_unlock(&dirty_i->seglist_lock);
1925 }
1926
1927 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1928                                                 struct cp_control *cpc)
1929 {
1930         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1931         struct list_head *head = &dcc->entry_list;
1932         struct discard_entry *entry, *this;
1933         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1934         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1935         unsigned int start = 0, end = -1;
1936         unsigned int secno, start_segno;
1937         bool force = (cpc->reason & CP_DISCARD);
1938         bool section_alignment = F2FS_OPTION(sbi).discard_unit ==
1939                                                 DISCARD_UNIT_SECTION;
1940
1941         if (f2fs_lfs_mode(sbi) && __is_large_section(sbi))
1942                 section_alignment = true;
1943
1944         mutex_lock(&dirty_i->seglist_lock);
1945
1946         while (1) {
1947                 int i;
1948
1949                 if (section_alignment && end != -1)
1950                         end--;
1951                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1952                 if (start >= MAIN_SEGS(sbi))
1953                         break;
1954                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1955                                                                 start + 1);
1956
1957                 if (section_alignment) {
1958                         start = rounddown(start, sbi->segs_per_sec);
1959                         end = roundup(end, sbi->segs_per_sec);
1960                 }
1961
1962                 for (i = start; i < end; i++) {
1963                         if (test_and_clear_bit(i, prefree_map))
1964                                 dirty_i->nr_dirty[PRE]--;
1965                 }
1966
1967                 if (!f2fs_realtime_discard_enable(sbi))
1968                         continue;
1969
1970                 if (force && start >= cpc->trim_start &&
1971                                         (end - 1) <= cpc->trim_end)
1972                                 continue;
1973
1974                 if (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi)) {
1975                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1976                                 (end - start) << sbi->log_blocks_per_seg);
1977                         continue;
1978                 }
1979 next:
1980                 secno = GET_SEC_FROM_SEG(sbi, start);
1981                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1982                 if (!IS_CURSEC(sbi, secno) &&
1983                         !get_valid_blocks(sbi, start, true))
1984                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1985                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1986
1987                 start = start_segno + sbi->segs_per_sec;
1988                 if (start < end)
1989                         goto next;
1990                 else
1991                         end = start - 1;
1992         }
1993         mutex_unlock(&dirty_i->seglist_lock);
1994
1995         if (!f2fs_block_unit_discard(sbi))
1996                 goto wakeup;
1997
1998         /* send small discards */
1999         list_for_each_entry_safe(entry, this, head, list) {
2000                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2001                 bool is_valid = test_bit_le(0, entry->discard_map);
2002
2003 find_next:
2004                 if (is_valid) {
2005                         next_pos = find_next_zero_bit_le(entry->discard_map,
2006                                         sbi->blocks_per_seg, cur_pos);
2007                         len = next_pos - cur_pos;
2008
2009                         if (f2fs_sb_has_blkzoned(sbi) ||
2010                             (force && len < cpc->trim_minlen))
2011                                 goto skip;
2012
2013                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2014                                                                         len);
2015                         total_len += len;
2016                 } else {
2017                         next_pos = find_next_bit_le(entry->discard_map,
2018                                         sbi->blocks_per_seg, cur_pos);
2019                 }
2020 skip:
2021                 cur_pos = next_pos;
2022                 is_valid = !is_valid;
2023
2024                 if (cur_pos < sbi->blocks_per_seg)
2025                         goto find_next;
2026
2027                 release_discard_addr(entry);
2028                 dcc->nr_discards -= total_len;
2029         }
2030
2031 wakeup:
2032         wake_up_discard_thread(sbi, false);
2033 }
2034
2035 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi)
2036 {
2037         dev_t dev = sbi->sb->s_bdev->bd_dev;
2038         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2039         int err = 0;
2040
2041         if (!f2fs_realtime_discard_enable(sbi))
2042                 return 0;
2043
2044         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2045                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2046         if (IS_ERR(dcc->f2fs_issue_discard)) {
2047                 err = PTR_ERR(dcc->f2fs_issue_discard);
2048                 dcc->f2fs_issue_discard = NULL;
2049         }
2050
2051         return err;
2052 }
2053
2054 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2055 {
2056         struct discard_cmd_control *dcc;
2057         int err = 0, i;
2058
2059         if (SM_I(sbi)->dcc_info) {
2060                 dcc = SM_I(sbi)->dcc_info;
2061                 goto init_thread;
2062         }
2063
2064         dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2065         if (!dcc)
2066                 return -ENOMEM;
2067
2068         dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2069         dcc->max_ordered_discard = DEFAULT_MAX_ORDERED_DISCARD_GRANULARITY;
2070         if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT)
2071                 dcc->discard_granularity = sbi->blocks_per_seg;
2072         else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION)
2073                 dcc->discard_granularity = BLKS_PER_SEC(sbi);
2074
2075         INIT_LIST_HEAD(&dcc->entry_list);
2076         for (i = 0; i < MAX_PLIST_NUM; i++)
2077                 INIT_LIST_HEAD(&dcc->pend_list[i]);
2078         INIT_LIST_HEAD(&dcc->wait_list);
2079         INIT_LIST_HEAD(&dcc->fstrim_list);
2080         mutex_init(&dcc->cmd_lock);
2081         atomic_set(&dcc->issued_discard, 0);
2082         atomic_set(&dcc->queued_discard, 0);
2083         atomic_set(&dcc->discard_cmd_cnt, 0);
2084         dcc->nr_discards = 0;
2085         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2086         dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST;
2087         dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME;
2088         dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME;
2089         dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME;
2090         dcc->discard_urgent_util = DEF_DISCARD_URGENT_UTIL;
2091         dcc->undiscard_blks = 0;
2092         dcc->next_pos = 0;
2093         dcc->root = RB_ROOT_CACHED;
2094         dcc->rbtree_check = false;
2095
2096         init_waitqueue_head(&dcc->discard_wait_queue);
2097         SM_I(sbi)->dcc_info = dcc;
2098 init_thread:
2099         err = f2fs_start_discard_thread(sbi);
2100         if (err) {
2101                 kfree(dcc);
2102                 SM_I(sbi)->dcc_info = NULL;
2103         }
2104
2105         return err;
2106 }
2107
2108 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2109 {
2110         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2111
2112         if (!dcc)
2113                 return;
2114
2115         f2fs_stop_discard_thread(sbi);
2116
2117         /*
2118          * Recovery can cache discard commands, so in error path of
2119          * fill_super(), it needs to give a chance to handle them.
2120          */
2121         f2fs_issue_discard_timeout(sbi);
2122
2123         kfree(dcc);
2124         SM_I(sbi)->dcc_info = NULL;
2125 }
2126
2127 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2128 {
2129         struct sit_info *sit_i = SIT_I(sbi);
2130
2131         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2132                 sit_i->dirty_sentries++;
2133                 return false;
2134         }
2135
2136         return true;
2137 }
2138
2139 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2140                                         unsigned int segno, int modified)
2141 {
2142         struct seg_entry *se = get_seg_entry(sbi, segno);
2143
2144         se->type = type;
2145         if (modified)
2146                 __mark_sit_entry_dirty(sbi, segno);
2147 }
2148
2149 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2150                                                                 block_t blkaddr)
2151 {
2152         unsigned int segno = GET_SEGNO(sbi, blkaddr);
2153
2154         if (segno == NULL_SEGNO)
2155                 return 0;
2156         return get_seg_entry(sbi, segno)->mtime;
2157 }
2158
2159 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2160                                                 unsigned long long old_mtime)
2161 {
2162         struct seg_entry *se;
2163         unsigned int segno = GET_SEGNO(sbi, blkaddr);
2164         unsigned long long ctime = get_mtime(sbi, false);
2165         unsigned long long mtime = old_mtime ? old_mtime : ctime;
2166
2167         if (segno == NULL_SEGNO)
2168                 return;
2169
2170         se = get_seg_entry(sbi, segno);
2171
2172         if (!se->mtime)
2173                 se->mtime = mtime;
2174         else
2175                 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2176                                                 se->valid_blocks + 1);
2177
2178         if (ctime > SIT_I(sbi)->max_mtime)
2179                 SIT_I(sbi)->max_mtime = ctime;
2180 }
2181
2182 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2183 {
2184         struct seg_entry *se;
2185         unsigned int segno, offset;
2186         long int new_vblocks;
2187         bool exist;
2188 #ifdef CONFIG_F2FS_CHECK_FS
2189         bool mir_exist;
2190 #endif
2191
2192         segno = GET_SEGNO(sbi, blkaddr);
2193
2194         se = get_seg_entry(sbi, segno);
2195         new_vblocks = se->valid_blocks + del;
2196         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2197
2198         f2fs_bug_on(sbi, (new_vblocks < 0 ||
2199                         (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2200
2201         se->valid_blocks = new_vblocks;
2202
2203         /* Update valid block bitmap */
2204         if (del > 0) {
2205                 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2206 #ifdef CONFIG_F2FS_CHECK_FS
2207                 mir_exist = f2fs_test_and_set_bit(offset,
2208                                                 se->cur_valid_map_mir);
2209                 if (unlikely(exist != mir_exist)) {
2210                         f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2211                                  blkaddr, exist);
2212                         f2fs_bug_on(sbi, 1);
2213                 }
2214 #endif
2215                 if (unlikely(exist)) {
2216                         f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2217                                  blkaddr);
2218                         f2fs_bug_on(sbi, 1);
2219                         se->valid_blocks--;
2220                         del = 0;
2221                 }
2222
2223                 if (f2fs_block_unit_discard(sbi) &&
2224                                 !f2fs_test_and_set_bit(offset, se->discard_map))
2225                         sbi->discard_blks--;
2226
2227                 /*
2228                  * SSR should never reuse block which is checkpointed
2229                  * or newly invalidated.
2230                  */
2231                 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2232                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2233                                 se->ckpt_valid_blocks++;
2234                 }
2235         } else {
2236                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2237 #ifdef CONFIG_F2FS_CHECK_FS
2238                 mir_exist = f2fs_test_and_clear_bit(offset,
2239                                                 se->cur_valid_map_mir);
2240                 if (unlikely(exist != mir_exist)) {
2241                         f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2242                                  blkaddr, exist);
2243                         f2fs_bug_on(sbi, 1);
2244                 }
2245 #endif
2246                 if (unlikely(!exist)) {
2247                         f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2248                                  blkaddr);
2249                         f2fs_bug_on(sbi, 1);
2250                         se->valid_blocks++;
2251                         del = 0;
2252                 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2253                         /*
2254                          * If checkpoints are off, we must not reuse data that
2255                          * was used in the previous checkpoint. If it was used
2256                          * before, we must track that to know how much space we
2257                          * really have.
2258                          */
2259                         if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2260                                 spin_lock(&sbi->stat_lock);
2261                                 sbi->unusable_block_count++;
2262                                 spin_unlock(&sbi->stat_lock);
2263                         }
2264                 }
2265
2266                 if (f2fs_block_unit_discard(sbi) &&
2267                         f2fs_test_and_clear_bit(offset, se->discard_map))
2268                         sbi->discard_blks++;
2269         }
2270         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2271                 se->ckpt_valid_blocks += del;
2272
2273         __mark_sit_entry_dirty(sbi, segno);
2274
2275         /* update total number of valid blocks to be written in ckpt area */
2276         SIT_I(sbi)->written_valid_blocks += del;
2277
2278         if (__is_large_section(sbi))
2279                 get_sec_entry(sbi, segno)->valid_blocks += del;
2280 }
2281
2282 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2283 {
2284         unsigned int segno = GET_SEGNO(sbi, addr);
2285         struct sit_info *sit_i = SIT_I(sbi);
2286
2287         f2fs_bug_on(sbi, addr == NULL_ADDR);
2288         if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2289                 return;
2290
2291         invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2292         f2fs_invalidate_compress_page(sbi, addr);
2293
2294         /* add it into sit main buffer */
2295         down_write(&sit_i->sentry_lock);
2296
2297         update_segment_mtime(sbi, addr, 0);
2298         update_sit_entry(sbi, addr, -1);
2299
2300         /* add it into dirty seglist */
2301         locate_dirty_segment(sbi, segno);
2302
2303         up_write(&sit_i->sentry_lock);
2304 }
2305
2306 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2307 {
2308         struct sit_info *sit_i = SIT_I(sbi);
2309         unsigned int segno, offset;
2310         struct seg_entry *se;
2311         bool is_cp = false;
2312
2313         if (!__is_valid_data_blkaddr(blkaddr))
2314                 return true;
2315
2316         down_read(&sit_i->sentry_lock);
2317
2318         segno = GET_SEGNO(sbi, blkaddr);
2319         se = get_seg_entry(sbi, segno);
2320         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2321
2322         if (f2fs_test_bit(offset, se->ckpt_valid_map))
2323                 is_cp = true;
2324
2325         up_read(&sit_i->sentry_lock);
2326
2327         return is_cp;
2328 }
2329
2330 /*
2331  * This function should be resided under the curseg_mutex lock
2332  */
2333 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2334                                         struct f2fs_summary *sum)
2335 {
2336         struct curseg_info *curseg = CURSEG_I(sbi, type);
2337         void *addr = curseg->sum_blk;
2338
2339         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2340         memcpy(addr, sum, sizeof(struct f2fs_summary));
2341 }
2342
2343 /*
2344  * Calculate the number of current summary pages for writing
2345  */
2346 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2347 {
2348         int valid_sum_count = 0;
2349         int i, sum_in_page;
2350
2351         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2352                 if (sbi->ckpt->alloc_type[i] == SSR)
2353                         valid_sum_count += sbi->blocks_per_seg;
2354                 else {
2355                         if (for_ra)
2356                                 valid_sum_count += le16_to_cpu(
2357                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2358                         else
2359                                 valid_sum_count += curseg_blkoff(sbi, i);
2360                 }
2361         }
2362
2363         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2364                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2365         if (valid_sum_count <= sum_in_page)
2366                 return 1;
2367         else if ((valid_sum_count - sum_in_page) <=
2368                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2369                 return 2;
2370         return 3;
2371 }
2372
2373 /*
2374  * Caller should put this summary page
2375  */
2376 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2377 {
2378         if (unlikely(f2fs_cp_error(sbi)))
2379                 return ERR_PTR(-EIO);
2380         return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno));
2381 }
2382
2383 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2384                                         void *src, block_t blk_addr)
2385 {
2386         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2387
2388         memcpy(page_address(page), src, PAGE_SIZE);
2389         set_page_dirty(page);
2390         f2fs_put_page(page, 1);
2391 }
2392
2393 static void write_sum_page(struct f2fs_sb_info *sbi,
2394                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
2395 {
2396         f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2397 }
2398
2399 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2400                                                 int type, block_t blk_addr)
2401 {
2402         struct curseg_info *curseg = CURSEG_I(sbi, type);
2403         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2404         struct f2fs_summary_block *src = curseg->sum_blk;
2405         struct f2fs_summary_block *dst;
2406
2407         dst = (struct f2fs_summary_block *)page_address(page);
2408         memset(dst, 0, PAGE_SIZE);
2409
2410         mutex_lock(&curseg->curseg_mutex);
2411
2412         down_read(&curseg->journal_rwsem);
2413         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2414         up_read(&curseg->journal_rwsem);
2415
2416         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2417         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2418
2419         mutex_unlock(&curseg->curseg_mutex);
2420
2421         set_page_dirty(page);
2422         f2fs_put_page(page, 1);
2423 }
2424
2425 static int is_next_segment_free(struct f2fs_sb_info *sbi,
2426                                 struct curseg_info *curseg, int type)
2427 {
2428         unsigned int segno = curseg->segno + 1;
2429         struct free_segmap_info *free_i = FREE_I(sbi);
2430
2431         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2432                 return !test_bit(segno, free_i->free_segmap);
2433         return 0;
2434 }
2435
2436 /*
2437  * Find a new segment from the free segments bitmap to right order
2438  * This function should be returned with success, otherwise BUG
2439  */
2440 static void get_new_segment(struct f2fs_sb_info *sbi,
2441                         unsigned int *newseg, bool new_sec, int dir)
2442 {
2443         struct free_segmap_info *free_i = FREE_I(sbi);
2444         unsigned int segno, secno, zoneno;
2445         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2446         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2447         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2448         unsigned int left_start = hint;
2449         bool init = true;
2450         int go_left = 0;
2451         int i;
2452
2453         spin_lock(&free_i->segmap_lock);
2454
2455         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2456                 segno = find_next_zero_bit(free_i->free_segmap,
2457                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2458                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2459                         goto got_it;
2460         }
2461 find_other_zone:
2462         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2463         if (secno >= MAIN_SECS(sbi)) {
2464                 if (dir == ALLOC_RIGHT) {
2465                         secno = find_first_zero_bit(free_i->free_secmap,
2466                                                         MAIN_SECS(sbi));
2467                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2468                 } else {
2469                         go_left = 1;
2470                         left_start = hint - 1;
2471                 }
2472         }
2473         if (go_left == 0)
2474                 goto skip_left;
2475
2476         while (test_bit(left_start, free_i->free_secmap)) {
2477                 if (left_start > 0) {
2478                         left_start--;
2479                         continue;
2480                 }
2481                 left_start = find_first_zero_bit(free_i->free_secmap,
2482                                                         MAIN_SECS(sbi));
2483                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2484                 break;
2485         }
2486         secno = left_start;
2487 skip_left:
2488         segno = GET_SEG_FROM_SEC(sbi, secno);
2489         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2490
2491         /* give up on finding another zone */
2492         if (!init)
2493                 goto got_it;
2494         if (sbi->secs_per_zone == 1)
2495                 goto got_it;
2496         if (zoneno == old_zoneno)
2497                 goto got_it;
2498         if (dir == ALLOC_LEFT) {
2499                 if (!go_left && zoneno + 1 >= total_zones)
2500                         goto got_it;
2501                 if (go_left && zoneno == 0)
2502                         goto got_it;
2503         }
2504         for (i = 0; i < NR_CURSEG_TYPE; i++)
2505                 if (CURSEG_I(sbi, i)->zone == zoneno)
2506                         break;
2507
2508         if (i < NR_CURSEG_TYPE) {
2509                 /* zone is in user, try another */
2510                 if (go_left)
2511                         hint = zoneno * sbi->secs_per_zone - 1;
2512                 else if (zoneno + 1 >= total_zones)
2513                         hint = 0;
2514                 else
2515                         hint = (zoneno + 1) * sbi->secs_per_zone;
2516                 init = false;
2517                 goto find_other_zone;
2518         }
2519 got_it:
2520         /* set it as dirty segment in free segmap */
2521         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2522         __set_inuse(sbi, segno);
2523         *newseg = segno;
2524         spin_unlock(&free_i->segmap_lock);
2525 }
2526
2527 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2528 {
2529         struct curseg_info *curseg = CURSEG_I(sbi, type);
2530         struct summary_footer *sum_footer;
2531         unsigned short seg_type = curseg->seg_type;
2532
2533         curseg->inited = true;
2534         curseg->segno = curseg->next_segno;
2535         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2536         curseg->next_blkoff = 0;
2537         curseg->next_segno = NULL_SEGNO;
2538
2539         sum_footer = &(curseg->sum_blk->footer);
2540         memset(sum_footer, 0, sizeof(struct summary_footer));
2541
2542         sanity_check_seg_type(sbi, seg_type);
2543
2544         if (IS_DATASEG(seg_type))
2545                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2546         if (IS_NODESEG(seg_type))
2547                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2548         __set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
2549 }
2550
2551 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2552 {
2553         struct curseg_info *curseg = CURSEG_I(sbi, type);
2554         unsigned short seg_type = curseg->seg_type;
2555
2556         sanity_check_seg_type(sbi, seg_type);
2557         if (f2fs_need_rand_seg(sbi))
2558                 return get_random_u32_below(MAIN_SECS(sbi) * sbi->segs_per_sec);
2559
2560         /* if segs_per_sec is large than 1, we need to keep original policy. */
2561         if (__is_large_section(sbi))
2562                 return curseg->segno;
2563
2564         /* inmem log may not locate on any segment after mount */
2565         if (!curseg->inited)
2566                 return 0;
2567
2568         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2569                 return 0;
2570
2571         if (test_opt(sbi, NOHEAP) &&
2572                 (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type)))
2573                 return 0;
2574
2575         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2576                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2577
2578         /* find segments from 0 to reuse freed segments */
2579         if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2580                 return 0;
2581
2582         return curseg->segno;
2583 }
2584
2585 /*
2586  * Allocate a current working segment.
2587  * This function always allocates a free segment in LFS manner.
2588  */
2589 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2590 {
2591         struct curseg_info *curseg = CURSEG_I(sbi, type);
2592         unsigned short seg_type = curseg->seg_type;
2593         unsigned int segno = curseg->segno;
2594         int dir = ALLOC_LEFT;
2595
2596         if (curseg->inited)
2597                 write_sum_page(sbi, curseg->sum_blk,
2598                                 GET_SUM_BLOCK(sbi, segno));
2599         if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA)
2600                 dir = ALLOC_RIGHT;
2601
2602         if (test_opt(sbi, NOHEAP))
2603                 dir = ALLOC_RIGHT;
2604
2605         segno = __get_next_segno(sbi, type);
2606         get_new_segment(sbi, &segno, new_sec, dir);
2607         curseg->next_segno = segno;
2608         reset_curseg(sbi, type, 1);
2609         curseg->alloc_type = LFS;
2610         if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
2611                 curseg->fragment_remained_chunk =
2612                                 get_random_u32_inclusive(1, sbi->max_fragment_chunk);
2613 }
2614
2615 static int __next_free_blkoff(struct f2fs_sb_info *sbi,
2616                                         int segno, block_t start)
2617 {
2618         struct seg_entry *se = get_seg_entry(sbi, segno);
2619         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2620         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2621         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2622         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2623         int i;
2624
2625         for (i = 0; i < entries; i++)
2626                 target_map[i] = ckpt_map[i] | cur_map[i];
2627
2628         return __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2629 }
2630
2631 /*
2632  * If a segment is written by LFS manner, next block offset is just obtained
2633  * by increasing the current block offset. However, if a segment is written by
2634  * SSR manner, next block offset obtained by calling __next_free_blkoff
2635  */
2636 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2637                                 struct curseg_info *seg)
2638 {
2639         if (seg->alloc_type == SSR) {
2640                 seg->next_blkoff =
2641                         __next_free_blkoff(sbi, seg->segno,
2642                                                 seg->next_blkoff + 1);
2643         } else {
2644                 seg->next_blkoff++;
2645                 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) {
2646                         /* To allocate block chunks in different sizes, use random number */
2647                         if (--seg->fragment_remained_chunk <= 0) {
2648                                 seg->fragment_remained_chunk =
2649                                    get_random_u32_inclusive(1, sbi->max_fragment_chunk);
2650                                 seg->next_blkoff +=
2651                                    get_random_u32_inclusive(1, sbi->max_fragment_hole);
2652                         }
2653                 }
2654         }
2655 }
2656
2657 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno)
2658 {
2659         return __next_free_blkoff(sbi, segno, 0) < sbi->blocks_per_seg;
2660 }
2661
2662 /*
2663  * This function always allocates a used segment(from dirty seglist) by SSR
2664  * manner, so it should recover the existing segment information of valid blocks
2665  */
2666 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2667 {
2668         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2669         struct curseg_info *curseg = CURSEG_I(sbi, type);
2670         unsigned int new_segno = curseg->next_segno;
2671         struct f2fs_summary_block *sum_node;
2672         struct page *sum_page;
2673
2674         write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, curseg->segno));
2675
2676         __set_test_and_inuse(sbi, new_segno);
2677
2678         mutex_lock(&dirty_i->seglist_lock);
2679         __remove_dirty_segment(sbi, new_segno, PRE);
2680         __remove_dirty_segment(sbi, new_segno, DIRTY);
2681         mutex_unlock(&dirty_i->seglist_lock);
2682
2683         reset_curseg(sbi, type, 1);
2684         curseg->alloc_type = SSR;
2685         curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0);
2686
2687         sum_page = f2fs_get_sum_page(sbi, new_segno);
2688         if (IS_ERR(sum_page)) {
2689                 /* GC won't be able to use stale summary pages by cp_error */
2690                 memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
2691                 return;
2692         }
2693         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2694         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2695         f2fs_put_page(sum_page, 1);
2696 }
2697
2698 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2699                                 int alloc_mode, unsigned long long age);
2700
2701 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type,
2702                                         int target_type, int alloc_mode,
2703                                         unsigned long long age)
2704 {
2705         struct curseg_info *curseg = CURSEG_I(sbi, type);
2706
2707         curseg->seg_type = target_type;
2708
2709         if (get_ssr_segment(sbi, type, alloc_mode, age)) {
2710                 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
2711
2712                 curseg->seg_type = se->type;
2713                 change_curseg(sbi, type);
2714         } else {
2715                 /* allocate cold segment by default */
2716                 curseg->seg_type = CURSEG_COLD_DATA;
2717                 new_curseg(sbi, type, true);
2718         }
2719         stat_inc_seg_type(sbi, curseg);
2720 }
2721
2722 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
2723 {
2724         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
2725
2726         if (!sbi->am.atgc_enabled)
2727                 return;
2728
2729         f2fs_down_read(&SM_I(sbi)->curseg_lock);
2730
2731         mutex_lock(&curseg->curseg_mutex);
2732         down_write(&SIT_I(sbi)->sentry_lock);
2733
2734         get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0);
2735
2736         up_write(&SIT_I(sbi)->sentry_lock);
2737         mutex_unlock(&curseg->curseg_mutex);
2738
2739         f2fs_up_read(&SM_I(sbi)->curseg_lock);
2740
2741 }
2742 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
2743 {
2744         __f2fs_init_atgc_curseg(sbi);
2745 }
2746
2747 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2748 {
2749         struct curseg_info *curseg = CURSEG_I(sbi, type);
2750
2751         mutex_lock(&curseg->curseg_mutex);
2752         if (!curseg->inited)
2753                 goto out;
2754
2755         if (get_valid_blocks(sbi, curseg->segno, false)) {
2756                 write_sum_page(sbi, curseg->sum_blk,
2757                                 GET_SUM_BLOCK(sbi, curseg->segno));
2758         } else {
2759                 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2760                 __set_test_and_free(sbi, curseg->segno, true);
2761                 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2762         }
2763 out:
2764         mutex_unlock(&curseg->curseg_mutex);
2765 }
2766
2767 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
2768 {
2769         __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2770
2771         if (sbi->am.atgc_enabled)
2772                 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2773 }
2774
2775 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2776 {
2777         struct curseg_info *curseg = CURSEG_I(sbi, type);
2778
2779         mutex_lock(&curseg->curseg_mutex);
2780         if (!curseg->inited)
2781                 goto out;
2782         if (get_valid_blocks(sbi, curseg->segno, false))
2783                 goto out;
2784
2785         mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2786         __set_test_and_inuse(sbi, curseg->segno);
2787         mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2788 out:
2789         mutex_unlock(&curseg->curseg_mutex);
2790 }
2791
2792 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
2793 {
2794         __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2795
2796         if (sbi->am.atgc_enabled)
2797                 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2798 }
2799
2800 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2801                                 int alloc_mode, unsigned long long age)
2802 {
2803         struct curseg_info *curseg = CURSEG_I(sbi, type);
2804         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2805         unsigned segno = NULL_SEGNO;
2806         unsigned short seg_type = curseg->seg_type;
2807         int i, cnt;
2808         bool reversed = false;
2809
2810         sanity_check_seg_type(sbi, seg_type);
2811
2812         /* f2fs_need_SSR() already forces to do this */
2813         if (!v_ops->get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) {
2814                 curseg->next_segno = segno;
2815                 return 1;
2816         }
2817
2818         /* For node segments, let's do SSR more intensively */
2819         if (IS_NODESEG(seg_type)) {
2820                 if (seg_type >= CURSEG_WARM_NODE) {
2821                         reversed = true;
2822                         i = CURSEG_COLD_NODE;
2823                 } else {
2824                         i = CURSEG_HOT_NODE;
2825                 }
2826                 cnt = NR_CURSEG_NODE_TYPE;
2827         } else {
2828                 if (seg_type >= CURSEG_WARM_DATA) {
2829                         reversed = true;
2830                         i = CURSEG_COLD_DATA;
2831                 } else {
2832                         i = CURSEG_HOT_DATA;
2833                 }
2834                 cnt = NR_CURSEG_DATA_TYPE;
2835         }
2836
2837         for (; cnt-- > 0; reversed ? i-- : i++) {
2838                 if (i == seg_type)
2839                         continue;
2840                 if (!v_ops->get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) {
2841                         curseg->next_segno = segno;
2842                         return 1;
2843                 }
2844         }
2845
2846         /* find valid_blocks=0 in dirty list */
2847         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2848                 segno = get_free_segment(sbi);
2849                 if (segno != NULL_SEGNO) {
2850                         curseg->next_segno = segno;
2851                         return 1;
2852                 }
2853         }
2854         return 0;
2855 }
2856
2857 static bool need_new_seg(struct f2fs_sb_info *sbi, int type)
2858 {
2859         struct curseg_info *curseg = CURSEG_I(sbi, type);
2860
2861         if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2862             curseg->seg_type == CURSEG_WARM_NODE)
2863                 return true;
2864         if (curseg->alloc_type == LFS &&
2865             is_next_segment_free(sbi, curseg, type) &&
2866             likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2867                 return true;
2868         if (!f2fs_need_SSR(sbi) || !get_ssr_segment(sbi, type, SSR, 0))
2869                 return true;
2870         return false;
2871 }
2872
2873 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2874                                         unsigned int start, unsigned int end)
2875 {
2876         struct curseg_info *curseg = CURSEG_I(sbi, type);
2877         unsigned int segno;
2878
2879         f2fs_down_read(&SM_I(sbi)->curseg_lock);
2880         mutex_lock(&curseg->curseg_mutex);
2881         down_write(&SIT_I(sbi)->sentry_lock);
2882
2883         segno = CURSEG_I(sbi, type)->segno;
2884         if (segno < start || segno > end)
2885                 goto unlock;
2886
2887         if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
2888                 change_curseg(sbi, type);
2889         else
2890                 new_curseg(sbi, type, true);
2891
2892         stat_inc_seg_type(sbi, curseg);
2893
2894         locate_dirty_segment(sbi, segno);
2895 unlock:
2896         up_write(&SIT_I(sbi)->sentry_lock);
2897
2898         if (segno != curseg->segno)
2899                 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2900                             type, segno, curseg->segno);
2901
2902         mutex_unlock(&curseg->curseg_mutex);
2903         f2fs_up_read(&SM_I(sbi)->curseg_lock);
2904 }
2905
2906 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type,
2907                                                 bool new_sec, bool force)
2908 {
2909         struct curseg_info *curseg = CURSEG_I(sbi, type);
2910         unsigned int old_segno;
2911
2912         if (!curseg->inited)
2913                 goto alloc;
2914
2915         if (force || curseg->next_blkoff ||
2916                 get_valid_blocks(sbi, curseg->segno, new_sec))
2917                 goto alloc;
2918
2919         if (!get_ckpt_valid_blocks(sbi, curseg->segno, new_sec))
2920                 return;
2921 alloc:
2922         old_segno = curseg->segno;
2923         new_curseg(sbi, type, true);
2924         stat_inc_seg_type(sbi, curseg);
2925         locate_dirty_segment(sbi, old_segno);
2926 }
2927
2928 static void __allocate_new_section(struct f2fs_sb_info *sbi,
2929                                                 int type, bool force)
2930 {
2931         __allocate_new_segment(sbi, type, true, force);
2932 }
2933
2934 void f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force)
2935 {
2936         f2fs_down_read(&SM_I(sbi)->curseg_lock);
2937         down_write(&SIT_I(sbi)->sentry_lock);
2938         __allocate_new_section(sbi, type, force);
2939         up_write(&SIT_I(sbi)->sentry_lock);
2940         f2fs_up_read(&SM_I(sbi)->curseg_lock);
2941 }
2942
2943 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2944 {
2945         int i;
2946
2947         f2fs_down_read(&SM_I(sbi)->curseg_lock);
2948         down_write(&SIT_I(sbi)->sentry_lock);
2949         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
2950                 __allocate_new_segment(sbi, i, false, false);
2951         up_write(&SIT_I(sbi)->sentry_lock);
2952         f2fs_up_read(&SM_I(sbi)->curseg_lock);
2953 }
2954
2955 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2956                                                 struct cp_control *cpc)
2957 {
2958         __u64 trim_start = cpc->trim_start;
2959         bool has_candidate = false;
2960
2961         down_write(&SIT_I(sbi)->sentry_lock);
2962         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2963                 if (add_discard_addrs(sbi, cpc, true)) {
2964                         has_candidate = true;
2965                         break;
2966                 }
2967         }
2968         up_write(&SIT_I(sbi)->sentry_lock);
2969
2970         cpc->trim_start = trim_start;
2971         return has_candidate;
2972 }
2973
2974 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2975                                         struct discard_policy *dpolicy,
2976                                         unsigned int start, unsigned int end)
2977 {
2978         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2979         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2980         struct rb_node **insert_p = NULL, *insert_parent = NULL;
2981         struct discard_cmd *dc;
2982         struct blk_plug plug;
2983         int issued;
2984         unsigned int trimmed = 0;
2985
2986 next:
2987         issued = 0;
2988
2989         mutex_lock(&dcc->cmd_lock);
2990         if (unlikely(dcc->rbtree_check))
2991                 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2992                                                         &dcc->root, false));
2993
2994         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2995                                         NULL, start,
2996                                         (struct rb_entry **)&prev_dc,
2997                                         (struct rb_entry **)&next_dc,
2998                                         &insert_p, &insert_parent, true, NULL);
2999         if (!dc)
3000                 dc = next_dc;
3001
3002         blk_start_plug(&plug);
3003
3004         while (dc && dc->lstart <= end) {
3005                 struct rb_node *node;
3006                 int err = 0;
3007
3008                 if (dc->len < dpolicy->granularity)
3009                         goto skip;
3010
3011                 if (dc->state != D_PREP) {
3012                         list_move_tail(&dc->list, &dcc->fstrim_list);
3013                         goto skip;
3014                 }
3015
3016                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
3017
3018                 if (issued >= dpolicy->max_requests) {
3019                         start = dc->lstart + dc->len;
3020
3021                         if (err)
3022                                 __remove_discard_cmd(sbi, dc);
3023
3024                         blk_finish_plug(&plug);
3025                         mutex_unlock(&dcc->cmd_lock);
3026                         trimmed += __wait_all_discard_cmd(sbi, NULL);
3027                         f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
3028                         goto next;
3029                 }
3030 skip:
3031                 node = rb_next(&dc->rb_node);
3032                 if (err)
3033                         __remove_discard_cmd(sbi, dc);
3034                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3035
3036                 if (fatal_signal_pending(current))
3037                         break;
3038         }
3039
3040         blk_finish_plug(&plug);
3041         mutex_unlock(&dcc->cmd_lock);
3042
3043         return trimmed;
3044 }
3045
3046 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3047 {
3048         __u64 start = F2FS_BYTES_TO_BLK(range->start);
3049         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
3050         unsigned int start_segno, end_segno;
3051         block_t start_block, end_block;
3052         struct cp_control cpc;
3053         struct discard_policy dpolicy;
3054         unsigned long long trimmed = 0;
3055         int err = 0;
3056         bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
3057
3058         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
3059                 return -EINVAL;
3060
3061         if (end < MAIN_BLKADDR(sbi))
3062                 goto out;
3063
3064         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
3065                 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
3066                 return -EFSCORRUPTED;
3067         }
3068
3069         /* start/end segment number in main_area */
3070         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3071         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3072                                                 GET_SEGNO(sbi, end);
3073         if (need_align) {
3074                 start_segno = rounddown(start_segno, sbi->segs_per_sec);
3075                 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
3076         }
3077
3078         cpc.reason = CP_DISCARD;
3079         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
3080         cpc.trim_start = start_segno;
3081         cpc.trim_end = end_segno;
3082
3083         if (sbi->discard_blks == 0)
3084                 goto out;
3085
3086         f2fs_down_write(&sbi->gc_lock);
3087         err = f2fs_write_checkpoint(sbi, &cpc);
3088         f2fs_up_write(&sbi->gc_lock);
3089         if (err)
3090                 goto out;
3091
3092         /*
3093          * We filed discard candidates, but actually we don't need to wait for
3094          * all of them, since they'll be issued in idle time along with runtime
3095          * discard option. User configuration looks like using runtime discard
3096          * or periodic fstrim instead of it.
3097          */
3098         if (f2fs_realtime_discard_enable(sbi))
3099                 goto out;
3100
3101         start_block = START_BLOCK(sbi, start_segno);
3102         end_block = START_BLOCK(sbi, end_segno + 1);
3103
3104         __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3105         trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3106                                         start_block, end_block);
3107
3108         trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3109                                         start_block, end_block);
3110 out:
3111         if (!err)
3112                 range->len = F2FS_BLK_TO_BYTES(trimmed);
3113         return err;
3114 }
3115
3116 static bool __has_curseg_space(struct f2fs_sb_info *sbi,
3117                                         struct curseg_info *curseg)
3118 {
3119         return curseg->next_blkoff < f2fs_usable_blks_in_seg(sbi,
3120                                                         curseg->segno);
3121 }
3122
3123 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
3124 {
3125         switch (hint) {
3126         case WRITE_LIFE_SHORT:
3127                 return CURSEG_HOT_DATA;
3128         case WRITE_LIFE_EXTREME:
3129                 return CURSEG_COLD_DATA;
3130         default:
3131                 return CURSEG_WARM_DATA;
3132         }
3133 }
3134
3135 static int __get_segment_type_2(struct f2fs_io_info *fio)
3136 {
3137         if (fio->type == DATA)
3138                 return CURSEG_HOT_DATA;
3139         else
3140                 return CURSEG_HOT_NODE;
3141 }
3142
3143 static int __get_segment_type_4(struct f2fs_io_info *fio)
3144 {
3145         if (fio->type == DATA) {
3146                 struct inode *inode = fio->page->mapping->host;
3147
3148                 if (S_ISDIR(inode->i_mode))
3149                         return CURSEG_HOT_DATA;
3150                 else
3151                         return CURSEG_COLD_DATA;
3152         } else {
3153                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3154                         return CURSEG_WARM_NODE;
3155                 else
3156                         return CURSEG_COLD_NODE;
3157         }
3158 }
3159
3160 static int __get_age_segment_type(struct inode *inode, pgoff_t pgofs)
3161 {
3162         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3163         struct extent_info ei = {};
3164
3165         if (f2fs_lookup_age_extent_cache(inode, pgofs, &ei)) {
3166                 if (!ei.age)
3167                         return NO_CHECK_TYPE;
3168                 if (ei.age <= sbi->hot_data_age_threshold)
3169                         return CURSEG_HOT_DATA;
3170                 if (ei.age <= sbi->warm_data_age_threshold)
3171                         return CURSEG_WARM_DATA;
3172                 return CURSEG_COLD_DATA;
3173         }
3174         return NO_CHECK_TYPE;
3175 }
3176
3177 static int __get_segment_type_6(struct f2fs_io_info *fio)
3178 {
3179         if (fio->type == DATA) {
3180                 struct inode *inode = fio->page->mapping->host;
3181                 int type;
3182
3183                 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
3184                         return CURSEG_COLD_DATA_PINNED;
3185
3186                 if (page_private_gcing(fio->page)) {
3187                         if (fio->sbi->am.atgc_enabled &&
3188                                 (fio->io_type == FS_DATA_IO) &&
3189                                 (fio->sbi->gc_mode != GC_URGENT_HIGH))
3190                                 return CURSEG_ALL_DATA_ATGC;
3191                         else
3192                                 return CURSEG_COLD_DATA;
3193                 }
3194                 if (file_is_cold(inode) || f2fs_need_compress_data(inode))
3195                         return CURSEG_COLD_DATA;
3196
3197                 type = __get_age_segment_type(inode, fio->page->index);
3198                 if (type != NO_CHECK_TYPE)
3199                         return type;
3200
3201                 if (file_is_hot(inode) ||
3202                                 is_inode_flag_set(inode, FI_HOT_DATA) ||
3203                                 f2fs_is_cow_file(inode))
3204                         return CURSEG_HOT_DATA;
3205                 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3206         } else {
3207                 if (IS_DNODE(fio->page))
3208                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3209                                                 CURSEG_HOT_NODE;
3210                 return CURSEG_COLD_NODE;
3211         }
3212 }
3213
3214 static int __get_segment_type(struct f2fs_io_info *fio)
3215 {
3216         int type = 0;
3217
3218         switch (F2FS_OPTION(fio->sbi).active_logs) {
3219         case 2:
3220                 type = __get_segment_type_2(fio);
3221                 break;
3222         case 4:
3223                 type = __get_segment_type_4(fio);
3224                 break;
3225         case 6:
3226                 type = __get_segment_type_6(fio);
3227                 break;
3228         default:
3229                 f2fs_bug_on(fio->sbi, true);
3230         }
3231
3232         if (IS_HOT(type))
3233                 fio->temp = HOT;
3234         else if (IS_WARM(type))
3235                 fio->temp = WARM;
3236         else
3237                 fio->temp = COLD;
3238         return type;
3239 }
3240
3241 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3242                 block_t old_blkaddr, block_t *new_blkaddr,
3243                 struct f2fs_summary *sum, int type,
3244                 struct f2fs_io_info *fio)
3245 {
3246         struct sit_info *sit_i = SIT_I(sbi);
3247         struct curseg_info *curseg = CURSEG_I(sbi, type);
3248         unsigned long long old_mtime;
3249         bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3250         struct seg_entry *se = NULL;
3251
3252         f2fs_down_read(&SM_I(sbi)->curseg_lock);
3253
3254         mutex_lock(&curseg->curseg_mutex);
3255         down_write(&sit_i->sentry_lock);
3256
3257         if (from_gc) {
3258                 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3259                 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3260                 sanity_check_seg_type(sbi, se->type);
3261                 f2fs_bug_on(sbi, IS_NODESEG(se->type));
3262         }
3263         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3264
3265         f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg);
3266
3267         f2fs_wait_discard_bio(sbi, *new_blkaddr);
3268
3269         /*
3270          * __add_sum_entry should be resided under the curseg_mutex
3271          * because, this function updates a summary entry in the
3272          * current summary block.
3273          */
3274         __add_sum_entry(sbi, type, sum);
3275
3276         __refresh_next_blkoff(sbi, curseg);
3277
3278         stat_inc_block_count(sbi, curseg);
3279
3280         if (from_gc) {
3281                 old_mtime = get_segment_mtime(sbi, old_blkaddr);
3282         } else {
3283                 update_segment_mtime(sbi, old_blkaddr, 0);
3284                 old_mtime = 0;
3285         }
3286         update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3287
3288         /*
3289          * SIT information should be updated before segment allocation,
3290          * since SSR needs latest valid block information.
3291          */
3292         update_sit_entry(sbi, *new_blkaddr, 1);
3293         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3294                 update_sit_entry(sbi, old_blkaddr, -1);
3295
3296         if (!__has_curseg_space(sbi, curseg)) {
3297                 /*
3298                  * Flush out current segment and replace it with new segment.
3299                  */
3300                 if (from_gc) {
3301                         get_atssr_segment(sbi, type, se->type,
3302                                                 AT_SSR, se->mtime);
3303                 } else {
3304                         if (need_new_seg(sbi, type))
3305                                 new_curseg(sbi, type, false);
3306                         else
3307                                 change_curseg(sbi, type);
3308                         stat_inc_seg_type(sbi, curseg);
3309                 }
3310         }
3311         /*
3312          * segment dirty status should be updated after segment allocation,
3313          * so we just need to update status only one time after previous
3314          * segment being closed.
3315          */
3316         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3317         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3318
3319         if (IS_DATASEG(type))
3320                 atomic64_inc(&sbi->allocated_data_blocks);
3321
3322         up_write(&sit_i->sentry_lock);
3323
3324         if (page && IS_NODESEG(type)) {
3325                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3326
3327                 f2fs_inode_chksum_set(sbi, page);
3328         }
3329
3330         if (fio) {
3331                 struct f2fs_bio_info *io;
3332
3333                 if (F2FS_IO_ALIGNED(sbi))
3334                         fio->retry = false;
3335
3336                 INIT_LIST_HEAD(&fio->list);
3337                 fio->in_list = true;
3338                 io = sbi->write_io[fio->type] + fio->temp;
3339                 spin_lock(&io->io_lock);
3340                 list_add_tail(&fio->list, &io->io_list);
3341                 spin_unlock(&io->io_lock);
3342         }
3343
3344         mutex_unlock(&curseg->curseg_mutex);
3345
3346         f2fs_up_read(&SM_I(sbi)->curseg_lock);
3347 }
3348
3349 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino,
3350                                         block_t blkaddr, unsigned int blkcnt)
3351 {
3352         if (!f2fs_is_multi_device(sbi))
3353                 return;
3354
3355         while (1) {
3356                 unsigned int devidx = f2fs_target_device_index(sbi, blkaddr);
3357                 unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1;
3358
3359                 /* update device state for fsync */
3360                 f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO);
3361
3362                 /* update device state for checkpoint */
3363                 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3364                         spin_lock(&sbi->dev_lock);
3365                         f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3366                         spin_unlock(&sbi->dev_lock);
3367                 }
3368
3369                 if (blkcnt <= blks)
3370                         break;
3371                 blkcnt -= blks;
3372                 blkaddr += blks;
3373         }
3374 }
3375
3376 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3377 {
3378         int type = __get_segment_type(fio);
3379         bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3380
3381         if (keep_order)
3382                 f2fs_down_read(&fio->sbi->io_order_lock);
3383 reallocate:
3384         f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3385                         &fio->new_blkaddr, sum, type, fio);
3386         if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO) {
3387                 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3388                                         fio->old_blkaddr, fio->old_blkaddr);
3389                 f2fs_invalidate_compress_page(fio->sbi, fio->old_blkaddr);
3390         }
3391
3392         /* writeout dirty page into bdev */
3393         f2fs_submit_page_write(fio);
3394         if (fio->retry) {
3395                 fio->old_blkaddr = fio->new_blkaddr;
3396                 goto reallocate;
3397         }
3398
3399         f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1);
3400
3401         if (keep_order)
3402                 f2fs_up_read(&fio->sbi->io_order_lock);
3403 }
3404
3405 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3406                                         enum iostat_type io_type)
3407 {
3408         struct f2fs_io_info fio = {
3409                 .sbi = sbi,
3410                 .type = META,
3411                 .temp = HOT,
3412                 .op = REQ_OP_WRITE,
3413                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3414                 .old_blkaddr = page->index,
3415                 .new_blkaddr = page->index,
3416                 .page = page,
3417                 .encrypted_page = NULL,
3418                 .in_list = false,
3419         };
3420
3421         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3422                 fio.op_flags &= ~REQ_META;
3423
3424         set_page_writeback(page);
3425         ClearPageError(page);
3426         f2fs_submit_page_write(&fio);
3427
3428         stat_inc_meta_count(sbi, page->index);
3429         f2fs_update_iostat(sbi, NULL, io_type, F2FS_BLKSIZE);
3430 }
3431
3432 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3433 {
3434         struct f2fs_summary sum;
3435
3436         set_summary(&sum, nid, 0, 0);
3437         do_write_page(&sum, fio);
3438
3439         f2fs_update_iostat(fio->sbi, NULL, fio->io_type, F2FS_BLKSIZE);
3440 }
3441
3442 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3443                                         struct f2fs_io_info *fio)
3444 {
3445         struct f2fs_sb_info *sbi = fio->sbi;
3446         struct f2fs_summary sum;
3447
3448         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3449         if (fio->io_type == FS_DATA_IO || fio->io_type == FS_CP_DATA_IO)
3450                 f2fs_update_age_extent_cache(dn);
3451         set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3452         do_write_page(&sum, fio);
3453         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3454
3455         f2fs_update_iostat(sbi, dn->inode, fio->io_type, F2FS_BLKSIZE);
3456 }
3457
3458 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3459 {
3460         int err;
3461         struct f2fs_sb_info *sbi = fio->sbi;
3462         unsigned int segno;
3463
3464         fio->new_blkaddr = fio->old_blkaddr;
3465         /* i/o temperature is needed for passing down write hints */
3466         __get_segment_type(fio);
3467
3468         segno = GET_SEGNO(sbi, fio->new_blkaddr);
3469
3470         if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3471                 set_sbi_flag(sbi, SBI_NEED_FSCK);
3472                 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3473                           __func__, segno);
3474                 err = -EFSCORRUPTED;
3475                 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
3476                 goto drop_bio;
3477         }
3478
3479         if (f2fs_cp_error(sbi)) {
3480                 err = -EIO;
3481                 goto drop_bio;
3482         }
3483
3484         if (fio->post_read)
3485                 invalidate_mapping_pages(META_MAPPING(sbi),
3486                                 fio->new_blkaddr, fio->new_blkaddr);
3487
3488         stat_inc_inplace_blocks(fio->sbi);
3489
3490         if (fio->bio && !(SM_I(sbi)->ipu_policy & (1 << F2FS_IPU_NOCACHE)))
3491                 err = f2fs_merge_page_bio(fio);
3492         else
3493                 err = f2fs_submit_page_bio(fio);
3494         if (!err) {
3495                 f2fs_update_device_state(fio->sbi, fio->ino,
3496                                                 fio->new_blkaddr, 1);
3497                 f2fs_update_iostat(fio->sbi, fio->page->mapping->host,
3498                                                 fio->io_type, F2FS_BLKSIZE);
3499         }
3500
3501         return err;
3502 drop_bio:
3503         if (fio->bio && *(fio->bio)) {
3504                 struct bio *bio = *(fio->bio);
3505
3506                 bio->bi_status = BLK_STS_IOERR;
3507                 bio_endio(bio);
3508                 *(fio->bio) = NULL;
3509         }
3510         return err;
3511 }
3512
3513 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3514                                                 unsigned int segno)
3515 {
3516         int i;
3517
3518         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3519                 if (CURSEG_I(sbi, i)->segno == segno)
3520                         break;
3521         }
3522         return i;
3523 }
3524
3525 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3526                                 block_t old_blkaddr, block_t new_blkaddr,
3527                                 bool recover_curseg, bool recover_newaddr,
3528                                 bool from_gc)
3529 {
3530         struct sit_info *sit_i = SIT_I(sbi);
3531         struct curseg_info *curseg;
3532         unsigned int segno, old_cursegno;
3533         struct seg_entry *se;
3534         int type;
3535         unsigned short old_blkoff;
3536         unsigned char old_alloc_type;
3537
3538         segno = GET_SEGNO(sbi, new_blkaddr);
3539         se = get_seg_entry(sbi, segno);
3540         type = se->type;
3541
3542         f2fs_down_write(&SM_I(sbi)->curseg_lock);
3543
3544         if (!recover_curseg) {
3545                 /* for recovery flow */
3546                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3547                         if (old_blkaddr == NULL_ADDR)
3548                                 type = CURSEG_COLD_DATA;
3549                         else
3550                                 type = CURSEG_WARM_DATA;
3551                 }
3552         } else {
3553                 if (IS_CURSEG(sbi, segno)) {
3554                         /* se->type is volatile as SSR allocation */
3555                         type = __f2fs_get_curseg(sbi, segno);
3556                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3557                 } else {
3558                         type = CURSEG_WARM_DATA;
3559                 }
3560         }
3561
3562         f2fs_bug_on(sbi, !IS_DATASEG(type));
3563         curseg = CURSEG_I(sbi, type);
3564
3565         mutex_lock(&curseg->curseg_mutex);
3566         down_write(&sit_i->sentry_lock);
3567
3568         old_cursegno = curseg->segno;
3569         old_blkoff = curseg->next_blkoff;
3570         old_alloc_type = curseg->alloc_type;
3571
3572         /* change the current segment */
3573         if (segno != curseg->segno) {
3574                 curseg->next_segno = segno;
3575                 change_curseg(sbi, type);
3576         }
3577
3578         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3579         __add_sum_entry(sbi, type, sum);
3580
3581         if (!recover_curseg || recover_newaddr) {
3582                 if (!from_gc)
3583                         update_segment_mtime(sbi, new_blkaddr, 0);
3584                 update_sit_entry(sbi, new_blkaddr, 1);
3585         }
3586         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3587                 invalidate_mapping_pages(META_MAPPING(sbi),
3588                                         old_blkaddr, old_blkaddr);
3589                 f2fs_invalidate_compress_page(sbi, old_blkaddr);
3590                 if (!from_gc)
3591                         update_segment_mtime(sbi, old_blkaddr, 0);
3592                 update_sit_entry(sbi, old_blkaddr, -1);
3593         }
3594
3595         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3596         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3597
3598         locate_dirty_segment(sbi, old_cursegno);
3599
3600         if (recover_curseg) {
3601                 if (old_cursegno != curseg->segno) {
3602                         curseg->next_segno = old_cursegno;
3603                         change_curseg(sbi, type);
3604                 }
3605                 curseg->next_blkoff = old_blkoff;
3606                 curseg->alloc_type = old_alloc_type;
3607         }
3608
3609         up_write(&sit_i->sentry_lock);
3610         mutex_unlock(&curseg->curseg_mutex);
3611         f2fs_up_write(&SM_I(sbi)->curseg_lock);
3612 }
3613
3614 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3615                                 block_t old_addr, block_t new_addr,
3616                                 unsigned char version, bool recover_curseg,
3617                                 bool recover_newaddr)
3618 {
3619         struct f2fs_summary sum;
3620
3621         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3622
3623         f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3624                                         recover_curseg, recover_newaddr, false);
3625
3626         f2fs_update_data_blkaddr(dn, new_addr);
3627 }
3628
3629 void f2fs_wait_on_page_writeback(struct page *page,
3630                                 enum page_type type, bool ordered, bool locked)
3631 {
3632         if (PageWriteback(page)) {
3633                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3634
3635                 /* submit cached LFS IO */
3636                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3637                 /* sbumit cached IPU IO */
3638                 f2fs_submit_merged_ipu_write(sbi, NULL, page);
3639                 if (ordered) {
3640                         wait_on_page_writeback(page);
3641                         f2fs_bug_on(sbi, locked && PageWriteback(page));
3642                 } else {
3643                         wait_for_stable_page(page);
3644                 }
3645         }
3646 }
3647
3648 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3649 {
3650         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3651         struct page *cpage;
3652
3653         if (!f2fs_post_read_required(inode))
3654                 return;
3655
3656         if (!__is_valid_data_blkaddr(blkaddr))
3657                 return;
3658
3659         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3660         if (cpage) {
3661                 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3662                 f2fs_put_page(cpage, 1);
3663         }
3664 }
3665
3666 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3667                                                                 block_t len)
3668 {
3669         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3670         block_t i;
3671
3672         if (!f2fs_post_read_required(inode))
3673                 return;
3674
3675         for (i = 0; i < len; i++)
3676                 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3677
3678         invalidate_mapping_pages(META_MAPPING(sbi), blkaddr, blkaddr + len - 1);
3679 }
3680
3681 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3682 {
3683         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3684         struct curseg_info *seg_i;
3685         unsigned char *kaddr;
3686         struct page *page;
3687         block_t start;
3688         int i, j, offset;
3689
3690         start = start_sum_block(sbi);
3691
3692         page = f2fs_get_meta_page(sbi, start++);
3693         if (IS_ERR(page))
3694                 return PTR_ERR(page);
3695         kaddr = (unsigned char *)page_address(page);
3696
3697         /* Step 1: restore nat cache */
3698         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3699         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3700
3701         /* Step 2: restore sit cache */
3702         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3703         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3704         offset = 2 * SUM_JOURNAL_SIZE;
3705
3706         /* Step 3: restore summary entries */
3707         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3708                 unsigned short blk_off;
3709                 unsigned int segno;
3710
3711                 seg_i = CURSEG_I(sbi, i);
3712                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3713                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3714                 seg_i->next_segno = segno;
3715                 reset_curseg(sbi, i, 0);
3716                 seg_i->alloc_type = ckpt->alloc_type[i];
3717                 seg_i->next_blkoff = blk_off;
3718
3719                 if (seg_i->alloc_type == SSR)
3720                         blk_off = sbi->blocks_per_seg;
3721
3722                 for (j = 0; j < blk_off; j++) {
3723                         struct f2fs_summary *s;
3724
3725                         s = (struct f2fs_summary *)(kaddr + offset);
3726                         seg_i->sum_blk->entries[j] = *s;
3727                         offset += SUMMARY_SIZE;
3728                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3729                                                 SUM_FOOTER_SIZE)
3730                                 continue;
3731
3732                         f2fs_put_page(page, 1);
3733                         page = NULL;
3734
3735                         page = f2fs_get_meta_page(sbi, start++);
3736                         if (IS_ERR(page))
3737                                 return PTR_ERR(page);
3738                         kaddr = (unsigned char *)page_address(page);
3739                         offset = 0;
3740                 }
3741         }
3742         f2fs_put_page(page, 1);
3743         return 0;
3744 }
3745
3746 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3747 {
3748         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3749         struct f2fs_summary_block *sum;
3750         struct curseg_info *curseg;
3751         struct page *new;
3752         unsigned short blk_off;
3753         unsigned int segno = 0;
3754         block_t blk_addr = 0;
3755         int err = 0;
3756
3757         /* get segment number and block addr */
3758         if (IS_DATASEG(type)) {
3759                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3760                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3761                                                         CURSEG_HOT_DATA]);
3762                 if (__exist_node_summaries(sbi))
3763                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
3764                 else
3765                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3766         } else {
3767                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3768                                                         CURSEG_HOT_NODE]);
3769                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3770                                                         CURSEG_HOT_NODE]);
3771                 if (__exist_node_summaries(sbi))
3772                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3773                                                         type - CURSEG_HOT_NODE);
3774                 else
3775                         blk_addr = GET_SUM_BLOCK(sbi, segno);
3776         }
3777
3778         new = f2fs_get_meta_page(sbi, blk_addr);
3779         if (IS_ERR(new))
3780                 return PTR_ERR(new);
3781         sum = (struct f2fs_summary_block *)page_address(new);
3782
3783         if (IS_NODESEG(type)) {
3784                 if (__exist_node_summaries(sbi)) {
3785                         struct f2fs_summary *ns = &sum->entries[0];
3786                         int i;
3787
3788                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3789                                 ns->version = 0;
3790                                 ns->ofs_in_node = 0;
3791                         }
3792                 } else {
3793                         err = f2fs_restore_node_summary(sbi, segno, sum);
3794                         if (err)
3795                                 goto out;
3796                 }
3797         }
3798
3799         /* set uncompleted segment to curseg */
3800         curseg = CURSEG_I(sbi, type);
3801         mutex_lock(&curseg->curseg_mutex);
3802
3803         /* update journal info */
3804         down_write(&curseg->journal_rwsem);
3805         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3806         up_write(&curseg->journal_rwsem);
3807
3808         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3809         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3810         curseg->next_segno = segno;
3811         reset_curseg(sbi, type, 0);
3812         curseg->alloc_type = ckpt->alloc_type[type];
3813         curseg->next_blkoff = blk_off;
3814         mutex_unlock(&curseg->curseg_mutex);
3815 out:
3816         f2fs_put_page(new, 1);
3817         return err;
3818 }
3819
3820 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3821 {
3822         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3823         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3824         int type = CURSEG_HOT_DATA;
3825         int err;
3826
3827         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3828                 int npages = f2fs_npages_for_summary_flush(sbi, true);
3829
3830                 if (npages >= 2)
3831                         f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3832                                                         META_CP, true);
3833
3834                 /* restore for compacted data summary */
3835                 err = read_compacted_summaries(sbi);
3836                 if (err)
3837                         return err;
3838                 type = CURSEG_HOT_NODE;
3839         }
3840
3841         if (__exist_node_summaries(sbi))
3842                 f2fs_ra_meta_pages(sbi,
3843                                 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
3844                                 NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
3845
3846         for (; type <= CURSEG_COLD_NODE; type++) {
3847                 err = read_normal_summaries(sbi, type);
3848                 if (err)
3849                         return err;
3850         }
3851
3852         /* sanity check for summary blocks */
3853         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3854                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3855                 f2fs_err(sbi, "invalid journal entries nats %u sits %u",
3856                          nats_in_cursum(nat_j), sits_in_cursum(sit_j));
3857                 return -EINVAL;
3858         }
3859
3860         return 0;
3861 }
3862
3863 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3864 {
3865         struct page *page;
3866         unsigned char *kaddr;
3867         struct f2fs_summary *summary;
3868         struct curseg_info *seg_i;
3869         int written_size = 0;
3870         int i, j;
3871
3872         page = f2fs_grab_meta_page(sbi, blkaddr++);
3873         kaddr = (unsigned char *)page_address(page);
3874         memset(kaddr, 0, PAGE_SIZE);
3875
3876         /* Step 1: write nat cache */
3877         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3878         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3879         written_size += SUM_JOURNAL_SIZE;
3880
3881         /* Step 2: write sit cache */
3882         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3883         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3884         written_size += SUM_JOURNAL_SIZE;
3885
3886         /* Step 3: write summary entries */
3887         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3888                 unsigned short blkoff;
3889
3890                 seg_i = CURSEG_I(sbi, i);
3891                 if (sbi->ckpt->alloc_type[i] == SSR)
3892                         blkoff = sbi->blocks_per_seg;
3893                 else
3894                         blkoff = curseg_blkoff(sbi, i);
3895
3896                 for (j = 0; j < blkoff; j++) {
3897                         if (!page) {
3898                                 page = f2fs_grab_meta_page(sbi, blkaddr++);
3899                                 kaddr = (unsigned char *)page_address(page);
3900                                 memset(kaddr, 0, PAGE_SIZE);
3901                                 written_size = 0;
3902                         }
3903                         summary = (struct f2fs_summary *)(kaddr + written_size);
3904                         *summary = seg_i->sum_blk->entries[j];
3905                         written_size += SUMMARY_SIZE;
3906
3907                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3908                                                         SUM_FOOTER_SIZE)
3909                                 continue;
3910
3911                         set_page_dirty(page);
3912                         f2fs_put_page(page, 1);
3913                         page = NULL;
3914                 }
3915         }
3916         if (page) {
3917                 set_page_dirty(page);
3918                 f2fs_put_page(page, 1);
3919         }
3920 }
3921
3922 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3923                                         block_t blkaddr, int type)
3924 {
3925         int i, end;
3926
3927         if (IS_DATASEG(type))
3928                 end = type + NR_CURSEG_DATA_TYPE;
3929         else
3930                 end = type + NR_CURSEG_NODE_TYPE;
3931
3932         for (i = type; i < end; i++)
3933                 write_current_sum_page(sbi, i, blkaddr + (i - type));
3934 }
3935
3936 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3937 {
3938         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3939                 write_compacted_summaries(sbi, start_blk);
3940         else
3941                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3942 }
3943
3944 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3945 {
3946         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3947 }
3948
3949 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3950                                         unsigned int val, int alloc)
3951 {
3952         int i;
3953
3954         if (type == NAT_JOURNAL) {
3955                 for (i = 0; i < nats_in_cursum(journal); i++) {
3956                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3957                                 return i;
3958                 }
3959                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3960                         return update_nats_in_cursum(journal, 1);
3961         } else if (type == SIT_JOURNAL) {
3962                 for (i = 0; i < sits_in_cursum(journal); i++)
3963                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3964                                 return i;
3965                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3966                         return update_sits_in_cursum(journal, 1);
3967         }
3968         return -1;
3969 }
3970
3971 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3972                                         unsigned int segno)
3973 {
3974         return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
3975 }
3976
3977 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3978                                         unsigned int start)
3979 {
3980         struct sit_info *sit_i = SIT_I(sbi);
3981         struct page *page;
3982         pgoff_t src_off, dst_off;
3983
3984         src_off = current_sit_addr(sbi, start);
3985         dst_off = next_sit_addr(sbi, src_off);
3986
3987         page = f2fs_grab_meta_page(sbi, dst_off);
3988         seg_info_to_sit_page(sbi, page, start);
3989
3990         set_page_dirty(page);
3991         set_to_next_sit(sit_i, start);
3992
3993         return page;
3994 }
3995
3996 static struct sit_entry_set *grab_sit_entry_set(void)
3997 {
3998         struct sit_entry_set *ses =
3999                         f2fs_kmem_cache_alloc(sit_entry_set_slab,
4000                                                 GFP_NOFS, true, NULL);
4001
4002         ses->entry_cnt = 0;
4003         INIT_LIST_HEAD(&ses->set_list);
4004         return ses;
4005 }
4006
4007 static void release_sit_entry_set(struct sit_entry_set *ses)
4008 {
4009         list_del(&ses->set_list);
4010         kmem_cache_free(sit_entry_set_slab, ses);
4011 }
4012
4013 static void adjust_sit_entry_set(struct sit_entry_set *ses,
4014                                                 struct list_head *head)
4015 {
4016         struct sit_entry_set *next = ses;
4017
4018         if (list_is_last(&ses->set_list, head))
4019                 return;
4020
4021         list_for_each_entry_continue(next, head, set_list)
4022                 if (ses->entry_cnt <= next->entry_cnt) {
4023                         list_move_tail(&ses->set_list, &next->set_list);
4024                         return;
4025                 }
4026
4027         list_move_tail(&ses->set_list, head);
4028 }
4029
4030 static void add_sit_entry(unsigned int segno, struct list_head *head)
4031 {
4032         struct sit_entry_set *ses;
4033         unsigned int start_segno = START_SEGNO(segno);
4034
4035         list_for_each_entry(ses, head, set_list) {
4036                 if (ses->start_segno == start_segno) {
4037                         ses->entry_cnt++;
4038                         adjust_sit_entry_set(ses, head);
4039                         return;
4040                 }
4041         }
4042
4043         ses = grab_sit_entry_set();
4044
4045         ses->start_segno = start_segno;
4046         ses->entry_cnt++;
4047         list_add(&ses->set_list, head);
4048 }
4049
4050 static void add_sits_in_set(struct f2fs_sb_info *sbi)
4051 {
4052         struct f2fs_sm_info *sm_info = SM_I(sbi);
4053         struct list_head *set_list = &sm_info->sit_entry_set;
4054         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
4055         unsigned int segno;
4056
4057         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
4058                 add_sit_entry(segno, set_list);
4059 }
4060
4061 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
4062 {
4063         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4064         struct f2fs_journal *journal = curseg->journal;
4065         int i;
4066
4067         down_write(&curseg->journal_rwsem);
4068         for (i = 0; i < sits_in_cursum(journal); i++) {
4069                 unsigned int segno;
4070                 bool dirtied;
4071
4072                 segno = le32_to_cpu(segno_in_journal(journal, i));
4073                 dirtied = __mark_sit_entry_dirty(sbi, segno);
4074
4075                 if (!dirtied)
4076                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
4077         }
4078         update_sits_in_cursum(journal, -i);
4079         up_write(&curseg->journal_rwsem);
4080 }
4081
4082 /*
4083  * CP calls this function, which flushes SIT entries including sit_journal,
4084  * and moves prefree segs to free segs.
4085  */
4086 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
4087 {
4088         struct sit_info *sit_i = SIT_I(sbi);
4089         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4090         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4091         struct f2fs_journal *journal = curseg->journal;
4092         struct sit_entry_set *ses, *tmp;
4093         struct list_head *head = &SM_I(sbi)->sit_entry_set;
4094         bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4095         struct seg_entry *se;
4096
4097         down_write(&sit_i->sentry_lock);
4098
4099         if (!sit_i->dirty_sentries)
4100                 goto out;
4101
4102         /*
4103          * add and account sit entries of dirty bitmap in sit entry
4104          * set temporarily
4105          */
4106         add_sits_in_set(sbi);
4107
4108         /*
4109          * if there are no enough space in journal to store dirty sit
4110          * entries, remove all entries from journal and add and account
4111          * them in sit entry set.
4112          */
4113         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4114                                                                 !to_journal)
4115                 remove_sits_in_journal(sbi);
4116
4117         /*
4118          * there are two steps to flush sit entries:
4119          * #1, flush sit entries to journal in current cold data summary block.
4120          * #2, flush sit entries to sit page.
4121          */
4122         list_for_each_entry_safe(ses, tmp, head, set_list) {
4123                 struct page *page = NULL;
4124                 struct f2fs_sit_block *raw_sit = NULL;
4125                 unsigned int start_segno = ses->start_segno;
4126                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4127                                                 (unsigned long)MAIN_SEGS(sbi));
4128                 unsigned int segno = start_segno;
4129
4130                 if (to_journal &&
4131                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4132                         to_journal = false;
4133
4134                 if (to_journal) {
4135                         down_write(&curseg->journal_rwsem);
4136                 } else {
4137                         page = get_next_sit_page(sbi, start_segno);
4138                         raw_sit = page_address(page);
4139                 }
4140
4141                 /* flush dirty sit entries in region of current sit set */
4142                 for_each_set_bit_from(segno, bitmap, end) {
4143                         int offset, sit_offset;
4144
4145                         se = get_seg_entry(sbi, segno);
4146 #ifdef CONFIG_F2FS_CHECK_FS
4147                         if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4148                                                 SIT_VBLOCK_MAP_SIZE))
4149                                 f2fs_bug_on(sbi, 1);
4150 #endif
4151
4152                         /* add discard candidates */
4153                         if (!(cpc->reason & CP_DISCARD)) {
4154                                 cpc->trim_start = segno;
4155                                 add_discard_addrs(sbi, cpc, false);
4156                         }
4157
4158                         if (to_journal) {
4159                                 offset = f2fs_lookup_journal_in_cursum(journal,
4160                                                         SIT_JOURNAL, segno, 1);
4161                                 f2fs_bug_on(sbi, offset < 0);
4162                                 segno_in_journal(journal, offset) =
4163                                                         cpu_to_le32(segno);
4164                                 seg_info_to_raw_sit(se,
4165                                         &sit_in_journal(journal, offset));
4166                                 check_block_count(sbi, segno,
4167                                         &sit_in_journal(journal, offset));
4168                         } else {
4169                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4170                                 seg_info_to_raw_sit(se,
4171                                                 &raw_sit->entries[sit_offset]);
4172                                 check_block_count(sbi, segno,
4173                                                 &raw_sit->entries[sit_offset]);
4174                         }
4175
4176                         __clear_bit(segno, bitmap);
4177                         sit_i->dirty_sentries--;
4178                         ses->entry_cnt--;
4179                 }
4180
4181                 if (to_journal)
4182                         up_write(&curseg->journal_rwsem);
4183                 else
4184                         f2fs_put_page(page, 1);
4185
4186                 f2fs_bug_on(sbi, ses->entry_cnt);
4187                 release_sit_entry_set(ses);
4188         }
4189
4190         f2fs_bug_on(sbi, !list_empty(head));
4191         f2fs_bug_on(sbi, sit_i->dirty_sentries);
4192 out:
4193         if (cpc->reason & CP_DISCARD) {
4194                 __u64 trim_start = cpc->trim_start;
4195
4196                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4197                         add_discard_addrs(sbi, cpc, false);
4198
4199                 cpc->trim_start = trim_start;
4200         }
4201         up_write(&sit_i->sentry_lock);
4202
4203         set_prefree_as_free_segments(sbi);
4204 }
4205
4206 static int build_sit_info(struct f2fs_sb_info *sbi)
4207 {
4208         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4209         struct sit_info *sit_i;
4210         unsigned int sit_segs, start;
4211         char *src_bitmap, *bitmap;
4212         unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4213         unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0;
4214
4215         /* allocate memory for SIT information */
4216         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4217         if (!sit_i)
4218                 return -ENOMEM;
4219
4220         SM_I(sbi)->sit_info = sit_i;
4221
4222         sit_i->sentries =
4223                 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4224                                               MAIN_SEGS(sbi)),
4225                               GFP_KERNEL);
4226         if (!sit_i->sentries)
4227                 return -ENOMEM;
4228
4229         main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4230         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4231                                                                 GFP_KERNEL);
4232         if (!sit_i->dirty_sentries_bitmap)
4233                 return -ENOMEM;
4234
4235 #ifdef CONFIG_F2FS_CHECK_FS
4236         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map);
4237 #else
4238         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map);
4239 #endif
4240         sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4241         if (!sit_i->bitmap)
4242                 return -ENOMEM;
4243
4244         bitmap = sit_i->bitmap;
4245
4246         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4247                 sit_i->sentries[start].cur_valid_map = bitmap;
4248                 bitmap += SIT_VBLOCK_MAP_SIZE;
4249
4250                 sit_i->sentries[start].ckpt_valid_map = bitmap;
4251                 bitmap += SIT_VBLOCK_MAP_SIZE;
4252
4253 #ifdef CONFIG_F2FS_CHECK_FS
4254                 sit_i->sentries[start].cur_valid_map_mir = bitmap;
4255                 bitmap += SIT_VBLOCK_MAP_SIZE;
4256 #endif
4257
4258                 if (discard_map) {
4259                         sit_i->sentries[start].discard_map = bitmap;
4260                         bitmap += SIT_VBLOCK_MAP_SIZE;
4261                 }
4262         }
4263
4264         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4265         if (!sit_i->tmp_map)
4266                 return -ENOMEM;
4267
4268         if (__is_large_section(sbi)) {
4269                 sit_i->sec_entries =
4270                         f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4271                                                       MAIN_SECS(sbi)),
4272                                       GFP_KERNEL);
4273                 if (!sit_i->sec_entries)
4274                         return -ENOMEM;
4275         }
4276
4277         /* get information related with SIT */
4278         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4279
4280         /* setup SIT bitmap from ckeckpoint pack */
4281         sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4282         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4283
4284         sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4285         if (!sit_i->sit_bitmap)
4286                 return -ENOMEM;
4287
4288 #ifdef CONFIG_F2FS_CHECK_FS
4289         sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4290                                         sit_bitmap_size, GFP_KERNEL);
4291         if (!sit_i->sit_bitmap_mir)
4292                 return -ENOMEM;
4293
4294         sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4295                                         main_bitmap_size, GFP_KERNEL);
4296         if (!sit_i->invalid_segmap)
4297                 return -ENOMEM;
4298 #endif
4299
4300         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4301         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4302         sit_i->written_valid_blocks = 0;
4303         sit_i->bitmap_size = sit_bitmap_size;
4304         sit_i->dirty_sentries = 0;
4305         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4306         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4307         sit_i->mounted_time = ktime_get_boottime_seconds();
4308         init_rwsem(&sit_i->sentry_lock);
4309         return 0;
4310 }
4311
4312 static int build_free_segmap(struct f2fs_sb_info *sbi)
4313 {
4314         struct free_segmap_info *free_i;
4315         unsigned int bitmap_size, sec_bitmap_size;
4316
4317         /* allocate memory for free segmap information */
4318         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4319         if (!free_i)
4320                 return -ENOMEM;
4321
4322         SM_I(sbi)->free_info = free_i;
4323
4324         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4325         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4326         if (!free_i->free_segmap)
4327                 return -ENOMEM;
4328
4329         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4330         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4331         if (!free_i->free_secmap)
4332                 return -ENOMEM;
4333
4334         /* set all segments as dirty temporarily */
4335         memset(free_i->free_segmap, 0xff, bitmap_size);
4336         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4337
4338         /* init free segmap information */
4339         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4340         free_i->free_segments = 0;
4341         free_i->free_sections = 0;
4342         spin_lock_init(&free_i->segmap_lock);
4343         return 0;
4344 }
4345
4346 static int build_curseg(struct f2fs_sb_info *sbi)
4347 {
4348         struct curseg_info *array;
4349         int i;
4350
4351         array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4352                                         sizeof(*array)), GFP_KERNEL);
4353         if (!array)
4354                 return -ENOMEM;
4355
4356         SM_I(sbi)->curseg_array = array;
4357
4358         for (i = 0; i < NO_CHECK_TYPE; i++) {
4359                 mutex_init(&array[i].curseg_mutex);
4360                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4361                 if (!array[i].sum_blk)
4362                         return -ENOMEM;
4363                 init_rwsem(&array[i].journal_rwsem);
4364                 array[i].journal = f2fs_kzalloc(sbi,
4365                                 sizeof(struct f2fs_journal), GFP_KERNEL);
4366                 if (!array[i].journal)
4367                         return -ENOMEM;
4368                 if (i < NR_PERSISTENT_LOG)
4369                         array[i].seg_type = CURSEG_HOT_DATA + i;
4370                 else if (i == CURSEG_COLD_DATA_PINNED)
4371                         array[i].seg_type = CURSEG_COLD_DATA;
4372                 else if (i == CURSEG_ALL_DATA_ATGC)
4373                         array[i].seg_type = CURSEG_COLD_DATA;
4374                 array[i].segno = NULL_SEGNO;
4375                 array[i].next_blkoff = 0;
4376                 array[i].inited = false;
4377         }
4378         return restore_curseg_summaries(sbi);
4379 }
4380
4381 static int build_sit_entries(struct f2fs_sb_info *sbi)
4382 {
4383         struct sit_info *sit_i = SIT_I(sbi);
4384         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4385         struct f2fs_journal *journal = curseg->journal;
4386         struct seg_entry *se;
4387         struct f2fs_sit_entry sit;
4388         int sit_blk_cnt = SIT_BLK_CNT(sbi);
4389         unsigned int i, start, end;
4390         unsigned int readed, start_blk = 0;
4391         int err = 0;
4392         block_t sit_valid_blocks[2] = {0, 0};
4393
4394         do {
4395                 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS,
4396                                                         META_SIT, true);
4397
4398                 start = start_blk * sit_i->sents_per_block;
4399                 end = (start_blk + readed) * sit_i->sents_per_block;
4400
4401                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4402                         struct f2fs_sit_block *sit_blk;
4403                         struct page *page;
4404
4405                         se = &sit_i->sentries[start];
4406                         page = get_current_sit_page(sbi, start);
4407                         if (IS_ERR(page))
4408                                 return PTR_ERR(page);
4409                         sit_blk = (struct f2fs_sit_block *)page_address(page);
4410                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4411                         f2fs_put_page(page, 1);
4412
4413                         err = check_block_count(sbi, start, &sit);
4414                         if (err)
4415                                 return err;
4416                         seg_info_from_raw_sit(se, &sit);
4417
4418                         if (se->type >= NR_PERSISTENT_LOG) {
4419                                 f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4420                                                         se->type, start);
4421                                 f2fs_handle_error(sbi,
4422                                                 ERROR_INCONSISTENT_SUM_TYPE);
4423                                 return -EFSCORRUPTED;
4424                         }
4425
4426                         sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4427
4428                         if (f2fs_block_unit_discard(sbi)) {
4429                                 /* build discard map only one time */
4430                                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4431                                         memset(se->discard_map, 0xff,
4432                                                 SIT_VBLOCK_MAP_SIZE);
4433                                 } else {
4434                                         memcpy(se->discard_map,
4435                                                 se->cur_valid_map,
4436                                                 SIT_VBLOCK_MAP_SIZE);
4437                                         sbi->discard_blks +=
4438                                                 sbi->blocks_per_seg -
4439                                                 se->valid_blocks;
4440                                 }
4441                         }
4442
4443                         if (__is_large_section(sbi))
4444                                 get_sec_entry(sbi, start)->valid_blocks +=
4445                                                         se->valid_blocks;
4446                 }
4447                 start_blk += readed;
4448         } while (start_blk < sit_blk_cnt);
4449
4450         down_read(&curseg->journal_rwsem);
4451         for (i = 0; i < sits_in_cursum(journal); i++) {
4452                 unsigned int old_valid_blocks;
4453
4454                 start = le32_to_cpu(segno_in_journal(journal, i));
4455                 if (start >= MAIN_SEGS(sbi)) {
4456                         f2fs_err(sbi, "Wrong journal entry on segno %u",
4457                                  start);
4458                         err = -EFSCORRUPTED;
4459                         f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL);
4460                         break;
4461                 }
4462
4463                 se = &sit_i->sentries[start];
4464                 sit = sit_in_journal(journal, i);
4465
4466                 old_valid_blocks = se->valid_blocks;
4467
4468                 sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks;
4469
4470                 err = check_block_count(sbi, start, &sit);
4471                 if (err)
4472                         break;
4473                 seg_info_from_raw_sit(se, &sit);
4474
4475                 if (se->type >= NR_PERSISTENT_LOG) {
4476                         f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4477                                                         se->type, start);
4478                         err = -EFSCORRUPTED;
4479                         f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
4480                         break;
4481                 }
4482
4483                 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4484
4485                 if (f2fs_block_unit_discard(sbi)) {
4486                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4487                                 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4488                         } else {
4489                                 memcpy(se->discard_map, se->cur_valid_map,
4490                                                         SIT_VBLOCK_MAP_SIZE);
4491                                 sbi->discard_blks += old_valid_blocks;
4492                                 sbi->discard_blks -= se->valid_blocks;
4493                         }
4494                 }
4495
4496                 if (__is_large_section(sbi)) {
4497                         get_sec_entry(sbi, start)->valid_blocks +=
4498                                                         se->valid_blocks;
4499                         get_sec_entry(sbi, start)->valid_blocks -=
4500                                                         old_valid_blocks;
4501                 }
4502         }
4503         up_read(&curseg->journal_rwsem);
4504
4505         if (err)
4506                 return err;
4507
4508         if (sit_valid_blocks[NODE] != valid_node_count(sbi)) {
4509                 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4510                          sit_valid_blocks[NODE], valid_node_count(sbi));
4511                 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT);
4512                 return -EFSCORRUPTED;
4513         }
4514
4515         if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] >
4516                                 valid_user_blocks(sbi)) {
4517                 f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u",
4518                          sit_valid_blocks[DATA], sit_valid_blocks[NODE],
4519                          valid_user_blocks(sbi));
4520                 f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT);
4521                 return -EFSCORRUPTED;
4522         }
4523
4524         return 0;
4525 }
4526
4527 static void init_free_segmap(struct f2fs_sb_info *sbi)
4528 {
4529         unsigned int start;
4530         int type;
4531         struct seg_entry *sentry;
4532
4533         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4534                 if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4535                         continue;
4536                 sentry = get_seg_entry(sbi, start);
4537                 if (!sentry->valid_blocks)
4538                         __set_free(sbi, start);
4539                 else
4540                         SIT_I(sbi)->written_valid_blocks +=
4541                                                 sentry->valid_blocks;
4542         }
4543
4544         /* set use the current segments */
4545         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4546                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4547
4548                 __set_test_and_inuse(sbi, curseg_t->segno);
4549         }
4550 }
4551
4552 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4553 {
4554         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4555         struct free_segmap_info *free_i = FREE_I(sbi);
4556         unsigned int segno = 0, offset = 0, secno;
4557         block_t valid_blocks, usable_blks_in_seg;
4558
4559         while (1) {
4560                 /* find dirty segment based on free segmap */
4561                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4562                 if (segno >= MAIN_SEGS(sbi))
4563                         break;
4564                 offset = segno + 1;
4565                 valid_blocks = get_valid_blocks(sbi, segno, false);
4566                 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4567                 if (valid_blocks == usable_blks_in_seg || !valid_blocks)
4568                         continue;
4569                 if (valid_blocks > usable_blks_in_seg) {
4570                         f2fs_bug_on(sbi, 1);
4571                         continue;
4572                 }
4573                 mutex_lock(&dirty_i->seglist_lock);
4574                 __locate_dirty_segment(sbi, segno, DIRTY);
4575                 mutex_unlock(&dirty_i->seglist_lock);
4576         }
4577
4578         if (!__is_large_section(sbi))
4579                 return;
4580
4581         mutex_lock(&dirty_i->seglist_lock);
4582         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4583                 valid_blocks = get_valid_blocks(sbi, segno, true);
4584                 secno = GET_SEC_FROM_SEG(sbi, segno);
4585
4586                 if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi))
4587                         continue;
4588                 if (IS_CURSEC(sbi, secno))
4589                         continue;
4590                 set_bit(secno, dirty_i->dirty_secmap);
4591         }
4592         mutex_unlock(&dirty_i->seglist_lock);
4593 }
4594
4595 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4596 {
4597         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4598         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4599
4600         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4601         if (!dirty_i->victim_secmap)
4602                 return -ENOMEM;
4603
4604         dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4605         if (!dirty_i->pinned_secmap)
4606                 return -ENOMEM;
4607
4608         dirty_i->pinned_secmap_cnt = 0;
4609         dirty_i->enable_pin_section = true;
4610         return 0;
4611 }
4612
4613 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4614 {
4615         struct dirty_seglist_info *dirty_i;
4616         unsigned int bitmap_size, i;
4617
4618         /* allocate memory for dirty segments list information */
4619         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4620                                                                 GFP_KERNEL);
4621         if (!dirty_i)
4622                 return -ENOMEM;
4623
4624         SM_I(sbi)->dirty_info = dirty_i;
4625         mutex_init(&dirty_i->seglist_lock);
4626
4627         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4628
4629         for (i = 0; i < NR_DIRTY_TYPE; i++) {
4630                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4631                                                                 GFP_KERNEL);
4632                 if (!dirty_i->dirty_segmap[i])
4633                         return -ENOMEM;
4634         }
4635
4636         if (__is_large_section(sbi)) {
4637                 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4638                 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
4639                                                 bitmap_size, GFP_KERNEL);
4640                 if (!dirty_i->dirty_secmap)
4641                         return -ENOMEM;
4642         }
4643
4644         init_dirty_segmap(sbi);
4645         return init_victim_secmap(sbi);
4646 }
4647
4648 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4649 {
4650         int i;
4651
4652         /*
4653          * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4654          * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4655          */
4656         for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4657                 struct curseg_info *curseg = CURSEG_I(sbi, i);
4658                 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4659                 unsigned int blkofs = curseg->next_blkoff;
4660
4661                 if (f2fs_sb_has_readonly(sbi) &&
4662                         i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE)
4663                         continue;
4664
4665                 sanity_check_seg_type(sbi, curseg->seg_type);
4666
4667                 if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) {
4668                         f2fs_err(sbi,
4669                                  "Current segment has invalid alloc_type:%d",
4670                                  curseg->alloc_type);
4671                         f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
4672                         return -EFSCORRUPTED;
4673                 }
4674
4675                 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4676                         goto out;
4677
4678                 if (curseg->alloc_type == SSR)
4679                         continue;
4680
4681                 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4682                         if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4683                                 continue;
4684 out:
4685                         f2fs_err(sbi,
4686                                  "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4687                                  i, curseg->segno, curseg->alloc_type,
4688                                  curseg->next_blkoff, blkofs);
4689                         f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
4690                         return -EFSCORRUPTED;
4691                 }
4692         }
4693         return 0;
4694 }
4695
4696 #ifdef CONFIG_BLK_DEV_ZONED
4697
4698 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4699                                     struct f2fs_dev_info *fdev,
4700                                     struct blk_zone *zone)
4701 {
4702         unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno;
4703         block_t zone_block, wp_block, last_valid_block;
4704         unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4705         int i, s, b, ret;
4706         struct seg_entry *se;
4707
4708         if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4709                 return 0;
4710
4711         wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block);
4712         wp_segno = GET_SEGNO(sbi, wp_block);
4713         wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4714         zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4715         zone_segno = GET_SEGNO(sbi, zone_block);
4716         zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno);
4717
4718         if (zone_segno >= MAIN_SEGS(sbi))
4719                 return 0;
4720
4721         /*
4722          * Skip check of zones cursegs point to, since
4723          * fix_curseg_write_pointer() checks them.
4724          */
4725         for (i = 0; i < NO_CHECK_TYPE; i++)
4726                 if (zone_secno == GET_SEC_FROM_SEG(sbi,
4727                                                    CURSEG_I(sbi, i)->segno))
4728                         return 0;
4729
4730         /*
4731          * Get last valid block of the zone.
4732          */
4733         last_valid_block = zone_block - 1;
4734         for (s = sbi->segs_per_sec - 1; s >= 0; s--) {
4735                 segno = zone_segno + s;
4736                 se = get_seg_entry(sbi, segno);
4737                 for (b = sbi->blocks_per_seg - 1; b >= 0; b--)
4738                         if (f2fs_test_bit(b, se->cur_valid_map)) {
4739                                 last_valid_block = START_BLOCK(sbi, segno) + b;
4740                                 break;
4741                         }
4742                 if (last_valid_block >= zone_block)
4743                         break;
4744         }
4745
4746         /*
4747          * If last valid block is beyond the write pointer, report the
4748          * inconsistency. This inconsistency does not cause write error
4749          * because the zone will not be selected for write operation until
4750          * it get discarded. Just report it.
4751          */
4752         if (last_valid_block >= wp_block) {
4753                 f2fs_notice(sbi, "Valid block beyond write pointer: "
4754                             "valid block[0x%x,0x%x] wp[0x%x,0x%x]",
4755                             GET_SEGNO(sbi, last_valid_block),
4756                             GET_BLKOFF_FROM_SEG0(sbi, last_valid_block),
4757                             wp_segno, wp_blkoff);
4758                 return 0;
4759         }
4760
4761         /*
4762          * If there is no valid block in the zone and if write pointer is
4763          * not at zone start, reset the write pointer.
4764          */
4765         if (last_valid_block + 1 == zone_block && zone->wp != zone->start) {
4766                 f2fs_notice(sbi,
4767                             "Zone without valid block has non-zero write "
4768                             "pointer. Reset the write pointer: wp[0x%x,0x%x]",
4769                             wp_segno, wp_blkoff);
4770                 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
4771                                         zone->len >> log_sectors_per_block);
4772                 if (ret) {
4773                         f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4774                                  fdev->path, ret);
4775                         return ret;
4776                 }
4777         }
4778
4779         return 0;
4780 }
4781
4782 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
4783                                                   block_t zone_blkaddr)
4784 {
4785         int i;
4786
4787         for (i = 0; i < sbi->s_ndevs; i++) {
4788                 if (!bdev_is_zoned(FDEV(i).bdev))
4789                         continue;
4790                 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
4791                                 zone_blkaddr <= FDEV(i).end_blk))
4792                         return &FDEV(i);
4793         }
4794
4795         return NULL;
4796 }
4797
4798 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
4799                               void *data)
4800 {
4801         memcpy(data, zone, sizeof(struct blk_zone));
4802         return 0;
4803 }
4804
4805 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
4806 {
4807         struct curseg_info *cs = CURSEG_I(sbi, type);
4808         struct f2fs_dev_info *zbd;
4809         struct blk_zone zone;
4810         unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
4811         block_t cs_zone_block, wp_block;
4812         unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4813         sector_t zone_sector;
4814         int err;
4815
4816         cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4817         cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4818
4819         zbd = get_target_zoned_dev(sbi, cs_zone_block);
4820         if (!zbd)
4821                 return 0;
4822
4823         /* report zone for the sector the curseg points to */
4824         zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4825                 << log_sectors_per_block;
4826         err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4827                                   report_one_zone_cb, &zone);
4828         if (err != 1) {
4829                 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4830                          zbd->path, err);
4831                 return err;
4832         }
4833
4834         if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4835                 return 0;
4836
4837         wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
4838         wp_segno = GET_SEGNO(sbi, wp_block);
4839         wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4840         wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
4841
4842         if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
4843                 wp_sector_off == 0)
4844                 return 0;
4845
4846         f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
4847                     "curseg[0x%x,0x%x] wp[0x%x,0x%x]",
4848                     type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff);
4849
4850         f2fs_notice(sbi, "Assign new section to curseg[%d]: "
4851                     "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff);
4852
4853         f2fs_allocate_new_section(sbi, type, true);
4854
4855         /* check consistency of the zone curseg pointed to */
4856         if (check_zone_write_pointer(sbi, zbd, &zone))
4857                 return -EIO;
4858
4859         /* check newly assigned zone */
4860         cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4861         cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4862
4863         zbd = get_target_zoned_dev(sbi, cs_zone_block);
4864         if (!zbd)
4865                 return 0;
4866
4867         zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4868                 << log_sectors_per_block;
4869         err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4870                                   report_one_zone_cb, &zone);
4871         if (err != 1) {
4872                 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4873                          zbd->path, err);
4874                 return err;
4875         }
4876
4877         if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4878                 return 0;
4879
4880         if (zone.wp != zone.start) {
4881                 f2fs_notice(sbi,
4882                             "New zone for curseg[%d] is not yet discarded. "
4883                             "Reset the zone: curseg[0x%x,0x%x]",
4884                             type, cs->segno, cs->next_blkoff);
4885                 err = __f2fs_issue_discard_zone(sbi, zbd->bdev,
4886                                 zone_sector >> log_sectors_per_block,
4887                                 zone.len >> log_sectors_per_block);
4888                 if (err) {
4889                         f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4890                                  zbd->path, err);
4891                         return err;
4892                 }
4893         }
4894
4895         return 0;
4896 }
4897
4898 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4899 {
4900         int i, ret;
4901
4902         for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4903                 ret = fix_curseg_write_pointer(sbi, i);
4904                 if (ret)
4905                         return ret;
4906         }
4907
4908         return 0;
4909 }
4910
4911 struct check_zone_write_pointer_args {
4912         struct f2fs_sb_info *sbi;
4913         struct f2fs_dev_info *fdev;
4914 };
4915
4916 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
4917                                       void *data)
4918 {
4919         struct check_zone_write_pointer_args *args;
4920
4921         args = (struct check_zone_write_pointer_args *)data;
4922
4923         return check_zone_write_pointer(args->sbi, args->fdev, zone);
4924 }
4925
4926 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4927 {
4928         int i, ret;
4929         struct check_zone_write_pointer_args args;
4930
4931         for (i = 0; i < sbi->s_ndevs; i++) {
4932                 if (!bdev_is_zoned(FDEV(i).bdev))
4933                         continue;
4934
4935                 args.sbi = sbi;
4936                 args.fdev = &FDEV(i);
4937                 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
4938                                           check_zone_write_pointer_cb, &args);
4939                 if (ret < 0)
4940                         return ret;
4941         }
4942
4943         return 0;
4944 }
4945
4946 static bool is_conv_zone(struct f2fs_sb_info *sbi, unsigned int zone_idx,
4947                                                 unsigned int dev_idx)
4948 {
4949         if (!bdev_is_zoned(FDEV(dev_idx).bdev))
4950                 return true;
4951         return !test_bit(zone_idx, FDEV(dev_idx).blkz_seq);
4952 }
4953
4954 /* Return the zone index in the given device */
4955 static unsigned int get_zone_idx(struct f2fs_sb_info *sbi, unsigned int secno,
4956                                         int dev_idx)
4957 {
4958         block_t sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
4959
4960         return (sec_start_blkaddr - FDEV(dev_idx).start_blk) >>
4961                                                 sbi->log_blocks_per_blkz;
4962 }
4963
4964 /*
4965  * Return the usable segments in a section based on the zone's
4966  * corresponding zone capacity. Zone is equal to a section.
4967  */
4968 static inline unsigned int f2fs_usable_zone_segs_in_sec(
4969                 struct f2fs_sb_info *sbi, unsigned int segno)
4970 {
4971         unsigned int dev_idx, zone_idx;
4972
4973         dev_idx = f2fs_target_device_index(sbi, START_BLOCK(sbi, segno));
4974         zone_idx = get_zone_idx(sbi, GET_SEC_FROM_SEG(sbi, segno), dev_idx);
4975
4976         /* Conventional zone's capacity is always equal to zone size */
4977         if (is_conv_zone(sbi, zone_idx, dev_idx))
4978                 return sbi->segs_per_sec;
4979
4980         if (!sbi->unusable_blocks_per_sec)
4981                 return sbi->segs_per_sec;
4982
4983         /* Get the segment count beyond zone capacity block */
4984         return sbi->segs_per_sec - (sbi->unusable_blocks_per_sec >>
4985                                                 sbi->log_blocks_per_seg);
4986 }
4987
4988 /*
4989  * Return the number of usable blocks in a segment. The number of blocks
4990  * returned is always equal to the number of blocks in a segment for
4991  * segments fully contained within a sequential zone capacity or a
4992  * conventional zone. For segments partially contained in a sequential
4993  * zone capacity, the number of usable blocks up to the zone capacity
4994  * is returned. 0 is returned in all other cases.
4995  */
4996 static inline unsigned int f2fs_usable_zone_blks_in_seg(
4997                         struct f2fs_sb_info *sbi, unsigned int segno)
4998 {
4999         block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
5000         unsigned int zone_idx, dev_idx, secno;
5001
5002         secno = GET_SEC_FROM_SEG(sbi, segno);
5003         seg_start = START_BLOCK(sbi, segno);
5004         dev_idx = f2fs_target_device_index(sbi, seg_start);
5005         zone_idx = get_zone_idx(sbi, secno, dev_idx);
5006
5007         /*
5008          * Conventional zone's capacity is always equal to zone size,
5009          * so, blocks per segment is unchanged.
5010          */
5011         if (is_conv_zone(sbi, zone_idx, dev_idx))
5012                 return sbi->blocks_per_seg;
5013
5014         if (!sbi->unusable_blocks_per_sec)
5015                 return sbi->blocks_per_seg;
5016
5017         sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
5018         sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi);
5019
5020         /*
5021          * If segment starts before zone capacity and spans beyond
5022          * zone capacity, then usable blocks are from seg start to
5023          * zone capacity. If the segment starts after the zone capacity,
5024          * then there are no usable blocks.
5025          */
5026         if (seg_start >= sec_cap_blkaddr)
5027                 return 0;
5028         if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr)
5029                 return sec_cap_blkaddr - seg_start;
5030
5031         return sbi->blocks_per_seg;
5032 }
5033 #else
5034 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5035 {
5036         return 0;
5037 }
5038
5039 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5040 {
5041         return 0;
5042 }
5043
5044 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
5045                                                         unsigned int segno)
5046 {
5047         return 0;
5048 }
5049
5050 static inline unsigned int f2fs_usable_zone_segs_in_sec(struct f2fs_sb_info *sbi,
5051                                                         unsigned int segno)
5052 {
5053         return 0;
5054 }
5055 #endif
5056 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
5057                                         unsigned int segno)
5058 {
5059         if (f2fs_sb_has_blkzoned(sbi))
5060                 return f2fs_usable_zone_blks_in_seg(sbi, segno);
5061
5062         return sbi->blocks_per_seg;
5063 }
5064
5065 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi,
5066                                         unsigned int segno)
5067 {
5068         if (f2fs_sb_has_blkzoned(sbi))
5069                 return f2fs_usable_zone_segs_in_sec(sbi, segno);
5070
5071         return sbi->segs_per_sec;
5072 }
5073
5074 /*
5075  * Update min, max modified time for cost-benefit GC algorithm
5076  */
5077 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5078 {
5079         struct sit_info *sit_i = SIT_I(sbi);
5080         unsigned int segno;
5081
5082         down_write(&sit_i->sentry_lock);
5083
5084         sit_i->min_mtime = ULLONG_MAX;
5085
5086         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
5087                 unsigned int i;
5088                 unsigned long long mtime = 0;
5089
5090                 for (i = 0; i < sbi->segs_per_sec; i++)
5091                         mtime += get_seg_entry(sbi, segno + i)->mtime;
5092
5093                 mtime = div_u64(mtime, sbi->segs_per_sec);
5094
5095                 if (sit_i->min_mtime > mtime)
5096                         sit_i->min_mtime = mtime;
5097         }
5098         sit_i->max_mtime = get_mtime(sbi, false);
5099         sit_i->dirty_max_mtime = 0;
5100         up_write(&sit_i->sentry_lock);
5101 }
5102
5103 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
5104 {
5105         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5106         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
5107         struct f2fs_sm_info *sm_info;
5108         int err;
5109
5110         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
5111         if (!sm_info)
5112                 return -ENOMEM;
5113
5114         /* init sm info */
5115         sbi->sm_info = sm_info;
5116         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5117         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5118         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5119         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5120         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5121         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5122         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
5123         sm_info->rec_prefree_segments = sm_info->main_segments *
5124                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
5125         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5126                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5127
5128         if (!f2fs_lfs_mode(sbi))
5129                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
5130         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
5131         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
5132         sm_info->min_seq_blocks = sbi->blocks_per_seg;
5133         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
5134         sm_info->min_ssr_sections = reserved_sections(sbi);
5135
5136         INIT_LIST_HEAD(&sm_info->sit_entry_set);
5137
5138         init_f2fs_rwsem(&sm_info->curseg_lock);
5139
5140         err = f2fs_create_flush_cmd_control(sbi);
5141         if (err)
5142                 return err;
5143
5144         err = create_discard_cmd_control(sbi);
5145         if (err)
5146                 return err;
5147
5148         err = build_sit_info(sbi);
5149         if (err)
5150                 return err;
5151         err = build_free_segmap(sbi);
5152         if (err)
5153                 return err;
5154         err = build_curseg(sbi);
5155         if (err)
5156                 return err;
5157
5158         /* reinit free segmap based on SIT */
5159         err = build_sit_entries(sbi);
5160         if (err)
5161                 return err;
5162
5163         init_free_segmap(sbi);
5164         err = build_dirty_segmap(sbi);
5165         if (err)
5166                 return err;
5167
5168         err = sanity_check_curseg(sbi);
5169         if (err)
5170                 return err;
5171
5172         init_min_max_mtime(sbi);
5173         return 0;
5174 }
5175
5176 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5177                 enum dirty_type dirty_type)
5178 {
5179         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5180
5181         mutex_lock(&dirty_i->seglist_lock);
5182         kvfree(dirty_i->dirty_segmap[dirty_type]);
5183         dirty_i->nr_dirty[dirty_type] = 0;
5184         mutex_unlock(&dirty_i->seglist_lock);
5185 }
5186
5187 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5188 {
5189         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5190
5191         kvfree(dirty_i->pinned_secmap);
5192         kvfree(dirty_i->victim_secmap);
5193 }
5194
5195 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5196 {
5197         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5198         int i;
5199
5200         if (!dirty_i)
5201                 return;
5202
5203         /* discard pre-free/dirty segments list */
5204         for (i = 0; i < NR_DIRTY_TYPE; i++)
5205                 discard_dirty_segmap(sbi, i);
5206
5207         if (__is_large_section(sbi)) {
5208                 mutex_lock(&dirty_i->seglist_lock);
5209                 kvfree(dirty_i->dirty_secmap);
5210                 mutex_unlock(&dirty_i->seglist_lock);
5211         }
5212
5213         destroy_victim_secmap(sbi);
5214         SM_I(sbi)->dirty_info = NULL;
5215         kfree(dirty_i);
5216 }
5217
5218 static void destroy_curseg(struct f2fs_sb_info *sbi)
5219 {
5220         struct curseg_info *array = SM_I(sbi)->curseg_array;
5221         int i;
5222
5223         if (!array)
5224                 return;
5225         SM_I(sbi)->curseg_array = NULL;
5226         for (i = 0; i < NR_CURSEG_TYPE; i++) {
5227                 kfree(array[i].sum_blk);
5228                 kfree(array[i].journal);
5229         }
5230         kfree(array);
5231 }
5232
5233 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5234 {
5235         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5236
5237         if (!free_i)
5238                 return;
5239         SM_I(sbi)->free_info = NULL;
5240         kvfree(free_i->free_segmap);
5241         kvfree(free_i->free_secmap);
5242         kfree(free_i);
5243 }
5244
5245 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5246 {
5247         struct sit_info *sit_i = SIT_I(sbi);
5248
5249         if (!sit_i)
5250                 return;
5251
5252         if (sit_i->sentries)
5253                 kvfree(sit_i->bitmap);
5254         kfree(sit_i->tmp_map);
5255
5256         kvfree(sit_i->sentries);
5257         kvfree(sit_i->sec_entries);
5258         kvfree(sit_i->dirty_sentries_bitmap);
5259
5260         SM_I(sbi)->sit_info = NULL;
5261         kvfree(sit_i->sit_bitmap);
5262 #ifdef CONFIG_F2FS_CHECK_FS
5263         kvfree(sit_i->sit_bitmap_mir);
5264         kvfree(sit_i->invalid_segmap);
5265 #endif
5266         kfree(sit_i);
5267 }
5268
5269 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5270 {
5271         struct f2fs_sm_info *sm_info = SM_I(sbi);
5272
5273         if (!sm_info)
5274                 return;
5275         f2fs_destroy_flush_cmd_control(sbi, true);
5276         destroy_discard_cmd_control(sbi);
5277         destroy_dirty_segmap(sbi);
5278         destroy_curseg(sbi);
5279         destroy_free_segmap(sbi);
5280         destroy_sit_info(sbi);
5281         sbi->sm_info = NULL;
5282         kfree(sm_info);
5283 }
5284
5285 int __init f2fs_create_segment_manager_caches(void)
5286 {
5287         discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5288                         sizeof(struct discard_entry));
5289         if (!discard_entry_slab)
5290                 goto fail;
5291
5292         discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5293                         sizeof(struct discard_cmd));
5294         if (!discard_cmd_slab)
5295                 goto destroy_discard_entry;
5296
5297         sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5298                         sizeof(struct sit_entry_set));
5299         if (!sit_entry_set_slab)
5300                 goto destroy_discard_cmd;
5301
5302         revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry",
5303                         sizeof(struct revoke_entry));
5304         if (!revoke_entry_slab)
5305                 goto destroy_sit_entry_set;
5306         return 0;
5307
5308 destroy_sit_entry_set:
5309         kmem_cache_destroy(sit_entry_set_slab);
5310 destroy_discard_cmd:
5311         kmem_cache_destroy(discard_cmd_slab);
5312 destroy_discard_entry:
5313         kmem_cache_destroy(discard_entry_slab);
5314 fail:
5315         return -ENOMEM;
5316 }
5317
5318 void f2fs_destroy_segment_manager_caches(void)
5319 {
5320         kmem_cache_destroy(sit_entry_set_slab);
5321         kmem_cache_destroy(discard_cmd_slab);
5322         kmem_cache_destroy(discard_entry_slab);
5323         kmem_cache_destroy(revoke_entry_slab);
5324 }