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         if (clean)
 196                 truncate_inode_pages_final(inode->i_mapping);
 197         clear_inode_flag(fi->cow_inode, FI_COW_FILE);
 198         iput(fi->cow_inode);
 199         fi->cow_inode = NULL;
 200         release_atomic_write_cnt(inode);
 201         clear_inode_flag(inode, FI_ATOMIC_FILE);
 202         stat_dec_atomic_inode(inode);
 203 }
 204
 205 static int __replace_atomic_write_block(struct inode *inode, pgoff_t index,
 206                         block_t new_addr, block_t *old_addr, bool recover)
 207 {
 208         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 209         struct dnode_of_data dn;
 210         struct node_info ni;
 211         int err;
 212
 213 retry:
 214         set_new_dnode(&dn, inode, NULL, NULL, 0);
 215         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE_RA);
 216         if (err) {
 217                 if (err == -ENOMEM) {
 218                         f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
 219                         goto retry;
 220                 }
 221                 return err;
 222         }
 223
 224         err = f2fs_get_node_info(sbi, dn.nid, &ni, false);
 225         if (err) {
 226                 f2fs_put_dnode(&dn);
 227                 return err;
 228         }
 229
 230         if (recover) {
 231                 /* dn.data_blkaddr is always valid */
 232                 if (!__is_valid_data_blkaddr(new_addr)) {
 233                         if (new_addr == NULL_ADDR)
 234                                 dec_valid_block_count(sbi, inode, 1);
 235                         f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
 236                         f2fs_update_data_blkaddr(&dn, new_addr);
 237                 } else {
 238                         f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 239                                 new_addr, ni.version, true, true);
 240                 }
 241         } else {
 242                 blkcnt_t count = 1;
 243
 244                 *old_addr = dn.data_blkaddr;
 245                 f2fs_truncate_data_blocks_range(&dn, 1);
 246                 dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count);
 247                 inc_valid_block_count(sbi, inode, &count);
 248                 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr,
 249                                         ni.version, true, false);
 250         }
 251
 252         f2fs_put_dnode(&dn);
 253         return 0;
 254 }
 255
 256 static void __complete_revoke_list(struct inode *inode, struct list_head *head,
 257                                         bool revoke)
 258 {
 259         struct revoke_entry *cur, *tmp;
 260
 261         list_for_each_entry_safe(cur, tmp, head, list) {
 262                 if (revoke)
 263                         __replace_atomic_write_block(inode, cur->index,
 264                                                 cur->old_addr, NULL, true);
 265                 list_del(&cur->list);
 266                 kmem_cache_free(revoke_entry_slab, cur);
 267         }
 268 }
 269
 270 static int __f2fs_commit_atomic_write(struct inode *inode)
 271 {
 272         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 273         struct f2fs_inode_info *fi = F2FS_I(inode);
 274         struct inode *cow_inode = fi->cow_inode;
 275         struct revoke_entry *new;
 276         struct list_head revoke_list;
 277         block_t blkaddr;
 278         struct dnode_of_data dn;
 279         pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
 280         pgoff_t off = 0, blen, index;
 281         int ret = 0, i;
 282
 283         INIT_LIST_HEAD(&revoke_list);
 284
 285         while (len) {
 286                 blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len);
 287
 288                 set_new_dnode(&dn, cow_inode, NULL, NULL, 0);
 289                 ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA);
 290                 if (ret && ret != -ENOENT) {
 291                         goto out;
 292                 } else if (ret == -ENOENT) {
 293                         ret = 0;
 294                         if (dn.max_level == 0)
 295                                 goto out;
 296                         goto next;
 297                 }
 298
 299                 blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_page, cow_inode),
 300                                 len);
 301                 index = off;
 302                 for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) {
 303                         blkaddr = f2fs_data_blkaddr(&dn);
 304
 305                         if (!__is_valid_data_blkaddr(blkaddr)) {
 306                                 continue;
 307                         } else if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
 308                                         DATA_GENERIC_ENHANCE)) {
 309                                 f2fs_put_dnode(&dn);
 310                                 ret = -EFSCORRUPTED;
 311                                 f2fs_handle_error(sbi,
 312                                                 ERROR_INVALID_BLKADDR);
 313                                 goto out;
 314                         }
 315
 316                         new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS,
 317                                                         true, NULL);
 318
 319                         ret = __replace_atomic_write_block(inode, index, blkaddr,
 320                                                         &new->old_addr, false);
 321                         if (ret) {
 322                                 f2fs_put_dnode(&dn);
 323                                 kmem_cache_free(revoke_entry_slab, new);
 324                                 goto out;
 325                         }
 326
 327                         f2fs_update_data_blkaddr(&dn, NULL_ADDR);
 328                         new->index = index;
 329                         list_add_tail(&new->list, &revoke_list);
 330                 }
 331                 f2fs_put_dnode(&dn);
 332 next:
 333                 off += blen;
 334                 len -= blen;
 335         }
 336
 337 out:
 338         if (ret)
 339                 sbi->revoked_atomic_block += fi->atomic_write_cnt;
 340         else
 341                 sbi->committed_atomic_block += fi->atomic_write_cnt;
 342
 343         __complete_revoke_list(inode, &revoke_list, ret ? true : false);
 344
 345         return ret;
 346 }
 347
 348 int f2fs_commit_atomic_write(struct inode *inode)
 349 {
 350         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 351         struct f2fs_inode_info *fi = F2FS_I(inode);
 352         int err;
 353
 354         err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX);
 355         if (err)
 356                 return err;
 357
 358         f2fs_down_write(&fi->i_gc_rwsem[WRITE]);
 359         f2fs_lock_op(sbi);
 360
 361         err = __f2fs_commit_atomic_write(inode);
 362
 363         f2fs_unlock_op(sbi);
 364         f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
 365
 366         return err;
 367 }
 368
 369 /*
 370  * This function balances dirty node and dentry pages.
 371  * In addition, it controls garbage collection.
 372  */
 373 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 374 {
 375         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 376                 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
 377                 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT);
 378         }
 379
 380         /* balance_fs_bg is able to be pending */
 381         if (need && excess_cached_nats(sbi))
 382                 f2fs_balance_fs_bg(sbi, false);
 383
 384         if (!f2fs_is_checkpoint_ready(sbi))
 385                 return;
 386
 387         /*
 388          * We should do GC or end up with checkpoint, if there are so many dirty
 389          * dir/node pages without enough free segments.
 390          */
 391         if (has_not_enough_free_secs(sbi, 0, 0)) {
 392                 if (test_opt(sbi, GC_MERGE) && sbi->gc_thread &&
 393                                         sbi->gc_thread->f2fs_gc_task) {
 394                         DEFINE_WAIT(wait);
 395
 396                         prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait,
 397                                                 TASK_UNINTERRUPTIBLE);
 398                         wake_up(&sbi->gc_thread->gc_wait_queue_head);
 399                         io_schedule();
 400                         finish_wait(&sbi->gc_thread->fggc_wq, &wait);
 401                 } else {
 402                         struct f2fs_gc_control gc_control = {
 403                                 .victim_segno = NULL_SEGNO,
 404                                 .init_gc_type = BG_GC,
 405                                 .no_bg_gc = true,
 406                                 .should_migrate_blocks = false,
 407                                 .err_gc_skipped = false,
 408                                 .nr_free_secs = 1 };
 409                         f2fs_down_write(&sbi->gc_lock);
 410                         f2fs_gc(sbi, &gc_control);
 411                 }
 412         }
 413 }
 414
 415 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi)
 416 {
 417         int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2;
 418         unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS);
 419         unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA);
 420         unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES);
 421         unsigned int meta = get_pages(sbi, F2FS_DIRTY_META);
 422         unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA);
 423         unsigned int threshold = sbi->blocks_per_seg * factor *
 424                                         DEFAULT_DIRTY_THRESHOLD;
 425         unsigned int global_threshold = threshold * 3 / 2;
 426
 427         if (dents >= threshold || qdata >= threshold ||
 428                 nodes >= threshold || meta >= threshold ||
 429                 imeta >= threshold)
 430                 return true;
 431         return dents + qdata + nodes + meta + imeta >  global_threshold;
 432 }
 433
 434 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
 435 {
 436         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 437                 return;
 438
 439         /* try to shrink extent cache when there is no enough memory */
 440         if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
 441                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 442
 443         /* check the # of cached NAT entries */
 444         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
 445                 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 446
 447         if (!f2fs_available_free_memory(sbi, FREE_NIDS))
 448                 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
 449         else
 450                 f2fs_build_free_nids(sbi, false, false);
 451
 452         if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) ||
 453                 excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi))
 454                 goto do_sync;
 455
 456         /* there is background inflight IO or foreground operation recently */
 457         if (is_inflight_io(sbi, REQ_TIME) ||
 458                 (!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem)))
 459                 return;
 460
 461         /* exceed periodical checkpoint timeout threshold */
 462         if (f2fs_time_over(sbi, CP_TIME))
 463                 goto do_sync;
 464
 465         /* checkpoint is the only way to shrink partial cached entries */
 466         if (f2fs_available_free_memory(sbi, NAT_ENTRIES) &&
 467                 f2fs_available_free_memory(sbi, INO_ENTRIES))
 468                 return;
 469
 470 do_sync:
 471         if (test_opt(sbi, DATA_FLUSH) && from_bg) {
 472                 struct blk_plug plug;
 473
 474                 mutex_lock(&sbi->flush_lock);
 475
 476                 blk_start_plug(&plug);
 477                 f2fs_sync_dirty_inodes(sbi, FILE_INODE, false);
 478                 blk_finish_plug(&plug);
 479
 480                 mutex_unlock(&sbi->flush_lock);
 481         }
 482         f2fs_sync_fs(sbi->sb, 1);
 483         stat_inc_bg_cp_count(sbi->stat_info);
 484 }
 485
 486 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 487                                 struct block_device *bdev)
 488 {
 489         int ret = blkdev_issue_flush(bdev);
 490
 491         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 492                                 test_opt(sbi, FLUSH_MERGE), ret);
 493         return ret;
 494 }
 495
 496 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 497 {
 498         int ret = 0;
 499         int i;
 500
 501         if (!f2fs_is_multi_device(sbi))
 502                 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 503
 504         for (i = 0; i < sbi->s_ndevs; i++) {
 505                 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
 506                         continue;
 507                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 508                 if (ret)
 509                         break;
 510         }
 511         return ret;
 512 }
 513
 514 static int issue_flush_thread(void *data)
 515 {
 516         struct f2fs_sb_info *sbi = data;
 517         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 518         wait_queue_head_t *q = &fcc->flush_wait_queue;
 519 repeat:
 520         if (kthread_should_stop())
 521                 return 0;
 522
 523         if (!llist_empty(&fcc->issue_list)) {
 524                 struct flush_cmd *cmd, *next;
 525                 int ret;
 526
 527                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 528                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 529
 530                 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 531
 532                 ret = submit_flush_wait(sbi, cmd->ino);
 533                 atomic_inc(&fcc->issued_flush);
 534
 535                 llist_for_each_entry_safe(cmd, next,
 536                                           fcc->dispatch_list, llnode) {
 537                         cmd->ret = ret;
 538                         complete(&cmd->wait);
 539                 }
 540                 fcc->dispatch_list = NULL;
 541         }
 542
 543         wait_event_interruptible(*q,
 544                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 545         goto repeat;
 546 }
 547
 548 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 549 {
 550         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 551         struct flush_cmd cmd;
 552         int ret;
 553
 554         if (test_opt(sbi, NOBARRIER))
 555                 return 0;
 556
 557         if (!test_opt(sbi, FLUSH_MERGE)) {
 558                 atomic_inc(&fcc->queued_flush);
 559                 ret = submit_flush_wait(sbi, ino);
 560                 atomic_dec(&fcc->queued_flush);
 561                 atomic_inc(&fcc->issued_flush);
 562                 return ret;
 563         }
 564
 565         if (atomic_inc_return(&fcc->queued_flush) == 1 ||
 566             f2fs_is_multi_device(sbi)) {
 567                 ret = submit_flush_wait(sbi, ino);
 568                 atomic_dec(&fcc->queued_flush);
 569
 570                 atomic_inc(&fcc->issued_flush);
 571                 return ret;
 572         }
 573
 574         cmd.ino = ino;
 575         init_completion(&cmd.wait);
 576
 577         llist_add(&cmd.llnode, &fcc->issue_list);
 578
 579         /*
 580          * update issue_list before we wake up issue_flush thread, this
 581          * smp_mb() pairs with another barrier in ___wait_event(), see
 582          * more details in comments of waitqueue_active().
 583          */
 584         smp_mb();
 585
 586         if (waitqueue_active(&fcc->flush_wait_queue))
 587                 wake_up(&fcc->flush_wait_queue);
 588
 589         if (fcc->f2fs_issue_flush) {
 590                 wait_for_completion(&cmd.wait);
 591                 atomic_dec(&fcc->queued_flush);
 592         } else {
 593                 struct llist_node *list;
 594
 595                 list = llist_del_all(&fcc->issue_list);
 596                 if (!list) {
 597                         wait_for_completion(&cmd.wait);
 598                         atomic_dec(&fcc->queued_flush);
 599                 } else {
 600                         struct flush_cmd *tmp, *next;
 601
 602                         ret = submit_flush_wait(sbi, ino);
 603
 604                         llist_for_each_entry_safe(tmp, next, list, llnode) {
 605                                 if (tmp == &cmd) {
 606                                         cmd.ret = ret;
 607                                         atomic_dec(&fcc->queued_flush);
 608                                         continue;
 609                                 }
 610                                 tmp->ret = ret;
 611                                 complete(&tmp->wait);
 612                         }
 613                 }
 614         }
 615
 616         return cmd.ret;
 617 }
 618
 619 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
 620 {
 621         dev_t dev = sbi->sb->s_bdev->bd_dev;
 622         struct flush_cmd_control *fcc;
 623         int err = 0;
 624
 625         if (SM_I(sbi)->fcc_info) {
 626                 fcc = SM_I(sbi)->fcc_info;
 627                 if (fcc->f2fs_issue_flush)
 628                         return err;
 629                 goto init_thread;
 630         }
 631
 632         fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
 633         if (!fcc)
 634                 return -ENOMEM;
 635         atomic_set(&fcc->issued_flush, 0);
 636         atomic_set(&fcc->queued_flush, 0);
 637         init_waitqueue_head(&fcc->flush_wait_queue);
 638         init_llist_head(&fcc->issue_list);
 639         SM_I(sbi)->fcc_info = fcc;
 640         if (!test_opt(sbi, FLUSH_MERGE))
 641                 return err;
 642
 643 init_thread:
 644         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 645                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 646         if (IS_ERR(fcc->f2fs_issue_flush)) {
 647                 err = PTR_ERR(fcc->f2fs_issue_flush);
 648                 kfree(fcc);
 649                 SM_I(sbi)->fcc_info = NULL;
 650                 return err;
 651         }
 652
 653         return err;
 654 }
 655
 656 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 657 {
 658         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 659
 660         if (fcc && fcc->f2fs_issue_flush) {
 661                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 662
 663                 fcc->f2fs_issue_flush = NULL;
 664                 kthread_stop(flush_thread);
 665         }
 666         if (free) {
 667                 kfree(fcc);
 668                 SM_I(sbi)->fcc_info = NULL;
 669         }
 670 }
 671
 672 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
 673 {
 674         int ret = 0, i;
 675
 676         if (!f2fs_is_multi_device(sbi))
 677                 return 0;
 678
 679         if (test_opt(sbi, NOBARRIER))
 680                 return 0;
 681
 682         for (i = 1; i < sbi->s_ndevs; i++) {
 683                 int count = DEFAULT_RETRY_IO_COUNT;
 684
 685                 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
 686                         continue;
 687
 688                 do {
 689                         ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 690                         if (ret)
 691                                 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
 692                 } while (ret && --count);
 693
 694                 if (ret) {
 695                         f2fs_stop_checkpoint(sbi, false,
 696                                         STOP_CP_REASON_FLUSH_FAIL);
 697                         break;
 698                 }
 699
 700                 spin_lock(&sbi->dev_lock);
 701                 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
 702                 spin_unlock(&sbi->dev_lock);
 703         }
 704
 705         return ret;
 706 }
 707
 708 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 709                 enum dirty_type dirty_type)
 710 {
 711         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 712
 713         /* need not be added */
 714         if (IS_CURSEG(sbi, segno))
 715                 return;
 716
 717         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 718                 dirty_i->nr_dirty[dirty_type]++;
 719
 720         if (dirty_type == DIRTY) {
 721                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 722                 enum dirty_type t = sentry->type;
 723
 724                 if (unlikely(t >= DIRTY)) {
 725                         f2fs_bug_on(sbi, 1);
 726                         return;
 727                 }
 728                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 729                         dirty_i->nr_dirty[t]++;
 730
 731                 if (__is_large_section(sbi)) {
 732                         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 733                         block_t valid_blocks =
 734                                 get_valid_blocks(sbi, segno, true);
 735
 736                         f2fs_bug_on(sbi, unlikely(!valid_blocks ||
 737                                         valid_blocks == CAP_BLKS_PER_SEC(sbi)));
 738
 739                         if (!IS_CURSEC(sbi, secno))
 740                                 set_bit(secno, dirty_i->dirty_secmap);
 741                 }
 742         }
 743 }
 744
 745 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 746                 enum dirty_type dirty_type)
 747 {
 748         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 749         block_t valid_blocks;
 750
 751         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 752                 dirty_i->nr_dirty[dirty_type]--;
 753
 754         if (dirty_type == DIRTY) {
 755                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 756                 enum dirty_type t = sentry->type;
 757
 758                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 759                         dirty_i->nr_dirty[t]--;
 760
 761                 valid_blocks = get_valid_blocks(sbi, segno, true);
 762                 if (valid_blocks == 0) {
 763                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 764                                                 dirty_i->victim_secmap);
 765 #ifdef CONFIG_F2FS_CHECK_FS
 766                         clear_bit(segno, SIT_I(sbi)->invalid_segmap);
 767 #endif
 768                 }
 769                 if (__is_large_section(sbi)) {
 770                         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 771
 772                         if (!valid_blocks ||
 773                                         valid_blocks == CAP_BLKS_PER_SEC(sbi)) {
 774                                 clear_bit(secno, dirty_i->dirty_secmap);
 775                                 return;
 776                         }
 777
 778                         if (!IS_CURSEC(sbi, secno))
 779                                 set_bit(secno, dirty_i->dirty_secmap);
 780                 }
 781         }
 782 }
 783
 784 /*
 785  * Should not occur error such as -ENOMEM.
 786  * Adding dirty entry into seglist is not critical operation.
 787  * If a given segment is one of current working segments, it won't be added.
 788  */
 789 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 790 {
 791         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 792         unsigned short valid_blocks, ckpt_valid_blocks;
 793         unsigned int usable_blocks;
 794
 795         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 796                 return;
 797
 798         usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
 799         mutex_lock(&dirty_i->seglist_lock);
 800
 801         valid_blocks = get_valid_blocks(sbi, segno, false);
 802         ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false);
 803
 804         if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
 805                 ckpt_valid_blocks == usable_blocks)) {
 806                 __locate_dirty_segment(sbi, segno, PRE);
 807                 __remove_dirty_segment(sbi, segno, DIRTY);
 808         } else if (valid_blocks < usable_blocks) {
 809                 __locate_dirty_segment(sbi, segno, DIRTY);
 810         } else {
 811                 /* Recovery routine with SSR needs this */
 812                 __remove_dirty_segment(sbi, segno, DIRTY);
 813         }
 814
 815         mutex_unlock(&dirty_i->seglist_lock);
 816 }
 817
 818 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
 819 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
 820 {
 821         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 822         unsigned int segno;
 823
 824         mutex_lock(&dirty_i->seglist_lock);
 825         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 826                 if (get_valid_blocks(sbi, segno, false))
 827                         continue;
 828                 if (IS_CURSEG(sbi, segno))
 829                         continue;
 830                 __locate_dirty_segment(sbi, segno, PRE);
 831                 __remove_dirty_segment(sbi, segno, DIRTY);
 832         }
 833         mutex_unlock(&dirty_i->seglist_lock);
 834 }
 835
 836 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
 837 {
 838         int ovp_hole_segs =
 839                 (overprovision_segments(sbi) - reserved_segments(sbi));
 840         block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
 841         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 842         block_t holes[2] = {0, 0};      /* DATA and NODE */
 843         block_t unusable;
 844         struct seg_entry *se;
 845         unsigned int segno;
 846
 847         mutex_lock(&dirty_i->seglist_lock);
 848         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 849                 se = get_seg_entry(sbi, segno);
 850                 if (IS_NODESEG(se->type))
 851                         holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
 852                                                         se->valid_blocks;
 853                 else
 854                         holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
 855                                                         se->valid_blocks;
 856         }
 857         mutex_unlock(&dirty_i->seglist_lock);
 858
 859         unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
 860         if (unusable > ovp_holes)
 861                 return unusable - ovp_holes;
 862         return 0;
 863 }
 864
 865 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
 866 {
 867         int ovp_hole_segs =
 868                 (overprovision_segments(sbi) - reserved_segments(sbi));
 869         if (unusable > F2FS_OPTION(sbi).unusable_cap)
 870                 return -EAGAIN;
 871         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
 872                 dirty_segments(sbi) > ovp_hole_segs)
 873                 return -EAGAIN;
 874         return 0;
 875 }
 876
 877 /* This is only used by SBI_CP_DISABLED */
 878 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
 879 {
 880         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 881         unsigned int segno = 0;
 882
 883         mutex_lock(&dirty_i->seglist_lock);
 884         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 885                 if (get_valid_blocks(sbi, segno, false))
 886                         continue;
 887                 if (get_ckpt_valid_blocks(sbi, segno, false))
 888                         continue;
 889                 mutex_unlock(&dirty_i->seglist_lock);
 890                 return segno;
 891         }
 892         mutex_unlock(&dirty_i->seglist_lock);
 893         return NULL_SEGNO;
 894 }
 895
 896 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 897                 struct block_device *bdev, block_t lstart,
 898                 block_t start, block_t len)
 899 {
 900         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 901         struct list_head *pend_list;
 902         struct discard_cmd *dc;
 903
 904         f2fs_bug_on(sbi, !len);
 905
 906         pend_list = &dcc->pend_list[plist_idx(len)];
 907
 908         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL);
 909         INIT_LIST_HEAD(&dc->list);
 910         dc->bdev = bdev;
 911         dc->lstart = lstart;
 912         dc->start = start;
 913         dc->len = len;
 914         dc->ref = 0;
 915         dc->state = D_PREP;
 916         dc->queued = 0;
 917         dc->error = 0;
 918         init_completion(&dc->wait);
 919         list_add_tail(&dc->list, pend_list);
 920         spin_lock_init(&dc->lock);
 921         dc->bio_ref = 0;
 922         atomic_inc(&dcc->discard_cmd_cnt);
 923         dcc->undiscard_blks += len;
 924
 925         return dc;
 926 }
 927
 928 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 929                                 struct block_device *bdev, block_t lstart,
 930                                 block_t start, block_t len,
 931                                 struct rb_node *parent, struct rb_node **p,
 932                                 bool leftmost)
 933 {
 934         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 935         struct discard_cmd *dc;
 936
 937         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
 938
 939         rb_link_node(&dc->rb_node, parent, p);
 940         rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
 941
 942         return dc;
 943 }
 944
 945 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
 946                                                         struct discard_cmd *dc)
 947 {
 948         if (dc->state == D_DONE)
 949                 atomic_sub(dc->queued, &dcc->queued_discard);
 950
 951         list_del(&dc->list);
 952         rb_erase_cached(&dc->rb_node, &dcc->root);
 953         dcc->undiscard_blks -= dc->len;
 954
 955         kmem_cache_free(discard_cmd_slab, dc);
 956
 957         atomic_dec(&dcc->discard_cmd_cnt);
 958 }
 959
 960 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
 961                                                         struct discard_cmd *dc)
 962 {
 963         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 964         unsigned long flags;
 965
 966         trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
 967
 968         spin_lock_irqsave(&dc->lock, flags);
 969         if (dc->bio_ref) {
 970                 spin_unlock_irqrestore(&dc->lock, flags);
 971                 return;
 972         }
 973         spin_unlock_irqrestore(&dc->lock, flags);
 974
 975         f2fs_bug_on(sbi, dc->ref);
 976
 977         if (dc->error == -EOPNOTSUPP)
 978                 dc->error = 0;
 979
 980         if (dc->error)
 981                 printk_ratelimited(
 982                         "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
 983                         KERN_INFO, sbi->sb->s_id,
 984                         dc->lstart, dc->start, dc->len, dc->error);
 985         __detach_discard_cmd(dcc, dc);
 986 }
 987
 988 static void f2fs_submit_discard_endio(struct bio *bio)
 989 {
 990         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
 991         unsigned long flags;
 992
 993         spin_lock_irqsave(&dc->lock, flags);
 994         if (!dc->error)
 995                 dc->error = blk_status_to_errno(bio->bi_status);
 996         dc->bio_ref--;
 997         if (!dc->bio_ref && dc->state == D_SUBMIT) {
 998                 dc->state = D_DONE;
 999                 complete_all(&dc->wait);
1000         }
1001         spin_unlock_irqrestore(&dc->lock, flags);
1002         bio_put(bio);
1003 }
1004
1005 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1006                                 block_t start, block_t end)
1007 {
1008 #ifdef CONFIG_F2FS_CHECK_FS
1009         struct seg_entry *sentry;
1010         unsigned int segno;
1011         block_t blk = start;
1012         unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1013         unsigned long *map;
1014
1015         while (blk < end) {
1016                 segno = GET_SEGNO(sbi, blk);
1017                 sentry = get_seg_entry(sbi, segno);
1018                 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1019
1020                 if (end < START_BLOCK(sbi, segno + 1))
1021                         size = GET_BLKOFF_FROM_SEG0(sbi, end);
1022                 else
1023                         size = max_blocks;
1024                 map = (unsigned long *)(sentry->cur_valid_map);
1025                 offset = __find_rev_next_bit(map, size, offset);
1026                 f2fs_bug_on(sbi, offset != size);
1027                 blk = START_BLOCK(sbi, segno + 1);
1028         }
1029 #endif
1030 }
1031
1032 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1033                                 struct discard_policy *dpolicy,
1034                                 int discard_type, unsigned int granularity)
1035 {
1036         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1037
1038         /* common policy */
1039         dpolicy->type = discard_type;
1040         dpolicy->sync = true;
1041         dpolicy->ordered = false;
1042         dpolicy->granularity = granularity;
1043
1044         dpolicy->max_requests = dcc->max_discard_request;
1045         dpolicy->io_aware_gran = MAX_PLIST_NUM;
1046         dpolicy->timeout = false;
1047
1048         if (discard_type == DPOLICY_BG) {
1049                 dpolicy->min_interval = dcc->min_discard_issue_time;
1050                 dpolicy->mid_interval = dcc->mid_discard_issue_time;
1051                 dpolicy->max_interval = dcc->max_discard_issue_time;
1052                 dpolicy->io_aware = true;
1053                 dpolicy->sync = false;
1054                 dpolicy->ordered = true;
1055                 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1056                         dpolicy->granularity = 1;
1057                         if (atomic_read(&dcc->discard_cmd_cnt))
1058                                 dpolicy->max_interval =
1059                                         dcc->min_discard_issue_time;
1060                 }
1061         } else if (discard_type == DPOLICY_FORCE) {
1062                 dpolicy->min_interval = dcc->min_discard_issue_time;
1063                 dpolicy->mid_interval = dcc->mid_discard_issue_time;
1064                 dpolicy->max_interval = dcc->max_discard_issue_time;
1065                 dpolicy->io_aware = false;
1066         } else if (discard_type == DPOLICY_FSTRIM) {
1067                 dpolicy->io_aware = false;
1068         } else if (discard_type == DPOLICY_UMOUNT) {
1069                 dpolicy->io_aware = false;
1070                 /* we need to issue all to keep CP_TRIMMED_FLAG */
1071                 dpolicy->granularity = 1;
1072                 dpolicy->timeout = true;
1073         }
1074 }
1075
1076 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1077                                 struct block_device *bdev, block_t lstart,
1078                                 block_t start, block_t len);
1079 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1080 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1081                                                 struct discard_policy *dpolicy,
1082                                                 struct discard_cmd *dc,
1083                                                 unsigned int *issued)
1084 {
1085         struct block_device *bdev = dc->bdev;
1086         unsigned int max_discard_blocks =
1087                         SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1088         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1089         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1090                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1091         blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0;
1092         block_t lstart, start, len, total_len;
1093         int err = 0;
1094
1095         if (dc->state != D_PREP)
1096                 return 0;
1097
1098         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1099                 return 0;
1100
1101         trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1102
1103         lstart = dc->lstart;
1104         start = dc->start;
1105         len = dc->len;
1106         total_len = len;
1107
1108         dc->len = 0;
1109
1110         while (total_len && *issued < dpolicy->max_requests && !err) {
1111                 struct bio *bio = NULL;
1112                 unsigned long flags;
1113                 bool last = true;
1114
1115                 if (len > max_discard_blocks) {
1116                         len = max_discard_blocks;
1117                         last = false;
1118                 }
1119
1120                 (*issued)++;
1121                 if (*issued == dpolicy->max_requests)
1122                         last = true;
1123
1124                 dc->len += len;
1125
1126                 if (time_to_inject(sbi, FAULT_DISCARD)) {
1127                         f2fs_show_injection_info(sbi, FAULT_DISCARD);
1128                         err = -EIO;
1129                         goto submit;
1130                 }
1131                 err = __blkdev_issue_discard(bdev,
1132                                         SECTOR_FROM_BLOCK(start),
1133                                         SECTOR_FROM_BLOCK(len),
1134                                         GFP_NOFS, &bio);
1135 submit:
1136                 if (err) {
1137                         spin_lock_irqsave(&dc->lock, flags);
1138                         if (dc->state == D_PARTIAL)
1139                                 dc->state = D_SUBMIT;
1140                         spin_unlock_irqrestore(&dc->lock, flags);
1141
1142                         break;
1143                 }
1144
1145                 f2fs_bug_on(sbi, !bio);
1146
1147                 /*
1148                  * should keep before submission to avoid D_DONE
1149                  * right away
1150                  */
1151                 spin_lock_irqsave(&dc->lock, flags);
1152                 if (last)
1153                         dc->state = D_SUBMIT;
1154                 else
1155                         dc->state = D_PARTIAL;
1156                 dc->bio_ref++;
1157                 spin_unlock_irqrestore(&dc->lock, flags);
1158
1159                 atomic_inc(&dcc->queued_discard);
1160                 dc->queued++;
1161                 list_move_tail(&dc->list, wait_list);
1162
1163                 /* sanity check on discard range */
1164                 __check_sit_bitmap(sbi, lstart, lstart + len);
1165
1166                 bio->bi_private = dc;
1167                 bio->bi_end_io = f2fs_submit_discard_endio;
1168                 bio->bi_opf |= flag;
1169                 submit_bio(bio);
1170
1171                 atomic_inc(&dcc->issued_discard);
1172
1173                 f2fs_update_iostat(sbi, NULL, FS_DISCARD, 1);
1174
1175                 lstart += len;
1176                 start += len;
1177                 total_len -= len;
1178                 len = total_len;
1179         }
1180
1181         if (!err && len) {
1182                 dcc->undiscard_blks -= len;
1183                 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1184         }
1185         return err;
1186 }
1187
1188 static void __insert_discard_tree(struct f2fs_sb_info *sbi,
1189                                 struct block_device *bdev, block_t lstart,
1190                                 block_t start, block_t len,
1191                                 struct rb_node **insert_p,
1192                                 struct rb_node *insert_parent)
1193 {
1194         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1195         struct rb_node **p;
1196         struct rb_node *parent = NULL;
1197         bool leftmost = true;
1198
1199         if (insert_p && insert_parent) {
1200                 parent = insert_parent;
1201                 p = insert_p;
1202                 goto do_insert;
1203         }
1204
1205         p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1206                                                         lstart, &leftmost);
1207 do_insert:
1208         __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1209                                                                 p, leftmost);
1210 }
1211
1212 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1213                                                 struct discard_cmd *dc)
1214 {
1215         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1216 }
1217
1218 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1219                                 struct discard_cmd *dc, block_t blkaddr)
1220 {
1221         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1222         struct discard_info di = dc->di;
1223         bool modified = false;
1224
1225         if (dc->state == D_DONE || dc->len == 1) {
1226                 __remove_discard_cmd(sbi, dc);
1227                 return;
1228         }
1229
1230         dcc->undiscard_blks -= di.len;
1231
1232         if (blkaddr > di.lstart) {
1233                 dc->len = blkaddr - dc->lstart;
1234                 dcc->undiscard_blks += dc->len;
1235                 __relocate_discard_cmd(dcc, dc);
1236                 modified = true;
1237         }
1238
1239         if (blkaddr < di.lstart + di.len - 1) {
1240                 if (modified) {
1241                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1242                                         di.start + blkaddr + 1 - di.lstart,
1243                                         di.lstart + di.len - 1 - blkaddr,
1244                                         NULL, NULL);
1245                 } else {
1246                         dc->lstart++;
1247                         dc->len--;
1248                         dc->start++;
1249                         dcc->undiscard_blks += dc->len;
1250                         __relocate_discard_cmd(dcc, dc);
1251                 }
1252         }
1253 }
1254
1255 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1256                                 struct block_device *bdev, block_t lstart,
1257                                 block_t start, block_t len)
1258 {
1259         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1260         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1261         struct discard_cmd *dc;
1262         struct discard_info di = {0};
1263         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1264         unsigned int max_discard_blocks =
1265                         SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1266         block_t end = lstart + len;
1267
1268         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1269                                         NULL, lstart,
1270                                         (struct rb_entry **)&prev_dc,
1271                                         (struct rb_entry **)&next_dc,
1272                                         &insert_p, &insert_parent, true, NULL);
1273         if (dc)
1274                 prev_dc = dc;
1275
1276         if (!prev_dc) {
1277                 di.lstart = lstart;
1278                 di.len = next_dc ? next_dc->lstart - lstart : len;
1279                 di.len = min(di.len, len);
1280                 di.start = start;
1281         }
1282
1283         while (1) {
1284                 struct rb_node *node;
1285                 bool merged = false;
1286                 struct discard_cmd *tdc = NULL;
1287
1288                 if (prev_dc) {
1289                         di.lstart = prev_dc->lstart + prev_dc->len;
1290                         if (di.lstart < lstart)
1291                                 di.lstart = lstart;
1292                         if (di.lstart >= end)
1293                                 break;
1294
1295                         if (!next_dc || next_dc->lstart > end)
1296                                 di.len = end - di.lstart;
1297                         else
1298                                 di.len = next_dc->lstart - di.lstart;
1299                         di.start = start + di.lstart - lstart;
1300                 }
1301
1302                 if (!di.len)
1303                         goto next;
1304
1305                 if (prev_dc && prev_dc->state == D_PREP &&
1306                         prev_dc->bdev == bdev &&
1307                         __is_discard_back_mergeable(&di, &prev_dc->di,
1308                                                         max_discard_blocks)) {
1309                         prev_dc->di.len += di.len;
1310                         dcc->undiscard_blks += di.len;
1311                         __relocate_discard_cmd(dcc, prev_dc);
1312                         di = prev_dc->di;
1313                         tdc = prev_dc;
1314                         merged = true;
1315                 }
1316
1317                 if (next_dc && next_dc->state == D_PREP &&
1318                         next_dc->bdev == bdev &&
1319                         __is_discard_front_mergeable(&di, &next_dc->di,
1320                                                         max_discard_blocks)) {
1321                         next_dc->di.lstart = di.lstart;
1322                         next_dc->di.len += di.len;
1323                         next_dc->di.start = di.start;
1324                         dcc->undiscard_blks += di.len;
1325                         __relocate_discard_cmd(dcc, next_dc);
1326                         if (tdc)
1327                                 __remove_discard_cmd(sbi, tdc);
1328                         merged = true;
1329                 }
1330
1331                 if (!merged) {
1332                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1333                                                         di.len, NULL, NULL);
1334                 }
1335  next:
1336                 prev_dc = next_dc;
1337                 if (!prev_dc)
1338                         break;
1339
1340                 node = rb_next(&prev_dc->rb_node);
1341                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1342         }
1343 }
1344
1345 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1346                 struct block_device *bdev, block_t blkstart, block_t blklen)
1347 {
1348         block_t lblkstart = blkstart;
1349
1350         if (!f2fs_bdev_support_discard(bdev))
1351                 return 0;
1352
1353         trace_f2fs_queue_discard(bdev, blkstart, blklen);
1354
1355         if (f2fs_is_multi_device(sbi)) {
1356                 int devi = f2fs_target_device_index(sbi, blkstart);
1357
1358                 blkstart -= FDEV(devi).start_blk;
1359         }
1360         mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1361         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1362         mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1363         return 0;
1364 }
1365
1366 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1367                                         struct discard_policy *dpolicy)
1368 {
1369         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1370         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1371         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1372         struct discard_cmd *dc;
1373         struct blk_plug plug;
1374         unsigned int pos = dcc->next_pos;
1375         unsigned int issued = 0;
1376         bool io_interrupted = false;
1377
1378         mutex_lock(&dcc->cmd_lock);
1379         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1380                                         NULL, pos,
1381                                         (struct rb_entry **)&prev_dc,
1382                                         (struct rb_entry **)&next_dc,
1383                                         &insert_p, &insert_parent, true, NULL);
1384         if (!dc)
1385                 dc = next_dc;
1386
1387         blk_start_plug(&plug);
1388
1389         while (dc) {
1390                 struct rb_node *node;
1391                 int err = 0;
1392
1393                 if (dc->state != D_PREP)
1394                         goto next;
1395
1396                 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1397                         io_interrupted = true;
1398                         break;
1399                 }
1400
1401                 dcc->next_pos = dc->lstart + dc->len;
1402                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1403
1404                 if (issued >= dpolicy->max_requests)
1405                         break;
1406 next:
1407                 node = rb_next(&dc->rb_node);
1408                 if (err)
1409                         __remove_discard_cmd(sbi, dc);
1410                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1411         }
1412
1413         blk_finish_plug(&plug);
1414
1415         if (!dc)
1416                 dcc->next_pos = 0;
1417
1418         mutex_unlock(&dcc->cmd_lock);
1419
1420         if (!issued && io_interrupted)
1421                 issued = -1;
1422
1423         return issued;
1424 }
1425 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1426                                         struct discard_policy *dpolicy);
1427
1428 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1429                                         struct discard_policy *dpolicy)
1430 {
1431         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1432         struct list_head *pend_list;
1433         struct discard_cmd *dc, *tmp;
1434         struct blk_plug plug;
1435         int i, issued;
1436         bool io_interrupted = false;
1437
1438         if (dpolicy->timeout)
1439                 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1440
1441 retry:
1442         issued = 0;
1443         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1444                 if (dpolicy->timeout &&
1445                                 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1446                         break;
1447
1448                 if (i + 1 < dpolicy->granularity)
1449                         break;
1450
1451                 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1452                         return __issue_discard_cmd_orderly(sbi, dpolicy);
1453
1454                 pend_list = &dcc->pend_list[i];
1455
1456                 mutex_lock(&dcc->cmd_lock);
1457                 if (list_empty(pend_list))
1458                         goto next;
1459                 if (unlikely(dcc->rbtree_check))
1460                         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1461                                                         &dcc->root, false));
1462                 blk_start_plug(&plug);
1463                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1464                         f2fs_bug_on(sbi, dc->state != D_PREP);
1465
1466                         if (dpolicy->timeout &&
1467                                 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1468                                 break;
1469
1470                         if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1471                                                 !is_idle(sbi, DISCARD_TIME)) {
1472                                 io_interrupted = true;
1473                                 break;
1474                         }
1475
1476                         __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1477
1478                         if (issued >= dpolicy->max_requests)
1479                                 break;
1480                 }
1481                 blk_finish_plug(&plug);
1482 next:
1483                 mutex_unlock(&dcc->cmd_lock);
1484
1485                 if (issued >= dpolicy->max_requests || io_interrupted)
1486                         break;
1487         }
1488
1489         if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1490                 __wait_all_discard_cmd(sbi, dpolicy);
1491                 goto retry;
1492         }
1493
1494         if (!issued && io_interrupted)
1495                 issued = -1;
1496
1497         return issued;
1498 }
1499
1500 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1501 {
1502         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1503         struct list_head *pend_list;
1504         struct discard_cmd *dc, *tmp;
1505         int i;
1506         bool dropped = false;
1507
1508         mutex_lock(&dcc->cmd_lock);
1509         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1510                 pend_list = &dcc->pend_list[i];
1511                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1512                         f2fs_bug_on(sbi, dc->state != D_PREP);
1513                         __remove_discard_cmd(sbi, dc);
1514                         dropped = true;
1515                 }
1516         }
1517         mutex_unlock(&dcc->cmd_lock);
1518
1519         return dropped;
1520 }
1521
1522 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1523 {
1524         __drop_discard_cmd(sbi);
1525 }
1526
1527 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1528                                                         struct discard_cmd *dc)
1529 {
1530         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1531         unsigned int len = 0;
1532
1533         wait_for_completion_io(&dc->wait);
1534         mutex_lock(&dcc->cmd_lock);
1535         f2fs_bug_on(sbi, dc->state != D_DONE);
1536         dc->ref--;
1537         if (!dc->ref) {
1538                 if (!dc->error)
1539                         len = dc->len;
1540                 __remove_discard_cmd(sbi, dc);
1541         }
1542         mutex_unlock(&dcc->cmd_lock);
1543
1544         return len;
1545 }
1546
1547 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1548                                                 struct discard_policy *dpolicy,
1549                                                 block_t start, block_t end)
1550 {
1551         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1552         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1553                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1554         struct discard_cmd *dc = NULL, *iter, *tmp;
1555         unsigned int trimmed = 0;
1556
1557 next:
1558         dc = NULL;
1559
1560         mutex_lock(&dcc->cmd_lock);
1561         list_for_each_entry_safe(iter, tmp, wait_list, list) {
1562                 if (iter->lstart + iter->len <= start || end <= iter->lstart)
1563                         continue;
1564                 if (iter->len < dpolicy->granularity)
1565                         continue;
1566                 if (iter->state == D_DONE && !iter->ref) {
1567                         wait_for_completion_io(&iter->wait);
1568                         if (!iter->error)
1569                                 trimmed += iter->len;
1570                         __remove_discard_cmd(sbi, iter);
1571                 } else {
1572                         iter->ref++;
1573                         dc = iter;
1574                         break;
1575                 }
1576         }
1577         mutex_unlock(&dcc->cmd_lock);
1578
1579         if (dc) {
1580                 trimmed += __wait_one_discard_bio(sbi, dc);
1581                 goto next;
1582         }
1583
1584         return trimmed;
1585 }
1586
1587 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1588                                                 struct discard_policy *dpolicy)
1589 {
1590         struct discard_policy dp;
1591         unsigned int discard_blks;
1592
1593         if (dpolicy)
1594                 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1595
1596         /* wait all */
1597         __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1598         discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1599         __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1600         discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1601
1602         return discard_blks;
1603 }
1604
1605 /* This should be covered by global mutex, &sit_i->sentry_lock */
1606 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1607 {
1608         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1609         struct discard_cmd *dc;
1610         bool need_wait = false;
1611
1612         mutex_lock(&dcc->cmd_lock);
1613         dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1614                                                         NULL, blkaddr);
1615         if (dc) {
1616                 if (dc->state == D_PREP) {
1617                         __punch_discard_cmd(sbi, dc, blkaddr);
1618                 } else {
1619                         dc->ref++;
1620                         need_wait = true;
1621                 }
1622         }
1623         mutex_unlock(&dcc->cmd_lock);
1624
1625         if (need_wait)
1626                 __wait_one_discard_bio(sbi, dc);
1627 }
1628
1629 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1630 {
1631         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1632
1633         if (dcc && dcc->f2fs_issue_discard) {
1634                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1635
1636                 dcc->f2fs_issue_discard = NULL;
1637                 kthread_stop(discard_thread);
1638         }
1639 }
1640
1641 /* This comes from f2fs_put_super */
1642 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1643 {
1644         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1645         struct discard_policy dpolicy;
1646         bool dropped;
1647
1648         __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1649                                         dcc->discard_granularity);
1650         __issue_discard_cmd(sbi, &dpolicy);
1651         dropped = __drop_discard_cmd(sbi);
1652
1653         /* just to make sure there is no pending discard commands */
1654         __wait_all_discard_cmd(sbi, NULL);
1655
1656         f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1657         return dropped;
1658 }
1659
1660 static int issue_discard_thread(void *data)
1661 {
1662         struct f2fs_sb_info *sbi = data;
1663         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1664         wait_queue_head_t *q = &dcc->discard_wait_queue;
1665         struct discard_policy dpolicy;
1666         unsigned int wait_ms = dcc->min_discard_issue_time;
1667         int issued;
1668
1669         set_freezable();
1670
1671         do {
1672                 if (sbi->gc_mode == GC_URGENT_HIGH ||
1673                         !f2fs_available_free_memory(sbi, DISCARD_CACHE))
1674                         __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1675                 else
1676                         __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1677                                                 dcc->discard_granularity);
1678
1679                 if (!atomic_read(&dcc->discard_cmd_cnt))
1680                        wait_ms = dpolicy.max_interval;
1681
1682                 wait_event_interruptible_timeout(*q,
1683                                 kthread_should_stop() || freezing(current) ||
1684                                 dcc->discard_wake,
1685                                 msecs_to_jiffies(wait_ms));
1686
1687                 if (dcc->discard_wake)
1688                         dcc->discard_wake = 0;
1689
1690                 /* clean up pending candidates before going to sleep */
1691                 if (atomic_read(&dcc->queued_discard))
1692                         __wait_all_discard_cmd(sbi, NULL);
1693
1694                 if (try_to_freeze())
1695                         continue;
1696                 if (f2fs_readonly(sbi->sb))
1697                         continue;
1698                 if (kthread_should_stop())
1699                         return 0;
1700                 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1701                         wait_ms = dpolicy.max_interval;
1702                         continue;
1703                 }
1704                 if (!atomic_read(&dcc->discard_cmd_cnt))
1705                         continue;
1706
1707                 sb_start_intwrite(sbi->sb);
1708
1709                 issued = __issue_discard_cmd(sbi, &dpolicy);
1710                 if (issued > 0) {
1711                         __wait_all_discard_cmd(sbi, &dpolicy);
1712                         wait_ms = dpolicy.min_interval;
1713                 } else if (issued == -1) {
1714                         wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1715                         if (!wait_ms)
1716                                 wait_ms = dpolicy.mid_interval;
1717                 } else {
1718                         wait_ms = dpolicy.max_interval;
1719                 }
1720
1721                 sb_end_intwrite(sbi->sb);
1722
1723         } while (!kthread_should_stop());
1724         return 0;
1725 }
1726
1727 #ifdef CONFIG_BLK_DEV_ZONED
1728 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1729                 struct block_device *bdev, block_t blkstart, block_t blklen)
1730 {
1731         sector_t sector, nr_sects;
1732         block_t lblkstart = blkstart;
1733         int devi = 0;
1734
1735         if (f2fs_is_multi_device(sbi)) {
1736                 devi = f2fs_target_device_index(sbi, blkstart);
1737                 if (blkstart < FDEV(devi).start_blk ||
1738                     blkstart > FDEV(devi).end_blk) {
1739                         f2fs_err(sbi, "Invalid block %x", blkstart);
1740                         return -EIO;
1741                 }
1742                 blkstart -= FDEV(devi).start_blk;
1743         }
1744
1745         /* For sequential zones, reset the zone write pointer */
1746         if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1747                 sector = SECTOR_FROM_BLOCK(blkstart);
1748                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1749
1750                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1751                                 nr_sects != bdev_zone_sectors(bdev)) {
1752                         f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1753                                  devi, sbi->s_ndevs ? FDEV(devi).path : "",
1754                                  blkstart, blklen);
1755                         return -EIO;
1756                 }
1757                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1758                 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1759                                         sector, nr_sects, GFP_NOFS);
1760         }
1761
1762         /* For conventional zones, use regular discard if supported */
1763         return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1764 }
1765 #endif
1766
1767 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1768                 struct block_device *bdev, block_t blkstart, block_t blklen)
1769 {
1770 #ifdef CONFIG_BLK_DEV_ZONED
1771         if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1772                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1773 #endif
1774         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1775 }
1776
1777 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1778                                 block_t blkstart, block_t blklen)
1779 {
1780         sector_t start = blkstart, len = 0;
1781         struct block_device *bdev;
1782         struct seg_entry *se;
1783         unsigned int offset;
1784         block_t i;
1785         int err = 0;
1786
1787         bdev = f2fs_target_device(sbi, blkstart, NULL);
1788
1789         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1790                 if (i != start) {
1791                         struct block_device *bdev2 =
1792                                 f2fs_target_device(sbi, i, NULL);
1793
1794                         if (bdev2 != bdev) {
1795                                 err = __issue_discard_async(sbi, bdev,
1796                                                 start, len);
1797                                 if (err)
1798                                         return err;
1799                                 bdev = bdev2;
1800                                 start = i;
1801                                 len = 0;
1802                         }
1803                 }
1804
1805                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1806                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1807
1808                 if (f2fs_block_unit_discard(sbi) &&
1809                                 !f2fs_test_and_set_bit(offset, se->discard_map))
1810                         sbi->discard_blks--;
1811         }
1812
1813         if (len)
1814                 err = __issue_discard_async(sbi, bdev, start, len);
1815         return err;
1816 }
1817
1818 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1819                                                         bool check_only)
1820 {
1821         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1822         int max_blocks = sbi->blocks_per_seg;
1823         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1824         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1825         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1826         unsigned long *discard_map = (unsigned long *)se->discard_map;
1827         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1828         unsigned int start = 0, end = -1;
1829         bool force = (cpc->reason & CP_DISCARD);
1830         struct discard_entry *de = NULL;
1831         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1832         int i;
1833
1834         if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi) ||
1835                         !f2fs_block_unit_discard(sbi))
1836                 return false;
1837
1838         if (!force) {
1839                 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1840                         SM_I(sbi)->dcc_info->nr_discards >=
1841                                 SM_I(sbi)->dcc_info->max_discards)
1842                         return false;
1843         }
1844
1845         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1846         for (i = 0; i < entries; i++)
1847                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1848                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1849
1850         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1851                                 SM_I(sbi)->dcc_info->max_discards) {
1852                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1853                 if (start >= max_blocks)
1854                         break;
1855
1856                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1857                 if (force && start && end != max_blocks
1858                                         && (end - start) < cpc->trim_minlen)
1859                         continue;
1860
1861                 if (check_only)
1862                         return true;
1863
1864                 if (!de) {
1865                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1866                                                 GFP_F2FS_ZERO, true, NULL);
1867                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1868                         list_add_tail(&de->list, head);
1869                 }
1870
1871                 for (i = start; i < end; i++)
1872                         __set_bit_le(i, (void *)de->discard_map);
1873
1874                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1875         }
1876         return false;
1877 }
1878
1879 static void release_discard_addr(struct discard_entry *entry)
1880 {
1881         list_del(&entry->list);
1882         kmem_cache_free(discard_entry_slab, entry);
1883 }
1884
1885 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1886 {
1887         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1888         struct discard_entry *entry, *this;
1889
1890         /* drop caches */
1891         list_for_each_entry_safe(entry, this, head, list)
1892                 release_discard_addr(entry);
1893 }
1894
1895 /*
1896  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1897  */
1898 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1899 {
1900         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1901         unsigned int segno;
1902
1903         mutex_lock(&dirty_i->seglist_lock);
1904         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1905                 __set_test_and_free(sbi, segno, false);
1906         mutex_unlock(&dirty_i->seglist_lock);
1907 }
1908
1909 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1910                                                 struct cp_control *cpc)
1911 {
1912         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1913         struct list_head *head = &dcc->entry_list;
1914         struct discard_entry *entry, *this;
1915         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1916         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1917         unsigned int start = 0, end = -1;
1918         unsigned int secno, start_segno;
1919         bool force = (cpc->reason & CP_DISCARD);
1920         bool section_alignment = F2FS_OPTION(sbi).discard_unit ==
1921                                                 DISCARD_UNIT_SECTION;
1922
1923         if (f2fs_lfs_mode(sbi) && __is_large_section(sbi))
1924                 section_alignment = true;
1925
1926         mutex_lock(&dirty_i->seglist_lock);
1927
1928         while (1) {
1929                 int i;
1930
1931                 if (section_alignment && end != -1)
1932                         end--;
1933                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1934                 if (start >= MAIN_SEGS(sbi))
1935                         break;
1936                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1937                                                                 start + 1);
1938
1939                 if (section_alignment) {
1940                         start = rounddown(start, sbi->segs_per_sec);
1941                         end = roundup(end, sbi->segs_per_sec);
1942                 }
1943
1944                 for (i = start; i < end; i++) {
1945                         if (test_and_clear_bit(i, prefree_map))
1946                                 dirty_i->nr_dirty[PRE]--;
1947                 }
1948
1949                 if (!f2fs_realtime_discard_enable(sbi))
1950                         continue;
1951
1952                 if (force && start >= cpc->trim_start &&
1953                                         (end - 1) <= cpc->trim_end)
1954                                 continue;
1955
1956                 if (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi)) {
1957                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1958                                 (end - start) << sbi->log_blocks_per_seg);
1959                         continue;
1960                 }
1961 next:
1962                 secno = GET_SEC_FROM_SEG(sbi, start);
1963                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1964                 if (!IS_CURSEC(sbi, secno) &&
1965                         !get_valid_blocks(sbi, start, true))
1966                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1967                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1968
1969                 start = start_segno + sbi->segs_per_sec;
1970                 if (start < end)
1971                         goto next;
1972                 else
1973                         end = start - 1;
1974         }
1975         mutex_unlock(&dirty_i->seglist_lock);
1976
1977         if (!f2fs_block_unit_discard(sbi))
1978                 goto wakeup;
1979
1980         /* send small discards */
1981         list_for_each_entry_safe(entry, this, head, list) {
1982                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1983                 bool is_valid = test_bit_le(0, entry->discard_map);
1984
1985 find_next:
1986                 if (is_valid) {
1987                         next_pos = find_next_zero_bit_le(entry->discard_map,
1988                                         sbi->blocks_per_seg, cur_pos);
1989                         len = next_pos - cur_pos;
1990
1991                         if (f2fs_sb_has_blkzoned(sbi) ||
1992                             (force && len < cpc->trim_minlen))
1993                                 goto skip;
1994
1995                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1996                                                                         len);
1997                         total_len += len;
1998                 } else {
1999                         next_pos = find_next_bit_le(entry->discard_map,
2000                                         sbi->blocks_per_seg, cur_pos);
2001                 }
2002 skip:
2003                 cur_pos = next_pos;
2004                 is_valid = !is_valid;
2005
2006                 if (cur_pos < sbi->blocks_per_seg)
2007                         goto find_next;
2008
2009                 release_discard_addr(entry);
2010                 dcc->nr_discards -= total_len;
2011         }
2012
2013 wakeup:
2014         wake_up_discard_thread(sbi, false);
2015 }
2016
2017 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi)
2018 {
2019         dev_t dev = sbi->sb->s_bdev->bd_dev;
2020         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2021         int err = 0;
2022
2023         if (!f2fs_realtime_discard_enable(sbi))
2024                 return 0;
2025
2026         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2027                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2028         if (IS_ERR(dcc->f2fs_issue_discard))
2029                 err = PTR_ERR(dcc->f2fs_issue_discard);
2030
2031         return err;
2032 }
2033
2034 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2035 {
2036         struct discard_cmd_control *dcc;
2037         int err = 0, i;
2038
2039         if (SM_I(sbi)->dcc_info) {
2040                 dcc = SM_I(sbi)->dcc_info;
2041                 goto init_thread;
2042         }
2043
2044         dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2045         if (!dcc)
2046                 return -ENOMEM;
2047
2048         dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2049         if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT)
2050                 dcc->discard_granularity = sbi->blocks_per_seg;
2051         else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION)
2052                 dcc->discard_granularity = BLKS_PER_SEC(sbi);
2053
2054         INIT_LIST_HEAD(&dcc->entry_list);
2055         for (i = 0; i < MAX_PLIST_NUM; i++)
2056                 INIT_LIST_HEAD(&dcc->pend_list[i]);
2057         INIT_LIST_HEAD(&dcc->wait_list);
2058         INIT_LIST_HEAD(&dcc->fstrim_list);
2059         mutex_init(&dcc->cmd_lock);
2060         atomic_set(&dcc->issued_discard, 0);
2061         atomic_set(&dcc->queued_discard, 0);
2062         atomic_set(&dcc->discard_cmd_cnt, 0);
2063         dcc->nr_discards = 0;
2064         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2065         dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST;
2066         dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME;
2067         dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME;
2068         dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME;
2069         dcc->undiscard_blks = 0;
2070         dcc->next_pos = 0;
2071         dcc->root = RB_ROOT_CACHED;
2072         dcc->rbtree_check = false;
2073
2074         init_waitqueue_head(&dcc->discard_wait_queue);
2075         SM_I(sbi)->dcc_info = dcc;
2076 init_thread:
2077         err = f2fs_start_discard_thread(sbi);
2078         if (err) {
2079                 kfree(dcc);
2080                 SM_I(sbi)->dcc_info = NULL;
2081         }
2082
2083         return err;
2084 }
2085
2086 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2087 {
2088         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2089
2090         if (!dcc)
2091                 return;
2092
2093         f2fs_stop_discard_thread(sbi);
2094
2095         /*
2096          * Recovery can cache discard commands, so in error path of
2097          * fill_super(), it needs to give a chance to handle them.
2098          */
2099         if (unlikely(atomic_read(&dcc->discard_cmd_cnt)))
2100                 f2fs_issue_discard_timeout(sbi);
2101
2102         kfree(dcc);
2103         SM_I(sbi)->dcc_info = NULL;
2104 }
2105
2106 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2107 {
2108         struct sit_info *sit_i = SIT_I(sbi);
2109
2110         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2111                 sit_i->dirty_sentries++;
2112                 return false;
2113         }
2114
2115         return true;
2116 }
2117
2118 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2119                                         unsigned int segno, int modified)
2120 {
2121         struct seg_entry *se = get_seg_entry(sbi, segno);
2122
2123         se->type = type;
2124         if (modified)
2125                 __mark_sit_entry_dirty(sbi, segno);
2126 }
2127
2128 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2129                                                                 block_t blkaddr)
2130 {
2131         unsigned int segno = GET_SEGNO(sbi, blkaddr);
2132
2133         if (segno == NULL_SEGNO)
2134                 return 0;
2135         return get_seg_entry(sbi, segno)->mtime;
2136 }
2137
2138 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2139                                                 unsigned long long old_mtime)
2140 {
2141         struct seg_entry *se;
2142         unsigned int segno = GET_SEGNO(sbi, blkaddr);
2143         unsigned long long ctime = get_mtime(sbi, false);
2144         unsigned long long mtime = old_mtime ? old_mtime : ctime;
2145
2146         if (segno == NULL_SEGNO)
2147                 return;
2148
2149         se = get_seg_entry(sbi, segno);
2150
2151         if (!se->mtime)
2152                 se->mtime = mtime;
2153         else
2154                 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2155                                                 se->valid_blocks + 1);
2156
2157         if (ctime > SIT_I(sbi)->max_mtime)
2158                 SIT_I(sbi)->max_mtime = ctime;
2159 }
2160
2161 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2162 {
2163         struct seg_entry *se;
2164         unsigned int segno, offset;
2165         long int new_vblocks;
2166         bool exist;
2167 #ifdef CONFIG_F2FS_CHECK_FS
2168         bool mir_exist;
2169 #endif
2170
2171         segno = GET_SEGNO(sbi, blkaddr);
2172
2173         se = get_seg_entry(sbi, segno);
2174         new_vblocks = se->valid_blocks + del;
2175         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2176
2177         f2fs_bug_on(sbi, (new_vblocks < 0 ||
2178                         (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2179
2180         se->valid_blocks = new_vblocks;
2181
2182         /* Update valid block bitmap */
2183         if (del > 0) {
2184                 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2185 #ifdef CONFIG_F2FS_CHECK_FS
2186                 mir_exist = f2fs_test_and_set_bit(offset,
2187                                                 se->cur_valid_map_mir);
2188                 if (unlikely(exist != mir_exist)) {
2189                         f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2190                                  blkaddr, exist);
2191                         f2fs_bug_on(sbi, 1);
2192                 }
2193 #endif
2194                 if (unlikely(exist)) {
2195                         f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2196                                  blkaddr);
2197                         f2fs_bug_on(sbi, 1);
2198                         se->valid_blocks--;
2199                         del = 0;
2200                 }
2201
2202                 if (f2fs_block_unit_discard(sbi) &&
2203                                 !f2fs_test_and_set_bit(offset, se->discard_map))
2204                         sbi->discard_blks--;
2205
2206                 /*
2207                  * SSR should never reuse block which is checkpointed
2208                  * or newly invalidated.
2209                  */
2210                 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2211                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2212                                 se->ckpt_valid_blocks++;
2213                 }
2214         } else {
2215                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2216 #ifdef CONFIG_F2FS_CHECK_FS
2217                 mir_exist = f2fs_test_and_clear_bit(offset,
2218                                                 se->cur_valid_map_mir);
2219                 if (unlikely(exist != mir_exist)) {
2220                         f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2221                                  blkaddr, exist);
2222                         f2fs_bug_on(sbi, 1);
2223                 }
2224 #endif
2225                 if (unlikely(!exist)) {
2226                         f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2227                                  blkaddr);
2228                         f2fs_bug_on(sbi, 1);
2229                         se->valid_blocks++;
2230                         del = 0;
2231                 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2232                         /*
2233                          * If checkpoints are off, we must not reuse data that
2234                          * was used in the previous checkpoint. If it was used
2235                          * before, we must track that to know how much space we
2236                          * really have.
2237                          */
2238                         if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2239                                 spin_lock(&sbi->stat_lock);
2240                                 sbi->unusable_block_count++;
2241                                 spin_unlock(&sbi->stat_lock);
2242                         }
2243                 }
2244
2245                 if (f2fs_block_unit_discard(sbi) &&
2246                         f2fs_test_and_clear_bit(offset, se->discard_map))
2247                         sbi->discard_blks++;
2248         }
2249         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2250                 se->ckpt_valid_blocks += del;
2251
2252         __mark_sit_entry_dirty(sbi, segno);
2253
2254         /* update total number of valid blocks to be written in ckpt area */
2255         SIT_I(sbi)->written_valid_blocks += del;
2256
2257         if (__is_large_section(sbi))
2258                 get_sec_entry(sbi, segno)->valid_blocks += del;
2259 }
2260
2261 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2262 {
2263         unsigned int segno = GET_SEGNO(sbi, addr);
2264         struct sit_info *sit_i = SIT_I(sbi);
2265
2266         f2fs_bug_on(sbi, addr == NULL_ADDR);
2267         if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2268                 return;
2269
2270         invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2271         f2fs_invalidate_compress_page(sbi, addr);
2272
2273         /* add it into sit main buffer */
2274         down_write(&sit_i->sentry_lock);
2275
2276         update_segment_mtime(sbi, addr, 0);
2277         update_sit_entry(sbi, addr, -1);
2278
2279         /* add it into dirty seglist */
2280         locate_dirty_segment(sbi, segno);
2281
2282         up_write(&sit_i->sentry_lock);
2283 }
2284
2285 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2286 {
2287         struct sit_info *sit_i = SIT_I(sbi);
2288         unsigned int segno, offset;
2289         struct seg_entry *se;
2290         bool is_cp = false;
2291
2292         if (!__is_valid_data_blkaddr(blkaddr))
2293                 return true;
2294
2295         down_read(&sit_i->sentry_lock);
2296
2297         segno = GET_SEGNO(sbi, blkaddr);
2298         se = get_seg_entry(sbi, segno);
2299         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2300
2301         if (f2fs_test_bit(offset, se->ckpt_valid_map))
2302                 is_cp = true;
2303
2304         up_read(&sit_i->sentry_lock);
2305
2306         return is_cp;
2307 }
2308
2309 /*
2310  * This function should be resided under the curseg_mutex lock
2311  */
2312 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2313                                         struct f2fs_summary *sum)
2314 {
2315         struct curseg_info *curseg = CURSEG_I(sbi, type);
2316         void *addr = curseg->sum_blk;
2317
2318         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2319         memcpy(addr, sum, sizeof(struct f2fs_summary));
2320 }
2321
2322 /*
2323  * Calculate the number of current summary pages for writing
2324  */
2325 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2326 {
2327         int valid_sum_count = 0;
2328         int i, sum_in_page;
2329
2330         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2331                 if (sbi->ckpt->alloc_type[i] == SSR)
2332                         valid_sum_count += sbi->blocks_per_seg;
2333                 else {
2334                         if (for_ra)
2335                                 valid_sum_count += le16_to_cpu(
2336                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2337                         else
2338                                 valid_sum_count += curseg_blkoff(sbi, i);
2339                 }
2340         }
2341
2342         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2343                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2344         if (valid_sum_count <= sum_in_page)
2345                 return 1;
2346         else if ((valid_sum_count - sum_in_page) <=
2347                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2348                 return 2;
2349         return 3;
2350 }
2351
2352 /*
2353  * Caller should put this summary page
2354  */
2355 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2356 {
2357         if (unlikely(f2fs_cp_error(sbi)))
2358                 return ERR_PTR(-EIO);
2359         return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno));
2360 }
2361
2362 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2363                                         void *src, block_t blk_addr)
2364 {
2365         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2366
2367         memcpy(page_address(page), src, PAGE_SIZE);
2368         set_page_dirty(page);
2369         f2fs_put_page(page, 1);
2370 }
2371
2372 static void write_sum_page(struct f2fs_sb_info *sbi,
2373                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
2374 {
2375         f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2376 }
2377
2378 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2379                                                 int type, block_t blk_addr)
2380 {
2381         struct curseg_info *curseg = CURSEG_I(sbi, type);
2382         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2383         struct f2fs_summary_block *src = curseg->sum_blk;
2384         struct f2fs_summary_block *dst;
2385
2386         dst = (struct f2fs_summary_block *)page_address(page);
2387         memset(dst, 0, PAGE_SIZE);
2388
2389         mutex_lock(&curseg->curseg_mutex);
2390
2391         down_read(&curseg->journal_rwsem);
2392         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2393         up_read(&curseg->journal_rwsem);
2394
2395         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2396         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2397
2398         mutex_unlock(&curseg->curseg_mutex);
2399
2400         set_page_dirty(page);
2401         f2fs_put_page(page, 1);
2402 }
2403
2404 static int is_next_segment_free(struct f2fs_sb_info *sbi,
2405                                 struct curseg_info *curseg, int type)
2406 {
2407         unsigned int segno = curseg->segno + 1;
2408         struct free_segmap_info *free_i = FREE_I(sbi);
2409
2410         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2411                 return !test_bit(segno, free_i->free_segmap);
2412         return 0;
2413 }
2414
2415 /*
2416  * Find a new segment from the free segments bitmap to right order
2417  * This function should be returned with success, otherwise BUG
2418  */
2419 static void get_new_segment(struct f2fs_sb_info *sbi,
2420                         unsigned int *newseg, bool new_sec, int dir)
2421 {
2422         struct free_segmap_info *free_i = FREE_I(sbi);
2423         unsigned int segno, secno, zoneno;
2424         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2425         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2426         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2427         unsigned int left_start = hint;
2428         bool init = true;
2429         int go_left = 0;
2430         int i;
2431
2432         spin_lock(&free_i->segmap_lock);
2433
2434         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2435                 segno = find_next_zero_bit(free_i->free_segmap,
2436                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2437                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2438                         goto got_it;
2439         }
2440 find_other_zone:
2441         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2442         if (secno >= MAIN_SECS(sbi)) {
2443                 if (dir == ALLOC_RIGHT) {
2444                         secno = find_first_zero_bit(free_i->free_secmap,
2445                                                         MAIN_SECS(sbi));
2446                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2447                 } else {
2448                         go_left = 1;
2449                         left_start = hint - 1;
2450                 }
2451         }
2452         if (go_left == 0)
2453                 goto skip_left;
2454
2455         while (test_bit(left_start, free_i->free_secmap)) {
2456                 if (left_start > 0) {
2457                         left_start--;
2458                         continue;
2459                 }
2460                 left_start = find_first_zero_bit(free_i->free_secmap,
2461                                                         MAIN_SECS(sbi));
2462                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2463                 break;
2464         }
2465         secno = left_start;
2466 skip_left:
2467         segno = GET_SEG_FROM_SEC(sbi, secno);
2468         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2469
2470         /* give up on finding another zone */
2471         if (!init)
2472                 goto got_it;
2473         if (sbi->secs_per_zone == 1)
2474                 goto got_it;
2475         if (zoneno == old_zoneno)
2476                 goto got_it;
2477         if (dir == ALLOC_LEFT) {
2478                 if (!go_left && zoneno + 1 >= total_zones)
2479                         goto got_it;
2480                 if (go_left && zoneno == 0)
2481                         goto got_it;
2482         }
2483         for (i = 0; i < NR_CURSEG_TYPE; i++)
2484                 if (CURSEG_I(sbi, i)->zone == zoneno)
2485                         break;
2486
2487         if (i < NR_CURSEG_TYPE) {
2488                 /* zone is in user, try another */
2489                 if (go_left)
2490                         hint = zoneno * sbi->secs_per_zone - 1;
2491                 else if (zoneno + 1 >= total_zones)
2492                         hint = 0;
2493                 else
2494                         hint = (zoneno + 1) * sbi->secs_per_zone;
2495                 init = false;
2496                 goto find_other_zone;
2497         }
2498 got_it:
2499         /* set it as dirty segment in free segmap */
2500         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2501         __set_inuse(sbi, segno);
2502         *newseg = segno;
2503         spin_unlock(&free_i->segmap_lock);
2504 }
2505
2506 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2507 {
2508         struct curseg_info *curseg = CURSEG_I(sbi, type);
2509         struct summary_footer *sum_footer;
2510         unsigned short seg_type = curseg->seg_type;
2511
2512         curseg->inited = true;
2513         curseg->segno = curseg->next_segno;
2514         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2515         curseg->next_blkoff = 0;
2516         curseg->next_segno = NULL_SEGNO;
2517
2518         sum_footer = &(curseg->sum_blk->footer);
2519         memset(sum_footer, 0, sizeof(struct summary_footer));
2520
2521         sanity_check_seg_type(sbi, seg_type);
2522
2523         if (IS_DATASEG(seg_type))
2524                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2525         if (IS_NODESEG(seg_type))
2526                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2527         __set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
2528 }
2529
2530 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2531 {
2532         struct curseg_info *curseg = CURSEG_I(sbi, type);
2533         unsigned short seg_type = curseg->seg_type;
2534
2535         sanity_check_seg_type(sbi, seg_type);
2536         if (f2fs_need_rand_seg(sbi))
2537                 return prandom_u32_max(MAIN_SECS(sbi) * sbi->segs_per_sec);
2538
2539         /* if segs_per_sec is large than 1, we need to keep original policy. */
2540         if (__is_large_section(sbi))
2541                 return curseg->segno;
2542
2543         /* inmem log may not locate on any segment after mount */
2544         if (!curseg->inited)
2545                 return 0;
2546
2547         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2548                 return 0;
2549
2550         if (test_opt(sbi, NOHEAP) &&
2551                 (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type)))
2552                 return 0;
2553
2554         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2555                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2556
2557         /* find segments from 0 to reuse freed segments */
2558         if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2559                 return 0;
2560
2561         return curseg->segno;
2562 }
2563
2564 /*
2565  * Allocate a current working segment.
2566  * This function always allocates a free segment in LFS manner.
2567  */
2568 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2569 {
2570         struct curseg_info *curseg = CURSEG_I(sbi, type);
2571         unsigned short seg_type = curseg->seg_type;
2572         unsigned int segno = curseg->segno;
2573         int dir = ALLOC_LEFT;
2574
2575         if (curseg->inited)
2576                 write_sum_page(sbi, curseg->sum_blk,
2577                                 GET_SUM_BLOCK(sbi, segno));
2578         if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA)
2579                 dir = ALLOC_RIGHT;
2580
2581         if (test_opt(sbi, NOHEAP))
2582                 dir = ALLOC_RIGHT;
2583
2584         segno = __get_next_segno(sbi, type);
2585         get_new_segment(sbi, &segno, new_sec, dir);
2586         curseg->next_segno = segno;
2587         reset_curseg(sbi, type, 1);
2588         curseg->alloc_type = LFS;
2589         if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
2590                 curseg->fragment_remained_chunk =
2591                                 prandom_u32_max(sbi->max_fragment_chunk) + 1;
2592 }
2593
2594 static int __next_free_blkoff(struct f2fs_sb_info *sbi,
2595                                         int segno, block_t start)
2596 {
2597         struct seg_entry *se = get_seg_entry(sbi, segno);
2598         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2599         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2600         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2601         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2602         int i;
2603
2604         for (i = 0; i < entries; i++)
2605                 target_map[i] = ckpt_map[i] | cur_map[i];
2606
2607         return __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2608 }
2609
2610 /*
2611  * If a segment is written by LFS manner, next block offset is just obtained
2612  * by increasing the current block offset. However, if a segment is written by
2613  * SSR manner, next block offset obtained by calling __next_free_blkoff
2614  */
2615 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2616                                 struct curseg_info *seg)
2617 {
2618         if (seg->alloc_type == SSR) {
2619                 seg->next_blkoff =
2620                         __next_free_blkoff(sbi, seg->segno,
2621                                                 seg->next_blkoff + 1);
2622         } else {
2623                 seg->next_blkoff++;
2624                 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) {
2625                         /* To allocate block chunks in different sizes, use random number */
2626                         if (--seg->fragment_remained_chunk <= 0) {
2627                                 seg->fragment_remained_chunk =
2628                                    prandom_u32_max(sbi->max_fragment_chunk) + 1;
2629                                 seg->next_blkoff +=
2630                                    prandom_u32_max(sbi->max_fragment_hole) + 1;
2631                         }
2632                 }
2633         }
2634 }
2635
2636 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno)
2637 {
2638         return __next_free_blkoff(sbi, segno, 0) < sbi->blocks_per_seg;
2639 }
2640
2641 /*
2642  * This function always allocates a used segment(from dirty seglist) by SSR
2643  * manner, so it should recover the existing segment information of valid blocks
2644  */
2645 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool flush)
2646 {
2647         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2648         struct curseg_info *curseg = CURSEG_I(sbi, type);
2649         unsigned int new_segno = curseg->next_segno;
2650         struct f2fs_summary_block *sum_node;
2651         struct page *sum_page;
2652
2653         if (flush)
2654                 write_sum_page(sbi, curseg->sum_blk,
2655                                         GET_SUM_BLOCK(sbi, curseg->segno));
2656
2657         __set_test_and_inuse(sbi, new_segno);
2658
2659         mutex_lock(&dirty_i->seglist_lock);
2660         __remove_dirty_segment(sbi, new_segno, PRE);
2661         __remove_dirty_segment(sbi, new_segno, DIRTY);
2662         mutex_unlock(&dirty_i->seglist_lock);
2663
2664         reset_curseg(sbi, type, 1);
2665         curseg->alloc_type = SSR;
2666         curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0);
2667
2668         sum_page = f2fs_get_sum_page(sbi, new_segno);
2669         if (IS_ERR(sum_page)) {
2670                 /* GC won't be able to use stale summary pages by cp_error */
2671                 memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
2672                 return;
2673         }
2674         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2675         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2676         f2fs_put_page(sum_page, 1);
2677 }
2678
2679 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2680                                 int alloc_mode, unsigned long long age);
2681
2682 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type,
2683                                         int target_type, int alloc_mode,
2684                                         unsigned long long age)
2685 {
2686         struct curseg_info *curseg = CURSEG_I(sbi, type);
2687
2688         curseg->seg_type = target_type;
2689
2690         if (get_ssr_segment(sbi, type, alloc_mode, age)) {
2691                 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
2692
2693                 curseg->seg_type = se->type;
2694                 change_curseg(sbi, type, true);
2695         } else {
2696                 /* allocate cold segment by default */
2697                 curseg->seg_type = CURSEG_COLD_DATA;
2698                 new_curseg(sbi, type, true);
2699         }
2700         stat_inc_seg_type(sbi, curseg);
2701 }
2702
2703 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
2704 {
2705         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
2706
2707         if (!sbi->am.atgc_enabled)
2708                 return;
2709
2710         f2fs_down_read(&SM_I(sbi)->curseg_lock);
2711
2712         mutex_lock(&curseg->curseg_mutex);
2713         down_write(&SIT_I(sbi)->sentry_lock);
2714
2715         get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0);
2716
2717         up_write(&SIT_I(sbi)->sentry_lock);
2718         mutex_unlock(&curseg->curseg_mutex);
2719
2720         f2fs_up_read(&SM_I(sbi)->curseg_lock);
2721
2722 }
2723 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
2724 {
2725         __f2fs_init_atgc_curseg(sbi);
2726 }
2727
2728 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2729 {
2730         struct curseg_info *curseg = CURSEG_I(sbi, type);
2731
2732         mutex_lock(&curseg->curseg_mutex);
2733         if (!curseg->inited)
2734                 goto out;
2735
2736         if (get_valid_blocks(sbi, curseg->segno, false)) {
2737                 write_sum_page(sbi, curseg->sum_blk,
2738                                 GET_SUM_BLOCK(sbi, curseg->segno));
2739         } else {
2740                 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2741                 __set_test_and_free(sbi, curseg->segno, true);
2742                 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2743         }
2744 out:
2745         mutex_unlock(&curseg->curseg_mutex);
2746 }
2747
2748 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
2749 {
2750         __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2751
2752         if (sbi->am.atgc_enabled)
2753                 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2754 }
2755
2756 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2757 {
2758         struct curseg_info *curseg = CURSEG_I(sbi, type);
2759
2760         mutex_lock(&curseg->curseg_mutex);
2761         if (!curseg->inited)
2762                 goto out;
2763         if (get_valid_blocks(sbi, curseg->segno, false))
2764                 goto out;
2765
2766         mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2767         __set_test_and_inuse(sbi, curseg->segno);
2768         mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2769 out:
2770         mutex_unlock(&curseg->curseg_mutex);
2771 }
2772
2773 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
2774 {
2775         __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2776
2777         if (sbi->am.atgc_enabled)
2778                 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2779 }
2780
2781 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2782                                 int alloc_mode, unsigned long long age)
2783 {
2784         struct curseg_info *curseg = CURSEG_I(sbi, type);
2785         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2786         unsigned segno = NULL_SEGNO;
2787         unsigned short seg_type = curseg->seg_type;
2788         int i, cnt;
2789         bool reversed = false;
2790
2791         sanity_check_seg_type(sbi, seg_type);
2792
2793         /* f2fs_need_SSR() already forces to do this */
2794         if (!v_ops->get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) {
2795                 curseg->next_segno = segno;
2796                 return 1;
2797         }
2798
2799         /* For node segments, let's do SSR more intensively */
2800         if (IS_NODESEG(seg_type)) {
2801                 if (seg_type >= CURSEG_WARM_NODE) {
2802                         reversed = true;
2803                         i = CURSEG_COLD_NODE;
2804                 } else {
2805                         i = CURSEG_HOT_NODE;
2806                 }
2807                 cnt = NR_CURSEG_NODE_TYPE;
2808         } else {
2809                 if (seg_type >= CURSEG_WARM_DATA) {
2810                         reversed = true;
2811                         i = CURSEG_COLD_DATA;
2812                 } else {
2813                         i = CURSEG_HOT_DATA;
2814                 }
2815                 cnt = NR_CURSEG_DATA_TYPE;
2816         }
2817
2818         for (; cnt-- > 0; reversed ? i-- : i++) {
2819                 if (i == seg_type)
2820                         continue;
2821                 if (!v_ops->get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) {
2822                         curseg->next_segno = segno;
2823                         return 1;
2824                 }
2825         }
2826
2827         /* find valid_blocks=0 in dirty list */
2828         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2829                 segno = get_free_segment(sbi);
2830                 if (segno != NULL_SEGNO) {
2831                         curseg->next_segno = segno;
2832                         return 1;
2833                 }
2834         }
2835         return 0;
2836 }
2837
2838 /*
2839  * flush out current segment and replace it with new segment
2840  * This function should be returned with success, otherwise BUG
2841  */
2842 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2843                                                 int type, bool force)
2844 {
2845         struct curseg_info *curseg = CURSEG_I(sbi, type);
2846
2847         if (force)
2848                 new_curseg(sbi, type, true);
2849         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2850                                         curseg->seg_type == CURSEG_WARM_NODE)
2851                 new_curseg(sbi, type, false);
2852         else if (curseg->alloc_type == LFS &&
2853                         is_next_segment_free(sbi, curseg, type) &&
2854                         likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2855                 new_curseg(sbi, type, false);
2856         else if (f2fs_need_SSR(sbi) &&
2857                         get_ssr_segment(sbi, type, SSR, 0))
2858                 change_curseg(sbi, type, true);
2859         else
2860                 new_curseg(sbi, type, false);
2861
2862         stat_inc_seg_type(sbi, curseg);
2863 }
2864
2865 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2866                                         unsigned int start, unsigned int end)
2867 {
2868         struct curseg_info *curseg = CURSEG_I(sbi, type);
2869         unsigned int segno;
2870
2871         f2fs_down_read(&SM_I(sbi)->curseg_lock);
2872         mutex_lock(&curseg->curseg_mutex);
2873         down_write(&SIT_I(sbi)->sentry_lock);
2874
2875         segno = CURSEG_I(sbi, type)->segno;
2876         if (segno < start || segno > end)
2877                 goto unlock;
2878
2879         if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
2880                 change_curseg(sbi, type, true);
2881         else
2882                 new_curseg(sbi, type, true);
2883
2884         stat_inc_seg_type(sbi, curseg);
2885
2886         locate_dirty_segment(sbi, segno);
2887 unlock:
2888         up_write(&SIT_I(sbi)->sentry_lock);
2889
2890         if (segno != curseg->segno)
2891                 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2892                             type, segno, curseg->segno);
2893
2894         mutex_unlock(&curseg->curseg_mutex);
2895         f2fs_up_read(&SM_I(sbi)->curseg_lock);
2896 }
2897
2898 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type,
2899                                                 bool new_sec, bool force)
2900 {
2901         struct curseg_info *curseg = CURSEG_I(sbi, type);
2902         unsigned int old_segno;
2903
2904         if (!curseg->inited)
2905                 goto alloc;
2906
2907         if (force || curseg->next_blkoff ||
2908                 get_valid_blocks(sbi, curseg->segno, new_sec))
2909                 goto alloc;
2910
2911         if (!get_ckpt_valid_blocks(sbi, curseg->segno, new_sec))
2912                 return;
2913 alloc:
2914         old_segno = curseg->segno;
2915         SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
2916         locate_dirty_segment(sbi, old_segno);
2917 }
2918
2919 static void __allocate_new_section(struct f2fs_sb_info *sbi,
2920                                                 int type, bool force)
2921 {
2922         __allocate_new_segment(sbi, type, true, force);
2923 }
2924
2925 void f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force)
2926 {
2927         f2fs_down_read(&SM_I(sbi)->curseg_lock);
2928         down_write(&SIT_I(sbi)->sentry_lock);
2929         __allocate_new_section(sbi, type, force);
2930         up_write(&SIT_I(sbi)->sentry_lock);
2931         f2fs_up_read(&SM_I(sbi)->curseg_lock);
2932 }
2933
2934 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2935 {
2936         int i;
2937
2938         f2fs_down_read(&SM_I(sbi)->curseg_lock);
2939         down_write(&SIT_I(sbi)->sentry_lock);
2940         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
2941                 __allocate_new_segment(sbi, i, false, false);
2942         up_write(&SIT_I(sbi)->sentry_lock);
2943         f2fs_up_read(&SM_I(sbi)->curseg_lock);
2944 }
2945
2946 static const struct segment_allocation default_salloc_ops = {
2947         .allocate_segment = allocate_segment_by_default,
2948 };
2949
2950 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2951                                                 struct cp_control *cpc)
2952 {
2953         __u64 trim_start = cpc->trim_start;
2954         bool has_candidate = false;
2955
2956         down_write(&SIT_I(sbi)->sentry_lock);
2957         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2958                 if (add_discard_addrs(sbi, cpc, true)) {
2959                         has_candidate = true;
2960                         break;
2961                 }
2962         }
2963         up_write(&SIT_I(sbi)->sentry_lock);
2964
2965         cpc->trim_start = trim_start;
2966         return has_candidate;
2967 }
2968
2969 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2970                                         struct discard_policy *dpolicy,
2971                                         unsigned int start, unsigned int end)
2972 {
2973         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2974         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2975         struct rb_node **insert_p = NULL, *insert_parent = NULL;
2976         struct discard_cmd *dc;
2977         struct blk_plug plug;
2978         int issued;
2979         unsigned int trimmed = 0;
2980
2981 next:
2982         issued = 0;
2983
2984         mutex_lock(&dcc->cmd_lock);
2985         if (unlikely(dcc->rbtree_check))
2986                 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2987                                                         &dcc->root, false));
2988
2989         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2990                                         NULL, start,
2991                                         (struct rb_entry **)&prev_dc,
2992                                         (struct rb_entry **)&next_dc,
2993                                         &insert_p, &insert_parent, true, NULL);
2994         if (!dc)
2995                 dc = next_dc;
2996
2997         blk_start_plug(&plug);
2998
2999         while (dc && dc->lstart <= end) {
3000                 struct rb_node *node;
3001                 int err = 0;
3002
3003                 if (dc->len < dpolicy->granularity)
3004                         goto skip;
3005
3006                 if (dc->state != D_PREP) {
3007                         list_move_tail(&dc->list, &dcc->fstrim_list);
3008                         goto skip;
3009                 }
3010
3011                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
3012
3013                 if (issued >= dpolicy->max_requests) {
3014                         start = dc->lstart + dc->len;
3015
3016                         if (err)
3017                                 __remove_discard_cmd(sbi, dc);
3018
3019                         blk_finish_plug(&plug);
3020                         mutex_unlock(&dcc->cmd_lock);
3021                         trimmed += __wait_all_discard_cmd(sbi, NULL);
3022                         f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
3023                         goto next;
3024                 }
3025 skip:
3026                 node = rb_next(&dc->rb_node);
3027                 if (err)
3028                         __remove_discard_cmd(sbi, dc);
3029                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3030
3031                 if (fatal_signal_pending(current))
3032                         break;
3033         }
3034
3035         blk_finish_plug(&plug);
3036         mutex_unlock(&dcc->cmd_lock);
3037
3038         return trimmed;
3039 }
3040
3041 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3042 {
3043         __u64 start = F2FS_BYTES_TO_BLK(range->start);
3044         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
3045         unsigned int start_segno, end_segno;
3046         block_t start_block, end_block;
3047         struct cp_control cpc;
3048         struct discard_policy dpolicy;
3049         unsigned long long trimmed = 0;
3050         int err = 0;
3051         bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
3052
3053         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
3054                 return -EINVAL;
3055
3056         if (end < MAIN_BLKADDR(sbi))
3057                 goto out;
3058
3059         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
3060                 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
3061                 return -EFSCORRUPTED;
3062         }
3063
3064         /* start/end segment number in main_area */
3065         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3066         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3067                                                 GET_SEGNO(sbi, end);
3068         if (need_align) {
3069                 start_segno = rounddown(start_segno, sbi->segs_per_sec);
3070                 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
3071         }
3072
3073         cpc.reason = CP_DISCARD;
3074         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
3075         cpc.trim_start = start_segno;
3076         cpc.trim_end = end_segno;
3077
3078         if (sbi->discard_blks == 0)
3079                 goto out;
3080
3081         f2fs_down_write(&sbi->gc_lock);
3082         err = f2fs_write_checkpoint(sbi, &cpc);
3083         f2fs_up_write(&sbi->gc_lock);
3084         if (err)
3085                 goto out;
3086
3087         /*
3088          * We filed discard candidates, but actually we don't need to wait for
3089          * all of them, since they'll be issued in idle time along with runtime
3090          * discard option. User configuration looks like using runtime discard
3091          * or periodic fstrim instead of it.
3092          */
3093         if (f2fs_realtime_discard_enable(sbi))
3094                 goto out;
3095
3096         start_block = START_BLOCK(sbi, start_segno);
3097         end_block = START_BLOCK(sbi, end_segno + 1);
3098
3099         __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3100         trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3101                                         start_block, end_block);
3102
3103         trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3104                                         start_block, end_block);
3105 out:
3106         if (!err)
3107                 range->len = F2FS_BLK_TO_BYTES(trimmed);
3108         return err;
3109 }
3110
3111 static bool __has_curseg_space(struct f2fs_sb_info *sbi,
3112                                         struct curseg_info *curseg)
3113 {
3114         return curseg->next_blkoff < f2fs_usable_blks_in_seg(sbi,
3115                                                         curseg->segno);
3116 }
3117
3118 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
3119 {
3120         switch (hint) {
3121         case WRITE_LIFE_SHORT:
3122                 return CURSEG_HOT_DATA;
3123         case WRITE_LIFE_EXTREME:
3124                 return CURSEG_COLD_DATA;
3125         default:
3126                 return CURSEG_WARM_DATA;
3127         }
3128 }
3129
3130 static int __get_segment_type_2(struct f2fs_io_info *fio)
3131 {
3132         if (fio->type == DATA)
3133                 return CURSEG_HOT_DATA;
3134         else
3135                 return CURSEG_HOT_NODE;
3136 }
3137
3138 static int __get_segment_type_4(struct f2fs_io_info *fio)
3139 {
3140         if (fio->type == DATA) {
3141                 struct inode *inode = fio->page->mapping->host;
3142
3143                 if (S_ISDIR(inode->i_mode))
3144                         return CURSEG_HOT_DATA;
3145                 else
3146                         return CURSEG_COLD_DATA;
3147         } else {
3148                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3149                         return CURSEG_WARM_NODE;
3150                 else
3151                         return CURSEG_COLD_NODE;
3152         }
3153 }
3154
3155 static int __get_segment_type_6(struct f2fs_io_info *fio)
3156 {
3157         if (fio->type == DATA) {
3158                 struct inode *inode = fio->page->mapping->host;
3159
3160                 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
3161                         return CURSEG_COLD_DATA_PINNED;
3162
3163                 if (page_private_gcing(fio->page)) {
3164                         if (fio->sbi->am.atgc_enabled &&
3165                                 (fio->io_type == FS_DATA_IO) &&
3166                                 (fio->sbi->gc_mode != GC_URGENT_HIGH))
3167                                 return CURSEG_ALL_DATA_ATGC;
3168                         else
3169                                 return CURSEG_COLD_DATA;
3170                 }
3171                 if (file_is_cold(inode) || f2fs_need_compress_data(inode))
3172                         return CURSEG_COLD_DATA;
3173                 if (file_is_hot(inode) ||
3174                                 is_inode_flag_set(inode, FI_HOT_DATA) ||
3175                                 f2fs_is_cow_file(inode))
3176                         return CURSEG_HOT_DATA;
3177                 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3178         } else {
3179                 if (IS_DNODE(fio->page))
3180                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3181                                                 CURSEG_HOT_NODE;
3182                 return CURSEG_COLD_NODE;
3183         }
3184 }
3185
3186 static int __get_segment_type(struct f2fs_io_info *fio)
3187 {
3188         int type = 0;
3189
3190         switch (F2FS_OPTION(fio->sbi).active_logs) {
3191         case 2:
3192                 type = __get_segment_type_2(fio);
3193                 break;
3194         case 4:
3195                 type = __get_segment_type_4(fio);
3196                 break;
3197         case 6:
3198                 type = __get_segment_type_6(fio);
3199                 break;
3200         default:
3201                 f2fs_bug_on(fio->sbi, true);
3202         }
3203
3204         if (IS_HOT(type))
3205                 fio->temp = HOT;
3206         else if (IS_WARM(type))
3207                 fio->temp = WARM;
3208         else
3209                 fio->temp = COLD;
3210         return type;
3211 }
3212
3213 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3214                 block_t old_blkaddr, block_t *new_blkaddr,
3215                 struct f2fs_summary *sum, int type,
3216                 struct f2fs_io_info *fio)
3217 {
3218         struct sit_info *sit_i = SIT_I(sbi);
3219         struct curseg_info *curseg = CURSEG_I(sbi, type);
3220         unsigned long long old_mtime;
3221         bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3222         struct seg_entry *se = NULL;
3223
3224         f2fs_down_read(&SM_I(sbi)->curseg_lock);
3225
3226         mutex_lock(&curseg->curseg_mutex);
3227         down_write(&sit_i->sentry_lock);
3228
3229         if (from_gc) {
3230                 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3231                 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3232                 sanity_check_seg_type(sbi, se->type);
3233                 f2fs_bug_on(sbi, IS_NODESEG(se->type));
3234         }
3235         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3236
3237         f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg);
3238
3239         f2fs_wait_discard_bio(sbi, *new_blkaddr);
3240
3241         /*
3242          * __add_sum_entry should be resided under the curseg_mutex
3243          * because, this function updates a summary entry in the
3244          * current summary block.
3245          */
3246         __add_sum_entry(sbi, type, sum);
3247
3248         __refresh_next_blkoff(sbi, curseg);
3249
3250         stat_inc_block_count(sbi, curseg);
3251
3252         if (from_gc) {
3253                 old_mtime = get_segment_mtime(sbi, old_blkaddr);
3254         } else {
3255                 update_segment_mtime(sbi, old_blkaddr, 0);
3256                 old_mtime = 0;
3257         }
3258         update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3259
3260         /*
3261          * SIT information should be updated before segment allocation,
3262          * since SSR needs latest valid block information.
3263          */
3264         update_sit_entry(sbi, *new_blkaddr, 1);
3265         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3266                 update_sit_entry(sbi, old_blkaddr, -1);
3267
3268         if (!__has_curseg_space(sbi, curseg)) {
3269                 if (from_gc)
3270                         get_atssr_segment(sbi, type, se->type,
3271                                                 AT_SSR, se->mtime);
3272                 else
3273                         sit_i->s_ops->allocate_segment(sbi, type, false);
3274         }
3275         /*
3276          * segment dirty status should be updated after segment allocation,
3277          * so we just need to update status only one time after previous
3278          * segment being closed.
3279          */
3280         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3281         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3282
3283         up_write(&sit_i->sentry_lock);
3284
3285         if (page && IS_NODESEG(type)) {
3286                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3287
3288                 f2fs_inode_chksum_set(sbi, page);
3289         }
3290
3291         if (fio) {
3292                 struct f2fs_bio_info *io;
3293
3294                 if (F2FS_IO_ALIGNED(sbi))
3295                         fio->retry = false;
3296
3297                 INIT_LIST_HEAD(&fio->list);
3298                 fio->in_list = true;
3299                 io = sbi->write_io[fio->type] + fio->temp;
3300                 spin_lock(&io->io_lock);
3301                 list_add_tail(&fio->list, &io->io_list);
3302                 spin_unlock(&io->io_lock);
3303         }
3304
3305         mutex_unlock(&curseg->curseg_mutex);
3306
3307         f2fs_up_read(&SM_I(sbi)->curseg_lock);
3308 }
3309
3310 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino,
3311                                         block_t blkaddr, unsigned int blkcnt)
3312 {
3313         if (!f2fs_is_multi_device(sbi))
3314                 return;
3315
3316         while (1) {
3317                 unsigned int devidx = f2fs_target_device_index(sbi, blkaddr);
3318                 unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1;
3319
3320                 /* update device state for fsync */
3321                 f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO);
3322
3323                 /* update device state for checkpoint */
3324                 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3325                         spin_lock(&sbi->dev_lock);
3326                         f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3327                         spin_unlock(&sbi->dev_lock);
3328                 }
3329
3330                 if (blkcnt <= blks)
3331                         break;
3332                 blkcnt -= blks;
3333                 blkaddr += blks;
3334         }
3335 }
3336
3337 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3338 {
3339         int type = __get_segment_type(fio);
3340         bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3341
3342         if (keep_order)
3343                 f2fs_down_read(&fio->sbi->io_order_lock);
3344 reallocate:
3345         f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3346                         &fio->new_blkaddr, sum, type, fio);
3347         if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO) {
3348                 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3349                                         fio->old_blkaddr, fio->old_blkaddr);
3350                 f2fs_invalidate_compress_page(fio->sbi, fio->old_blkaddr);
3351         }
3352
3353         /* writeout dirty page into bdev */
3354         f2fs_submit_page_write(fio);
3355         if (fio->retry) {
3356                 fio->old_blkaddr = fio->new_blkaddr;
3357                 goto reallocate;
3358         }
3359
3360         f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1);
3361
3362         if (keep_order)
3363                 f2fs_up_read(&fio->sbi->io_order_lock);
3364 }
3365
3366 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3367                                         enum iostat_type io_type)
3368 {
3369         struct f2fs_io_info fio = {
3370                 .sbi = sbi,
3371                 .type = META,
3372                 .temp = HOT,
3373                 .op = REQ_OP_WRITE,
3374                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3375                 .old_blkaddr = page->index,
3376                 .new_blkaddr = page->index,
3377                 .page = page,
3378                 .encrypted_page = NULL,
3379                 .in_list = false,
3380         };
3381
3382         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3383                 fio.op_flags &= ~REQ_META;
3384
3385         set_page_writeback(page);
3386         ClearPageError(page);
3387         f2fs_submit_page_write(&fio);
3388
3389         stat_inc_meta_count(sbi, page->index);
3390         f2fs_update_iostat(sbi, NULL, io_type, F2FS_BLKSIZE);
3391 }
3392
3393 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3394 {
3395         struct f2fs_summary sum;
3396
3397         set_summary(&sum, nid, 0, 0);
3398         do_write_page(&sum, fio);
3399
3400         f2fs_update_iostat(fio->sbi, NULL, fio->io_type, F2FS_BLKSIZE);
3401 }
3402
3403 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3404                                         struct f2fs_io_info *fio)
3405 {
3406         struct f2fs_sb_info *sbi = fio->sbi;
3407         struct f2fs_summary sum;
3408
3409         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3410         set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3411         do_write_page(&sum, fio);
3412         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3413
3414         f2fs_update_iostat(sbi, dn->inode, fio->io_type, F2FS_BLKSIZE);
3415 }
3416
3417 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3418 {
3419         int err;
3420         struct f2fs_sb_info *sbi = fio->sbi;
3421         unsigned int segno;
3422
3423         fio->new_blkaddr = fio->old_blkaddr;
3424         /* i/o temperature is needed for passing down write hints */
3425         __get_segment_type(fio);
3426
3427         segno = GET_SEGNO(sbi, fio->new_blkaddr);
3428
3429         if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3430                 set_sbi_flag(sbi, SBI_NEED_FSCK);
3431                 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3432                           __func__, segno);
3433                 err = -EFSCORRUPTED;
3434                 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
3435                 goto drop_bio;
3436         }
3437
3438         if (f2fs_cp_error(sbi)) {
3439                 err = -EIO;
3440                 goto drop_bio;
3441         }
3442
3443         if (fio->post_read)
3444                 invalidate_mapping_pages(META_MAPPING(sbi),
3445                                 fio->new_blkaddr, fio->new_blkaddr);
3446
3447         stat_inc_inplace_blocks(fio->sbi);
3448
3449         if (fio->bio && !(SM_I(sbi)->ipu_policy & (1 << F2FS_IPU_NOCACHE)))
3450                 err = f2fs_merge_page_bio(fio);
3451         else
3452                 err = f2fs_submit_page_bio(fio);
3453         if (!err) {
3454                 f2fs_update_device_state(fio->sbi, fio->ino,
3455                                                 fio->new_blkaddr, 1);
3456                 f2fs_update_iostat(fio->sbi, fio->page->mapping->host,
3457                                                 fio->io_type, F2FS_BLKSIZE);
3458         }
3459
3460         return err;
3461 drop_bio:
3462         if (fio->bio && *(fio->bio)) {
3463                 struct bio *bio = *(fio->bio);
3464
3465                 bio->bi_status = BLK_STS_IOERR;
3466                 bio_endio(bio);
3467                 *(fio->bio) = NULL;
3468         }
3469         return err;
3470 }
3471
3472 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3473                                                 unsigned int segno)
3474 {
3475         int i;
3476
3477         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3478                 if (CURSEG_I(sbi, i)->segno == segno)
3479                         break;
3480         }
3481         return i;
3482 }
3483
3484 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3485                                 block_t old_blkaddr, block_t new_blkaddr,
3486                                 bool recover_curseg, bool recover_newaddr,
3487                                 bool from_gc)
3488 {
3489         struct sit_info *sit_i = SIT_I(sbi);
3490         struct curseg_info *curseg;
3491         unsigned int segno, old_cursegno;
3492         struct seg_entry *se;
3493         int type;
3494         unsigned short old_blkoff;
3495         unsigned char old_alloc_type;
3496
3497         segno = GET_SEGNO(sbi, new_blkaddr);
3498         se = get_seg_entry(sbi, segno);
3499         type = se->type;
3500
3501         f2fs_down_write(&SM_I(sbi)->curseg_lock);
3502
3503         if (!recover_curseg) {
3504                 /* for recovery flow */
3505                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3506                         if (old_blkaddr == NULL_ADDR)
3507                                 type = CURSEG_COLD_DATA;
3508                         else
3509                                 type = CURSEG_WARM_DATA;
3510                 }
3511         } else {
3512                 if (IS_CURSEG(sbi, segno)) {
3513                         /* se->type is volatile as SSR allocation */
3514                         type = __f2fs_get_curseg(sbi, segno);
3515                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3516                 } else {
3517                         type = CURSEG_WARM_DATA;
3518                 }
3519         }
3520
3521         f2fs_bug_on(sbi, !IS_DATASEG(type));
3522         curseg = CURSEG_I(sbi, type);
3523
3524         mutex_lock(&curseg->curseg_mutex);
3525         down_write(&sit_i->sentry_lock);
3526
3527         old_cursegno = curseg->segno;
3528         old_blkoff = curseg->next_blkoff;
3529         old_alloc_type = curseg->alloc_type;
3530
3531         /* change the current segment */
3532         if (segno != curseg->segno) {
3533                 curseg->next_segno = segno;
3534                 change_curseg(sbi, type, true);
3535         }
3536
3537         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3538         __add_sum_entry(sbi, type, sum);
3539
3540         if (!recover_curseg || recover_newaddr) {
3541                 if (!from_gc)
3542                         update_segment_mtime(sbi, new_blkaddr, 0);
3543                 update_sit_entry(sbi, new_blkaddr, 1);
3544         }
3545         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3546                 invalidate_mapping_pages(META_MAPPING(sbi),
3547                                         old_blkaddr, old_blkaddr);
3548                 f2fs_invalidate_compress_page(sbi, old_blkaddr);
3549                 if (!from_gc)
3550                         update_segment_mtime(sbi, old_blkaddr, 0);
3551                 update_sit_entry(sbi, old_blkaddr, -1);
3552         }
3553
3554         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3555         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3556
3557         locate_dirty_segment(sbi, old_cursegno);
3558
3559         if (recover_curseg) {
3560                 if (old_cursegno != curseg->segno) {
3561                         curseg->next_segno = old_cursegno;
3562                         change_curseg(sbi, type, true);
3563                 }
3564                 curseg->next_blkoff = old_blkoff;
3565                 curseg->alloc_type = old_alloc_type;
3566         }
3567
3568         up_write(&sit_i->sentry_lock);
3569         mutex_unlock(&curseg->curseg_mutex);
3570         f2fs_up_write(&SM_I(sbi)->curseg_lock);
3571 }
3572
3573 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3574                                 block_t old_addr, block_t new_addr,
3575                                 unsigned char version, bool recover_curseg,
3576                                 bool recover_newaddr)
3577 {
3578         struct f2fs_summary sum;
3579
3580         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3581
3582         f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3583                                         recover_curseg, recover_newaddr, false);
3584
3585         f2fs_update_data_blkaddr(dn, new_addr);
3586 }
3587
3588 void f2fs_wait_on_page_writeback(struct page *page,
3589                                 enum page_type type, bool ordered, bool locked)
3590 {
3591         if (PageWriteback(page)) {
3592                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3593
3594                 /* submit cached LFS IO */
3595                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3596                 /* sbumit cached IPU IO */
3597                 f2fs_submit_merged_ipu_write(sbi, NULL, page);
3598                 if (ordered) {
3599                         wait_on_page_writeback(page);
3600                         f2fs_bug_on(sbi, locked && PageWriteback(page));
3601                 } else {
3602                         wait_for_stable_page(page);
3603                 }
3604         }
3605 }
3606
3607 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3608 {
3609         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3610         struct page *cpage;
3611
3612         if (!f2fs_post_read_required(inode))
3613                 return;
3614
3615         if (!__is_valid_data_blkaddr(blkaddr))
3616                 return;
3617
3618         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3619         if (cpage) {
3620                 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3621                 f2fs_put_page(cpage, 1);
3622         }
3623 }
3624
3625 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3626                                                                 block_t len)
3627 {
3628         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3629         block_t i;
3630
3631         if (!f2fs_post_read_required(inode))
3632                 return;
3633
3634         for (i = 0; i < len; i++)
3635                 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3636
3637         invalidate_mapping_pages(META_MAPPING(sbi), blkaddr, blkaddr + len - 1);
3638 }
3639
3640 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3641 {
3642         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3643         struct curseg_info *seg_i;
3644         unsigned char *kaddr;
3645         struct page *page;
3646         block_t start;
3647         int i, j, offset;
3648
3649         start = start_sum_block(sbi);
3650
3651         page = f2fs_get_meta_page(sbi, start++);
3652         if (IS_ERR(page))
3653                 return PTR_ERR(page);
3654         kaddr = (unsigned char *)page_address(page);
3655
3656         /* Step 1: restore nat cache */
3657         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3658         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3659
3660         /* Step 2: restore sit cache */
3661         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3662         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3663         offset = 2 * SUM_JOURNAL_SIZE;
3664
3665         /* Step 3: restore summary entries */
3666         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3667                 unsigned short blk_off;
3668                 unsigned int segno;
3669
3670                 seg_i = CURSEG_I(sbi, i);
3671                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3672                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3673                 seg_i->next_segno = segno;
3674                 reset_curseg(sbi, i, 0);
3675                 seg_i->alloc_type = ckpt->alloc_type[i];
3676                 seg_i->next_blkoff = blk_off;
3677
3678                 if (seg_i->alloc_type == SSR)
3679                         blk_off = sbi->blocks_per_seg;
3680
3681                 for (j = 0; j < blk_off; j++) {
3682                         struct f2fs_summary *s;
3683
3684                         s = (struct f2fs_summary *)(kaddr + offset);
3685                         seg_i->sum_blk->entries[j] = *s;
3686                         offset += SUMMARY_SIZE;
3687                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3688                                                 SUM_FOOTER_SIZE)
3689                                 continue;
3690
3691                         f2fs_put_page(page, 1);
3692                         page = NULL;
3693
3694                         page = f2fs_get_meta_page(sbi, start++);
3695                         if (IS_ERR(page))
3696                                 return PTR_ERR(page);
3697                         kaddr = (unsigned char *)page_address(page);
3698                         offset = 0;
3699                 }
3700         }
3701         f2fs_put_page(page, 1);
3702         return 0;
3703 }
3704
3705 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3706 {
3707         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3708         struct f2fs_summary_block *sum;
3709         struct curseg_info *curseg;
3710         struct page *new;
3711         unsigned short blk_off;
3712         unsigned int segno = 0;
3713         block_t blk_addr = 0;
3714         int err = 0;
3715
3716         /* get segment number and block addr */
3717         if (IS_DATASEG(type)) {
3718                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3719                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3720                                                         CURSEG_HOT_DATA]);
3721                 if (__exist_node_summaries(sbi))
3722                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
3723                 else
3724                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3725         } else {
3726                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3727                                                         CURSEG_HOT_NODE]);
3728                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3729                                                         CURSEG_HOT_NODE]);
3730                 if (__exist_node_summaries(sbi))
3731                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3732                                                         type - CURSEG_HOT_NODE);
3733                 else
3734                         blk_addr = GET_SUM_BLOCK(sbi, segno);
3735         }
3736
3737         new = f2fs_get_meta_page(sbi, blk_addr);
3738         if (IS_ERR(new))
3739                 return PTR_ERR(new);
3740         sum = (struct f2fs_summary_block *)page_address(new);
3741
3742         if (IS_NODESEG(type)) {
3743                 if (__exist_node_summaries(sbi)) {
3744                         struct f2fs_summary *ns = &sum->entries[0];
3745                         int i;
3746
3747                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3748                                 ns->version = 0;
3749                                 ns->ofs_in_node = 0;
3750                         }
3751                 } else {
3752                         err = f2fs_restore_node_summary(sbi, segno, sum);
3753                         if (err)
3754                                 goto out;
3755                 }
3756         }
3757
3758         /* set uncompleted segment to curseg */
3759         curseg = CURSEG_I(sbi, type);
3760         mutex_lock(&curseg->curseg_mutex);
3761
3762         /* update journal info */
3763         down_write(&curseg->journal_rwsem);
3764         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3765         up_write(&curseg->journal_rwsem);
3766
3767         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3768         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3769         curseg->next_segno = segno;
3770         reset_curseg(sbi, type, 0);
3771         curseg->alloc_type = ckpt->alloc_type[type];
3772         curseg->next_blkoff = blk_off;
3773         mutex_unlock(&curseg->curseg_mutex);
3774 out:
3775         f2fs_put_page(new, 1);
3776         return err;
3777 }
3778
3779 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3780 {
3781         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3782         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3783         int type = CURSEG_HOT_DATA;
3784         int err;
3785
3786         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3787                 int npages = f2fs_npages_for_summary_flush(sbi, true);
3788
3789                 if (npages >= 2)
3790                         f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3791                                                         META_CP, true);
3792
3793                 /* restore for compacted data summary */
3794                 err = read_compacted_summaries(sbi);
3795                 if (err)
3796                         return err;
3797                 type = CURSEG_HOT_NODE;
3798         }
3799
3800         if (__exist_node_summaries(sbi))
3801                 f2fs_ra_meta_pages(sbi,
3802                                 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
3803                                 NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
3804
3805         for (; type <= CURSEG_COLD_NODE; type++) {
3806                 err = read_normal_summaries(sbi, type);
3807                 if (err)
3808                         return err;
3809         }
3810
3811         /* sanity check for summary blocks */
3812         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3813                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3814                 f2fs_err(sbi, "invalid journal entries nats %u sits %u",
3815                          nats_in_cursum(nat_j), sits_in_cursum(sit_j));
3816                 return -EINVAL;
3817         }
3818
3819         return 0;
3820 }
3821
3822 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3823 {
3824         struct page *page;
3825         unsigned char *kaddr;
3826         struct f2fs_summary *summary;
3827         struct curseg_info *seg_i;
3828         int written_size = 0;
3829         int i, j;
3830
3831         page = f2fs_grab_meta_page(sbi, blkaddr++);
3832         kaddr = (unsigned char *)page_address(page);
3833         memset(kaddr, 0, PAGE_SIZE);
3834
3835         /* Step 1: write nat cache */
3836         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3837         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3838         written_size += SUM_JOURNAL_SIZE;
3839
3840         /* Step 2: write sit cache */
3841         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3842         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3843         written_size += SUM_JOURNAL_SIZE;
3844
3845         /* Step 3: write summary entries */
3846         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3847                 unsigned short blkoff;
3848
3849                 seg_i = CURSEG_I(sbi, i);
3850                 if (sbi->ckpt->alloc_type[i] == SSR)
3851                         blkoff = sbi->blocks_per_seg;
3852                 else
3853                         blkoff = curseg_blkoff(sbi, i);
3854
3855                 for (j = 0; j < blkoff; j++) {
3856                         if (!page) {
3857                                 page = f2fs_grab_meta_page(sbi, blkaddr++);
3858                                 kaddr = (unsigned char *)page_address(page);
3859                                 memset(kaddr, 0, PAGE_SIZE);
3860                                 written_size = 0;
3861                         }
3862                         summary = (struct f2fs_summary *)(kaddr + written_size);
3863                         *summary = seg_i->sum_blk->entries[j];
3864                         written_size += SUMMARY_SIZE;
3865
3866                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3867                                                         SUM_FOOTER_SIZE)
3868                                 continue;
3869
3870                         set_page_dirty(page);
3871                         f2fs_put_page(page, 1);
3872                         page = NULL;
3873                 }
3874         }
3875         if (page) {
3876                 set_page_dirty(page);
3877                 f2fs_put_page(page, 1);
3878         }
3879 }
3880
3881 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3882                                         block_t blkaddr, int type)
3883 {
3884         int i, end;
3885
3886         if (IS_DATASEG(type))
3887                 end = type + NR_CURSEG_DATA_TYPE;
3888         else
3889                 end = type + NR_CURSEG_NODE_TYPE;
3890
3891         for (i = type; i < end; i++)
3892                 write_current_sum_page(sbi, i, blkaddr + (i - type));
3893 }
3894
3895 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3896 {
3897         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3898                 write_compacted_summaries(sbi, start_blk);
3899         else
3900                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3901 }
3902
3903 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3904 {
3905         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3906 }
3907
3908 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3909                                         unsigned int val, int alloc)
3910 {
3911         int i;
3912
3913         if (type == NAT_JOURNAL) {
3914                 for (i = 0; i < nats_in_cursum(journal); i++) {
3915                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3916                                 return i;
3917                 }
3918                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3919                         return update_nats_in_cursum(journal, 1);
3920         } else if (type == SIT_JOURNAL) {
3921                 for (i = 0; i < sits_in_cursum(journal); i++)
3922                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3923                                 return i;
3924                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3925                         return update_sits_in_cursum(journal, 1);
3926         }
3927         return -1;
3928 }
3929
3930 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3931                                         unsigned int segno)
3932 {
3933         return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
3934 }
3935
3936 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3937                                         unsigned int start)
3938 {
3939         struct sit_info *sit_i = SIT_I(sbi);
3940         struct page *page;
3941         pgoff_t src_off, dst_off;
3942
3943         src_off = current_sit_addr(sbi, start);
3944         dst_off = next_sit_addr(sbi, src_off);
3945
3946         page = f2fs_grab_meta_page(sbi, dst_off);
3947         seg_info_to_sit_page(sbi, page, start);
3948
3949         set_page_dirty(page);
3950         set_to_next_sit(sit_i, start);
3951
3952         return page;
3953 }
3954
3955 static struct sit_entry_set *grab_sit_entry_set(void)
3956 {
3957         struct sit_entry_set *ses =
3958                         f2fs_kmem_cache_alloc(sit_entry_set_slab,
3959                                                 GFP_NOFS, true, NULL);
3960
3961         ses->entry_cnt = 0;
3962         INIT_LIST_HEAD(&ses->set_list);
3963         return ses;
3964 }
3965
3966 static void release_sit_entry_set(struct sit_entry_set *ses)
3967 {
3968         list_del(&ses->set_list);
3969         kmem_cache_free(sit_entry_set_slab, ses);
3970 }
3971
3972 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3973                                                 struct list_head *head)
3974 {
3975         struct sit_entry_set *next = ses;
3976
3977         if (list_is_last(&ses->set_list, head))
3978                 return;
3979
3980         list_for_each_entry_continue(next, head, set_list)
3981                 if (ses->entry_cnt <= next->entry_cnt) {
3982                         list_move_tail(&ses->set_list, &next->set_list);
3983                         return;
3984                 }
3985
3986         list_move_tail(&ses->set_list, head);
3987 }
3988
3989 static void add_sit_entry(unsigned int segno, struct list_head *head)
3990 {
3991         struct sit_entry_set *ses;
3992         unsigned int start_segno = START_SEGNO(segno);
3993
3994         list_for_each_entry(ses, head, set_list) {
3995                 if (ses->start_segno == start_segno) {
3996                         ses->entry_cnt++;
3997                         adjust_sit_entry_set(ses, head);
3998                         return;
3999                 }
4000         }
4001
4002         ses = grab_sit_entry_set();
4003
4004         ses->start_segno = start_segno;
4005         ses->entry_cnt++;
4006         list_add(&ses->set_list, head);
4007 }
4008
4009 static void add_sits_in_set(struct f2fs_sb_info *sbi)
4010 {
4011         struct f2fs_sm_info *sm_info = SM_I(sbi);
4012         struct list_head *set_list = &sm_info->sit_entry_set;
4013         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
4014         unsigned int segno;
4015
4016         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
4017                 add_sit_entry(segno, set_list);
4018 }
4019
4020 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
4021 {
4022         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4023         struct f2fs_journal *journal = curseg->journal;
4024         int i;
4025
4026         down_write(&curseg->journal_rwsem);
4027         for (i = 0; i < sits_in_cursum(journal); i++) {
4028                 unsigned int segno;
4029                 bool dirtied;
4030
4031                 segno = le32_to_cpu(segno_in_journal(journal, i));
4032                 dirtied = __mark_sit_entry_dirty(sbi, segno);
4033
4034                 if (!dirtied)
4035                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
4036         }
4037         update_sits_in_cursum(journal, -i);
4038         up_write(&curseg->journal_rwsem);
4039 }
4040
4041 /*
4042  * CP calls this function, which flushes SIT entries including sit_journal,
4043  * and moves prefree segs to free segs.
4044  */
4045 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
4046 {
4047         struct sit_info *sit_i = SIT_I(sbi);
4048         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4049         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4050         struct f2fs_journal *journal = curseg->journal;
4051         struct sit_entry_set *ses, *tmp;
4052         struct list_head *head = &SM_I(sbi)->sit_entry_set;
4053         bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4054         struct seg_entry *se;
4055
4056         down_write(&sit_i->sentry_lock);
4057
4058         if (!sit_i->dirty_sentries)
4059                 goto out;
4060
4061         /*
4062          * add and account sit entries of dirty bitmap in sit entry
4063          * set temporarily
4064          */
4065         add_sits_in_set(sbi);
4066
4067         /*
4068          * if there are no enough space in journal to store dirty sit
4069          * entries, remove all entries from journal and add and account
4070          * them in sit entry set.
4071          */
4072         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4073                                                                 !to_journal)
4074                 remove_sits_in_journal(sbi);
4075
4076         /*
4077          * there are two steps to flush sit entries:
4078          * #1, flush sit entries to journal in current cold data summary block.
4079          * #2, flush sit entries to sit page.
4080          */
4081         list_for_each_entry_safe(ses, tmp, head, set_list) {
4082                 struct page *page = NULL;
4083                 struct f2fs_sit_block *raw_sit = NULL;
4084                 unsigned int start_segno = ses->start_segno;
4085                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4086                                                 (unsigned long)MAIN_SEGS(sbi));
4087                 unsigned int segno = start_segno;
4088
4089                 if (to_journal &&
4090                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4091                         to_journal = false;
4092
4093                 if (to_journal) {
4094                         down_write(&curseg->journal_rwsem);
4095                 } else {
4096                         page = get_next_sit_page(sbi, start_segno);
4097                         raw_sit = page_address(page);
4098                 }
4099
4100                 /* flush dirty sit entries in region of current sit set */
4101                 for_each_set_bit_from(segno, bitmap, end) {
4102                         int offset, sit_offset;
4103
4104                         se = get_seg_entry(sbi, segno);
4105 #ifdef CONFIG_F2FS_CHECK_FS
4106                         if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4107                                                 SIT_VBLOCK_MAP_SIZE))
4108                                 f2fs_bug_on(sbi, 1);
4109 #endif
4110
4111                         /* add discard candidates */
4112                         if (!(cpc->reason & CP_DISCARD)) {
4113                                 cpc->trim_start = segno;
4114                                 add_discard_addrs(sbi, cpc, false);
4115                         }
4116
4117                         if (to_journal) {
4118                                 offset = f2fs_lookup_journal_in_cursum(journal,
4119                                                         SIT_JOURNAL, segno, 1);
4120                                 f2fs_bug_on(sbi, offset < 0);
4121                                 segno_in_journal(journal, offset) =
4122                                                         cpu_to_le32(segno);
4123                                 seg_info_to_raw_sit(se,
4124                                         &sit_in_journal(journal, offset));
4125                                 check_block_count(sbi, segno,
4126                                         &sit_in_journal(journal, offset));
4127                         } else {
4128                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4129                                 seg_info_to_raw_sit(se,
4130                                                 &raw_sit->entries[sit_offset]);
4131                                 check_block_count(sbi, segno,
4132                                                 &raw_sit->entries[sit_offset]);
4133                         }
4134
4135                         __clear_bit(segno, bitmap);
4136                         sit_i->dirty_sentries--;
4137                         ses->entry_cnt--;
4138                 }
4139
4140                 if (to_journal)
4141                         up_write(&curseg->journal_rwsem);
4142                 else
4143                         f2fs_put_page(page, 1);
4144
4145                 f2fs_bug_on(sbi, ses->entry_cnt);
4146                 release_sit_entry_set(ses);
4147         }
4148
4149         f2fs_bug_on(sbi, !list_empty(head));
4150         f2fs_bug_on(sbi, sit_i->dirty_sentries);
4151 out:
4152         if (cpc->reason & CP_DISCARD) {
4153                 __u64 trim_start = cpc->trim_start;
4154
4155                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4156                         add_discard_addrs(sbi, cpc, false);
4157
4158                 cpc->trim_start = trim_start;
4159         }
4160         up_write(&sit_i->sentry_lock);
4161
4162         set_prefree_as_free_segments(sbi);
4163 }
4164
4165 static int build_sit_info(struct f2fs_sb_info *sbi)
4166 {
4167         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4168         struct sit_info *sit_i;
4169         unsigned int sit_segs, start;
4170         char *src_bitmap, *bitmap;
4171         unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4172         unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0;
4173
4174         /* allocate memory for SIT information */
4175         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4176         if (!sit_i)
4177                 return -ENOMEM;
4178
4179         SM_I(sbi)->sit_info = sit_i;
4180
4181         sit_i->sentries =
4182                 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4183                                               MAIN_SEGS(sbi)),
4184                               GFP_KERNEL);
4185         if (!sit_i->sentries)
4186                 return -ENOMEM;
4187
4188         main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4189         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4190                                                                 GFP_KERNEL);
4191         if (!sit_i->dirty_sentries_bitmap)
4192                 return -ENOMEM;
4193
4194 #ifdef CONFIG_F2FS_CHECK_FS
4195         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map);
4196 #else
4197         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map);
4198 #endif
4199         sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4200         if (!sit_i->bitmap)
4201                 return -ENOMEM;
4202
4203         bitmap = sit_i->bitmap;
4204
4205         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4206                 sit_i->sentries[start].cur_valid_map = bitmap;
4207                 bitmap += SIT_VBLOCK_MAP_SIZE;
4208
4209                 sit_i->sentries[start].ckpt_valid_map = bitmap;
4210                 bitmap += SIT_VBLOCK_MAP_SIZE;
4211
4212 #ifdef CONFIG_F2FS_CHECK_FS
4213                 sit_i->sentries[start].cur_valid_map_mir = bitmap;
4214                 bitmap += SIT_VBLOCK_MAP_SIZE;
4215 #endif
4216
4217                 if (discard_map) {
4218                         sit_i->sentries[start].discard_map = bitmap;
4219                         bitmap += SIT_VBLOCK_MAP_SIZE;
4220                 }
4221         }
4222
4223         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4224         if (!sit_i->tmp_map)
4225                 return -ENOMEM;
4226
4227         if (__is_large_section(sbi)) {
4228                 sit_i->sec_entries =
4229                         f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4230                                                       MAIN_SECS(sbi)),
4231                                       GFP_KERNEL);
4232                 if (!sit_i->sec_entries)
4233                         return -ENOMEM;
4234         }
4235
4236         /* get information related with SIT */
4237         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4238
4239         /* setup SIT bitmap from ckeckpoint pack */
4240         sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4241         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4242
4243         sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4244         if (!sit_i->sit_bitmap)
4245                 return -ENOMEM;
4246
4247 #ifdef CONFIG_F2FS_CHECK_FS
4248         sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4249                                         sit_bitmap_size, GFP_KERNEL);
4250         if (!sit_i->sit_bitmap_mir)
4251                 return -ENOMEM;
4252
4253         sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4254                                         main_bitmap_size, GFP_KERNEL);
4255         if (!sit_i->invalid_segmap)
4256                 return -ENOMEM;
4257 #endif
4258
4259         /* init SIT information */
4260         sit_i->s_ops = &default_salloc_ops;
4261
4262         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4263         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4264         sit_i->written_valid_blocks = 0;
4265         sit_i->bitmap_size = sit_bitmap_size;
4266         sit_i->dirty_sentries = 0;
4267         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4268         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4269         sit_i->mounted_time = ktime_get_boottime_seconds();
4270         init_rwsem(&sit_i->sentry_lock);
4271         return 0;
4272 }
4273
4274 static int build_free_segmap(struct f2fs_sb_info *sbi)
4275 {
4276         struct free_segmap_info *free_i;
4277         unsigned int bitmap_size, sec_bitmap_size;
4278
4279         /* allocate memory for free segmap information */
4280         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4281         if (!free_i)
4282                 return -ENOMEM;
4283
4284         SM_I(sbi)->free_info = free_i;
4285
4286         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4287         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4288         if (!free_i->free_segmap)
4289                 return -ENOMEM;
4290
4291         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4292         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4293         if (!free_i->free_secmap)
4294                 return -ENOMEM;
4295
4296         /* set all segments as dirty temporarily */
4297         memset(free_i->free_segmap, 0xff, bitmap_size);
4298         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4299
4300         /* init free segmap information */
4301         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4302         free_i->free_segments = 0;
4303         free_i->free_sections = 0;
4304         spin_lock_init(&free_i->segmap_lock);
4305         return 0;
4306 }
4307
4308 static int build_curseg(struct f2fs_sb_info *sbi)
4309 {
4310         struct curseg_info *array;
4311         int i;
4312
4313         array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4314                                         sizeof(*array)), GFP_KERNEL);
4315         if (!array)
4316                 return -ENOMEM;
4317
4318         SM_I(sbi)->curseg_array = array;
4319
4320         for (i = 0; i < NO_CHECK_TYPE; i++) {
4321                 mutex_init(&array[i].curseg_mutex);
4322                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4323                 if (!array[i].sum_blk)
4324                         return -ENOMEM;
4325                 init_rwsem(&array[i].journal_rwsem);
4326                 array[i].journal = f2fs_kzalloc(sbi,
4327                                 sizeof(struct f2fs_journal), GFP_KERNEL);
4328                 if (!array[i].journal)
4329                         return -ENOMEM;
4330                 if (i < NR_PERSISTENT_LOG)
4331                         array[i].seg_type = CURSEG_HOT_DATA + i;
4332                 else if (i == CURSEG_COLD_DATA_PINNED)
4333                         array[i].seg_type = CURSEG_COLD_DATA;
4334                 else if (i == CURSEG_ALL_DATA_ATGC)
4335                         array[i].seg_type = CURSEG_COLD_DATA;
4336                 array[i].segno = NULL_SEGNO;
4337                 array[i].next_blkoff = 0;
4338                 array[i].inited = false;
4339         }
4340         return restore_curseg_summaries(sbi);
4341 }
4342
4343 static int build_sit_entries(struct f2fs_sb_info *sbi)
4344 {
4345         struct sit_info *sit_i = SIT_I(sbi);
4346         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4347         struct f2fs_journal *journal = curseg->journal;
4348         struct seg_entry *se;
4349         struct f2fs_sit_entry sit;
4350         int sit_blk_cnt = SIT_BLK_CNT(sbi);
4351         unsigned int i, start, end;
4352         unsigned int readed, start_blk = 0;
4353         int err = 0;
4354         block_t sit_valid_blocks[2] = {0, 0};
4355
4356         do {
4357                 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS,
4358                                                         META_SIT, true);
4359
4360                 start = start_blk * sit_i->sents_per_block;
4361                 end = (start_blk + readed) * sit_i->sents_per_block;
4362
4363                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4364                         struct f2fs_sit_block *sit_blk;
4365                         struct page *page;
4366
4367                         se = &sit_i->sentries[start];
4368                         page = get_current_sit_page(sbi, start);
4369                         if (IS_ERR(page))
4370                                 return PTR_ERR(page);
4371                         sit_blk = (struct f2fs_sit_block *)page_address(page);
4372                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4373                         f2fs_put_page(page, 1);
4374
4375                         err = check_block_count(sbi, start, &sit);
4376                         if (err)
4377                                 return err;
4378                         seg_info_from_raw_sit(se, &sit);
4379
4380                         if (se->type >= NR_PERSISTENT_LOG) {
4381                                 f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4382                                                         se->type, start);
4383                                 f2fs_handle_error(sbi,
4384                                                 ERROR_INCONSISTENT_SUM_TYPE);
4385                                 return -EFSCORRUPTED;
4386                         }
4387
4388                         sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4389
4390                         if (f2fs_block_unit_discard(sbi)) {
4391                                 /* build discard map only one time */
4392                                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4393                                         memset(se->discard_map, 0xff,
4394                                                 SIT_VBLOCK_MAP_SIZE);
4395                                 } else {
4396                                         memcpy(se->discard_map,
4397                                                 se->cur_valid_map,
4398                                                 SIT_VBLOCK_MAP_SIZE);
4399                                         sbi->discard_blks +=
4400                                                 sbi->blocks_per_seg -
4401                                                 se->valid_blocks;
4402                                 }
4403                         }
4404
4405                         if (__is_large_section(sbi))
4406                                 get_sec_entry(sbi, start)->valid_blocks +=
4407                                                         se->valid_blocks;
4408                 }
4409                 start_blk += readed;
4410         } while (start_blk < sit_blk_cnt);
4411
4412         down_read(&curseg->journal_rwsem);
4413         for (i = 0; i < sits_in_cursum(journal); i++) {
4414                 unsigned int old_valid_blocks;
4415
4416                 start = le32_to_cpu(segno_in_journal(journal, i));
4417                 if (start >= MAIN_SEGS(sbi)) {
4418                         f2fs_err(sbi, "Wrong journal entry on segno %u",
4419                                  start);
4420                         err = -EFSCORRUPTED;
4421                         f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL);
4422                         break;
4423                 }
4424
4425                 se = &sit_i->sentries[start];
4426                 sit = sit_in_journal(journal, i);
4427
4428                 old_valid_blocks = se->valid_blocks;
4429
4430                 sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks;
4431
4432                 err = check_block_count(sbi, start, &sit);
4433                 if (err)
4434                         break;
4435                 seg_info_from_raw_sit(se, &sit);
4436
4437                 if (se->type >= NR_PERSISTENT_LOG) {
4438                         f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4439                                                         se->type, start);
4440                         err = -EFSCORRUPTED;
4441                         f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
4442                         break;
4443                 }
4444
4445                 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4446
4447                 if (f2fs_block_unit_discard(sbi)) {
4448                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4449                                 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4450                         } else {
4451                                 memcpy(se->discard_map, se->cur_valid_map,
4452                                                         SIT_VBLOCK_MAP_SIZE);
4453                                 sbi->discard_blks += old_valid_blocks;
4454                                 sbi->discard_blks -= se->valid_blocks;
4455                         }
4456                 }
4457
4458                 if (__is_large_section(sbi)) {
4459                         get_sec_entry(sbi, start)->valid_blocks +=
4460                                                         se->valid_blocks;
4461                         get_sec_entry(sbi, start)->valid_blocks -=
4462                                                         old_valid_blocks;
4463                 }
4464         }
4465         up_read(&curseg->journal_rwsem);
4466
4467         if (err)
4468                 return err;
4469
4470         if (sit_valid_blocks[NODE] != valid_node_count(sbi)) {
4471                 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4472                          sit_valid_blocks[NODE], valid_node_count(sbi));
4473                 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT);
4474                 return -EFSCORRUPTED;
4475         }
4476
4477         if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] >
4478                                 valid_user_blocks(sbi)) {
4479                 f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u",
4480                          sit_valid_blocks[DATA], sit_valid_blocks[NODE],
4481                          valid_user_blocks(sbi));
4482                 f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT);
4483                 return -EFSCORRUPTED;
4484         }
4485
4486         return 0;
4487 }
4488
4489 static void init_free_segmap(struct f2fs_sb_info *sbi)
4490 {
4491         unsigned int start;
4492         int type;
4493         struct seg_entry *sentry;
4494
4495         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4496                 if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4497                         continue;
4498                 sentry = get_seg_entry(sbi, start);
4499                 if (!sentry->valid_blocks)
4500                         __set_free(sbi, start);
4501                 else
4502                         SIT_I(sbi)->written_valid_blocks +=
4503                                                 sentry->valid_blocks;
4504         }
4505
4506         /* set use the current segments */
4507         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4508                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4509
4510                 __set_test_and_inuse(sbi, curseg_t->segno);
4511         }
4512 }
4513
4514 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4515 {
4516         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4517         struct free_segmap_info *free_i = FREE_I(sbi);
4518         unsigned int segno = 0, offset = 0, secno;
4519         block_t valid_blocks, usable_blks_in_seg;
4520
4521         while (1) {
4522                 /* find dirty segment based on free segmap */
4523                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4524                 if (segno >= MAIN_SEGS(sbi))
4525                         break;
4526                 offset = segno + 1;
4527                 valid_blocks = get_valid_blocks(sbi, segno, false);
4528                 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4529                 if (valid_blocks == usable_blks_in_seg || !valid_blocks)
4530                         continue;
4531                 if (valid_blocks > usable_blks_in_seg) {
4532                         f2fs_bug_on(sbi, 1);
4533                         continue;
4534                 }
4535                 mutex_lock(&dirty_i->seglist_lock);
4536                 __locate_dirty_segment(sbi, segno, DIRTY);
4537                 mutex_unlock(&dirty_i->seglist_lock);
4538         }
4539
4540         if (!__is_large_section(sbi))
4541                 return;
4542
4543         mutex_lock(&dirty_i->seglist_lock);
4544         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4545                 valid_blocks = get_valid_blocks(sbi, segno, true);
4546                 secno = GET_SEC_FROM_SEG(sbi, segno);
4547
4548                 if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi))
4549                         continue;
4550                 if (IS_CURSEC(sbi, secno))
4551                         continue;
4552                 set_bit(secno, dirty_i->dirty_secmap);
4553         }
4554         mutex_unlock(&dirty_i->seglist_lock);
4555 }
4556
4557 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4558 {
4559         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4560         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4561
4562         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4563         if (!dirty_i->victim_secmap)
4564                 return -ENOMEM;
4565
4566         dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4567         if (!dirty_i->pinned_secmap)
4568                 return -ENOMEM;
4569
4570         dirty_i->pinned_secmap_cnt = 0;
4571         dirty_i->enable_pin_section = true;
4572         return 0;
4573 }
4574
4575 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4576 {
4577         struct dirty_seglist_info *dirty_i;
4578         unsigned int bitmap_size, i;
4579
4580         /* allocate memory for dirty segments list information */
4581         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4582                                                                 GFP_KERNEL);
4583         if (!dirty_i)
4584                 return -ENOMEM;
4585
4586         SM_I(sbi)->dirty_info = dirty_i;
4587         mutex_init(&dirty_i->seglist_lock);
4588
4589         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4590
4591         for (i = 0; i < NR_DIRTY_TYPE; i++) {
4592                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4593                                                                 GFP_KERNEL);
4594                 if (!dirty_i->dirty_segmap[i])
4595                         return -ENOMEM;
4596         }
4597
4598         if (__is_large_section(sbi)) {
4599                 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4600                 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
4601                                                 bitmap_size, GFP_KERNEL);
4602                 if (!dirty_i->dirty_secmap)
4603                         return -ENOMEM;
4604         }
4605
4606         init_dirty_segmap(sbi);
4607         return init_victim_secmap(sbi);
4608 }
4609
4610 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4611 {
4612         int i;
4613
4614         /*
4615          * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4616          * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4617          */
4618         for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4619                 struct curseg_info *curseg = CURSEG_I(sbi, i);
4620                 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4621                 unsigned int blkofs = curseg->next_blkoff;
4622
4623                 if (f2fs_sb_has_readonly(sbi) &&
4624                         i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE)
4625                         continue;
4626
4627                 sanity_check_seg_type(sbi, curseg->seg_type);
4628
4629                 if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) {
4630                         f2fs_err(sbi,
4631                                  "Current segment has invalid alloc_type:%d",
4632                                  curseg->alloc_type);
4633                         f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
4634                         return -EFSCORRUPTED;
4635                 }
4636
4637                 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4638                         goto out;
4639
4640                 if (curseg->alloc_type == SSR)
4641                         continue;
4642
4643                 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4644                         if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4645                                 continue;
4646 out:
4647                         f2fs_err(sbi,
4648                                  "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4649                                  i, curseg->segno, curseg->alloc_type,
4650                                  curseg->next_blkoff, blkofs);
4651                         f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
4652                         return -EFSCORRUPTED;
4653                 }
4654         }
4655         return 0;
4656 }
4657
4658 #ifdef CONFIG_BLK_DEV_ZONED
4659
4660 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4661                                     struct f2fs_dev_info *fdev,
4662                                     struct blk_zone *zone)
4663 {
4664         unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno;
4665         block_t zone_block, wp_block, last_valid_block;
4666         unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4667         int i, s, b, ret;
4668         struct seg_entry *se;
4669
4670         if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4671                 return 0;
4672
4673         wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block);
4674         wp_segno = GET_SEGNO(sbi, wp_block);
4675         wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4676         zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4677         zone_segno = GET_SEGNO(sbi, zone_block);
4678         zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno);
4679
4680         if (zone_segno >= MAIN_SEGS(sbi))
4681                 return 0;
4682
4683         /*
4684          * Skip check of zones cursegs point to, since
4685          * fix_curseg_write_pointer() checks them.
4686          */
4687         for (i = 0; i < NO_CHECK_TYPE; i++)
4688                 if (zone_secno == GET_SEC_FROM_SEG(sbi,
4689                                                    CURSEG_I(sbi, i)->segno))
4690                         return 0;
4691
4692         /*
4693          * Get last valid block of the zone.
4694          */
4695         last_valid_block = zone_block - 1;
4696         for (s = sbi->segs_per_sec - 1; s >= 0; s--) {
4697                 segno = zone_segno + s;
4698                 se = get_seg_entry(sbi, segno);
4699                 for (b = sbi->blocks_per_seg - 1; b >= 0; b--)
4700                         if (f2fs_test_bit(b, se->cur_valid_map)) {
4701                                 last_valid_block = START_BLOCK(sbi, segno) + b;
4702                                 break;
4703                         }
4704                 if (last_valid_block >= zone_block)
4705                         break;
4706         }
4707
4708         /*
4709          * If last valid block is beyond the write pointer, report the
4710          * inconsistency. This inconsistency does not cause write error
4711          * because the zone will not be selected for write operation until
4712          * it get discarded. Just report it.
4713          */
4714         if (last_valid_block >= wp_block) {
4715                 f2fs_notice(sbi, "Valid block beyond write pointer: "
4716                             "valid block[0x%x,0x%x] wp[0x%x,0x%x]",
4717                             GET_SEGNO(sbi, last_valid_block),
4718                             GET_BLKOFF_FROM_SEG0(sbi, last_valid_block),
4719                             wp_segno, wp_blkoff);
4720                 return 0;
4721         }
4722
4723         /*
4724          * If there is no valid block in the zone and if write pointer is
4725          * not at zone start, reset the write pointer.
4726          */
4727         if (last_valid_block + 1 == zone_block && zone->wp != zone->start) {
4728                 f2fs_notice(sbi,
4729                             "Zone without valid block has non-zero write "
4730                             "pointer. Reset the write pointer: wp[0x%x,0x%x]",
4731                             wp_segno, wp_blkoff);
4732                 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
4733                                         zone->len >> log_sectors_per_block);
4734                 if (ret) {
4735                         f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4736                                  fdev->path, ret);
4737                         return ret;
4738                 }
4739         }
4740
4741         return 0;
4742 }
4743
4744 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
4745                                                   block_t zone_blkaddr)
4746 {
4747         int i;
4748
4749         for (i = 0; i < sbi->s_ndevs; i++) {
4750                 if (!bdev_is_zoned(FDEV(i).bdev))
4751                         continue;
4752                 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
4753                                 zone_blkaddr <= FDEV(i).end_blk))
4754                         return &FDEV(i);
4755         }
4756
4757         return NULL;
4758 }
4759
4760 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
4761                               void *data)
4762 {
4763         memcpy(data, zone, sizeof(struct blk_zone));
4764         return 0;
4765 }
4766
4767 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
4768 {
4769         struct curseg_info *cs = CURSEG_I(sbi, type);
4770         struct f2fs_dev_info *zbd;
4771         struct blk_zone zone;
4772         unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
4773         block_t cs_zone_block, wp_block;
4774         unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4775         sector_t zone_sector;
4776         int err;
4777
4778         cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4779         cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4780
4781         zbd = get_target_zoned_dev(sbi, cs_zone_block);
4782         if (!zbd)
4783                 return 0;
4784
4785         /* report zone for the sector the curseg points to */
4786         zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4787                 << log_sectors_per_block;
4788         err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4789                                   report_one_zone_cb, &zone);
4790         if (err != 1) {
4791                 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4792                          zbd->path, err);
4793                 return err;
4794         }
4795
4796         if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4797                 return 0;
4798
4799         wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
4800         wp_segno = GET_SEGNO(sbi, wp_block);
4801         wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4802         wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
4803
4804         if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
4805                 wp_sector_off == 0)
4806                 return 0;
4807
4808         f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
4809                     "curseg[0x%x,0x%x] wp[0x%x,0x%x]",
4810                     type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff);
4811
4812         f2fs_notice(sbi, "Assign new section to curseg[%d]: "
4813                     "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff);
4814
4815         f2fs_allocate_new_section(sbi, type, true);
4816
4817         /* check consistency of the zone curseg pointed to */
4818         if (check_zone_write_pointer(sbi, zbd, &zone))
4819                 return -EIO;
4820
4821         /* check newly assigned zone */
4822         cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4823         cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4824
4825         zbd = get_target_zoned_dev(sbi, cs_zone_block);
4826         if (!zbd)
4827                 return 0;
4828
4829         zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4830                 << log_sectors_per_block;
4831         err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4832                                   report_one_zone_cb, &zone);
4833         if (err != 1) {
4834                 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4835                          zbd->path, err);
4836                 return err;
4837         }
4838
4839         if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4840                 return 0;
4841
4842         if (zone.wp != zone.start) {
4843                 f2fs_notice(sbi,
4844                             "New zone for curseg[%d] is not yet discarded. "
4845                             "Reset the zone: curseg[0x%x,0x%x]",
4846                             type, cs->segno, cs->next_blkoff);
4847                 err = __f2fs_issue_discard_zone(sbi, zbd->bdev,
4848                                 zone_sector >> log_sectors_per_block,
4849                                 zone.len >> log_sectors_per_block);
4850                 if (err) {
4851                         f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4852                                  zbd->path, err);
4853                         return err;
4854                 }
4855         }
4856
4857         return 0;
4858 }
4859
4860 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4861 {
4862         int i, ret;
4863
4864         for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4865                 ret = fix_curseg_write_pointer(sbi, i);
4866                 if (ret)
4867                         return ret;
4868         }
4869
4870         return 0;
4871 }
4872
4873 struct check_zone_write_pointer_args {
4874         struct f2fs_sb_info *sbi;
4875         struct f2fs_dev_info *fdev;
4876 };
4877
4878 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
4879                                       void *data)
4880 {
4881         struct check_zone_write_pointer_args *args;
4882
4883         args = (struct check_zone_write_pointer_args *)data;
4884
4885         return check_zone_write_pointer(args->sbi, args->fdev, zone);
4886 }
4887
4888 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4889 {
4890         int i, ret;
4891         struct check_zone_write_pointer_args args;
4892
4893         for (i = 0; i < sbi->s_ndevs; i++) {
4894                 if (!bdev_is_zoned(FDEV(i).bdev))
4895                         continue;
4896
4897                 args.sbi = sbi;
4898                 args.fdev = &FDEV(i);
4899                 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
4900                                           check_zone_write_pointer_cb, &args);
4901                 if (ret < 0)
4902                         return ret;
4903         }
4904
4905         return 0;
4906 }
4907
4908 static bool is_conv_zone(struct f2fs_sb_info *sbi, unsigned int zone_idx,
4909                                                 unsigned int dev_idx)
4910 {
4911         if (!bdev_is_zoned(FDEV(dev_idx).bdev))
4912                 return true;
4913         return !test_bit(zone_idx, FDEV(dev_idx).blkz_seq);
4914 }
4915
4916 /* Return the zone index in the given device */
4917 static unsigned int get_zone_idx(struct f2fs_sb_info *sbi, unsigned int secno,
4918                                         int dev_idx)
4919 {
4920         block_t sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
4921
4922         return (sec_start_blkaddr - FDEV(dev_idx).start_blk) >>
4923                                                 sbi->log_blocks_per_blkz;
4924 }
4925
4926 /*
4927  * Return the usable segments in a section based on the zone's
4928  * corresponding zone capacity. Zone is equal to a section.
4929  */
4930 static inline unsigned int f2fs_usable_zone_segs_in_sec(
4931                 struct f2fs_sb_info *sbi, unsigned int segno)
4932 {
4933         unsigned int dev_idx, zone_idx;
4934
4935         dev_idx = f2fs_target_device_index(sbi, START_BLOCK(sbi, segno));
4936         zone_idx = get_zone_idx(sbi, GET_SEC_FROM_SEG(sbi, segno), dev_idx);
4937
4938         /* Conventional zone's capacity is always equal to zone size */
4939         if (is_conv_zone(sbi, zone_idx, dev_idx))
4940                 return sbi->segs_per_sec;
4941
4942         if (!sbi->unusable_blocks_per_sec)
4943                 return sbi->segs_per_sec;
4944
4945         /* Get the segment count beyond zone capacity block */
4946         return sbi->segs_per_sec - (sbi->unusable_blocks_per_sec >>
4947                                                 sbi->log_blocks_per_seg);
4948 }
4949
4950 /*
4951  * Return the number of usable blocks in a segment. The number of blocks
4952  * returned is always equal to the number of blocks in a segment for
4953  * segments fully contained within a sequential zone capacity or a
4954  * conventional zone. For segments partially contained in a sequential
4955  * zone capacity, the number of usable blocks up to the zone capacity
4956  * is returned. 0 is returned in all other cases.
4957  */
4958 static inline unsigned int f2fs_usable_zone_blks_in_seg(
4959                         struct f2fs_sb_info *sbi, unsigned int segno)
4960 {
4961         block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
4962         unsigned int zone_idx, dev_idx, secno;
4963
4964         secno = GET_SEC_FROM_SEG(sbi, segno);
4965         seg_start = START_BLOCK(sbi, segno);
4966         dev_idx = f2fs_target_device_index(sbi, seg_start);
4967         zone_idx = get_zone_idx(sbi, secno, dev_idx);
4968
4969         /*
4970          * Conventional zone's capacity is always equal to zone size,
4971          * so, blocks per segment is unchanged.
4972          */
4973         if (is_conv_zone(sbi, zone_idx, dev_idx))
4974                 return sbi->blocks_per_seg;
4975
4976         if (!sbi->unusable_blocks_per_sec)
4977                 return sbi->blocks_per_seg;
4978
4979         sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
4980         sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi);
4981
4982         /*
4983          * If segment starts before zone capacity and spans beyond
4984          * zone capacity, then usable blocks are from seg start to
4985          * zone capacity. If the segment starts after the zone capacity,
4986          * then there are no usable blocks.
4987          */
4988         if (seg_start >= sec_cap_blkaddr)
4989                 return 0;
4990         if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr)
4991                 return sec_cap_blkaddr - seg_start;
4992
4993         return sbi->blocks_per_seg;
4994 }
4995 #else
4996 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4997 {
4998         return 0;
4999 }
5000
5001 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5002 {
5003         return 0;
5004 }
5005
5006 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
5007                                                         unsigned int segno)
5008 {
5009         return 0;
5010 }
5011
5012 static inline unsigned int f2fs_usable_zone_segs_in_sec(struct f2fs_sb_info *sbi,
5013                                                         unsigned int segno)
5014 {
5015         return 0;
5016 }
5017 #endif
5018 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
5019                                         unsigned int segno)
5020 {
5021         if (f2fs_sb_has_blkzoned(sbi))
5022                 return f2fs_usable_zone_blks_in_seg(sbi, segno);
5023
5024         return sbi->blocks_per_seg;
5025 }
5026
5027 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi,
5028                                         unsigned int segno)
5029 {
5030         if (f2fs_sb_has_blkzoned(sbi))
5031                 return f2fs_usable_zone_segs_in_sec(sbi, segno);
5032
5033         return sbi->segs_per_sec;
5034 }
5035
5036 /*
5037  * Update min, max modified time for cost-benefit GC algorithm
5038  */
5039 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5040 {
5041         struct sit_info *sit_i = SIT_I(sbi);
5042         unsigned int segno;
5043
5044         down_write(&sit_i->sentry_lock);
5045
5046         sit_i->min_mtime = ULLONG_MAX;
5047
5048         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
5049                 unsigned int i;
5050                 unsigned long long mtime = 0;
5051
5052                 for (i = 0; i < sbi->segs_per_sec; i++)
5053                         mtime += get_seg_entry(sbi, segno + i)->mtime;
5054
5055                 mtime = div_u64(mtime, sbi->segs_per_sec);
5056
5057                 if (sit_i->min_mtime > mtime)
5058                         sit_i->min_mtime = mtime;
5059         }
5060         sit_i->max_mtime = get_mtime(sbi, false);
5061         sit_i->dirty_max_mtime = 0;
5062         up_write(&sit_i->sentry_lock);
5063 }
5064
5065 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
5066 {
5067         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5068         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
5069         struct f2fs_sm_info *sm_info;
5070         int err;
5071
5072         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
5073         if (!sm_info)
5074                 return -ENOMEM;
5075
5076         /* init sm info */
5077         sbi->sm_info = sm_info;
5078         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5079         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5080         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5081         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5082         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5083         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5084         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
5085         sm_info->rec_prefree_segments = sm_info->main_segments *
5086                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
5087         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5088                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5089
5090         if (!f2fs_lfs_mode(sbi))
5091                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
5092         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
5093         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
5094         sm_info->min_seq_blocks = sbi->blocks_per_seg;
5095         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
5096         sm_info->min_ssr_sections = reserved_sections(sbi);
5097
5098         INIT_LIST_HEAD(&sm_info->sit_entry_set);
5099
5100         init_f2fs_rwsem(&sm_info->curseg_lock);
5101
5102         if (!f2fs_readonly(sbi->sb)) {
5103                 err = f2fs_create_flush_cmd_control(sbi);
5104                 if (err)
5105                         return err;
5106         }
5107
5108         err = create_discard_cmd_control(sbi);
5109         if (err)
5110                 return err;
5111
5112         err = build_sit_info(sbi);
5113         if (err)
5114                 return err;
5115         err = build_free_segmap(sbi);
5116         if (err)
5117                 return err;
5118         err = build_curseg(sbi);
5119         if (err)
5120                 return err;
5121
5122         /* reinit free segmap based on SIT */
5123         err = build_sit_entries(sbi);
5124         if (err)
5125                 return err;
5126
5127         init_free_segmap(sbi);
5128         err = build_dirty_segmap(sbi);
5129         if (err)
5130                 return err;
5131
5132         err = sanity_check_curseg(sbi);
5133         if (err)
5134                 return err;
5135
5136         init_min_max_mtime(sbi);
5137         return 0;
5138 }
5139
5140 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5141                 enum dirty_type dirty_type)
5142 {
5143         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5144
5145         mutex_lock(&dirty_i->seglist_lock);
5146         kvfree(dirty_i->dirty_segmap[dirty_type]);
5147         dirty_i->nr_dirty[dirty_type] = 0;
5148         mutex_unlock(&dirty_i->seglist_lock);
5149 }
5150
5151 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5152 {
5153         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5154
5155         kvfree(dirty_i->pinned_secmap);
5156         kvfree(dirty_i->victim_secmap);
5157 }
5158
5159 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5160 {
5161         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5162         int i;
5163
5164         if (!dirty_i)
5165                 return;
5166
5167         /* discard pre-free/dirty segments list */
5168         for (i = 0; i < NR_DIRTY_TYPE; i++)
5169                 discard_dirty_segmap(sbi, i);
5170
5171         if (__is_large_section(sbi)) {
5172                 mutex_lock(&dirty_i->seglist_lock);
5173                 kvfree(dirty_i->dirty_secmap);
5174                 mutex_unlock(&dirty_i->seglist_lock);
5175         }
5176
5177         destroy_victim_secmap(sbi);
5178         SM_I(sbi)->dirty_info = NULL;
5179         kfree(dirty_i);
5180 }
5181
5182 static void destroy_curseg(struct f2fs_sb_info *sbi)
5183 {
5184         struct curseg_info *array = SM_I(sbi)->curseg_array;
5185         int i;
5186
5187         if (!array)
5188                 return;
5189         SM_I(sbi)->curseg_array = NULL;
5190         for (i = 0; i < NR_CURSEG_TYPE; i++) {
5191                 kfree(array[i].sum_blk);
5192                 kfree(array[i].journal);
5193         }
5194         kfree(array);
5195 }
5196
5197 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5198 {
5199         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5200
5201         if (!free_i)
5202                 return;
5203         SM_I(sbi)->free_info = NULL;
5204         kvfree(free_i->free_segmap);
5205         kvfree(free_i->free_secmap);
5206         kfree(free_i);
5207 }
5208
5209 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5210 {
5211         struct sit_info *sit_i = SIT_I(sbi);
5212
5213         if (!sit_i)
5214                 return;
5215
5216         if (sit_i->sentries)
5217                 kvfree(sit_i->bitmap);
5218         kfree(sit_i->tmp_map);
5219
5220         kvfree(sit_i->sentries);
5221         kvfree(sit_i->sec_entries);
5222         kvfree(sit_i->dirty_sentries_bitmap);
5223
5224         SM_I(sbi)->sit_info = NULL;
5225         kvfree(sit_i->sit_bitmap);
5226 #ifdef CONFIG_F2FS_CHECK_FS
5227         kvfree(sit_i->sit_bitmap_mir);
5228         kvfree(sit_i->invalid_segmap);
5229 #endif
5230         kfree(sit_i);
5231 }
5232
5233 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5234 {
5235         struct f2fs_sm_info *sm_info = SM_I(sbi);
5236
5237         if (!sm_info)
5238                 return;
5239         f2fs_destroy_flush_cmd_control(sbi, true);
5240         destroy_discard_cmd_control(sbi);
5241         destroy_dirty_segmap(sbi);
5242         destroy_curseg(sbi);
5243         destroy_free_segmap(sbi);
5244         destroy_sit_info(sbi);
5245         sbi->sm_info = NULL;
5246         kfree(sm_info);
5247 }
5248
5249 int __init f2fs_create_segment_manager_caches(void)
5250 {
5251         discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5252                         sizeof(struct discard_entry));
5253         if (!discard_entry_slab)
5254                 goto fail;
5255
5256         discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5257                         sizeof(struct discard_cmd));
5258         if (!discard_cmd_slab)
5259                 goto destroy_discard_entry;
5260
5261         sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5262                         sizeof(struct sit_entry_set));
5263         if (!sit_entry_set_slab)
5264                 goto destroy_discard_cmd;
5265
5266         revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry",
5267                         sizeof(struct revoke_entry));
5268         if (!revoke_entry_slab)
5269                 goto destroy_sit_entry_set;
5270         return 0;
5271
5272 destroy_sit_entry_set:
5273         kmem_cache_destroy(sit_entry_set_slab);
5274 destroy_discard_cmd:
5275         kmem_cache_destroy(discard_cmd_slab);
5276 destroy_discard_entry:
5277         kmem_cache_destroy(discard_entry_slab);
5278 fail:
5279         return -ENOMEM;
5280 }
5281
5282 void f2fs_destroy_segment_manager_caches(void)
5283 {
5284         kmem_cache_destroy(sit_entry_set_slab);
5285         kmem_cache_destroy(discard_cmd_slab);
5286         kmem_cache_destroy(discard_entry_slab);
5287         kmem_cache_destroy(revoke_entry_slab);
5288 }