fs/f2fs/segment.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/f2fs/segment.c
   4  *
   5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6  *             http://www.samsung.com/
   7  */
   8 #include <linux/fs.h>
   9 #include <linux/f2fs_fs.h>
  10 #include <linux/bio.h>
  11 #include <linux/blkdev.h>
  12 #include <linux/prefetch.h>
  13 #include <linux/kthread.h>
  14 #include <linux/swap.h>
  15 #include <linux/timer.h>
  16 #include <linux/freezer.h>
  17 #include <linux/sched/signal.h>
  18
  19 #include "f2fs.h"
  20 #include "segment.h"
  21 #include "node.h"
  22 #include "gc.h"
  23 #include "trace.h"
  24 #include <trace/events/f2fs.h>
  25
  26 #define __reverse_ffz(x) __reverse_ffs(~(x))
  27
  28 static struct kmem_cache *discard_entry_slab;
  29 static struct kmem_cache *discard_cmd_slab;
  30 static struct kmem_cache *sit_entry_set_slab;
  31 static struct kmem_cache *inmem_entry_slab;
  32
  33 static unsigned long __reverse_ulong(unsigned char *str)
  34 {
  35         unsigned long tmp = 0;
  36         int shift = 24, idx = 0;
  37
  38 #if BITS_PER_LONG == 64
  39         shift = 56;
  40 #endif
  41         while (shift >= 0) {
  42                 tmp |= (unsigned long)str[idx++] << shift;
  43                 shift -= BITS_PER_BYTE;
  44         }
  45         return tmp;
  46 }
  47
  48 /*
  49  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  50  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  51  */
  52 static inline unsigned long __reverse_ffs(unsigned long word)
  53 {
  54         int num = 0;
  55
  56 #if BITS_PER_LONG == 64
  57         if ((word & 0xffffffff00000000UL) == 0)
  58                 num += 32;
  59         else
  60                 word >>= 32;
  61 #endif
  62         if ((word & 0xffff0000) == 0)
  63                 num += 16;
  64         else
  65                 word >>= 16;
  66
  67         if ((word & 0xff00) == 0)
  68                 num += 8;
  69         else
  70                 word >>= 8;
  71
  72         if ((word & 0xf0) == 0)
  73                 num += 4;
  74         else
  75                 word >>= 4;
  76
  77         if ((word & 0xc) == 0)
  78                 num += 2;
  79         else
  80                 word >>= 2;
  81
  82         if ((word & 0x2) == 0)
  83                 num += 1;
  84         return num;
  85 }
  86
  87 /*
  88  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  89  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  90  * @size must be integral times of unsigned long.
  91  * Example:
  92  *                             MSB <--> LSB
  93  *   f2fs_set_bit(0, bitmap) => 1000 0000
  94  *   f2fs_set_bit(7, bitmap) => 0000 0001
  95  */
  96 static unsigned long __find_rev_next_bit(const unsigned long *addr,
  97                         unsigned long size, unsigned long offset)
  98 {
  99         const unsigned long *p = addr + BIT_WORD(offset);
 100         unsigned long result = size;
 101         unsigned long tmp;
 102
 103         if (offset >= size)
 104                 return size;
 105
 106         size -= (offset & ~(BITS_PER_LONG - 1));
 107         offset %= BITS_PER_LONG;
 108
 109         while (1) {
 110                 if (*p == 0)
 111                         goto pass;
 112
 113                 tmp = __reverse_ulong((unsigned char *)p);
 114
 115                 tmp &= ~0UL >> offset;
 116                 if (size < BITS_PER_LONG)
 117                         tmp &= (~0UL << (BITS_PER_LONG - size));
 118                 if (tmp)
 119                         goto found;
 120 pass:
 121                 if (size <= BITS_PER_LONG)
 122                         break;
 123                 size -= BITS_PER_LONG;
 124                 offset = 0;
 125                 p++;
 126         }
 127         return result;
 128 found:
 129         return result - size + __reverse_ffs(tmp);
 130 }
 131
 132 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 133                         unsigned long size, unsigned long offset)
 134 {
 135         const unsigned long *p = addr + BIT_WORD(offset);
 136         unsigned long result = size;
 137         unsigned long tmp;
 138
 139         if (offset >= size)
 140                 return size;
 141
 142         size -= (offset & ~(BITS_PER_LONG - 1));
 143         offset %= BITS_PER_LONG;
 144
 145         while (1) {
 146                 if (*p == ~0UL)
 147                         goto pass;
 148
 149                 tmp = __reverse_ulong((unsigned char *)p);
 150
 151                 if (offset)
 152                         tmp |= ~0UL << (BITS_PER_LONG - offset);
 153                 if (size < BITS_PER_LONG)
 154                         tmp |= ~0UL >> size;
 155                 if (tmp != ~0UL)
 156                         goto found;
 157 pass:
 158                 if (size <= BITS_PER_LONG)
 159                         break;
 160                 size -= BITS_PER_LONG;
 161                 offset = 0;
 162                 p++;
 163         }
 164         return result;
 165 found:
 166         return result - size + __reverse_ffz(tmp);
 167 }
 168
 169 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
 170 {
 171         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 172         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 173         int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
 174
 175         if (test_opt(sbi, LFS))
 176                 return false;
 177         if (sbi->gc_mode == GC_URGENT)
 178                 return true;
 179         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
 180                 return true;
 181
 182         return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
 183                         SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
 184 }
 185
 186 void f2fs_register_inmem_page(struct inode *inode, struct page *page)
 187 {
 188         struct inmem_pages *new;
 189
 190         f2fs_trace_pid(page);
 191
 192         f2fs_set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
 193
 194         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 195
 196         /* add atomic page indices to the list */
 197         new->page = page;
 198         INIT_LIST_HEAD(&new->list);
 199
 200         /* increase reference count with clean state */
 201         get_page(page);
 202         mutex_lock(&F2FS_I(inode)->inmem_lock);
 203         list_add_tail(&new->list, &F2FS_I(inode)->inmem_pages);
 204         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 205         mutex_unlock(&F2FS_I(inode)->inmem_lock);
 206
 207         trace_f2fs_register_inmem_page(page, INMEM);
 208 }
 209
 210 static int __revoke_inmem_pages(struct inode *inode,
 211                                 struct list_head *head, bool drop, bool recover,
 212                                 bool trylock)
 213 {
 214         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 215         struct inmem_pages *cur, *tmp;
 216         int err = 0;
 217
 218         list_for_each_entry_safe(cur, tmp, head, list) {
 219                 struct page *page = cur->page;
 220
 221                 if (drop)
 222                         trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 223
 224                 if (trylock) {
 225                         /*
 226                          * to avoid deadlock in between page lock and
 227                          * inmem_lock.
 228                          */
 229                         if (!trylock_page(page))
 230                                 continue;
 231                 } else {
 232                         lock_page(page);
 233                 }
 234
 235                 f2fs_wait_on_page_writeback(page, DATA, true, true);
 236
 237                 if (recover) {
 238                         struct dnode_of_data dn;
 239                         struct node_info ni;
 240
 241                         trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 242 retry:
 243                         set_new_dnode(&dn, inode, NULL, NULL, 0);
 244                         err = f2fs_get_dnode_of_data(&dn, page->index,
 245                                                                 LOOKUP_NODE);
 246                         if (err) {
 247                                 if (err == -ENOMEM) {
 248                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 249                                         cond_resched();
 250                                         goto retry;
 251                                 }
 252                                 err = -EAGAIN;
 253                                 goto next;
 254                         }
 255
 256                         err = f2fs_get_node_info(sbi, dn.nid, &ni);
 257                         if (err) {
 258                                 f2fs_put_dnode(&dn);
 259                                 return err;
 260                         }
 261
 262                         if (cur->old_addr == NEW_ADDR) {
 263                                 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
 264                                 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
 265                         } else
 266                                 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 267                                         cur->old_addr, ni.version, true, true);
 268                         f2fs_put_dnode(&dn);
 269                 }
 270 next:
 271                 /* we don't need to invalidate this in the sccessful status */
 272                 if (drop || recover) {
 273                         ClearPageUptodate(page);
 274                         clear_cold_data(page);
 275                 }
 276                 f2fs_clear_page_private(page);
 277                 f2fs_put_page(page, 1);
 278
 279                 list_del(&cur->list);
 280                 kmem_cache_free(inmem_entry_slab, cur);
 281                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 282         }
 283         return err;
 284 }
 285
 286 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
 287 {
 288         struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
 289         struct inode *inode;
 290         struct f2fs_inode_info *fi;
 291 next:
 292         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 293         if (list_empty(head)) {
 294                 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 295                 return;
 296         }
 297         fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
 298         inode = igrab(&fi->vfs_inode);
 299         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 300
 301         if (inode) {
 302                 if (gc_failure) {
 303                         if (fi->i_gc_failures[GC_FAILURE_ATOMIC])
 304                                 goto drop;
 305                         goto skip;
 306                 }
 307 drop:
 308                 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
 309                 f2fs_drop_inmem_pages(inode);
 310                 iput(inode);
 311         }
 312 skip:
 313         congestion_wait(BLK_RW_ASYNC, HZ/50);
 314         cond_resched();
 315         goto next;
 316 }
 317
 318 void f2fs_drop_inmem_pages(struct inode *inode)
 319 {
 320         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 321         struct f2fs_inode_info *fi = F2FS_I(inode);
 322
 323         while (!list_empty(&fi->inmem_pages)) {
 324                 mutex_lock(&fi->inmem_lock);
 325                 __revoke_inmem_pages(inode, &fi->inmem_pages,
 326                                                 true, false, true);
 327                 mutex_unlock(&fi->inmem_lock);
 328         }
 329
 330         clear_inode_flag(inode, FI_ATOMIC_FILE);
 331         fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
 332         stat_dec_atomic_write(inode);
 333
 334         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 335         if (!list_empty(&fi->inmem_ilist))
 336                 list_del_init(&fi->inmem_ilist);
 337         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 338 }
 339
 340 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
 341 {
 342         struct f2fs_inode_info *fi = F2FS_I(inode);
 343         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 344         struct list_head *head = &fi->inmem_pages;
 345         struct inmem_pages *cur = NULL;
 346
 347         f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
 348
 349         mutex_lock(&fi->inmem_lock);
 350         list_for_each_entry(cur, head, list) {
 351                 if (cur->page == page)
 352                         break;
 353         }
 354
 355         f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
 356         list_del(&cur->list);
 357         mutex_unlock(&fi->inmem_lock);
 358
 359         dec_page_count(sbi, F2FS_INMEM_PAGES);
 360         kmem_cache_free(inmem_entry_slab, cur);
 361
 362         ClearPageUptodate(page);
 363         f2fs_clear_page_private(page);
 364         f2fs_put_page(page, 0);
 365
 366         trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 367 }
 368
 369 static int __f2fs_commit_inmem_pages(struct inode *inode)
 370 {
 371         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 372         struct f2fs_inode_info *fi = F2FS_I(inode);
 373         struct inmem_pages *cur, *tmp;
 374         struct f2fs_io_info fio = {
 375                 .sbi = sbi,
 376                 .ino = inode->i_ino,
 377                 .type = DATA,
 378                 .op = REQ_OP_WRITE,
 379                 .op_flags = REQ_SYNC | REQ_PRIO,
 380                 .io_type = FS_DATA_IO,
 381         };
 382         struct list_head revoke_list;
 383         bool submit_bio = false;
 384         int err = 0;
 385
 386         INIT_LIST_HEAD(&revoke_list);
 387
 388         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 389                 struct page *page = cur->page;
 390
 391                 lock_page(page);
 392                 if (page->mapping == inode->i_mapping) {
 393                         trace_f2fs_commit_inmem_page(page, INMEM);
 394
 395                         f2fs_wait_on_page_writeback(page, DATA, true, true);
 396
 397                         set_page_dirty(page);
 398                         if (clear_page_dirty_for_io(page)) {
 399                                 inode_dec_dirty_pages(inode);
 400                                 f2fs_remove_dirty_inode(inode);
 401                         }
 402 retry:
 403                         fio.page = page;
 404                         fio.old_blkaddr = NULL_ADDR;
 405                         fio.encrypted_page = NULL;
 406                         fio.need_lock = LOCK_DONE;
 407                         err = f2fs_do_write_data_page(&fio);
 408                         if (err) {
 409                                 if (err == -ENOMEM) {
 410                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 411                                         cond_resched();
 412                                         goto retry;
 413                                 }
 414                                 unlock_page(page);
 415                                 break;
 416                         }
 417                         /* record old blkaddr for revoking */
 418                         cur->old_addr = fio.old_blkaddr;
 419                         submit_bio = true;
 420                 }
 421                 unlock_page(page);
 422                 list_move_tail(&cur->list, &revoke_list);
 423         }
 424
 425         if (submit_bio)
 426                 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
 427
 428         if (err) {
 429                 /*
 430                  * try to revoke all committed pages, but still we could fail
 431                  * due to no memory or other reason, if that happened, EAGAIN
 432                  * will be returned, which means in such case, transaction is
 433                  * already not integrity, caller should use journal to do the
 434                  * recovery or rewrite & commit last transaction. For other
 435                  * error number, revoking was done by filesystem itself.
 436                  */
 437                 err = __revoke_inmem_pages(inode, &revoke_list,
 438                                                 false, true, false);
 439
 440                 /* drop all uncommitted pages */
 441                 __revoke_inmem_pages(inode, &fi->inmem_pages,
 442                                                 true, false, false);
 443         } else {
 444                 __revoke_inmem_pages(inode, &revoke_list,
 445                                                 false, false, false);
 446         }
 447
 448         return err;
 449 }
 450
 451 int f2fs_commit_inmem_pages(struct inode *inode)
 452 {
 453         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 454         struct f2fs_inode_info *fi = F2FS_I(inode);
 455         int err;
 456
 457         f2fs_balance_fs(sbi, true);
 458
 459         down_write(&fi->i_gc_rwsem[WRITE]);
 460
 461         f2fs_lock_op(sbi);
 462         set_inode_flag(inode, FI_ATOMIC_COMMIT);
 463
 464         mutex_lock(&fi->inmem_lock);
 465         err = __f2fs_commit_inmem_pages(inode);
 466         mutex_unlock(&fi->inmem_lock);
 467
 468         clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 469
 470         f2fs_unlock_op(sbi);
 471         up_write(&fi->i_gc_rwsem[WRITE]);
 472
 473         return err;
 474 }
 475
 476 /*
 477  * This function balances dirty node and dentry pages.
 478  * In addition, it controls garbage collection.
 479  */
 480 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 481 {
 482         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 483                 f2fs_show_injection_info(FAULT_CHECKPOINT);
 484                 f2fs_stop_checkpoint(sbi, false);
 485         }
 486
 487         /* balance_fs_bg is able to be pending */
 488         if (need && excess_cached_nats(sbi))
 489                 f2fs_balance_fs_bg(sbi);
 490
 491         if (!f2fs_is_checkpoint_ready(sbi))
 492                 return;
 493
 494         /*
 495          * We should do GC or end up with checkpoint, if there are so many dirty
 496          * dir/node pages without enough free segments.
 497          */
 498         if (has_not_enough_free_secs(sbi, 0, 0)) {
 499                 mutex_lock(&sbi->gc_mutex);
 500                 f2fs_gc(sbi, false, false, NULL_SEGNO);
 501         }
 502 }
 503
 504 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 505 {
 506         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 507                 return;
 508
 509         /* try to shrink extent cache when there is no enough memory */
 510         if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
 511                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 512
 513         /* check the # of cached NAT entries */
 514         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
 515                 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 516
 517         if (!f2fs_available_free_memory(sbi, FREE_NIDS))
 518                 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
 519         else
 520                 f2fs_build_free_nids(sbi, false, false);
 521
 522         if (!is_idle(sbi, REQ_TIME) &&
 523                 (!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
 524                 return;
 525
 526         /* checkpoint is the only way to shrink partial cached entries */
 527         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
 528                         !f2fs_available_free_memory(sbi, INO_ENTRIES) ||
 529                         excess_prefree_segs(sbi) ||
 530                         excess_dirty_nats(sbi) ||
 531                         excess_dirty_nodes(sbi) ||
 532                         f2fs_time_over(sbi, CP_TIME)) {
 533                 if (test_opt(sbi, DATA_FLUSH)) {
 534                         struct blk_plug plug;
 535
 536                         mutex_lock(&sbi->flush_lock);
 537
 538                         blk_start_plug(&plug);
 539                         f2fs_sync_dirty_inodes(sbi, FILE_INODE);
 540                         blk_finish_plug(&plug);
 541
 542                         mutex_unlock(&sbi->flush_lock);
 543                 }
 544                 f2fs_sync_fs(sbi->sb, true);
 545                 stat_inc_bg_cp_count(sbi->stat_info);
 546         }
 547 }
 548
 549 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 550                                 struct block_device *bdev)
 551 {
 552         struct bio *bio;
 553         int ret;
 554
 555         bio = f2fs_bio_alloc(sbi, 0, false);
 556         if (!bio)
 557                 return -ENOMEM;
 558
 559         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
 560         bio_set_dev(bio, bdev);
 561         ret = submit_bio_wait(bio);
 562         bio_put(bio);
 563
 564         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 565                                 test_opt(sbi, FLUSH_MERGE), ret);
 566         return ret;
 567 }
 568
 569 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 570 {
 571         int ret = 0;
 572         int i;
 573
 574         if (!f2fs_is_multi_device(sbi))
 575                 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 576
 577         for (i = 0; i < sbi->s_ndevs; i++) {
 578                 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
 579                         continue;
 580                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 581                 if (ret)
 582                         break;
 583         }
 584         return ret;
 585 }
 586
 587 static int issue_flush_thread(void *data)
 588 {
 589         struct f2fs_sb_info *sbi = data;
 590         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 591         wait_queue_head_t *q = &fcc->flush_wait_queue;
 592 repeat:
 593         if (kthread_should_stop())
 594                 return 0;
 595
 596         sb_start_intwrite(sbi->sb);
 597
 598         if (!llist_empty(&fcc->issue_list)) {
 599                 struct flush_cmd *cmd, *next;
 600                 int ret;
 601
 602                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 603                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 604
 605                 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 606
 607                 ret = submit_flush_wait(sbi, cmd->ino);
 608                 atomic_inc(&fcc->issued_flush);
 609
 610                 llist_for_each_entry_safe(cmd, next,
 611                                           fcc->dispatch_list, llnode) {
 612                         cmd->ret = ret;
 613                         complete(&cmd->wait);
 614                 }
 615                 fcc->dispatch_list = NULL;
 616         }
 617
 618         sb_end_intwrite(sbi->sb);
 619
 620         wait_event_interruptible(*q,
 621                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 622         goto repeat;
 623 }
 624
 625 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 626 {
 627         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 628         struct flush_cmd cmd;
 629         int ret;
 630
 631         if (test_opt(sbi, NOBARRIER))
 632                 return 0;
 633
 634         if (!test_opt(sbi, FLUSH_MERGE)) {
 635                 atomic_inc(&fcc->queued_flush);
 636                 ret = submit_flush_wait(sbi, ino);
 637                 atomic_dec(&fcc->queued_flush);
 638                 atomic_inc(&fcc->issued_flush);
 639                 return ret;
 640         }
 641
 642         if (atomic_inc_return(&fcc->queued_flush) == 1 ||
 643             f2fs_is_multi_device(sbi)) {
 644                 ret = submit_flush_wait(sbi, ino);
 645                 atomic_dec(&fcc->queued_flush);
 646
 647                 atomic_inc(&fcc->issued_flush);
 648                 return ret;
 649         }
 650
 651         cmd.ino = ino;
 652         init_completion(&cmd.wait);
 653
 654         llist_add(&cmd.llnode, &fcc->issue_list);
 655
 656         /* update issue_list before we wake up issue_flush thread */
 657         smp_mb();
 658
 659         if (waitqueue_active(&fcc->flush_wait_queue))
 660                 wake_up(&fcc->flush_wait_queue);
 661
 662         if (fcc->f2fs_issue_flush) {
 663                 wait_for_completion(&cmd.wait);
 664                 atomic_dec(&fcc->queued_flush);
 665         } else {
 666                 struct llist_node *list;
 667
 668                 list = llist_del_all(&fcc->issue_list);
 669                 if (!list) {
 670                         wait_for_completion(&cmd.wait);
 671                         atomic_dec(&fcc->queued_flush);
 672                 } else {
 673                         struct flush_cmd *tmp, *next;
 674
 675                         ret = submit_flush_wait(sbi, ino);
 676
 677                         llist_for_each_entry_safe(tmp, next, list, llnode) {
 678                                 if (tmp == &cmd) {
 679                                         cmd.ret = ret;
 680                                         atomic_dec(&fcc->queued_flush);
 681                                         continue;
 682                                 }
 683                                 tmp->ret = ret;
 684                                 complete(&tmp->wait);
 685                         }
 686                 }
 687         }
 688
 689         return cmd.ret;
 690 }
 691
 692 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
 693 {
 694         dev_t dev = sbi->sb->s_bdev->bd_dev;
 695         struct flush_cmd_control *fcc;
 696         int err = 0;
 697
 698         if (SM_I(sbi)->fcc_info) {
 699                 fcc = SM_I(sbi)->fcc_info;
 700                 if (fcc->f2fs_issue_flush)
 701                         return err;
 702                 goto init_thread;
 703         }
 704
 705         fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
 706         if (!fcc)
 707                 return -ENOMEM;
 708         atomic_set(&fcc->issued_flush, 0);
 709         atomic_set(&fcc->queued_flush, 0);
 710         init_waitqueue_head(&fcc->flush_wait_queue);
 711         init_llist_head(&fcc->issue_list);
 712         SM_I(sbi)->fcc_info = fcc;
 713         if (!test_opt(sbi, FLUSH_MERGE))
 714                 return err;
 715
 716 init_thread:
 717         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 718                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 719         if (IS_ERR(fcc->f2fs_issue_flush)) {
 720                 err = PTR_ERR(fcc->f2fs_issue_flush);
 721                 kvfree(fcc);
 722                 SM_I(sbi)->fcc_info = NULL;
 723                 return err;
 724         }
 725
 726         return err;
 727 }
 728
 729 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 730 {
 731         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 732
 733         if (fcc && fcc->f2fs_issue_flush) {
 734                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 735
 736                 fcc->f2fs_issue_flush = NULL;
 737                 kthread_stop(flush_thread);
 738         }
 739         if (free) {
 740                 kvfree(fcc);
 741                 SM_I(sbi)->fcc_info = NULL;
 742         }
 743 }
 744
 745 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
 746 {
 747         int ret = 0, i;
 748
 749         if (!f2fs_is_multi_device(sbi))
 750                 return 0;
 751
 752         for (i = 1; i < sbi->s_ndevs; i++) {
 753                 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
 754                         continue;
 755                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 756                 if (ret)
 757                         break;
 758
 759                 spin_lock(&sbi->dev_lock);
 760                 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
 761                 spin_unlock(&sbi->dev_lock);
 762         }
 763
 764         return ret;
 765 }
 766
 767 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 768                 enum dirty_type dirty_type)
 769 {
 770         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 771
 772         /* need not be added */
 773         if (IS_CURSEG(sbi, segno))
 774                 return;
 775
 776         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 777                 dirty_i->nr_dirty[dirty_type]++;
 778
 779         if (dirty_type == DIRTY) {
 780                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 781                 enum dirty_type t = sentry->type;
 782
 783                 if (unlikely(t >= DIRTY)) {
 784                         f2fs_bug_on(sbi, 1);
 785                         return;
 786                 }
 787                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 788                         dirty_i->nr_dirty[t]++;
 789         }
 790 }
 791
 792 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 793                 enum dirty_type dirty_type)
 794 {
 795         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 796
 797         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 798                 dirty_i->nr_dirty[dirty_type]--;
 799
 800         if (dirty_type == DIRTY) {
 801                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 802                 enum dirty_type t = sentry->type;
 803
 804                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 805                         dirty_i->nr_dirty[t]--;
 806
 807                 if (get_valid_blocks(sbi, segno, true) == 0) {
 808                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 809                                                 dirty_i->victim_secmap);
 810 #ifdef CONFIG_F2FS_CHECK_FS
 811                         clear_bit(segno, SIT_I(sbi)->invalid_segmap);
 812 #endif
 813                 }
 814         }
 815 }
 816
 817 /*
 818  * Should not occur error such as -ENOMEM.
 819  * Adding dirty entry into seglist is not critical operation.
 820  * If a given segment is one of current working segments, it won't be added.
 821  */
 822 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 823 {
 824         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 825         unsigned short valid_blocks, ckpt_valid_blocks;
 826
 827         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 828                 return;
 829
 830         mutex_lock(&dirty_i->seglist_lock);
 831
 832         valid_blocks = get_valid_blocks(sbi, segno, false);
 833         ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno);
 834
 835         if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
 836                                 ckpt_valid_blocks == sbi->blocks_per_seg)) {
 837                 __locate_dirty_segment(sbi, segno, PRE);
 838                 __remove_dirty_segment(sbi, segno, DIRTY);
 839         } else if (valid_blocks < sbi->blocks_per_seg) {
 840                 __locate_dirty_segment(sbi, segno, DIRTY);
 841         } else {
 842                 /* Recovery routine with SSR needs this */
 843                 __remove_dirty_segment(sbi, segno, DIRTY);
 844         }
 845
 846         mutex_unlock(&dirty_i->seglist_lock);
 847 }
 848
 849 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
 850 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
 851 {
 852         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 853         unsigned int segno;
 854
 855         mutex_lock(&dirty_i->seglist_lock);
 856         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 857                 if (get_valid_blocks(sbi, segno, false))
 858                         continue;
 859                 if (IS_CURSEG(sbi, segno))
 860                         continue;
 861                 __locate_dirty_segment(sbi, segno, PRE);
 862                 __remove_dirty_segment(sbi, segno, DIRTY);
 863         }
 864         mutex_unlock(&dirty_i->seglist_lock);
 865 }
 866
 867 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
 868 {
 869         int ovp_hole_segs =
 870                 (overprovision_segments(sbi) - reserved_segments(sbi));
 871         block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
 872         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 873         block_t holes[2] = {0, 0};      /* DATA and NODE */
 874         block_t unusable;
 875         struct seg_entry *se;
 876         unsigned int segno;
 877
 878         mutex_lock(&dirty_i->seglist_lock);
 879         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 880                 se = get_seg_entry(sbi, segno);
 881                 if (IS_NODESEG(se->type))
 882                         holes[NODE] += sbi->blocks_per_seg - se->valid_blocks;
 883                 else
 884                         holes[DATA] += sbi->blocks_per_seg - se->valid_blocks;
 885         }
 886         mutex_unlock(&dirty_i->seglist_lock);
 887
 888         unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
 889         if (unusable > ovp_holes)
 890                 return unusable - ovp_holes;
 891         return 0;
 892 }
 893
 894 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
 895 {
 896         int ovp_hole_segs =
 897                 (overprovision_segments(sbi) - reserved_segments(sbi));
 898         if (unusable > F2FS_OPTION(sbi).unusable_cap)
 899                 return -EAGAIN;
 900         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
 901                 dirty_segments(sbi) > ovp_hole_segs)
 902                 return -EAGAIN;
 903         return 0;
 904 }
 905
 906 /* This is only used by SBI_CP_DISABLED */
 907 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
 908 {
 909         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 910         unsigned int segno = 0;
 911
 912         mutex_lock(&dirty_i->seglist_lock);
 913         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 914                 if (get_valid_blocks(sbi, segno, false))
 915                         continue;
 916                 if (get_ckpt_valid_blocks(sbi, segno))
 917                         continue;
 918                 mutex_unlock(&dirty_i->seglist_lock);
 919                 return segno;
 920         }
 921         mutex_unlock(&dirty_i->seglist_lock);
 922         return NULL_SEGNO;
 923 }
 924
 925 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 926                 struct block_device *bdev, block_t lstart,
 927                 block_t start, block_t len)
 928 {
 929         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 930         struct list_head *pend_list;
 931         struct discard_cmd *dc;
 932
 933         f2fs_bug_on(sbi, !len);
 934
 935         pend_list = &dcc->pend_list[plist_idx(len)];
 936
 937         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 938         INIT_LIST_HEAD(&dc->list);
 939         dc->bdev = bdev;
 940         dc->lstart = lstart;
 941         dc->start = start;
 942         dc->len = len;
 943         dc->ref = 0;
 944         dc->state = D_PREP;
 945         dc->queued = 0;
 946         dc->error = 0;
 947         init_completion(&dc->wait);
 948         list_add_tail(&dc->list, pend_list);
 949         spin_lock_init(&dc->lock);
 950         dc->bio_ref = 0;
 951         atomic_inc(&dcc->discard_cmd_cnt);
 952         dcc->undiscard_blks += len;
 953
 954         return dc;
 955 }
 956
 957 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 958                                 struct block_device *bdev, block_t lstart,
 959                                 block_t start, block_t len,
 960                                 struct rb_node *parent, struct rb_node **p,
 961                                 bool leftmost)
 962 {
 963         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 964         struct discard_cmd *dc;
 965
 966         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
 967
 968         rb_link_node(&dc->rb_node, parent, p);
 969         rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
 970
 971         return dc;
 972 }
 973
 974 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
 975                                                         struct discard_cmd *dc)
 976 {
 977         if (dc->state == D_DONE)
 978                 atomic_sub(dc->queued, &dcc->queued_discard);
 979
 980         list_del(&dc->list);
 981         rb_erase_cached(&dc->rb_node, &dcc->root);
 982         dcc->undiscard_blks -= dc->len;
 983
 984         kmem_cache_free(discard_cmd_slab, dc);
 985
 986         atomic_dec(&dcc->discard_cmd_cnt);
 987 }
 988
 989 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
 990                                                         struct discard_cmd *dc)
 991 {
 992         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 993         unsigned long flags;
 994
 995         trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
 996
 997         spin_lock_irqsave(&dc->lock, flags);
 998         if (dc->bio_ref) {
 999                 spin_unlock_irqrestore(&dc->lock, flags);
1000                 return;
1001         }
1002         spin_unlock_irqrestore(&dc->lock, flags);
1003
1004         f2fs_bug_on(sbi, dc->ref);
1005
1006         if (dc->error == -EOPNOTSUPP)
1007                 dc->error = 0;
1008
1009         if (dc->error)
1010                 printk_ratelimited(
1011                         "%sF2FS-fs: Issue discard(%u, %u, %u) failed, ret: %d",
1012                         KERN_INFO, dc->lstart, dc->start, dc->len, dc->error);
1013         __detach_discard_cmd(dcc, dc);
1014 }
1015
1016 static void f2fs_submit_discard_endio(struct bio *bio)
1017 {
1018         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1019         unsigned long flags;
1020
1021         dc->error = blk_status_to_errno(bio->bi_status);
1022
1023         spin_lock_irqsave(&dc->lock, flags);
1024         dc->bio_ref--;
1025         if (!dc->bio_ref && dc->state == D_SUBMIT) {
1026                 dc->state = D_DONE;
1027                 complete_all(&dc->wait);
1028         }
1029         spin_unlock_irqrestore(&dc->lock, flags);
1030         bio_put(bio);
1031 }
1032
1033 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1034                                 block_t start, block_t end)
1035 {
1036 #ifdef CONFIG_F2FS_CHECK_FS
1037         struct seg_entry *sentry;
1038         unsigned int segno;
1039         block_t blk = start;
1040         unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1041         unsigned long *map;
1042
1043         while (blk < end) {
1044                 segno = GET_SEGNO(sbi, blk);
1045                 sentry = get_seg_entry(sbi, segno);
1046                 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1047
1048                 if (end < START_BLOCK(sbi, segno + 1))
1049                         size = GET_BLKOFF_FROM_SEG0(sbi, end);
1050                 else
1051                         size = max_blocks;
1052                 map = (unsigned long *)(sentry->cur_valid_map);
1053                 offset = __find_rev_next_bit(map, size, offset);
1054                 f2fs_bug_on(sbi, offset != size);
1055                 blk = START_BLOCK(sbi, segno + 1);
1056         }
1057 #endif
1058 }
1059
1060 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1061                                 struct discard_policy *dpolicy,
1062                                 int discard_type, unsigned int granularity)
1063 {
1064         /* common policy */
1065         dpolicy->type = discard_type;
1066         dpolicy->sync = true;
1067         dpolicy->ordered = false;
1068         dpolicy->granularity = granularity;
1069
1070         dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1071         dpolicy->io_aware_gran = MAX_PLIST_NUM;
1072         dpolicy->timeout = 0;
1073
1074         if (discard_type == DPOLICY_BG) {
1075                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1076                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1077                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1078                 dpolicy->io_aware = true;
1079                 dpolicy->sync = false;
1080                 dpolicy->ordered = true;
1081                 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1082                         dpolicy->granularity = 1;
1083                         dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1084                 }
1085         } else if (discard_type == DPOLICY_FORCE) {
1086                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1087                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1088                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1089                 dpolicy->io_aware = false;
1090         } else if (discard_type == DPOLICY_FSTRIM) {
1091                 dpolicy->io_aware = false;
1092         } else if (discard_type == DPOLICY_UMOUNT) {
1093                 dpolicy->max_requests = UINT_MAX;
1094                 dpolicy->io_aware = false;
1095                 /* we need to issue all to keep CP_TRIMMED_FLAG */
1096                 dpolicy->granularity = 1;
1097         }
1098 }
1099
1100 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1101                                 struct block_device *bdev, block_t lstart,
1102                                 block_t start, block_t len);
1103 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1104 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1105                                                 struct discard_policy *dpolicy,
1106                                                 struct discard_cmd *dc,
1107                                                 unsigned int *issued)
1108 {
1109         struct block_device *bdev = dc->bdev;
1110         struct request_queue *q = bdev_get_queue(bdev);
1111         unsigned int max_discard_blocks =
1112                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1113         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1114         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1115                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1116         int flag = dpolicy->sync ? REQ_SYNC : 0;
1117         block_t lstart, start, len, total_len;
1118         int err = 0;
1119
1120         if (dc->state != D_PREP)
1121                 return 0;
1122
1123         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1124                 return 0;
1125
1126         trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1127
1128         lstart = dc->lstart;
1129         start = dc->start;
1130         len = dc->len;
1131         total_len = len;
1132
1133         dc->len = 0;
1134
1135         while (total_len && *issued < dpolicy->max_requests && !err) {
1136                 struct bio *bio = NULL;
1137                 unsigned long flags;
1138                 bool last = true;
1139
1140                 if (len > max_discard_blocks) {
1141                         len = max_discard_blocks;
1142                         last = false;
1143                 }
1144
1145                 (*issued)++;
1146                 if (*issued == dpolicy->max_requests)
1147                         last = true;
1148
1149                 dc->len += len;
1150
1151                 if (time_to_inject(sbi, FAULT_DISCARD)) {
1152                         f2fs_show_injection_info(FAULT_DISCARD);
1153                         err = -EIO;
1154                         goto submit;
1155                 }
1156                 err = __blkdev_issue_discard(bdev,
1157                                         SECTOR_FROM_BLOCK(start),
1158                                         SECTOR_FROM_BLOCK(len),
1159                                         GFP_NOFS, 0, &bio);
1160 submit:
1161                 if (err) {
1162                         spin_lock_irqsave(&dc->lock, flags);
1163                         if (dc->state == D_PARTIAL)
1164                                 dc->state = D_SUBMIT;
1165                         spin_unlock_irqrestore(&dc->lock, flags);
1166
1167                         break;
1168                 }
1169
1170                 f2fs_bug_on(sbi, !bio);
1171
1172                 /*
1173                  * should keep before submission to avoid D_DONE
1174                  * right away
1175                  */
1176                 spin_lock_irqsave(&dc->lock, flags);
1177                 if (last)
1178                         dc->state = D_SUBMIT;
1179                 else
1180                         dc->state = D_PARTIAL;
1181                 dc->bio_ref++;
1182                 spin_unlock_irqrestore(&dc->lock, flags);
1183
1184                 atomic_inc(&dcc->queued_discard);
1185                 dc->queued++;
1186                 list_move_tail(&dc->list, wait_list);
1187
1188                 /* sanity check on discard range */
1189                 __check_sit_bitmap(sbi, lstart, lstart + len);
1190
1191                 bio->bi_private = dc;
1192                 bio->bi_end_io = f2fs_submit_discard_endio;
1193                 bio->bi_opf |= flag;
1194                 submit_bio(bio);
1195
1196                 atomic_inc(&dcc->issued_discard);
1197
1198                 f2fs_update_iostat(sbi, FS_DISCARD, 1);
1199
1200                 lstart += len;
1201                 start += len;
1202                 total_len -= len;
1203                 len = total_len;
1204         }
1205
1206         if (!err && len)
1207                 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1208         return err;
1209 }
1210
1211 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1212                                 struct block_device *bdev, block_t lstart,
1213                                 block_t start, block_t len,
1214                                 struct rb_node **insert_p,
1215                                 struct rb_node *insert_parent)
1216 {
1217         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1218         struct rb_node **p;
1219         struct rb_node *parent = NULL;
1220         struct discard_cmd *dc = NULL;
1221         bool leftmost = true;
1222
1223         if (insert_p && insert_parent) {
1224                 parent = insert_parent;
1225                 p = insert_p;
1226                 goto do_insert;
1227         }
1228
1229         p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1230                                                         lstart, &leftmost);
1231 do_insert:
1232         dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1233                                                                 p, leftmost);
1234         if (!dc)
1235                 return NULL;
1236
1237         return dc;
1238 }
1239
1240 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1241                                                 struct discard_cmd *dc)
1242 {
1243         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1244 }
1245
1246 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1247                                 struct discard_cmd *dc, block_t blkaddr)
1248 {
1249         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1250         struct discard_info di = dc->di;
1251         bool modified = false;
1252
1253         if (dc->state == D_DONE || dc->len == 1) {
1254                 __remove_discard_cmd(sbi, dc);
1255                 return;
1256         }
1257
1258         dcc->undiscard_blks -= di.len;
1259
1260         if (blkaddr > di.lstart) {
1261                 dc->len = blkaddr - dc->lstart;
1262                 dcc->undiscard_blks += dc->len;
1263                 __relocate_discard_cmd(dcc, dc);
1264                 modified = true;
1265         }
1266
1267         if (blkaddr < di.lstart + di.len - 1) {
1268                 if (modified) {
1269                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1270                                         di.start + blkaddr + 1 - di.lstart,
1271                                         di.lstart + di.len - 1 - blkaddr,
1272                                         NULL, NULL);
1273                 } else {
1274                         dc->lstart++;
1275                         dc->len--;
1276                         dc->start++;
1277                         dcc->undiscard_blks += dc->len;
1278                         __relocate_discard_cmd(dcc, dc);
1279                 }
1280         }
1281 }
1282
1283 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1284                                 struct block_device *bdev, block_t lstart,
1285                                 block_t start, block_t len)
1286 {
1287         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1288         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1289         struct discard_cmd *dc;
1290         struct discard_info di = {0};
1291         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1292         struct request_queue *q = bdev_get_queue(bdev);
1293         unsigned int max_discard_blocks =
1294                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1295         block_t end = lstart + len;
1296
1297         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1298                                         NULL, lstart,
1299                                         (struct rb_entry **)&prev_dc,
1300                                         (struct rb_entry **)&next_dc,
1301                                         &insert_p, &insert_parent, true, NULL);
1302         if (dc)
1303                 prev_dc = dc;
1304
1305         if (!prev_dc) {
1306                 di.lstart = lstart;
1307                 di.len = next_dc ? next_dc->lstart - lstart : len;
1308                 di.len = min(di.len, len);
1309                 di.start = start;
1310         }
1311
1312         while (1) {
1313                 struct rb_node *node;
1314                 bool merged = false;
1315                 struct discard_cmd *tdc = NULL;
1316
1317                 if (prev_dc) {
1318                         di.lstart = prev_dc->lstart + prev_dc->len;
1319                         if (di.lstart < lstart)
1320                                 di.lstart = lstart;
1321                         if (di.lstart >= end)
1322                                 break;
1323
1324                         if (!next_dc || next_dc->lstart > end)
1325                                 di.len = end - di.lstart;
1326                         else
1327                                 di.len = next_dc->lstart - di.lstart;
1328                         di.start = start + di.lstart - lstart;
1329                 }
1330
1331                 if (!di.len)
1332                         goto next;
1333
1334                 if (prev_dc && prev_dc->state == D_PREP &&
1335                         prev_dc->bdev == bdev &&
1336                         __is_discard_back_mergeable(&di, &prev_dc->di,
1337                                                         max_discard_blocks)) {
1338                         prev_dc->di.len += di.len;
1339                         dcc->undiscard_blks += di.len;
1340                         __relocate_discard_cmd(dcc, prev_dc);
1341                         di = prev_dc->di;
1342                         tdc = prev_dc;
1343                         merged = true;
1344                 }
1345
1346                 if (next_dc && next_dc->state == D_PREP &&
1347                         next_dc->bdev == bdev &&
1348                         __is_discard_front_mergeable(&di, &next_dc->di,
1349                                                         max_discard_blocks)) {
1350                         next_dc->di.lstart = di.lstart;
1351                         next_dc->di.len += di.len;
1352                         next_dc->di.start = di.start;
1353                         dcc->undiscard_blks += di.len;
1354                         __relocate_discard_cmd(dcc, next_dc);
1355                         if (tdc)
1356                                 __remove_discard_cmd(sbi, tdc);
1357                         merged = true;
1358                 }
1359
1360                 if (!merged) {
1361                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1362                                                         di.len, NULL, NULL);
1363                 }
1364  next:
1365                 prev_dc = next_dc;
1366                 if (!prev_dc)
1367                         break;
1368
1369                 node = rb_next(&prev_dc->rb_node);
1370                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1371         }
1372 }
1373
1374 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1375                 struct block_device *bdev, block_t blkstart, block_t blklen)
1376 {
1377         block_t lblkstart = blkstart;
1378
1379         if (!f2fs_bdev_support_discard(bdev))
1380                 return 0;
1381
1382         trace_f2fs_queue_discard(bdev, blkstart, blklen);
1383
1384         if (f2fs_is_multi_device(sbi)) {
1385                 int devi = f2fs_target_device_index(sbi, blkstart);
1386
1387                 blkstart -= FDEV(devi).start_blk;
1388         }
1389         mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1390         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1391         mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1392         return 0;
1393 }
1394
1395 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1396                                         struct discard_policy *dpolicy)
1397 {
1398         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1399         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1400         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1401         struct discard_cmd *dc;
1402         struct blk_plug plug;
1403         unsigned int pos = dcc->next_pos;
1404         unsigned int issued = 0;
1405         bool io_interrupted = false;
1406
1407         mutex_lock(&dcc->cmd_lock);
1408         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1409                                         NULL, pos,
1410                                         (struct rb_entry **)&prev_dc,
1411                                         (struct rb_entry **)&next_dc,
1412                                         &insert_p, &insert_parent, true, NULL);
1413         if (!dc)
1414                 dc = next_dc;
1415
1416         blk_start_plug(&plug);
1417
1418         while (dc) {
1419                 struct rb_node *node;
1420                 int err = 0;
1421
1422                 if (dc->state != D_PREP)
1423                         goto next;
1424
1425                 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1426                         io_interrupted = true;
1427                         break;
1428                 }
1429
1430                 dcc->next_pos = dc->lstart + dc->len;
1431                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1432
1433                 if (issued >= dpolicy->max_requests)
1434                         break;
1435 next:
1436                 node = rb_next(&dc->rb_node);
1437                 if (err)
1438                         __remove_discard_cmd(sbi, dc);
1439                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1440         }
1441
1442         blk_finish_plug(&plug);
1443
1444         if (!dc)
1445                 dcc->next_pos = 0;
1446
1447         mutex_unlock(&dcc->cmd_lock);
1448
1449         if (!issued && io_interrupted)
1450                 issued = -1;
1451
1452         return issued;
1453 }
1454
1455 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1456                                         struct discard_policy *dpolicy)
1457 {
1458         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1459         struct list_head *pend_list;
1460         struct discard_cmd *dc, *tmp;
1461         struct blk_plug plug;
1462         int i, issued = 0;
1463         bool io_interrupted = false;
1464
1465         if (dpolicy->timeout != 0)
1466                 f2fs_update_time(sbi, dpolicy->timeout);
1467
1468         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1469                 if (dpolicy->timeout != 0 &&
1470                                 f2fs_time_over(sbi, dpolicy->timeout))
1471                         break;
1472
1473                 if (i + 1 < dpolicy->granularity)
1474                         break;
1475
1476                 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1477                         return __issue_discard_cmd_orderly(sbi, dpolicy);
1478
1479                 pend_list = &dcc->pend_list[i];
1480
1481                 mutex_lock(&dcc->cmd_lock);
1482                 if (list_empty(pend_list))
1483                         goto next;
1484                 if (unlikely(dcc->rbtree_check))
1485                         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1486                                                                 &dcc->root));
1487                 blk_start_plug(&plug);
1488                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1489                         f2fs_bug_on(sbi, dc->state != D_PREP);
1490
1491                         if (dpolicy->timeout != 0 &&
1492                                 f2fs_time_over(sbi, dpolicy->timeout))
1493                                 break;
1494
1495                         if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1496                                                 !is_idle(sbi, DISCARD_TIME)) {
1497                                 io_interrupted = true;
1498                                 break;
1499                         }
1500
1501                         __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1502
1503                         if (issued >= dpolicy->max_requests)
1504                                 break;
1505                 }
1506                 blk_finish_plug(&plug);
1507 next:
1508                 mutex_unlock(&dcc->cmd_lock);
1509
1510                 if (issued >= dpolicy->max_requests || io_interrupted)
1511                         break;
1512         }
1513
1514         if (!issued && io_interrupted)
1515                 issued = -1;
1516
1517         return issued;
1518 }
1519
1520 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1521 {
1522         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1523         struct list_head *pend_list;
1524         struct discard_cmd *dc, *tmp;
1525         int i;
1526         bool dropped = false;
1527
1528         mutex_lock(&dcc->cmd_lock);
1529         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1530                 pend_list = &dcc->pend_list[i];
1531                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1532                         f2fs_bug_on(sbi, dc->state != D_PREP);
1533                         __remove_discard_cmd(sbi, dc);
1534                         dropped = true;
1535                 }
1536         }
1537         mutex_unlock(&dcc->cmd_lock);
1538
1539         return dropped;
1540 }
1541
1542 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1543 {
1544         __drop_discard_cmd(sbi);
1545 }
1546
1547 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1548                                                         struct discard_cmd *dc)
1549 {
1550         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1551         unsigned int len = 0;
1552
1553         wait_for_completion_io(&dc->wait);
1554         mutex_lock(&dcc->cmd_lock);
1555         f2fs_bug_on(sbi, dc->state != D_DONE);
1556         dc->ref--;
1557         if (!dc->ref) {
1558                 if (!dc->error)
1559                         len = dc->len;
1560                 __remove_discard_cmd(sbi, dc);
1561         }
1562         mutex_unlock(&dcc->cmd_lock);
1563
1564         return len;
1565 }
1566
1567 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1568                                                 struct discard_policy *dpolicy,
1569                                                 block_t start, block_t end)
1570 {
1571         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1572         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1573                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1574         struct discard_cmd *dc, *tmp;
1575         bool need_wait;
1576         unsigned int trimmed = 0;
1577
1578 next:
1579         need_wait = false;
1580
1581         mutex_lock(&dcc->cmd_lock);
1582         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1583                 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1584                         continue;
1585                 if (dc->len < dpolicy->granularity)
1586                         continue;
1587                 if (dc->state == D_DONE && !dc->ref) {
1588                         wait_for_completion_io(&dc->wait);
1589                         if (!dc->error)
1590                                 trimmed += dc->len;
1591                         __remove_discard_cmd(sbi, dc);
1592                 } else {
1593                         dc->ref++;
1594                         need_wait = true;
1595                         break;
1596                 }
1597         }
1598         mutex_unlock(&dcc->cmd_lock);
1599
1600         if (need_wait) {
1601                 trimmed += __wait_one_discard_bio(sbi, dc);
1602                 goto next;
1603         }
1604
1605         return trimmed;
1606 }
1607
1608 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1609                                                 struct discard_policy *dpolicy)
1610 {
1611         struct discard_policy dp;
1612         unsigned int discard_blks;
1613
1614         if (dpolicy)
1615                 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1616
1617         /* wait all */
1618         __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1619         discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1620         __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1621         discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1622
1623         return discard_blks;
1624 }
1625
1626 /* This should be covered by global mutex, &sit_i->sentry_lock */
1627 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1628 {
1629         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1630         struct discard_cmd *dc;
1631         bool need_wait = false;
1632
1633         mutex_lock(&dcc->cmd_lock);
1634         dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1635                                                         NULL, blkaddr);
1636         if (dc) {
1637                 if (dc->state == D_PREP) {
1638                         __punch_discard_cmd(sbi, dc, blkaddr);
1639                 } else {
1640                         dc->ref++;
1641                         need_wait = true;
1642                 }
1643         }
1644         mutex_unlock(&dcc->cmd_lock);
1645
1646         if (need_wait)
1647                 __wait_one_discard_bio(sbi, dc);
1648 }
1649
1650 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1651 {
1652         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1653
1654         if (dcc && dcc->f2fs_issue_discard) {
1655                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1656
1657                 dcc->f2fs_issue_discard = NULL;
1658                 kthread_stop(discard_thread);
1659         }
1660 }
1661
1662 /* This comes from f2fs_put_super */
1663 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1664 {
1665         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1666         struct discard_policy dpolicy;
1667         bool dropped;
1668
1669         __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1670                                         dcc->discard_granularity);
1671         dpolicy.timeout = UMOUNT_DISCARD_TIMEOUT;
1672         __issue_discard_cmd(sbi, &dpolicy);
1673         dropped = __drop_discard_cmd(sbi);
1674
1675         /* just to make sure there is no pending discard commands */
1676         __wait_all_discard_cmd(sbi, NULL);
1677
1678         f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1679         return dropped;
1680 }
1681
1682 static int issue_discard_thread(void *data)
1683 {
1684         struct f2fs_sb_info *sbi = data;
1685         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1686         wait_queue_head_t *q = &dcc->discard_wait_queue;
1687         struct discard_policy dpolicy;
1688         unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1689         int issued;
1690
1691         set_freezable();
1692
1693         do {
1694                 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1695                                         dcc->discard_granularity);
1696
1697                 wait_event_interruptible_timeout(*q,
1698                                 kthread_should_stop() || freezing(current) ||
1699                                 dcc->discard_wake,
1700                                 msecs_to_jiffies(wait_ms));
1701
1702                 if (dcc->discard_wake)
1703                         dcc->discard_wake = 0;
1704
1705                 /* clean up pending candidates before going to sleep */
1706                 if (atomic_read(&dcc->queued_discard))
1707                         __wait_all_discard_cmd(sbi, NULL);
1708
1709                 if (try_to_freeze())
1710                         continue;
1711                 if (f2fs_readonly(sbi->sb))
1712                         continue;
1713                 if (kthread_should_stop())
1714                         return 0;
1715                 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1716                         wait_ms = dpolicy.max_interval;
1717                         continue;
1718                 }
1719
1720                 if (sbi->gc_mode == GC_URGENT)
1721                         __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1722
1723                 sb_start_intwrite(sbi->sb);
1724
1725                 issued = __issue_discard_cmd(sbi, &dpolicy);
1726                 if (issued > 0) {
1727                         __wait_all_discard_cmd(sbi, &dpolicy);
1728                         wait_ms = dpolicy.min_interval;
1729                 } else if (issued == -1){
1730                         wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1731                         if (!wait_ms)
1732                                 wait_ms = dpolicy.mid_interval;
1733                 } else {
1734                         wait_ms = dpolicy.max_interval;
1735                 }
1736
1737                 sb_end_intwrite(sbi->sb);
1738
1739         } while (!kthread_should_stop());
1740         return 0;
1741 }
1742
1743 #ifdef CONFIG_BLK_DEV_ZONED
1744 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1745                 struct block_device *bdev, block_t blkstart, block_t blklen)
1746 {
1747         sector_t sector, nr_sects;
1748         block_t lblkstart = blkstart;
1749         int devi = 0;
1750
1751         if (f2fs_is_multi_device(sbi)) {
1752                 devi = f2fs_target_device_index(sbi, blkstart);
1753                 if (blkstart < FDEV(devi).start_blk ||
1754                     blkstart > FDEV(devi).end_blk) {
1755                         f2fs_err(sbi, "Invalid block %x", blkstart);
1756                         return -EIO;
1757                 }
1758                 blkstart -= FDEV(devi).start_blk;
1759         }
1760
1761         /* For sequential zones, reset the zone write pointer */
1762         if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1763                 sector = SECTOR_FROM_BLOCK(blkstart);
1764                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1765
1766                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1767                                 nr_sects != bdev_zone_sectors(bdev)) {
1768                         f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1769                                  devi, sbi->s_ndevs ? FDEV(devi).path : "",
1770                                  blkstart, blklen);
1771                         return -EIO;
1772                 }
1773                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1774                 return blkdev_reset_zones(bdev, sector, nr_sects, GFP_NOFS);
1775         }
1776
1777         /* For conventional zones, use regular discard if supported */
1778         return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1779 }
1780 #endif
1781
1782 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1783                 struct block_device *bdev, block_t blkstart, block_t blklen)
1784 {
1785 #ifdef CONFIG_BLK_DEV_ZONED
1786         if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1787                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1788 #endif
1789         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1790 }
1791
1792 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1793                                 block_t blkstart, block_t blklen)
1794 {
1795         sector_t start = blkstart, len = 0;
1796         struct block_device *bdev;
1797         struct seg_entry *se;
1798         unsigned int offset;
1799         block_t i;
1800         int err = 0;
1801
1802         bdev = f2fs_target_device(sbi, blkstart, NULL);
1803
1804         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1805                 if (i != start) {
1806                         struct block_device *bdev2 =
1807                                 f2fs_target_device(sbi, i, NULL);
1808
1809                         if (bdev2 != bdev) {
1810                                 err = __issue_discard_async(sbi, bdev,
1811                                                 start, len);
1812                                 if (err)
1813                                         return err;
1814                                 bdev = bdev2;
1815                                 start = i;
1816                                 len = 0;
1817                         }
1818                 }
1819
1820                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1821                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1822
1823                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1824                         sbi->discard_blks--;
1825         }
1826
1827         if (len)
1828                 err = __issue_discard_async(sbi, bdev, start, len);
1829         return err;
1830 }
1831
1832 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1833                                                         bool check_only)
1834 {
1835         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1836         int max_blocks = sbi->blocks_per_seg;
1837         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1838         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1839         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1840         unsigned long *discard_map = (unsigned long *)se->discard_map;
1841         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1842         unsigned int start = 0, end = -1;
1843         bool force = (cpc->reason & CP_DISCARD);
1844         struct discard_entry *de = NULL;
1845         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1846         int i;
1847
1848         if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
1849                 return false;
1850
1851         if (!force) {
1852                 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1853                         SM_I(sbi)->dcc_info->nr_discards >=
1854                                 SM_I(sbi)->dcc_info->max_discards)
1855                         return false;
1856         }
1857
1858         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1859         for (i = 0; i < entries; i++)
1860                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1861                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1862
1863         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1864                                 SM_I(sbi)->dcc_info->max_discards) {
1865                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1866                 if (start >= max_blocks)
1867                         break;
1868
1869                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1870                 if (force && start && end != max_blocks
1871                                         && (end - start) < cpc->trim_minlen)
1872                         continue;
1873
1874                 if (check_only)
1875                         return true;
1876
1877                 if (!de) {
1878                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1879                                                                 GFP_F2FS_ZERO);
1880                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1881                         list_add_tail(&de->list, head);
1882                 }
1883
1884                 for (i = start; i < end; i++)
1885                         __set_bit_le(i, (void *)de->discard_map);
1886
1887                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1888         }
1889         return false;
1890 }
1891
1892 static void release_discard_addr(struct discard_entry *entry)
1893 {
1894         list_del(&entry->list);
1895         kmem_cache_free(discard_entry_slab, entry);
1896 }
1897
1898 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1899 {
1900         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1901         struct discard_entry *entry, *this;
1902
1903         /* drop caches */
1904         list_for_each_entry_safe(entry, this, head, list)
1905                 release_discard_addr(entry);
1906 }
1907
1908 /*
1909  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1910  */
1911 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1912 {
1913         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1914         unsigned int segno;
1915
1916         mutex_lock(&dirty_i->seglist_lock);
1917         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1918                 __set_test_and_free(sbi, segno);
1919         mutex_unlock(&dirty_i->seglist_lock);
1920 }
1921
1922 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1923                                                 struct cp_control *cpc)
1924 {
1925         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1926         struct list_head *head = &dcc->entry_list;
1927         struct discard_entry *entry, *this;
1928         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1929         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1930         unsigned int start = 0, end = -1;
1931         unsigned int secno, start_segno;
1932         bool force = (cpc->reason & CP_DISCARD);
1933         bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
1934
1935         mutex_lock(&dirty_i->seglist_lock);
1936
1937         while (1) {
1938                 int i;
1939
1940                 if (need_align && end != -1)
1941                         end--;
1942                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1943                 if (start >= MAIN_SEGS(sbi))
1944                         break;
1945                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1946                                                                 start + 1);
1947
1948                 if (need_align) {
1949                         start = rounddown(start, sbi->segs_per_sec);
1950                         end = roundup(end, sbi->segs_per_sec);
1951                 }
1952
1953                 for (i = start; i < end; i++) {
1954                         if (test_and_clear_bit(i, prefree_map))
1955                                 dirty_i->nr_dirty[PRE]--;
1956                 }
1957
1958                 if (!f2fs_realtime_discard_enable(sbi))
1959                         continue;
1960
1961                 if (force && start >= cpc->trim_start &&
1962                                         (end - 1) <= cpc->trim_end)
1963                                 continue;
1964
1965                 if (!test_opt(sbi, LFS) || !__is_large_section(sbi)) {
1966                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1967                                 (end - start) << sbi->log_blocks_per_seg);
1968                         continue;
1969                 }
1970 next:
1971                 secno = GET_SEC_FROM_SEG(sbi, start);
1972                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1973                 if (!IS_CURSEC(sbi, secno) &&
1974                         !get_valid_blocks(sbi, start, true))
1975                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1976                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1977
1978                 start = start_segno + sbi->segs_per_sec;
1979                 if (start < end)
1980                         goto next;
1981                 else
1982                         end = start - 1;
1983         }
1984         mutex_unlock(&dirty_i->seglist_lock);
1985
1986         /* send small discards */
1987         list_for_each_entry_safe(entry, this, head, list) {
1988                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1989                 bool is_valid = test_bit_le(0, entry->discard_map);
1990
1991 find_next:
1992                 if (is_valid) {
1993                         next_pos = find_next_zero_bit_le(entry->discard_map,
1994                                         sbi->blocks_per_seg, cur_pos);
1995                         len = next_pos - cur_pos;
1996
1997                         if (f2fs_sb_has_blkzoned(sbi) ||
1998                             (force && len < cpc->trim_minlen))
1999                                 goto skip;
2000
2001                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2002                                                                         len);
2003                         total_len += len;
2004                 } else {
2005                         next_pos = find_next_bit_le(entry->discard_map,
2006                                         sbi->blocks_per_seg, cur_pos);
2007                 }
2008 skip:
2009                 cur_pos = next_pos;
2010                 is_valid = !is_valid;
2011
2012                 if (cur_pos < sbi->blocks_per_seg)
2013                         goto find_next;
2014
2015                 release_discard_addr(entry);
2016                 dcc->nr_discards -= total_len;
2017         }
2018
2019         wake_up_discard_thread(sbi, false);
2020 }
2021
2022 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2023 {
2024         dev_t dev = sbi->sb->s_bdev->bd_dev;
2025         struct discard_cmd_control *dcc;
2026         int err = 0, i;
2027
2028         if (SM_I(sbi)->dcc_info) {
2029                 dcc = SM_I(sbi)->dcc_info;
2030                 goto init_thread;
2031         }
2032
2033         dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2034         if (!dcc)
2035                 return -ENOMEM;
2036
2037         dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2038         INIT_LIST_HEAD(&dcc->entry_list);
2039         for (i = 0; i < MAX_PLIST_NUM; i++)
2040                 INIT_LIST_HEAD(&dcc->pend_list[i]);
2041         INIT_LIST_HEAD(&dcc->wait_list);
2042         INIT_LIST_HEAD(&dcc->fstrim_list);
2043         mutex_init(&dcc->cmd_lock);
2044         atomic_set(&dcc->issued_discard, 0);
2045         atomic_set(&dcc->queued_discard, 0);
2046         atomic_set(&dcc->discard_cmd_cnt, 0);
2047         dcc->nr_discards = 0;
2048         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2049         dcc->undiscard_blks = 0;
2050         dcc->next_pos = 0;
2051         dcc->root = RB_ROOT_CACHED;
2052         dcc->rbtree_check = false;
2053
2054         init_waitqueue_head(&dcc->discard_wait_queue);
2055         SM_I(sbi)->dcc_info = dcc;
2056 init_thread:
2057         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2058                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2059         if (IS_ERR(dcc->f2fs_issue_discard)) {
2060                 err = PTR_ERR(dcc->f2fs_issue_discard);
2061                 kvfree(dcc);
2062                 SM_I(sbi)->dcc_info = NULL;
2063                 return err;
2064         }
2065
2066         return err;
2067 }
2068
2069 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2070 {
2071         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2072
2073         if (!dcc)
2074                 return;
2075
2076         f2fs_stop_discard_thread(sbi);
2077
2078         /*
2079          * Recovery can cache discard commands, so in error path of
2080          * fill_super(), it needs to give a chance to handle them.
2081          */
2082         if (unlikely(atomic_read(&dcc->discard_cmd_cnt)))
2083                 f2fs_issue_discard_timeout(sbi);
2084
2085         kvfree(dcc);
2086         SM_I(sbi)->dcc_info = NULL;
2087 }
2088
2089 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2090 {
2091         struct sit_info *sit_i = SIT_I(sbi);
2092
2093         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2094                 sit_i->dirty_sentries++;
2095                 return false;
2096         }
2097
2098         return true;
2099 }
2100
2101 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2102                                         unsigned int segno, int modified)
2103 {
2104         struct seg_entry *se = get_seg_entry(sbi, segno);
2105         se->type = type;
2106         if (modified)
2107                 __mark_sit_entry_dirty(sbi, segno);
2108 }
2109
2110 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2111 {
2112         struct seg_entry *se;
2113         unsigned int segno, offset;
2114         long int new_vblocks;
2115         bool exist;
2116 #ifdef CONFIG_F2FS_CHECK_FS
2117         bool mir_exist;
2118 #endif
2119
2120         segno = GET_SEGNO(sbi, blkaddr);
2121
2122         se = get_seg_entry(sbi, segno);
2123         new_vblocks = se->valid_blocks + del;
2124         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2125
2126         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
2127                                 (new_vblocks > sbi->blocks_per_seg)));
2128
2129         se->valid_blocks = new_vblocks;
2130         se->mtime = get_mtime(sbi, false);
2131         if (se->mtime > SIT_I(sbi)->max_mtime)
2132                 SIT_I(sbi)->max_mtime = se->mtime;
2133
2134         /* Update valid block bitmap */
2135         if (del > 0) {
2136                 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2137 #ifdef CONFIG_F2FS_CHECK_FS
2138                 mir_exist = f2fs_test_and_set_bit(offset,
2139                                                 se->cur_valid_map_mir);
2140                 if (unlikely(exist != mir_exist)) {
2141                         f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2142                                  blkaddr, exist);
2143                         f2fs_bug_on(sbi, 1);
2144                 }
2145 #endif
2146                 if (unlikely(exist)) {
2147                         f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2148                                  blkaddr);
2149                         f2fs_bug_on(sbi, 1);
2150                         se->valid_blocks--;
2151                         del = 0;
2152                 }
2153
2154                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
2155                         sbi->discard_blks--;
2156
2157                 /*
2158                  * SSR should never reuse block which is checkpointed
2159                  * or newly invalidated.
2160                  */
2161                 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2162                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2163                                 se->ckpt_valid_blocks++;
2164                 }
2165         } else {
2166                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2167 #ifdef CONFIG_F2FS_CHECK_FS
2168                 mir_exist = f2fs_test_and_clear_bit(offset,
2169                                                 se->cur_valid_map_mir);
2170                 if (unlikely(exist != mir_exist)) {
2171                         f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2172                                  blkaddr, exist);
2173                         f2fs_bug_on(sbi, 1);
2174                 }
2175 #endif
2176                 if (unlikely(!exist)) {
2177                         f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2178                                  blkaddr);
2179                         f2fs_bug_on(sbi, 1);
2180                         se->valid_blocks++;
2181                         del = 0;
2182                 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2183                         /*
2184                          * If checkpoints are off, we must not reuse data that
2185                          * was used in the previous checkpoint. If it was used
2186                          * before, we must track that to know how much space we
2187                          * really have.
2188                          */
2189                         if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2190                                 spin_lock(&sbi->stat_lock);
2191                                 sbi->unusable_block_count++;
2192                                 spin_unlock(&sbi->stat_lock);
2193                         }
2194                 }
2195
2196                 if (f2fs_test_and_clear_bit(offset, se->discard_map))
2197                         sbi->discard_blks++;
2198         }
2199         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2200                 se->ckpt_valid_blocks += del;
2201
2202         __mark_sit_entry_dirty(sbi, segno);
2203
2204         /* update total number of valid blocks to be written in ckpt area */
2205         SIT_I(sbi)->written_valid_blocks += del;
2206
2207         if (__is_large_section(sbi))
2208                 get_sec_entry(sbi, segno)->valid_blocks += del;
2209 }
2210
2211 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2212 {
2213         unsigned int segno = GET_SEGNO(sbi, addr);
2214         struct sit_info *sit_i = SIT_I(sbi);
2215
2216         f2fs_bug_on(sbi, addr == NULL_ADDR);
2217         if (addr == NEW_ADDR)
2218                 return;
2219
2220         invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2221
2222         /* add it into sit main buffer */
2223         down_write(&sit_i->sentry_lock);
2224
2225         update_sit_entry(sbi, addr, -1);
2226
2227         /* add it into dirty seglist */
2228         locate_dirty_segment(sbi, segno);
2229
2230         up_write(&sit_i->sentry_lock);
2231 }
2232
2233 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2234 {
2235         struct sit_info *sit_i = SIT_I(sbi);
2236         unsigned int segno, offset;
2237         struct seg_entry *se;
2238         bool is_cp = false;
2239
2240         if (!__is_valid_data_blkaddr(blkaddr))
2241                 return true;
2242
2243         down_read(&sit_i->sentry_lock);
2244
2245         segno = GET_SEGNO(sbi, blkaddr);
2246         se = get_seg_entry(sbi, segno);
2247         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2248
2249         if (f2fs_test_bit(offset, se->ckpt_valid_map))
2250                 is_cp = true;
2251
2252         up_read(&sit_i->sentry_lock);
2253
2254         return is_cp;
2255 }
2256
2257 /*
2258  * This function should be resided under the curseg_mutex lock
2259  */
2260 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2261                                         struct f2fs_summary *sum)
2262 {
2263         struct curseg_info *curseg = CURSEG_I(sbi, type);
2264         void *addr = curseg->sum_blk;
2265         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2266         memcpy(addr, sum, sizeof(struct f2fs_summary));
2267 }
2268
2269 /*
2270  * Calculate the number of current summary pages for writing
2271  */
2272 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2273 {
2274         int valid_sum_count = 0;
2275         int i, sum_in_page;
2276
2277         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2278                 if (sbi->ckpt->alloc_type[i] == SSR)
2279                         valid_sum_count += sbi->blocks_per_seg;
2280                 else {
2281                         if (for_ra)
2282                                 valid_sum_count += le16_to_cpu(
2283                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2284                         else
2285                                 valid_sum_count += curseg_blkoff(sbi, i);
2286                 }
2287         }
2288
2289         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2290                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2291         if (valid_sum_count <= sum_in_page)
2292                 return 1;
2293         else if ((valid_sum_count - sum_in_page) <=
2294                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2295                 return 2;
2296         return 3;
2297 }
2298
2299 /*
2300  * Caller should put this summary page
2301  */
2302 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2303 {
2304         return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
2305 }
2306
2307 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2308                                         void *src, block_t blk_addr)
2309 {
2310         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2311
2312         memcpy(page_address(page), src, PAGE_SIZE);
2313         set_page_dirty(page);
2314         f2fs_put_page(page, 1);
2315 }
2316
2317 static void write_sum_page(struct f2fs_sb_info *sbi,
2318                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
2319 {
2320         f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2321 }
2322
2323 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2324                                                 int type, block_t blk_addr)
2325 {
2326         struct curseg_info *curseg = CURSEG_I(sbi, type);
2327         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2328         struct f2fs_summary_block *src = curseg->sum_blk;
2329         struct f2fs_summary_block *dst;
2330
2331         dst = (struct f2fs_summary_block *)page_address(page);
2332         memset(dst, 0, PAGE_SIZE);
2333
2334         mutex_lock(&curseg->curseg_mutex);
2335
2336         down_read(&curseg->journal_rwsem);
2337         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2338         up_read(&curseg->journal_rwsem);
2339
2340         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2341         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2342
2343         mutex_unlock(&curseg->curseg_mutex);
2344
2345         set_page_dirty(page);
2346         f2fs_put_page(page, 1);
2347 }
2348
2349 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2350 {
2351         struct curseg_info *curseg = CURSEG_I(sbi, type);
2352         unsigned int segno = curseg->segno + 1;
2353         struct free_segmap_info *free_i = FREE_I(sbi);
2354
2355         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2356                 return !test_bit(segno, free_i->free_segmap);
2357         return 0;
2358 }
2359
2360 /*
2361  * Find a new segment from the free segments bitmap to right order
2362  * This function should be returned with success, otherwise BUG
2363  */
2364 static void get_new_segment(struct f2fs_sb_info *sbi,
2365                         unsigned int *newseg, bool new_sec, int dir)
2366 {
2367         struct free_segmap_info *free_i = FREE_I(sbi);
2368         unsigned int segno, secno, zoneno;
2369         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2370         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2371         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2372         unsigned int left_start = hint;
2373         bool init = true;
2374         int go_left = 0;
2375         int i;
2376
2377         spin_lock(&free_i->segmap_lock);
2378
2379         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2380                 segno = find_next_zero_bit(free_i->free_segmap,
2381                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2382                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2383                         goto got_it;
2384         }
2385 find_other_zone:
2386         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2387         if (secno >= MAIN_SECS(sbi)) {
2388                 if (dir == ALLOC_RIGHT) {
2389                         secno = find_next_zero_bit(free_i->free_secmap,
2390                                                         MAIN_SECS(sbi), 0);
2391                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2392                 } else {
2393                         go_left = 1;
2394                         left_start = hint - 1;
2395                 }
2396         }
2397         if (go_left == 0)
2398                 goto skip_left;
2399
2400         while (test_bit(left_start, free_i->free_secmap)) {
2401                 if (left_start > 0) {
2402                         left_start--;
2403                         continue;
2404                 }
2405                 left_start = find_next_zero_bit(free_i->free_secmap,
2406                                                         MAIN_SECS(sbi), 0);
2407                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2408                 break;
2409         }
2410         secno = left_start;
2411 skip_left:
2412         segno = GET_SEG_FROM_SEC(sbi, secno);
2413         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2414
2415         /* give up on finding another zone */
2416         if (!init)
2417                 goto got_it;
2418         if (sbi->secs_per_zone == 1)
2419                 goto got_it;
2420         if (zoneno == old_zoneno)
2421                 goto got_it;
2422         if (dir == ALLOC_LEFT) {
2423                 if (!go_left && zoneno + 1 >= total_zones)
2424                         goto got_it;
2425                 if (go_left && zoneno == 0)
2426                         goto got_it;
2427         }
2428         for (i = 0; i < NR_CURSEG_TYPE; i++)
2429                 if (CURSEG_I(sbi, i)->zone == zoneno)
2430                         break;
2431
2432         if (i < NR_CURSEG_TYPE) {
2433                 /* zone is in user, try another */
2434                 if (go_left)
2435                         hint = zoneno * sbi->secs_per_zone - 1;
2436                 else if (zoneno + 1 >= total_zones)
2437                         hint = 0;
2438                 else
2439                         hint = (zoneno + 1) * sbi->secs_per_zone;
2440                 init = false;
2441                 goto find_other_zone;
2442         }
2443 got_it:
2444         /* set it as dirty segment in free segmap */
2445         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2446         __set_inuse(sbi, segno);
2447         *newseg = segno;
2448         spin_unlock(&free_i->segmap_lock);
2449 }
2450
2451 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2452 {
2453         struct curseg_info *curseg = CURSEG_I(sbi, type);
2454         struct summary_footer *sum_footer;
2455
2456         curseg->segno = curseg->next_segno;
2457         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2458         curseg->next_blkoff = 0;
2459         curseg->next_segno = NULL_SEGNO;
2460
2461         sum_footer = &(curseg->sum_blk->footer);
2462         memset(sum_footer, 0, sizeof(struct summary_footer));
2463         if (IS_DATASEG(type))
2464                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2465         if (IS_NODESEG(type))
2466                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2467         __set_sit_entry_type(sbi, type, curseg->segno, modified);
2468 }
2469
2470 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2471 {
2472         /* if segs_per_sec is large than 1, we need to keep original policy. */
2473         if (__is_large_section(sbi))
2474                 return CURSEG_I(sbi, type)->segno;
2475
2476         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2477                 return 0;
2478
2479         if (test_opt(sbi, NOHEAP) &&
2480                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2481                 return 0;
2482
2483         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2484                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2485
2486         /* find segments from 0 to reuse freed segments */
2487         if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2488                 return 0;
2489
2490         return CURSEG_I(sbi, type)->segno;
2491 }
2492
2493 /*
2494  * Allocate a current working segment.
2495  * This function always allocates a free segment in LFS manner.
2496  */
2497 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2498 {
2499         struct curseg_info *curseg = CURSEG_I(sbi, type);
2500         unsigned int segno = curseg->segno;
2501         int dir = ALLOC_LEFT;
2502
2503         write_sum_page(sbi, curseg->sum_blk,
2504                                 GET_SUM_BLOCK(sbi, segno));
2505         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2506                 dir = ALLOC_RIGHT;
2507
2508         if (test_opt(sbi, NOHEAP))
2509                 dir = ALLOC_RIGHT;
2510
2511         segno = __get_next_segno(sbi, type);
2512         get_new_segment(sbi, &segno, new_sec, dir);
2513         curseg->next_segno = segno;
2514         reset_curseg(sbi, type, 1);
2515         curseg->alloc_type = LFS;
2516 }
2517
2518 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2519                         struct curseg_info *seg, block_t start)
2520 {
2521         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2522         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2523         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2524         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2525         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2526         int i, pos;
2527
2528         for (i = 0; i < entries; i++)
2529                 target_map[i] = ckpt_map[i] | cur_map[i];
2530
2531         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2532
2533         seg->next_blkoff = pos;
2534 }
2535
2536 /*
2537  * If a segment is written by LFS manner, next block offset is just obtained
2538  * by increasing the current block offset. However, if a segment is written by
2539  * SSR manner, next block offset obtained by calling __next_free_blkoff
2540  */
2541 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2542                                 struct curseg_info *seg)
2543 {
2544         if (seg->alloc_type == SSR)
2545                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2546         else
2547                 seg->next_blkoff++;
2548 }
2549
2550 /*
2551  * This function always allocates a used segment(from dirty seglist) by SSR
2552  * manner, so it should recover the existing segment information of valid blocks
2553  */
2554 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2555 {
2556         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2557         struct curseg_info *curseg = CURSEG_I(sbi, type);
2558         unsigned int new_segno = curseg->next_segno;
2559         struct f2fs_summary_block *sum_node;
2560         struct page *sum_page;
2561
2562         write_sum_page(sbi, curseg->sum_blk,
2563                                 GET_SUM_BLOCK(sbi, curseg->segno));
2564         __set_test_and_inuse(sbi, new_segno);
2565
2566         mutex_lock(&dirty_i->seglist_lock);
2567         __remove_dirty_segment(sbi, new_segno, PRE);
2568         __remove_dirty_segment(sbi, new_segno, DIRTY);
2569         mutex_unlock(&dirty_i->seglist_lock);
2570
2571         reset_curseg(sbi, type, 1);
2572         curseg->alloc_type = SSR;
2573         __next_free_blkoff(sbi, curseg, 0);
2574
2575         sum_page = f2fs_get_sum_page(sbi, new_segno);
2576         f2fs_bug_on(sbi, IS_ERR(sum_page));
2577         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2578         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2579         f2fs_put_page(sum_page, 1);
2580 }
2581
2582 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2583 {
2584         struct curseg_info *curseg = CURSEG_I(sbi, type);
2585         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2586         unsigned segno = NULL_SEGNO;
2587         int i, cnt;
2588         bool reversed = false;
2589
2590         /* f2fs_need_SSR() already forces to do this */
2591         if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2592                 curseg->next_segno = segno;
2593                 return 1;
2594         }
2595
2596         /* For node segments, let's do SSR more intensively */
2597         if (IS_NODESEG(type)) {
2598                 if (type >= CURSEG_WARM_NODE) {
2599                         reversed = true;
2600                         i = CURSEG_COLD_NODE;
2601                 } else {
2602                         i = CURSEG_HOT_NODE;
2603                 }
2604                 cnt = NR_CURSEG_NODE_TYPE;
2605         } else {
2606                 if (type >= CURSEG_WARM_DATA) {
2607                         reversed = true;
2608                         i = CURSEG_COLD_DATA;
2609                 } else {
2610                         i = CURSEG_HOT_DATA;
2611                 }
2612                 cnt = NR_CURSEG_DATA_TYPE;
2613         }
2614
2615         for (; cnt-- > 0; reversed ? i-- : i++) {
2616                 if (i == type)
2617                         continue;
2618                 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2619                         curseg->next_segno = segno;
2620                         return 1;
2621                 }
2622         }
2623
2624         /* find valid_blocks=0 in dirty list */
2625         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2626                 segno = get_free_segment(sbi);
2627                 if (segno != NULL_SEGNO) {
2628                         curseg->next_segno = segno;
2629                         return 1;
2630                 }
2631         }
2632         return 0;
2633 }
2634
2635 /*
2636  * flush out current segment and replace it with new segment
2637  * This function should be returned with success, otherwise BUG
2638  */
2639 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2640                                                 int type, bool force)
2641 {
2642         struct curseg_info *curseg = CURSEG_I(sbi, type);
2643
2644         if (force)
2645                 new_curseg(sbi, type, true);
2646         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2647                                         type == CURSEG_WARM_NODE)
2648                 new_curseg(sbi, type, false);
2649         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2650                         likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2651                 new_curseg(sbi, type, false);
2652         else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2653                 change_curseg(sbi, type);
2654         else
2655                 new_curseg(sbi, type, false);
2656
2657         stat_inc_seg_type(sbi, curseg);
2658 }
2659
2660 void allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2661                                         unsigned int start, unsigned int end)
2662 {
2663         struct curseg_info *curseg = CURSEG_I(sbi, type);
2664         unsigned int segno;
2665
2666         down_read(&SM_I(sbi)->curseg_lock);
2667         mutex_lock(&curseg->curseg_mutex);
2668         down_write(&SIT_I(sbi)->sentry_lock);
2669
2670         segno = CURSEG_I(sbi, type)->segno;
2671         if (segno < start || segno > end)
2672                 goto unlock;
2673
2674         if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2675                 change_curseg(sbi, type);
2676         else
2677                 new_curseg(sbi, type, true);
2678
2679         stat_inc_seg_type(sbi, curseg);
2680
2681         locate_dirty_segment(sbi, segno);
2682 unlock:
2683         up_write(&SIT_I(sbi)->sentry_lock);
2684
2685         if (segno != curseg->segno)
2686                 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2687                             type, segno, curseg->segno);
2688
2689         mutex_unlock(&curseg->curseg_mutex);
2690         up_read(&SM_I(sbi)->curseg_lock);
2691 }
2692
2693 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2694 {
2695         struct curseg_info *curseg;
2696         unsigned int old_segno;
2697         int i;
2698
2699         down_write(&SIT_I(sbi)->sentry_lock);
2700
2701         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2702                 curseg = CURSEG_I(sbi, i);
2703                 old_segno = curseg->segno;
2704                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2705                 locate_dirty_segment(sbi, old_segno);
2706         }
2707
2708         up_write(&SIT_I(sbi)->sentry_lock);
2709 }
2710
2711 static const struct segment_allocation default_salloc_ops = {
2712         .allocate_segment = allocate_segment_by_default,
2713 };
2714
2715 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2716                                                 struct cp_control *cpc)
2717 {
2718         __u64 trim_start = cpc->trim_start;
2719         bool has_candidate = false;
2720
2721         down_write(&SIT_I(sbi)->sentry_lock);
2722         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2723                 if (add_discard_addrs(sbi, cpc, true)) {
2724                         has_candidate = true;
2725                         break;
2726                 }
2727         }
2728         up_write(&SIT_I(sbi)->sentry_lock);
2729
2730         cpc->trim_start = trim_start;
2731         return has_candidate;
2732 }
2733
2734 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2735                                         struct discard_policy *dpolicy,
2736                                         unsigned int start, unsigned int end)
2737 {
2738         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2739         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2740         struct rb_node **insert_p = NULL, *insert_parent = NULL;
2741         struct discard_cmd *dc;
2742         struct blk_plug plug;
2743         int issued;
2744         unsigned int trimmed = 0;
2745
2746 next:
2747         issued = 0;
2748
2749         mutex_lock(&dcc->cmd_lock);
2750         if (unlikely(dcc->rbtree_check))
2751                 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2752                                                                 &dcc->root));
2753
2754         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2755                                         NULL, start,
2756                                         (struct rb_entry **)&prev_dc,
2757                                         (struct rb_entry **)&next_dc,
2758                                         &insert_p, &insert_parent, true, NULL);
2759         if (!dc)
2760                 dc = next_dc;
2761
2762         blk_start_plug(&plug);
2763
2764         while (dc && dc->lstart <= end) {
2765                 struct rb_node *node;
2766                 int err = 0;
2767
2768                 if (dc->len < dpolicy->granularity)
2769                         goto skip;
2770
2771                 if (dc->state != D_PREP) {
2772                         list_move_tail(&dc->list, &dcc->fstrim_list);
2773                         goto skip;
2774                 }
2775
2776                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2777
2778                 if (issued >= dpolicy->max_requests) {
2779                         start = dc->lstart + dc->len;
2780
2781                         if (err)
2782                                 __remove_discard_cmd(sbi, dc);
2783
2784                         blk_finish_plug(&plug);
2785                         mutex_unlock(&dcc->cmd_lock);
2786                         trimmed += __wait_all_discard_cmd(sbi, NULL);
2787                         congestion_wait(BLK_RW_ASYNC, HZ/50);
2788                         goto next;
2789                 }
2790 skip:
2791                 node = rb_next(&dc->rb_node);
2792                 if (err)
2793                         __remove_discard_cmd(sbi, dc);
2794                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2795
2796                 if (fatal_signal_pending(current))
2797                         break;
2798         }
2799
2800         blk_finish_plug(&plug);
2801         mutex_unlock(&dcc->cmd_lock);
2802
2803         return trimmed;
2804 }
2805
2806 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2807 {
2808         __u64 start = F2FS_BYTES_TO_BLK(range->start);
2809         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2810         unsigned int start_segno, end_segno;
2811         block_t start_block, end_block;
2812         struct cp_control cpc;
2813         struct discard_policy dpolicy;
2814         unsigned long long trimmed = 0;
2815         int err = 0;
2816         bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
2817
2818         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2819                 return -EINVAL;
2820
2821         if (end < MAIN_BLKADDR(sbi))
2822                 goto out;
2823
2824         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2825                 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
2826                 return -EFSCORRUPTED;
2827         }
2828
2829         /* start/end segment number in main_area */
2830         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2831         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2832                                                 GET_SEGNO(sbi, end);
2833         if (need_align) {
2834                 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2835                 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2836         }
2837
2838         cpc.reason = CP_DISCARD;
2839         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2840         cpc.trim_start = start_segno;
2841         cpc.trim_end = end_segno;
2842
2843         if (sbi->discard_blks == 0)
2844                 goto out;
2845
2846         mutex_lock(&sbi->gc_mutex);
2847         err = f2fs_write_checkpoint(sbi, &cpc);
2848         mutex_unlock(&sbi->gc_mutex);
2849         if (err)
2850                 goto out;
2851
2852         /*
2853          * We filed discard candidates, but actually we don't need to wait for
2854          * all of them, since they'll be issued in idle time along with runtime
2855          * discard option. User configuration looks like using runtime discard
2856          * or periodic fstrim instead of it.
2857          */
2858         if (f2fs_realtime_discard_enable(sbi))
2859                 goto out;
2860
2861         start_block = START_BLOCK(sbi, start_segno);
2862         end_block = START_BLOCK(sbi, end_segno + 1);
2863
2864         __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2865         trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2866                                         start_block, end_block);
2867
2868         trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
2869                                         start_block, end_block);
2870 out:
2871         if (!err)
2872                 range->len = F2FS_BLK_TO_BYTES(trimmed);
2873         return err;
2874 }
2875
2876 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2877 {
2878         struct curseg_info *curseg = CURSEG_I(sbi, type);
2879         if (curseg->next_blkoff < sbi->blocks_per_seg)
2880                 return true;
2881         return false;
2882 }
2883
2884 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
2885 {
2886         switch (hint) {
2887         case WRITE_LIFE_SHORT:
2888                 return CURSEG_HOT_DATA;
2889         case WRITE_LIFE_EXTREME:
2890                 return CURSEG_COLD_DATA;
2891         default:
2892                 return CURSEG_WARM_DATA;
2893         }
2894 }
2895
2896 /* This returns write hints for each segment type. This hints will be
2897  * passed down to block layer. There are mapping tables which depend on
2898  * the mount option 'whint_mode'.
2899  *
2900  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2901  *
2902  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2903  *
2904  * User                  F2FS                     Block
2905  * ----                  ----                     -----
2906  *                       META                     WRITE_LIFE_NOT_SET
2907  *                       HOT_NODE                 "
2908  *                       WARM_NODE                "
2909  *                       COLD_NODE                "
2910  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2911  * extension list        "                        "
2912  *
2913  * -- buffered io
2914  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2915  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2916  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2917  * WRITE_LIFE_NONE       "                        "
2918  * WRITE_LIFE_MEDIUM     "                        "
2919  * WRITE_LIFE_LONG       "                        "
2920  *
2921  * -- direct io
2922  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2923  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2924  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2925  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2926  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2927  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2928  *
2929  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2930  *
2931  * User                  F2FS                     Block
2932  * ----                  ----                     -----
2933  *                       META                     WRITE_LIFE_MEDIUM;
2934  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
2935  *                       WARM_NODE                "
2936  *                       COLD_NODE                WRITE_LIFE_NONE
2937  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2938  * extension list        "                        "
2939  *
2940  * -- buffered io
2941  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2942  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2943  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
2944  * WRITE_LIFE_NONE       "                        "
2945  * WRITE_LIFE_MEDIUM     "                        "
2946  * WRITE_LIFE_LONG       "                        "
2947  *
2948  * -- direct io
2949  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2950  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2951  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2952  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2953  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2954  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2955  */
2956
2957 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2958                                 enum page_type type, enum temp_type temp)
2959 {
2960         if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2961                 if (type == DATA) {
2962                         if (temp == WARM)
2963                                 return WRITE_LIFE_NOT_SET;
2964                         else if (temp == HOT)
2965                                 return WRITE_LIFE_SHORT;
2966                         else if (temp == COLD)
2967                                 return WRITE_LIFE_EXTREME;
2968                 } else {
2969                         return WRITE_LIFE_NOT_SET;
2970                 }
2971         } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2972                 if (type == DATA) {
2973                         if (temp == WARM)
2974                                 return WRITE_LIFE_LONG;
2975                         else if (temp == HOT)
2976                                 return WRITE_LIFE_SHORT;
2977                         else if (temp == COLD)
2978                                 return WRITE_LIFE_EXTREME;
2979                 } else if (type == NODE) {
2980                         if (temp == WARM || temp == HOT)
2981                                 return WRITE_LIFE_NOT_SET;
2982                         else if (temp == COLD)
2983                                 return WRITE_LIFE_NONE;
2984                 } else if (type == META) {
2985                         return WRITE_LIFE_MEDIUM;
2986                 }
2987         }
2988         return WRITE_LIFE_NOT_SET;
2989 }
2990
2991 static int __get_segment_type_2(struct f2fs_io_info *fio)
2992 {
2993         if (fio->type == DATA)
2994                 return CURSEG_HOT_DATA;
2995         else
2996                 return CURSEG_HOT_NODE;
2997 }
2998
2999 static int __get_segment_type_4(struct f2fs_io_info *fio)
3000 {
3001         if (fio->type == DATA) {
3002                 struct inode *inode = fio->page->mapping->host;
3003
3004                 if (S_ISDIR(inode->i_mode))
3005                         return CURSEG_HOT_DATA;
3006                 else
3007                         return CURSEG_COLD_DATA;
3008         } else {
3009                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3010                         return CURSEG_WARM_NODE;
3011                 else
3012                         return CURSEG_COLD_NODE;
3013         }
3014 }
3015
3016 static int __get_segment_type_6(struct f2fs_io_info *fio)
3017 {
3018         if (fio->type == DATA) {
3019                 struct inode *inode = fio->page->mapping->host;
3020
3021                 if (is_cold_data(fio->page) || file_is_cold(inode))
3022                         return CURSEG_COLD_DATA;
3023                 if (file_is_hot(inode) ||
3024                                 is_inode_flag_set(inode, FI_HOT_DATA) ||
3025                                 f2fs_is_atomic_file(inode) ||
3026                                 f2fs_is_volatile_file(inode))
3027                         return CURSEG_HOT_DATA;
3028                 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3029         } else {
3030                 if (IS_DNODE(fio->page))
3031                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3032                                                 CURSEG_HOT_NODE;
3033                 return CURSEG_COLD_NODE;
3034         }
3035 }
3036
3037 static int __get_segment_type(struct f2fs_io_info *fio)
3038 {
3039         int type = 0;
3040
3041         switch (F2FS_OPTION(fio->sbi).active_logs) {
3042         case 2:
3043                 type = __get_segment_type_2(fio);
3044                 break;
3045         case 4:
3046                 type = __get_segment_type_4(fio);
3047                 break;
3048         case 6:
3049                 type = __get_segment_type_6(fio);
3050                 break;
3051         default:
3052                 f2fs_bug_on(fio->sbi, true);
3053         }
3054
3055         if (IS_HOT(type))
3056                 fio->temp = HOT;
3057         else if (IS_WARM(type))
3058                 fio->temp = WARM;
3059         else
3060                 fio->temp = COLD;
3061         return type;
3062 }
3063
3064 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3065                 block_t old_blkaddr, block_t *new_blkaddr,
3066                 struct f2fs_summary *sum, int type,
3067                 struct f2fs_io_info *fio, bool add_list)
3068 {
3069         struct sit_info *sit_i = SIT_I(sbi);
3070         struct curseg_info *curseg = CURSEG_I(sbi, type);
3071
3072         down_read(&SM_I(sbi)->curseg_lock);
3073
3074         mutex_lock(&curseg->curseg_mutex);
3075         down_write(&sit_i->sentry_lock);
3076
3077         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3078
3079         f2fs_wait_discard_bio(sbi, *new_blkaddr);
3080
3081         /*
3082          * __add_sum_entry should be resided under the curseg_mutex
3083          * because, this function updates a summary entry in the
3084          * current summary block.
3085          */
3086         __add_sum_entry(sbi, type, sum);
3087
3088         __refresh_next_blkoff(sbi, curseg);
3089
3090         stat_inc_block_count(sbi, curseg);
3091
3092         /*
3093          * SIT information should be updated before segment allocation,
3094          * since SSR needs latest valid block information.
3095          */
3096         update_sit_entry(sbi, *new_blkaddr, 1);
3097         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3098                 update_sit_entry(sbi, old_blkaddr, -1);
3099
3100         if (!__has_curseg_space(sbi, type))
3101                 sit_i->s_ops->allocate_segment(sbi, type, false);
3102
3103         /*
3104          * segment dirty status should be updated after segment allocation,
3105          * so we just need to update status only one time after previous
3106          * segment being closed.
3107          */
3108         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3109         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3110
3111         up_write(&sit_i->sentry_lock);
3112
3113         if (page && IS_NODESEG(type)) {
3114                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3115
3116                 f2fs_inode_chksum_set(sbi, page);
3117         }
3118
3119         if (F2FS_IO_ALIGNED(sbi))
3120                 fio->retry = false;
3121
3122         if (add_list) {
3123                 struct f2fs_bio_info *io;
3124
3125                 INIT_LIST_HEAD(&fio->list);
3126                 fio->in_list = true;
3127                 io = sbi->write_io[fio->type] + fio->temp;
3128                 spin_lock(&io->io_lock);
3129                 list_add_tail(&fio->list, &io->io_list);
3130                 spin_unlock(&io->io_lock);
3131         }
3132
3133         mutex_unlock(&curseg->curseg_mutex);
3134
3135         up_read(&SM_I(sbi)->curseg_lock);
3136 }
3137
3138 static void update_device_state(struct f2fs_io_info *fio)
3139 {
3140         struct f2fs_sb_info *sbi = fio->sbi;
3141         unsigned int devidx;
3142
3143         if (!f2fs_is_multi_device(sbi))
3144                 return;
3145
3146         devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3147
3148         /* update device state for fsync */
3149         f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
3150
3151         /* update device state for checkpoint */
3152         if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3153                 spin_lock(&sbi->dev_lock);
3154                 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3155                 spin_unlock(&sbi->dev_lock);
3156         }
3157 }
3158
3159 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3160 {
3161         int type = __get_segment_type(fio);
3162         bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
3163
3164         if (keep_order)
3165                 down_read(&fio->sbi->io_order_lock);
3166 reallocate:
3167         f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3168                         &fio->new_blkaddr, sum, type, fio, true);
3169         if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3170                 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3171                                         fio->old_blkaddr, fio->old_blkaddr);
3172
3173         /* writeout dirty page into bdev */
3174         f2fs_submit_page_write(fio);
3175         if (fio->retry) {
3176                 fio->old_blkaddr = fio->new_blkaddr;
3177                 goto reallocate;
3178         }
3179
3180         update_device_state(fio);
3181
3182         if (keep_order)
3183                 up_read(&fio->sbi->io_order_lock);
3184 }
3185
3186 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3187                                         enum iostat_type io_type)
3188 {
3189         struct f2fs_io_info fio = {
3190                 .sbi = sbi,
3191                 .type = META,
3192                 .temp = HOT,
3193                 .op = REQ_OP_WRITE,
3194                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3195                 .old_blkaddr = page->index,
3196                 .new_blkaddr = page->index,
3197                 .page = page,
3198                 .encrypted_page = NULL,
3199                 .in_list = false,
3200         };
3201
3202         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3203                 fio.op_flags &= ~REQ_META;
3204
3205         set_page_writeback(page);
3206         ClearPageError(page);
3207         f2fs_submit_page_write(&fio);
3208
3209         stat_inc_meta_count(sbi, page->index);
3210         f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3211 }
3212
3213 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3214 {
3215         struct f2fs_summary sum;
3216
3217         set_summary(&sum, nid, 0, 0);
3218         do_write_page(&sum, fio);
3219
3220         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3221 }
3222
3223 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3224                                         struct f2fs_io_info *fio)
3225 {
3226         struct f2fs_sb_info *sbi = fio->sbi;
3227         struct f2fs_summary sum;
3228
3229         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3230         set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3231         do_write_page(&sum, fio);
3232         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3233
3234         f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3235 }
3236
3237 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3238 {
3239         int err;
3240         struct f2fs_sb_info *sbi = fio->sbi;
3241         unsigned int segno;
3242
3243         fio->new_blkaddr = fio->old_blkaddr;
3244         /* i/o temperature is needed for passing down write hints */
3245         __get_segment_type(fio);
3246
3247         segno = GET_SEGNO(sbi, fio->new_blkaddr);
3248
3249         if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3250                 set_sbi_flag(sbi, SBI_NEED_FSCK);
3251                 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3252                           __func__, segno);
3253                 return -EFSCORRUPTED;
3254         }
3255
3256         stat_inc_inplace_blocks(fio->sbi);
3257
3258         if (fio->bio)
3259                 err = f2fs_merge_page_bio(fio);
3260         else
3261                 err = f2fs_submit_page_bio(fio);
3262         if (!err) {
3263                 update_device_state(fio);
3264                 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3265         }
3266
3267         return err;
3268 }
3269
3270 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3271                                                 unsigned int segno)
3272 {
3273         int i;
3274
3275         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3276                 if (CURSEG_I(sbi, i)->segno == segno)
3277                         break;
3278         }
3279         return i;
3280 }
3281
3282 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3283                                 block_t old_blkaddr, block_t new_blkaddr,
3284                                 bool recover_curseg, bool recover_newaddr)
3285 {
3286         struct sit_info *sit_i = SIT_I(sbi);
3287         struct curseg_info *curseg;
3288         unsigned int segno, old_cursegno;
3289         struct seg_entry *se;
3290         int type;
3291         unsigned short old_blkoff;
3292
3293         segno = GET_SEGNO(sbi, new_blkaddr);
3294         se = get_seg_entry(sbi, segno);
3295         type = se->type;
3296
3297         down_write(&SM_I(sbi)->curseg_lock);
3298
3299         if (!recover_curseg) {
3300                 /* for recovery flow */
3301                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3302                         if (old_blkaddr == NULL_ADDR)
3303                                 type = CURSEG_COLD_DATA;
3304                         else
3305                                 type = CURSEG_WARM_DATA;
3306                 }
3307         } else {
3308                 if (IS_CURSEG(sbi, segno)) {
3309                         /* se->type is volatile as SSR allocation */
3310                         type = __f2fs_get_curseg(sbi, segno);
3311                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3312                 } else {
3313                         type = CURSEG_WARM_DATA;
3314                 }
3315         }
3316
3317         f2fs_bug_on(sbi, !IS_DATASEG(type));
3318         curseg = CURSEG_I(sbi, type);
3319
3320         mutex_lock(&curseg->curseg_mutex);
3321         down_write(&sit_i->sentry_lock);
3322
3323         old_cursegno = curseg->segno;
3324         old_blkoff = curseg->next_blkoff;
3325
3326         /* change the current segment */
3327         if (segno != curseg->segno) {
3328                 curseg->next_segno = segno;
3329                 change_curseg(sbi, type);
3330         }
3331
3332         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3333         __add_sum_entry(sbi, type, sum);
3334
3335         if (!recover_curseg || recover_newaddr)
3336                 update_sit_entry(sbi, new_blkaddr, 1);
3337         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3338                 invalidate_mapping_pages(META_MAPPING(sbi),
3339                                         old_blkaddr, old_blkaddr);
3340                 update_sit_entry(sbi, old_blkaddr, -1);
3341         }
3342
3343         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3344         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3345
3346         locate_dirty_segment(sbi, old_cursegno);
3347
3348         if (recover_curseg) {
3349                 if (old_cursegno != curseg->segno) {
3350                         curseg->next_segno = old_cursegno;
3351                         change_curseg(sbi, type);
3352                 }
3353                 curseg->next_blkoff = old_blkoff;
3354         }
3355
3356         up_write(&sit_i->sentry_lock);
3357         mutex_unlock(&curseg->curseg_mutex);
3358         up_write(&SM_I(sbi)->curseg_lock);
3359 }
3360
3361 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3362                                 block_t old_addr, block_t new_addr,
3363                                 unsigned char version, bool recover_curseg,
3364                                 bool recover_newaddr)
3365 {
3366         struct f2fs_summary sum;
3367
3368         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3369
3370         f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3371                                         recover_curseg, recover_newaddr);
3372
3373         f2fs_update_data_blkaddr(dn, new_addr);
3374 }
3375
3376 void f2fs_wait_on_page_writeback(struct page *page,
3377                                 enum page_type type, bool ordered, bool locked)
3378 {
3379         if (PageWriteback(page)) {
3380                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3381
3382                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3383                 if (ordered) {
3384                         wait_on_page_writeback(page);
3385                         f2fs_bug_on(sbi, locked && PageWriteback(page));
3386                 } else {
3387                         wait_for_stable_page(page);
3388                 }
3389         }
3390 }
3391
3392 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3393 {
3394         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3395         struct page *cpage;
3396
3397         if (!f2fs_post_read_required(inode))
3398                 return;
3399
3400         if (!__is_valid_data_blkaddr(blkaddr))
3401                 return;
3402
3403         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3404         if (cpage) {
3405                 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3406                 f2fs_put_page(cpage, 1);
3407         }
3408 }
3409
3410 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3411                                                                 block_t len)
3412 {
3413         block_t i;
3414
3415         for (i = 0; i < len; i++)
3416                 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3417 }
3418
3419 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3420 {
3421         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3422         struct curseg_info *seg_i;
3423         unsigned char *kaddr;
3424         struct page *page;
3425         block_t start;
3426         int i, j, offset;
3427
3428         start = start_sum_block(sbi);
3429
3430         page = f2fs_get_meta_page(sbi, start++);
3431         if (IS_ERR(page))
3432                 return PTR_ERR(page);
3433         kaddr = (unsigned char *)page_address(page);
3434
3435         /* Step 1: restore nat cache */
3436         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3437         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3438
3439         /* Step 2: restore sit cache */
3440         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3441         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3442         offset = 2 * SUM_JOURNAL_SIZE;
3443
3444         /* Step 3: restore summary entries */
3445         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3446                 unsigned short blk_off;
3447                 unsigned int segno;
3448
3449                 seg_i = CURSEG_I(sbi, i);
3450                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3451                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3452                 seg_i->next_segno = segno;
3453                 reset_curseg(sbi, i, 0);
3454                 seg_i->alloc_type = ckpt->alloc_type[i];
3455                 seg_i->next_blkoff = blk_off;
3456
3457                 if (seg_i->alloc_type == SSR)
3458                         blk_off = sbi->blocks_per_seg;
3459
3460                 for (j = 0; j < blk_off; j++) {
3461                         struct f2fs_summary *s;
3462                         s = (struct f2fs_summary *)(kaddr + offset);
3463                         seg_i->sum_blk->entries[j] = *s;
3464                         offset += SUMMARY_SIZE;
3465                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3466                                                 SUM_FOOTER_SIZE)
3467                                 continue;
3468
3469                         f2fs_put_page(page, 1);
3470                         page = NULL;
3471
3472                         page = f2fs_get_meta_page(sbi, start++);
3473                         if (IS_ERR(page))
3474                                 return PTR_ERR(page);
3475                         kaddr = (unsigned char *)page_address(page);
3476                         offset = 0;
3477                 }
3478         }
3479         f2fs_put_page(page, 1);
3480         return 0;
3481 }
3482
3483 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3484 {
3485         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3486         struct f2fs_summary_block *sum;
3487         struct curseg_info *curseg;
3488         struct page *new;
3489         unsigned short blk_off;
3490         unsigned int segno = 0;
3491         block_t blk_addr = 0;
3492         int err = 0;
3493
3494         /* get segment number and block addr */
3495         if (IS_DATASEG(type)) {
3496                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3497                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3498                                                         CURSEG_HOT_DATA]);
3499                 if (__exist_node_summaries(sbi))
3500                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3501                 else
3502                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3503         } else {
3504                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3505                                                         CURSEG_HOT_NODE]);
3506                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3507                                                         CURSEG_HOT_NODE]);
3508                 if (__exist_node_summaries(sbi))
3509                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3510                                                         type - CURSEG_HOT_NODE);
3511                 else
3512                         blk_addr = GET_SUM_BLOCK(sbi, segno);
3513         }
3514
3515         new = f2fs_get_meta_page(sbi, blk_addr);
3516         if (IS_ERR(new))
3517                 return PTR_ERR(new);
3518         sum = (struct f2fs_summary_block *)page_address(new);
3519
3520         if (IS_NODESEG(type)) {
3521                 if (__exist_node_summaries(sbi)) {
3522                         struct f2fs_summary *ns = &sum->entries[0];
3523                         int i;
3524                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3525                                 ns->version = 0;
3526                                 ns->ofs_in_node = 0;
3527                         }
3528                 } else {
3529                         err = f2fs_restore_node_summary(sbi, segno, sum);
3530                         if (err)
3531                                 goto out;
3532                 }
3533         }
3534
3535         /* set uncompleted segment to curseg */
3536         curseg = CURSEG_I(sbi, type);
3537         mutex_lock(&curseg->curseg_mutex);
3538
3539         /* update journal info */
3540         down_write(&curseg->journal_rwsem);
3541         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3542         up_write(&curseg->journal_rwsem);
3543
3544         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3545         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3546         curseg->next_segno = segno;
3547         reset_curseg(sbi, type, 0);
3548         curseg->alloc_type = ckpt->alloc_type[type];
3549         curseg->next_blkoff = blk_off;
3550         mutex_unlock(&curseg->curseg_mutex);
3551 out:
3552         f2fs_put_page(new, 1);
3553         return err;
3554 }
3555
3556 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3557 {
3558         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3559         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3560         int type = CURSEG_HOT_DATA;
3561         int err;
3562
3563         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3564                 int npages = f2fs_npages_for_summary_flush(sbi, true);
3565
3566                 if (npages >= 2)
3567                         f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3568                                                         META_CP, true);
3569
3570                 /* restore for compacted data summary */
3571                 err = read_compacted_summaries(sbi);
3572                 if (err)
3573                         return err;
3574                 type = CURSEG_HOT_NODE;
3575         }
3576
3577         if (__exist_node_summaries(sbi))
3578                 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3579                                         NR_CURSEG_TYPE - type, META_CP, true);
3580
3581         for (; type <= CURSEG_COLD_NODE; type++) {
3582                 err = read_normal_summaries(sbi, type);
3583                 if (err)
3584                         return err;
3585         }
3586
3587         /* sanity check for summary blocks */
3588         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3589                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3590                 f2fs_err(sbi, "invalid journal entries nats %u sits %u\n",
3591                          nats_in_cursum(nat_j), sits_in_cursum(sit_j));
3592                 return -EINVAL;
3593         }
3594
3595         return 0;
3596 }
3597
3598 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3599 {
3600         struct page *page;
3601         unsigned char *kaddr;
3602         struct f2fs_summary *summary;
3603         struct curseg_info *seg_i;
3604         int written_size = 0;
3605         int i, j;
3606
3607         page = f2fs_grab_meta_page(sbi, blkaddr++);
3608         kaddr = (unsigned char *)page_address(page);
3609         memset(kaddr, 0, PAGE_SIZE);
3610
3611         /* Step 1: write nat cache */
3612         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3613         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3614         written_size += SUM_JOURNAL_SIZE;
3615
3616         /* Step 2: write sit cache */
3617         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3618         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3619         written_size += SUM_JOURNAL_SIZE;
3620
3621         /* Step 3: write summary entries */
3622         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3623                 unsigned short blkoff;
3624                 seg_i = CURSEG_I(sbi, i);
3625                 if (sbi->ckpt->alloc_type[i] == SSR)
3626                         blkoff = sbi->blocks_per_seg;
3627                 else
3628                         blkoff = curseg_blkoff(sbi, i);
3629
3630                 for (j = 0; j < blkoff; j++) {
3631                         if (!page) {
3632                                 page = f2fs_grab_meta_page(sbi, blkaddr++);
3633                                 kaddr = (unsigned char *)page_address(page);
3634                                 memset(kaddr, 0, PAGE_SIZE);
3635                                 written_size = 0;
3636                         }
3637                         summary = (struct f2fs_summary *)(kaddr + written_size);
3638                         *summary = seg_i->sum_blk->entries[j];
3639                         written_size += SUMMARY_SIZE;
3640
3641                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3642                                                         SUM_FOOTER_SIZE)
3643                                 continue;
3644
3645                         set_page_dirty(page);
3646                         f2fs_put_page(page, 1);
3647                         page = NULL;
3648                 }
3649         }
3650         if (page) {
3651                 set_page_dirty(page);
3652                 f2fs_put_page(page, 1);
3653         }
3654 }
3655
3656 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3657                                         block_t blkaddr, int type)
3658 {
3659         int i, end;
3660         if (IS_DATASEG(type))
3661                 end = type + NR_CURSEG_DATA_TYPE;
3662         else
3663                 end = type + NR_CURSEG_NODE_TYPE;
3664
3665         for (i = type; i < end; i++)
3666                 write_current_sum_page(sbi, i, blkaddr + (i - type));
3667 }
3668
3669 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3670 {
3671         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3672                 write_compacted_summaries(sbi, start_blk);
3673         else
3674                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3675 }
3676
3677 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3678 {
3679         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3680 }
3681
3682 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3683                                         unsigned int val, int alloc)
3684 {
3685         int i;
3686
3687         if (type == NAT_JOURNAL) {
3688                 for (i = 0; i < nats_in_cursum(journal); i++) {
3689                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3690                                 return i;
3691                 }
3692                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3693                         return update_nats_in_cursum(journal, 1);
3694         } else if (type == SIT_JOURNAL) {
3695                 for (i = 0; i < sits_in_cursum(journal); i++)
3696                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3697                                 return i;
3698                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3699                         return update_sits_in_cursum(journal, 1);
3700         }
3701         return -1;
3702 }
3703
3704 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3705                                         unsigned int segno)
3706 {
3707         return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
3708 }
3709
3710 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3711                                         unsigned int start)
3712 {
3713         struct sit_info *sit_i = SIT_I(sbi);
3714         struct page *page;
3715         pgoff_t src_off, dst_off;
3716
3717         src_off = current_sit_addr(sbi, start);
3718         dst_off = next_sit_addr(sbi, src_off);
3719
3720         page = f2fs_grab_meta_page(sbi, dst_off);
3721         seg_info_to_sit_page(sbi, page, start);
3722
3723         set_page_dirty(page);
3724         set_to_next_sit(sit_i, start);
3725
3726         return page;
3727 }
3728
3729 static struct sit_entry_set *grab_sit_entry_set(void)
3730 {
3731         struct sit_entry_set *ses =
3732                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3733
3734         ses->entry_cnt = 0;
3735         INIT_LIST_HEAD(&ses->set_list);
3736         return ses;
3737 }
3738
3739 static void release_sit_entry_set(struct sit_entry_set *ses)
3740 {
3741         list_del(&ses->set_list);
3742         kmem_cache_free(sit_entry_set_slab, ses);
3743 }
3744
3745 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3746                                                 struct list_head *head)
3747 {
3748         struct sit_entry_set *next = ses;
3749
3750         if (list_is_last(&ses->set_list, head))
3751                 return;
3752
3753         list_for_each_entry_continue(next, head, set_list)
3754                 if (ses->entry_cnt <= next->entry_cnt)
3755                         break;
3756
3757         list_move_tail(&ses->set_list, &next->set_list);
3758 }
3759
3760 static void add_sit_entry(unsigned int segno, struct list_head *head)
3761 {
3762         struct sit_entry_set *ses;
3763         unsigned int start_segno = START_SEGNO(segno);
3764
3765         list_for_each_entry(ses, head, set_list) {
3766                 if (ses->start_segno == start_segno) {
3767                         ses->entry_cnt++;
3768                         adjust_sit_entry_set(ses, head);
3769                         return;
3770                 }
3771         }
3772
3773         ses = grab_sit_entry_set();
3774
3775         ses->start_segno = start_segno;
3776         ses->entry_cnt++;
3777         list_add(&ses->set_list, head);
3778 }
3779
3780 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3781 {
3782         struct f2fs_sm_info *sm_info = SM_I(sbi);
3783         struct list_head *set_list = &sm_info->sit_entry_set;
3784         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3785         unsigned int segno;
3786
3787         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3788                 add_sit_entry(segno, set_list);
3789 }
3790
3791 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3792 {
3793         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3794         struct f2fs_journal *journal = curseg->journal;
3795         int i;
3796
3797         down_write(&curseg->journal_rwsem);
3798         for (i = 0; i < sits_in_cursum(journal); i++) {
3799                 unsigned int segno;
3800                 bool dirtied;
3801
3802                 segno = le32_to_cpu(segno_in_journal(journal, i));
3803                 dirtied = __mark_sit_entry_dirty(sbi, segno);
3804
3805                 if (!dirtied)
3806                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3807         }
3808         update_sits_in_cursum(journal, -i);
3809         up_write(&curseg->journal_rwsem);
3810 }
3811
3812 /*
3813  * CP calls this function, which flushes SIT entries including sit_journal,
3814  * and moves prefree segs to free segs.
3815  */
3816 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3817 {
3818         struct sit_info *sit_i = SIT_I(sbi);
3819         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3820         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3821         struct f2fs_journal *journal = curseg->journal;
3822         struct sit_entry_set *ses, *tmp;
3823         struct list_head *head = &SM_I(sbi)->sit_entry_set;
3824         bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
3825         struct seg_entry *se;
3826
3827         down_write(&sit_i->sentry_lock);
3828
3829         if (!sit_i->dirty_sentries)
3830                 goto out;
3831
3832         /*
3833          * add and account sit entries of dirty bitmap in sit entry
3834          * set temporarily
3835          */
3836         add_sits_in_set(sbi);
3837
3838         /*
3839          * if there are no enough space in journal to store dirty sit
3840          * entries, remove all entries from journal and add and account
3841          * them in sit entry set.
3842          */
3843         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
3844                                                                 !to_journal)
3845                 remove_sits_in_journal(sbi);
3846
3847         /*
3848          * there are two steps to flush sit entries:
3849          * #1, flush sit entries to journal in current cold data summary block.
3850          * #2, flush sit entries to sit page.
3851          */
3852         list_for_each_entry_safe(ses, tmp, head, set_list) {
3853                 struct page *page = NULL;
3854                 struct f2fs_sit_block *raw_sit = NULL;
3855                 unsigned int start_segno = ses->start_segno;
3856                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3857                                                 (unsigned long)MAIN_SEGS(sbi));
3858                 unsigned int segno = start_segno;
3859
3860                 if (to_journal &&
3861                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3862                         to_journal = false;
3863
3864                 if (to_journal) {
3865                         down_write(&curseg->journal_rwsem);
3866                 } else {
3867                         page = get_next_sit_page(sbi, start_segno);
3868                         raw_sit = page_address(page);
3869                 }
3870
3871                 /* flush dirty sit entries in region of current sit set */
3872                 for_each_set_bit_from(segno, bitmap, end) {
3873                         int offset, sit_offset;
3874
3875                         se = get_seg_entry(sbi, segno);
3876 #ifdef CONFIG_F2FS_CHECK_FS
3877                         if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3878                                                 SIT_VBLOCK_MAP_SIZE))
3879                                 f2fs_bug_on(sbi, 1);
3880 #endif
3881
3882                         /* add discard candidates */
3883                         if (!(cpc->reason & CP_DISCARD)) {
3884                                 cpc->trim_start = segno;
3885                                 add_discard_addrs(sbi, cpc, false);
3886                         }
3887
3888                         if (to_journal) {
3889                                 offset = f2fs_lookup_journal_in_cursum(journal,
3890                                                         SIT_JOURNAL, segno, 1);
3891                                 f2fs_bug_on(sbi, offset < 0);
3892                                 segno_in_journal(journal, offset) =
3893                                                         cpu_to_le32(segno);
3894                                 seg_info_to_raw_sit(se,
3895                                         &sit_in_journal(journal, offset));
3896                                 check_block_count(sbi, segno,
3897                                         &sit_in_journal(journal, offset));
3898                         } else {
3899                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3900                                 seg_info_to_raw_sit(se,
3901                                                 &raw_sit->entries[sit_offset]);
3902                                 check_block_count(sbi, segno,
3903                                                 &raw_sit->entries[sit_offset]);
3904                         }
3905
3906                         __clear_bit(segno, bitmap);
3907                         sit_i->dirty_sentries--;
3908                         ses->entry_cnt--;
3909                 }
3910
3911                 if (to_journal)
3912                         up_write(&curseg->journal_rwsem);
3913                 else
3914                         f2fs_put_page(page, 1);
3915
3916                 f2fs_bug_on(sbi, ses->entry_cnt);
3917                 release_sit_entry_set(ses);
3918         }
3919
3920         f2fs_bug_on(sbi, !list_empty(head));
3921         f2fs_bug_on(sbi, sit_i->dirty_sentries);
3922 out:
3923         if (cpc->reason & CP_DISCARD) {
3924                 __u64 trim_start = cpc->trim_start;
3925
3926                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3927                         add_discard_addrs(sbi, cpc, false);
3928
3929                 cpc->trim_start = trim_start;
3930         }
3931         up_write(&sit_i->sentry_lock);
3932
3933         set_prefree_as_free_segments(sbi);
3934 }
3935
3936 static int build_sit_info(struct f2fs_sb_info *sbi)
3937 {
3938         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3939         struct sit_info *sit_i;
3940         unsigned int sit_segs, start;
3941         char *src_bitmap, *bitmap;
3942         unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
3943
3944         /* allocate memory for SIT information */
3945         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3946         if (!sit_i)
3947                 return -ENOMEM;
3948
3949         SM_I(sbi)->sit_info = sit_i;
3950
3951         sit_i->sentries =
3952                 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3953                                               MAIN_SEGS(sbi)),
3954                               GFP_KERNEL);
3955         if (!sit_i->sentries)
3956                 return -ENOMEM;
3957
3958         main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3959         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
3960                                                                 GFP_KERNEL);
3961         if (!sit_i->dirty_sentries_bitmap)
3962                 return -ENOMEM;
3963
3964 #ifdef CONFIG_F2FS_CHECK_FS
3965         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 4;
3966 #else
3967         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 3;
3968 #endif
3969         sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
3970         if (!sit_i->bitmap)
3971                 return -ENOMEM;
3972
3973         bitmap = sit_i->bitmap;
3974
3975         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3976                 sit_i->sentries[start].cur_valid_map = bitmap;
3977                 bitmap += SIT_VBLOCK_MAP_SIZE;
3978
3979                 sit_i->sentries[start].ckpt_valid_map = bitmap;
3980                 bitmap += SIT_VBLOCK_MAP_SIZE;
3981
3982 #ifdef CONFIG_F2FS_CHECK_FS
3983                 sit_i->sentries[start].cur_valid_map_mir = bitmap;
3984                 bitmap += SIT_VBLOCK_MAP_SIZE;
3985 #endif
3986
3987                 sit_i->sentries[start].discard_map = bitmap;
3988                 bitmap += SIT_VBLOCK_MAP_SIZE;
3989         }
3990
3991         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3992         if (!sit_i->tmp_map)
3993                 return -ENOMEM;
3994
3995         if (__is_large_section(sbi)) {
3996                 sit_i->sec_entries =
3997                         f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
3998                                                       MAIN_SECS(sbi)),
3999                                       GFP_KERNEL);
4000                 if (!sit_i->sec_entries)
4001                         return -ENOMEM;
4002         }
4003
4004         /* get information related with SIT */
4005         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4006
4007         /* setup SIT bitmap from ckeckpoint pack */
4008         sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4009         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4010
4011         sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4012         if (!sit_i->sit_bitmap)
4013                 return -ENOMEM;
4014
4015 #ifdef CONFIG_F2FS_CHECK_FS
4016         sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4017                                         sit_bitmap_size, GFP_KERNEL);
4018         if (!sit_i->sit_bitmap_mir)
4019                 return -ENOMEM;
4020
4021         sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4022                                         main_bitmap_size, GFP_KERNEL);
4023         if (!sit_i->invalid_segmap)
4024                 return -ENOMEM;
4025 #endif
4026
4027         /* init SIT information */
4028         sit_i->s_ops = &default_salloc_ops;
4029
4030         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4031         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4032         sit_i->written_valid_blocks = 0;
4033         sit_i->bitmap_size = sit_bitmap_size;
4034         sit_i->dirty_sentries = 0;
4035         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4036         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4037         sit_i->mounted_time = ktime_get_real_seconds();
4038         init_rwsem(&sit_i->sentry_lock);
4039         return 0;
4040 }
4041
4042 static int build_free_segmap(struct f2fs_sb_info *sbi)
4043 {
4044         struct free_segmap_info *free_i;
4045         unsigned int bitmap_size, sec_bitmap_size;
4046
4047         /* allocate memory for free segmap information */
4048         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4049         if (!free_i)
4050                 return -ENOMEM;
4051
4052         SM_I(sbi)->free_info = free_i;
4053
4054         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4055         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4056         if (!free_i->free_segmap)
4057                 return -ENOMEM;
4058
4059         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4060         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4061         if (!free_i->free_secmap)
4062                 return -ENOMEM;
4063
4064         /* set all segments as dirty temporarily */
4065         memset(free_i->free_segmap, 0xff, bitmap_size);
4066         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4067
4068         /* init free segmap information */
4069         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4070         free_i->free_segments = 0;
4071         free_i->free_sections = 0;
4072         spin_lock_init(&free_i->segmap_lock);
4073         return 0;
4074 }
4075
4076 static int build_curseg(struct f2fs_sb_info *sbi)
4077 {
4078         struct curseg_info *array;
4079         int i;
4080
4081         array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
4082                              GFP_KERNEL);
4083         if (!array)
4084                 return -ENOMEM;
4085
4086         SM_I(sbi)->curseg_array = array;
4087
4088         for (i = 0; i < NR_CURSEG_TYPE; i++) {
4089                 mutex_init(&array[i].curseg_mutex);
4090                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4091                 if (!array[i].sum_blk)
4092                         return -ENOMEM;
4093                 init_rwsem(&array[i].journal_rwsem);
4094                 array[i].journal = f2fs_kzalloc(sbi,
4095                                 sizeof(struct f2fs_journal), GFP_KERNEL);
4096                 if (!array[i].journal)
4097                         return -ENOMEM;
4098                 array[i].segno = NULL_SEGNO;
4099                 array[i].next_blkoff = 0;
4100         }
4101         return restore_curseg_summaries(sbi);
4102 }
4103
4104 static int build_sit_entries(struct f2fs_sb_info *sbi)
4105 {
4106         struct sit_info *sit_i = SIT_I(sbi);
4107         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4108         struct f2fs_journal *journal = curseg->journal;
4109         struct seg_entry *se;
4110         struct f2fs_sit_entry sit;
4111         int sit_blk_cnt = SIT_BLK_CNT(sbi);
4112         unsigned int i, start, end;
4113         unsigned int readed, start_blk = 0;
4114         int err = 0;
4115         block_t total_node_blocks = 0;
4116
4117         do {
4118                 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
4119                                                         META_SIT, true);
4120
4121                 start = start_blk * sit_i->sents_per_block;
4122                 end = (start_blk + readed) * sit_i->sents_per_block;
4123
4124                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4125                         struct f2fs_sit_block *sit_blk;
4126                         struct page *page;
4127
4128                         se = &sit_i->sentries[start];
4129                         page = get_current_sit_page(sbi, start);
4130                         if (IS_ERR(page))
4131                                 return PTR_ERR(page);
4132                         sit_blk = (struct f2fs_sit_block *)page_address(page);
4133                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4134                         f2fs_put_page(page, 1);
4135
4136                         err = check_block_count(sbi, start, &sit);
4137                         if (err)
4138                                 return err;
4139                         seg_info_from_raw_sit(se, &sit);
4140                         if (IS_NODESEG(se->type))
4141                                 total_node_blocks += se->valid_blocks;
4142
4143                         /* build discard map only one time */
4144                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4145                                 memset(se->discard_map, 0xff,
4146                                         SIT_VBLOCK_MAP_SIZE);
4147                         } else {
4148                                 memcpy(se->discard_map,
4149                                         se->cur_valid_map,
4150                                         SIT_VBLOCK_MAP_SIZE);
4151                                 sbi->discard_blks +=
4152                                         sbi->blocks_per_seg -
4153                                         se->valid_blocks;
4154                         }
4155
4156                         if (__is_large_section(sbi))
4157                                 get_sec_entry(sbi, start)->valid_blocks +=
4158                                                         se->valid_blocks;
4159                 }
4160                 start_blk += readed;
4161         } while (start_blk < sit_blk_cnt);
4162
4163         down_read(&curseg->journal_rwsem);
4164         for (i = 0; i < sits_in_cursum(journal); i++) {
4165                 unsigned int old_valid_blocks;
4166
4167                 start = le32_to_cpu(segno_in_journal(journal, i));
4168                 if (start >= MAIN_SEGS(sbi)) {
4169                         f2fs_err(sbi, "Wrong journal entry on segno %u",
4170                                  start);
4171                         err = -EFSCORRUPTED;
4172                         break;
4173                 }
4174
4175                 se = &sit_i->sentries[start];
4176                 sit = sit_in_journal(journal, i);
4177
4178                 old_valid_blocks = se->valid_blocks;
4179                 if (IS_NODESEG(se->type))
4180                         total_node_blocks -= old_valid_blocks;
4181
4182                 err = check_block_count(sbi, start, &sit);
4183                 if (err)
4184                         break;
4185                 seg_info_from_raw_sit(se, &sit);
4186                 if (IS_NODESEG(se->type))
4187                         total_node_blocks += se->valid_blocks;
4188
4189                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4190                         memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4191                 } else {
4192                         memcpy(se->discard_map, se->cur_valid_map,
4193                                                 SIT_VBLOCK_MAP_SIZE);
4194                         sbi->discard_blks += old_valid_blocks;
4195                         sbi->discard_blks -= se->valid_blocks;
4196                 }
4197
4198                 if (__is_large_section(sbi)) {
4199                         get_sec_entry(sbi, start)->valid_blocks +=
4200                                                         se->valid_blocks;
4201                         get_sec_entry(sbi, start)->valid_blocks -=
4202                                                         old_valid_blocks;
4203                 }
4204         }
4205         up_read(&curseg->journal_rwsem);
4206
4207         if (!err && total_node_blocks != valid_node_count(sbi)) {
4208                 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4209                          total_node_blocks, valid_node_count(sbi));
4210                 err = -EFSCORRUPTED;
4211         }
4212
4213         return err;
4214 }
4215
4216 static void init_free_segmap(struct f2fs_sb_info *sbi)
4217 {
4218         unsigned int start;
4219         int type;
4220
4221         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4222                 struct seg_entry *sentry = get_seg_entry(sbi, start);
4223                 if (!sentry->valid_blocks)
4224                         __set_free(sbi, start);
4225                 else
4226                         SIT_I(sbi)->written_valid_blocks +=
4227                                                 sentry->valid_blocks;
4228         }
4229
4230         /* set use the current segments */
4231         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4232                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4233                 __set_test_and_inuse(sbi, curseg_t->segno);
4234         }
4235 }
4236
4237 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4238 {
4239         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4240         struct free_segmap_info *free_i = FREE_I(sbi);
4241         unsigned int segno = 0, offset = 0;
4242         unsigned short valid_blocks;
4243
4244         while (1) {
4245                 /* find dirty segment based on free segmap */
4246                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4247                 if (segno >= MAIN_SEGS(sbi))
4248                         break;
4249                 offset = segno + 1;
4250                 valid_blocks = get_valid_blocks(sbi, segno, false);
4251                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
4252                         continue;
4253                 if (valid_blocks > sbi->blocks_per_seg) {
4254                         f2fs_bug_on(sbi, 1);
4255                         continue;
4256                 }
4257                 mutex_lock(&dirty_i->seglist_lock);
4258                 __locate_dirty_segment(sbi, segno, DIRTY);
4259                 mutex_unlock(&dirty_i->seglist_lock);
4260         }
4261 }
4262
4263 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4264 {
4265         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4266         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4267
4268         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4269         if (!dirty_i->victim_secmap)
4270                 return -ENOMEM;
4271         return 0;
4272 }
4273
4274 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4275 {
4276         struct dirty_seglist_info *dirty_i;
4277         unsigned int bitmap_size, i;
4278
4279         /* allocate memory for dirty segments list information */
4280         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4281                                                                 GFP_KERNEL);
4282         if (!dirty_i)
4283                 return -ENOMEM;
4284
4285         SM_I(sbi)->dirty_info = dirty_i;
4286         mutex_init(&dirty_i->seglist_lock);
4287
4288         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4289
4290         for (i = 0; i < NR_DIRTY_TYPE; i++) {
4291                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4292                                                                 GFP_KERNEL);
4293                 if (!dirty_i->dirty_segmap[i])
4294                         return -ENOMEM;
4295         }
4296
4297         init_dirty_segmap(sbi);
4298         return init_victim_secmap(sbi);
4299 }
4300
4301 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4302 {
4303         int i;
4304
4305         /*
4306          * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4307          * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4308          */
4309         for (i = 0; i < NO_CHECK_TYPE; i++) {
4310                 struct curseg_info *curseg = CURSEG_I(sbi, i);
4311                 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4312                 unsigned int blkofs = curseg->next_blkoff;
4313
4314                 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4315                         goto out;
4316
4317                 if (curseg->alloc_type == SSR)
4318                         continue;
4319
4320                 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4321                         if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4322                                 continue;
4323 out:
4324                         f2fs_err(sbi,
4325                                  "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4326                                  i, curseg->segno, curseg->alloc_type,
4327                                  curseg->next_blkoff, blkofs);
4328                         return -EFSCORRUPTED;
4329                 }
4330         }
4331         return 0;
4332 }
4333
4334 /*
4335  * Update min, max modified time for cost-benefit GC algorithm
4336  */
4337 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4338 {
4339         struct sit_info *sit_i = SIT_I(sbi);
4340         unsigned int segno;
4341
4342         down_write(&sit_i->sentry_lock);
4343
4344         sit_i->min_mtime = ULLONG_MAX;
4345
4346         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4347                 unsigned int i;
4348                 unsigned long long mtime = 0;
4349
4350                 for (i = 0; i < sbi->segs_per_sec; i++)
4351                         mtime += get_seg_entry(sbi, segno + i)->mtime;
4352
4353                 mtime = div_u64(mtime, sbi->segs_per_sec);
4354
4355                 if (sit_i->min_mtime > mtime)
4356                         sit_i->min_mtime = mtime;
4357         }
4358         sit_i->max_mtime = get_mtime(sbi, false);
4359         up_write(&sit_i->sentry_lock);
4360 }
4361
4362 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
4363 {
4364         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4365         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4366         struct f2fs_sm_info *sm_info;
4367         int err;
4368
4369         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
4370         if (!sm_info)
4371                 return -ENOMEM;
4372
4373         /* init sm info */
4374         sbi->sm_info = sm_info;
4375         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4376         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4377         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4378         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4379         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4380         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4381         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
4382         sm_info->rec_prefree_segments = sm_info->main_segments *
4383                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
4384         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4385                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4386
4387         if (!test_opt(sbi, LFS))
4388                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
4389         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
4390         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
4391         sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
4392         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
4393         sm_info->min_ssr_sections = reserved_sections(sbi);
4394
4395         INIT_LIST_HEAD(&sm_info->sit_entry_set);
4396
4397         init_rwsem(&sm_info->curseg_lock);
4398
4399         if (!f2fs_readonly(sbi->sb)) {
4400                 err = f2fs_create_flush_cmd_control(sbi);
4401                 if (err)
4402                         return err;
4403         }
4404
4405         err = create_discard_cmd_control(sbi);
4406         if (err)
4407                 return err;
4408
4409         err = build_sit_info(sbi);
4410         if (err)
4411                 return err;
4412         err = build_free_segmap(sbi);
4413         if (err)
4414                 return err;
4415         err = build_curseg(sbi);
4416         if (err)
4417                 return err;
4418
4419         /* reinit free segmap based on SIT */
4420         err = build_sit_entries(sbi);
4421         if (err)
4422                 return err;
4423
4424         init_free_segmap(sbi);
4425         err = build_dirty_segmap(sbi);
4426         if (err)
4427                 return err;
4428
4429         err = sanity_check_curseg(sbi);
4430         if (err)
4431                 return err;
4432
4433         init_min_max_mtime(sbi);
4434         return 0;
4435 }
4436
4437 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4438                 enum dirty_type dirty_type)
4439 {
4440         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4441
4442         mutex_lock(&dirty_i->seglist_lock);
4443         kvfree(dirty_i->dirty_segmap[dirty_type]);
4444         dirty_i->nr_dirty[dirty_type] = 0;
4445         mutex_unlock(&dirty_i->seglist_lock);
4446 }
4447
4448 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
4449 {
4450         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4451         kvfree(dirty_i->victim_secmap);
4452 }
4453
4454 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4455 {
4456         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4457         int i;
4458
4459         if (!dirty_i)
4460                 return;
4461
4462         /* discard pre-free/dirty segments list */
4463         for (i = 0; i < NR_DIRTY_TYPE; i++)
4464                 discard_dirty_segmap(sbi, i);
4465
4466         destroy_victim_secmap(sbi);
4467         SM_I(sbi)->dirty_info = NULL;
4468         kvfree(dirty_i);
4469 }
4470
4471 static void destroy_curseg(struct f2fs_sb_info *sbi)
4472 {
4473         struct curseg_info *array = SM_I(sbi)->curseg_array;
4474         int i;
4475
4476         if (!array)
4477                 return;
4478         SM_I(sbi)->curseg_array = NULL;
4479         for (i = 0; i < NR_CURSEG_TYPE; i++) {
4480                 kvfree(array[i].sum_blk);
4481                 kvfree(array[i].journal);
4482         }
4483         kvfree(array);
4484 }
4485
4486 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4487 {
4488         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4489         if (!free_i)
4490                 return;
4491         SM_I(sbi)->free_info = NULL;
4492         kvfree(free_i->free_segmap);
4493         kvfree(free_i->free_secmap);
4494         kvfree(free_i);
4495 }
4496
4497 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4498 {
4499         struct sit_info *sit_i = SIT_I(sbi);
4500
4501         if (!sit_i)
4502                 return;
4503
4504         if (sit_i->sentries)
4505                 kvfree(sit_i->bitmap);
4506         kvfree(sit_i->tmp_map);
4507
4508         kvfree(sit_i->sentries);
4509         kvfree(sit_i->sec_entries);
4510         kvfree(sit_i->dirty_sentries_bitmap);
4511
4512         SM_I(sbi)->sit_info = NULL;
4513         kvfree(sit_i->sit_bitmap);
4514 #ifdef CONFIG_F2FS_CHECK_FS
4515         kvfree(sit_i->sit_bitmap_mir);
4516         kvfree(sit_i->invalid_segmap);
4517 #endif
4518         kvfree(sit_i);
4519 }
4520
4521 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
4522 {
4523         struct f2fs_sm_info *sm_info = SM_I(sbi);
4524
4525         if (!sm_info)
4526                 return;
4527         f2fs_destroy_flush_cmd_control(sbi, true);
4528         destroy_discard_cmd_control(sbi);
4529         destroy_dirty_segmap(sbi);
4530         destroy_curseg(sbi);
4531         destroy_free_segmap(sbi);
4532         destroy_sit_info(sbi);
4533         sbi->sm_info = NULL;
4534         kvfree(sm_info);
4535 }
4536
4537 int __init f2fs_create_segment_manager_caches(void)
4538 {
4539         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4540                         sizeof(struct discard_entry));
4541         if (!discard_entry_slab)
4542                 goto fail;
4543
4544         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4545                         sizeof(struct discard_cmd));
4546         if (!discard_cmd_slab)
4547                 goto destroy_discard_entry;
4548
4549         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4550                         sizeof(struct sit_entry_set));
4551         if (!sit_entry_set_slab)
4552                 goto destroy_discard_cmd;
4553
4554         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4555                         sizeof(struct inmem_pages));
4556         if (!inmem_entry_slab)
4557                 goto destroy_sit_entry_set;
4558         return 0;
4559
4560 destroy_sit_entry_set:
4561         kmem_cache_destroy(sit_entry_set_slab);
4562 destroy_discard_cmd:
4563         kmem_cache_destroy(discard_cmd_slab);
4564 destroy_discard_entry:
4565         kmem_cache_destroy(discard_entry_slab);
4566 fail:
4567         return -ENOMEM;
4568 }
4569
4570 void f2fs_destroy_segment_manager_caches(void)
4571 {
4572         kmem_cache_destroy(sit_entry_set_slab);
4573         kmem_cache_destroy(discard_cmd_slab);
4574         kmem_cache_destroy(discard_entry_slab);
4575         kmem_cache_destroy(inmem_entry_slab);
4576 }