fs/f2fs/segment.c

   1 /*
   2  * fs/f2fs/segment.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/bio.h>
  14 #include <linux/blkdev.h>
  15 #include <linux/prefetch.h>
  16 #include <linux/kthread.h>
  17 #include <linux/swap.h>
  18 #include <linux/timer.h>
  19 #include <linux/freezer.h>
  20 #include <linux/sched/signal.h>
  21
  22 #include "f2fs.h"
  23 #include "segment.h"
  24 #include "node.h"
  25 #include "gc.h"
  26 #include "trace.h"
  27 #include <trace/events/f2fs.h>
  28
  29 #define __reverse_ffz(x) __reverse_ffs(~(x))
  30
  31 static struct kmem_cache *discard_entry_slab;
  32 static struct kmem_cache *discard_cmd_slab;
  33 static struct kmem_cache *sit_entry_set_slab;
  34 static struct kmem_cache *inmem_entry_slab;
  35
  36 static unsigned long __reverse_ulong(unsigned char *str)
  37 {
  38         unsigned long tmp = 0;
  39         int shift = 24, idx = 0;
  40
  41 #if BITS_PER_LONG == 64
  42         shift = 56;
  43 #endif
  44         while (shift >= 0) {
  45                 tmp |= (unsigned long)str[idx++] << shift;
  46                 shift -= BITS_PER_BYTE;
  47         }
  48         return tmp;
  49 }
  50
  51 /*
  52  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  53  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  54  */
  55 static inline unsigned long __reverse_ffs(unsigned long word)
  56 {
  57         int num = 0;
  58
  59 #if BITS_PER_LONG == 64
  60         if ((word & 0xffffffff00000000UL) == 0)
  61                 num += 32;
  62         else
  63                 word >>= 32;
  64 #endif
  65         if ((word & 0xffff0000) == 0)
  66                 num += 16;
  67         else
  68                 word >>= 16;
  69
  70         if ((word & 0xff00) == 0)
  71                 num += 8;
  72         else
  73                 word >>= 8;
  74
  75         if ((word & 0xf0) == 0)
  76                 num += 4;
  77         else
  78                 word >>= 4;
  79
  80         if ((word & 0xc) == 0)
  81                 num += 2;
  82         else
  83                 word >>= 2;
  84
  85         if ((word & 0x2) == 0)
  86                 num += 1;
  87         return num;
  88 }
  89
  90 /*
  91  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  92  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  93  * @size must be integral times of unsigned long.
  94  * Example:
  95  *                             MSB <--> LSB
  96  *   f2fs_set_bit(0, bitmap) => 1000 0000
  97  *   f2fs_set_bit(7, bitmap) => 0000 0001
  98  */
  99 static unsigned long __find_rev_next_bit(const unsigned long *addr,
 100                         unsigned long size, unsigned long offset)
 101 {
 102         const unsigned long *p = addr + BIT_WORD(offset);
 103         unsigned long result = size;
 104         unsigned long tmp;
 105
 106         if (offset >= size)
 107                 return size;
 108
 109         size -= (offset & ~(BITS_PER_LONG - 1));
 110         offset %= BITS_PER_LONG;
 111
 112         while (1) {
 113                 if (*p == 0)
 114                         goto pass;
 115
 116                 tmp = __reverse_ulong((unsigned char *)p);
 117
 118                 tmp &= ~0UL >> offset;
 119                 if (size < BITS_PER_LONG)
 120                         tmp &= (~0UL << (BITS_PER_LONG - size));
 121                 if (tmp)
 122                         goto found;
 123 pass:
 124                 if (size <= BITS_PER_LONG)
 125                         break;
 126                 size -= BITS_PER_LONG;
 127                 offset = 0;
 128                 p++;
 129         }
 130         return result;
 131 found:
 132         return result - size + __reverse_ffs(tmp);
 133 }
 134
 135 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 136                         unsigned long size, unsigned long offset)
 137 {
 138         const unsigned long *p = addr + BIT_WORD(offset);
 139         unsigned long result = size;
 140         unsigned long tmp;
 141
 142         if (offset >= size)
 143                 return size;
 144
 145         size -= (offset & ~(BITS_PER_LONG - 1));
 146         offset %= BITS_PER_LONG;
 147
 148         while (1) {
 149                 if (*p == ~0UL)
 150                         goto pass;
 151
 152                 tmp = __reverse_ulong((unsigned char *)p);
 153
 154                 if (offset)
 155                         tmp |= ~0UL << (BITS_PER_LONG - offset);
 156                 if (size < BITS_PER_LONG)
 157                         tmp |= ~0UL >> size;
 158                 if (tmp != ~0UL)
 159                         goto found;
 160 pass:
 161                 if (size <= BITS_PER_LONG)
 162                         break;
 163                 size -= BITS_PER_LONG;
 164                 offset = 0;
 165                 p++;
 166         }
 167         return result;
 168 found:
 169         return result - size + __reverse_ffz(tmp);
 170 }
 171
 172 bool need_SSR(struct f2fs_sb_info *sbi)
 173 {
 174         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 175         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 176         int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
 177
 178         if (test_opt(sbi, LFS))
 179                 return false;
 180         if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
 181                 return true;
 182
 183         return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
 184                         SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
 185 }
 186
 187 void register_inmem_page(struct inode *inode, struct page *page)
 188 {
 189         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 190         struct f2fs_inode_info *fi = F2FS_I(inode);
 191         struct inmem_pages *new;
 192
 193         f2fs_trace_pid(page);
 194
 195         set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
 196         SetPagePrivate(page);
 197
 198         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 199
 200         /* add atomic page indices to the list */
 201         new->page = page;
 202         INIT_LIST_HEAD(&new->list);
 203
 204         /* increase reference count with clean state */
 205         mutex_lock(&fi->inmem_lock);
 206         get_page(page);
 207         list_add_tail(&new->list, &fi->inmem_pages);
 208         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 209         if (list_empty(&fi->inmem_ilist))
 210                 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
 211         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 212         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 213         mutex_unlock(&fi->inmem_lock);
 214
 215         trace_f2fs_register_inmem_page(page, INMEM);
 216 }
 217
 218 static int __revoke_inmem_pages(struct inode *inode,
 219                                 struct list_head *head, bool drop, bool recover)
 220 {
 221         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 222         struct inmem_pages *cur, *tmp;
 223         int err = 0;
 224
 225         list_for_each_entry_safe(cur, tmp, head, list) {
 226                 struct page *page = cur->page;
 227
 228                 if (drop)
 229                         trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 230
 231                 lock_page(page);
 232
 233                 if (recover) {
 234                         struct dnode_of_data dn;
 235                         struct node_info ni;
 236
 237                         trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 238 retry:
 239                         set_new_dnode(&dn, inode, NULL, NULL, 0);
 240                         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
 241                         if (err) {
 242                                 if (err == -ENOMEM) {
 243                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 244                                         cond_resched();
 245                                         goto retry;
 246                                 }
 247                                 err = -EAGAIN;
 248                                 goto next;
 249                         }
 250                         get_node_info(sbi, dn.nid, &ni);
 251                         if (cur->old_addr == NEW_ADDR) {
 252                                 invalidate_blocks(sbi, dn.data_blkaddr);
 253                                 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
 254                         } else
 255                                 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 256                                         cur->old_addr, ni.version, true, true);
 257                         f2fs_put_dnode(&dn);
 258                 }
 259 next:
 260                 /* we don't need to invalidate this in the sccessful status */
 261                 if (drop || recover)
 262                         ClearPageUptodate(page);
 263                 set_page_private(page, 0);
 264                 ClearPagePrivate(page);
 265                 f2fs_put_page(page, 1);
 266
 267                 list_del(&cur->list);
 268                 kmem_cache_free(inmem_entry_slab, cur);
 269                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 270         }
 271         return err;
 272 }
 273
 274 void drop_inmem_pages_all(struct f2fs_sb_info *sbi)
 275 {
 276         struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
 277         struct inode *inode;
 278         struct f2fs_inode_info *fi;
 279 next:
 280         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 281         if (list_empty(head)) {
 282                 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 283                 return;
 284         }
 285         fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
 286         inode = igrab(&fi->vfs_inode);
 287         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 288
 289         if (inode) {
 290                 drop_inmem_pages(inode);
 291                 iput(inode);
 292         }
 293         congestion_wait(BLK_RW_ASYNC, HZ/50);
 294         cond_resched();
 295         goto next;
 296 }
 297
 298 void drop_inmem_pages(struct inode *inode)
 299 {
 300         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 301         struct f2fs_inode_info *fi = F2FS_I(inode);
 302
 303         mutex_lock(&fi->inmem_lock);
 304         __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 305         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 306         if (!list_empty(&fi->inmem_ilist))
 307                 list_del_init(&fi->inmem_ilist);
 308         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 309         mutex_unlock(&fi->inmem_lock);
 310
 311         clear_inode_flag(inode, FI_ATOMIC_FILE);
 312         clear_inode_flag(inode, FI_HOT_DATA);
 313         stat_dec_atomic_write(inode);
 314 }
 315
 316 void drop_inmem_page(struct inode *inode, struct page *page)
 317 {
 318         struct f2fs_inode_info *fi = F2FS_I(inode);
 319         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 320         struct list_head *head = &fi->inmem_pages;
 321         struct inmem_pages *cur = NULL;
 322
 323         f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
 324
 325         mutex_lock(&fi->inmem_lock);
 326         list_for_each_entry(cur, head, list) {
 327                 if (cur->page == page)
 328                         break;
 329         }
 330
 331         f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
 332         list_del(&cur->list);
 333         mutex_unlock(&fi->inmem_lock);
 334
 335         dec_page_count(sbi, F2FS_INMEM_PAGES);
 336         kmem_cache_free(inmem_entry_slab, cur);
 337
 338         ClearPageUptodate(page);
 339         set_page_private(page, 0);
 340         ClearPagePrivate(page);
 341         f2fs_put_page(page, 0);
 342
 343         trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 344 }
 345
 346 static int __commit_inmem_pages(struct inode *inode)
 347 {
 348         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 349         struct f2fs_inode_info *fi = F2FS_I(inode);
 350         struct inmem_pages *cur, *tmp;
 351         struct f2fs_io_info fio = {
 352                 .sbi = sbi,
 353                 .ino = inode->i_ino,
 354                 .type = DATA,
 355                 .op = REQ_OP_WRITE,
 356                 .op_flags = REQ_SYNC | REQ_PRIO,
 357                 .io_type = FS_DATA_IO,
 358         };
 359         struct list_head revoke_list;
 360         pgoff_t last_idx = ULONG_MAX;
 361         int err = 0;
 362
 363         INIT_LIST_HEAD(&revoke_list);
 364
 365         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 366                 struct page *page = cur->page;
 367
 368                 lock_page(page);
 369                 if (page->mapping == inode->i_mapping) {
 370                         trace_f2fs_commit_inmem_page(page, INMEM);
 371
 372                         set_page_dirty(page);
 373                         f2fs_wait_on_page_writeback(page, DATA, true);
 374                         if (clear_page_dirty_for_io(page)) {
 375                                 inode_dec_dirty_pages(inode);
 376                                 remove_dirty_inode(inode);
 377                         }
 378 retry:
 379                         fio.page = page;
 380                         fio.old_blkaddr = NULL_ADDR;
 381                         fio.encrypted_page = NULL;
 382                         fio.need_lock = LOCK_DONE;
 383                         err = do_write_data_page(&fio);
 384                         if (err) {
 385                                 if (err == -ENOMEM) {
 386                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 387                                         cond_resched();
 388                                         goto retry;
 389                                 }
 390                                 unlock_page(page);
 391                                 break;
 392                         }
 393                         /* record old blkaddr for revoking */
 394                         cur->old_addr = fio.old_blkaddr;
 395                         last_idx = page->index;
 396                 }
 397                 unlock_page(page);
 398                 list_move_tail(&cur->list, &revoke_list);
 399         }
 400
 401         if (last_idx != ULONG_MAX)
 402                 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
 403
 404         if (err) {
 405                 /*
 406                  * try to revoke all committed pages, but still we could fail
 407                  * due to no memory or other reason, if that happened, EAGAIN
 408                  * will be returned, which means in such case, transaction is
 409                  * already not integrity, caller should use journal to do the
 410                  * recovery or rewrite & commit last transaction. For other
 411                  * error number, revoking was done by filesystem itself.
 412                  */
 413                 err = __revoke_inmem_pages(inode, &revoke_list, false, true);
 414
 415                 /* drop all uncommitted pages */
 416                 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 417         } else {
 418                 __revoke_inmem_pages(inode, &revoke_list, false, false);
 419         }
 420
 421         return err;
 422 }
 423
 424 int commit_inmem_pages(struct inode *inode)
 425 {
 426         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 427         struct f2fs_inode_info *fi = F2FS_I(inode);
 428         int err;
 429
 430         f2fs_balance_fs(sbi, true);
 431         f2fs_lock_op(sbi);
 432
 433         set_inode_flag(inode, FI_ATOMIC_COMMIT);
 434
 435         mutex_lock(&fi->inmem_lock);
 436         err = __commit_inmem_pages(inode);
 437
 438         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 439         if (!list_empty(&fi->inmem_ilist))
 440                 list_del_init(&fi->inmem_ilist);
 441         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 442         mutex_unlock(&fi->inmem_lock);
 443
 444         clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 445
 446         f2fs_unlock_op(sbi);
 447         return err;
 448 }
 449
 450 /*
 451  * This function balances dirty node and dentry pages.
 452  * In addition, it controls garbage collection.
 453  */
 454 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 455 {
 456 #ifdef CONFIG_F2FS_FAULT_INJECTION
 457         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 458                 f2fs_show_injection_info(FAULT_CHECKPOINT);
 459                 f2fs_stop_checkpoint(sbi, false);
 460         }
 461 #endif
 462
 463         /* balance_fs_bg is able to be pending */
 464         if (need && excess_cached_nats(sbi))
 465                 f2fs_balance_fs_bg(sbi);
 466
 467         /*
 468          * We should do GC or end up with checkpoint, if there are so many dirty
 469          * dir/node pages without enough free segments.
 470          */
 471         if (has_not_enough_free_secs(sbi, 0, 0)) {
 472                 mutex_lock(&sbi->gc_mutex);
 473                 f2fs_gc(sbi, false, false, NULL_SEGNO);
 474         }
 475 }
 476
 477 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 478 {
 479         /* try to shrink extent cache when there is no enough memory */
 480         if (!available_free_memory(sbi, EXTENT_CACHE))
 481                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 482
 483         /* check the # of cached NAT entries */
 484         if (!available_free_memory(sbi, NAT_ENTRIES))
 485                 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 486
 487         if (!available_free_memory(sbi, FREE_NIDS))
 488                 try_to_free_nids(sbi, MAX_FREE_NIDS);
 489         else
 490                 build_free_nids(sbi, false, false);
 491
 492         if (!is_idle(sbi) && !excess_dirty_nats(sbi))
 493                 return;
 494
 495         /* checkpoint is the only way to shrink partial cached entries */
 496         if (!available_free_memory(sbi, NAT_ENTRIES) ||
 497                         !available_free_memory(sbi, INO_ENTRIES) ||
 498                         excess_prefree_segs(sbi) ||
 499                         excess_dirty_nats(sbi) ||
 500                         f2fs_time_over(sbi, CP_TIME)) {
 501                 if (test_opt(sbi, DATA_FLUSH)) {
 502                         struct blk_plug plug;
 503
 504                         blk_start_plug(&plug);
 505                         sync_dirty_inodes(sbi, FILE_INODE);
 506                         blk_finish_plug(&plug);
 507                 }
 508                 f2fs_sync_fs(sbi->sb, true);
 509                 stat_inc_bg_cp_count(sbi->stat_info);
 510         }
 511 }
 512
 513 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 514                                 struct block_device *bdev)
 515 {
 516         struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
 517         int ret;
 518
 519         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
 520         bio_set_dev(bio, bdev);
 521         ret = submit_bio_wait(bio);
 522         bio_put(bio);
 523
 524         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 525                                 test_opt(sbi, FLUSH_MERGE), ret);
 526         return ret;
 527 }
 528
 529 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 530 {
 531         int ret = 0;
 532         int i;
 533
 534         if (!sbi->s_ndevs)
 535                 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 536
 537         for (i = 0; i < sbi->s_ndevs; i++) {
 538                 if (!is_dirty_device(sbi, ino, i, FLUSH_INO))
 539                         continue;
 540                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 541                 if (ret)
 542                         break;
 543         }
 544         return ret;
 545 }
 546
 547 static int issue_flush_thread(void *data)
 548 {
 549         struct f2fs_sb_info *sbi = data;
 550         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 551         wait_queue_head_t *q = &fcc->flush_wait_queue;
 552 repeat:
 553         if (kthread_should_stop())
 554                 return 0;
 555
 556         sb_start_intwrite(sbi->sb);
 557
 558         if (!llist_empty(&fcc->issue_list)) {
 559                 struct flush_cmd *cmd, *next;
 560                 int ret;
 561
 562                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 563                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 564
 565                 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 566
 567                 ret = submit_flush_wait(sbi, cmd->ino);
 568                 atomic_inc(&fcc->issued_flush);
 569
 570                 llist_for_each_entry_safe(cmd, next,
 571                                           fcc->dispatch_list, llnode) {
 572                         cmd->ret = ret;
 573                         complete(&cmd->wait);
 574                 }
 575                 fcc->dispatch_list = NULL;
 576         }
 577
 578         sb_end_intwrite(sbi->sb);
 579
 580         wait_event_interruptible(*q,
 581                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 582         goto repeat;
 583 }
 584
 585 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 586 {
 587         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 588         struct flush_cmd cmd;
 589         int ret;
 590
 591         if (test_opt(sbi, NOBARRIER))
 592                 return 0;
 593
 594         if (!test_opt(sbi, FLUSH_MERGE)) {
 595                 ret = submit_flush_wait(sbi, ino);
 596                 atomic_inc(&fcc->issued_flush);
 597                 return ret;
 598         }
 599
 600         if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
 601                 ret = submit_flush_wait(sbi, ino);
 602                 atomic_dec(&fcc->issing_flush);
 603
 604                 atomic_inc(&fcc->issued_flush);
 605                 return ret;
 606         }
 607
 608         cmd.ino = ino;
 609         init_completion(&cmd.wait);
 610
 611         llist_add(&cmd.llnode, &fcc->issue_list);
 612
 613         /* update issue_list before we wake up issue_flush thread */
 614         smp_mb();
 615
 616         if (waitqueue_active(&fcc->flush_wait_queue))
 617                 wake_up(&fcc->flush_wait_queue);
 618
 619         if (fcc->f2fs_issue_flush) {
 620                 wait_for_completion(&cmd.wait);
 621                 atomic_dec(&fcc->issing_flush);
 622         } else {
 623                 struct llist_node *list;
 624
 625                 list = llist_del_all(&fcc->issue_list);
 626                 if (!list) {
 627                         wait_for_completion(&cmd.wait);
 628                         atomic_dec(&fcc->issing_flush);
 629                 } else {
 630                         struct flush_cmd *tmp, *next;
 631
 632                         ret = submit_flush_wait(sbi, ino);
 633
 634                         llist_for_each_entry_safe(tmp, next, list, llnode) {
 635                                 if (tmp == &cmd) {
 636                                         cmd.ret = ret;
 637                                         atomic_dec(&fcc->issing_flush);
 638                                         continue;
 639                                 }
 640                                 tmp->ret = ret;
 641                                 complete(&tmp->wait);
 642                         }
 643                 }
 644         }
 645
 646         return cmd.ret;
 647 }
 648
 649 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
 650 {
 651         dev_t dev = sbi->sb->s_bdev->bd_dev;
 652         struct flush_cmd_control *fcc;
 653         int err = 0;
 654
 655         if (SM_I(sbi)->fcc_info) {
 656                 fcc = SM_I(sbi)->fcc_info;
 657                 if (fcc->f2fs_issue_flush)
 658                         return err;
 659                 goto init_thread;
 660         }
 661
 662         fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
 663         if (!fcc)
 664                 return -ENOMEM;
 665         atomic_set(&fcc->issued_flush, 0);
 666         atomic_set(&fcc->issing_flush, 0);
 667         init_waitqueue_head(&fcc->flush_wait_queue);
 668         init_llist_head(&fcc->issue_list);
 669         SM_I(sbi)->fcc_info = fcc;
 670         if (!test_opt(sbi, FLUSH_MERGE))
 671                 return err;
 672
 673 init_thread:
 674         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 675                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 676         if (IS_ERR(fcc->f2fs_issue_flush)) {
 677                 err = PTR_ERR(fcc->f2fs_issue_flush);
 678                 kfree(fcc);
 679                 SM_I(sbi)->fcc_info = NULL;
 680                 return err;
 681         }
 682
 683         return err;
 684 }
 685
 686 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 687 {
 688         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 689
 690         if (fcc && fcc->f2fs_issue_flush) {
 691                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 692
 693                 fcc->f2fs_issue_flush = NULL;
 694                 kthread_stop(flush_thread);
 695         }
 696         if (free) {
 697                 kfree(fcc);
 698                 SM_I(sbi)->fcc_info = NULL;
 699         }
 700 }
 701
 702 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
 703 {
 704         int ret = 0, i;
 705
 706         if (!sbi->s_ndevs)
 707                 return 0;
 708
 709         for (i = 1; i < sbi->s_ndevs; i++) {
 710                 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
 711                         continue;
 712                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 713                 if (ret)
 714                         break;
 715
 716                 spin_lock(&sbi->dev_lock);
 717                 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
 718                 spin_unlock(&sbi->dev_lock);
 719         }
 720
 721         return ret;
 722 }
 723
 724 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 725                 enum dirty_type dirty_type)
 726 {
 727         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 728
 729         /* need not be added */
 730         if (IS_CURSEG(sbi, segno))
 731                 return;
 732
 733         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 734                 dirty_i->nr_dirty[dirty_type]++;
 735
 736         if (dirty_type == DIRTY) {
 737                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 738                 enum dirty_type t = sentry->type;
 739
 740                 if (unlikely(t >= DIRTY)) {
 741                         f2fs_bug_on(sbi, 1);
 742                         return;
 743                 }
 744                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 745                         dirty_i->nr_dirty[t]++;
 746         }
 747 }
 748
 749 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 750                 enum dirty_type dirty_type)
 751 {
 752         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 753
 754         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 755                 dirty_i->nr_dirty[dirty_type]--;
 756
 757         if (dirty_type == DIRTY) {
 758                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 759                 enum dirty_type t = sentry->type;
 760
 761                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 762                         dirty_i->nr_dirty[t]--;
 763
 764                 if (get_valid_blocks(sbi, segno, true) == 0)
 765                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 766                                                 dirty_i->victim_secmap);
 767         }
 768 }
 769
 770 /*
 771  * Should not occur error such as -ENOMEM.
 772  * Adding dirty entry into seglist is not critical operation.
 773  * If a given segment is one of current working segments, it won't be added.
 774  */
 775 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 776 {
 777         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 778         unsigned short valid_blocks;
 779
 780         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 781                 return;
 782
 783         mutex_lock(&dirty_i->seglist_lock);
 784
 785         valid_blocks = get_valid_blocks(sbi, segno, false);
 786
 787         if (valid_blocks == 0) {
 788                 __locate_dirty_segment(sbi, segno, PRE);
 789                 __remove_dirty_segment(sbi, segno, DIRTY);
 790         } else if (valid_blocks < sbi->blocks_per_seg) {
 791                 __locate_dirty_segment(sbi, segno, DIRTY);
 792         } else {
 793                 /* Recovery routine with SSR needs this */
 794                 __remove_dirty_segment(sbi, segno, DIRTY);
 795         }
 796
 797         mutex_unlock(&dirty_i->seglist_lock);
 798 }
 799
 800 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 801                 struct block_device *bdev, block_t lstart,
 802                 block_t start, block_t len)
 803 {
 804         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 805         struct list_head *pend_list;
 806         struct discard_cmd *dc;
 807
 808         f2fs_bug_on(sbi, !len);
 809
 810         pend_list = &dcc->pend_list[plist_idx(len)];
 811
 812         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 813         INIT_LIST_HEAD(&dc->list);
 814         dc->bdev = bdev;
 815         dc->lstart = lstart;
 816         dc->start = start;
 817         dc->len = len;
 818         dc->ref = 0;
 819         dc->state = D_PREP;
 820         dc->error = 0;
 821         init_completion(&dc->wait);
 822         list_add_tail(&dc->list, pend_list);
 823         atomic_inc(&dcc->discard_cmd_cnt);
 824         dcc->undiscard_blks += len;
 825
 826         return dc;
 827 }
 828
 829 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 830                                 struct block_device *bdev, block_t lstart,
 831                                 block_t start, block_t len,
 832                                 struct rb_node *parent, struct rb_node **p)
 833 {
 834         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 835         struct discard_cmd *dc;
 836
 837         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
 838
 839         rb_link_node(&dc->rb_node, parent, p);
 840         rb_insert_color(&dc->rb_node, &dcc->root);
 841
 842         return dc;
 843 }
 844
 845 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
 846                                                         struct discard_cmd *dc)
 847 {
 848         if (dc->state == D_DONE)
 849                 atomic_dec(&dcc->issing_discard);
 850
 851         list_del(&dc->list);
 852         rb_erase(&dc->rb_node, &dcc->root);
 853         dcc->undiscard_blks -= dc->len;
 854
 855         kmem_cache_free(discard_cmd_slab, dc);
 856
 857         atomic_dec(&dcc->discard_cmd_cnt);
 858 }
 859
 860 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
 861                                                         struct discard_cmd *dc)
 862 {
 863         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 864
 865         trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
 866
 867         f2fs_bug_on(sbi, dc->ref);
 868
 869         if (dc->error == -EOPNOTSUPP)
 870                 dc->error = 0;
 871
 872         if (dc->error)
 873                 f2fs_msg(sbi->sb, KERN_INFO,
 874                         "Issue discard(%u, %u, %u) failed, ret: %d",
 875                         dc->lstart, dc->start, dc->len, dc->error);
 876         __detach_discard_cmd(dcc, dc);
 877 }
 878
 879 static void f2fs_submit_discard_endio(struct bio *bio)
 880 {
 881         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
 882
 883         dc->error = blk_status_to_errno(bio->bi_status);
 884         dc->state = D_DONE;
 885         complete_all(&dc->wait);
 886         bio_put(bio);
 887 }
 888
 889 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
 890                                 block_t start, block_t end)
 891 {
 892 #ifdef CONFIG_F2FS_CHECK_FS
 893         struct seg_entry *sentry;
 894         unsigned int segno;
 895         block_t blk = start;
 896         unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
 897         unsigned long *map;
 898
 899         while (blk < end) {
 900                 segno = GET_SEGNO(sbi, blk);
 901                 sentry = get_seg_entry(sbi, segno);
 902                 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
 903
 904                 if (end < START_BLOCK(sbi, segno + 1))
 905                         size = GET_BLKOFF_FROM_SEG0(sbi, end);
 906                 else
 907                         size = max_blocks;
 908                 map = (unsigned long *)(sentry->cur_valid_map);
 909                 offset = __find_rev_next_bit(map, size, offset);
 910                 f2fs_bug_on(sbi, offset != size);
 911                 blk = START_BLOCK(sbi, segno + 1);
 912         }
 913 #endif
 914 }
 915
 916 static void __init_discard_policy(struct f2fs_sb_info *sbi,
 917                                 struct discard_policy *dpolicy,
 918                                 int discard_type, unsigned int granularity)
 919 {
 920         /* common policy */
 921         dpolicy->type = discard_type;
 922         dpolicy->sync = true;
 923         dpolicy->granularity = granularity;
 924
 925         dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
 926         dpolicy->io_aware_gran = MAX_PLIST_NUM;
 927
 928         if (discard_type == DPOLICY_BG) {
 929                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
 930                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
 931                 dpolicy->io_aware = true;
 932                 dpolicy->sync = false;
 933                 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
 934                         dpolicy->granularity = 1;
 935                         dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
 936                 }
 937         } else if (discard_type == DPOLICY_FORCE) {
 938                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
 939                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
 940                 dpolicy->io_aware = false;
 941         } else if (discard_type == DPOLICY_FSTRIM) {
 942                 dpolicy->io_aware = false;
 943         } else if (discard_type == DPOLICY_UMOUNT) {
 944                 dpolicy->max_requests = UINT_MAX;
 945                 dpolicy->io_aware = false;
 946         }
 947 }
 948
 949
 950 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
 951 static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
 952                                                 struct discard_policy *dpolicy,
 953                                                 struct discard_cmd *dc)
 954 {
 955         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 956         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
 957                                         &(dcc->fstrim_list) : &(dcc->wait_list);
 958         struct bio *bio = NULL;
 959         int flag = dpolicy->sync ? REQ_SYNC : 0;
 960
 961         if (dc->state != D_PREP)
 962                 return;
 963
 964         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
 965                 return;
 966
 967         trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
 968
 969         dc->error = __blkdev_issue_discard(dc->bdev,
 970                                 SECTOR_FROM_BLOCK(dc->start),
 971                                 SECTOR_FROM_BLOCK(dc->len),
 972                                 GFP_NOFS, 0, &bio);
 973         if (!dc->error) {
 974                 /* should keep before submission to avoid D_DONE right away */
 975                 dc->state = D_SUBMIT;
 976                 atomic_inc(&dcc->issued_discard);
 977                 atomic_inc(&dcc->issing_discard);
 978                 if (bio) {
 979                         bio->bi_private = dc;
 980                         bio->bi_end_io = f2fs_submit_discard_endio;
 981                         bio->bi_opf |= flag;
 982                         submit_bio(bio);
 983                         list_move_tail(&dc->list, wait_list);
 984                         __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
 985
 986                         f2fs_update_iostat(sbi, FS_DISCARD, 1);
 987                 }
 988         } else {
 989                 __remove_discard_cmd(sbi, dc);
 990         }
 991 }
 992
 993 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
 994                                 struct block_device *bdev, block_t lstart,
 995                                 block_t start, block_t len,
 996                                 struct rb_node **insert_p,
 997                                 struct rb_node *insert_parent)
 998 {
 999         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1000         struct rb_node **p;
1001         struct rb_node *parent = NULL;
1002         struct discard_cmd *dc = NULL;
1003
1004         if (insert_p && insert_parent) {
1005                 parent = insert_parent;
1006                 p = insert_p;
1007                 goto do_insert;
1008         }
1009
1010         p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
1011 do_insert:
1012         dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
1013         if (!dc)
1014                 return NULL;
1015
1016         return dc;
1017 }
1018
1019 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1020                                                 struct discard_cmd *dc)
1021 {
1022         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1023 }
1024
1025 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1026                                 struct discard_cmd *dc, block_t blkaddr)
1027 {
1028         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1029         struct discard_info di = dc->di;
1030         bool modified = false;
1031
1032         if (dc->state == D_DONE || dc->len == 1) {
1033                 __remove_discard_cmd(sbi, dc);
1034                 return;
1035         }
1036
1037         dcc->undiscard_blks -= di.len;
1038
1039         if (blkaddr > di.lstart) {
1040                 dc->len = blkaddr - dc->lstart;
1041                 dcc->undiscard_blks += dc->len;
1042                 __relocate_discard_cmd(dcc, dc);
1043                 modified = true;
1044         }
1045
1046         if (blkaddr < di.lstart + di.len - 1) {
1047                 if (modified) {
1048                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1049                                         di.start + blkaddr + 1 - di.lstart,
1050                                         di.lstart + di.len - 1 - blkaddr,
1051                                         NULL, NULL);
1052                 } else {
1053                         dc->lstart++;
1054                         dc->len--;
1055                         dc->start++;
1056                         dcc->undiscard_blks += dc->len;
1057                         __relocate_discard_cmd(dcc, dc);
1058                 }
1059         }
1060 }
1061
1062 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1063                                 struct block_device *bdev, block_t lstart,
1064                                 block_t start, block_t len)
1065 {
1066         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1067         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1068         struct discard_cmd *dc;
1069         struct discard_info di = {0};
1070         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1071         block_t end = lstart + len;
1072
1073         mutex_lock(&dcc->cmd_lock);
1074
1075         dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1076                                         NULL, lstart,
1077                                         (struct rb_entry **)&prev_dc,
1078                                         (struct rb_entry **)&next_dc,
1079                                         &insert_p, &insert_parent, true);
1080         if (dc)
1081                 prev_dc = dc;
1082
1083         if (!prev_dc) {
1084                 di.lstart = lstart;
1085                 di.len = next_dc ? next_dc->lstart - lstart : len;
1086                 di.len = min(di.len, len);
1087                 di.start = start;
1088         }
1089
1090         while (1) {
1091                 struct rb_node *node;
1092                 bool merged = false;
1093                 struct discard_cmd *tdc = NULL;
1094
1095                 if (prev_dc) {
1096                         di.lstart = prev_dc->lstart + prev_dc->len;
1097                         if (di.lstart < lstart)
1098                                 di.lstart = lstart;
1099                         if (di.lstart >= end)
1100                                 break;
1101
1102                         if (!next_dc || next_dc->lstart > end)
1103                                 di.len = end - di.lstart;
1104                         else
1105                                 di.len = next_dc->lstart - di.lstart;
1106                         di.start = start + di.lstart - lstart;
1107                 }
1108
1109                 if (!di.len)
1110                         goto next;
1111
1112                 if (prev_dc && prev_dc->state == D_PREP &&
1113                         prev_dc->bdev == bdev &&
1114                         __is_discard_back_mergeable(&di, &prev_dc->di)) {
1115                         prev_dc->di.len += di.len;
1116                         dcc->undiscard_blks += di.len;
1117                         __relocate_discard_cmd(dcc, prev_dc);
1118                         di = prev_dc->di;
1119                         tdc = prev_dc;
1120                         merged = true;
1121                 }
1122
1123                 if (next_dc && next_dc->state == D_PREP &&
1124                         next_dc->bdev == bdev &&
1125                         __is_discard_front_mergeable(&di, &next_dc->di)) {
1126                         next_dc->di.lstart = di.lstart;
1127                         next_dc->di.len += di.len;
1128                         next_dc->di.start = di.start;
1129                         dcc->undiscard_blks += di.len;
1130                         __relocate_discard_cmd(dcc, next_dc);
1131                         if (tdc)
1132                                 __remove_discard_cmd(sbi, tdc);
1133                         merged = true;
1134                 }
1135
1136                 if (!merged) {
1137                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1138                                                         di.len, NULL, NULL);
1139                 }
1140  next:
1141                 prev_dc = next_dc;
1142                 if (!prev_dc)
1143                         break;
1144
1145                 node = rb_next(&prev_dc->rb_node);
1146                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1147         }
1148
1149         mutex_unlock(&dcc->cmd_lock);
1150 }
1151
1152 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1153                 struct block_device *bdev, block_t blkstart, block_t blklen)
1154 {
1155         block_t lblkstart = blkstart;
1156
1157         trace_f2fs_queue_discard(bdev, blkstart, blklen);
1158
1159         if (sbi->s_ndevs) {
1160                 int devi = f2fs_target_device_index(sbi, blkstart);
1161
1162                 blkstart -= FDEV(devi).start_blk;
1163         }
1164         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1165         return 0;
1166 }
1167
1168 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1169                                         struct discard_policy *dpolicy)
1170 {
1171         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1172         struct list_head *pend_list;
1173         struct discard_cmd *dc, *tmp;
1174         struct blk_plug plug;
1175         int i, iter = 0, issued = 0;
1176         bool io_interrupted = false;
1177
1178         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1179                 if (i + 1 < dpolicy->granularity)
1180                         break;
1181                 pend_list = &dcc->pend_list[i];
1182
1183                 mutex_lock(&dcc->cmd_lock);
1184                 if (list_empty(pend_list))
1185                         goto next;
1186                 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1187                 blk_start_plug(&plug);
1188                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1189                         f2fs_bug_on(sbi, dc->state != D_PREP);
1190
1191                         if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1192                                                                 !is_idle(sbi)) {
1193                                 io_interrupted = true;
1194                                 goto skip;
1195                         }
1196
1197                         __submit_discard_cmd(sbi, dpolicy, dc);
1198                         issued++;
1199 skip:
1200                         if (++iter >= dpolicy->max_requests)
1201                                 break;
1202                 }
1203                 blk_finish_plug(&plug);
1204 next:
1205                 mutex_unlock(&dcc->cmd_lock);
1206
1207                 if (iter >= dpolicy->max_requests)
1208                         break;
1209         }
1210
1211         if (!issued && io_interrupted)
1212                 issued = -1;
1213
1214         return issued;
1215 }
1216
1217 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1218 {
1219         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1220         struct list_head *pend_list;
1221         struct discard_cmd *dc, *tmp;
1222         int i;
1223         bool dropped = false;
1224
1225         mutex_lock(&dcc->cmd_lock);
1226         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1227                 pend_list = &dcc->pend_list[i];
1228                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1229                         f2fs_bug_on(sbi, dc->state != D_PREP);
1230                         __remove_discard_cmd(sbi, dc);
1231                         dropped = true;
1232                 }
1233         }
1234         mutex_unlock(&dcc->cmd_lock);
1235
1236         return dropped;
1237 }
1238
1239 void drop_discard_cmd(struct f2fs_sb_info *sbi)
1240 {
1241         __drop_discard_cmd(sbi);
1242 }
1243
1244 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1245                                                         struct discard_cmd *dc)
1246 {
1247         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1248         unsigned int len = 0;
1249
1250         wait_for_completion_io(&dc->wait);
1251         mutex_lock(&dcc->cmd_lock);
1252         f2fs_bug_on(sbi, dc->state != D_DONE);
1253         dc->ref--;
1254         if (!dc->ref) {
1255                 if (!dc->error)
1256                         len = dc->len;
1257                 __remove_discard_cmd(sbi, dc);
1258         }
1259         mutex_unlock(&dcc->cmd_lock);
1260
1261         return len;
1262 }
1263
1264 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1265                                                 struct discard_policy *dpolicy,
1266                                                 block_t start, block_t end)
1267 {
1268         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1269         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1270                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1271         struct discard_cmd *dc, *tmp;
1272         bool need_wait;
1273         unsigned int trimmed = 0;
1274
1275 next:
1276         need_wait = false;
1277
1278         mutex_lock(&dcc->cmd_lock);
1279         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1280                 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1281                         continue;
1282                 if (dc->len < dpolicy->granularity)
1283                         continue;
1284                 if (dc->state == D_DONE && !dc->ref) {
1285                         wait_for_completion_io(&dc->wait);
1286                         if (!dc->error)
1287                                 trimmed += dc->len;
1288                         __remove_discard_cmd(sbi, dc);
1289                 } else {
1290                         dc->ref++;
1291                         need_wait = true;
1292                         break;
1293                 }
1294         }
1295         mutex_unlock(&dcc->cmd_lock);
1296
1297         if (need_wait) {
1298                 trimmed += __wait_one_discard_bio(sbi, dc);
1299                 goto next;
1300         }
1301
1302         return trimmed;
1303 }
1304
1305 static void __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1306                                                 struct discard_policy *dpolicy)
1307 {
1308         struct discard_policy dp;
1309
1310         if (dpolicy) {
1311                 __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1312                 return;
1313         }
1314
1315         /* wait all */
1316         __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1317         __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1318         __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1319         __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1320 }
1321
1322 /* This should be covered by global mutex, &sit_i->sentry_lock */
1323 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1324 {
1325         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1326         struct discard_cmd *dc;
1327         bool need_wait = false;
1328
1329         mutex_lock(&dcc->cmd_lock);
1330         dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1331         if (dc) {
1332                 if (dc->state == D_PREP) {
1333                         __punch_discard_cmd(sbi, dc, blkaddr);
1334                 } else {
1335                         dc->ref++;
1336                         need_wait = true;
1337                 }
1338         }
1339         mutex_unlock(&dcc->cmd_lock);
1340
1341         if (need_wait)
1342                 __wait_one_discard_bio(sbi, dc);
1343 }
1344
1345 void stop_discard_thread(struct f2fs_sb_info *sbi)
1346 {
1347         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1348
1349         if (dcc && dcc->f2fs_issue_discard) {
1350                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1351
1352                 dcc->f2fs_issue_discard = NULL;
1353                 kthread_stop(discard_thread);
1354         }
1355 }
1356
1357 /* This comes from f2fs_put_super */
1358 bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1359 {
1360         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1361         struct discard_policy dpolicy;
1362         bool dropped;
1363
1364         __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1365                                         dcc->discard_granularity);
1366         __issue_discard_cmd(sbi, &dpolicy);
1367         dropped = __drop_discard_cmd(sbi);
1368
1369         /* just to make sure there is no pending discard commands */
1370         __wait_all_discard_cmd(sbi, NULL);
1371         return dropped;
1372 }
1373
1374 static int issue_discard_thread(void *data)
1375 {
1376         struct f2fs_sb_info *sbi = data;
1377         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1378         wait_queue_head_t *q = &dcc->discard_wait_queue;
1379         struct discard_policy dpolicy;
1380         unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1381         int issued;
1382
1383         set_freezable();
1384
1385         do {
1386                 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1387                                         dcc->discard_granularity);
1388
1389                 wait_event_interruptible_timeout(*q,
1390                                 kthread_should_stop() || freezing(current) ||
1391                                 dcc->discard_wake,
1392                                 msecs_to_jiffies(wait_ms));
1393                 if (try_to_freeze())
1394                         continue;
1395                 if (f2fs_readonly(sbi->sb))
1396                         continue;
1397                 if (kthread_should_stop())
1398                         return 0;
1399                 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1400                         wait_ms = dpolicy.max_interval;
1401                         continue;
1402                 }
1403
1404                 if (dcc->discard_wake)
1405                         dcc->discard_wake = 0;
1406
1407                 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
1408                         __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1409
1410                 sb_start_intwrite(sbi->sb);
1411
1412                 issued = __issue_discard_cmd(sbi, &dpolicy);
1413                 if (issued) {
1414                         __wait_all_discard_cmd(sbi, &dpolicy);
1415                         wait_ms = dpolicy.min_interval;
1416                 } else {
1417                         wait_ms = dpolicy.max_interval;
1418                 }
1419
1420                 sb_end_intwrite(sbi->sb);
1421
1422         } while (!kthread_should_stop());
1423         return 0;
1424 }
1425
1426 #ifdef CONFIG_BLK_DEV_ZONED
1427 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1428                 struct block_device *bdev, block_t blkstart, block_t blklen)
1429 {
1430         sector_t sector, nr_sects;
1431         block_t lblkstart = blkstart;
1432         int devi = 0;
1433
1434         if (sbi->s_ndevs) {
1435                 devi = f2fs_target_device_index(sbi, blkstart);
1436                 blkstart -= FDEV(devi).start_blk;
1437         }
1438
1439         /*
1440          * We need to know the type of the zone: for conventional zones,
1441          * use regular discard if the drive supports it. For sequential
1442          * zones, reset the zone write pointer.
1443          */
1444         switch (get_blkz_type(sbi, bdev, blkstart)) {
1445
1446         case BLK_ZONE_TYPE_CONVENTIONAL:
1447                 if (!blk_queue_discard(bdev_get_queue(bdev)))
1448                         return 0;
1449                 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1450         case BLK_ZONE_TYPE_SEQWRITE_REQ:
1451         case BLK_ZONE_TYPE_SEQWRITE_PREF:
1452                 sector = SECTOR_FROM_BLOCK(blkstart);
1453                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1454
1455                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1456                                 nr_sects != bdev_zone_sectors(bdev)) {
1457                         f2fs_msg(sbi->sb, KERN_INFO,
1458                                 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1459                                 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1460                                 blkstart, blklen);
1461                         return -EIO;
1462                 }
1463                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1464                 return blkdev_reset_zones(bdev, sector,
1465                                           nr_sects, GFP_NOFS);
1466         default:
1467                 /* Unknown zone type: broken device ? */
1468                 return -EIO;
1469         }
1470 }
1471 #endif
1472
1473 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1474                 struct block_device *bdev, block_t blkstart, block_t blklen)
1475 {
1476 #ifdef CONFIG_BLK_DEV_ZONED
1477         if (f2fs_sb_has_blkzoned(sbi->sb) &&
1478                                 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1479                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1480 #endif
1481         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1482 }
1483
1484 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1485                                 block_t blkstart, block_t blklen)
1486 {
1487         sector_t start = blkstart, len = 0;
1488         struct block_device *bdev;
1489         struct seg_entry *se;
1490         unsigned int offset;
1491         block_t i;
1492         int err = 0;
1493
1494         bdev = f2fs_target_device(sbi, blkstart, NULL);
1495
1496         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1497                 if (i != start) {
1498                         struct block_device *bdev2 =
1499                                 f2fs_target_device(sbi, i, NULL);
1500
1501                         if (bdev2 != bdev) {
1502                                 err = __issue_discard_async(sbi, bdev,
1503                                                 start, len);
1504                                 if (err)
1505                                         return err;
1506                                 bdev = bdev2;
1507                                 start = i;
1508                                 len = 0;
1509                         }
1510                 }
1511
1512                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1513                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1514
1515                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1516                         sbi->discard_blks--;
1517         }
1518
1519         if (len)
1520                 err = __issue_discard_async(sbi, bdev, start, len);
1521         return err;
1522 }
1523
1524 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1525                                                         bool check_only)
1526 {
1527         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1528         int max_blocks = sbi->blocks_per_seg;
1529         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1530         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1531         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1532         unsigned long *discard_map = (unsigned long *)se->discard_map;
1533         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1534         unsigned int start = 0, end = -1;
1535         bool force = (cpc->reason & CP_DISCARD);
1536         struct discard_entry *de = NULL;
1537         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1538         int i;
1539
1540         if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
1541                 return false;
1542
1543         if (!force) {
1544                 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
1545                         SM_I(sbi)->dcc_info->nr_discards >=
1546                                 SM_I(sbi)->dcc_info->max_discards)
1547                         return false;
1548         }
1549
1550         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1551         for (i = 0; i < entries; i++)
1552                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1553                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1554
1555         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1556                                 SM_I(sbi)->dcc_info->max_discards) {
1557                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1558                 if (start >= max_blocks)
1559                         break;
1560
1561                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1562                 if (force && start && end != max_blocks
1563                                         && (end - start) < cpc->trim_minlen)
1564                         continue;
1565
1566                 if (check_only)
1567                         return true;
1568
1569                 if (!de) {
1570                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1571                                                                 GFP_F2FS_ZERO);
1572                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1573                         list_add_tail(&de->list, head);
1574                 }
1575
1576                 for (i = start; i < end; i++)
1577                         __set_bit_le(i, (void *)de->discard_map);
1578
1579                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1580         }
1581         return false;
1582 }
1583
1584 void release_discard_addrs(struct f2fs_sb_info *sbi)
1585 {
1586         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1587         struct discard_entry *entry, *this;
1588
1589         /* drop caches */
1590         list_for_each_entry_safe(entry, this, head, list) {
1591                 list_del(&entry->list);
1592                 kmem_cache_free(discard_entry_slab, entry);
1593         }
1594 }
1595
1596 /*
1597  * Should call clear_prefree_segments after checkpoint is done.
1598  */
1599 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1600 {
1601         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1602         unsigned int segno;
1603
1604         mutex_lock(&dirty_i->seglist_lock);
1605         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1606                 __set_test_and_free(sbi, segno);
1607         mutex_unlock(&dirty_i->seglist_lock);
1608 }
1609
1610 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1611 {
1612         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1613         struct list_head *head = &dcc->entry_list;
1614         struct discard_entry *entry, *this;
1615         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1616         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1617         unsigned int start = 0, end = -1;
1618         unsigned int secno, start_segno;
1619         bool force = (cpc->reason & CP_DISCARD);
1620
1621         mutex_lock(&dirty_i->seglist_lock);
1622
1623         while (1) {
1624                 int i;
1625                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1626                 if (start >= MAIN_SEGS(sbi))
1627                         break;
1628                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1629                                                                 start + 1);
1630
1631                 for (i = start; i < end; i++)
1632                         clear_bit(i, prefree_map);
1633
1634                 dirty_i->nr_dirty[PRE] -= end - start;
1635
1636                 if (!test_opt(sbi, DISCARD))
1637                         continue;
1638
1639                 if (force && start >= cpc->trim_start &&
1640                                         (end - 1) <= cpc->trim_end)
1641                                 continue;
1642
1643                 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1644                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1645                                 (end - start) << sbi->log_blocks_per_seg);
1646                         continue;
1647                 }
1648 next:
1649                 secno = GET_SEC_FROM_SEG(sbi, start);
1650                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1651                 if (!IS_CURSEC(sbi, secno) &&
1652                         !get_valid_blocks(sbi, start, true))
1653                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1654                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1655
1656                 start = start_segno + sbi->segs_per_sec;
1657                 if (start < end)
1658                         goto next;
1659                 else
1660                         end = start - 1;
1661         }
1662         mutex_unlock(&dirty_i->seglist_lock);
1663
1664         /* send small discards */
1665         list_for_each_entry_safe(entry, this, head, list) {
1666                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1667                 bool is_valid = test_bit_le(0, entry->discard_map);
1668
1669 find_next:
1670                 if (is_valid) {
1671                         next_pos = find_next_zero_bit_le(entry->discard_map,
1672                                         sbi->blocks_per_seg, cur_pos);
1673                         len = next_pos - cur_pos;
1674
1675                         if (f2fs_sb_has_blkzoned(sbi->sb) ||
1676                             (force && len < cpc->trim_minlen))
1677                                 goto skip;
1678
1679                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1680                                                                         len);
1681                         total_len += len;
1682                 } else {
1683                         next_pos = find_next_bit_le(entry->discard_map,
1684                                         sbi->blocks_per_seg, cur_pos);
1685                 }
1686 skip:
1687                 cur_pos = next_pos;
1688                 is_valid = !is_valid;
1689
1690                 if (cur_pos < sbi->blocks_per_seg)
1691                         goto find_next;
1692
1693                 list_del(&entry->list);
1694                 dcc->nr_discards -= total_len;
1695                 kmem_cache_free(discard_entry_slab, entry);
1696         }
1697
1698         wake_up_discard_thread(sbi, false);
1699 }
1700
1701 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1702 {
1703         dev_t dev = sbi->sb->s_bdev->bd_dev;
1704         struct discard_cmd_control *dcc;
1705         int err = 0, i;
1706
1707         if (SM_I(sbi)->dcc_info) {
1708                 dcc = SM_I(sbi)->dcc_info;
1709                 goto init_thread;
1710         }
1711
1712         dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
1713         if (!dcc)
1714                 return -ENOMEM;
1715
1716         dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
1717         INIT_LIST_HEAD(&dcc->entry_list);
1718         for (i = 0; i < MAX_PLIST_NUM; i++)
1719                 INIT_LIST_HEAD(&dcc->pend_list[i]);
1720         INIT_LIST_HEAD(&dcc->wait_list);
1721         INIT_LIST_HEAD(&dcc->fstrim_list);
1722         mutex_init(&dcc->cmd_lock);
1723         atomic_set(&dcc->issued_discard, 0);
1724         atomic_set(&dcc->issing_discard, 0);
1725         atomic_set(&dcc->discard_cmd_cnt, 0);
1726         dcc->nr_discards = 0;
1727         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1728         dcc->undiscard_blks = 0;
1729         dcc->root = RB_ROOT;
1730
1731         init_waitqueue_head(&dcc->discard_wait_queue);
1732         SM_I(sbi)->dcc_info = dcc;
1733 init_thread:
1734         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1735                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1736         if (IS_ERR(dcc->f2fs_issue_discard)) {
1737                 err = PTR_ERR(dcc->f2fs_issue_discard);
1738                 kfree(dcc);
1739                 SM_I(sbi)->dcc_info = NULL;
1740                 return err;
1741         }
1742
1743         return err;
1744 }
1745
1746 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1747 {
1748         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1749
1750         if (!dcc)
1751                 return;
1752
1753         stop_discard_thread(sbi);
1754
1755         kfree(dcc);
1756         SM_I(sbi)->dcc_info = NULL;
1757 }
1758
1759 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1760 {
1761         struct sit_info *sit_i = SIT_I(sbi);
1762
1763         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1764                 sit_i->dirty_sentries++;
1765                 return false;
1766         }
1767
1768         return true;
1769 }
1770
1771 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1772                                         unsigned int segno, int modified)
1773 {
1774         struct seg_entry *se = get_seg_entry(sbi, segno);
1775         se->type = type;
1776         if (modified)
1777                 __mark_sit_entry_dirty(sbi, segno);
1778 }
1779
1780 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1781 {
1782         struct seg_entry *se;
1783         unsigned int segno, offset;
1784         long int new_vblocks;
1785         bool exist;
1786 #ifdef CONFIG_F2FS_CHECK_FS
1787         bool mir_exist;
1788 #endif
1789
1790         segno = GET_SEGNO(sbi, blkaddr);
1791
1792         se = get_seg_entry(sbi, segno);
1793         new_vblocks = se->valid_blocks + del;
1794         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1795
1796         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
1797                                 (new_vblocks > sbi->blocks_per_seg)));
1798
1799         se->valid_blocks = new_vblocks;
1800         se->mtime = get_mtime(sbi);
1801         SIT_I(sbi)->max_mtime = se->mtime;
1802
1803         /* Update valid block bitmap */
1804         if (del > 0) {
1805                 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
1806 #ifdef CONFIG_F2FS_CHECK_FS
1807                 mir_exist = f2fs_test_and_set_bit(offset,
1808                                                 se->cur_valid_map_mir);
1809                 if (unlikely(exist != mir_exist)) {
1810                         f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1811                                 "when setting bitmap, blk:%u, old bit:%d",
1812                                 blkaddr, exist);
1813                         f2fs_bug_on(sbi, 1);
1814                 }
1815 #endif
1816                 if (unlikely(exist)) {
1817                         f2fs_msg(sbi->sb, KERN_ERR,
1818                                 "Bitmap was wrongly set, blk:%u", blkaddr);
1819                         f2fs_bug_on(sbi, 1);
1820                         se->valid_blocks--;
1821                         del = 0;
1822                 }
1823
1824                 if (f2fs_discard_en(sbi) &&
1825                         !f2fs_test_and_set_bit(offset, se->discard_map))
1826                         sbi->discard_blks--;
1827
1828                 /* don't overwrite by SSR to keep node chain */
1829                 if (IS_NODESEG(se->type)) {
1830                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1831                                 se->ckpt_valid_blocks++;
1832                 }
1833         } else {
1834                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
1835 #ifdef CONFIG_F2FS_CHECK_FS
1836                 mir_exist = f2fs_test_and_clear_bit(offset,
1837                                                 se->cur_valid_map_mir);
1838                 if (unlikely(exist != mir_exist)) {
1839                         f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1840                                 "when clearing bitmap, blk:%u, old bit:%d",
1841                                 blkaddr, exist);
1842                         f2fs_bug_on(sbi, 1);
1843                 }
1844 #endif
1845                 if (unlikely(!exist)) {
1846                         f2fs_msg(sbi->sb, KERN_ERR,
1847                                 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1848                         f2fs_bug_on(sbi, 1);
1849                         se->valid_blocks++;
1850                         del = 0;
1851                 }
1852
1853                 if (f2fs_discard_en(sbi) &&
1854                         f2fs_test_and_clear_bit(offset, se->discard_map))
1855                         sbi->discard_blks++;
1856         }
1857         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1858                 se->ckpt_valid_blocks += del;
1859
1860         __mark_sit_entry_dirty(sbi, segno);
1861
1862         /* update total number of valid blocks to be written in ckpt area */
1863         SIT_I(sbi)->written_valid_blocks += del;
1864
1865         if (sbi->segs_per_sec > 1)
1866                 get_sec_entry(sbi, segno)->valid_blocks += del;
1867 }
1868
1869 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1870 {
1871         unsigned int segno = GET_SEGNO(sbi, addr);
1872         struct sit_info *sit_i = SIT_I(sbi);
1873
1874         f2fs_bug_on(sbi, addr == NULL_ADDR);
1875         if (addr == NEW_ADDR)
1876                 return;
1877
1878         /* add it into sit main buffer */
1879         down_write(&sit_i->sentry_lock);
1880
1881         update_sit_entry(sbi, addr, -1);
1882
1883         /* add it into dirty seglist */
1884         locate_dirty_segment(sbi, segno);
1885
1886         up_write(&sit_i->sentry_lock);
1887 }
1888
1889 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1890 {
1891         struct sit_info *sit_i = SIT_I(sbi);
1892         unsigned int segno, offset;
1893         struct seg_entry *se;
1894         bool is_cp = false;
1895
1896         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1897                 return true;
1898
1899         down_read(&sit_i->sentry_lock);
1900
1901         segno = GET_SEGNO(sbi, blkaddr);
1902         se = get_seg_entry(sbi, segno);
1903         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1904
1905         if (f2fs_test_bit(offset, se->ckpt_valid_map))
1906                 is_cp = true;
1907
1908         up_read(&sit_i->sentry_lock);
1909
1910         return is_cp;
1911 }
1912
1913 /*
1914  * This function should be resided under the curseg_mutex lock
1915  */
1916 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1917                                         struct f2fs_summary *sum)
1918 {
1919         struct curseg_info *curseg = CURSEG_I(sbi, type);
1920         void *addr = curseg->sum_blk;
1921         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1922         memcpy(addr, sum, sizeof(struct f2fs_summary));
1923 }
1924
1925 /*
1926  * Calculate the number of current summary pages for writing
1927  */
1928 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1929 {
1930         int valid_sum_count = 0;
1931         int i, sum_in_page;
1932
1933         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1934                 if (sbi->ckpt->alloc_type[i] == SSR)
1935                         valid_sum_count += sbi->blocks_per_seg;
1936                 else {
1937                         if (for_ra)
1938                                 valid_sum_count += le16_to_cpu(
1939                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1940                         else
1941                                 valid_sum_count += curseg_blkoff(sbi, i);
1942                 }
1943         }
1944
1945         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1946                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1947         if (valid_sum_count <= sum_in_page)
1948                 return 1;
1949         else if ((valid_sum_count - sum_in_page) <=
1950                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1951                 return 2;
1952         return 3;
1953 }
1954
1955 /*
1956  * Caller should put this summary page
1957  */
1958 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1959 {
1960         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1961 }
1962
1963 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1964 {
1965         struct page *page = grab_meta_page(sbi, blk_addr);
1966
1967         memcpy(page_address(page), src, PAGE_SIZE);
1968         set_page_dirty(page);
1969         f2fs_put_page(page, 1);
1970 }
1971
1972 static void write_sum_page(struct f2fs_sb_info *sbi,
1973                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
1974 {
1975         update_meta_page(sbi, (void *)sum_blk, blk_addr);
1976 }
1977
1978 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1979                                                 int type, block_t blk_addr)
1980 {
1981         struct curseg_info *curseg = CURSEG_I(sbi, type);
1982         struct page *page = grab_meta_page(sbi, blk_addr);
1983         struct f2fs_summary_block *src = curseg->sum_blk;
1984         struct f2fs_summary_block *dst;
1985
1986         dst = (struct f2fs_summary_block *)page_address(page);
1987         memset(dst, 0, PAGE_SIZE);
1988
1989         mutex_lock(&curseg->curseg_mutex);
1990
1991         down_read(&curseg->journal_rwsem);
1992         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1993         up_read(&curseg->journal_rwsem);
1994
1995         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1996         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1997
1998         mutex_unlock(&curseg->curseg_mutex);
1999
2000         set_page_dirty(page);
2001         f2fs_put_page(page, 1);
2002 }
2003
2004 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2005 {
2006         struct curseg_info *curseg = CURSEG_I(sbi, type);
2007         unsigned int segno = curseg->segno + 1;
2008         struct free_segmap_info *free_i = FREE_I(sbi);
2009
2010         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2011                 return !test_bit(segno, free_i->free_segmap);
2012         return 0;
2013 }
2014
2015 /*
2016  * Find a new segment from the free segments bitmap to right order
2017  * This function should be returned with success, otherwise BUG
2018  */
2019 static void get_new_segment(struct f2fs_sb_info *sbi,
2020                         unsigned int *newseg, bool new_sec, int dir)
2021 {
2022         struct free_segmap_info *free_i = FREE_I(sbi);
2023         unsigned int segno, secno, zoneno;
2024         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2025         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2026         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2027         unsigned int left_start = hint;
2028         bool init = true;
2029         int go_left = 0;
2030         int i;
2031
2032         spin_lock(&free_i->segmap_lock);
2033
2034         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2035                 segno = find_next_zero_bit(free_i->free_segmap,
2036                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2037                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2038                         goto got_it;
2039         }
2040 find_other_zone:
2041         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2042         if (secno >= MAIN_SECS(sbi)) {
2043                 if (dir == ALLOC_RIGHT) {
2044                         secno = find_next_zero_bit(free_i->free_secmap,
2045                                                         MAIN_SECS(sbi), 0);
2046                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2047                 } else {
2048                         go_left = 1;
2049                         left_start = hint - 1;
2050                 }
2051         }
2052         if (go_left == 0)
2053                 goto skip_left;
2054
2055         while (test_bit(left_start, free_i->free_secmap)) {
2056                 if (left_start > 0) {
2057                         left_start--;
2058                         continue;
2059                 }
2060                 left_start = find_next_zero_bit(free_i->free_secmap,
2061                                                         MAIN_SECS(sbi), 0);
2062                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2063                 break;
2064         }
2065         secno = left_start;
2066 skip_left:
2067         segno = GET_SEG_FROM_SEC(sbi, secno);
2068         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2069
2070         /* give up on finding another zone */
2071         if (!init)
2072                 goto got_it;
2073         if (sbi->secs_per_zone == 1)
2074                 goto got_it;
2075         if (zoneno == old_zoneno)
2076                 goto got_it;
2077         if (dir == ALLOC_LEFT) {
2078                 if (!go_left && zoneno + 1 >= total_zones)
2079                         goto got_it;
2080                 if (go_left && zoneno == 0)
2081                         goto got_it;
2082         }
2083         for (i = 0; i < NR_CURSEG_TYPE; i++)
2084                 if (CURSEG_I(sbi, i)->zone == zoneno)
2085                         break;
2086
2087         if (i < NR_CURSEG_TYPE) {
2088                 /* zone is in user, try another */
2089                 if (go_left)
2090                         hint = zoneno * sbi->secs_per_zone - 1;
2091                 else if (zoneno + 1 >= total_zones)
2092                         hint = 0;
2093                 else
2094                         hint = (zoneno + 1) * sbi->secs_per_zone;
2095                 init = false;
2096                 goto find_other_zone;
2097         }
2098 got_it:
2099         /* set it as dirty segment in free segmap */
2100         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2101         __set_inuse(sbi, segno);
2102         *newseg = segno;
2103         spin_unlock(&free_i->segmap_lock);
2104 }
2105
2106 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2107 {
2108         struct curseg_info *curseg = CURSEG_I(sbi, type);
2109         struct summary_footer *sum_footer;
2110
2111         curseg->segno = curseg->next_segno;
2112         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2113         curseg->next_blkoff = 0;
2114         curseg->next_segno = NULL_SEGNO;
2115
2116         sum_footer = &(curseg->sum_blk->footer);
2117         memset(sum_footer, 0, sizeof(struct summary_footer));
2118         if (IS_DATASEG(type))
2119                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2120         if (IS_NODESEG(type))
2121                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2122         __set_sit_entry_type(sbi, type, curseg->segno, modified);
2123 }
2124
2125 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2126 {
2127         /* if segs_per_sec is large than 1, we need to keep original policy. */
2128         if (sbi->segs_per_sec != 1)
2129                 return CURSEG_I(sbi, type)->segno;
2130
2131         if (test_opt(sbi, NOHEAP) &&
2132                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2133                 return 0;
2134
2135         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2136                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2137
2138         /* find segments from 0 to reuse freed segments */
2139         if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2140                 return 0;
2141
2142         return CURSEG_I(sbi, type)->segno;
2143 }
2144
2145 /*
2146  * Allocate a current working segment.
2147  * This function always allocates a free segment in LFS manner.
2148  */
2149 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2150 {
2151         struct curseg_info *curseg = CURSEG_I(sbi, type);
2152         unsigned int segno = curseg->segno;
2153         int dir = ALLOC_LEFT;
2154
2155         write_sum_page(sbi, curseg->sum_blk,
2156                                 GET_SUM_BLOCK(sbi, segno));
2157         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2158                 dir = ALLOC_RIGHT;
2159
2160         if (test_opt(sbi, NOHEAP))
2161                 dir = ALLOC_RIGHT;
2162
2163         segno = __get_next_segno(sbi, type);
2164         get_new_segment(sbi, &segno, new_sec, dir);
2165         curseg->next_segno = segno;
2166         reset_curseg(sbi, type, 1);
2167         curseg->alloc_type = LFS;
2168 }
2169
2170 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2171                         struct curseg_info *seg, block_t start)
2172 {
2173         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2174         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2175         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2176         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2177         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2178         int i, pos;
2179
2180         for (i = 0; i < entries; i++)
2181                 target_map[i] = ckpt_map[i] | cur_map[i];
2182
2183         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2184
2185         seg->next_blkoff = pos;
2186 }
2187
2188 /*
2189  * If a segment is written by LFS manner, next block offset is just obtained
2190  * by increasing the current block offset. However, if a segment is written by
2191  * SSR manner, next block offset obtained by calling __next_free_blkoff
2192  */
2193 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2194                                 struct curseg_info *seg)
2195 {
2196         if (seg->alloc_type == SSR)
2197                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2198         else
2199                 seg->next_blkoff++;
2200 }
2201
2202 /*
2203  * This function always allocates a used segment(from dirty seglist) by SSR
2204  * manner, so it should recover the existing segment information of valid blocks
2205  */
2206 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2207 {
2208         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2209         struct curseg_info *curseg = CURSEG_I(sbi, type);
2210         unsigned int new_segno = curseg->next_segno;
2211         struct f2fs_summary_block *sum_node;
2212         struct page *sum_page;
2213
2214         write_sum_page(sbi, curseg->sum_blk,
2215                                 GET_SUM_BLOCK(sbi, curseg->segno));
2216         __set_test_and_inuse(sbi, new_segno);
2217
2218         mutex_lock(&dirty_i->seglist_lock);
2219         __remove_dirty_segment(sbi, new_segno, PRE);
2220         __remove_dirty_segment(sbi, new_segno, DIRTY);
2221         mutex_unlock(&dirty_i->seglist_lock);
2222
2223         reset_curseg(sbi, type, 1);
2224         curseg->alloc_type = SSR;
2225         __next_free_blkoff(sbi, curseg, 0);
2226
2227         sum_page = get_sum_page(sbi, new_segno);
2228         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2229         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2230         f2fs_put_page(sum_page, 1);
2231 }
2232
2233 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2234 {
2235         struct curseg_info *curseg = CURSEG_I(sbi, type);
2236         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2237         unsigned segno = NULL_SEGNO;
2238         int i, cnt;
2239         bool reversed = false;
2240
2241         /* need_SSR() already forces to do this */
2242         if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2243                 curseg->next_segno = segno;
2244                 return 1;
2245         }
2246
2247         /* For node segments, let's do SSR more intensively */
2248         if (IS_NODESEG(type)) {
2249                 if (type >= CURSEG_WARM_NODE) {
2250                         reversed = true;
2251                         i = CURSEG_COLD_NODE;
2252                 } else {
2253                         i = CURSEG_HOT_NODE;
2254                 }
2255                 cnt = NR_CURSEG_NODE_TYPE;
2256         } else {
2257                 if (type >= CURSEG_WARM_DATA) {
2258                         reversed = true;
2259                         i = CURSEG_COLD_DATA;
2260                 } else {
2261                         i = CURSEG_HOT_DATA;
2262                 }
2263                 cnt = NR_CURSEG_DATA_TYPE;
2264         }
2265
2266         for (; cnt-- > 0; reversed ? i-- : i++) {
2267                 if (i == type)
2268                         continue;
2269                 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2270                         curseg->next_segno = segno;
2271                         return 1;
2272                 }
2273         }
2274         return 0;
2275 }
2276
2277 /*
2278  * flush out current segment and replace it with new segment
2279  * This function should be returned with success, otherwise BUG
2280  */
2281 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2282                                                 int type, bool force)
2283 {
2284         struct curseg_info *curseg = CURSEG_I(sbi, type);
2285
2286         if (force)
2287                 new_curseg(sbi, type, true);
2288         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2289                                         type == CURSEG_WARM_NODE)
2290                 new_curseg(sbi, type, false);
2291         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2292                 new_curseg(sbi, type, false);
2293         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
2294                 change_curseg(sbi, type);
2295         else
2296                 new_curseg(sbi, type, false);
2297
2298         stat_inc_seg_type(sbi, curseg);
2299 }
2300
2301 void allocate_new_segments(struct f2fs_sb_info *sbi)
2302 {
2303         struct curseg_info *curseg;
2304         unsigned int old_segno;
2305         int i;
2306
2307         down_write(&SIT_I(sbi)->sentry_lock);
2308
2309         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2310                 curseg = CURSEG_I(sbi, i);
2311                 old_segno = curseg->segno;
2312                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2313                 locate_dirty_segment(sbi, old_segno);
2314         }
2315
2316         up_write(&SIT_I(sbi)->sentry_lock);
2317 }
2318
2319 static const struct segment_allocation default_salloc_ops = {
2320         .allocate_segment = allocate_segment_by_default,
2321 };
2322
2323 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2324 {
2325         __u64 trim_start = cpc->trim_start;
2326         bool has_candidate = false;
2327
2328         down_write(&SIT_I(sbi)->sentry_lock);
2329         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2330                 if (add_discard_addrs(sbi, cpc, true)) {
2331                         has_candidate = true;
2332                         break;
2333                 }
2334         }
2335         up_write(&SIT_I(sbi)->sentry_lock);
2336
2337         cpc->trim_start = trim_start;
2338         return has_candidate;
2339 }
2340
2341 static void __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2342                                         struct discard_policy *dpolicy,
2343                                         unsigned int start, unsigned int end)
2344 {
2345         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2346         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2347         struct rb_node **insert_p = NULL, *insert_parent = NULL;
2348         struct discard_cmd *dc;
2349         struct blk_plug plug;
2350         int issued;
2351
2352 next:
2353         issued = 0;
2354
2355         mutex_lock(&dcc->cmd_lock);
2356         f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
2357
2358         dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
2359                                         NULL, start,
2360                                         (struct rb_entry **)&prev_dc,
2361                                         (struct rb_entry **)&next_dc,
2362                                         &insert_p, &insert_parent, true);
2363         if (!dc)
2364                 dc = next_dc;
2365
2366         blk_start_plug(&plug);
2367
2368         while (dc && dc->lstart <= end) {
2369                 struct rb_node *node;
2370
2371                 if (dc->len < dpolicy->granularity)
2372                         goto skip;
2373
2374                 if (dc->state != D_PREP) {
2375                         list_move_tail(&dc->list, &dcc->fstrim_list);
2376                         goto skip;
2377                 }
2378
2379                 __submit_discard_cmd(sbi, dpolicy, dc);
2380
2381                 if (++issued >= dpolicy->max_requests) {
2382                         start = dc->lstart + dc->len;
2383
2384                         blk_finish_plug(&plug);
2385                         mutex_unlock(&dcc->cmd_lock);
2386                         __wait_all_discard_cmd(sbi, NULL);
2387                         congestion_wait(BLK_RW_ASYNC, HZ/50);
2388                         goto next;
2389                 }
2390 skip:
2391                 node = rb_next(&dc->rb_node);
2392                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2393
2394                 if (fatal_signal_pending(current))
2395                         break;
2396         }
2397
2398         blk_finish_plug(&plug);
2399         mutex_unlock(&dcc->cmd_lock);
2400 }
2401
2402 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2403 {
2404         __u64 start = F2FS_BYTES_TO_BLK(range->start);
2405         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2406         unsigned int start_segno, end_segno;
2407         block_t start_block, end_block;
2408         struct cp_control cpc;
2409         struct discard_policy dpolicy;
2410         unsigned long long trimmed = 0;
2411         int err = 0;
2412
2413         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2414                 return -EINVAL;
2415
2416         if (end <= MAIN_BLKADDR(sbi))
2417                 return -EINVAL;
2418
2419         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2420                 f2fs_msg(sbi->sb, KERN_WARNING,
2421                         "Found FS corruption, run fsck to fix.");
2422                 return -EIO;
2423         }
2424
2425         /* start/end segment number in main_area */
2426         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2427         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2428                                                 GET_SEGNO(sbi, end);
2429
2430         cpc.reason = CP_DISCARD;
2431         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2432         cpc.trim_start = start_segno;
2433         cpc.trim_end = end_segno;
2434
2435         if (sbi->discard_blks == 0)
2436                 goto out;
2437
2438         mutex_lock(&sbi->gc_mutex);
2439         err = write_checkpoint(sbi, &cpc);
2440         mutex_unlock(&sbi->gc_mutex);
2441         if (err)
2442                 goto out;
2443
2444         start_block = START_BLOCK(sbi, start_segno);
2445         end_block = START_BLOCK(sbi, end_segno + 1);
2446
2447         __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2448         __issue_discard_cmd_range(sbi, &dpolicy, start_block, end_block);
2449
2450         /*
2451          * We filed discard candidates, but actually we don't need to wait for
2452          * all of them, since they'll be issued in idle time along with runtime
2453          * discard option. User configuration looks like using runtime discard
2454          * or periodic fstrim instead of it.
2455          */
2456         if (!test_opt(sbi, DISCARD)) {
2457                 trimmed = __wait_discard_cmd_range(sbi, &dpolicy,
2458                                         start_block, end_block);
2459                 range->len = F2FS_BLK_TO_BYTES(trimmed);
2460         }
2461 out:
2462         return err;
2463 }
2464
2465 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2466 {
2467         struct curseg_info *curseg = CURSEG_I(sbi, type);
2468         if (curseg->next_blkoff < sbi->blocks_per_seg)
2469                 return true;
2470         return false;
2471 }
2472
2473 int rw_hint_to_seg_type(enum rw_hint hint)
2474 {
2475         switch (hint) {
2476         case WRITE_LIFE_SHORT:
2477                 return CURSEG_HOT_DATA;
2478         case WRITE_LIFE_EXTREME:
2479                 return CURSEG_COLD_DATA;
2480         default:
2481                 return CURSEG_WARM_DATA;
2482         }
2483 }
2484
2485 /* This returns write hints for each segment type. This hints will be
2486  * passed down to block layer. There are mapping tables which depend on
2487  * the mount option 'whint_mode'.
2488  *
2489  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2490  *
2491  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2492  *
2493  * User                  F2FS                     Block
2494  * ----                  ----                     -----
2495  *                       META                     WRITE_LIFE_NOT_SET
2496  *                       HOT_NODE                 "
2497  *                       WARM_NODE                "
2498  *                       COLD_NODE                "
2499  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2500  * extension list        "                        "
2501  *
2502  * -- buffered io
2503  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2504  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2505  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2506  * WRITE_LIFE_NONE       "                        "
2507  * WRITE_LIFE_MEDIUM     "                        "
2508  * WRITE_LIFE_LONG       "                        "
2509  *
2510  * -- direct io
2511  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2512  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2513  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2514  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2515  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2516  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2517  *
2518  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2519  *
2520  * User                  F2FS                     Block
2521  * ----                  ----                     -----
2522  *                       META                     WRITE_LIFE_MEDIUM;
2523  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
2524  *                       WARM_NODE                "
2525  *                       COLD_NODE                WRITE_LIFE_NONE
2526  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2527  * extension list        "                        "
2528  *
2529  * -- buffered io
2530  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2531  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2532  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
2533  * WRITE_LIFE_NONE       "                        "
2534  * WRITE_LIFE_MEDIUM     "                        "
2535  * WRITE_LIFE_LONG       "                        "
2536  *
2537  * -- direct io
2538  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2539  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2540  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2541  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2542  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2543  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2544  */
2545
2546 enum rw_hint io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2547                                 enum page_type type, enum temp_type temp)
2548 {
2549         if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2550                 if (type == DATA) {
2551                         if (temp == WARM)
2552                                 return WRITE_LIFE_NOT_SET;
2553                         else if (temp == HOT)
2554                                 return WRITE_LIFE_SHORT;
2555                         else if (temp == COLD)
2556                                 return WRITE_LIFE_EXTREME;
2557                 } else {
2558                         return WRITE_LIFE_NOT_SET;
2559                 }
2560         } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2561                 if (type == DATA) {
2562                         if (temp == WARM)
2563                                 return WRITE_LIFE_LONG;
2564                         else if (temp == HOT)
2565                                 return WRITE_LIFE_SHORT;
2566                         else if (temp == COLD)
2567                                 return WRITE_LIFE_EXTREME;
2568                 } else if (type == NODE) {
2569                         if (temp == WARM || temp == HOT)
2570                                 return WRITE_LIFE_NOT_SET;
2571                         else if (temp == COLD)
2572                                 return WRITE_LIFE_NONE;
2573                 } else if (type == META) {
2574                         return WRITE_LIFE_MEDIUM;
2575                 }
2576         }
2577         return WRITE_LIFE_NOT_SET;
2578 }
2579
2580 static int __get_segment_type_2(struct f2fs_io_info *fio)
2581 {
2582         if (fio->type == DATA)
2583                 return CURSEG_HOT_DATA;
2584         else
2585                 return CURSEG_HOT_NODE;
2586 }
2587
2588 static int __get_segment_type_4(struct f2fs_io_info *fio)
2589 {
2590         if (fio->type == DATA) {
2591                 struct inode *inode = fio->page->mapping->host;
2592
2593                 if (S_ISDIR(inode->i_mode))
2594                         return CURSEG_HOT_DATA;
2595                 else
2596                         return CURSEG_COLD_DATA;
2597         } else {
2598                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2599                         return CURSEG_WARM_NODE;
2600                 else
2601                         return CURSEG_COLD_NODE;
2602         }
2603 }
2604
2605 static int __get_segment_type_6(struct f2fs_io_info *fio)
2606 {
2607         if (fio->type == DATA) {
2608                 struct inode *inode = fio->page->mapping->host;
2609
2610                 if (is_cold_data(fio->page) || file_is_cold(inode))
2611                         return CURSEG_COLD_DATA;
2612                 if (file_is_hot(inode) ||
2613                                 is_inode_flag_set(inode, FI_HOT_DATA))
2614                         return CURSEG_HOT_DATA;
2615                 return rw_hint_to_seg_type(inode->i_write_hint);
2616         } else {
2617                 if (IS_DNODE(fio->page))
2618                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2619                                                 CURSEG_HOT_NODE;
2620                 return CURSEG_COLD_NODE;
2621         }
2622 }
2623
2624 static int __get_segment_type(struct f2fs_io_info *fio)
2625 {
2626         int type = 0;
2627
2628         switch (F2FS_OPTION(fio->sbi).active_logs) {
2629         case 2:
2630                 type = __get_segment_type_2(fio);
2631                 break;
2632         case 4:
2633                 type = __get_segment_type_4(fio);
2634                 break;
2635         case 6:
2636                 type = __get_segment_type_6(fio);
2637                 break;
2638         default:
2639                 f2fs_bug_on(fio->sbi, true);
2640         }
2641
2642         if (IS_HOT(type))
2643                 fio->temp = HOT;
2644         else if (IS_WARM(type))
2645                 fio->temp = WARM;
2646         else
2647                 fio->temp = COLD;
2648         return type;
2649 }
2650
2651 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2652                 block_t old_blkaddr, block_t *new_blkaddr,
2653                 struct f2fs_summary *sum, int type,
2654                 struct f2fs_io_info *fio, bool add_list)
2655 {
2656         struct sit_info *sit_i = SIT_I(sbi);
2657         struct curseg_info *curseg = CURSEG_I(sbi, type);
2658
2659         down_read(&SM_I(sbi)->curseg_lock);
2660
2661         mutex_lock(&curseg->curseg_mutex);
2662         down_write(&sit_i->sentry_lock);
2663
2664         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2665
2666         f2fs_wait_discard_bio(sbi, *new_blkaddr);
2667
2668         /*
2669          * __add_sum_entry should be resided under the curseg_mutex
2670          * because, this function updates a summary entry in the
2671          * current summary block.
2672          */
2673         __add_sum_entry(sbi, type, sum);
2674
2675         __refresh_next_blkoff(sbi, curseg);
2676
2677         stat_inc_block_count(sbi, curseg);
2678
2679         /*
2680          * SIT information should be updated before segment allocation,
2681          * since SSR needs latest valid block information.
2682          */
2683         update_sit_entry(sbi, *new_blkaddr, 1);
2684         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2685                 update_sit_entry(sbi, old_blkaddr, -1);
2686
2687         if (!__has_curseg_space(sbi, type))
2688                 sit_i->s_ops->allocate_segment(sbi, type, false);
2689
2690         /*
2691          * segment dirty status should be updated after segment allocation,
2692          * so we just need to update status only one time after previous
2693          * segment being closed.
2694          */
2695         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2696         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
2697
2698         up_write(&sit_i->sentry_lock);
2699
2700         if (page && IS_NODESEG(type)) {
2701                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2702
2703                 f2fs_inode_chksum_set(sbi, page);
2704         }
2705
2706         if (add_list) {
2707                 struct f2fs_bio_info *io;
2708
2709                 INIT_LIST_HEAD(&fio->list);
2710                 fio->in_list = true;
2711                 io = sbi->write_io[fio->type] + fio->temp;
2712                 spin_lock(&io->io_lock);
2713                 list_add_tail(&fio->list, &io->io_list);
2714                 spin_unlock(&io->io_lock);
2715         }
2716
2717         mutex_unlock(&curseg->curseg_mutex);
2718
2719         up_read(&SM_I(sbi)->curseg_lock);
2720 }
2721
2722 static void update_device_state(struct f2fs_io_info *fio)
2723 {
2724         struct f2fs_sb_info *sbi = fio->sbi;
2725         unsigned int devidx;
2726
2727         if (!sbi->s_ndevs)
2728                 return;
2729
2730         devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2731
2732         /* update device state for fsync */
2733         set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
2734
2735         /* update device state for checkpoint */
2736         if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2737                 spin_lock(&sbi->dev_lock);
2738                 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2739                 spin_unlock(&sbi->dev_lock);
2740         }
2741 }
2742
2743 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2744 {
2745         int type = __get_segment_type(fio);
2746         int err;
2747
2748 reallocate:
2749         allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
2750                         &fio->new_blkaddr, sum, type, fio, true);
2751
2752         /* writeout dirty page into bdev */
2753         err = f2fs_submit_page_write(fio);
2754         if (err == -EAGAIN) {
2755                 fio->old_blkaddr = fio->new_blkaddr;
2756                 goto reallocate;
2757         } else if (!err) {
2758                 update_device_state(fio);
2759         }
2760 }
2761
2762 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2763                                         enum iostat_type io_type)
2764 {
2765         struct f2fs_io_info fio = {
2766                 .sbi = sbi,
2767                 .type = META,
2768                 .temp = HOT,
2769                 .op = REQ_OP_WRITE,
2770                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
2771                 .old_blkaddr = page->index,
2772                 .new_blkaddr = page->index,
2773                 .page = page,
2774                 .encrypted_page = NULL,
2775                 .in_list = false,
2776         };
2777
2778         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
2779                 fio.op_flags &= ~REQ_META;
2780
2781         set_page_writeback(page);
2782         ClearPageError(page);
2783         f2fs_submit_page_write(&fio);
2784
2785         f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
2786 }
2787
2788 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
2789 {
2790         struct f2fs_summary sum;
2791
2792         set_summary(&sum, nid, 0, 0);
2793         do_write_page(&sum, fio);
2794
2795         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2796 }
2797
2798 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
2799 {
2800         struct f2fs_sb_info *sbi = fio->sbi;
2801         struct f2fs_summary sum;
2802         struct node_info ni;
2803
2804         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
2805         get_node_info(sbi, dn->nid, &ni);
2806         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
2807         do_write_page(&sum, fio);
2808         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
2809
2810         f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
2811 }
2812
2813 int rewrite_data_page(struct f2fs_io_info *fio)
2814 {
2815         int err;
2816         struct f2fs_sb_info *sbi = fio->sbi;
2817
2818         fio->new_blkaddr = fio->old_blkaddr;
2819         /* i/o temperature is needed for passing down write hints */
2820         __get_segment_type(fio);
2821
2822         f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
2823                         GET_SEGNO(sbi, fio->new_blkaddr))->type));
2824
2825         stat_inc_inplace_blocks(fio->sbi);
2826
2827         err = f2fs_submit_page_bio(fio);
2828         if (!err)
2829                 update_device_state(fio);
2830
2831         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2832
2833         return err;
2834 }
2835
2836 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
2837                                                 unsigned int segno)
2838 {
2839         int i;
2840
2841         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
2842                 if (CURSEG_I(sbi, i)->segno == segno)
2843                         break;
2844         }
2845         return i;
2846 }
2847
2848 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
2849                                 block_t old_blkaddr, block_t new_blkaddr,
2850                                 bool recover_curseg, bool recover_newaddr)
2851 {
2852         struct sit_info *sit_i = SIT_I(sbi);
2853         struct curseg_info *curseg;
2854         unsigned int segno, old_cursegno;
2855         struct seg_entry *se;
2856         int type;
2857         unsigned short old_blkoff;
2858
2859         segno = GET_SEGNO(sbi, new_blkaddr);
2860         se = get_seg_entry(sbi, segno);
2861         type = se->type;
2862
2863         down_write(&SM_I(sbi)->curseg_lock);
2864
2865         if (!recover_curseg) {
2866                 /* for recovery flow */
2867                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2868                         if (old_blkaddr == NULL_ADDR)
2869                                 type = CURSEG_COLD_DATA;
2870                         else
2871                                 type = CURSEG_WARM_DATA;
2872                 }
2873         } else {
2874                 if (IS_CURSEG(sbi, segno)) {
2875                         /* se->type is volatile as SSR allocation */
2876                         type = __f2fs_get_curseg(sbi, segno);
2877                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
2878                 } else {
2879                         type = CURSEG_WARM_DATA;
2880                 }
2881         }
2882
2883         f2fs_bug_on(sbi, !IS_DATASEG(type));
2884         curseg = CURSEG_I(sbi, type);
2885
2886         mutex_lock(&curseg->curseg_mutex);
2887         down_write(&sit_i->sentry_lock);
2888
2889         old_cursegno = curseg->segno;
2890         old_blkoff = curseg->next_blkoff;
2891
2892         /* change the current segment */
2893         if (segno != curseg->segno) {
2894                 curseg->next_segno = segno;
2895                 change_curseg(sbi, type);
2896         }
2897
2898         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
2899         __add_sum_entry(sbi, type, sum);
2900
2901         if (!recover_curseg || recover_newaddr)
2902                 update_sit_entry(sbi, new_blkaddr, 1);
2903         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2904                 update_sit_entry(sbi, old_blkaddr, -1);
2905
2906         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2907         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2908
2909         locate_dirty_segment(sbi, old_cursegno);
2910
2911         if (recover_curseg) {
2912                 if (old_cursegno != curseg->segno) {
2913                         curseg->next_segno = old_cursegno;
2914                         change_curseg(sbi, type);
2915                 }
2916                 curseg->next_blkoff = old_blkoff;
2917         }
2918
2919         up_write(&sit_i->sentry_lock);
2920         mutex_unlock(&curseg->curseg_mutex);
2921         up_write(&SM_I(sbi)->curseg_lock);
2922 }
2923
2924 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2925                                 block_t old_addr, block_t new_addr,
2926                                 unsigned char version, bool recover_curseg,
2927                                 bool recover_newaddr)
2928 {
2929         struct f2fs_summary sum;
2930
2931         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2932
2933         __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2934                                         recover_curseg, recover_newaddr);
2935
2936         f2fs_update_data_blkaddr(dn, new_addr);
2937 }
2938
2939 void f2fs_wait_on_page_writeback(struct page *page,
2940                                 enum page_type type, bool ordered)
2941 {
2942         if (PageWriteback(page)) {
2943                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2944
2945                 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2946                                                 0, page->index, type);
2947                 if (ordered)
2948                         wait_on_page_writeback(page);
2949                 else
2950                         wait_for_stable_page(page);
2951         }
2952 }
2953
2954 void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
2955 {
2956         struct page *cpage;
2957
2958         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
2959                 return;
2960
2961         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2962         if (cpage) {
2963                 f2fs_wait_on_page_writeback(cpage, DATA, true);
2964                 f2fs_put_page(cpage, 1);
2965         }
2966 }
2967
2968 static void read_compacted_summaries(struct f2fs_sb_info *sbi)
2969 {
2970         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2971         struct curseg_info *seg_i;
2972         unsigned char *kaddr;
2973         struct page *page;
2974         block_t start;
2975         int i, j, offset;
2976
2977         start = start_sum_block(sbi);
2978
2979         page = get_meta_page(sbi, start++);
2980         kaddr = (unsigned char *)page_address(page);
2981
2982         /* Step 1: restore nat cache */
2983         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2984         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
2985
2986         /* Step 2: restore sit cache */
2987         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2988         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
2989         offset = 2 * SUM_JOURNAL_SIZE;
2990
2991         /* Step 3: restore summary entries */
2992         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2993                 unsigned short blk_off;
2994                 unsigned int segno;
2995
2996                 seg_i = CURSEG_I(sbi, i);
2997                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2998                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2999                 seg_i->next_segno = segno;
3000                 reset_curseg(sbi, i, 0);
3001                 seg_i->alloc_type = ckpt->alloc_type[i];
3002                 seg_i->next_blkoff = blk_off;
3003
3004                 if (seg_i->alloc_type == SSR)
3005                         blk_off = sbi->blocks_per_seg;
3006
3007                 for (j = 0; j < blk_off; j++) {
3008                         struct f2fs_summary *s;
3009                         s = (struct f2fs_summary *)(kaddr + offset);
3010                         seg_i->sum_blk->entries[j] = *s;
3011                         offset += SUMMARY_SIZE;
3012                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3013                                                 SUM_FOOTER_SIZE)
3014                                 continue;
3015
3016                         f2fs_put_page(page, 1);
3017                         page = NULL;
3018
3019                         page = get_meta_page(sbi, start++);
3020                         kaddr = (unsigned char *)page_address(page);
3021                         offset = 0;
3022                 }
3023         }
3024         f2fs_put_page(page, 1);
3025 }
3026
3027 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3028 {
3029         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3030         struct f2fs_summary_block *sum;
3031         struct curseg_info *curseg;
3032         struct page *new;
3033         unsigned short blk_off;
3034         unsigned int segno = 0;
3035         block_t blk_addr = 0;
3036
3037         /* get segment number and block addr */
3038         if (IS_DATASEG(type)) {
3039                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3040                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3041                                                         CURSEG_HOT_DATA]);
3042                 if (__exist_node_summaries(sbi))
3043                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3044                 else
3045                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3046         } else {
3047                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3048                                                         CURSEG_HOT_NODE]);
3049                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3050                                                         CURSEG_HOT_NODE]);
3051                 if (__exist_node_summaries(sbi))
3052                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3053                                                         type - CURSEG_HOT_NODE);
3054                 else
3055                         blk_addr = GET_SUM_BLOCK(sbi, segno);
3056         }
3057
3058         new = get_meta_page(sbi, blk_addr);
3059         sum = (struct f2fs_summary_block *)page_address(new);
3060
3061         if (IS_NODESEG(type)) {
3062                 if (__exist_node_summaries(sbi)) {
3063                         struct f2fs_summary *ns = &sum->entries[0];
3064                         int i;
3065                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3066                                 ns->version = 0;
3067                                 ns->ofs_in_node = 0;
3068                         }
3069                 } else {
3070                         restore_node_summary(sbi, segno, sum);
3071                 }
3072         }
3073
3074         /* set uncompleted segment to curseg */
3075         curseg = CURSEG_I(sbi, type);
3076         mutex_lock(&curseg->curseg_mutex);
3077
3078         /* update journal info */
3079         down_write(&curseg->journal_rwsem);
3080         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3081         up_write(&curseg->journal_rwsem);
3082
3083         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3084         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3085         curseg->next_segno = segno;
3086         reset_curseg(sbi, type, 0);
3087         curseg->alloc_type = ckpt->alloc_type[type];
3088         curseg->next_blkoff = blk_off;
3089         mutex_unlock(&curseg->curseg_mutex);
3090         f2fs_put_page(new, 1);
3091         return 0;
3092 }
3093
3094 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3095 {
3096         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3097         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3098         int type = CURSEG_HOT_DATA;
3099         int err;
3100
3101         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3102                 int npages = npages_for_summary_flush(sbi, true);
3103
3104                 if (npages >= 2)
3105                         ra_meta_pages(sbi, start_sum_block(sbi), npages,
3106                                                         META_CP, true);
3107
3108                 /* restore for compacted data summary */
3109                 read_compacted_summaries(sbi);
3110                 type = CURSEG_HOT_NODE;
3111         }
3112
3113         if (__exist_node_summaries(sbi))
3114                 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3115                                         NR_CURSEG_TYPE - type, META_CP, true);
3116
3117         for (; type <= CURSEG_COLD_NODE; type++) {
3118                 err = read_normal_summaries(sbi, type);
3119                 if (err)
3120                         return err;
3121         }
3122
3123         /* sanity check for summary blocks */
3124         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3125                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3126                 return -EINVAL;
3127
3128         return 0;
3129 }
3130
3131 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3132 {
3133         struct page *page;
3134         unsigned char *kaddr;
3135         struct f2fs_summary *summary;
3136         struct curseg_info *seg_i;
3137         int written_size = 0;
3138         int i, j;
3139
3140         page = grab_meta_page(sbi, blkaddr++);
3141         kaddr = (unsigned char *)page_address(page);
3142         memset(kaddr, 0, PAGE_SIZE);
3143
3144         /* Step 1: write nat cache */
3145         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3146         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3147         written_size += SUM_JOURNAL_SIZE;
3148
3149         /* Step 2: write sit cache */
3150         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3151         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3152         written_size += SUM_JOURNAL_SIZE;
3153
3154         /* Step 3: write summary entries */
3155         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3156                 unsigned short blkoff;
3157                 seg_i = CURSEG_I(sbi, i);
3158                 if (sbi->ckpt->alloc_type[i] == SSR)
3159                         blkoff = sbi->blocks_per_seg;
3160                 else
3161                         blkoff = curseg_blkoff(sbi, i);
3162
3163                 for (j = 0; j < blkoff; j++) {
3164                         if (!page) {
3165                                 page = grab_meta_page(sbi, blkaddr++);
3166                                 kaddr = (unsigned char *)page_address(page);
3167                                 memset(kaddr, 0, PAGE_SIZE);
3168                                 written_size = 0;
3169                         }
3170                         summary = (struct f2fs_summary *)(kaddr + written_size);
3171                         *summary = seg_i->sum_blk->entries[j];
3172                         written_size += SUMMARY_SIZE;
3173
3174                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3175                                                         SUM_FOOTER_SIZE)
3176                                 continue;
3177
3178                         set_page_dirty(page);
3179                         f2fs_put_page(page, 1);
3180                         page = NULL;
3181                 }
3182         }
3183         if (page) {
3184                 set_page_dirty(page);
3185                 f2fs_put_page(page, 1);
3186         }
3187 }
3188
3189 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3190                                         block_t blkaddr, int type)
3191 {
3192         int i, end;
3193         if (IS_DATASEG(type))
3194                 end = type + NR_CURSEG_DATA_TYPE;
3195         else
3196                 end = type + NR_CURSEG_NODE_TYPE;
3197
3198         for (i = type; i < end; i++)
3199                 write_current_sum_page(sbi, i, blkaddr + (i - type));
3200 }
3201
3202 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3203 {
3204         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3205                 write_compacted_summaries(sbi, start_blk);
3206         else
3207                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3208 }
3209
3210 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3211 {
3212         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3213 }
3214
3215 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3216                                         unsigned int val, int alloc)
3217 {
3218         int i;
3219
3220         if (type == NAT_JOURNAL) {
3221                 for (i = 0; i < nats_in_cursum(journal); i++) {
3222                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3223                                 return i;
3224                 }
3225                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3226                         return update_nats_in_cursum(journal, 1);
3227         } else if (type == SIT_JOURNAL) {
3228                 for (i = 0; i < sits_in_cursum(journal); i++)
3229                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3230                                 return i;
3231                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3232                         return update_sits_in_cursum(journal, 1);
3233         }
3234         return -1;
3235 }
3236
3237 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3238                                         unsigned int segno)
3239 {
3240         return get_meta_page(sbi, current_sit_addr(sbi, segno));
3241 }
3242
3243 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3244                                         unsigned int start)
3245 {
3246         struct sit_info *sit_i = SIT_I(sbi);
3247         struct page *page;
3248         pgoff_t src_off, dst_off;
3249
3250         src_off = current_sit_addr(sbi, start);
3251         dst_off = next_sit_addr(sbi, src_off);
3252
3253         page = grab_meta_page(sbi, dst_off);
3254         seg_info_to_sit_page(sbi, page, start);
3255
3256         set_page_dirty(page);
3257         set_to_next_sit(sit_i, start);
3258
3259         return page;
3260 }
3261
3262 static struct sit_entry_set *grab_sit_entry_set(void)
3263 {
3264         struct sit_entry_set *ses =
3265                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3266
3267         ses->entry_cnt = 0;
3268         INIT_LIST_HEAD(&ses->set_list);
3269         return ses;
3270 }
3271
3272 static void release_sit_entry_set(struct sit_entry_set *ses)
3273 {
3274         list_del(&ses->set_list);
3275         kmem_cache_free(sit_entry_set_slab, ses);
3276 }
3277
3278 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3279                                                 struct list_head *head)
3280 {
3281         struct sit_entry_set *next = ses;
3282
3283         if (list_is_last(&ses->set_list, head))
3284                 return;
3285
3286         list_for_each_entry_continue(next, head, set_list)
3287                 if (ses->entry_cnt <= next->entry_cnt)
3288                         break;
3289
3290         list_move_tail(&ses->set_list, &next->set_list);
3291 }
3292
3293 static void add_sit_entry(unsigned int segno, struct list_head *head)
3294 {
3295         struct sit_entry_set *ses;
3296         unsigned int start_segno = START_SEGNO(segno);
3297
3298         list_for_each_entry(ses, head, set_list) {
3299                 if (ses->start_segno == start_segno) {
3300                         ses->entry_cnt++;
3301                         adjust_sit_entry_set(ses, head);
3302                         return;
3303                 }
3304         }
3305
3306         ses = grab_sit_entry_set();
3307
3308         ses->start_segno = start_segno;
3309         ses->entry_cnt++;
3310         list_add(&ses->set_list, head);
3311 }
3312
3313 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3314 {
3315         struct f2fs_sm_info *sm_info = SM_I(sbi);
3316         struct list_head *set_list = &sm_info->sit_entry_set;
3317         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3318         unsigned int segno;
3319
3320         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3321                 add_sit_entry(segno, set_list);
3322 }
3323
3324 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3325 {
3326         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3327         struct f2fs_journal *journal = curseg->journal;
3328         int i;
3329
3330         down_write(&curseg->journal_rwsem);
3331         for (i = 0; i < sits_in_cursum(journal); i++) {
3332                 unsigned int segno;
3333                 bool dirtied;
3334
3335                 segno = le32_to_cpu(segno_in_journal(journal, i));
3336                 dirtied = __mark_sit_entry_dirty(sbi, segno);
3337
3338                 if (!dirtied)
3339                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3340         }
3341         update_sits_in_cursum(journal, -i);
3342         up_write(&curseg->journal_rwsem);
3343 }
3344
3345 /*
3346  * CP calls this function, which flushes SIT entries including sit_journal,
3347  * and moves prefree segs to free segs.
3348  */
3349 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3350 {
3351         struct sit_info *sit_i = SIT_I(sbi);
3352         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3353         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3354         struct f2fs_journal *journal = curseg->journal;
3355         struct sit_entry_set *ses, *tmp;
3356         struct list_head *head = &SM_I(sbi)->sit_entry_set;
3357         bool to_journal = true;
3358         struct seg_entry *se;
3359
3360         down_write(&sit_i->sentry_lock);
3361
3362         if (!sit_i->dirty_sentries)
3363                 goto out;
3364
3365         /*
3366          * add and account sit entries of dirty bitmap in sit entry
3367          * set temporarily
3368          */
3369         add_sits_in_set(sbi);
3370
3371         /*
3372          * if there are no enough space in journal to store dirty sit
3373          * entries, remove all entries from journal and add and account
3374          * them in sit entry set.
3375          */
3376         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3377                 remove_sits_in_journal(sbi);
3378
3379         /*
3380          * there are two steps to flush sit entries:
3381          * #1, flush sit entries to journal in current cold data summary block.
3382          * #2, flush sit entries to sit page.
3383          */
3384         list_for_each_entry_safe(ses, tmp, head, set_list) {
3385                 struct page *page = NULL;
3386                 struct f2fs_sit_block *raw_sit = NULL;
3387                 unsigned int start_segno = ses->start_segno;
3388                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3389                                                 (unsigned long)MAIN_SEGS(sbi));
3390                 unsigned int segno = start_segno;
3391
3392                 if (to_journal &&
3393                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3394                         to_journal = false;
3395
3396                 if (to_journal) {
3397                         down_write(&curseg->journal_rwsem);
3398                 } else {
3399                         page = get_next_sit_page(sbi, start_segno);
3400                         raw_sit = page_address(page);
3401                 }
3402
3403                 /* flush dirty sit entries in region of current sit set */
3404                 for_each_set_bit_from(segno, bitmap, end) {
3405                         int offset, sit_offset;
3406
3407                         se = get_seg_entry(sbi, segno);
3408 #ifdef CONFIG_F2FS_CHECK_FS
3409                         if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3410                                                 SIT_VBLOCK_MAP_SIZE))
3411                                 f2fs_bug_on(sbi, 1);
3412 #endif
3413
3414                         /* add discard candidates */
3415                         if (!(cpc->reason & CP_DISCARD)) {
3416                                 cpc->trim_start = segno;
3417                                 add_discard_addrs(sbi, cpc, false);
3418                         }
3419
3420                         if (to_journal) {
3421                                 offset = lookup_journal_in_cursum(journal,
3422                                                         SIT_JOURNAL, segno, 1);
3423                                 f2fs_bug_on(sbi, offset < 0);
3424                                 segno_in_journal(journal, offset) =
3425                                                         cpu_to_le32(segno);
3426                                 seg_info_to_raw_sit(se,
3427                                         &sit_in_journal(journal, offset));
3428                                 check_block_count(sbi, segno,
3429                                         &sit_in_journal(journal, offset));
3430                         } else {
3431                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3432                                 seg_info_to_raw_sit(se,
3433                                                 &raw_sit->entries[sit_offset]);
3434                                 check_block_count(sbi, segno,
3435                                                 &raw_sit->entries[sit_offset]);
3436                         }
3437
3438                         __clear_bit(segno, bitmap);
3439                         sit_i->dirty_sentries--;
3440                         ses->entry_cnt--;
3441                 }
3442
3443                 if (to_journal)
3444                         up_write(&curseg->journal_rwsem);
3445                 else
3446                         f2fs_put_page(page, 1);
3447
3448                 f2fs_bug_on(sbi, ses->entry_cnt);
3449                 release_sit_entry_set(ses);
3450         }
3451
3452         f2fs_bug_on(sbi, !list_empty(head));
3453         f2fs_bug_on(sbi, sit_i->dirty_sentries);
3454 out:
3455         if (cpc->reason & CP_DISCARD) {
3456                 __u64 trim_start = cpc->trim_start;
3457
3458                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3459                         add_discard_addrs(sbi, cpc, false);
3460
3461                 cpc->trim_start = trim_start;
3462         }
3463         up_write(&sit_i->sentry_lock);
3464
3465         set_prefree_as_free_segments(sbi);
3466 }
3467
3468 static int build_sit_info(struct f2fs_sb_info *sbi)
3469 {
3470         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3471         struct sit_info *sit_i;
3472         unsigned int sit_segs, start;
3473         char *src_bitmap;
3474         unsigned int bitmap_size;
3475
3476         /* allocate memory for SIT information */
3477         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3478         if (!sit_i)
3479                 return -ENOMEM;
3480
3481         SM_I(sbi)->sit_info = sit_i;
3482
3483         sit_i->sentries = f2fs_kvzalloc(sbi, MAIN_SEGS(sbi) *
3484                                         sizeof(struct seg_entry), GFP_KERNEL);
3485         if (!sit_i->sentries)
3486                 return -ENOMEM;
3487
3488         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3489         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3490                                                                 GFP_KERNEL);
3491         if (!sit_i->dirty_sentries_bitmap)
3492                 return -ENOMEM;
3493
3494         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3495                 sit_i->sentries[start].cur_valid_map
3496                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3497                 sit_i->sentries[start].ckpt_valid_map
3498                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3499                 if (!sit_i->sentries[start].cur_valid_map ||
3500                                 !sit_i->sentries[start].ckpt_valid_map)
3501                         return -ENOMEM;
3502
3503 #ifdef CONFIG_F2FS_CHECK_FS
3504                 sit_i->sentries[start].cur_valid_map_mir
3505                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3506                 if (!sit_i->sentries[start].cur_valid_map_mir)
3507                         return -ENOMEM;
3508 #endif
3509
3510                 if (f2fs_discard_en(sbi)) {
3511                         sit_i->sentries[start].discard_map
3512                                 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3513                                                                 GFP_KERNEL);
3514                         if (!sit_i->sentries[start].discard_map)
3515                                 return -ENOMEM;
3516                 }
3517         }
3518
3519         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3520         if (!sit_i->tmp_map)
3521                 return -ENOMEM;
3522
3523         if (sbi->segs_per_sec > 1) {
3524                 sit_i->sec_entries = f2fs_kvzalloc(sbi, MAIN_SECS(sbi) *
3525                                         sizeof(struct sec_entry), GFP_KERNEL);
3526                 if (!sit_i->sec_entries)
3527                         return -ENOMEM;
3528         }
3529
3530         /* get information related with SIT */
3531         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3532
3533         /* setup SIT bitmap from ckeckpoint pack */
3534         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3535         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3536
3537         sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3538         if (!sit_i->sit_bitmap)
3539                 return -ENOMEM;
3540
3541 #ifdef CONFIG_F2FS_CHECK_FS
3542         sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3543         if (!sit_i->sit_bitmap_mir)
3544                 return -ENOMEM;
3545 #endif
3546
3547         /* init SIT information */
3548         sit_i->s_ops = &default_salloc_ops;
3549
3550         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3551         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
3552         sit_i->written_valid_blocks = 0;
3553         sit_i->bitmap_size = bitmap_size;
3554         sit_i->dirty_sentries = 0;
3555         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3556         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3557         sit_i->mounted_time = ktime_get_real_seconds();
3558         init_rwsem(&sit_i->sentry_lock);
3559         return 0;
3560 }
3561
3562 static int build_free_segmap(struct f2fs_sb_info *sbi)
3563 {
3564         struct free_segmap_info *free_i;
3565         unsigned int bitmap_size, sec_bitmap_size;
3566
3567         /* allocate memory for free segmap information */
3568         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
3569         if (!free_i)
3570                 return -ENOMEM;
3571
3572         SM_I(sbi)->free_info = free_i;
3573
3574         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3575         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
3576         if (!free_i->free_segmap)
3577                 return -ENOMEM;
3578
3579         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3580         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
3581         if (!free_i->free_secmap)
3582                 return -ENOMEM;
3583
3584         /* set all segments as dirty temporarily */
3585         memset(free_i->free_segmap, 0xff, bitmap_size);
3586         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3587
3588         /* init free segmap information */
3589         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
3590         free_i->free_segments = 0;
3591         free_i->free_sections = 0;
3592         spin_lock_init(&free_i->segmap_lock);
3593         return 0;
3594 }
3595
3596 static int build_curseg(struct f2fs_sb_info *sbi)
3597 {
3598         struct curseg_info *array;
3599         int i;
3600
3601         array = f2fs_kzalloc(sbi, sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
3602         if (!array)
3603                 return -ENOMEM;
3604
3605         SM_I(sbi)->curseg_array = array;
3606
3607         for (i = 0; i < NR_CURSEG_TYPE; i++) {
3608                 mutex_init(&array[i].curseg_mutex);
3609                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
3610                 if (!array[i].sum_blk)
3611                         return -ENOMEM;
3612                 init_rwsem(&array[i].journal_rwsem);
3613                 array[i].journal = f2fs_kzalloc(sbi,
3614                                 sizeof(struct f2fs_journal), GFP_KERNEL);
3615                 if (!array[i].journal)
3616                         return -ENOMEM;
3617                 array[i].segno = NULL_SEGNO;
3618                 array[i].next_blkoff = 0;
3619         }
3620         return restore_curseg_summaries(sbi);
3621 }
3622
3623 static int build_sit_entries(struct f2fs_sb_info *sbi)
3624 {
3625         struct sit_info *sit_i = SIT_I(sbi);
3626         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3627         struct f2fs_journal *journal = curseg->journal;
3628         struct seg_entry *se;
3629         struct f2fs_sit_entry sit;
3630         int sit_blk_cnt = SIT_BLK_CNT(sbi);
3631         unsigned int i, start, end;
3632         unsigned int readed, start_blk = 0;
3633         int err = 0;
3634
3635         do {
3636                 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3637                                                         META_SIT, true);
3638
3639                 start = start_blk * sit_i->sents_per_block;
3640                 end = (start_blk + readed) * sit_i->sents_per_block;
3641
3642                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
3643                         struct f2fs_sit_block *sit_blk;
3644                         struct page *page;
3645
3646                         se = &sit_i->sentries[start];
3647                         page = get_current_sit_page(sbi, start);
3648                         sit_blk = (struct f2fs_sit_block *)page_address(page);
3649                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3650                         f2fs_put_page(page, 1);
3651
3652                         err = check_block_count(sbi, start, &sit);
3653                         if (err)
3654                                 return err;
3655                         seg_info_from_raw_sit(se, &sit);
3656
3657                         /* build discard map only one time */
3658                         if (f2fs_discard_en(sbi)) {
3659                                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3660                                         memset(se->discard_map, 0xff,
3661                                                 SIT_VBLOCK_MAP_SIZE);
3662                                 } else {
3663                                         memcpy(se->discard_map,
3664                                                 se->cur_valid_map,
3665                                                 SIT_VBLOCK_MAP_SIZE);
3666                                         sbi->discard_blks +=
3667                                                 sbi->blocks_per_seg -
3668                                                 se->valid_blocks;
3669                                 }
3670                         }
3671
3672                         if (sbi->segs_per_sec > 1)
3673                                 get_sec_entry(sbi, start)->valid_blocks +=
3674                                                         se->valid_blocks;
3675                 }
3676                 start_blk += readed;
3677         } while (start_blk < sit_blk_cnt);
3678
3679         down_read(&curseg->journal_rwsem);
3680         for (i = 0; i < sits_in_cursum(journal); i++) {
3681                 unsigned int old_valid_blocks;
3682
3683                 start = le32_to_cpu(segno_in_journal(journal, i));
3684                 se = &sit_i->sentries[start];
3685                 sit = sit_in_journal(journal, i);
3686
3687                 old_valid_blocks = se->valid_blocks;
3688
3689                 err = check_block_count(sbi, start, &sit);
3690                 if (err)
3691                         break;
3692                 seg_info_from_raw_sit(se, &sit);
3693
3694                 if (f2fs_discard_en(sbi)) {
3695                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3696                                 memset(se->discard_map, 0xff,
3697                                                         SIT_VBLOCK_MAP_SIZE);
3698                         } else {
3699                                 memcpy(se->discard_map, se->cur_valid_map,
3700                                                         SIT_VBLOCK_MAP_SIZE);
3701                                 sbi->discard_blks += old_valid_blocks -
3702                                                         se->valid_blocks;
3703                         }
3704                 }
3705
3706                 if (sbi->segs_per_sec > 1)
3707                         get_sec_entry(sbi, start)->valid_blocks +=
3708                                 se->valid_blocks - old_valid_blocks;
3709         }
3710         up_read(&curseg->journal_rwsem);
3711         return err;
3712 }
3713
3714 static void init_free_segmap(struct f2fs_sb_info *sbi)
3715 {
3716         unsigned int start;
3717         int type;
3718
3719         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3720                 struct seg_entry *sentry = get_seg_entry(sbi, start);
3721                 if (!sentry->valid_blocks)
3722                         __set_free(sbi, start);
3723                 else
3724                         SIT_I(sbi)->written_valid_blocks +=
3725                                                 sentry->valid_blocks;
3726         }
3727
3728         /* set use the current segments */
3729         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3730                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3731                 __set_test_and_inuse(sbi, curseg_t->segno);
3732         }
3733 }
3734
3735 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3736 {
3737         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3738         struct free_segmap_info *free_i = FREE_I(sbi);
3739         unsigned int segno = 0, offset = 0;
3740         unsigned short valid_blocks;
3741
3742         while (1) {
3743                 /* find dirty segment based on free segmap */
3744                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3745                 if (segno >= MAIN_SEGS(sbi))
3746                         break;
3747                 offset = segno + 1;
3748                 valid_blocks = get_valid_blocks(sbi, segno, false);
3749                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
3750                         continue;
3751                 if (valid_blocks > sbi->blocks_per_seg) {
3752                         f2fs_bug_on(sbi, 1);
3753                         continue;
3754                 }
3755                 mutex_lock(&dirty_i->seglist_lock);
3756                 __locate_dirty_segment(sbi, segno, DIRTY);
3757                 mutex_unlock(&dirty_i->seglist_lock);
3758         }
3759 }
3760
3761 static int init_victim_secmap(struct f2fs_sb_info *sbi)
3762 {
3763         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3764         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3765
3766         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
3767         if (!dirty_i->victim_secmap)
3768                 return -ENOMEM;
3769         return 0;
3770 }
3771
3772 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3773 {
3774         struct dirty_seglist_info *dirty_i;
3775         unsigned int bitmap_size, i;
3776
3777         /* allocate memory for dirty segments list information */
3778         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
3779                                                                 GFP_KERNEL);
3780         if (!dirty_i)
3781                 return -ENOMEM;
3782
3783         SM_I(sbi)->dirty_info = dirty_i;
3784         mutex_init(&dirty_i->seglist_lock);
3785
3786         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3787
3788         for (i = 0; i < NR_DIRTY_TYPE; i++) {
3789                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
3790                                                                 GFP_KERNEL);
3791                 if (!dirty_i->dirty_segmap[i])
3792                         return -ENOMEM;
3793         }
3794
3795         init_dirty_segmap(sbi);
3796         return init_victim_secmap(sbi);
3797 }
3798
3799 /*
3800  * Update min, max modified time for cost-benefit GC algorithm
3801  */
3802 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3803 {
3804         struct sit_info *sit_i = SIT_I(sbi);
3805         unsigned int segno;
3806
3807         down_write(&sit_i->sentry_lock);
3808
3809         sit_i->min_mtime = LLONG_MAX;
3810
3811         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
3812                 unsigned int i;
3813                 unsigned long long mtime = 0;
3814
3815                 for (i = 0; i < sbi->segs_per_sec; i++)
3816                         mtime += get_seg_entry(sbi, segno + i)->mtime;
3817
3818                 mtime = div_u64(mtime, sbi->segs_per_sec);
3819
3820                 if (sit_i->min_mtime > mtime)
3821                         sit_i->min_mtime = mtime;
3822         }
3823         sit_i->max_mtime = get_mtime(sbi);
3824         up_write(&sit_i->sentry_lock);
3825 }
3826
3827 int build_segment_manager(struct f2fs_sb_info *sbi)
3828 {
3829         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3830         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3831         struct f2fs_sm_info *sm_info;
3832         int err;
3833
3834         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
3835         if (!sm_info)
3836                 return -ENOMEM;
3837
3838         /* init sm info */
3839         sbi->sm_info = sm_info;
3840         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3841         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3842         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3843         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3844         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3845         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3846         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
3847         sm_info->rec_prefree_segments = sm_info->main_segments *
3848                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
3849         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3850                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3851
3852         if (!test_opt(sbi, LFS))
3853                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
3854         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
3855         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
3856         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
3857         sm_info->min_ssr_sections = reserved_sections(sbi);
3858
3859         INIT_LIST_HEAD(&sm_info->sit_entry_set);
3860
3861         init_rwsem(&sm_info->curseg_lock);
3862
3863         if (!f2fs_readonly(sbi->sb)) {
3864                 err = create_flush_cmd_control(sbi);
3865                 if (err)
3866                         return err;
3867         }
3868
3869         err = create_discard_cmd_control(sbi);
3870         if (err)
3871                 return err;
3872
3873         err = build_sit_info(sbi);
3874         if (err)
3875                 return err;
3876         err = build_free_segmap(sbi);
3877         if (err)
3878                 return err;
3879         err = build_curseg(sbi);
3880         if (err)
3881                 return err;
3882
3883         /* reinit free segmap based on SIT */
3884         err = build_sit_entries(sbi);
3885         if (err)
3886                 return err;
3887
3888         init_free_segmap(sbi);
3889         err = build_dirty_segmap(sbi);
3890         if (err)
3891                 return err;
3892
3893         init_min_max_mtime(sbi);
3894         return 0;
3895 }
3896
3897 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3898                 enum dirty_type dirty_type)
3899 {
3900         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3901
3902         mutex_lock(&dirty_i->seglist_lock);
3903         kvfree(dirty_i->dirty_segmap[dirty_type]);
3904         dirty_i->nr_dirty[dirty_type] = 0;
3905         mutex_unlock(&dirty_i->seglist_lock);
3906 }
3907
3908 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
3909 {
3910         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3911         kvfree(dirty_i->victim_secmap);
3912 }
3913
3914 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3915 {
3916         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3917         int i;
3918
3919         if (!dirty_i)
3920                 return;
3921
3922         /* discard pre-free/dirty segments list */
3923         for (i = 0; i < NR_DIRTY_TYPE; i++)
3924                 discard_dirty_segmap(sbi, i);
3925
3926         destroy_victim_secmap(sbi);
3927         SM_I(sbi)->dirty_info = NULL;
3928         kfree(dirty_i);
3929 }
3930
3931 static void destroy_curseg(struct f2fs_sb_info *sbi)
3932 {
3933         struct curseg_info *array = SM_I(sbi)->curseg_array;
3934         int i;
3935
3936         if (!array)
3937                 return;
3938         SM_I(sbi)->curseg_array = NULL;
3939         for (i = 0; i < NR_CURSEG_TYPE; i++) {
3940                 kfree(array[i].sum_blk);
3941                 kfree(array[i].journal);
3942         }
3943         kfree(array);
3944 }
3945
3946 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3947 {
3948         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3949         if (!free_i)
3950                 return;
3951         SM_I(sbi)->free_info = NULL;
3952         kvfree(free_i->free_segmap);
3953         kvfree(free_i->free_secmap);
3954         kfree(free_i);
3955 }
3956
3957 static void destroy_sit_info(struct f2fs_sb_info *sbi)
3958 {
3959         struct sit_info *sit_i = SIT_I(sbi);
3960         unsigned int start;
3961
3962         if (!sit_i)
3963                 return;
3964
3965         if (sit_i->sentries) {
3966                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3967                         kfree(sit_i->sentries[start].cur_valid_map);
3968 #ifdef CONFIG_F2FS_CHECK_FS
3969                         kfree(sit_i->sentries[start].cur_valid_map_mir);
3970 #endif
3971                         kfree(sit_i->sentries[start].ckpt_valid_map);
3972                         kfree(sit_i->sentries[start].discard_map);
3973                 }
3974         }
3975         kfree(sit_i->tmp_map);
3976
3977         kvfree(sit_i->sentries);
3978         kvfree(sit_i->sec_entries);
3979         kvfree(sit_i->dirty_sentries_bitmap);
3980
3981         SM_I(sbi)->sit_info = NULL;
3982         kfree(sit_i->sit_bitmap);
3983 #ifdef CONFIG_F2FS_CHECK_FS
3984         kfree(sit_i->sit_bitmap_mir);
3985 #endif
3986         kfree(sit_i);
3987 }
3988
3989 void destroy_segment_manager(struct f2fs_sb_info *sbi)
3990 {
3991         struct f2fs_sm_info *sm_info = SM_I(sbi);
3992
3993         if (!sm_info)
3994                 return;
3995         destroy_flush_cmd_control(sbi, true);
3996         destroy_discard_cmd_control(sbi);
3997         destroy_dirty_segmap(sbi);
3998         destroy_curseg(sbi);
3999         destroy_free_segmap(sbi);
4000         destroy_sit_info(sbi);
4001         sbi->sm_info = NULL;
4002         kfree(sm_info);
4003 }
4004
4005 int __init create_segment_manager_caches(void)
4006 {
4007         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4008                         sizeof(struct discard_entry));
4009         if (!discard_entry_slab)
4010                 goto fail;
4011
4012         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4013                         sizeof(struct discard_cmd));
4014         if (!discard_cmd_slab)
4015                 goto destroy_discard_entry;
4016
4017         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4018                         sizeof(struct sit_entry_set));
4019         if (!sit_entry_set_slab)
4020                 goto destroy_discard_cmd;
4021
4022         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4023                         sizeof(struct inmem_pages));
4024         if (!inmem_entry_slab)
4025                 goto destroy_sit_entry_set;
4026         return 0;
4027
4028 destroy_sit_entry_set:
4029         kmem_cache_destroy(sit_entry_set_slab);
4030 destroy_discard_cmd:
4031         kmem_cache_destroy(discard_cmd_slab);
4032 destroy_discard_entry:
4033         kmem_cache_destroy(discard_entry_slab);
4034 fail:
4035         return -ENOMEM;
4036 }
4037
4038 void destroy_segment_manager_caches(void)
4039 {
4040         kmem_cache_destroy(sit_entry_set_slab);
4041         kmem_cache_destroy(discard_cmd_slab);
4042         kmem_cache_destroy(discard_entry_slab);
4043         kmem_cache_destroy(inmem_entry_slab);
4044 }