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