fs/f2fs/segment.c

   1 /*
   2  * fs/f2fs/segment.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/bio.h>
  14 #include <linux/blkdev.h>
  15 #include <linux/prefetch.h>
  16 #include <linux/kthread.h>
  17 #include <linux/swap.h>
  18 #include <linux/timer.h>
  19 #include <linux/freezer.h>
  20 #include <linux/sched/signal.h>
  21
  22 #include "f2fs.h"
  23 #include "segment.h"
  24 #include "node.h"
  25 #include "gc.h"
  26 #include "trace.h"
  27 #include <trace/events/f2fs.h>
  28
  29 #define __reverse_ffz(x) __reverse_ffs(~(x))
  30
  31 static struct kmem_cache *discard_entry_slab;
  32 static struct kmem_cache *discard_cmd_slab;
  33 static struct kmem_cache *sit_entry_set_slab;
  34 static struct kmem_cache *inmem_entry_slab;
  35
  36 static unsigned long __reverse_ulong(unsigned char *str)
  37 {
  38         unsigned long tmp = 0;
  39         int shift = 24, idx = 0;
  40
  41 #if BITS_PER_LONG == 64
  42         shift = 56;
  43 #endif
  44         while (shift >= 0) {
  45                 tmp |= (unsigned long)str[idx++] << shift;
  46                 shift -= BITS_PER_BYTE;
  47         }
  48         return tmp;
  49 }
  50
  51 /*
  52  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  53  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  54  */
  55 static inline unsigned long __reverse_ffs(unsigned long word)
  56 {
  57         int num = 0;
  58
  59 #if BITS_PER_LONG == 64
  60         if ((word & 0xffffffff00000000UL) == 0)
  61                 num += 32;
  62         else
  63                 word >>= 32;
  64 #endif
  65         if ((word & 0xffff0000) == 0)
  66                 num += 16;
  67         else
  68                 word >>= 16;
  69
  70         if ((word & 0xff00) == 0)
  71                 num += 8;
  72         else
  73                 word >>= 8;
  74
  75         if ((word & 0xf0) == 0)
  76                 num += 4;
  77         else
  78                 word >>= 4;
  79
  80         if ((word & 0xc) == 0)
  81                 num += 2;
  82         else
  83                 word >>= 2;
  84
  85         if ((word & 0x2) == 0)
  86                 num += 1;
  87         return num;
  88 }
  89
  90 /*
  91  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  92  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  93  * @size must be integral times of unsigned long.
  94  * Example:
  95  *                             MSB <--> LSB
  96  *   f2fs_set_bit(0, bitmap) => 1000 0000
  97  *   f2fs_set_bit(7, bitmap) => 0000 0001
  98  */
  99 static unsigned long __find_rev_next_bit(const unsigned long *addr,
 100                         unsigned long size, unsigned long offset)
 101 {
 102         const unsigned long *p = addr + BIT_WORD(offset);
 103         unsigned long result = size;
 104         unsigned long tmp;
 105
 106         if (offset >= size)
 107                 return size;
 108
 109         size -= (offset & ~(BITS_PER_LONG - 1));
 110         offset %= BITS_PER_LONG;
 111
 112         while (1) {
 113                 if (*p == 0)
 114                         goto pass;
 115
 116                 tmp = __reverse_ulong((unsigned char *)p);
 117
 118                 tmp &= ~0UL >> offset;
 119                 if (size < BITS_PER_LONG)
 120                         tmp &= (~0UL << (BITS_PER_LONG - size));
 121                 if (tmp)
 122                         goto found;
 123 pass:
 124                 if (size <= BITS_PER_LONG)
 125                         break;
 126                 size -= BITS_PER_LONG;
 127                 offset = 0;
 128                 p++;
 129         }
 130         return result;
 131 found:
 132         return result - size + __reverse_ffs(tmp);
 133 }
 134
 135 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 136                         unsigned long size, unsigned long offset)
 137 {
 138         const unsigned long *p = addr + BIT_WORD(offset);
 139         unsigned long result = size;
 140         unsigned long tmp;
 141
 142         if (offset >= size)
 143                 return size;
 144
 145         size -= (offset & ~(BITS_PER_LONG - 1));
 146         offset %= BITS_PER_LONG;
 147
 148         while (1) {
 149                 if (*p == ~0UL)
 150                         goto pass;
 151
 152                 tmp = __reverse_ulong((unsigned char *)p);
 153
 154                 if (offset)
 155                         tmp |= ~0UL << (BITS_PER_LONG - offset);
 156                 if (size < BITS_PER_LONG)
 157                         tmp |= ~0UL >> size;
 158                 if (tmp != ~0UL)
 159                         goto found;
 160 pass:
 161                 if (size <= BITS_PER_LONG)
 162                         break;
 163                 size -= BITS_PER_LONG;
 164                 offset = 0;
 165                 p++;
 166         }
 167         return result;
 168 found:
 169         return result - size + __reverse_ffz(tmp);
 170 }
 171
 172 bool need_SSR(struct f2fs_sb_info *sbi)
 173 {
 174         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 175         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 176         int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
 177
 178         if (test_opt(sbi, LFS))
 179                 return false;
 180         if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
 181                 return true;
 182
 183         return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
 184                         SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
 185 }
 186
 187 void register_inmem_page(struct inode *inode, struct page *page)
 188 {
 189         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 190         struct f2fs_inode_info *fi = F2FS_I(inode);
 191         struct inmem_pages *new;
 192
 193         f2fs_trace_pid(page);
 194
 195         set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
 196         SetPagePrivate(page);
 197
 198         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 199
 200         /* add atomic page indices to the list */
 201         new->page = page;
 202         INIT_LIST_HEAD(&new->list);
 203
 204         /* increase reference count with clean state */
 205         mutex_lock(&fi->inmem_lock);
 206         get_page(page);
 207         list_add_tail(&new->list, &fi->inmem_pages);
 208         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 209         if (list_empty(&fi->inmem_ilist))
 210                 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
 211         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 212         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 213         mutex_unlock(&fi->inmem_lock);
 214
 215         trace_f2fs_register_inmem_page(page, INMEM);
 216 }
 217
 218 static int __revoke_inmem_pages(struct inode *inode,
 219                                 struct list_head *head, bool drop, bool recover)
 220 {
 221         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 222         struct inmem_pages *cur, *tmp;
 223         int err = 0;
 224
 225         list_for_each_entry_safe(cur, tmp, head, list) {
 226                 struct page *page = cur->page;
 227
 228                 if (drop)
 229                         trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 230
 231                 lock_page(page);
 232
 233                 if (recover) {
 234                         struct dnode_of_data dn;
 235                         struct node_info ni;
 236
 237                         trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 238 retry:
 239                         set_new_dnode(&dn, inode, NULL, NULL, 0);
 240                         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
 241                         if (err) {
 242                                 if (err == -ENOMEM) {
 243                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 244                                         cond_resched();
 245                                         goto retry;
 246                                 }
 247                                 err = -EAGAIN;
 248                                 goto next;
 249                         }
 250                         get_node_info(sbi, dn.nid, &ni);
 251                         if (cur->old_addr == NEW_ADDR) {
 252                                 invalidate_blocks(sbi, dn.data_blkaddr);
 253                                 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
 254                         } else
 255                                 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 256                                         cur->old_addr, ni.version, true, true);
 257                         f2fs_put_dnode(&dn);
 258                 }
 259 next:
 260                 /* we don't need to invalidate this in the sccessful status */
 261                 if (drop || recover)
 262                         ClearPageUptodate(page);
 263                 set_page_private(page, 0);
 264                 ClearPagePrivate(page);
 265                 f2fs_put_page(page, 1);
 266
 267                 list_del(&cur->list);
 268                 kmem_cache_free(inmem_entry_slab, cur);
 269                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 270         }
 271         return err;
 272 }
 273
 274 void drop_inmem_pages_all(struct f2fs_sb_info *sbi)
 275 {
 276         struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
 277         struct inode *inode;
 278         struct f2fs_inode_info *fi;
 279 next:
 280         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 281         if (list_empty(head)) {
 282                 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 283                 return;
 284         }
 285         fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
 286         inode = igrab(&fi->vfs_inode);
 287         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 288
 289         if (inode) {
 290                 drop_inmem_pages(inode);
 291                 iput(inode);
 292         }
 293         congestion_wait(BLK_RW_ASYNC, HZ/50);
 294         cond_resched();
 295         goto next;
 296 }
 297
 298 void drop_inmem_pages(struct inode *inode)
 299 {
 300         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 301         struct f2fs_inode_info *fi = F2FS_I(inode);
 302
 303         mutex_lock(&fi->inmem_lock);
 304         __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 305         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 306         if (!list_empty(&fi->inmem_ilist))
 307                 list_del_init(&fi->inmem_ilist);
 308         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 309         mutex_unlock(&fi->inmem_lock);
 310
 311         clear_inode_flag(inode, FI_ATOMIC_FILE);
 312         clear_inode_flag(inode, FI_HOT_DATA);
 313         stat_dec_atomic_write(inode);
 314 }
 315
 316 void drop_inmem_page(struct inode *inode, struct page *page)
 317 {
 318         struct f2fs_inode_info *fi = F2FS_I(inode);
 319         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 320         struct list_head *head = &fi->inmem_pages;
 321         struct inmem_pages *cur = NULL;
 322
 323         f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
 324
 325         mutex_lock(&fi->inmem_lock);
 326         list_for_each_entry(cur, head, list) {
 327                 if (cur->page == page)
 328                         break;
 329         }
 330
 331         f2fs_bug_on(sbi, !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_set_dev(bio, 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 = blk_status_to_errno(bio->bi_status);
 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         trimmed = __wait_discard_cmd_range(sbi, &dpolicy,
2443                                         start_block, end_block);
2444         range->len = F2FS_BLK_TO_BYTES(trimmed);
2445 out:
2446         return err;
2447 }
2448
2449 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2450 {
2451         struct curseg_info *curseg = CURSEG_I(sbi, type);
2452         if (curseg->next_blkoff < sbi->blocks_per_seg)
2453                 return true;
2454         return false;
2455 }
2456
2457 int rw_hint_to_seg_type(enum rw_hint hint)
2458 {
2459         switch (hint) {
2460         case WRITE_LIFE_SHORT:
2461                 return CURSEG_HOT_DATA;
2462         case WRITE_LIFE_EXTREME:
2463                 return CURSEG_COLD_DATA;
2464         default:
2465                 return CURSEG_WARM_DATA;
2466         }
2467 }
2468
2469 /* This returns write hints for each segment type. This hints will be
2470  * passed down to block layer. There are mapping tables which depend on
2471  * the mount option 'whint_mode'.
2472  *
2473  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2474  *
2475  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2476  *
2477  * User                  F2FS                     Block
2478  * ----                  ----                     -----
2479  *                       META                     WRITE_LIFE_NOT_SET
2480  *                       HOT_NODE                 "
2481  *                       WARM_NODE                "
2482  *                       COLD_NODE                "
2483  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2484  * extension list        "                        "
2485  *
2486  * -- buffered io
2487  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2488  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2489  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2490  * WRITE_LIFE_NONE       "                        "
2491  * WRITE_LIFE_MEDIUM     "                        "
2492  * WRITE_LIFE_LONG       "                        "
2493  *
2494  * -- direct io
2495  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2496  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2497  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2498  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2499  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2500  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2501  *
2502  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2503  *
2504  * User                  F2FS                     Block
2505  * ----                  ----                     -----
2506  *                       META                     WRITE_LIFE_MEDIUM;
2507  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
2508  *                       WARM_NODE                "
2509  *                       COLD_NODE                WRITE_LIFE_NONE
2510  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2511  * extension list        "                        "
2512  *
2513  * -- buffered io
2514  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2515  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2516  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
2517  * WRITE_LIFE_NONE       "                        "
2518  * WRITE_LIFE_MEDIUM     "                        "
2519  * WRITE_LIFE_LONG       "                        "
2520  *
2521  * -- direct io
2522  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2523  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2524  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2525  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2526  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2527  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2528  */
2529
2530 enum rw_hint io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2531                                 enum page_type type, enum temp_type temp)
2532 {
2533         if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2534                 if (type == DATA) {
2535                         if (temp == WARM)
2536                                 return WRITE_LIFE_NOT_SET;
2537                         else if (temp == HOT)
2538                                 return WRITE_LIFE_SHORT;
2539                         else if (temp == COLD)
2540                                 return WRITE_LIFE_EXTREME;
2541                 } else {
2542                         return WRITE_LIFE_NOT_SET;
2543                 }
2544         } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2545                 if (type == DATA) {
2546                         if (temp == WARM)
2547                                 return WRITE_LIFE_LONG;
2548                         else if (temp == HOT)
2549                                 return WRITE_LIFE_SHORT;
2550                         else if (temp == COLD)
2551                                 return WRITE_LIFE_EXTREME;
2552                 } else if (type == NODE) {
2553                         if (temp == WARM || temp == HOT)
2554                                 return WRITE_LIFE_NOT_SET;
2555                         else if (temp == COLD)
2556                                 return WRITE_LIFE_NONE;
2557                 } else if (type == META) {
2558                         return WRITE_LIFE_MEDIUM;
2559                 }
2560         }
2561         return WRITE_LIFE_NOT_SET;
2562 }
2563
2564 static int __get_segment_type_2(struct f2fs_io_info *fio)
2565 {
2566         if (fio->type == DATA)
2567                 return CURSEG_HOT_DATA;
2568         else
2569                 return CURSEG_HOT_NODE;
2570 }
2571
2572 static int __get_segment_type_4(struct f2fs_io_info *fio)
2573 {
2574         if (fio->type == DATA) {
2575                 struct inode *inode = fio->page->mapping->host;
2576
2577                 if (S_ISDIR(inode->i_mode))
2578                         return CURSEG_HOT_DATA;
2579                 else
2580                         return CURSEG_COLD_DATA;
2581         } else {
2582                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2583                         return CURSEG_WARM_NODE;
2584                 else
2585                         return CURSEG_COLD_NODE;
2586         }
2587 }
2588
2589 static int __get_segment_type_6(struct f2fs_io_info *fio)
2590 {
2591         if (fio->type == DATA) {
2592                 struct inode *inode = fio->page->mapping->host;
2593
2594                 if (is_cold_data(fio->page) || file_is_cold(inode))
2595                         return CURSEG_COLD_DATA;
2596                 if (file_is_hot(inode) ||
2597                                 is_inode_flag_set(inode, FI_HOT_DATA))
2598                         return CURSEG_HOT_DATA;
2599                 return rw_hint_to_seg_type(inode->i_write_hint);
2600         } else {
2601                 if (IS_DNODE(fio->page))
2602                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2603                                                 CURSEG_HOT_NODE;
2604                 return CURSEG_COLD_NODE;
2605         }
2606 }
2607
2608 static int __get_segment_type(struct f2fs_io_info *fio)
2609 {
2610         int type = 0;
2611
2612         switch (F2FS_OPTION(fio->sbi).active_logs) {
2613         case 2:
2614                 type = __get_segment_type_2(fio);
2615                 break;
2616         case 4:
2617                 type = __get_segment_type_4(fio);
2618                 break;
2619         case 6:
2620                 type = __get_segment_type_6(fio);
2621                 break;
2622         default:
2623                 f2fs_bug_on(fio->sbi, true);
2624         }
2625
2626         if (IS_HOT(type))
2627                 fio->temp = HOT;
2628         else if (IS_WARM(type))
2629                 fio->temp = WARM;
2630         else
2631                 fio->temp = COLD;
2632         return type;
2633 }
2634
2635 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2636                 block_t old_blkaddr, block_t *new_blkaddr,
2637                 struct f2fs_summary *sum, int type,
2638                 struct f2fs_io_info *fio, bool add_list)
2639 {
2640         struct sit_info *sit_i = SIT_I(sbi);
2641         struct curseg_info *curseg = CURSEG_I(sbi, type);
2642
2643         down_read(&SM_I(sbi)->curseg_lock);
2644
2645         mutex_lock(&curseg->curseg_mutex);
2646         down_write(&sit_i->sentry_lock);
2647
2648         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2649
2650         f2fs_wait_discard_bio(sbi, *new_blkaddr);
2651
2652         /*
2653          * __add_sum_entry should be resided under the curseg_mutex
2654          * because, this function updates a summary entry in the
2655          * current summary block.
2656          */
2657         __add_sum_entry(sbi, type, sum);
2658
2659         __refresh_next_blkoff(sbi, curseg);
2660
2661         stat_inc_block_count(sbi, curseg);
2662
2663         /*
2664          * SIT information should be updated before segment allocation,
2665          * since SSR needs latest valid block information.
2666          */
2667         update_sit_entry(sbi, *new_blkaddr, 1);
2668         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2669                 update_sit_entry(sbi, old_blkaddr, -1);
2670
2671         if (!__has_curseg_space(sbi, type))
2672                 sit_i->s_ops->allocate_segment(sbi, type, false);
2673
2674         /*
2675          * segment dirty status should be updated after segment allocation,
2676          * so we just need to update status only one time after previous
2677          * segment being closed.
2678          */
2679         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2680         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
2681
2682         up_write(&sit_i->sentry_lock);
2683
2684         if (page && IS_NODESEG(type)) {
2685                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2686
2687                 f2fs_inode_chksum_set(sbi, page);
2688         }
2689
2690         if (add_list) {
2691                 struct f2fs_bio_info *io;
2692
2693                 INIT_LIST_HEAD(&fio->list);
2694                 fio->in_list = true;
2695                 io = sbi->write_io[fio->type] + fio->temp;
2696                 spin_lock(&io->io_lock);
2697                 list_add_tail(&fio->list, &io->io_list);
2698                 spin_unlock(&io->io_lock);
2699         }
2700
2701         mutex_unlock(&curseg->curseg_mutex);
2702
2703         up_read(&SM_I(sbi)->curseg_lock);
2704 }
2705
2706 static void update_device_state(struct f2fs_io_info *fio)
2707 {
2708         struct f2fs_sb_info *sbi = fio->sbi;
2709         unsigned int devidx;
2710
2711         if (!sbi->s_ndevs)
2712                 return;
2713
2714         devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2715
2716         /* update device state for fsync */
2717         set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
2718
2719         /* update device state for checkpoint */
2720         if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2721                 spin_lock(&sbi->dev_lock);
2722                 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2723                 spin_unlock(&sbi->dev_lock);
2724         }
2725 }
2726
2727 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2728 {
2729         int type = __get_segment_type(fio);
2730         int err;
2731
2732 reallocate:
2733         allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
2734                         &fio->new_blkaddr, sum, type, fio, true);
2735
2736         /* writeout dirty page into bdev */
2737         err = f2fs_submit_page_write(fio);
2738         if (err == -EAGAIN) {
2739                 fio->old_blkaddr = fio->new_blkaddr;
2740                 goto reallocate;
2741         } else if (!err) {
2742                 update_device_state(fio);
2743         }
2744 }
2745
2746 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2747                                         enum iostat_type io_type)
2748 {
2749         struct f2fs_io_info fio = {
2750                 .sbi = sbi,
2751                 .type = META,
2752                 .temp = HOT,
2753                 .op = REQ_OP_WRITE,
2754                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
2755                 .old_blkaddr = page->index,
2756                 .new_blkaddr = page->index,
2757                 .page = page,
2758                 .encrypted_page = NULL,
2759                 .in_list = false,
2760         };
2761
2762         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
2763                 fio.op_flags &= ~REQ_META;
2764
2765         set_page_writeback(page);
2766         ClearPageError(page);
2767         f2fs_submit_page_write(&fio);
2768
2769         f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
2770 }
2771
2772 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
2773 {
2774         struct f2fs_summary sum;
2775
2776         set_summary(&sum, nid, 0, 0);
2777         do_write_page(&sum, fio);
2778
2779         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2780 }
2781
2782 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
2783 {
2784         struct f2fs_sb_info *sbi = fio->sbi;
2785         struct f2fs_summary sum;
2786         struct node_info ni;
2787
2788         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
2789         get_node_info(sbi, dn->nid, &ni);
2790         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
2791         do_write_page(&sum, fio);
2792         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
2793
2794         f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
2795 }
2796
2797 int rewrite_data_page(struct f2fs_io_info *fio)
2798 {
2799         int err;
2800         struct f2fs_sb_info *sbi = fio->sbi;
2801
2802         fio->new_blkaddr = fio->old_blkaddr;
2803         /* i/o temperature is needed for passing down write hints */
2804         __get_segment_type(fio);
2805
2806         f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
2807                         GET_SEGNO(sbi, fio->new_blkaddr))->type));
2808
2809         stat_inc_inplace_blocks(fio->sbi);
2810
2811         err = f2fs_submit_page_bio(fio);
2812         if (!err)
2813                 update_device_state(fio);
2814
2815         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2816
2817         return err;
2818 }
2819
2820 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
2821                                                 unsigned int segno)
2822 {
2823         int i;
2824
2825         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
2826                 if (CURSEG_I(sbi, i)->segno == segno)
2827                         break;
2828         }
2829         return i;
2830 }
2831
2832 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
2833                                 block_t old_blkaddr, block_t new_blkaddr,
2834                                 bool recover_curseg, bool recover_newaddr)
2835 {
2836         struct sit_info *sit_i = SIT_I(sbi);
2837         struct curseg_info *curseg;
2838         unsigned int segno, old_cursegno;
2839         struct seg_entry *se;
2840         int type;
2841         unsigned short old_blkoff;
2842
2843         segno = GET_SEGNO(sbi, new_blkaddr);
2844         se = get_seg_entry(sbi, segno);
2845         type = se->type;
2846
2847         down_write(&SM_I(sbi)->curseg_lock);
2848
2849         if (!recover_curseg) {
2850                 /* for recovery flow */
2851                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2852                         if (old_blkaddr == NULL_ADDR)
2853                                 type = CURSEG_COLD_DATA;
2854                         else
2855                                 type = CURSEG_WARM_DATA;
2856                 }
2857         } else {
2858                 if (IS_CURSEG(sbi, segno)) {
2859                         /* se->type is volatile as SSR allocation */
2860                         type = __f2fs_get_curseg(sbi, segno);
2861                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
2862                 } else {
2863                         type = CURSEG_WARM_DATA;
2864                 }
2865         }
2866
2867         f2fs_bug_on(sbi, !IS_DATASEG(type));
2868         curseg = CURSEG_I(sbi, type);
2869
2870         mutex_lock(&curseg->curseg_mutex);
2871         down_write(&sit_i->sentry_lock);
2872
2873         old_cursegno = curseg->segno;
2874         old_blkoff = curseg->next_blkoff;
2875
2876         /* change the current segment */
2877         if (segno != curseg->segno) {
2878                 curseg->next_segno = segno;
2879                 change_curseg(sbi, type);
2880         }
2881
2882         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
2883         __add_sum_entry(sbi, type, sum);
2884
2885         if (!recover_curseg || recover_newaddr)
2886                 update_sit_entry(sbi, new_blkaddr, 1);
2887         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2888                 update_sit_entry(sbi, old_blkaddr, -1);
2889
2890         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2891         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2892
2893         locate_dirty_segment(sbi, old_cursegno);
2894
2895         if (recover_curseg) {
2896                 if (old_cursegno != curseg->segno) {
2897                         curseg->next_segno = old_cursegno;
2898                         change_curseg(sbi, type);
2899                 }
2900                 curseg->next_blkoff = old_blkoff;
2901         }
2902
2903         up_write(&sit_i->sentry_lock);
2904         mutex_unlock(&curseg->curseg_mutex);
2905         up_write(&SM_I(sbi)->curseg_lock);
2906 }
2907
2908 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2909                                 block_t old_addr, block_t new_addr,
2910                                 unsigned char version, bool recover_curseg,
2911                                 bool recover_newaddr)
2912 {
2913         struct f2fs_summary sum;
2914
2915         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2916
2917         __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2918                                         recover_curseg, recover_newaddr);
2919
2920         f2fs_update_data_blkaddr(dn, new_addr);
2921 }
2922
2923 void f2fs_wait_on_page_writeback(struct page *page,
2924                                 enum page_type type, bool ordered)
2925 {
2926         if (PageWriteback(page)) {
2927                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2928
2929                 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2930                                                 0, page->index, type);
2931                 if (ordered)
2932                         wait_on_page_writeback(page);
2933                 else
2934                         wait_for_stable_page(page);
2935         }
2936 }
2937
2938 void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
2939 {
2940         struct page *cpage;
2941
2942         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
2943                 return;
2944
2945         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2946         if (cpage) {
2947                 f2fs_wait_on_page_writeback(cpage, DATA, true);
2948                 f2fs_put_page(cpage, 1);
2949         }
2950 }
2951
2952 static void read_compacted_summaries(struct f2fs_sb_info *sbi)
2953 {
2954         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2955         struct curseg_info *seg_i;
2956         unsigned char *kaddr;
2957         struct page *page;
2958         block_t start;
2959         int i, j, offset;
2960
2961         start = start_sum_block(sbi);
2962
2963         page = get_meta_page(sbi, start++);
2964         kaddr = (unsigned char *)page_address(page);
2965
2966         /* Step 1: restore nat cache */
2967         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2968         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
2969
2970         /* Step 2: restore sit cache */
2971         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2972         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
2973         offset = 2 * SUM_JOURNAL_SIZE;
2974
2975         /* Step 3: restore summary entries */
2976         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2977                 unsigned short blk_off;
2978                 unsigned int segno;
2979
2980                 seg_i = CURSEG_I(sbi, i);
2981                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2982                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2983                 seg_i->next_segno = segno;
2984                 reset_curseg(sbi, i, 0);
2985                 seg_i->alloc_type = ckpt->alloc_type[i];
2986                 seg_i->next_blkoff = blk_off;
2987
2988                 if (seg_i->alloc_type == SSR)
2989                         blk_off = sbi->blocks_per_seg;
2990
2991                 for (j = 0; j < blk_off; j++) {
2992                         struct f2fs_summary *s;
2993                         s = (struct f2fs_summary *)(kaddr + offset);
2994                         seg_i->sum_blk->entries[j] = *s;
2995                         offset += SUMMARY_SIZE;
2996                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
2997                                                 SUM_FOOTER_SIZE)
2998                                 continue;
2999
3000                         f2fs_put_page(page, 1);
3001                         page = NULL;
3002
3003                         page = get_meta_page(sbi, start++);
3004                         kaddr = (unsigned char *)page_address(page);
3005                         offset = 0;
3006                 }
3007         }
3008         f2fs_put_page(page, 1);
3009 }
3010
3011 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3012 {
3013         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3014         struct f2fs_summary_block *sum;
3015         struct curseg_info *curseg;
3016         struct page *new;
3017         unsigned short blk_off;
3018         unsigned int segno = 0;
3019         block_t blk_addr = 0;
3020
3021         /* get segment number and block addr */
3022         if (IS_DATASEG(type)) {
3023                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3024                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3025                                                         CURSEG_HOT_DATA]);
3026                 if (__exist_node_summaries(sbi))
3027                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3028                 else
3029                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3030         } else {
3031                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3032                                                         CURSEG_HOT_NODE]);
3033                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3034                                                         CURSEG_HOT_NODE]);
3035                 if (__exist_node_summaries(sbi))
3036                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3037                                                         type - CURSEG_HOT_NODE);
3038                 else
3039                         blk_addr = GET_SUM_BLOCK(sbi, segno);
3040         }
3041
3042         new = get_meta_page(sbi, blk_addr);
3043         sum = (struct f2fs_summary_block *)page_address(new);
3044
3045         if (IS_NODESEG(type)) {
3046                 if (__exist_node_summaries(sbi)) {
3047                         struct f2fs_summary *ns = &sum->entries[0];
3048                         int i;
3049                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3050                                 ns->version = 0;
3051                                 ns->ofs_in_node = 0;
3052                         }
3053                 } else {
3054                         restore_node_summary(sbi, segno, sum);
3055                 }
3056         }
3057
3058         /* set uncompleted segment to curseg */
3059         curseg = CURSEG_I(sbi, type);
3060         mutex_lock(&curseg->curseg_mutex);
3061
3062         /* update journal info */
3063         down_write(&curseg->journal_rwsem);
3064         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3065         up_write(&curseg->journal_rwsem);
3066
3067         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3068         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3069         curseg->next_segno = segno;
3070         reset_curseg(sbi, type, 0);
3071         curseg->alloc_type = ckpt->alloc_type[type];
3072         curseg->next_blkoff = blk_off;
3073         mutex_unlock(&curseg->curseg_mutex);
3074         f2fs_put_page(new, 1);
3075         return 0;
3076 }
3077
3078 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3079 {
3080         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3081         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3082         int type = CURSEG_HOT_DATA;
3083         int err;
3084
3085         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3086                 int npages = npages_for_summary_flush(sbi, true);
3087
3088                 if (npages >= 2)
3089                         ra_meta_pages(sbi, start_sum_block(sbi), npages,
3090                                                         META_CP, true);
3091
3092                 /* restore for compacted data summary */
3093                 read_compacted_summaries(sbi);
3094                 type = CURSEG_HOT_NODE;
3095         }
3096
3097         if (__exist_node_summaries(sbi))
3098                 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3099                                         NR_CURSEG_TYPE - type, META_CP, true);
3100
3101         for (; type <= CURSEG_COLD_NODE; type++) {
3102                 err = read_normal_summaries(sbi, type);
3103                 if (err)
3104                         return err;
3105         }
3106
3107         /* sanity check for summary blocks */
3108         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3109                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3110                 return -EINVAL;
3111
3112         return 0;
3113 }
3114
3115 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3116 {
3117         struct page *page;
3118         unsigned char *kaddr;
3119         struct f2fs_summary *summary;
3120         struct curseg_info *seg_i;
3121         int written_size = 0;
3122         int i, j;
3123
3124         page = grab_meta_page(sbi, blkaddr++);
3125         kaddr = (unsigned char *)page_address(page);
3126
3127         /* Step 1: write nat cache */
3128         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3129         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3130         written_size += SUM_JOURNAL_SIZE;
3131
3132         /* Step 2: write sit cache */
3133         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3134         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3135         written_size += SUM_JOURNAL_SIZE;
3136
3137         /* Step 3: write summary entries */
3138         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3139                 unsigned short blkoff;
3140                 seg_i = CURSEG_I(sbi, i);
3141                 if (sbi->ckpt->alloc_type[i] == SSR)
3142                         blkoff = sbi->blocks_per_seg;
3143                 else
3144                         blkoff = curseg_blkoff(sbi, i);
3145
3146                 for (j = 0; j < blkoff; j++) {
3147                         if (!page) {
3148                                 page = grab_meta_page(sbi, blkaddr++);
3149                                 kaddr = (unsigned char *)page_address(page);
3150                                 written_size = 0;
3151                         }
3152                         summary = (struct f2fs_summary *)(kaddr + written_size);
3153                         *summary = seg_i->sum_blk->entries[j];
3154                         written_size += SUMMARY_SIZE;
3155
3156                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3157                                                         SUM_FOOTER_SIZE)
3158                                 continue;
3159
3160                         set_page_dirty(page);
3161                         f2fs_put_page(page, 1);
3162                         page = NULL;
3163                 }
3164         }
3165         if (page) {
3166                 set_page_dirty(page);
3167                 f2fs_put_page(page, 1);
3168         }
3169 }
3170
3171 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3172                                         block_t blkaddr, int type)
3173 {
3174         int i, end;
3175         if (IS_DATASEG(type))
3176                 end = type + NR_CURSEG_DATA_TYPE;
3177         else
3178                 end = type + NR_CURSEG_NODE_TYPE;
3179
3180         for (i = type; i < end; i++)
3181                 write_current_sum_page(sbi, i, blkaddr + (i - type));
3182 }
3183
3184 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3185 {
3186         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3187                 write_compacted_summaries(sbi, start_blk);
3188         else
3189                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3190 }
3191
3192 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3193 {
3194         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3195 }
3196
3197 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3198                                         unsigned int val, int alloc)
3199 {
3200         int i;
3201
3202         if (type == NAT_JOURNAL) {
3203                 for (i = 0; i < nats_in_cursum(journal); i++) {
3204                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3205                                 return i;
3206                 }
3207                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3208                         return update_nats_in_cursum(journal, 1);
3209         } else if (type == SIT_JOURNAL) {
3210                 for (i = 0; i < sits_in_cursum(journal); i++)
3211                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3212                                 return i;
3213                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3214                         return update_sits_in_cursum(journal, 1);
3215         }
3216         return -1;
3217 }
3218
3219 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3220                                         unsigned int segno)
3221 {
3222         return get_meta_page(sbi, current_sit_addr(sbi, segno));
3223 }
3224
3225 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3226                                         unsigned int start)
3227 {
3228         struct sit_info *sit_i = SIT_I(sbi);
3229         struct page *page;
3230         pgoff_t src_off, dst_off;
3231
3232         src_off = current_sit_addr(sbi, start);
3233         dst_off = next_sit_addr(sbi, src_off);
3234
3235         page = grab_meta_page(sbi, dst_off);
3236         seg_info_to_sit_page(sbi, page, start);
3237
3238         set_page_dirty(page);
3239         set_to_next_sit(sit_i, start);
3240
3241         return page;
3242 }
3243
3244 static struct sit_entry_set *grab_sit_entry_set(void)
3245 {
3246         struct sit_entry_set *ses =
3247                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3248
3249         ses->entry_cnt = 0;
3250         INIT_LIST_HEAD(&ses->set_list);
3251         return ses;
3252 }
3253
3254 static void release_sit_entry_set(struct sit_entry_set *ses)
3255 {
3256         list_del(&ses->set_list);
3257         kmem_cache_free(sit_entry_set_slab, ses);
3258 }
3259
3260 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3261                                                 struct list_head *head)
3262 {
3263         struct sit_entry_set *next = ses;
3264
3265         if (list_is_last(&ses->set_list, head))
3266                 return;
3267
3268         list_for_each_entry_continue(next, head, set_list)
3269                 if (ses->entry_cnt <= next->entry_cnt)
3270                         break;
3271
3272         list_move_tail(&ses->set_list, &next->set_list);
3273 }
3274
3275 static void add_sit_entry(unsigned int segno, struct list_head *head)
3276 {
3277         struct sit_entry_set *ses;
3278         unsigned int start_segno = START_SEGNO(segno);
3279
3280         list_for_each_entry(ses, head, set_list) {
3281                 if (ses->start_segno == start_segno) {
3282                         ses->entry_cnt++;
3283                         adjust_sit_entry_set(ses, head);
3284                         return;
3285                 }
3286         }
3287
3288         ses = grab_sit_entry_set();
3289
3290         ses->start_segno = start_segno;
3291         ses->entry_cnt++;
3292         list_add(&ses->set_list, head);
3293 }
3294
3295 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3296 {
3297         struct f2fs_sm_info *sm_info = SM_I(sbi);
3298         struct list_head *set_list = &sm_info->sit_entry_set;
3299         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3300         unsigned int segno;
3301
3302         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3303                 add_sit_entry(segno, set_list);
3304 }
3305
3306 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3307 {
3308         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3309         struct f2fs_journal *journal = curseg->journal;
3310         int i;
3311
3312         down_write(&curseg->journal_rwsem);
3313         for (i = 0; i < sits_in_cursum(journal); i++) {
3314                 unsigned int segno;
3315                 bool dirtied;
3316
3317                 segno = le32_to_cpu(segno_in_journal(journal, i));
3318                 dirtied = __mark_sit_entry_dirty(sbi, segno);
3319
3320                 if (!dirtied)
3321                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3322         }
3323         update_sits_in_cursum(journal, -i);
3324         up_write(&curseg->journal_rwsem);
3325 }
3326
3327 /*
3328  * CP calls this function, which flushes SIT entries including sit_journal,
3329  * and moves prefree segs to free segs.
3330  */
3331 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3332 {
3333         struct sit_info *sit_i = SIT_I(sbi);
3334         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3335         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3336         struct f2fs_journal *journal = curseg->journal;
3337         struct sit_entry_set *ses, *tmp;
3338         struct list_head *head = &SM_I(sbi)->sit_entry_set;
3339         bool to_journal = true;
3340         struct seg_entry *se;
3341
3342         down_write(&sit_i->sentry_lock);
3343
3344         if (!sit_i->dirty_sentries)
3345                 goto out;
3346
3347         /*
3348          * add and account sit entries of dirty bitmap in sit entry
3349          * set temporarily
3350          */
3351         add_sits_in_set(sbi);
3352
3353         /*
3354          * if there are no enough space in journal to store dirty sit
3355          * entries, remove all entries from journal and add and account
3356          * them in sit entry set.
3357          */
3358         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3359                 remove_sits_in_journal(sbi);
3360
3361         /*
3362          * there are two steps to flush sit entries:
3363          * #1, flush sit entries to journal in current cold data summary block.
3364          * #2, flush sit entries to sit page.
3365          */
3366         list_for_each_entry_safe(ses, tmp, head, set_list) {
3367                 struct page *page = NULL;
3368                 struct f2fs_sit_block *raw_sit = NULL;
3369                 unsigned int start_segno = ses->start_segno;
3370                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3371                                                 (unsigned long)MAIN_SEGS(sbi));
3372                 unsigned int segno = start_segno;
3373
3374                 if (to_journal &&
3375                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3376                         to_journal = false;
3377
3378                 if (to_journal) {
3379                         down_write(&curseg->journal_rwsem);
3380                 } else {
3381                         page = get_next_sit_page(sbi, start_segno);
3382                         raw_sit = page_address(page);
3383                 }
3384
3385                 /* flush dirty sit entries in region of current sit set */
3386                 for_each_set_bit_from(segno, bitmap, end) {
3387                         int offset, sit_offset;
3388
3389                         se = get_seg_entry(sbi, segno);
3390
3391                         /* add discard candidates */
3392                         if (!(cpc->reason & CP_DISCARD)) {
3393                                 cpc->trim_start = segno;
3394                                 add_discard_addrs(sbi, cpc, false);
3395                         }
3396
3397                         if (to_journal) {
3398                                 offset = lookup_journal_in_cursum(journal,
3399                                                         SIT_JOURNAL, segno, 1);
3400                                 f2fs_bug_on(sbi, offset < 0);
3401                                 segno_in_journal(journal, offset) =
3402                                                         cpu_to_le32(segno);
3403                                 seg_info_to_raw_sit(se,
3404                                         &sit_in_journal(journal, offset));
3405                         } else {
3406                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3407                                 seg_info_to_raw_sit(se,
3408                                                 &raw_sit->entries[sit_offset]);
3409                         }
3410
3411                         __clear_bit(segno, bitmap);
3412                         sit_i->dirty_sentries--;
3413                         ses->entry_cnt--;
3414                 }
3415
3416                 if (to_journal)
3417                         up_write(&curseg->journal_rwsem);
3418                 else
3419                         f2fs_put_page(page, 1);
3420
3421                 f2fs_bug_on(sbi, ses->entry_cnt);
3422                 release_sit_entry_set(ses);
3423         }
3424
3425         f2fs_bug_on(sbi, !list_empty(head));
3426         f2fs_bug_on(sbi, sit_i->dirty_sentries);
3427 out:
3428         if (cpc->reason & CP_DISCARD) {
3429                 __u64 trim_start = cpc->trim_start;
3430
3431                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3432                         add_discard_addrs(sbi, cpc, false);
3433
3434                 cpc->trim_start = trim_start;
3435         }
3436         up_write(&sit_i->sentry_lock);
3437
3438         set_prefree_as_free_segments(sbi);
3439 }
3440
3441 static int build_sit_info(struct f2fs_sb_info *sbi)
3442 {
3443         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3444         struct sit_info *sit_i;
3445         unsigned int sit_segs, start;
3446         char *src_bitmap;
3447         unsigned int bitmap_size;
3448
3449         /* allocate memory for SIT information */
3450         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3451         if (!sit_i)
3452                 return -ENOMEM;
3453
3454         SM_I(sbi)->sit_info = sit_i;
3455
3456         sit_i->sentries = f2fs_kvzalloc(sbi, MAIN_SEGS(sbi) *
3457                                         sizeof(struct seg_entry), GFP_KERNEL);
3458         if (!sit_i->sentries)
3459                 return -ENOMEM;
3460
3461         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3462         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3463                                                                 GFP_KERNEL);
3464         if (!sit_i->dirty_sentries_bitmap)
3465                 return -ENOMEM;
3466
3467         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3468                 sit_i->sentries[start].cur_valid_map
3469                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3470                 sit_i->sentries[start].ckpt_valid_map
3471                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3472                 if (!sit_i->sentries[start].cur_valid_map ||
3473                                 !sit_i->sentries[start].ckpt_valid_map)
3474                         return -ENOMEM;
3475
3476 #ifdef CONFIG_F2FS_CHECK_FS
3477                 sit_i->sentries[start].cur_valid_map_mir
3478                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3479                 if (!sit_i->sentries[start].cur_valid_map_mir)
3480                         return -ENOMEM;
3481 #endif
3482
3483                 if (f2fs_discard_en(sbi)) {
3484                         sit_i->sentries[start].discard_map
3485                                 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3486                                                                 GFP_KERNEL);
3487                         if (!sit_i->sentries[start].discard_map)
3488                                 return -ENOMEM;
3489                 }
3490         }
3491
3492         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3493         if (!sit_i->tmp_map)
3494                 return -ENOMEM;
3495
3496         if (sbi->segs_per_sec > 1) {
3497                 sit_i->sec_entries = f2fs_kvzalloc(sbi, MAIN_SECS(sbi) *
3498                                         sizeof(struct sec_entry), GFP_KERNEL);
3499                 if (!sit_i->sec_entries)
3500                         return -ENOMEM;
3501         }
3502
3503         /* get information related with SIT */
3504         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3505
3506         /* setup SIT bitmap from ckeckpoint pack */
3507         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3508         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3509
3510         sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3511         if (!sit_i->sit_bitmap)
3512                 return -ENOMEM;
3513
3514 #ifdef CONFIG_F2FS_CHECK_FS
3515         sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3516         if (!sit_i->sit_bitmap_mir)
3517                 return -ENOMEM;
3518 #endif
3519
3520         /* init SIT information */
3521         sit_i->s_ops = &default_salloc_ops;
3522
3523         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3524         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
3525         sit_i->written_valid_blocks = 0;
3526         sit_i->bitmap_size = bitmap_size;
3527         sit_i->dirty_sentries = 0;
3528         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3529         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3530         sit_i->mounted_time = ktime_get_real_seconds();
3531         init_rwsem(&sit_i->sentry_lock);
3532         return 0;
3533 }
3534
3535 static int build_free_segmap(struct f2fs_sb_info *sbi)
3536 {
3537         struct free_segmap_info *free_i;
3538         unsigned int bitmap_size, sec_bitmap_size;
3539
3540         /* allocate memory for free segmap information */
3541         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
3542         if (!free_i)
3543                 return -ENOMEM;
3544
3545         SM_I(sbi)->free_info = free_i;
3546
3547         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3548         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
3549         if (!free_i->free_segmap)
3550                 return -ENOMEM;
3551
3552         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3553         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
3554         if (!free_i->free_secmap)
3555                 return -ENOMEM;
3556
3557         /* set all segments as dirty temporarily */
3558         memset(free_i->free_segmap, 0xff, bitmap_size);
3559         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3560
3561         /* init free segmap information */
3562         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
3563         free_i->free_segments = 0;
3564         free_i->free_sections = 0;
3565         spin_lock_init(&free_i->segmap_lock);
3566         return 0;
3567 }
3568
3569 static int build_curseg(struct f2fs_sb_info *sbi)
3570 {
3571         struct curseg_info *array;
3572         int i;
3573
3574         array = f2fs_kzalloc(sbi, sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
3575         if (!array)
3576                 return -ENOMEM;
3577
3578         SM_I(sbi)->curseg_array = array;
3579
3580         for (i = 0; i < NR_CURSEG_TYPE; i++) {
3581                 mutex_init(&array[i].curseg_mutex);
3582                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
3583                 if (!array[i].sum_blk)
3584                         return -ENOMEM;
3585                 init_rwsem(&array[i].journal_rwsem);
3586                 array[i].journal = f2fs_kzalloc(sbi,
3587                                 sizeof(struct f2fs_journal), GFP_KERNEL);
3588                 if (!array[i].journal)
3589                         return -ENOMEM;
3590                 array[i].segno = NULL_SEGNO;
3591                 array[i].next_blkoff = 0;
3592         }
3593         return restore_curseg_summaries(sbi);
3594 }
3595
3596 static int build_sit_entries(struct f2fs_sb_info *sbi)
3597 {
3598         struct sit_info *sit_i = SIT_I(sbi);
3599         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3600         struct f2fs_journal *journal = curseg->journal;
3601         struct seg_entry *se;
3602         struct f2fs_sit_entry sit;
3603         int sit_blk_cnt = SIT_BLK_CNT(sbi);
3604         unsigned int i, start, end;
3605         unsigned int readed, start_blk = 0;
3606         int err = 0;
3607
3608         do {
3609                 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3610                                                         META_SIT, true);
3611
3612                 start = start_blk * sit_i->sents_per_block;
3613                 end = (start_blk + readed) * sit_i->sents_per_block;
3614
3615                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
3616                         struct f2fs_sit_block *sit_blk;
3617                         struct page *page;
3618
3619                         se = &sit_i->sentries[start];
3620                         page = get_current_sit_page(sbi, start);
3621                         sit_blk = (struct f2fs_sit_block *)page_address(page);
3622                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3623                         f2fs_put_page(page, 1);
3624
3625                         err = check_block_count(sbi, start, &sit);
3626                         if (err)
3627                                 return err;
3628                         seg_info_from_raw_sit(se, &sit);
3629
3630                         /* build discard map only one time */
3631                         if (f2fs_discard_en(sbi)) {
3632                                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3633                                         memset(se->discard_map, 0xff,
3634                                                 SIT_VBLOCK_MAP_SIZE);
3635                                 } else {
3636                                         memcpy(se->discard_map,
3637                                                 se->cur_valid_map,
3638                                                 SIT_VBLOCK_MAP_SIZE);
3639                                         sbi->discard_blks +=
3640                                                 sbi->blocks_per_seg -
3641                                                 se->valid_blocks;
3642                                 }
3643                         }
3644
3645                         if (sbi->segs_per_sec > 1)
3646                                 get_sec_entry(sbi, start)->valid_blocks +=
3647                                                         se->valid_blocks;
3648                 }
3649                 start_blk += readed;
3650         } while (start_blk < sit_blk_cnt);
3651
3652         down_read(&curseg->journal_rwsem);
3653         for (i = 0; i < sits_in_cursum(journal); i++) {
3654                 unsigned int old_valid_blocks;
3655
3656                 start = le32_to_cpu(segno_in_journal(journal, i));
3657                 se = &sit_i->sentries[start];
3658                 sit = sit_in_journal(journal, i);
3659
3660                 old_valid_blocks = se->valid_blocks;
3661
3662                 err = check_block_count(sbi, start, &sit);
3663                 if (err)
3664                         break;
3665                 seg_info_from_raw_sit(se, &sit);
3666
3667                 if (f2fs_discard_en(sbi)) {
3668                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3669                                 memset(se->discard_map, 0xff,
3670                                                         SIT_VBLOCK_MAP_SIZE);
3671                         } else {
3672                                 memcpy(se->discard_map, se->cur_valid_map,
3673                                                         SIT_VBLOCK_MAP_SIZE);
3674                                 sbi->discard_blks += old_valid_blocks -
3675                                                         se->valid_blocks;
3676                         }
3677                 }
3678
3679                 if (sbi->segs_per_sec > 1)
3680                         get_sec_entry(sbi, start)->valid_blocks +=
3681                                 se->valid_blocks - old_valid_blocks;
3682         }
3683         up_read(&curseg->journal_rwsem);
3684         return err;
3685 }
3686
3687 static void init_free_segmap(struct f2fs_sb_info *sbi)
3688 {
3689         unsigned int start;
3690         int type;
3691
3692         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3693                 struct seg_entry *sentry = get_seg_entry(sbi, start);
3694                 if (!sentry->valid_blocks)
3695                         __set_free(sbi, start);
3696                 else
3697                         SIT_I(sbi)->written_valid_blocks +=
3698                                                 sentry->valid_blocks;
3699         }
3700
3701         /* set use the current segments */
3702         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3703                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3704                 __set_test_and_inuse(sbi, curseg_t->segno);
3705         }
3706 }
3707
3708 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3709 {
3710         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3711         struct free_segmap_info *free_i = FREE_I(sbi);
3712         unsigned int segno = 0, offset = 0;
3713         unsigned short valid_blocks;
3714
3715         while (1) {
3716                 /* find dirty segment based on free segmap */
3717                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3718                 if (segno >= MAIN_SEGS(sbi))
3719                         break;
3720                 offset = segno + 1;
3721                 valid_blocks = get_valid_blocks(sbi, segno, false);
3722                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
3723                         continue;
3724                 if (valid_blocks > sbi->blocks_per_seg) {
3725                         f2fs_bug_on(sbi, 1);
3726                         continue;
3727                 }
3728                 mutex_lock(&dirty_i->seglist_lock);
3729                 __locate_dirty_segment(sbi, segno, DIRTY);
3730                 mutex_unlock(&dirty_i->seglist_lock);
3731         }
3732 }
3733
3734 static int init_victim_secmap(struct f2fs_sb_info *sbi)
3735 {
3736         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3737         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3738
3739         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
3740         if (!dirty_i->victim_secmap)
3741                 return -ENOMEM;
3742         return 0;
3743 }
3744
3745 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3746 {
3747         struct dirty_seglist_info *dirty_i;
3748         unsigned int bitmap_size, i;
3749
3750         /* allocate memory for dirty segments list information */
3751         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
3752                                                                 GFP_KERNEL);
3753         if (!dirty_i)
3754                 return -ENOMEM;
3755
3756         SM_I(sbi)->dirty_info = dirty_i;
3757         mutex_init(&dirty_i->seglist_lock);
3758
3759         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3760
3761         for (i = 0; i < NR_DIRTY_TYPE; i++) {
3762                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
3763                                                                 GFP_KERNEL);
3764                 if (!dirty_i->dirty_segmap[i])
3765                         return -ENOMEM;
3766         }
3767
3768         init_dirty_segmap(sbi);
3769         return init_victim_secmap(sbi);
3770 }
3771
3772 /*
3773  * Update min, max modified time for cost-benefit GC algorithm
3774  */
3775 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3776 {
3777         struct sit_info *sit_i = SIT_I(sbi);
3778         unsigned int segno;
3779
3780         down_write(&sit_i->sentry_lock);
3781
3782         sit_i->min_mtime = LLONG_MAX;
3783
3784         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
3785                 unsigned int i;
3786                 unsigned long long mtime = 0;
3787
3788                 for (i = 0; i < sbi->segs_per_sec; i++)
3789                         mtime += get_seg_entry(sbi, segno + i)->mtime;
3790
3791                 mtime = div_u64(mtime, sbi->segs_per_sec);
3792
3793                 if (sit_i->min_mtime > mtime)
3794                         sit_i->min_mtime = mtime;
3795         }
3796         sit_i->max_mtime = get_mtime(sbi);
3797         up_write(&sit_i->sentry_lock);
3798 }
3799
3800 int build_segment_manager(struct f2fs_sb_info *sbi)
3801 {
3802         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3803         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3804         struct f2fs_sm_info *sm_info;
3805         int err;
3806
3807         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
3808         if (!sm_info)
3809                 return -ENOMEM;
3810
3811         /* init sm info */
3812         sbi->sm_info = sm_info;
3813         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3814         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3815         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3816         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3817         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3818         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3819         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
3820         sm_info->rec_prefree_segments = sm_info->main_segments *
3821                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
3822         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3823                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3824
3825         if (!test_opt(sbi, LFS))
3826                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
3827         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
3828         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
3829         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
3830         sm_info->min_ssr_sections = reserved_sections(sbi);
3831
3832         INIT_LIST_HEAD(&sm_info->sit_entry_set);
3833
3834         init_rwsem(&sm_info->curseg_lock);
3835
3836         if (!f2fs_readonly(sbi->sb)) {
3837                 err = create_flush_cmd_control(sbi);
3838                 if (err)
3839                         return err;
3840         }
3841
3842         err = create_discard_cmd_control(sbi);
3843         if (err)
3844                 return err;
3845
3846         err = build_sit_info(sbi);
3847         if (err)
3848                 return err;
3849         err = build_free_segmap(sbi);
3850         if (err)
3851                 return err;
3852         err = build_curseg(sbi);
3853         if (err)
3854                 return err;
3855
3856         /* reinit free segmap based on SIT */
3857         err = build_sit_entries(sbi);
3858         if (err)
3859                 return err;
3860
3861         init_free_segmap(sbi);
3862         err = build_dirty_segmap(sbi);
3863         if (err)
3864                 return err;
3865
3866         init_min_max_mtime(sbi);
3867         return 0;
3868 }
3869
3870 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3871                 enum dirty_type dirty_type)
3872 {
3873         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3874
3875         mutex_lock(&dirty_i->seglist_lock);
3876         kvfree(dirty_i->dirty_segmap[dirty_type]);
3877         dirty_i->nr_dirty[dirty_type] = 0;
3878         mutex_unlock(&dirty_i->seglist_lock);
3879 }
3880
3881 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
3882 {
3883         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3884         kvfree(dirty_i->victim_secmap);
3885 }
3886
3887 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3888 {
3889         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3890         int i;
3891
3892         if (!dirty_i)
3893                 return;
3894
3895         /* discard pre-free/dirty segments list */
3896         for (i = 0; i < NR_DIRTY_TYPE; i++)
3897                 discard_dirty_segmap(sbi, i);
3898
3899         destroy_victim_secmap(sbi);
3900         SM_I(sbi)->dirty_info = NULL;
3901         kfree(dirty_i);
3902 }
3903
3904 static void destroy_curseg(struct f2fs_sb_info *sbi)
3905 {
3906         struct curseg_info *array = SM_I(sbi)->curseg_array;
3907         int i;
3908
3909         if (!array)
3910                 return;
3911         SM_I(sbi)->curseg_array = NULL;
3912         for (i = 0; i < NR_CURSEG_TYPE; i++) {
3913                 kfree(array[i].sum_blk);
3914                 kfree(array[i].journal);
3915         }
3916         kfree(array);
3917 }
3918
3919 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3920 {
3921         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3922         if (!free_i)
3923                 return;
3924         SM_I(sbi)->free_info = NULL;
3925         kvfree(free_i->free_segmap);
3926         kvfree(free_i->free_secmap);
3927         kfree(free_i);
3928 }
3929
3930 static void destroy_sit_info(struct f2fs_sb_info *sbi)
3931 {
3932         struct sit_info *sit_i = SIT_I(sbi);
3933         unsigned int start;
3934
3935         if (!sit_i)
3936                 return;
3937
3938         if (sit_i->sentries) {
3939                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3940                         kfree(sit_i->sentries[start].cur_valid_map);
3941 #ifdef CONFIG_F2FS_CHECK_FS
3942                         kfree(sit_i->sentries[start].cur_valid_map_mir);
3943 #endif
3944                         kfree(sit_i->sentries[start].ckpt_valid_map);
3945                         kfree(sit_i->sentries[start].discard_map);
3946                 }
3947         }
3948         kfree(sit_i->tmp_map);
3949
3950         kvfree(sit_i->sentries);
3951         kvfree(sit_i->sec_entries);
3952         kvfree(sit_i->dirty_sentries_bitmap);
3953
3954         SM_I(sbi)->sit_info = NULL;
3955         kfree(sit_i->sit_bitmap);
3956 #ifdef CONFIG_F2FS_CHECK_FS
3957         kfree(sit_i->sit_bitmap_mir);
3958 #endif
3959         kfree(sit_i);
3960 }
3961
3962 void destroy_segment_manager(struct f2fs_sb_info *sbi)
3963 {
3964         struct f2fs_sm_info *sm_info = SM_I(sbi);
3965
3966         if (!sm_info)
3967                 return;
3968         destroy_flush_cmd_control(sbi, true);
3969         destroy_discard_cmd_control(sbi);
3970         destroy_dirty_segmap(sbi);
3971         destroy_curseg(sbi);
3972         destroy_free_segmap(sbi);
3973         destroy_sit_info(sbi);
3974         sbi->sm_info = NULL;
3975         kfree(sm_info);
3976 }
3977
3978 int __init create_segment_manager_caches(void)
3979 {
3980         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
3981                         sizeof(struct discard_entry));
3982         if (!discard_entry_slab)
3983                 goto fail;
3984
3985         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3986                         sizeof(struct discard_cmd));
3987         if (!discard_cmd_slab)
3988                 goto destroy_discard_entry;
3989
3990         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
3991                         sizeof(struct sit_entry_set));
3992         if (!sit_entry_set_slab)
3993                 goto destroy_discard_cmd;
3994
3995         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3996                         sizeof(struct inmem_pages));
3997         if (!inmem_entry_slab)
3998                 goto destroy_sit_entry_set;
3999         return 0;
4000
4001 destroy_sit_entry_set:
4002         kmem_cache_destroy(sit_entry_set_slab);
4003 destroy_discard_cmd:
4004         kmem_cache_destroy(discard_cmd_slab);
4005 destroy_discard_entry:
4006         kmem_cache_destroy(discard_entry_slab);
4007 fail:
4008         return -ENOMEM;
4009 }
4010
4011 void destroy_segment_manager_caches(void)
4012 {
4013         kmem_cache_destroy(sit_entry_set_slab);
4014         kmem_cache_destroy(discard_cmd_slab);
4015         kmem_cache_destroy(discard_entry_slab);
4016         kmem_cache_destroy(inmem_entry_slab);
4017 }