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 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
 919 static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
 920                                                 struct discard_policy *dpolicy,
 921                                                 struct discard_cmd *dc)
 922 {
 923         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 924         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
 925                                         &(dcc->fstrim_list) : &(dcc->wait_list);
 926         struct bio *bio = NULL;
 927         int flag = dpolicy->sync ? REQ_SYNC : 0;
 928
 929         if (dc->state != D_PREP)
 930                 return;
 931
 932         trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
 933
 934         dc->error = __blkdev_issue_discard(dc->bdev,
 935                                 SECTOR_FROM_BLOCK(dc->start),
 936                                 SECTOR_FROM_BLOCK(dc->len),
 937                                 GFP_NOFS, 0, &bio);
 938         if (!dc->error) {
 939                 /* should keep before submission to avoid D_DONE right away */
 940                 dc->state = D_SUBMIT;
 941                 atomic_inc(&dcc->issued_discard);
 942                 atomic_inc(&dcc->issing_discard);
 943                 if (bio) {
 944                         bio->bi_private = dc;
 945                         bio->bi_end_io = f2fs_submit_discard_endio;
 946                         bio->bi_opf |= flag;
 947                         submit_bio(bio);
 948                         list_move_tail(&dc->list, wait_list);
 949                         __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
 950
 951                         f2fs_update_iostat(sbi, FS_DISCARD, 1);
 952                 }
 953         } else {
 954                 __remove_discard_cmd(sbi, dc);
 955         }
 956 }
 957
 958 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
 959                                 struct block_device *bdev, block_t lstart,
 960                                 block_t start, block_t len,
 961                                 struct rb_node **insert_p,
 962                                 struct rb_node *insert_parent)
 963 {
 964         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 965         struct rb_node **p;
 966         struct rb_node *parent = NULL;
 967         struct discard_cmd *dc = NULL;
 968
 969         if (insert_p && insert_parent) {
 970                 parent = insert_parent;
 971                 p = insert_p;
 972                 goto do_insert;
 973         }
 974
 975         p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
 976 do_insert:
 977         dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
 978         if (!dc)
 979                 return NULL;
 980
 981         return dc;
 982 }
 983
 984 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
 985                                                 struct discard_cmd *dc)
 986 {
 987         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
 988 }
 989
 990 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
 991                                 struct discard_cmd *dc, block_t blkaddr)
 992 {
 993         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 994         struct discard_info di = dc->di;
 995         bool modified = false;
 996
 997         if (dc->state == D_DONE || dc->len == 1) {
 998                 __remove_discard_cmd(sbi, dc);
 999                 return;
1000         }
1001
1002         dcc->undiscard_blks -= di.len;
1003
1004         if (blkaddr > di.lstart) {
1005                 dc->len = blkaddr - dc->lstart;
1006                 dcc->undiscard_blks += dc->len;
1007                 __relocate_discard_cmd(dcc, dc);
1008                 modified = true;
1009         }
1010
1011         if (blkaddr < di.lstart + di.len - 1) {
1012                 if (modified) {
1013                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1014                                         di.start + blkaddr + 1 - di.lstart,
1015                                         di.lstart + di.len - 1 - blkaddr,
1016                                         NULL, NULL);
1017                 } else {
1018                         dc->lstart++;
1019                         dc->len--;
1020                         dc->start++;
1021                         dcc->undiscard_blks += dc->len;
1022                         __relocate_discard_cmd(dcc, dc);
1023                 }
1024         }
1025 }
1026
1027 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1028                                 struct block_device *bdev, block_t lstart,
1029                                 block_t start, block_t len)
1030 {
1031         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1032         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1033         struct discard_cmd *dc;
1034         struct discard_info di = {0};
1035         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1036         block_t end = lstart + len;
1037
1038         mutex_lock(&dcc->cmd_lock);
1039
1040         dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1041                                         NULL, lstart,
1042                                         (struct rb_entry **)&prev_dc,
1043                                         (struct rb_entry **)&next_dc,
1044                                         &insert_p, &insert_parent, true);
1045         if (dc)
1046                 prev_dc = dc;
1047
1048         if (!prev_dc) {
1049                 di.lstart = lstart;
1050                 di.len = next_dc ? next_dc->lstart - lstart : len;
1051                 di.len = min(di.len, len);
1052                 di.start = start;
1053         }
1054
1055         while (1) {
1056                 struct rb_node *node;
1057                 bool merged = false;
1058                 struct discard_cmd *tdc = NULL;
1059
1060                 if (prev_dc) {
1061                         di.lstart = prev_dc->lstart + prev_dc->len;
1062                         if (di.lstart < lstart)
1063                                 di.lstart = lstart;
1064                         if (di.lstart >= end)
1065                                 break;
1066
1067                         if (!next_dc || next_dc->lstart > end)
1068                                 di.len = end - di.lstart;
1069                         else
1070                                 di.len = next_dc->lstart - di.lstart;
1071                         di.start = start + di.lstart - lstart;
1072                 }
1073
1074                 if (!di.len)
1075                         goto next;
1076
1077                 if (prev_dc && prev_dc->state == D_PREP &&
1078                         prev_dc->bdev == bdev &&
1079                         __is_discard_back_mergeable(&di, &prev_dc->di)) {
1080                         prev_dc->di.len += di.len;
1081                         dcc->undiscard_blks += di.len;
1082                         __relocate_discard_cmd(dcc, prev_dc);
1083                         di = prev_dc->di;
1084                         tdc = prev_dc;
1085                         merged = true;
1086                 }
1087
1088                 if (next_dc && next_dc->state == D_PREP &&
1089                         next_dc->bdev == bdev &&
1090                         __is_discard_front_mergeable(&di, &next_dc->di)) {
1091                         next_dc->di.lstart = di.lstart;
1092                         next_dc->di.len += di.len;
1093                         next_dc->di.start = di.start;
1094                         dcc->undiscard_blks += di.len;
1095                         __relocate_discard_cmd(dcc, next_dc);
1096                         if (tdc)
1097                                 __remove_discard_cmd(sbi, tdc);
1098                         merged = true;
1099                 }
1100
1101                 if (!merged) {
1102                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1103                                                         di.len, NULL, NULL);
1104                 }
1105  next:
1106                 prev_dc = next_dc;
1107                 if (!prev_dc)
1108                         break;
1109
1110                 node = rb_next(&prev_dc->rb_node);
1111                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1112         }
1113
1114         mutex_unlock(&dcc->cmd_lock);
1115 }
1116
1117 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1118                 struct block_device *bdev, block_t blkstart, block_t blklen)
1119 {
1120         block_t lblkstart = blkstart;
1121
1122         trace_f2fs_queue_discard(bdev, blkstart, blklen);
1123
1124         if (sbi->s_ndevs) {
1125                 int devi = f2fs_target_device_index(sbi, blkstart);
1126
1127                 blkstart -= FDEV(devi).start_blk;
1128         }
1129         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1130         return 0;
1131 }
1132
1133 static void __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
1134                                         struct discard_policy *dpolicy,
1135                                         unsigned int start, unsigned int end)
1136 {
1137         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1138         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1139         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1140         struct discard_cmd *dc;
1141         struct blk_plug plug;
1142         int issued;
1143
1144 next:
1145         issued = 0;
1146
1147         mutex_lock(&dcc->cmd_lock);
1148         f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1149
1150         dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1151                                         NULL, start,
1152                                         (struct rb_entry **)&prev_dc,
1153                                         (struct rb_entry **)&next_dc,
1154                                         &insert_p, &insert_parent, true);
1155         if (!dc)
1156                 dc = next_dc;
1157
1158         blk_start_plug(&plug);
1159
1160         while (dc && dc->lstart <= end) {
1161                 struct rb_node *node;
1162
1163                 if (dc->len < dpolicy->granularity)
1164                         goto skip;
1165
1166                 if (dc->state != D_PREP) {
1167                         list_move_tail(&dc->list, &dcc->fstrim_list);
1168                         goto skip;
1169                 }
1170
1171                 __submit_discard_cmd(sbi, dpolicy, dc);
1172
1173                 if (++issued >= dpolicy->max_requests) {
1174                         start = dc->lstart + dc->len;
1175
1176                         blk_finish_plug(&plug);
1177                         mutex_unlock(&dcc->cmd_lock);
1178
1179                         schedule();
1180
1181                         goto next;
1182                 }
1183 skip:
1184                 node = rb_next(&dc->rb_node);
1185                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1186
1187                 if (fatal_signal_pending(current))
1188                         break;
1189         }
1190
1191         blk_finish_plug(&plug);
1192         mutex_unlock(&dcc->cmd_lock);
1193 }
1194
1195 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1196                                         struct discard_policy *dpolicy)
1197 {
1198         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1199         struct list_head *pend_list;
1200         struct discard_cmd *dc, *tmp;
1201         struct blk_plug plug;
1202         int i, iter = 0, issued = 0;
1203         bool io_interrupted = false;
1204
1205         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1206                 if (i + 1 < dpolicy->granularity)
1207                         break;
1208                 pend_list = &dcc->pend_list[i];
1209
1210                 mutex_lock(&dcc->cmd_lock);
1211                 if (list_empty(pend_list))
1212                         goto next;
1213                 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1214                 blk_start_plug(&plug);
1215                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1216                         f2fs_bug_on(sbi, dc->state != D_PREP);
1217
1218                         if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1219                                                                 !is_idle(sbi)) {
1220                                 io_interrupted = true;
1221                                 goto skip;
1222                         }
1223
1224                         __submit_discard_cmd(sbi, dpolicy, dc);
1225                         issued++;
1226 skip:
1227                         if (++iter >= dpolicy->max_requests)
1228                                 break;
1229                 }
1230                 blk_finish_plug(&plug);
1231 next:
1232                 mutex_unlock(&dcc->cmd_lock);
1233
1234                 if (iter >= dpolicy->max_requests)
1235                         break;
1236         }
1237
1238         if (!issued && io_interrupted)
1239                 issued = -1;
1240
1241         return issued;
1242 }
1243
1244 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1245 {
1246         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1247         struct list_head *pend_list;
1248         struct discard_cmd *dc, *tmp;
1249         int i;
1250         bool dropped = false;
1251
1252         mutex_lock(&dcc->cmd_lock);
1253         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1254                 pend_list = &dcc->pend_list[i];
1255                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1256                         f2fs_bug_on(sbi, dc->state != D_PREP);
1257                         __remove_discard_cmd(sbi, dc);
1258                         dropped = true;
1259                 }
1260         }
1261         mutex_unlock(&dcc->cmd_lock);
1262
1263         return dropped;
1264 }
1265
1266 void drop_discard_cmd(struct f2fs_sb_info *sbi)
1267 {
1268         __drop_discard_cmd(sbi);
1269 }
1270
1271 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1272                                                         struct discard_cmd *dc)
1273 {
1274         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1275         unsigned int len = 0;
1276
1277         wait_for_completion_io(&dc->wait);
1278         mutex_lock(&dcc->cmd_lock);
1279         f2fs_bug_on(sbi, dc->state != D_DONE);
1280         dc->ref--;
1281         if (!dc->ref) {
1282                 if (!dc->error)
1283                         len = dc->len;
1284                 __remove_discard_cmd(sbi, dc);
1285         }
1286         mutex_unlock(&dcc->cmd_lock);
1287
1288         return len;
1289 }
1290
1291 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1292                                                 struct discard_policy *dpolicy,
1293                                                 block_t start, block_t end)
1294 {
1295         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1296         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1297                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1298         struct discard_cmd *dc, *tmp;
1299         bool need_wait;
1300         unsigned int trimmed = 0;
1301
1302 next:
1303         need_wait = false;
1304
1305         mutex_lock(&dcc->cmd_lock);
1306         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1307                 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1308                         continue;
1309                 if (dc->len < dpolicy->granularity)
1310                         continue;
1311                 if (dc->state == D_DONE && !dc->ref) {
1312                         wait_for_completion_io(&dc->wait);
1313                         if (!dc->error)
1314                                 trimmed += dc->len;
1315                         __remove_discard_cmd(sbi, dc);
1316                 } else {
1317                         dc->ref++;
1318                         need_wait = true;
1319                         break;
1320                 }
1321         }
1322         mutex_unlock(&dcc->cmd_lock);
1323
1324         if (need_wait) {
1325                 trimmed += __wait_one_discard_bio(sbi, dc);
1326                 goto next;
1327         }
1328
1329         return trimmed;
1330 }
1331
1332 static void __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1333                                                 struct discard_policy *dpolicy)
1334 {
1335         __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1336 }
1337
1338 /* This should be covered by global mutex, &sit_i->sentry_lock */
1339 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1340 {
1341         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1342         struct discard_cmd *dc;
1343         bool need_wait = false;
1344
1345         mutex_lock(&dcc->cmd_lock);
1346         dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1347         if (dc) {
1348                 if (dc->state == D_PREP) {
1349                         __punch_discard_cmd(sbi, dc, blkaddr);
1350                 } else {
1351                         dc->ref++;
1352                         need_wait = true;
1353                 }
1354         }
1355         mutex_unlock(&dcc->cmd_lock);
1356
1357         if (need_wait)
1358                 __wait_one_discard_bio(sbi, dc);
1359 }
1360
1361 void stop_discard_thread(struct f2fs_sb_info *sbi)
1362 {
1363         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1364
1365         if (dcc && dcc->f2fs_issue_discard) {
1366                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1367
1368                 dcc->f2fs_issue_discard = NULL;
1369                 kthread_stop(discard_thread);
1370         }
1371 }
1372
1373 /* This comes from f2fs_put_super */
1374 bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1375 {
1376         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1377         struct discard_policy dpolicy;
1378         bool dropped;
1379
1380         init_discard_policy(&dpolicy, DPOLICY_UMOUNT, dcc->discard_granularity);
1381         __issue_discard_cmd(sbi, &dpolicy);
1382         dropped = __drop_discard_cmd(sbi);
1383         __wait_all_discard_cmd(sbi, &dpolicy);
1384
1385         return dropped;
1386 }
1387
1388 static int issue_discard_thread(void *data)
1389 {
1390         struct f2fs_sb_info *sbi = data;
1391         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1392         wait_queue_head_t *q = &dcc->discard_wait_queue;
1393         struct discard_policy dpolicy;
1394         unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1395         int issued;
1396
1397         set_freezable();
1398
1399         do {
1400                 init_discard_policy(&dpolicy, DPOLICY_BG,
1401                                         dcc->discard_granularity);
1402
1403                 wait_event_interruptible_timeout(*q,
1404                                 kthread_should_stop() || freezing(current) ||
1405                                 dcc->discard_wake,
1406                                 msecs_to_jiffies(wait_ms));
1407                 if (try_to_freeze())
1408                         continue;
1409                 if (f2fs_readonly(sbi->sb))
1410                         continue;
1411                 if (kthread_should_stop())
1412                         return 0;
1413
1414                 if (dcc->discard_wake) {
1415                         dcc->discard_wake = 0;
1416                         if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
1417                                 init_discard_policy(&dpolicy,
1418                                                         DPOLICY_FORCE, 1);
1419                 }
1420
1421                 sb_start_intwrite(sbi->sb);
1422
1423                 issued = __issue_discard_cmd(sbi, &dpolicy);
1424                 if (issued) {
1425                         __wait_all_discard_cmd(sbi, &dpolicy);
1426                         wait_ms = dpolicy.min_interval;
1427                 } else {
1428                         wait_ms = dpolicy.max_interval;
1429                 }
1430
1431                 sb_end_intwrite(sbi->sb);
1432
1433         } while (!kthread_should_stop());
1434         return 0;
1435 }
1436
1437 #ifdef CONFIG_BLK_DEV_ZONED
1438 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1439                 struct block_device *bdev, block_t blkstart, block_t blklen)
1440 {
1441         sector_t sector, nr_sects;
1442         block_t lblkstart = blkstart;
1443         int devi = 0;
1444
1445         if (sbi->s_ndevs) {
1446                 devi = f2fs_target_device_index(sbi, blkstart);
1447                 blkstart -= FDEV(devi).start_blk;
1448         }
1449
1450         /*
1451          * We need to know the type of the zone: for conventional zones,
1452          * use regular discard if the drive supports it. For sequential
1453          * zones, reset the zone write pointer.
1454          */
1455         switch (get_blkz_type(sbi, bdev, blkstart)) {
1456
1457         case BLK_ZONE_TYPE_CONVENTIONAL:
1458                 if (!blk_queue_discard(bdev_get_queue(bdev)))
1459                         return 0;
1460                 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1461         case BLK_ZONE_TYPE_SEQWRITE_REQ:
1462         case BLK_ZONE_TYPE_SEQWRITE_PREF:
1463                 sector = SECTOR_FROM_BLOCK(blkstart);
1464                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1465
1466                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1467                                 nr_sects != bdev_zone_sectors(bdev)) {
1468                         f2fs_msg(sbi->sb, KERN_INFO,
1469                                 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1470                                 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1471                                 blkstart, blklen);
1472                         return -EIO;
1473                 }
1474                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1475                 return blkdev_reset_zones(bdev, sector,
1476                                           nr_sects, GFP_NOFS);
1477         default:
1478                 /* Unknown zone type: broken device ? */
1479                 return -EIO;
1480         }
1481 }
1482 #endif
1483
1484 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1485                 struct block_device *bdev, block_t blkstart, block_t blklen)
1486 {
1487 #ifdef CONFIG_BLK_DEV_ZONED
1488         if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
1489                                 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1490                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1491 #endif
1492         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1493 }
1494
1495 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1496                                 block_t blkstart, block_t blklen)
1497 {
1498         sector_t start = blkstart, len = 0;
1499         struct block_device *bdev;
1500         struct seg_entry *se;
1501         unsigned int offset;
1502         block_t i;
1503         int err = 0;
1504
1505         bdev = f2fs_target_device(sbi, blkstart, NULL);
1506
1507         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1508                 if (i != start) {
1509                         struct block_device *bdev2 =
1510                                 f2fs_target_device(sbi, i, NULL);
1511
1512                         if (bdev2 != bdev) {
1513                                 err = __issue_discard_async(sbi, bdev,
1514                                                 start, len);
1515                                 if (err)
1516                                         return err;
1517                                 bdev = bdev2;
1518                                 start = i;
1519                                 len = 0;
1520                         }
1521                 }
1522
1523                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1524                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1525
1526                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1527                         sbi->discard_blks--;
1528         }
1529
1530         if (len)
1531                 err = __issue_discard_async(sbi, bdev, start, len);
1532         return err;
1533 }
1534
1535 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1536                                                         bool check_only)
1537 {
1538         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1539         int max_blocks = sbi->blocks_per_seg;
1540         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1541         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1542         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1543         unsigned long *discard_map = (unsigned long *)se->discard_map;
1544         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1545         unsigned int start = 0, end = -1;
1546         bool force = (cpc->reason & CP_DISCARD);
1547         struct discard_entry *de = NULL;
1548         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1549         int i;
1550
1551         if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
1552                 return false;
1553
1554         if (!force) {
1555                 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
1556                         SM_I(sbi)->dcc_info->nr_discards >=
1557                                 SM_I(sbi)->dcc_info->max_discards)
1558                         return false;
1559         }
1560
1561         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1562         for (i = 0; i < entries; i++)
1563                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1564                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1565
1566         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1567                                 SM_I(sbi)->dcc_info->max_discards) {
1568                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1569                 if (start >= max_blocks)
1570                         break;
1571
1572                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1573                 if (force && start && end != max_blocks
1574                                         && (end - start) < cpc->trim_minlen)
1575                         continue;
1576
1577                 if (check_only)
1578                         return true;
1579
1580                 if (!de) {
1581                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1582                                                                 GFP_F2FS_ZERO);
1583                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1584                         list_add_tail(&de->list, head);
1585                 }
1586
1587                 for (i = start; i < end; i++)
1588                         __set_bit_le(i, (void *)de->discard_map);
1589
1590                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1591         }
1592         return false;
1593 }
1594
1595 void release_discard_addrs(struct f2fs_sb_info *sbi)
1596 {
1597         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1598         struct discard_entry *entry, *this;
1599
1600         /* drop caches */
1601         list_for_each_entry_safe(entry, this, head, list) {
1602                 list_del(&entry->list);
1603                 kmem_cache_free(discard_entry_slab, entry);
1604         }
1605 }
1606
1607 /*
1608  * Should call clear_prefree_segments after checkpoint is done.
1609  */
1610 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1611 {
1612         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1613         unsigned int segno;
1614
1615         mutex_lock(&dirty_i->seglist_lock);
1616         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1617                 __set_test_and_free(sbi, segno);
1618         mutex_unlock(&dirty_i->seglist_lock);
1619 }
1620
1621 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1622 {
1623         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1624         struct list_head *head = &dcc->entry_list;
1625         struct discard_entry *entry, *this;
1626         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1627         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1628         unsigned int start = 0, end = -1;
1629         unsigned int secno, start_segno;
1630         bool force = (cpc->reason & CP_DISCARD);
1631
1632         mutex_lock(&dirty_i->seglist_lock);
1633
1634         while (1) {
1635                 int i;
1636                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1637                 if (start >= MAIN_SEGS(sbi))
1638                         break;
1639                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1640                                                                 start + 1);
1641
1642                 for (i = start; i < end; i++)
1643                         clear_bit(i, prefree_map);
1644
1645                 dirty_i->nr_dirty[PRE] -= end - start;
1646
1647                 if (!test_opt(sbi, DISCARD))
1648                         continue;
1649
1650                 if (force && start >= cpc->trim_start &&
1651                                         (end - 1) <= cpc->trim_end)
1652                                 continue;
1653
1654                 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1655                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1656                                 (end - start) << sbi->log_blocks_per_seg);
1657                         continue;
1658                 }
1659 next:
1660                 secno = GET_SEC_FROM_SEG(sbi, start);
1661                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1662                 if (!IS_CURSEC(sbi, secno) &&
1663                         !get_valid_blocks(sbi, start, true))
1664                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1665                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1666
1667                 start = start_segno + sbi->segs_per_sec;
1668                 if (start < end)
1669                         goto next;
1670                 else
1671                         end = start - 1;
1672         }
1673         mutex_unlock(&dirty_i->seglist_lock);
1674
1675         /* send small discards */
1676         list_for_each_entry_safe(entry, this, head, list) {
1677                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1678                 bool is_valid = test_bit_le(0, entry->discard_map);
1679
1680 find_next:
1681                 if (is_valid) {
1682                         next_pos = find_next_zero_bit_le(entry->discard_map,
1683                                         sbi->blocks_per_seg, cur_pos);
1684                         len = next_pos - cur_pos;
1685
1686                         if (f2fs_sb_mounted_blkzoned(sbi->sb) ||
1687                             (force && len < cpc->trim_minlen))
1688                                 goto skip;
1689
1690                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1691                                                                         len);
1692                         total_len += len;
1693                 } else {
1694                         next_pos = find_next_bit_le(entry->discard_map,
1695                                         sbi->blocks_per_seg, cur_pos);
1696                 }
1697 skip:
1698                 cur_pos = next_pos;
1699                 is_valid = !is_valid;
1700
1701                 if (cur_pos < sbi->blocks_per_seg)
1702                         goto find_next;
1703
1704                 list_del(&entry->list);
1705                 dcc->nr_discards -= total_len;
1706                 kmem_cache_free(discard_entry_slab, entry);
1707         }
1708
1709         wake_up_discard_thread(sbi, false);
1710 }
1711
1712 void init_discard_policy(struct discard_policy *dpolicy,
1713                                 int discard_type, unsigned int granularity)
1714 {
1715         /* common policy */
1716         dpolicy->type = discard_type;
1717         dpolicy->sync = true;
1718         dpolicy->granularity = granularity;
1719
1720         dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1721         dpolicy->io_aware_gran = MAX_PLIST_NUM;
1722
1723         if (discard_type == DPOLICY_BG) {
1724                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1725                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1726                 dpolicy->io_aware = true;
1727         } else if (discard_type == DPOLICY_FORCE) {
1728                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1729                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1730                 dpolicy->io_aware = true;
1731         } else if (discard_type == DPOLICY_FSTRIM) {
1732                 dpolicy->io_aware = false;
1733         } else if (discard_type == DPOLICY_UMOUNT) {
1734                 dpolicy->io_aware = false;
1735         }
1736 }
1737
1738 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1739 {
1740         dev_t dev = sbi->sb->s_bdev->bd_dev;
1741         struct discard_cmd_control *dcc;
1742         int err = 0, i;
1743
1744         if (SM_I(sbi)->dcc_info) {
1745                 dcc = SM_I(sbi)->dcc_info;
1746                 goto init_thread;
1747         }
1748
1749         dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
1750         if (!dcc)
1751                 return -ENOMEM;
1752
1753         dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
1754         INIT_LIST_HEAD(&dcc->entry_list);
1755         for (i = 0; i < MAX_PLIST_NUM; i++)
1756                 INIT_LIST_HEAD(&dcc->pend_list[i]);
1757         INIT_LIST_HEAD(&dcc->wait_list);
1758         INIT_LIST_HEAD(&dcc->fstrim_list);
1759         mutex_init(&dcc->cmd_lock);
1760         atomic_set(&dcc->issued_discard, 0);
1761         atomic_set(&dcc->issing_discard, 0);
1762         atomic_set(&dcc->discard_cmd_cnt, 0);
1763         dcc->nr_discards = 0;
1764         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1765         dcc->undiscard_blks = 0;
1766         dcc->root = RB_ROOT;
1767
1768         init_waitqueue_head(&dcc->discard_wait_queue);
1769         SM_I(sbi)->dcc_info = dcc;
1770 init_thread:
1771         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1772                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1773         if (IS_ERR(dcc->f2fs_issue_discard)) {
1774                 err = PTR_ERR(dcc->f2fs_issue_discard);
1775                 kfree(dcc);
1776                 SM_I(sbi)->dcc_info = NULL;
1777                 return err;
1778         }
1779
1780         return err;
1781 }
1782
1783 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1784 {
1785         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1786
1787         if (!dcc)
1788                 return;
1789
1790         stop_discard_thread(sbi);
1791
1792         kfree(dcc);
1793         SM_I(sbi)->dcc_info = NULL;
1794 }
1795
1796 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1797 {
1798         struct sit_info *sit_i = SIT_I(sbi);
1799
1800         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1801                 sit_i->dirty_sentries++;
1802                 return false;
1803         }
1804
1805         return true;
1806 }
1807
1808 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1809                                         unsigned int segno, int modified)
1810 {
1811         struct seg_entry *se = get_seg_entry(sbi, segno);
1812         se->type = type;
1813         if (modified)
1814                 __mark_sit_entry_dirty(sbi, segno);
1815 }
1816
1817 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1818 {
1819         struct seg_entry *se;
1820         unsigned int segno, offset;
1821         long int new_vblocks;
1822         bool exist;
1823 #ifdef CONFIG_F2FS_CHECK_FS
1824         bool mir_exist;
1825 #endif
1826
1827         segno = GET_SEGNO(sbi, blkaddr);
1828
1829         se = get_seg_entry(sbi, segno);
1830         new_vblocks = se->valid_blocks + del;
1831         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1832
1833         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
1834                                 (new_vblocks > sbi->blocks_per_seg)));
1835
1836         se->valid_blocks = new_vblocks;
1837         se->mtime = get_mtime(sbi);
1838         SIT_I(sbi)->max_mtime = se->mtime;
1839
1840         /* Update valid block bitmap */
1841         if (del > 0) {
1842                 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
1843 #ifdef CONFIG_F2FS_CHECK_FS
1844                 mir_exist = f2fs_test_and_set_bit(offset,
1845                                                 se->cur_valid_map_mir);
1846                 if (unlikely(exist != mir_exist)) {
1847                         f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1848                                 "when setting bitmap, blk:%u, old bit:%d",
1849                                 blkaddr, exist);
1850                         f2fs_bug_on(sbi, 1);
1851                 }
1852 #endif
1853                 if (unlikely(exist)) {
1854                         f2fs_msg(sbi->sb, KERN_ERR,
1855                                 "Bitmap was wrongly set, blk:%u", blkaddr);
1856                         f2fs_bug_on(sbi, 1);
1857                         se->valid_blocks--;
1858                         del = 0;
1859                 }
1860
1861                 if (f2fs_discard_en(sbi) &&
1862                         !f2fs_test_and_set_bit(offset, se->discard_map))
1863                         sbi->discard_blks--;
1864
1865                 /* don't overwrite by SSR to keep node chain */
1866                 if (se->type == CURSEG_WARM_NODE) {
1867                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1868                                 se->ckpt_valid_blocks++;
1869                 }
1870         } else {
1871                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
1872 #ifdef CONFIG_F2FS_CHECK_FS
1873                 mir_exist = f2fs_test_and_clear_bit(offset,
1874                                                 se->cur_valid_map_mir);
1875                 if (unlikely(exist != mir_exist)) {
1876                         f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1877                                 "when clearing bitmap, blk:%u, old bit:%d",
1878                                 blkaddr, exist);
1879                         f2fs_bug_on(sbi, 1);
1880                 }
1881 #endif
1882                 if (unlikely(!exist)) {
1883                         f2fs_msg(sbi->sb, KERN_ERR,
1884                                 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1885                         f2fs_bug_on(sbi, 1);
1886                         se->valid_blocks++;
1887                         del = 0;
1888                 }
1889
1890                 if (f2fs_discard_en(sbi) &&
1891                         f2fs_test_and_clear_bit(offset, se->discard_map))
1892                         sbi->discard_blks++;
1893         }
1894         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1895                 se->ckpt_valid_blocks += del;
1896
1897         __mark_sit_entry_dirty(sbi, segno);
1898
1899         /* update total number of valid blocks to be written in ckpt area */
1900         SIT_I(sbi)->written_valid_blocks += del;
1901
1902         if (sbi->segs_per_sec > 1)
1903                 get_sec_entry(sbi, segno)->valid_blocks += del;
1904 }
1905
1906 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1907 {
1908         unsigned int segno = GET_SEGNO(sbi, addr);
1909         struct sit_info *sit_i = SIT_I(sbi);
1910
1911         f2fs_bug_on(sbi, addr == NULL_ADDR);
1912         if (addr == NEW_ADDR)
1913                 return;
1914
1915         /* add it into sit main buffer */
1916         down_write(&sit_i->sentry_lock);
1917
1918         update_sit_entry(sbi, addr, -1);
1919
1920         /* add it into dirty seglist */
1921         locate_dirty_segment(sbi, segno);
1922
1923         up_write(&sit_i->sentry_lock);
1924 }
1925
1926 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1927 {
1928         struct sit_info *sit_i = SIT_I(sbi);
1929         unsigned int segno, offset;
1930         struct seg_entry *se;
1931         bool is_cp = false;
1932
1933         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1934                 return true;
1935
1936         down_read(&sit_i->sentry_lock);
1937
1938         segno = GET_SEGNO(sbi, blkaddr);
1939         se = get_seg_entry(sbi, segno);
1940         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1941
1942         if (f2fs_test_bit(offset, se->ckpt_valid_map))
1943                 is_cp = true;
1944
1945         up_read(&sit_i->sentry_lock);
1946
1947         return is_cp;
1948 }
1949
1950 /*
1951  * This function should be resided under the curseg_mutex lock
1952  */
1953 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1954                                         struct f2fs_summary *sum)
1955 {
1956         struct curseg_info *curseg = CURSEG_I(sbi, type);
1957         void *addr = curseg->sum_blk;
1958         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1959         memcpy(addr, sum, sizeof(struct f2fs_summary));
1960 }
1961
1962 /*
1963  * Calculate the number of current summary pages for writing
1964  */
1965 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1966 {
1967         int valid_sum_count = 0;
1968         int i, sum_in_page;
1969
1970         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1971                 if (sbi->ckpt->alloc_type[i] == SSR)
1972                         valid_sum_count += sbi->blocks_per_seg;
1973                 else {
1974                         if (for_ra)
1975                                 valid_sum_count += le16_to_cpu(
1976                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1977                         else
1978                                 valid_sum_count += curseg_blkoff(sbi, i);
1979                 }
1980         }
1981
1982         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1983                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1984         if (valid_sum_count <= sum_in_page)
1985                 return 1;
1986         else if ((valid_sum_count - sum_in_page) <=
1987                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1988                 return 2;
1989         return 3;
1990 }
1991
1992 /*
1993  * Caller should put this summary page
1994  */
1995 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1996 {
1997         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1998 }
1999
2000 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
2001 {
2002         struct page *page = grab_meta_page(sbi, blk_addr);
2003
2004         memcpy(page_address(page), src, PAGE_SIZE);
2005         set_page_dirty(page);
2006         f2fs_put_page(page, 1);
2007 }
2008
2009 static void write_sum_page(struct f2fs_sb_info *sbi,
2010                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
2011 {
2012         update_meta_page(sbi, (void *)sum_blk, blk_addr);
2013 }
2014
2015 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2016                                                 int type, block_t blk_addr)
2017 {
2018         struct curseg_info *curseg = CURSEG_I(sbi, type);
2019         struct page *page = grab_meta_page(sbi, blk_addr);
2020         struct f2fs_summary_block *src = curseg->sum_blk;
2021         struct f2fs_summary_block *dst;
2022
2023         dst = (struct f2fs_summary_block *)page_address(page);
2024
2025         mutex_lock(&curseg->curseg_mutex);
2026
2027         down_read(&curseg->journal_rwsem);
2028         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2029         up_read(&curseg->journal_rwsem);
2030
2031         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2032         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2033
2034         mutex_unlock(&curseg->curseg_mutex);
2035
2036         set_page_dirty(page);
2037         f2fs_put_page(page, 1);
2038 }
2039
2040 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2041 {
2042         struct curseg_info *curseg = CURSEG_I(sbi, type);
2043         unsigned int segno = curseg->segno + 1;
2044         struct free_segmap_info *free_i = FREE_I(sbi);
2045
2046         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2047                 return !test_bit(segno, free_i->free_segmap);
2048         return 0;
2049 }
2050
2051 /*
2052  * Find a new segment from the free segments bitmap to right order
2053  * This function should be returned with success, otherwise BUG
2054  */
2055 static void get_new_segment(struct f2fs_sb_info *sbi,
2056                         unsigned int *newseg, bool new_sec, int dir)
2057 {
2058         struct free_segmap_info *free_i = FREE_I(sbi);
2059         unsigned int segno, secno, zoneno;
2060         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2061         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2062         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2063         unsigned int left_start = hint;
2064         bool init = true;
2065         int go_left = 0;
2066         int i;
2067
2068         spin_lock(&free_i->segmap_lock);
2069
2070         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2071                 segno = find_next_zero_bit(free_i->free_segmap,
2072                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2073                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2074                         goto got_it;
2075         }
2076 find_other_zone:
2077         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2078         if (secno >= MAIN_SECS(sbi)) {
2079                 if (dir == ALLOC_RIGHT) {
2080                         secno = find_next_zero_bit(free_i->free_secmap,
2081                                                         MAIN_SECS(sbi), 0);
2082                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2083                 } else {
2084                         go_left = 1;
2085                         left_start = hint - 1;
2086                 }
2087         }
2088         if (go_left == 0)
2089                 goto skip_left;
2090
2091         while (test_bit(left_start, free_i->free_secmap)) {
2092                 if (left_start > 0) {
2093                         left_start--;
2094                         continue;
2095                 }
2096                 left_start = find_next_zero_bit(free_i->free_secmap,
2097                                                         MAIN_SECS(sbi), 0);
2098                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2099                 break;
2100         }
2101         secno = left_start;
2102 skip_left:
2103         segno = GET_SEG_FROM_SEC(sbi, secno);
2104         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2105
2106         /* give up on finding another zone */
2107         if (!init)
2108                 goto got_it;
2109         if (sbi->secs_per_zone == 1)
2110                 goto got_it;
2111         if (zoneno == old_zoneno)
2112                 goto got_it;
2113         if (dir == ALLOC_LEFT) {
2114                 if (!go_left && zoneno + 1 >= total_zones)
2115                         goto got_it;
2116                 if (go_left && zoneno == 0)
2117                         goto got_it;
2118         }
2119         for (i = 0; i < NR_CURSEG_TYPE; i++)
2120                 if (CURSEG_I(sbi, i)->zone == zoneno)
2121                         break;
2122
2123         if (i < NR_CURSEG_TYPE) {
2124                 /* zone is in user, try another */
2125                 if (go_left)
2126                         hint = zoneno * sbi->secs_per_zone - 1;
2127                 else if (zoneno + 1 >= total_zones)
2128                         hint = 0;
2129                 else
2130                         hint = (zoneno + 1) * sbi->secs_per_zone;
2131                 init = false;
2132                 goto find_other_zone;
2133         }
2134 got_it:
2135         /* set it as dirty segment in free segmap */
2136         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2137         __set_inuse(sbi, segno);
2138         *newseg = segno;
2139         spin_unlock(&free_i->segmap_lock);
2140 }
2141
2142 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2143 {
2144         struct curseg_info *curseg = CURSEG_I(sbi, type);
2145         struct summary_footer *sum_footer;
2146
2147         curseg->segno = curseg->next_segno;
2148         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2149         curseg->next_blkoff = 0;
2150         curseg->next_segno = NULL_SEGNO;
2151
2152         sum_footer = &(curseg->sum_blk->footer);
2153         memset(sum_footer, 0, sizeof(struct summary_footer));
2154         if (IS_DATASEG(type))
2155                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2156         if (IS_NODESEG(type))
2157                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2158         __set_sit_entry_type(sbi, type, curseg->segno, modified);
2159 }
2160
2161 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2162 {
2163         /* if segs_per_sec is large than 1, we need to keep original policy. */
2164         if (sbi->segs_per_sec != 1)
2165                 return CURSEG_I(sbi, type)->segno;
2166
2167         if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
2168                 return 0;
2169
2170         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2171                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2172         return CURSEG_I(sbi, type)->segno;
2173 }
2174
2175 /*
2176  * Allocate a current working segment.
2177  * This function always allocates a free segment in LFS manner.
2178  */
2179 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2180 {
2181         struct curseg_info *curseg = CURSEG_I(sbi, type);
2182         unsigned int segno = curseg->segno;
2183         int dir = ALLOC_LEFT;
2184
2185         write_sum_page(sbi, curseg->sum_blk,
2186                                 GET_SUM_BLOCK(sbi, segno));
2187         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2188                 dir = ALLOC_RIGHT;
2189
2190         if (test_opt(sbi, NOHEAP))
2191                 dir = ALLOC_RIGHT;
2192
2193         segno = __get_next_segno(sbi, type);
2194         get_new_segment(sbi, &segno, new_sec, dir);
2195         curseg->next_segno = segno;
2196         reset_curseg(sbi, type, 1);
2197         curseg->alloc_type = LFS;
2198 }
2199
2200 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2201                         struct curseg_info *seg, block_t start)
2202 {
2203         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2204         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2205         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2206         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2207         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2208         int i, pos;
2209
2210         for (i = 0; i < entries; i++)
2211                 target_map[i] = ckpt_map[i] | cur_map[i];
2212
2213         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2214
2215         seg->next_blkoff = pos;
2216 }
2217
2218 /*
2219  * If a segment is written by LFS manner, next block offset is just obtained
2220  * by increasing the current block offset. However, if a segment is written by
2221  * SSR manner, next block offset obtained by calling __next_free_blkoff
2222  */
2223 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2224                                 struct curseg_info *seg)
2225 {
2226         if (seg->alloc_type == SSR)
2227                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2228         else
2229                 seg->next_blkoff++;
2230 }
2231
2232 /*
2233  * This function always allocates a used segment(from dirty seglist) by SSR
2234  * manner, so it should recover the existing segment information of valid blocks
2235  */
2236 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2237 {
2238         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2239         struct curseg_info *curseg = CURSEG_I(sbi, type);
2240         unsigned int new_segno = curseg->next_segno;
2241         struct f2fs_summary_block *sum_node;
2242         struct page *sum_page;
2243
2244         write_sum_page(sbi, curseg->sum_blk,
2245                                 GET_SUM_BLOCK(sbi, curseg->segno));
2246         __set_test_and_inuse(sbi, new_segno);
2247
2248         mutex_lock(&dirty_i->seglist_lock);
2249         __remove_dirty_segment(sbi, new_segno, PRE);
2250         __remove_dirty_segment(sbi, new_segno, DIRTY);
2251         mutex_unlock(&dirty_i->seglist_lock);
2252
2253         reset_curseg(sbi, type, 1);
2254         curseg->alloc_type = SSR;
2255         __next_free_blkoff(sbi, curseg, 0);
2256
2257         sum_page = get_sum_page(sbi, new_segno);
2258         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2259         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2260         f2fs_put_page(sum_page, 1);
2261 }
2262
2263 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2264 {
2265         struct curseg_info *curseg = CURSEG_I(sbi, type);
2266         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2267         unsigned segno = NULL_SEGNO;
2268         int i, cnt;
2269         bool reversed = false;
2270
2271         /* need_SSR() already forces to do this */
2272         if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2273                 curseg->next_segno = segno;
2274                 return 1;
2275         }
2276
2277         /* For node segments, let's do SSR more intensively */
2278         if (IS_NODESEG(type)) {
2279                 if (type >= CURSEG_WARM_NODE) {
2280                         reversed = true;
2281                         i = CURSEG_COLD_NODE;
2282                 } else {
2283                         i = CURSEG_HOT_NODE;
2284                 }
2285                 cnt = NR_CURSEG_NODE_TYPE;
2286         } else {
2287                 if (type >= CURSEG_WARM_DATA) {
2288                         reversed = true;
2289                         i = CURSEG_COLD_DATA;
2290                 } else {
2291                         i = CURSEG_HOT_DATA;
2292                 }
2293                 cnt = NR_CURSEG_DATA_TYPE;
2294         }
2295
2296         for (; cnt-- > 0; reversed ? i-- : i++) {
2297                 if (i == type)
2298                         continue;
2299                 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2300                         curseg->next_segno = segno;
2301                         return 1;
2302                 }
2303         }
2304         return 0;
2305 }
2306
2307 /*
2308  * flush out current segment and replace it with new segment
2309  * This function should be returned with success, otherwise BUG
2310  */
2311 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2312                                                 int type, bool force)
2313 {
2314         struct curseg_info *curseg = CURSEG_I(sbi, type);
2315
2316         if (force)
2317                 new_curseg(sbi, type, true);
2318         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2319                                         type == CURSEG_WARM_NODE)
2320                 new_curseg(sbi, type, false);
2321         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2322                 new_curseg(sbi, type, false);
2323         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
2324                 change_curseg(sbi, type);
2325         else
2326                 new_curseg(sbi, type, false);
2327
2328         stat_inc_seg_type(sbi, curseg);
2329 }
2330
2331 void allocate_new_segments(struct f2fs_sb_info *sbi)
2332 {
2333         struct curseg_info *curseg;
2334         unsigned int old_segno;
2335         int i;
2336
2337         down_write(&SIT_I(sbi)->sentry_lock);
2338
2339         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2340                 curseg = CURSEG_I(sbi, i);
2341                 old_segno = curseg->segno;
2342                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2343                 locate_dirty_segment(sbi, old_segno);
2344         }
2345
2346         up_write(&SIT_I(sbi)->sentry_lock);
2347 }
2348
2349 static const struct segment_allocation default_salloc_ops = {
2350         .allocate_segment = allocate_segment_by_default,
2351 };
2352
2353 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2354 {
2355         __u64 trim_start = cpc->trim_start;
2356         bool has_candidate = false;
2357
2358         down_write(&SIT_I(sbi)->sentry_lock);
2359         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2360                 if (add_discard_addrs(sbi, cpc, true)) {
2361                         has_candidate = true;
2362                         break;
2363                 }
2364         }
2365         up_write(&SIT_I(sbi)->sentry_lock);
2366
2367         cpc->trim_start = trim_start;
2368         return has_candidate;
2369 }
2370
2371 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2372 {
2373         __u64 start = F2FS_BYTES_TO_BLK(range->start);
2374         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2375         unsigned int start_segno, end_segno, cur_segno;
2376         block_t start_block, end_block;
2377         struct cp_control cpc;
2378         struct discard_policy dpolicy;
2379         unsigned long long trimmed = 0;
2380         int err = 0;
2381
2382         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2383                 return -EINVAL;
2384
2385         if (end <= MAIN_BLKADDR(sbi))
2386                 goto out;
2387
2388         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2389                 f2fs_msg(sbi->sb, KERN_WARNING,
2390                         "Found FS corruption, run fsck to fix.");
2391                 goto out;
2392         }
2393
2394         /* start/end segment number in main_area */
2395         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2396         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2397                                                 GET_SEGNO(sbi, end);
2398
2399         cpc.reason = CP_DISCARD;
2400         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2401
2402         /* do checkpoint to issue discard commands safely */
2403         for (cur_segno = start_segno; cur_segno <= end_segno;
2404                                         cur_segno = cpc.trim_end + 1) {
2405                 cpc.trim_start = cur_segno;
2406
2407                 if (sbi->discard_blks == 0)
2408                         break;
2409                 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
2410                         cpc.trim_end = end_segno;
2411                 else
2412                         cpc.trim_end = min_t(unsigned int,
2413                                 rounddown(cur_segno +
2414                                 BATCHED_TRIM_SEGMENTS(sbi),
2415                                 sbi->segs_per_sec) - 1, end_segno);
2416
2417                 mutex_lock(&sbi->gc_mutex);
2418                 err = write_checkpoint(sbi, &cpc);
2419                 mutex_unlock(&sbi->gc_mutex);
2420                 if (err)
2421                         break;
2422
2423                 schedule();
2424         }
2425
2426         start_block = START_BLOCK(sbi, start_segno);
2427         end_block = START_BLOCK(sbi, min(cur_segno, end_segno) + 1);
2428
2429         init_discard_policy(&dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2430         __issue_discard_cmd_range(sbi, &dpolicy, start_block, end_block);
2431         trimmed = __wait_discard_cmd_range(sbi, &dpolicy,
2432                                         start_block, end_block);
2433 out:
2434         range->len = F2FS_BLK_TO_BYTES(trimmed);
2435         return err;
2436 }
2437
2438 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2439 {
2440         struct curseg_info *curseg = CURSEG_I(sbi, type);
2441         if (curseg->next_blkoff < sbi->blocks_per_seg)
2442                 return true;
2443         return false;
2444 }
2445
2446 int rw_hint_to_seg_type(enum rw_hint hint)
2447 {
2448         switch (hint) {
2449         case WRITE_LIFE_SHORT:
2450                 return CURSEG_HOT_DATA;
2451         case WRITE_LIFE_EXTREME:
2452                 return CURSEG_COLD_DATA;
2453         default:
2454                 return CURSEG_WARM_DATA;
2455         }
2456 }
2457
2458 static int __get_segment_type_2(struct f2fs_io_info *fio)
2459 {
2460         if (fio->type == DATA)
2461                 return CURSEG_HOT_DATA;
2462         else
2463                 return CURSEG_HOT_NODE;
2464 }
2465
2466 static int __get_segment_type_4(struct f2fs_io_info *fio)
2467 {
2468         if (fio->type == DATA) {
2469                 struct inode *inode = fio->page->mapping->host;
2470
2471                 if (S_ISDIR(inode->i_mode))
2472                         return CURSEG_HOT_DATA;
2473                 else
2474                         return CURSEG_COLD_DATA;
2475         } else {
2476                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2477                         return CURSEG_WARM_NODE;
2478                 else
2479                         return CURSEG_COLD_NODE;
2480         }
2481 }
2482
2483 static int __get_segment_type_6(struct f2fs_io_info *fio)
2484 {
2485         if (fio->type == DATA) {
2486                 struct inode *inode = fio->page->mapping->host;
2487
2488                 if (is_cold_data(fio->page) || file_is_cold(inode))
2489                         return CURSEG_COLD_DATA;
2490                 if (is_inode_flag_set(inode, FI_HOT_DATA))
2491                         return CURSEG_HOT_DATA;
2492                 return rw_hint_to_seg_type(inode->i_write_hint);
2493         } else {
2494                 if (IS_DNODE(fio->page))
2495                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2496                                                 CURSEG_HOT_NODE;
2497                 return CURSEG_COLD_NODE;
2498         }
2499 }
2500
2501 static int __get_segment_type(struct f2fs_io_info *fio)
2502 {
2503         int type = 0;
2504
2505         switch (fio->sbi->active_logs) {
2506         case 2:
2507                 type = __get_segment_type_2(fio);
2508                 break;
2509         case 4:
2510                 type = __get_segment_type_4(fio);
2511                 break;
2512         case 6:
2513                 type = __get_segment_type_6(fio);
2514                 break;
2515         default:
2516                 f2fs_bug_on(fio->sbi, true);
2517         }
2518
2519         if (IS_HOT(type))
2520                 fio->temp = HOT;
2521         else if (IS_WARM(type))
2522                 fio->temp = WARM;
2523         else
2524                 fio->temp = COLD;
2525         return type;
2526 }
2527
2528 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2529                 block_t old_blkaddr, block_t *new_blkaddr,
2530                 struct f2fs_summary *sum, int type,
2531                 struct f2fs_io_info *fio, bool add_list)
2532 {
2533         struct sit_info *sit_i = SIT_I(sbi);
2534         struct curseg_info *curseg = CURSEG_I(sbi, type);
2535
2536         down_read(&SM_I(sbi)->curseg_lock);
2537
2538         mutex_lock(&curseg->curseg_mutex);
2539         down_write(&sit_i->sentry_lock);
2540
2541         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2542
2543         f2fs_wait_discard_bio(sbi, *new_blkaddr);
2544
2545         /*
2546          * __add_sum_entry should be resided under the curseg_mutex
2547          * because, this function updates a summary entry in the
2548          * current summary block.
2549          */
2550         __add_sum_entry(sbi, type, sum);
2551
2552         __refresh_next_blkoff(sbi, curseg);
2553
2554         stat_inc_block_count(sbi, curseg);
2555
2556         /*
2557          * SIT information should be updated before segment allocation,
2558          * since SSR needs latest valid block information.
2559          */
2560         update_sit_entry(sbi, *new_blkaddr, 1);
2561         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2562                 update_sit_entry(sbi, old_blkaddr, -1);
2563
2564         if (!__has_curseg_space(sbi, type))
2565                 sit_i->s_ops->allocate_segment(sbi, type, false);
2566
2567         /*
2568          * segment dirty status should be updated after segment allocation,
2569          * so we just need to update status only one time after previous
2570          * segment being closed.
2571          */
2572         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2573         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
2574
2575         up_write(&sit_i->sentry_lock);
2576
2577         if (page && IS_NODESEG(type)) {
2578                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2579
2580                 f2fs_inode_chksum_set(sbi, page);
2581         }
2582
2583         if (add_list) {
2584                 struct f2fs_bio_info *io;
2585
2586                 INIT_LIST_HEAD(&fio->list);
2587                 fio->in_list = true;
2588                 io = sbi->write_io[fio->type] + fio->temp;
2589                 spin_lock(&io->io_lock);
2590                 list_add_tail(&fio->list, &io->io_list);
2591                 spin_unlock(&io->io_lock);
2592         }
2593
2594         mutex_unlock(&curseg->curseg_mutex);
2595
2596         up_read(&SM_I(sbi)->curseg_lock);
2597 }
2598
2599 static void update_device_state(struct f2fs_io_info *fio)
2600 {
2601         struct f2fs_sb_info *sbi = fio->sbi;
2602         unsigned int devidx;
2603
2604         if (!sbi->s_ndevs)
2605                 return;
2606
2607         devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2608
2609         /* update device state for fsync */
2610         set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
2611
2612         /* update device state for checkpoint */
2613         if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2614                 spin_lock(&sbi->dev_lock);
2615                 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2616                 spin_unlock(&sbi->dev_lock);
2617         }
2618 }
2619
2620 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2621 {
2622         int type = __get_segment_type(fio);
2623         int err;
2624
2625 reallocate:
2626         allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
2627                         &fio->new_blkaddr, sum, type, fio, true);
2628
2629         /* writeout dirty page into bdev */
2630         err = f2fs_submit_page_write(fio);
2631         if (err == -EAGAIN) {
2632                 fio->old_blkaddr = fio->new_blkaddr;
2633                 goto reallocate;
2634         } else if (!err) {
2635                 update_device_state(fio);
2636         }
2637 }
2638
2639 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2640                                         enum iostat_type io_type)
2641 {
2642         struct f2fs_io_info fio = {
2643                 .sbi = sbi,
2644                 .type = META,
2645                 .op = REQ_OP_WRITE,
2646                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
2647                 .old_blkaddr = page->index,
2648                 .new_blkaddr = page->index,
2649                 .page = page,
2650                 .encrypted_page = NULL,
2651                 .in_list = false,
2652         };
2653
2654         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
2655                 fio.op_flags &= ~REQ_META;
2656
2657         set_page_writeback(page);
2658         f2fs_submit_page_write(&fio);
2659
2660         f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
2661 }
2662
2663 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
2664 {
2665         struct f2fs_summary sum;
2666
2667         set_summary(&sum, nid, 0, 0);
2668         do_write_page(&sum, fio);
2669
2670         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2671 }
2672
2673 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
2674 {
2675         struct f2fs_sb_info *sbi = fio->sbi;
2676         struct f2fs_summary sum;
2677         struct node_info ni;
2678
2679         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
2680         get_node_info(sbi, dn->nid, &ni);
2681         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
2682         do_write_page(&sum, fio);
2683         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
2684
2685         f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
2686 }
2687
2688 int rewrite_data_page(struct f2fs_io_info *fio)
2689 {
2690         int err;
2691
2692         fio->new_blkaddr = fio->old_blkaddr;
2693         stat_inc_inplace_blocks(fio->sbi);
2694
2695         err = f2fs_submit_page_bio(fio);
2696         if (!err)
2697                 update_device_state(fio);
2698
2699         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2700
2701         return err;
2702 }
2703
2704 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
2705                                                 unsigned int segno)
2706 {
2707         int i;
2708
2709         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
2710                 if (CURSEG_I(sbi, i)->segno == segno)
2711                         break;
2712         }
2713         return i;
2714 }
2715
2716 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
2717                                 block_t old_blkaddr, block_t new_blkaddr,
2718                                 bool recover_curseg, bool recover_newaddr)
2719 {
2720         struct sit_info *sit_i = SIT_I(sbi);
2721         struct curseg_info *curseg;
2722         unsigned int segno, old_cursegno;
2723         struct seg_entry *se;
2724         int type;
2725         unsigned short old_blkoff;
2726
2727         segno = GET_SEGNO(sbi, new_blkaddr);
2728         se = get_seg_entry(sbi, segno);
2729         type = se->type;
2730
2731         down_write(&SM_I(sbi)->curseg_lock);
2732
2733         if (!recover_curseg) {
2734                 /* for recovery flow */
2735                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2736                         if (old_blkaddr == NULL_ADDR)
2737                                 type = CURSEG_COLD_DATA;
2738                         else
2739                                 type = CURSEG_WARM_DATA;
2740                 }
2741         } else {
2742                 if (IS_CURSEG(sbi, segno)) {
2743                         /* se->type is volatile as SSR allocation */
2744                         type = __f2fs_get_curseg(sbi, segno);
2745                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
2746                 } else {
2747                         type = CURSEG_WARM_DATA;
2748                 }
2749         }
2750
2751         f2fs_bug_on(sbi, !IS_DATASEG(type));
2752         curseg = CURSEG_I(sbi, type);
2753
2754         mutex_lock(&curseg->curseg_mutex);
2755         down_write(&sit_i->sentry_lock);
2756
2757         old_cursegno = curseg->segno;
2758         old_blkoff = curseg->next_blkoff;
2759
2760         /* change the current segment */
2761         if (segno != curseg->segno) {
2762                 curseg->next_segno = segno;
2763                 change_curseg(sbi, type);
2764         }
2765
2766         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
2767         __add_sum_entry(sbi, type, sum);
2768
2769         if (!recover_curseg || recover_newaddr)
2770                 update_sit_entry(sbi, new_blkaddr, 1);
2771         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2772                 update_sit_entry(sbi, old_blkaddr, -1);
2773
2774         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2775         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2776
2777         locate_dirty_segment(sbi, old_cursegno);
2778
2779         if (recover_curseg) {
2780                 if (old_cursegno != curseg->segno) {
2781                         curseg->next_segno = old_cursegno;
2782                         change_curseg(sbi, type);
2783                 }
2784                 curseg->next_blkoff = old_blkoff;
2785         }
2786
2787         up_write(&sit_i->sentry_lock);
2788         mutex_unlock(&curseg->curseg_mutex);
2789         up_write(&SM_I(sbi)->curseg_lock);
2790 }
2791
2792 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2793                                 block_t old_addr, block_t new_addr,
2794                                 unsigned char version, bool recover_curseg,
2795                                 bool recover_newaddr)
2796 {
2797         struct f2fs_summary sum;
2798
2799         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2800
2801         __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2802                                         recover_curseg, recover_newaddr);
2803
2804         f2fs_update_data_blkaddr(dn, new_addr);
2805 }
2806
2807 void f2fs_wait_on_page_writeback(struct page *page,
2808                                 enum page_type type, bool ordered)
2809 {
2810         if (PageWriteback(page)) {
2811                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2812
2813                 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2814                                                 0, page->index, type);
2815                 if (ordered)
2816                         wait_on_page_writeback(page);
2817                 else
2818                         wait_for_stable_page(page);
2819         }
2820 }
2821
2822 void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
2823 {
2824         struct page *cpage;
2825
2826         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
2827                 return;
2828
2829         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2830         if (cpage) {
2831                 f2fs_wait_on_page_writeback(cpage, DATA, true);
2832                 f2fs_put_page(cpage, 1);
2833         }
2834 }
2835
2836 static void read_compacted_summaries(struct f2fs_sb_info *sbi)
2837 {
2838         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2839         struct curseg_info *seg_i;
2840         unsigned char *kaddr;
2841         struct page *page;
2842         block_t start;
2843         int i, j, offset;
2844
2845         start = start_sum_block(sbi);
2846
2847         page = get_meta_page(sbi, start++);
2848         kaddr = (unsigned char *)page_address(page);
2849
2850         /* Step 1: restore nat cache */
2851         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2852         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
2853
2854         /* Step 2: restore sit cache */
2855         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2856         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
2857         offset = 2 * SUM_JOURNAL_SIZE;
2858
2859         /* Step 3: restore summary entries */
2860         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2861                 unsigned short blk_off;
2862                 unsigned int segno;
2863
2864                 seg_i = CURSEG_I(sbi, i);
2865                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2866                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2867                 seg_i->next_segno = segno;
2868                 reset_curseg(sbi, i, 0);
2869                 seg_i->alloc_type = ckpt->alloc_type[i];
2870                 seg_i->next_blkoff = blk_off;
2871
2872                 if (seg_i->alloc_type == SSR)
2873                         blk_off = sbi->blocks_per_seg;
2874
2875                 for (j = 0; j < blk_off; j++) {
2876                         struct f2fs_summary *s;
2877                         s = (struct f2fs_summary *)(kaddr + offset);
2878                         seg_i->sum_blk->entries[j] = *s;
2879                         offset += SUMMARY_SIZE;
2880                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
2881                                                 SUM_FOOTER_SIZE)
2882                                 continue;
2883
2884                         f2fs_put_page(page, 1);
2885                         page = NULL;
2886
2887                         page = get_meta_page(sbi, start++);
2888                         kaddr = (unsigned char *)page_address(page);
2889                         offset = 0;
2890                 }
2891         }
2892         f2fs_put_page(page, 1);
2893 }
2894
2895 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2896 {
2897         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2898         struct f2fs_summary_block *sum;
2899         struct curseg_info *curseg;
2900         struct page *new;
2901         unsigned short blk_off;
2902         unsigned int segno = 0;
2903         block_t blk_addr = 0;
2904
2905         /* get segment number and block addr */
2906         if (IS_DATASEG(type)) {
2907                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2908                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2909                                                         CURSEG_HOT_DATA]);
2910                 if (__exist_node_summaries(sbi))
2911                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2912                 else
2913                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2914         } else {
2915                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2916                                                         CURSEG_HOT_NODE]);
2917                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2918                                                         CURSEG_HOT_NODE]);
2919                 if (__exist_node_summaries(sbi))
2920                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2921                                                         type - CURSEG_HOT_NODE);
2922                 else
2923                         blk_addr = GET_SUM_BLOCK(sbi, segno);
2924         }
2925
2926         new = get_meta_page(sbi, blk_addr);
2927         sum = (struct f2fs_summary_block *)page_address(new);
2928
2929         if (IS_NODESEG(type)) {
2930                 if (__exist_node_summaries(sbi)) {
2931                         struct f2fs_summary *ns = &sum->entries[0];
2932                         int i;
2933                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2934                                 ns->version = 0;
2935                                 ns->ofs_in_node = 0;
2936                         }
2937                 } else {
2938                         restore_node_summary(sbi, segno, sum);
2939                 }
2940         }
2941
2942         /* set uncompleted segment to curseg */
2943         curseg = CURSEG_I(sbi, type);
2944         mutex_lock(&curseg->curseg_mutex);
2945
2946         /* update journal info */
2947         down_write(&curseg->journal_rwsem);
2948         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2949         up_write(&curseg->journal_rwsem);
2950
2951         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2952         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
2953         curseg->next_segno = segno;
2954         reset_curseg(sbi, type, 0);
2955         curseg->alloc_type = ckpt->alloc_type[type];
2956         curseg->next_blkoff = blk_off;
2957         mutex_unlock(&curseg->curseg_mutex);
2958         f2fs_put_page(new, 1);
2959         return 0;
2960 }
2961
2962 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2963 {
2964         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
2965         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
2966         int type = CURSEG_HOT_DATA;
2967         int err;
2968
2969         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
2970                 int npages = npages_for_summary_flush(sbi, true);
2971
2972                 if (npages >= 2)
2973                         ra_meta_pages(sbi, start_sum_block(sbi), npages,
2974                                                         META_CP, true);
2975
2976                 /* restore for compacted data summary */
2977                 read_compacted_summaries(sbi);
2978                 type = CURSEG_HOT_NODE;
2979         }
2980
2981         if (__exist_node_summaries(sbi))
2982                 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
2983                                         NR_CURSEG_TYPE - type, META_CP, true);
2984
2985         for (; type <= CURSEG_COLD_NODE; type++) {
2986                 err = read_normal_summaries(sbi, type);
2987                 if (err)
2988                         return err;
2989         }
2990
2991         /* sanity check for summary blocks */
2992         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
2993                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
2994                 return -EINVAL;
2995
2996         return 0;
2997 }
2998
2999 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3000 {
3001         struct page *page;
3002         unsigned char *kaddr;
3003         struct f2fs_summary *summary;
3004         struct curseg_info *seg_i;
3005         int written_size = 0;
3006         int i, j;
3007
3008         page = grab_meta_page(sbi, blkaddr++);
3009         kaddr = (unsigned char *)page_address(page);
3010
3011         /* Step 1: write nat cache */
3012         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3013         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3014         written_size += SUM_JOURNAL_SIZE;
3015
3016         /* Step 2: write sit cache */
3017         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3018         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3019         written_size += SUM_JOURNAL_SIZE;
3020
3021         /* Step 3: write summary entries */
3022         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3023                 unsigned short blkoff;
3024                 seg_i = CURSEG_I(sbi, i);
3025                 if (sbi->ckpt->alloc_type[i] == SSR)
3026                         blkoff = sbi->blocks_per_seg;
3027                 else
3028                         blkoff = curseg_blkoff(sbi, i);
3029
3030                 for (j = 0; j < blkoff; j++) {
3031                         if (!page) {
3032                                 page = grab_meta_page(sbi, blkaddr++);
3033                                 kaddr = (unsigned char *)page_address(page);
3034                                 written_size = 0;
3035                         }
3036                         summary = (struct f2fs_summary *)(kaddr + written_size);
3037                         *summary = seg_i->sum_blk->entries[j];
3038                         written_size += SUMMARY_SIZE;
3039
3040                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3041                                                         SUM_FOOTER_SIZE)
3042                                 continue;
3043
3044                         set_page_dirty(page);
3045                         f2fs_put_page(page, 1);
3046                         page = NULL;
3047                 }
3048         }
3049         if (page) {
3050                 set_page_dirty(page);
3051                 f2fs_put_page(page, 1);
3052         }
3053 }
3054
3055 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3056                                         block_t blkaddr, int type)
3057 {
3058         int i, end;
3059         if (IS_DATASEG(type))
3060                 end = type + NR_CURSEG_DATA_TYPE;
3061         else
3062                 end = type + NR_CURSEG_NODE_TYPE;
3063
3064         for (i = type; i < end; i++)
3065                 write_current_sum_page(sbi, i, blkaddr + (i - type));
3066 }
3067
3068 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3069 {
3070         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3071                 write_compacted_summaries(sbi, start_blk);
3072         else
3073                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3074 }
3075
3076 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3077 {
3078         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3079 }
3080
3081 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3082                                         unsigned int val, int alloc)
3083 {
3084         int i;
3085
3086         if (type == NAT_JOURNAL) {
3087                 for (i = 0; i < nats_in_cursum(journal); i++) {
3088                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3089                                 return i;
3090                 }
3091                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3092                         return update_nats_in_cursum(journal, 1);
3093         } else if (type == SIT_JOURNAL) {
3094                 for (i = 0; i < sits_in_cursum(journal); i++)
3095                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3096                                 return i;
3097                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3098                         return update_sits_in_cursum(journal, 1);
3099         }
3100         return -1;
3101 }
3102
3103 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3104                                         unsigned int segno)
3105 {
3106         return get_meta_page(sbi, current_sit_addr(sbi, segno));
3107 }
3108
3109 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3110                                         unsigned int start)
3111 {
3112         struct sit_info *sit_i = SIT_I(sbi);
3113         struct page *page;
3114         pgoff_t src_off, dst_off;
3115
3116         src_off = current_sit_addr(sbi, start);
3117         dst_off = next_sit_addr(sbi, src_off);
3118
3119         page = grab_meta_page(sbi, dst_off);
3120         seg_info_to_sit_page(sbi, page, start);
3121
3122         set_page_dirty(page);
3123         set_to_next_sit(sit_i, start);
3124
3125         return page;
3126 }
3127
3128 static struct sit_entry_set *grab_sit_entry_set(void)
3129 {
3130         struct sit_entry_set *ses =
3131                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3132
3133         ses->entry_cnt = 0;
3134         INIT_LIST_HEAD(&ses->set_list);
3135         return ses;
3136 }
3137
3138 static void release_sit_entry_set(struct sit_entry_set *ses)
3139 {
3140         list_del(&ses->set_list);
3141         kmem_cache_free(sit_entry_set_slab, ses);
3142 }
3143
3144 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3145                                                 struct list_head *head)
3146 {
3147         struct sit_entry_set *next = ses;
3148
3149         if (list_is_last(&ses->set_list, head))
3150                 return;
3151
3152         list_for_each_entry_continue(next, head, set_list)
3153                 if (ses->entry_cnt <= next->entry_cnt)
3154                         break;
3155
3156         list_move_tail(&ses->set_list, &next->set_list);
3157 }
3158
3159 static void add_sit_entry(unsigned int segno, struct list_head *head)
3160 {
3161         struct sit_entry_set *ses;
3162         unsigned int start_segno = START_SEGNO(segno);
3163
3164         list_for_each_entry(ses, head, set_list) {
3165                 if (ses->start_segno == start_segno) {
3166                         ses->entry_cnt++;
3167                         adjust_sit_entry_set(ses, head);
3168                         return;
3169                 }
3170         }
3171
3172         ses = grab_sit_entry_set();
3173
3174         ses->start_segno = start_segno;
3175         ses->entry_cnt++;
3176         list_add(&ses->set_list, head);
3177 }
3178
3179 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3180 {
3181         struct f2fs_sm_info *sm_info = SM_I(sbi);
3182         struct list_head *set_list = &sm_info->sit_entry_set;
3183         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3184         unsigned int segno;
3185
3186         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3187                 add_sit_entry(segno, set_list);
3188 }
3189
3190 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3191 {
3192         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3193         struct f2fs_journal *journal = curseg->journal;
3194         int i;
3195
3196         down_write(&curseg->journal_rwsem);
3197         for (i = 0; i < sits_in_cursum(journal); i++) {
3198                 unsigned int segno;
3199                 bool dirtied;
3200
3201                 segno = le32_to_cpu(segno_in_journal(journal, i));
3202                 dirtied = __mark_sit_entry_dirty(sbi, segno);
3203
3204                 if (!dirtied)
3205                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3206         }
3207         update_sits_in_cursum(journal, -i);
3208         up_write(&curseg->journal_rwsem);
3209 }
3210
3211 /*
3212  * CP calls this function, which flushes SIT entries including sit_journal,
3213  * and moves prefree segs to free segs.
3214  */
3215 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3216 {
3217         struct sit_info *sit_i = SIT_I(sbi);
3218         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3219         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3220         struct f2fs_journal *journal = curseg->journal;
3221         struct sit_entry_set *ses, *tmp;
3222         struct list_head *head = &SM_I(sbi)->sit_entry_set;
3223         bool to_journal = true;
3224         struct seg_entry *se;
3225
3226         down_write(&sit_i->sentry_lock);
3227
3228         if (!sit_i->dirty_sentries)
3229                 goto out;
3230
3231         /*
3232          * add and account sit entries of dirty bitmap in sit entry
3233          * set temporarily
3234          */
3235         add_sits_in_set(sbi);
3236
3237         /*
3238          * if there are no enough space in journal to store dirty sit
3239          * entries, remove all entries from journal and add and account
3240          * them in sit entry set.
3241          */
3242         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3243                 remove_sits_in_journal(sbi);
3244
3245         /*
3246          * there are two steps to flush sit entries:
3247          * #1, flush sit entries to journal in current cold data summary block.
3248          * #2, flush sit entries to sit page.
3249          */
3250         list_for_each_entry_safe(ses, tmp, head, set_list) {
3251                 struct page *page = NULL;
3252                 struct f2fs_sit_block *raw_sit = NULL;
3253                 unsigned int start_segno = ses->start_segno;
3254                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3255                                                 (unsigned long)MAIN_SEGS(sbi));
3256                 unsigned int segno = start_segno;
3257
3258                 if (to_journal &&
3259                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3260                         to_journal = false;
3261
3262                 if (to_journal) {
3263                         down_write(&curseg->journal_rwsem);
3264                 } else {
3265                         page = get_next_sit_page(sbi, start_segno);
3266                         raw_sit = page_address(page);
3267                 }
3268
3269                 /* flush dirty sit entries in region of current sit set */
3270                 for_each_set_bit_from(segno, bitmap, end) {
3271                         int offset, sit_offset;
3272
3273                         se = get_seg_entry(sbi, segno);
3274
3275                         /* add discard candidates */
3276                         if (!(cpc->reason & CP_DISCARD)) {
3277                                 cpc->trim_start = segno;
3278                                 add_discard_addrs(sbi, cpc, false);
3279                         }
3280
3281                         if (to_journal) {
3282                                 offset = lookup_journal_in_cursum(journal,
3283                                                         SIT_JOURNAL, segno, 1);
3284                                 f2fs_bug_on(sbi, offset < 0);
3285                                 segno_in_journal(journal, offset) =
3286                                                         cpu_to_le32(segno);
3287                                 seg_info_to_raw_sit(se,
3288                                         &sit_in_journal(journal, offset));
3289                         } else {
3290                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3291                                 seg_info_to_raw_sit(se,
3292                                                 &raw_sit->entries[sit_offset]);
3293                         }
3294
3295                         __clear_bit(segno, bitmap);
3296                         sit_i->dirty_sentries--;
3297                         ses->entry_cnt--;
3298                 }
3299
3300                 if (to_journal)
3301                         up_write(&curseg->journal_rwsem);
3302                 else
3303                         f2fs_put_page(page, 1);
3304
3305                 f2fs_bug_on(sbi, ses->entry_cnt);
3306                 release_sit_entry_set(ses);
3307         }
3308
3309         f2fs_bug_on(sbi, !list_empty(head));
3310         f2fs_bug_on(sbi, sit_i->dirty_sentries);
3311 out:
3312         if (cpc->reason & CP_DISCARD) {
3313                 __u64 trim_start = cpc->trim_start;
3314
3315                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3316                         add_discard_addrs(sbi, cpc, false);
3317
3318                 cpc->trim_start = trim_start;
3319         }
3320         up_write(&sit_i->sentry_lock);
3321
3322         set_prefree_as_free_segments(sbi);
3323 }
3324
3325 static int build_sit_info(struct f2fs_sb_info *sbi)
3326 {
3327         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3328         struct sit_info *sit_i;
3329         unsigned int sit_segs, start;
3330         char *src_bitmap;
3331         unsigned int bitmap_size;
3332
3333         /* allocate memory for SIT information */
3334         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3335         if (!sit_i)
3336                 return -ENOMEM;
3337
3338         SM_I(sbi)->sit_info = sit_i;
3339
3340         sit_i->sentries = f2fs_kvzalloc(sbi, MAIN_SEGS(sbi) *
3341                                         sizeof(struct seg_entry), GFP_KERNEL);
3342         if (!sit_i->sentries)
3343                 return -ENOMEM;
3344
3345         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3346         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3347                                                                 GFP_KERNEL);
3348         if (!sit_i->dirty_sentries_bitmap)
3349                 return -ENOMEM;
3350
3351         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3352                 sit_i->sentries[start].cur_valid_map
3353                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3354                 sit_i->sentries[start].ckpt_valid_map
3355                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3356                 if (!sit_i->sentries[start].cur_valid_map ||
3357                                 !sit_i->sentries[start].ckpt_valid_map)
3358                         return -ENOMEM;
3359
3360 #ifdef CONFIG_F2FS_CHECK_FS
3361                 sit_i->sentries[start].cur_valid_map_mir
3362                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3363                 if (!sit_i->sentries[start].cur_valid_map_mir)
3364                         return -ENOMEM;
3365 #endif
3366
3367                 if (f2fs_discard_en(sbi)) {
3368                         sit_i->sentries[start].discard_map
3369                                 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3370                                                                 GFP_KERNEL);
3371                         if (!sit_i->sentries[start].discard_map)
3372                                 return -ENOMEM;
3373                 }
3374         }
3375
3376         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3377         if (!sit_i->tmp_map)
3378                 return -ENOMEM;
3379
3380         if (sbi->segs_per_sec > 1) {
3381                 sit_i->sec_entries = f2fs_kvzalloc(sbi, MAIN_SECS(sbi) *
3382                                         sizeof(struct sec_entry), GFP_KERNEL);
3383                 if (!sit_i->sec_entries)
3384                         return -ENOMEM;
3385         }
3386
3387         /* get information related with SIT */
3388         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3389
3390         /* setup SIT bitmap from ckeckpoint pack */
3391         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3392         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3393
3394         sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3395         if (!sit_i->sit_bitmap)
3396                 return -ENOMEM;
3397
3398 #ifdef CONFIG_F2FS_CHECK_FS
3399         sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3400         if (!sit_i->sit_bitmap_mir)
3401                 return -ENOMEM;
3402 #endif
3403
3404         /* init SIT information */
3405         sit_i->s_ops = &default_salloc_ops;
3406
3407         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3408         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
3409         sit_i->written_valid_blocks = 0;
3410         sit_i->bitmap_size = bitmap_size;
3411         sit_i->dirty_sentries = 0;
3412         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3413         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3414         sit_i->mounted_time = ktime_get_real_seconds();
3415         init_rwsem(&sit_i->sentry_lock);
3416         return 0;
3417 }
3418
3419 static int build_free_segmap(struct f2fs_sb_info *sbi)
3420 {
3421         struct free_segmap_info *free_i;
3422         unsigned int bitmap_size, sec_bitmap_size;
3423
3424         /* allocate memory for free segmap information */
3425         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
3426         if (!free_i)
3427                 return -ENOMEM;
3428
3429         SM_I(sbi)->free_info = free_i;
3430
3431         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3432         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
3433         if (!free_i->free_segmap)
3434                 return -ENOMEM;
3435
3436         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3437         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
3438         if (!free_i->free_secmap)
3439                 return -ENOMEM;
3440
3441         /* set all segments as dirty temporarily */
3442         memset(free_i->free_segmap, 0xff, bitmap_size);
3443         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3444
3445         /* init free segmap information */
3446         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
3447         free_i->free_segments = 0;
3448         free_i->free_sections = 0;
3449         spin_lock_init(&free_i->segmap_lock);
3450         return 0;
3451 }
3452
3453 static int build_curseg(struct f2fs_sb_info *sbi)
3454 {
3455         struct curseg_info *array;
3456         int i;
3457
3458         array = f2fs_kzalloc(sbi, sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
3459         if (!array)
3460                 return -ENOMEM;
3461
3462         SM_I(sbi)->curseg_array = array;
3463
3464         for (i = 0; i < NR_CURSEG_TYPE; i++) {
3465                 mutex_init(&array[i].curseg_mutex);
3466                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
3467                 if (!array[i].sum_blk)
3468                         return -ENOMEM;
3469                 init_rwsem(&array[i].journal_rwsem);
3470                 array[i].journal = f2fs_kzalloc(sbi,
3471                                 sizeof(struct f2fs_journal), GFP_KERNEL);
3472                 if (!array[i].journal)
3473                         return -ENOMEM;
3474                 array[i].segno = NULL_SEGNO;
3475                 array[i].next_blkoff = 0;
3476         }
3477         return restore_curseg_summaries(sbi);
3478 }
3479
3480 static int build_sit_entries(struct f2fs_sb_info *sbi)
3481 {
3482         struct sit_info *sit_i = SIT_I(sbi);
3483         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3484         struct f2fs_journal *journal = curseg->journal;
3485         struct seg_entry *se;
3486         struct f2fs_sit_entry sit;
3487         int sit_blk_cnt = SIT_BLK_CNT(sbi);
3488         unsigned int i, start, end;
3489         unsigned int readed, start_blk = 0;
3490         int err = 0;
3491
3492         do {
3493                 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3494                                                         META_SIT, true);
3495
3496                 start = start_blk * sit_i->sents_per_block;
3497                 end = (start_blk + readed) * sit_i->sents_per_block;
3498
3499                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
3500                         struct f2fs_sit_block *sit_blk;
3501                         struct page *page;
3502
3503                         se = &sit_i->sentries[start];
3504                         page = get_current_sit_page(sbi, start);
3505                         sit_blk = (struct f2fs_sit_block *)page_address(page);
3506                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3507                         f2fs_put_page(page, 1);
3508
3509                         err = check_block_count(sbi, start, &sit);
3510                         if (err)
3511                                 return err;
3512                         seg_info_from_raw_sit(se, &sit);
3513
3514                         /* build discard map only one time */
3515                         if (f2fs_discard_en(sbi)) {
3516                                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3517                                         memset(se->discard_map, 0xff,
3518                                                 SIT_VBLOCK_MAP_SIZE);
3519                                 } else {
3520                                         memcpy(se->discard_map,
3521                                                 se->cur_valid_map,
3522                                                 SIT_VBLOCK_MAP_SIZE);
3523                                         sbi->discard_blks +=
3524                                                 sbi->blocks_per_seg -
3525                                                 se->valid_blocks;
3526                                 }
3527                         }
3528
3529                         if (sbi->segs_per_sec > 1)
3530                                 get_sec_entry(sbi, start)->valid_blocks +=
3531                                                         se->valid_blocks;
3532                 }
3533                 start_blk += readed;
3534         } while (start_blk < sit_blk_cnt);
3535
3536         down_read(&curseg->journal_rwsem);
3537         for (i = 0; i < sits_in_cursum(journal); i++) {
3538                 unsigned int old_valid_blocks;
3539
3540                 start = le32_to_cpu(segno_in_journal(journal, i));
3541                 se = &sit_i->sentries[start];
3542                 sit = sit_in_journal(journal, i);
3543
3544                 old_valid_blocks = se->valid_blocks;
3545
3546                 err = check_block_count(sbi, start, &sit);
3547                 if (err)
3548                         break;
3549                 seg_info_from_raw_sit(se, &sit);
3550
3551                 if (f2fs_discard_en(sbi)) {
3552                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3553                                 memset(se->discard_map, 0xff,
3554                                                         SIT_VBLOCK_MAP_SIZE);
3555                         } else {
3556                                 memcpy(se->discard_map, se->cur_valid_map,
3557                                                         SIT_VBLOCK_MAP_SIZE);
3558                                 sbi->discard_blks += old_valid_blocks -
3559                                                         se->valid_blocks;
3560                         }
3561                 }
3562
3563                 if (sbi->segs_per_sec > 1)
3564                         get_sec_entry(sbi, start)->valid_blocks +=
3565                                 se->valid_blocks - old_valid_blocks;
3566         }
3567         up_read(&curseg->journal_rwsem);
3568         return err;
3569 }
3570
3571 static void init_free_segmap(struct f2fs_sb_info *sbi)
3572 {
3573         unsigned int start;
3574         int type;
3575
3576         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3577                 struct seg_entry *sentry = get_seg_entry(sbi, start);
3578                 if (!sentry->valid_blocks)
3579                         __set_free(sbi, start);
3580                 else
3581                         SIT_I(sbi)->written_valid_blocks +=
3582                                                 sentry->valid_blocks;
3583         }
3584
3585         /* set use the current segments */
3586         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3587                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3588                 __set_test_and_inuse(sbi, curseg_t->segno);
3589         }
3590 }
3591
3592 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3593 {
3594         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3595         struct free_segmap_info *free_i = FREE_I(sbi);
3596         unsigned int segno = 0, offset = 0;
3597         unsigned short valid_blocks;
3598
3599         while (1) {
3600                 /* find dirty segment based on free segmap */
3601                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3602                 if (segno >= MAIN_SEGS(sbi))
3603                         break;
3604                 offset = segno + 1;
3605                 valid_blocks = get_valid_blocks(sbi, segno, false);
3606                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
3607                         continue;
3608                 if (valid_blocks > sbi->blocks_per_seg) {
3609                         f2fs_bug_on(sbi, 1);
3610                         continue;
3611                 }
3612                 mutex_lock(&dirty_i->seglist_lock);
3613                 __locate_dirty_segment(sbi, segno, DIRTY);
3614                 mutex_unlock(&dirty_i->seglist_lock);
3615         }
3616 }
3617
3618 static int init_victim_secmap(struct f2fs_sb_info *sbi)
3619 {
3620         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3621         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3622
3623         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
3624         if (!dirty_i->victim_secmap)
3625                 return -ENOMEM;
3626         return 0;
3627 }
3628
3629 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3630 {
3631         struct dirty_seglist_info *dirty_i;
3632         unsigned int bitmap_size, i;
3633
3634         /* allocate memory for dirty segments list information */
3635         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
3636                                                                 GFP_KERNEL);
3637         if (!dirty_i)
3638                 return -ENOMEM;
3639
3640         SM_I(sbi)->dirty_info = dirty_i;
3641         mutex_init(&dirty_i->seglist_lock);
3642
3643         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3644
3645         for (i = 0; i < NR_DIRTY_TYPE; i++) {
3646                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
3647                                                                 GFP_KERNEL);
3648                 if (!dirty_i->dirty_segmap[i])
3649                         return -ENOMEM;
3650         }
3651
3652         init_dirty_segmap(sbi);
3653         return init_victim_secmap(sbi);
3654 }
3655
3656 /*
3657  * Update min, max modified time for cost-benefit GC algorithm
3658  */
3659 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3660 {
3661         struct sit_info *sit_i = SIT_I(sbi);
3662         unsigned int segno;
3663
3664         down_write(&sit_i->sentry_lock);
3665
3666         sit_i->min_mtime = LLONG_MAX;
3667
3668         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
3669                 unsigned int i;
3670                 unsigned long long mtime = 0;
3671
3672                 for (i = 0; i < sbi->segs_per_sec; i++)
3673                         mtime += get_seg_entry(sbi, segno + i)->mtime;
3674
3675                 mtime = div_u64(mtime, sbi->segs_per_sec);
3676
3677                 if (sit_i->min_mtime > mtime)
3678                         sit_i->min_mtime = mtime;
3679         }
3680         sit_i->max_mtime = get_mtime(sbi);
3681         up_write(&sit_i->sentry_lock);
3682 }
3683
3684 int build_segment_manager(struct f2fs_sb_info *sbi)
3685 {
3686         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3687         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3688         struct f2fs_sm_info *sm_info;
3689         int err;
3690
3691         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
3692         if (!sm_info)
3693                 return -ENOMEM;
3694
3695         /* init sm info */
3696         sbi->sm_info = sm_info;
3697         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3698         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3699         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3700         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3701         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3702         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3703         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
3704         sm_info->rec_prefree_segments = sm_info->main_segments *
3705                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
3706         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3707                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3708
3709         if (!test_opt(sbi, LFS))
3710                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
3711         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
3712         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
3713         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
3714         sm_info->min_ssr_sections = reserved_sections(sbi);
3715
3716         sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
3717
3718         INIT_LIST_HEAD(&sm_info->sit_entry_set);
3719
3720         init_rwsem(&sm_info->curseg_lock);
3721
3722         if (!f2fs_readonly(sbi->sb)) {
3723                 err = create_flush_cmd_control(sbi);
3724                 if (err)
3725                         return err;
3726         }
3727
3728         err = create_discard_cmd_control(sbi);
3729         if (err)
3730                 return err;
3731
3732         err = build_sit_info(sbi);
3733         if (err)
3734                 return err;
3735         err = build_free_segmap(sbi);
3736         if (err)
3737                 return err;
3738         err = build_curseg(sbi);
3739         if (err)
3740                 return err;
3741
3742         /* reinit free segmap based on SIT */
3743         err = build_sit_entries(sbi);
3744         if (err)
3745                 return err;
3746
3747         init_free_segmap(sbi);
3748         err = build_dirty_segmap(sbi);
3749         if (err)
3750                 return err;
3751
3752         init_min_max_mtime(sbi);
3753         return 0;
3754 }
3755
3756 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3757                 enum dirty_type dirty_type)
3758 {
3759         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3760
3761         mutex_lock(&dirty_i->seglist_lock);
3762         kvfree(dirty_i->dirty_segmap[dirty_type]);
3763         dirty_i->nr_dirty[dirty_type] = 0;
3764         mutex_unlock(&dirty_i->seglist_lock);
3765 }
3766
3767 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
3768 {
3769         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3770         kvfree(dirty_i->victim_secmap);
3771 }
3772
3773 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3774 {
3775         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3776         int i;
3777
3778         if (!dirty_i)
3779                 return;
3780
3781         /* discard pre-free/dirty segments list */
3782         for (i = 0; i < NR_DIRTY_TYPE; i++)
3783                 discard_dirty_segmap(sbi, i);
3784
3785         destroy_victim_secmap(sbi);
3786         SM_I(sbi)->dirty_info = NULL;
3787         kfree(dirty_i);
3788 }
3789
3790 static void destroy_curseg(struct f2fs_sb_info *sbi)
3791 {
3792         struct curseg_info *array = SM_I(sbi)->curseg_array;
3793         int i;
3794
3795         if (!array)
3796                 return;
3797         SM_I(sbi)->curseg_array = NULL;
3798         for (i = 0; i < NR_CURSEG_TYPE; i++) {
3799                 kfree(array[i].sum_blk);
3800                 kfree(array[i].journal);
3801         }
3802         kfree(array);
3803 }
3804
3805 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3806 {
3807         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3808         if (!free_i)
3809                 return;
3810         SM_I(sbi)->free_info = NULL;
3811         kvfree(free_i->free_segmap);
3812         kvfree(free_i->free_secmap);
3813         kfree(free_i);
3814 }
3815
3816 static void destroy_sit_info(struct f2fs_sb_info *sbi)
3817 {
3818         struct sit_info *sit_i = SIT_I(sbi);
3819         unsigned int start;
3820
3821         if (!sit_i)
3822                 return;
3823
3824         if (sit_i->sentries) {
3825                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3826                         kfree(sit_i->sentries[start].cur_valid_map);
3827 #ifdef CONFIG_F2FS_CHECK_FS
3828                         kfree(sit_i->sentries[start].cur_valid_map_mir);
3829 #endif
3830                         kfree(sit_i->sentries[start].ckpt_valid_map);
3831                         kfree(sit_i->sentries[start].discard_map);
3832                 }
3833         }
3834         kfree(sit_i->tmp_map);
3835
3836         kvfree(sit_i->sentries);
3837         kvfree(sit_i->sec_entries);
3838         kvfree(sit_i->dirty_sentries_bitmap);
3839
3840         SM_I(sbi)->sit_info = NULL;
3841         kfree(sit_i->sit_bitmap);
3842 #ifdef CONFIG_F2FS_CHECK_FS
3843         kfree(sit_i->sit_bitmap_mir);
3844 #endif
3845         kfree(sit_i);
3846 }
3847
3848 void destroy_segment_manager(struct f2fs_sb_info *sbi)
3849 {
3850         struct f2fs_sm_info *sm_info = SM_I(sbi);
3851
3852         if (!sm_info)
3853                 return;
3854         destroy_flush_cmd_control(sbi, true);
3855         destroy_discard_cmd_control(sbi);
3856         destroy_dirty_segmap(sbi);
3857         destroy_curseg(sbi);
3858         destroy_free_segmap(sbi);
3859         destroy_sit_info(sbi);
3860         sbi->sm_info = NULL;
3861         kfree(sm_info);
3862 }
3863
3864 int __init create_segment_manager_caches(void)
3865 {
3866         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
3867                         sizeof(struct discard_entry));
3868         if (!discard_entry_slab)
3869                 goto fail;
3870
3871         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3872                         sizeof(struct discard_cmd));
3873         if (!discard_cmd_slab)
3874                 goto destroy_discard_entry;
3875
3876         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
3877                         sizeof(struct sit_entry_set));
3878         if (!sit_entry_set_slab)
3879                 goto destroy_discard_cmd;
3880
3881         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3882                         sizeof(struct inmem_pages));
3883         if (!inmem_entry_slab)
3884                 goto destroy_sit_entry_set;
3885         return 0;
3886
3887 destroy_sit_entry_set:
3888         kmem_cache_destroy(sit_entry_set_slab);
3889 destroy_discard_cmd:
3890         kmem_cache_destroy(discard_cmd_slab);
3891 destroy_discard_entry:
3892         kmem_cache_destroy(discard_entry_slab);
3893 fail:
3894         return -ENOMEM;
3895 }
3896
3897 void destroy_segment_manager_caches(void)
3898 {
3899         kmem_cache_destroy(sit_entry_set_slab);
3900         kmem_cache_destroy(discard_cmd_slab);
3901         kmem_cache_destroy(discard_entry_slab);
3902         kmem_cache_destroy(inmem_entry_slab);
3903 }