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
  20 #include "f2fs.h"
  21 #include "segment.h"
  22 #include "node.h"
  23 #include "trace.h"
  24 #include <trace/events/f2fs.h>
  25
  26 #define __reverse_ffz(x) __reverse_ffs(~(x))
  27
  28 static struct kmem_cache *discard_entry_slab;
  29 static struct kmem_cache *bio_entry_slab;
  30 static struct kmem_cache *sit_entry_set_slab;
  31 static struct kmem_cache *inmem_entry_slab;
  32
  33 static unsigned long __reverse_ulong(unsigned char *str)
  34 {
  35         unsigned long tmp = 0;
  36         int shift = 24, idx = 0;
  37
  38 #if BITS_PER_LONG == 64
  39         shift = 56;
  40 #endif
  41         while (shift >= 0) {
  42                 tmp |= (unsigned long)str[idx++] << shift;
  43                 shift -= BITS_PER_BYTE;
  44         }
  45         return tmp;
  46 }
  47
  48 /*
  49  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  50  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  51  */
  52 static inline unsigned long __reverse_ffs(unsigned long word)
  53 {
  54         int num = 0;
  55
  56 #if BITS_PER_LONG == 64
  57         if ((word & 0xffffffff00000000UL) == 0)
  58                 num += 32;
  59         else
  60                 word >>= 32;
  61 #endif
  62         if ((word & 0xffff0000) == 0)
  63                 num += 16;
  64         else
  65                 word >>= 16;
  66
  67         if ((word & 0xff00) == 0)
  68                 num += 8;
  69         else
  70                 word >>= 8;
  71
  72         if ((word & 0xf0) == 0)
  73                 num += 4;
  74         else
  75                 word >>= 4;
  76
  77         if ((word & 0xc) == 0)
  78                 num += 2;
  79         else
  80                 word >>= 2;
  81
  82         if ((word & 0x2) == 0)
  83                 num += 1;
  84         return num;
  85 }
  86
  87 /*
  88  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  89  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  90  * @size must be integral times of unsigned long.
  91  * Example:
  92  *                             MSB <--> LSB
  93  *   f2fs_set_bit(0, bitmap) => 1000 0000
  94  *   f2fs_set_bit(7, bitmap) => 0000 0001
  95  */
  96 static unsigned long __find_rev_next_bit(const unsigned long *addr,
  97                         unsigned long size, unsigned long offset)
  98 {
  99         const unsigned long *p = addr + BIT_WORD(offset);
 100         unsigned long result = size;
 101         unsigned long tmp;
 102
 103         if (offset >= size)
 104                 return size;
 105
 106         size -= (offset & ~(BITS_PER_LONG - 1));
 107         offset %= BITS_PER_LONG;
 108
 109         while (1) {
 110                 if (*p == 0)
 111                         goto pass;
 112
 113                 tmp = __reverse_ulong((unsigned char *)p);
 114
 115                 tmp &= ~0UL >> offset;
 116                 if (size < BITS_PER_LONG)
 117                         tmp &= (~0UL << (BITS_PER_LONG - size));
 118                 if (tmp)
 119                         goto found;
 120 pass:
 121                 if (size <= BITS_PER_LONG)
 122                         break;
 123                 size -= BITS_PER_LONG;
 124                 offset = 0;
 125                 p++;
 126         }
 127         return result;
 128 found:
 129         return result - size + __reverse_ffs(tmp);
 130 }
 131
 132 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 133                         unsigned long size, unsigned long offset)
 134 {
 135         const unsigned long *p = addr + BIT_WORD(offset);
 136         unsigned long result = size;
 137         unsigned long tmp;
 138
 139         if (offset >= size)
 140                 return size;
 141
 142         size -= (offset & ~(BITS_PER_LONG - 1));
 143         offset %= BITS_PER_LONG;
 144
 145         while (1) {
 146                 if (*p == ~0UL)
 147                         goto pass;
 148
 149                 tmp = __reverse_ulong((unsigned char *)p);
 150
 151                 if (offset)
 152                         tmp |= ~0UL << (BITS_PER_LONG - offset);
 153                 if (size < BITS_PER_LONG)
 154                         tmp |= ~0UL >> size;
 155                 if (tmp != ~0UL)
 156                         goto found;
 157 pass:
 158                 if (size <= BITS_PER_LONG)
 159                         break;
 160                 size -= BITS_PER_LONG;
 161                 offset = 0;
 162                 p++;
 163         }
 164         return result;
 165 found:
 166         return result - size + __reverse_ffz(tmp);
 167 }
 168
 169 void register_inmem_page(struct inode *inode, struct page *page)
 170 {
 171         struct f2fs_inode_info *fi = F2FS_I(inode);
 172         struct inmem_pages *new;
 173
 174         f2fs_trace_pid(page);
 175
 176         set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
 177         SetPagePrivate(page);
 178
 179         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 180
 181         /* add atomic page indices to the list */
 182         new->page = page;
 183         INIT_LIST_HEAD(&new->list);
 184
 185         /* increase reference count with clean state */
 186         mutex_lock(&fi->inmem_lock);
 187         get_page(page);
 188         list_add_tail(&new->list, &fi->inmem_pages);
 189         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 190         mutex_unlock(&fi->inmem_lock);
 191
 192         trace_f2fs_register_inmem_page(page, INMEM);
 193 }
 194
 195 static int __revoke_inmem_pages(struct inode *inode,
 196                                 struct list_head *head, bool drop, bool recover)
 197 {
 198         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 199         struct inmem_pages *cur, *tmp;
 200         int err = 0;
 201
 202         list_for_each_entry_safe(cur, tmp, head, list) {
 203                 struct page *page = cur->page;
 204
 205                 if (drop)
 206                         trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 207
 208                 lock_page(page);
 209
 210                 if (recover) {
 211                         struct dnode_of_data dn;
 212                         struct node_info ni;
 213
 214                         trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 215
 216                         set_new_dnode(&dn, inode, NULL, NULL, 0);
 217                         if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) {
 218                                 err = -EAGAIN;
 219                                 goto next;
 220                         }
 221                         get_node_info(sbi, dn.nid, &ni);
 222                         f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 223                                         cur->old_addr, ni.version, true, true);
 224                         f2fs_put_dnode(&dn);
 225                 }
 226 next:
 227                 /* we don't need to invalidate this in the sccessful status */
 228                 if (drop || recover)
 229                         ClearPageUptodate(page);
 230                 set_page_private(page, 0);
 231                 ClearPagePrivate(page);
 232                 f2fs_put_page(page, 1);
 233
 234                 list_del(&cur->list);
 235                 kmem_cache_free(inmem_entry_slab, cur);
 236                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 237         }
 238         return err;
 239 }
 240
 241 void drop_inmem_pages(struct inode *inode)
 242 {
 243         struct f2fs_inode_info *fi = F2FS_I(inode);
 244
 245         clear_inode_flag(inode, FI_ATOMIC_FILE);
 246
 247         mutex_lock(&fi->inmem_lock);
 248         __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 249         mutex_unlock(&fi->inmem_lock);
 250 }
 251
 252 static int __commit_inmem_pages(struct inode *inode,
 253                                         struct list_head *revoke_list)
 254 {
 255         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 256         struct f2fs_inode_info *fi = F2FS_I(inode);
 257         struct inmem_pages *cur, *tmp;
 258         struct f2fs_io_info fio = {
 259                 .sbi = sbi,
 260                 .type = DATA,
 261                 .op = REQ_OP_WRITE,
 262                 .op_flags = REQ_SYNC | REQ_PRIO,
 263                 .encrypted_page = NULL,
 264         };
 265         bool submit_bio = false;
 266         int err = 0;
 267
 268         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 269                 struct page *page = cur->page;
 270
 271                 lock_page(page);
 272                 if (page->mapping == inode->i_mapping) {
 273                         trace_f2fs_commit_inmem_page(page, INMEM);
 274
 275                         set_page_dirty(page);
 276                         f2fs_wait_on_page_writeback(page, DATA, true);
 277                         if (clear_page_dirty_for_io(page)) {
 278                                 inode_dec_dirty_pages(inode);
 279                                 remove_dirty_inode(inode);
 280                         }
 281
 282                         fio.page = page;
 283                         err = do_write_data_page(&fio);
 284                         if (err) {
 285                                 unlock_page(page);
 286                                 break;
 287                         }
 288
 289                         /* record old blkaddr for revoking */
 290                         cur->old_addr = fio.old_blkaddr;
 291
 292                         submit_bio = true;
 293                 }
 294                 unlock_page(page);
 295                 list_move_tail(&cur->list, revoke_list);
 296         }
 297
 298         if (submit_bio)
 299                 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
 300
 301         if (!err)
 302                 __revoke_inmem_pages(inode, revoke_list, false, false);
 303
 304         return err;
 305 }
 306
 307 int commit_inmem_pages(struct inode *inode)
 308 {
 309         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 310         struct f2fs_inode_info *fi = F2FS_I(inode);
 311         struct list_head revoke_list;
 312         int err;
 313
 314         INIT_LIST_HEAD(&revoke_list);
 315         f2fs_balance_fs(sbi, true);
 316         f2fs_lock_op(sbi);
 317
 318         mutex_lock(&fi->inmem_lock);
 319         err = __commit_inmem_pages(inode, &revoke_list);
 320         if (err) {
 321                 int ret;
 322                 /*
 323                  * try to revoke all committed pages, but still we could fail
 324                  * due to no memory or other reason, if that happened, EAGAIN
 325                  * will be returned, which means in such case, transaction is
 326                  * already not integrity, caller should use journal to do the
 327                  * recovery or rewrite & commit last transaction. For other
 328                  * error number, revoking was done by filesystem itself.
 329                  */
 330                 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
 331                 if (ret)
 332                         err = ret;
 333
 334                 /* drop all uncommitted pages */
 335                 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 336         }
 337         mutex_unlock(&fi->inmem_lock);
 338
 339         f2fs_unlock_op(sbi);
 340         return err;
 341 }
 342
 343 /*
 344  * This function balances dirty node and dentry pages.
 345  * In addition, it controls garbage collection.
 346  */
 347 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 348 {
 349 #ifdef CONFIG_F2FS_FAULT_INJECTION
 350         if (time_to_inject(sbi, FAULT_CHECKPOINT))
 351                 f2fs_stop_checkpoint(sbi, false);
 352 #endif
 353
 354         if (!need)
 355                 return;
 356
 357         /* balance_fs_bg is able to be pending */
 358         if (excess_cached_nats(sbi))
 359                 f2fs_balance_fs_bg(sbi);
 360
 361         /*
 362          * We should do GC or end up with checkpoint, if there are so many dirty
 363          * dir/node pages without enough free segments.
 364          */
 365         if (has_not_enough_free_secs(sbi, 0, 0)) {
 366                 mutex_lock(&sbi->gc_mutex);
 367                 f2fs_gc(sbi, false, false);
 368         }
 369 }
 370
 371 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 372 {
 373         /* try to shrink extent cache when there is no enough memory */
 374         if (!available_free_memory(sbi, EXTENT_CACHE))
 375                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 376
 377         /* check the # of cached NAT entries */
 378         if (!available_free_memory(sbi, NAT_ENTRIES))
 379                 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 380
 381         if (!available_free_memory(sbi, FREE_NIDS))
 382                 try_to_free_nids(sbi, MAX_FREE_NIDS);
 383         else
 384                 build_free_nids(sbi, false);
 385
 386         if (!is_idle(sbi))
 387                 return;
 388
 389         /* checkpoint is the only way to shrink partial cached entries */
 390         if (!available_free_memory(sbi, NAT_ENTRIES) ||
 391                         !available_free_memory(sbi, INO_ENTRIES) ||
 392                         excess_prefree_segs(sbi) ||
 393                         excess_dirty_nats(sbi) ||
 394                         f2fs_time_over(sbi, CP_TIME)) {
 395                 if (test_opt(sbi, DATA_FLUSH)) {
 396                         struct blk_plug plug;
 397
 398                         blk_start_plug(&plug);
 399                         sync_dirty_inodes(sbi, FILE_INODE);
 400                         blk_finish_plug(&plug);
 401                 }
 402                 f2fs_sync_fs(sbi->sb, true);
 403                 stat_inc_bg_cp_count(sbi->stat_info);
 404         }
 405 }
 406
 407 static int __submit_flush_wait(struct block_device *bdev)
 408 {
 409         struct bio *bio = f2fs_bio_alloc(0);
 410         int ret;
 411
 412         bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
 413         bio->bi_bdev = bdev;
 414         ret = submit_bio_wait(bio);
 415         bio_put(bio);
 416         return ret;
 417 }
 418
 419 static int submit_flush_wait(struct f2fs_sb_info *sbi)
 420 {
 421         int ret = __submit_flush_wait(sbi->sb->s_bdev);
 422         int i;
 423
 424         if (sbi->s_ndevs && !ret) {
 425                 for (i = 1; i < sbi->s_ndevs; i++) {
 426                         ret = __submit_flush_wait(FDEV(i).bdev);
 427                         if (ret)
 428                                 break;
 429                 }
 430         }
 431         return ret;
 432 }
 433
 434 static int issue_flush_thread(void *data)
 435 {
 436         struct f2fs_sb_info *sbi = data;
 437         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 438         wait_queue_head_t *q = &fcc->flush_wait_queue;
 439 repeat:
 440         if (kthread_should_stop())
 441                 return 0;
 442
 443         if (!llist_empty(&fcc->issue_list)) {
 444                 struct flush_cmd *cmd, *next;
 445                 int ret;
 446
 447                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 448                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 449
 450                 ret = submit_flush_wait(sbi);
 451                 llist_for_each_entry_safe(cmd, next,
 452                                           fcc->dispatch_list, llnode) {
 453                         cmd->ret = ret;
 454                         complete(&cmd->wait);
 455                 }
 456                 fcc->dispatch_list = NULL;
 457         }
 458
 459         wait_event_interruptible(*q,
 460                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 461         goto repeat;
 462 }
 463
 464 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
 465 {
 466         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 467         struct flush_cmd cmd;
 468
 469         trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
 470                                         test_opt(sbi, FLUSH_MERGE));
 471
 472         if (test_opt(sbi, NOBARRIER))
 473                 return 0;
 474
 475         if (!test_opt(sbi, FLUSH_MERGE) || !atomic_read(&fcc->submit_flush)) {
 476                 int ret;
 477
 478                 atomic_inc(&fcc->submit_flush);
 479                 ret = submit_flush_wait(sbi);
 480                 atomic_dec(&fcc->submit_flush);
 481                 return ret;
 482         }
 483
 484         init_completion(&cmd.wait);
 485
 486         atomic_inc(&fcc->submit_flush);
 487         llist_add(&cmd.llnode, &fcc->issue_list);
 488
 489         if (!fcc->dispatch_list)
 490                 wake_up(&fcc->flush_wait_queue);
 491
 492         if (fcc->f2fs_issue_flush) {
 493                 wait_for_completion(&cmd.wait);
 494                 atomic_dec(&fcc->submit_flush);
 495         } else {
 496                 llist_del_all(&fcc->issue_list);
 497                 atomic_set(&fcc->submit_flush, 0);
 498         }
 499
 500         return cmd.ret;
 501 }
 502
 503 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
 504 {
 505         dev_t dev = sbi->sb->s_bdev->bd_dev;
 506         struct flush_cmd_control *fcc;
 507         int err = 0;
 508
 509         if (SM_I(sbi)->cmd_control_info) {
 510                 fcc = SM_I(sbi)->cmd_control_info;
 511                 goto init_thread;
 512         }
 513
 514         fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
 515         if (!fcc)
 516                 return -ENOMEM;
 517         atomic_set(&fcc->submit_flush, 0);
 518         init_waitqueue_head(&fcc->flush_wait_queue);
 519         init_llist_head(&fcc->issue_list);
 520         SM_I(sbi)->cmd_control_info = fcc;
 521 init_thread:
 522         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 523                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 524         if (IS_ERR(fcc->f2fs_issue_flush)) {
 525                 err = PTR_ERR(fcc->f2fs_issue_flush);
 526                 kfree(fcc);
 527                 SM_I(sbi)->cmd_control_info = NULL;
 528                 return err;
 529         }
 530
 531         return err;
 532 }
 533
 534 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 535 {
 536         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 537
 538         if (fcc && fcc->f2fs_issue_flush) {
 539                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 540
 541                 fcc->f2fs_issue_flush = NULL;
 542                 kthread_stop(flush_thread);
 543         }
 544         if (free) {
 545                 kfree(fcc);
 546                 SM_I(sbi)->cmd_control_info = NULL;
 547         }
 548 }
 549
 550 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 551                 enum dirty_type dirty_type)
 552 {
 553         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 554
 555         /* need not be added */
 556         if (IS_CURSEG(sbi, segno))
 557                 return;
 558
 559         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 560                 dirty_i->nr_dirty[dirty_type]++;
 561
 562         if (dirty_type == DIRTY) {
 563                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 564                 enum dirty_type t = sentry->type;
 565
 566                 if (unlikely(t >= DIRTY)) {
 567                         f2fs_bug_on(sbi, 1);
 568                         return;
 569                 }
 570                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 571                         dirty_i->nr_dirty[t]++;
 572         }
 573 }
 574
 575 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 576                 enum dirty_type dirty_type)
 577 {
 578         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 579
 580         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 581                 dirty_i->nr_dirty[dirty_type]--;
 582
 583         if (dirty_type == DIRTY) {
 584                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 585                 enum dirty_type t = sentry->type;
 586
 587                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 588                         dirty_i->nr_dirty[t]--;
 589
 590                 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
 591                         clear_bit(GET_SECNO(sbi, segno),
 592                                                 dirty_i->victim_secmap);
 593         }
 594 }
 595
 596 /*
 597  * Should not occur error such as -ENOMEM.
 598  * Adding dirty entry into seglist is not critical operation.
 599  * If a given segment is one of current working segments, it won't be added.
 600  */
 601 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 602 {
 603         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 604         unsigned short valid_blocks;
 605
 606         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 607                 return;
 608
 609         mutex_lock(&dirty_i->seglist_lock);
 610
 611         valid_blocks = get_valid_blocks(sbi, segno, 0);
 612
 613         if (valid_blocks == 0) {
 614                 __locate_dirty_segment(sbi, segno, PRE);
 615                 __remove_dirty_segment(sbi, segno, DIRTY);
 616         } else if (valid_blocks < sbi->blocks_per_seg) {
 617                 __locate_dirty_segment(sbi, segno, DIRTY);
 618         } else {
 619                 /* Recovery routine with SSR needs this */
 620                 __remove_dirty_segment(sbi, segno, DIRTY);
 621         }
 622
 623         mutex_unlock(&dirty_i->seglist_lock);
 624 }
 625
 626 static struct bio_entry *__add_bio_entry(struct f2fs_sb_info *sbi,
 627                                                         struct bio *bio)
 628 {
 629         struct list_head *wait_list = &(SM_I(sbi)->wait_list);
 630         struct bio_entry *be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS);
 631
 632         INIT_LIST_HEAD(&be->list);
 633         be->bio = bio;
 634         init_completion(&be->event);
 635         list_add_tail(&be->list, wait_list);
 636
 637         return be;
 638 }
 639
 640 void f2fs_wait_all_discard_bio(struct f2fs_sb_info *sbi)
 641 {
 642         struct list_head *wait_list = &(SM_I(sbi)->wait_list);
 643         struct bio_entry *be, *tmp;
 644
 645         list_for_each_entry_safe(be, tmp, wait_list, list) {
 646                 struct bio *bio = be->bio;
 647                 int err;
 648
 649                 wait_for_completion_io(&be->event);
 650                 err = be->error;
 651                 if (err == -EOPNOTSUPP)
 652                         err = 0;
 653
 654                 if (err)
 655                         f2fs_msg(sbi->sb, KERN_INFO,
 656                                 "Issue discard failed, ret: %d", err);
 657
 658                 bio_put(bio);
 659                 list_del(&be->list);
 660                 kmem_cache_free(bio_entry_slab, be);
 661         }
 662 }
 663
 664 static void f2fs_submit_bio_wait_endio(struct bio *bio)
 665 {
 666         struct bio_entry *be = (struct bio_entry *)bio->bi_private;
 667
 668         be->error = bio->bi_error;
 669         complete(&be->event);
 670 }
 671
 672 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
 673 static int __f2fs_issue_discard_async(struct f2fs_sb_info *sbi,
 674                 struct block_device *bdev, block_t blkstart, block_t blklen)
 675 {
 676         struct bio *bio = NULL;
 677         int err;
 678
 679         trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
 680
 681         if (sbi->s_ndevs) {
 682                 int devi = f2fs_target_device_index(sbi, blkstart);
 683
 684                 blkstart -= FDEV(devi).start_blk;
 685         }
 686         err = __blkdev_issue_discard(bdev,
 687                                 SECTOR_FROM_BLOCK(blkstart),
 688                                 SECTOR_FROM_BLOCK(blklen),
 689                                 GFP_NOFS, 0, &bio);
 690         if (!err && bio) {
 691                 struct bio_entry *be = __add_bio_entry(sbi, bio);
 692
 693                 bio->bi_private = be;
 694                 bio->bi_end_io = f2fs_submit_bio_wait_endio;
 695                 bio->bi_opf |= REQ_SYNC;
 696                 submit_bio(bio);
 697         }
 698
 699         return err;
 700 }
 701
 702 #ifdef CONFIG_BLK_DEV_ZONED
 703 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
 704                 struct block_device *bdev, block_t blkstart, block_t blklen)
 705 {
 706         sector_t nr_sects = SECTOR_FROM_BLOCK(blklen);
 707         sector_t sector;
 708         int devi = 0;
 709
 710         if (sbi->s_ndevs) {
 711                 devi = f2fs_target_device_index(sbi, blkstart);
 712                 blkstart -= FDEV(devi).start_blk;
 713         }
 714         sector = SECTOR_FROM_BLOCK(blkstart);
 715
 716         if (sector & (bdev_zone_sectors(bdev) - 1) ||
 717             nr_sects != bdev_zone_sectors(bdev)) {
 718                 f2fs_msg(sbi->sb, KERN_INFO,
 719                         "(%d) %s: Unaligned discard attempted (block %x + %x)",
 720                         devi, sbi->s_ndevs ? FDEV(devi).path: "",
 721                         blkstart, blklen);
 722                 return -EIO;
 723         }
 724
 725         /*
 726          * We need to know the type of the zone: for conventional zones,
 727          * use regular discard if the drive supports it. For sequential
 728          * zones, reset the zone write pointer.
 729          */
 730         switch (get_blkz_type(sbi, bdev, blkstart)) {
 731
 732         case BLK_ZONE_TYPE_CONVENTIONAL:
 733                 if (!blk_queue_discard(bdev_get_queue(bdev)))
 734                         return 0;
 735                 return __f2fs_issue_discard_async(sbi, bdev, blkstart, blklen);
 736         case BLK_ZONE_TYPE_SEQWRITE_REQ:
 737         case BLK_ZONE_TYPE_SEQWRITE_PREF:
 738                 trace_f2fs_issue_reset_zone(sbi->sb, blkstart);
 739                 return blkdev_reset_zones(bdev, sector,
 740                                           nr_sects, GFP_NOFS);
 741         default:
 742                 /* Unknown zone type: broken device ? */
 743                 return -EIO;
 744         }
 745 }
 746 #endif
 747
 748 static int __issue_discard_async(struct f2fs_sb_info *sbi,
 749                 struct block_device *bdev, block_t blkstart, block_t blklen)
 750 {
 751 #ifdef CONFIG_BLK_DEV_ZONED
 752         if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
 753                                 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
 754                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
 755 #endif
 756         return __f2fs_issue_discard_async(sbi, bdev, blkstart, blklen);
 757 }
 758
 759 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
 760                                 block_t blkstart, block_t blklen)
 761 {
 762         sector_t start = blkstart, len = 0;
 763         struct block_device *bdev;
 764         struct seg_entry *se;
 765         unsigned int offset;
 766         block_t i;
 767         int err = 0;
 768
 769         bdev = f2fs_target_device(sbi, blkstart, NULL);
 770
 771         for (i = blkstart; i < blkstart + blklen; i++, len++) {
 772                 if (i != start) {
 773                         struct block_device *bdev2 =
 774                                 f2fs_target_device(sbi, i, NULL);
 775
 776                         if (bdev2 != bdev) {
 777                                 err = __issue_discard_async(sbi, bdev,
 778                                                 start, len);
 779                                 if (err)
 780                                         return err;
 781                                 bdev = bdev2;
 782                                 start = i;
 783                                 len = 0;
 784                         }
 785                 }
 786
 787                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
 788                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
 789
 790                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
 791                         sbi->discard_blks--;
 792         }
 793
 794         if (len)
 795                 err = __issue_discard_async(sbi, bdev, start, len);
 796         return err;
 797 }
 798
 799 static void __add_discard_entry(struct f2fs_sb_info *sbi,
 800                 struct cp_control *cpc, struct seg_entry *se,
 801                 unsigned int start, unsigned int end)
 802 {
 803         struct list_head *head = &SM_I(sbi)->discard_list;
 804         struct discard_entry *new, *last;
 805
 806         if (!list_empty(head)) {
 807                 last = list_last_entry(head, struct discard_entry, list);
 808                 if (START_BLOCK(sbi, cpc->trim_start) + start ==
 809                                                 last->blkaddr + last->len) {
 810                         last->len += end - start;
 811                         goto done;
 812                 }
 813         }
 814
 815         new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
 816         INIT_LIST_HEAD(&new->list);
 817         new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
 818         new->len = end - start;
 819         list_add_tail(&new->list, head);
 820 done:
 821         SM_I(sbi)->nr_discards += end - start;
 822 }
 823
 824 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 825 {
 826         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
 827         int max_blocks = sbi->blocks_per_seg;
 828         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
 829         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
 830         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
 831         unsigned long *discard_map = (unsigned long *)se->discard_map;
 832         unsigned long *dmap = SIT_I(sbi)->tmp_map;
 833         unsigned int start = 0, end = -1;
 834         bool force = (cpc->reason == CP_DISCARD);
 835         int i;
 836
 837         if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
 838                 return;
 839
 840         if (!force) {
 841                 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
 842                     SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
 843                         return;
 844         }
 845
 846         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
 847         for (i = 0; i < entries; i++)
 848                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
 849                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
 850
 851         while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
 852                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
 853                 if (start >= max_blocks)
 854                         break;
 855
 856                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
 857                 if (force && start && end != max_blocks
 858                                         && (end - start) < cpc->trim_minlen)
 859                         continue;
 860
 861                 __add_discard_entry(sbi, cpc, se, start, end);
 862         }
 863 }
 864
 865 void release_discard_addrs(struct f2fs_sb_info *sbi)
 866 {
 867         struct list_head *head = &(SM_I(sbi)->discard_list);
 868         struct discard_entry *entry, *this;
 869
 870         /* drop caches */
 871         list_for_each_entry_safe(entry, this, head, list) {
 872                 list_del(&entry->list);
 873                 kmem_cache_free(discard_entry_slab, entry);
 874         }
 875 }
 876
 877 /*
 878  * Should call clear_prefree_segments after checkpoint is done.
 879  */
 880 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
 881 {
 882         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 883         unsigned int segno;
 884
 885         mutex_lock(&dirty_i->seglist_lock);
 886         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
 887                 __set_test_and_free(sbi, segno);
 888         mutex_unlock(&dirty_i->seglist_lock);
 889 }
 890
 891 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 892 {
 893         struct list_head *head = &(SM_I(sbi)->discard_list);
 894         struct discard_entry *entry, *this;
 895         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 896         struct blk_plug plug;
 897         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
 898         unsigned int start = 0, end = -1;
 899         unsigned int secno, start_segno;
 900         bool force = (cpc->reason == CP_DISCARD);
 901
 902         blk_start_plug(&plug);
 903
 904         mutex_lock(&dirty_i->seglist_lock);
 905
 906         while (1) {
 907                 int i;
 908                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
 909                 if (start >= MAIN_SEGS(sbi))
 910                         break;
 911                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
 912                                                                 start + 1);
 913
 914                 for (i = start; i < end; i++)
 915                         clear_bit(i, prefree_map);
 916
 917                 dirty_i->nr_dirty[PRE] -= end - start;
 918
 919                 if (force || !test_opt(sbi, DISCARD))
 920                         continue;
 921
 922                 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
 923                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
 924                                 (end - start) << sbi->log_blocks_per_seg);
 925                         continue;
 926                 }
 927 next:
 928                 secno = GET_SECNO(sbi, start);
 929                 start_segno = secno * sbi->segs_per_sec;
 930                 if (!IS_CURSEC(sbi, secno) &&
 931                         !get_valid_blocks(sbi, start, sbi->segs_per_sec))
 932                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
 933                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
 934
 935                 start = start_segno + sbi->segs_per_sec;
 936                 if (start < end)
 937                         goto next;
 938         }
 939         mutex_unlock(&dirty_i->seglist_lock);
 940
 941         /* send small discards */
 942         list_for_each_entry_safe(entry, this, head, list) {
 943                 if (force && entry->len < cpc->trim_minlen)
 944                         goto skip;
 945                 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
 946                 cpc->trimmed += entry->len;
 947 skip:
 948                 list_del(&entry->list);
 949                 SM_I(sbi)->nr_discards -= entry->len;
 950                 kmem_cache_free(discard_entry_slab, entry);
 951         }
 952
 953         blk_finish_plug(&plug);
 954 }
 955
 956 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
 957 {
 958         struct sit_info *sit_i = SIT_I(sbi);
 959
 960         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
 961                 sit_i->dirty_sentries++;
 962                 return false;
 963         }
 964
 965         return true;
 966 }
 967
 968 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
 969                                         unsigned int segno, int modified)
 970 {
 971         struct seg_entry *se = get_seg_entry(sbi, segno);
 972         se->type = type;
 973         if (modified)
 974                 __mark_sit_entry_dirty(sbi, segno);
 975 }
 976
 977 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
 978 {
 979         struct seg_entry *se;
 980         unsigned int segno, offset;
 981         long int new_vblocks;
 982
 983         segno = GET_SEGNO(sbi, blkaddr);
 984
 985         se = get_seg_entry(sbi, segno);
 986         new_vblocks = se->valid_blocks + del;
 987         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 988
 989         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
 990                                 (new_vblocks > sbi->blocks_per_seg)));
 991
 992         se->valid_blocks = new_vblocks;
 993         se->mtime = get_mtime(sbi);
 994         SIT_I(sbi)->max_mtime = se->mtime;
 995
 996         /* Update valid block bitmap */
 997         if (del > 0) {
 998                 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
 999                         f2fs_bug_on(sbi, 1);
1000                 if (f2fs_discard_en(sbi) &&
1001                         !f2fs_test_and_set_bit(offset, se->discard_map))
1002                         sbi->discard_blks--;
1003         } else {
1004                 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
1005                         f2fs_bug_on(sbi, 1);
1006                 if (f2fs_discard_en(sbi) &&
1007                         f2fs_test_and_clear_bit(offset, se->discard_map))
1008                         sbi->discard_blks++;
1009         }
1010         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1011                 se->ckpt_valid_blocks += del;
1012
1013         __mark_sit_entry_dirty(sbi, segno);
1014
1015         /* update total number of valid blocks to be written in ckpt area */
1016         SIT_I(sbi)->written_valid_blocks += del;
1017
1018         if (sbi->segs_per_sec > 1)
1019                 get_sec_entry(sbi, segno)->valid_blocks += del;
1020 }
1021
1022 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
1023 {
1024         update_sit_entry(sbi, new, 1);
1025         if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1026                 update_sit_entry(sbi, old, -1);
1027
1028         locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1029         locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
1030 }
1031
1032 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1033 {
1034         unsigned int segno = GET_SEGNO(sbi, addr);
1035         struct sit_info *sit_i = SIT_I(sbi);
1036
1037         f2fs_bug_on(sbi, addr == NULL_ADDR);
1038         if (addr == NEW_ADDR)
1039                 return;
1040
1041         /* add it into sit main buffer */
1042         mutex_lock(&sit_i->sentry_lock);
1043
1044         update_sit_entry(sbi, addr, -1);
1045
1046         /* add it into dirty seglist */
1047         locate_dirty_segment(sbi, segno);
1048
1049         mutex_unlock(&sit_i->sentry_lock);
1050 }
1051
1052 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1053 {
1054         struct sit_info *sit_i = SIT_I(sbi);
1055         unsigned int segno, offset;
1056         struct seg_entry *se;
1057         bool is_cp = false;
1058
1059         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1060                 return true;
1061
1062         mutex_lock(&sit_i->sentry_lock);
1063
1064         segno = GET_SEGNO(sbi, blkaddr);
1065         se = get_seg_entry(sbi, segno);
1066         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1067
1068         if (f2fs_test_bit(offset, se->ckpt_valid_map))
1069                 is_cp = true;
1070
1071         mutex_unlock(&sit_i->sentry_lock);
1072
1073         return is_cp;
1074 }
1075
1076 /*
1077  * This function should be resided under the curseg_mutex lock
1078  */
1079 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1080                                         struct f2fs_summary *sum)
1081 {
1082         struct curseg_info *curseg = CURSEG_I(sbi, type);
1083         void *addr = curseg->sum_blk;
1084         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1085         memcpy(addr, sum, sizeof(struct f2fs_summary));
1086 }
1087
1088 /*
1089  * Calculate the number of current summary pages for writing
1090  */
1091 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1092 {
1093         int valid_sum_count = 0;
1094         int i, sum_in_page;
1095
1096         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1097                 if (sbi->ckpt->alloc_type[i] == SSR)
1098                         valid_sum_count += sbi->blocks_per_seg;
1099                 else {
1100                         if (for_ra)
1101                                 valid_sum_count += le16_to_cpu(
1102                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1103                         else
1104                                 valid_sum_count += curseg_blkoff(sbi, i);
1105                 }
1106         }
1107
1108         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1109                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1110         if (valid_sum_count <= sum_in_page)
1111                 return 1;
1112         else if ((valid_sum_count - sum_in_page) <=
1113                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1114                 return 2;
1115         return 3;
1116 }
1117
1118 /*
1119  * Caller should put this summary page
1120  */
1121 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1122 {
1123         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1124 }
1125
1126 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1127 {
1128         struct page *page = grab_meta_page(sbi, blk_addr);
1129         void *dst = page_address(page);
1130
1131         if (src)
1132                 memcpy(dst, src, PAGE_SIZE);
1133         else
1134                 memset(dst, 0, PAGE_SIZE);
1135         set_page_dirty(page);
1136         f2fs_put_page(page, 1);
1137 }
1138
1139 static void write_sum_page(struct f2fs_sb_info *sbi,
1140                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
1141 {
1142         update_meta_page(sbi, (void *)sum_blk, blk_addr);
1143 }
1144
1145 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1146                                                 int type, block_t blk_addr)
1147 {
1148         struct curseg_info *curseg = CURSEG_I(sbi, type);
1149         struct page *page = grab_meta_page(sbi, blk_addr);
1150         struct f2fs_summary_block *src = curseg->sum_blk;
1151         struct f2fs_summary_block *dst;
1152
1153         dst = (struct f2fs_summary_block *)page_address(page);
1154
1155         mutex_lock(&curseg->curseg_mutex);
1156
1157         down_read(&curseg->journal_rwsem);
1158         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1159         up_read(&curseg->journal_rwsem);
1160
1161         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1162         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1163
1164         mutex_unlock(&curseg->curseg_mutex);
1165
1166         set_page_dirty(page);
1167         f2fs_put_page(page, 1);
1168 }
1169
1170 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1171 {
1172         struct curseg_info *curseg = CURSEG_I(sbi, type);
1173         unsigned int segno = curseg->segno + 1;
1174         struct free_segmap_info *free_i = FREE_I(sbi);
1175
1176         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
1177                 return !test_bit(segno, free_i->free_segmap);
1178         return 0;
1179 }
1180
1181 /*
1182  * Find a new segment from the free segments bitmap to right order
1183  * This function should be returned with success, otherwise BUG
1184  */
1185 static void get_new_segment(struct f2fs_sb_info *sbi,
1186                         unsigned int *newseg, bool new_sec, int dir)
1187 {
1188         struct free_segmap_info *free_i = FREE_I(sbi);
1189         unsigned int segno, secno, zoneno;
1190         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
1191         unsigned int hint = *newseg / sbi->segs_per_sec;
1192         unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1193         unsigned int left_start = hint;
1194         bool init = true;
1195         int go_left = 0;
1196         int i;
1197
1198         spin_lock(&free_i->segmap_lock);
1199
1200         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1201                 segno = find_next_zero_bit(free_i->free_segmap,
1202                                 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1203                 if (segno < (hint + 1) * sbi->segs_per_sec)
1204                         goto got_it;
1205         }
1206 find_other_zone:
1207         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1208         if (secno >= MAIN_SECS(sbi)) {
1209                 if (dir == ALLOC_RIGHT) {
1210                         secno = find_next_zero_bit(free_i->free_secmap,
1211                                                         MAIN_SECS(sbi), 0);
1212                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1213                 } else {
1214                         go_left = 1;
1215                         left_start = hint - 1;
1216                 }
1217         }
1218         if (go_left == 0)
1219                 goto skip_left;
1220
1221         while (test_bit(left_start, free_i->free_secmap)) {
1222                 if (left_start > 0) {
1223                         left_start--;
1224                         continue;
1225                 }
1226                 left_start = find_next_zero_bit(free_i->free_secmap,
1227                                                         MAIN_SECS(sbi), 0);
1228                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1229                 break;
1230         }
1231         secno = left_start;
1232 skip_left:
1233         hint = secno;
1234         segno = secno * sbi->segs_per_sec;
1235         zoneno = secno / sbi->secs_per_zone;
1236
1237         /* give up on finding another zone */
1238         if (!init)
1239                 goto got_it;
1240         if (sbi->secs_per_zone == 1)
1241                 goto got_it;
1242         if (zoneno == old_zoneno)
1243                 goto got_it;
1244         if (dir == ALLOC_LEFT) {
1245                 if (!go_left && zoneno + 1 >= total_zones)
1246                         goto got_it;
1247                 if (go_left && zoneno == 0)
1248                         goto got_it;
1249         }
1250         for (i = 0; i < NR_CURSEG_TYPE; i++)
1251                 if (CURSEG_I(sbi, i)->zone == zoneno)
1252                         break;
1253
1254         if (i < NR_CURSEG_TYPE) {
1255                 /* zone is in user, try another */
1256                 if (go_left)
1257                         hint = zoneno * sbi->secs_per_zone - 1;
1258                 else if (zoneno + 1 >= total_zones)
1259                         hint = 0;
1260                 else
1261                         hint = (zoneno + 1) * sbi->secs_per_zone;
1262                 init = false;
1263                 goto find_other_zone;
1264         }
1265 got_it:
1266         /* set it as dirty segment in free segmap */
1267         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1268         __set_inuse(sbi, segno);
1269         *newseg = segno;
1270         spin_unlock(&free_i->segmap_lock);
1271 }
1272
1273 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1274 {
1275         struct curseg_info *curseg = CURSEG_I(sbi, type);
1276         struct summary_footer *sum_footer;
1277
1278         curseg->segno = curseg->next_segno;
1279         curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1280         curseg->next_blkoff = 0;
1281         curseg->next_segno = NULL_SEGNO;
1282
1283         sum_footer = &(curseg->sum_blk->footer);
1284         memset(sum_footer, 0, sizeof(struct summary_footer));
1285         if (IS_DATASEG(type))
1286                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1287         if (IS_NODESEG(type))
1288                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1289         __set_sit_entry_type(sbi, type, curseg->segno, modified);
1290 }
1291
1292 /*
1293  * Allocate a current working segment.
1294  * This function always allocates a free segment in LFS manner.
1295  */
1296 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1297 {
1298         struct curseg_info *curseg = CURSEG_I(sbi, type);
1299         unsigned int segno = curseg->segno;
1300         int dir = ALLOC_LEFT;
1301
1302         write_sum_page(sbi, curseg->sum_blk,
1303                                 GET_SUM_BLOCK(sbi, segno));
1304         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1305                 dir = ALLOC_RIGHT;
1306
1307         if (test_opt(sbi, NOHEAP))
1308                 dir = ALLOC_RIGHT;
1309
1310         get_new_segment(sbi, &segno, new_sec, dir);
1311         curseg->next_segno = segno;
1312         reset_curseg(sbi, type, 1);
1313         curseg->alloc_type = LFS;
1314 }
1315
1316 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1317                         struct curseg_info *seg, block_t start)
1318 {
1319         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1320         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1321         unsigned long *target_map = SIT_I(sbi)->tmp_map;
1322         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1323         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1324         int i, pos;
1325
1326         for (i = 0; i < entries; i++)
1327                 target_map[i] = ckpt_map[i] | cur_map[i];
1328
1329         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1330
1331         seg->next_blkoff = pos;
1332 }
1333
1334 /*
1335  * If a segment is written by LFS manner, next block offset is just obtained
1336  * by increasing the current block offset. However, if a segment is written by
1337  * SSR manner, next block offset obtained by calling __next_free_blkoff
1338  */
1339 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1340                                 struct curseg_info *seg)
1341 {
1342         if (seg->alloc_type == SSR)
1343                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1344         else
1345                 seg->next_blkoff++;
1346 }
1347
1348 /*
1349  * This function always allocates a used segment(from dirty seglist) by SSR
1350  * manner, so it should recover the existing segment information of valid blocks
1351  */
1352 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1353 {
1354         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1355         struct curseg_info *curseg = CURSEG_I(sbi, type);
1356         unsigned int new_segno = curseg->next_segno;
1357         struct f2fs_summary_block *sum_node;
1358         struct page *sum_page;
1359
1360         write_sum_page(sbi, curseg->sum_blk,
1361                                 GET_SUM_BLOCK(sbi, curseg->segno));
1362         __set_test_and_inuse(sbi, new_segno);
1363
1364         mutex_lock(&dirty_i->seglist_lock);
1365         __remove_dirty_segment(sbi, new_segno, PRE);
1366         __remove_dirty_segment(sbi, new_segno, DIRTY);
1367         mutex_unlock(&dirty_i->seglist_lock);
1368
1369         reset_curseg(sbi, type, 1);
1370         curseg->alloc_type = SSR;
1371         __next_free_blkoff(sbi, curseg, 0);
1372
1373         if (reuse) {
1374                 sum_page = get_sum_page(sbi, new_segno);
1375                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1376                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1377                 f2fs_put_page(sum_page, 1);
1378         }
1379 }
1380
1381 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1382 {
1383         struct curseg_info *curseg = CURSEG_I(sbi, type);
1384         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1385
1386         if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0, 0))
1387                 return v_ops->get_victim(sbi,
1388                                 &(curseg)->next_segno, BG_GC, type, SSR);
1389
1390         /* For data segments, let's do SSR more intensively */
1391         for (; type >= CURSEG_HOT_DATA; type--)
1392                 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1393                                                 BG_GC, type, SSR))
1394                         return 1;
1395         return 0;
1396 }
1397
1398 /*
1399  * flush out current segment and replace it with new segment
1400  * This function should be returned with success, otherwise BUG
1401  */
1402 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1403                                                 int type, bool force)
1404 {
1405         struct curseg_info *curseg = CURSEG_I(sbi, type);
1406
1407         if (force)
1408                 new_curseg(sbi, type, true);
1409         else if (type == CURSEG_WARM_NODE)
1410                 new_curseg(sbi, type, false);
1411         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1412                 new_curseg(sbi, type, false);
1413         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1414                 change_curseg(sbi, type, true);
1415         else
1416                 new_curseg(sbi, type, false);
1417
1418         stat_inc_seg_type(sbi, curseg);
1419 }
1420
1421 void allocate_new_segments(struct f2fs_sb_info *sbi)
1422 {
1423         struct curseg_info *curseg;
1424         unsigned int old_segno;
1425         int i;
1426
1427         if (test_opt(sbi, LFS))
1428                 return;
1429
1430         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1431                 curseg = CURSEG_I(sbi, i);
1432                 old_segno = curseg->segno;
1433                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
1434                 locate_dirty_segment(sbi, old_segno);
1435         }
1436 }
1437
1438 static const struct segment_allocation default_salloc_ops = {
1439         .allocate_segment = allocate_segment_by_default,
1440 };
1441
1442 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1443 {
1444         __u64 start = F2FS_BYTES_TO_BLK(range->start);
1445         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1446         unsigned int start_segno, end_segno;
1447         struct cp_control cpc;
1448         int err = 0;
1449
1450         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1451                 return -EINVAL;
1452
1453         cpc.trimmed = 0;
1454         if (end <= MAIN_BLKADDR(sbi))
1455                 goto out;
1456
1457         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1458                 f2fs_msg(sbi->sb, KERN_WARNING,
1459                         "Found FS corruption, run fsck to fix.");
1460                 goto out;
1461         }
1462
1463         /* start/end segment number in main_area */
1464         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1465         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1466                                                 GET_SEGNO(sbi, end);
1467         cpc.reason = CP_DISCARD;
1468         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1469
1470         /* do checkpoint to issue discard commands safely */
1471         for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1472                 cpc.trim_start = start_segno;
1473
1474                 if (sbi->discard_blks == 0)
1475                         break;
1476                 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1477                         cpc.trim_end = end_segno;
1478                 else
1479                         cpc.trim_end = min_t(unsigned int,
1480                                 rounddown(start_segno +
1481                                 BATCHED_TRIM_SEGMENTS(sbi),
1482                                 sbi->segs_per_sec) - 1, end_segno);
1483
1484                 mutex_lock(&sbi->gc_mutex);
1485                 err = write_checkpoint(sbi, &cpc);
1486                 mutex_unlock(&sbi->gc_mutex);
1487                 if (err)
1488                         break;
1489
1490                 schedule();
1491         }
1492 out:
1493         range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1494         return err;
1495 }
1496
1497 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1498 {
1499         struct curseg_info *curseg = CURSEG_I(sbi, type);
1500         if (curseg->next_blkoff < sbi->blocks_per_seg)
1501                 return true;
1502         return false;
1503 }
1504
1505 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1506 {
1507         if (p_type == DATA)
1508                 return CURSEG_HOT_DATA;
1509         else
1510                 return CURSEG_HOT_NODE;
1511 }
1512
1513 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1514 {
1515         if (p_type == DATA) {
1516                 struct inode *inode = page->mapping->host;
1517
1518                 if (S_ISDIR(inode->i_mode))
1519                         return CURSEG_HOT_DATA;
1520                 else
1521                         return CURSEG_COLD_DATA;
1522         } else {
1523                 if (IS_DNODE(page) && is_cold_node(page))
1524                         return CURSEG_WARM_NODE;
1525                 else
1526                         return CURSEG_COLD_NODE;
1527         }
1528 }
1529
1530 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1531 {
1532         if (p_type == DATA) {
1533                 struct inode *inode = page->mapping->host;
1534
1535                 if (S_ISDIR(inode->i_mode))
1536                         return CURSEG_HOT_DATA;
1537                 else if (is_cold_data(page) || file_is_cold(inode))
1538                         return CURSEG_COLD_DATA;
1539                 else
1540                         return CURSEG_WARM_DATA;
1541         } else {
1542                 if (IS_DNODE(page))
1543                         return is_cold_node(page) ? CURSEG_WARM_NODE :
1544                                                 CURSEG_HOT_NODE;
1545                 else
1546                         return CURSEG_COLD_NODE;
1547         }
1548 }
1549
1550 static int __get_segment_type(struct page *page, enum page_type p_type)
1551 {
1552         switch (F2FS_P_SB(page)->active_logs) {
1553         case 2:
1554                 return __get_segment_type_2(page, p_type);
1555         case 4:
1556                 return __get_segment_type_4(page, p_type);
1557         }
1558         /* NR_CURSEG_TYPE(6) logs by default */
1559         f2fs_bug_on(F2FS_P_SB(page),
1560                 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1561         return __get_segment_type_6(page, p_type);
1562 }
1563
1564 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1565                 block_t old_blkaddr, block_t *new_blkaddr,
1566                 struct f2fs_summary *sum, int type)
1567 {
1568         struct sit_info *sit_i = SIT_I(sbi);
1569         struct curseg_info *curseg = CURSEG_I(sbi, type);
1570
1571         mutex_lock(&curseg->curseg_mutex);
1572         mutex_lock(&sit_i->sentry_lock);
1573
1574         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1575
1576         /*
1577          * __add_sum_entry should be resided under the curseg_mutex
1578          * because, this function updates a summary entry in the
1579          * current summary block.
1580          */
1581         __add_sum_entry(sbi, type, sum);
1582
1583         __refresh_next_blkoff(sbi, curseg);
1584
1585         stat_inc_block_count(sbi, curseg);
1586
1587         if (!__has_curseg_space(sbi, type))
1588                 sit_i->s_ops->allocate_segment(sbi, type, false);
1589         /*
1590          * SIT information should be updated before segment allocation,
1591          * since SSR needs latest valid block information.
1592          */
1593         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1594
1595         mutex_unlock(&sit_i->sentry_lock);
1596
1597         if (page && IS_NODESEG(type))
1598                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1599
1600         mutex_unlock(&curseg->curseg_mutex);
1601 }
1602
1603 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1604 {
1605         int type = __get_segment_type(fio->page, fio->type);
1606
1607         if (fio->type == NODE || fio->type == DATA)
1608                 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1609
1610         allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1611                                         &fio->new_blkaddr, sum, type);
1612
1613         /* writeout dirty page into bdev */
1614         f2fs_submit_page_mbio(fio);
1615
1616         if (fio->type == NODE || fio->type == DATA)
1617                 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
1618 }
1619
1620 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1621 {
1622         struct f2fs_io_info fio = {
1623                 .sbi = sbi,
1624                 .type = META,
1625                 .op = REQ_OP_WRITE,
1626                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
1627                 .old_blkaddr = page->index,
1628                 .new_blkaddr = page->index,
1629                 .page = page,
1630                 .encrypted_page = NULL,
1631         };
1632
1633         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1634                 fio.op_flags &= ~REQ_META;
1635
1636         set_page_writeback(page);
1637         f2fs_submit_page_mbio(&fio);
1638 }
1639
1640 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1641 {
1642         struct f2fs_summary sum;
1643
1644         set_summary(&sum, nid, 0, 0);
1645         do_write_page(&sum, fio);
1646 }
1647
1648 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1649 {
1650         struct f2fs_sb_info *sbi = fio->sbi;
1651         struct f2fs_summary sum;
1652         struct node_info ni;
1653
1654         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1655         get_node_info(sbi, dn->nid, &ni);
1656         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1657         do_write_page(&sum, fio);
1658         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
1659 }
1660
1661 void rewrite_data_page(struct f2fs_io_info *fio)
1662 {
1663         fio->new_blkaddr = fio->old_blkaddr;
1664         stat_inc_inplace_blocks(fio->sbi);
1665         f2fs_submit_page_mbio(fio);
1666 }
1667
1668 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1669                                 block_t old_blkaddr, block_t new_blkaddr,
1670                                 bool recover_curseg, bool recover_newaddr)
1671 {
1672         struct sit_info *sit_i = SIT_I(sbi);
1673         struct curseg_info *curseg;
1674         unsigned int segno, old_cursegno;
1675         struct seg_entry *se;
1676         int type;
1677         unsigned short old_blkoff;
1678
1679         segno = GET_SEGNO(sbi, new_blkaddr);
1680         se = get_seg_entry(sbi, segno);
1681         type = se->type;
1682
1683         if (!recover_curseg) {
1684                 /* for recovery flow */
1685                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1686                         if (old_blkaddr == NULL_ADDR)
1687                                 type = CURSEG_COLD_DATA;
1688                         else
1689                                 type = CURSEG_WARM_DATA;
1690                 }
1691         } else {
1692                 if (!IS_CURSEG(sbi, segno))
1693                         type = CURSEG_WARM_DATA;
1694         }
1695
1696         curseg = CURSEG_I(sbi, type);
1697
1698         mutex_lock(&curseg->curseg_mutex);
1699         mutex_lock(&sit_i->sentry_lock);
1700
1701         old_cursegno = curseg->segno;
1702         old_blkoff = curseg->next_blkoff;
1703
1704         /* change the current segment */
1705         if (segno != curseg->segno) {
1706                 curseg->next_segno = segno;
1707                 change_curseg(sbi, type, true);
1708         }
1709
1710         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1711         __add_sum_entry(sbi, type, sum);
1712
1713         if (!recover_curseg || recover_newaddr)
1714                 update_sit_entry(sbi, new_blkaddr, 1);
1715         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1716                 update_sit_entry(sbi, old_blkaddr, -1);
1717
1718         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1719         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1720
1721         locate_dirty_segment(sbi, old_cursegno);
1722
1723         if (recover_curseg) {
1724                 if (old_cursegno != curseg->segno) {
1725                         curseg->next_segno = old_cursegno;
1726                         change_curseg(sbi, type, true);
1727                 }
1728                 curseg->next_blkoff = old_blkoff;
1729         }
1730
1731         mutex_unlock(&sit_i->sentry_lock);
1732         mutex_unlock(&curseg->curseg_mutex);
1733 }
1734
1735 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1736                                 block_t old_addr, block_t new_addr,
1737                                 unsigned char version, bool recover_curseg,
1738                                 bool recover_newaddr)
1739 {
1740         struct f2fs_summary sum;
1741
1742         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1743
1744         __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1745                                         recover_curseg, recover_newaddr);
1746
1747         f2fs_update_data_blkaddr(dn, new_addr);
1748 }
1749
1750 void f2fs_wait_on_page_writeback(struct page *page,
1751                                 enum page_type type, bool ordered)
1752 {
1753         if (PageWriteback(page)) {
1754                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1755
1756                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
1757                 if (ordered)
1758                         wait_on_page_writeback(page);
1759                 else
1760                         wait_for_stable_page(page);
1761         }
1762 }
1763
1764 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1765                                                         block_t blkaddr)
1766 {
1767         struct page *cpage;
1768
1769         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1770                 return;
1771
1772         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1773         if (cpage) {
1774                 f2fs_wait_on_page_writeback(cpage, DATA, true);
1775                 f2fs_put_page(cpage, 1);
1776         }
1777 }
1778
1779 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1780 {
1781         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1782         struct curseg_info *seg_i;
1783         unsigned char *kaddr;
1784         struct page *page;
1785         block_t start;
1786         int i, j, offset;
1787
1788         start = start_sum_block(sbi);
1789
1790         page = get_meta_page(sbi, start++);
1791         kaddr = (unsigned char *)page_address(page);
1792
1793         /* Step 1: restore nat cache */
1794         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1795         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
1796
1797         /* Step 2: restore sit cache */
1798         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1799         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
1800         offset = 2 * SUM_JOURNAL_SIZE;
1801
1802         /* Step 3: restore summary entries */
1803         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1804                 unsigned short blk_off;
1805                 unsigned int segno;
1806
1807                 seg_i = CURSEG_I(sbi, i);
1808                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1809                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1810                 seg_i->next_segno = segno;
1811                 reset_curseg(sbi, i, 0);
1812                 seg_i->alloc_type = ckpt->alloc_type[i];
1813                 seg_i->next_blkoff = blk_off;
1814
1815                 if (seg_i->alloc_type == SSR)
1816                         blk_off = sbi->blocks_per_seg;
1817
1818                 for (j = 0; j < blk_off; j++) {
1819                         struct f2fs_summary *s;
1820                         s = (struct f2fs_summary *)(kaddr + offset);
1821                         seg_i->sum_blk->entries[j] = *s;
1822                         offset += SUMMARY_SIZE;
1823                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
1824                                                 SUM_FOOTER_SIZE)
1825                                 continue;
1826
1827                         f2fs_put_page(page, 1);
1828                         page = NULL;
1829
1830                         page = get_meta_page(sbi, start++);
1831                         kaddr = (unsigned char *)page_address(page);
1832                         offset = 0;
1833                 }
1834         }
1835         f2fs_put_page(page, 1);
1836         return 0;
1837 }
1838
1839 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1840 {
1841         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1842         struct f2fs_summary_block *sum;
1843         struct curseg_info *curseg;
1844         struct page *new;
1845         unsigned short blk_off;
1846         unsigned int segno = 0;
1847         block_t blk_addr = 0;
1848
1849         /* get segment number and block addr */
1850         if (IS_DATASEG(type)) {
1851                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1852                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1853                                                         CURSEG_HOT_DATA]);
1854                 if (__exist_node_summaries(sbi))
1855                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1856                 else
1857                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1858         } else {
1859                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1860                                                         CURSEG_HOT_NODE]);
1861                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1862                                                         CURSEG_HOT_NODE]);
1863                 if (__exist_node_summaries(sbi))
1864                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1865                                                         type - CURSEG_HOT_NODE);
1866                 else
1867                         blk_addr = GET_SUM_BLOCK(sbi, segno);
1868         }
1869
1870         new = get_meta_page(sbi, blk_addr);
1871         sum = (struct f2fs_summary_block *)page_address(new);
1872
1873         if (IS_NODESEG(type)) {
1874                 if (__exist_node_summaries(sbi)) {
1875                         struct f2fs_summary *ns = &sum->entries[0];
1876                         int i;
1877                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1878                                 ns->version = 0;
1879                                 ns->ofs_in_node = 0;
1880                         }
1881                 } else {
1882                         int err;
1883
1884                         err = restore_node_summary(sbi, segno, sum);
1885                         if (err) {
1886                                 f2fs_put_page(new, 1);
1887                                 return err;
1888                         }
1889                 }
1890         }
1891
1892         /* set uncompleted segment to curseg */
1893         curseg = CURSEG_I(sbi, type);
1894         mutex_lock(&curseg->curseg_mutex);
1895
1896         /* update journal info */
1897         down_write(&curseg->journal_rwsem);
1898         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1899         up_write(&curseg->journal_rwsem);
1900
1901         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1902         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
1903         curseg->next_segno = segno;
1904         reset_curseg(sbi, type, 0);
1905         curseg->alloc_type = ckpt->alloc_type[type];
1906         curseg->next_blkoff = blk_off;
1907         mutex_unlock(&curseg->curseg_mutex);
1908         f2fs_put_page(new, 1);
1909         return 0;
1910 }
1911
1912 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1913 {
1914         int type = CURSEG_HOT_DATA;
1915         int err;
1916
1917         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
1918                 int npages = npages_for_summary_flush(sbi, true);
1919
1920                 if (npages >= 2)
1921                         ra_meta_pages(sbi, start_sum_block(sbi), npages,
1922                                                         META_CP, true);
1923
1924                 /* restore for compacted data summary */
1925                 if (read_compacted_summaries(sbi))
1926                         return -EINVAL;
1927                 type = CURSEG_HOT_NODE;
1928         }
1929
1930         if (__exist_node_summaries(sbi))
1931                 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1932                                         NR_CURSEG_TYPE - type, META_CP, true);
1933
1934         for (; type <= CURSEG_COLD_NODE; type++) {
1935                 err = read_normal_summaries(sbi, type);
1936                 if (err)
1937                         return err;
1938         }
1939
1940         return 0;
1941 }
1942
1943 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1944 {
1945         struct page *page;
1946         unsigned char *kaddr;
1947         struct f2fs_summary *summary;
1948         struct curseg_info *seg_i;
1949         int written_size = 0;
1950         int i, j;
1951
1952         page = grab_meta_page(sbi, blkaddr++);
1953         kaddr = (unsigned char *)page_address(page);
1954
1955         /* Step 1: write nat cache */
1956         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1957         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
1958         written_size += SUM_JOURNAL_SIZE;
1959
1960         /* Step 2: write sit cache */
1961         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1962         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
1963         written_size += SUM_JOURNAL_SIZE;
1964
1965         /* Step 3: write summary entries */
1966         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1967                 unsigned short blkoff;
1968                 seg_i = CURSEG_I(sbi, i);
1969                 if (sbi->ckpt->alloc_type[i] == SSR)
1970                         blkoff = sbi->blocks_per_seg;
1971                 else
1972                         blkoff = curseg_blkoff(sbi, i);
1973
1974                 for (j = 0; j < blkoff; j++) {
1975                         if (!page) {
1976                                 page = grab_meta_page(sbi, blkaddr++);
1977                                 kaddr = (unsigned char *)page_address(page);
1978                                 written_size = 0;
1979                         }
1980                         summary = (struct f2fs_summary *)(kaddr + written_size);
1981                         *summary = seg_i->sum_blk->entries[j];
1982                         written_size += SUMMARY_SIZE;
1983
1984                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
1985                                                         SUM_FOOTER_SIZE)
1986                                 continue;
1987
1988                         set_page_dirty(page);
1989                         f2fs_put_page(page, 1);
1990                         page = NULL;
1991                 }
1992         }
1993         if (page) {
1994                 set_page_dirty(page);
1995                 f2fs_put_page(page, 1);
1996         }
1997 }
1998
1999 static void write_normal_summaries(struct f2fs_sb_info *sbi,
2000                                         block_t blkaddr, int type)
2001 {
2002         int i, end;
2003         if (IS_DATASEG(type))
2004                 end = type + NR_CURSEG_DATA_TYPE;
2005         else
2006                 end = type + NR_CURSEG_NODE_TYPE;
2007
2008         for (i = type; i < end; i++)
2009                 write_current_sum_page(sbi, i, blkaddr + (i - type));
2010 }
2011
2012 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2013 {
2014         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
2015                 write_compacted_summaries(sbi, start_blk);
2016         else
2017                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2018 }
2019
2020 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2021 {
2022         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
2023 }
2024
2025 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
2026                                         unsigned int val, int alloc)
2027 {
2028         int i;
2029
2030         if (type == NAT_JOURNAL) {
2031                 for (i = 0; i < nats_in_cursum(journal); i++) {
2032                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
2033                                 return i;
2034                 }
2035                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2036                         return update_nats_in_cursum(journal, 1);
2037         } else if (type == SIT_JOURNAL) {
2038                 for (i = 0; i < sits_in_cursum(journal); i++)
2039                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
2040                                 return i;
2041                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2042                         return update_sits_in_cursum(journal, 1);
2043         }
2044         return -1;
2045 }
2046
2047 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2048                                         unsigned int segno)
2049 {
2050         return get_meta_page(sbi, current_sit_addr(sbi, segno));
2051 }
2052
2053 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2054                                         unsigned int start)
2055 {
2056         struct sit_info *sit_i = SIT_I(sbi);
2057         struct page *src_page, *dst_page;
2058         pgoff_t src_off, dst_off;
2059         void *src_addr, *dst_addr;
2060
2061         src_off = current_sit_addr(sbi, start);
2062         dst_off = next_sit_addr(sbi, src_off);
2063
2064         /* get current sit block page without lock */
2065         src_page = get_meta_page(sbi, src_off);
2066         dst_page = grab_meta_page(sbi, dst_off);
2067         f2fs_bug_on(sbi, PageDirty(src_page));
2068
2069         src_addr = page_address(src_page);
2070         dst_addr = page_address(dst_page);
2071         memcpy(dst_addr, src_addr, PAGE_SIZE);
2072
2073         set_page_dirty(dst_page);
2074         f2fs_put_page(src_page, 1);
2075
2076         set_to_next_sit(sit_i, start);
2077
2078         return dst_page;
2079 }
2080
2081 static struct sit_entry_set *grab_sit_entry_set(void)
2082 {
2083         struct sit_entry_set *ses =
2084                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
2085
2086         ses->entry_cnt = 0;
2087         INIT_LIST_HEAD(&ses->set_list);
2088         return ses;
2089 }
2090
2091 static void release_sit_entry_set(struct sit_entry_set *ses)
2092 {
2093         list_del(&ses->set_list);
2094         kmem_cache_free(sit_entry_set_slab, ses);
2095 }
2096
2097 static void adjust_sit_entry_set(struct sit_entry_set *ses,
2098                                                 struct list_head *head)
2099 {
2100         struct sit_entry_set *next = ses;
2101
2102         if (list_is_last(&ses->set_list, head))
2103                 return;
2104
2105         list_for_each_entry_continue(next, head, set_list)
2106                 if (ses->entry_cnt <= next->entry_cnt)
2107                         break;
2108
2109         list_move_tail(&ses->set_list, &next->set_list);
2110 }
2111
2112 static void add_sit_entry(unsigned int segno, struct list_head *head)
2113 {
2114         struct sit_entry_set *ses;
2115         unsigned int start_segno = START_SEGNO(segno);
2116
2117         list_for_each_entry(ses, head, set_list) {
2118                 if (ses->start_segno == start_segno) {
2119                         ses->entry_cnt++;
2120                         adjust_sit_entry_set(ses, head);
2121                         return;
2122                 }
2123         }
2124
2125         ses = grab_sit_entry_set();
2126
2127         ses->start_segno = start_segno;
2128         ses->entry_cnt++;
2129         list_add(&ses->set_list, head);
2130 }
2131
2132 static void add_sits_in_set(struct f2fs_sb_info *sbi)
2133 {
2134         struct f2fs_sm_info *sm_info = SM_I(sbi);
2135         struct list_head *set_list = &sm_info->sit_entry_set;
2136         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
2137         unsigned int segno;
2138
2139         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
2140                 add_sit_entry(segno, set_list);
2141 }
2142
2143 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
2144 {
2145         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2146         struct f2fs_journal *journal = curseg->journal;
2147         int i;
2148
2149         down_write(&curseg->journal_rwsem);
2150         for (i = 0; i < sits_in_cursum(journal); i++) {
2151                 unsigned int segno;
2152                 bool dirtied;
2153
2154                 segno = le32_to_cpu(segno_in_journal(journal, i));
2155                 dirtied = __mark_sit_entry_dirty(sbi, segno);
2156
2157                 if (!dirtied)
2158                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
2159         }
2160         update_sits_in_cursum(journal, -i);
2161         up_write(&curseg->journal_rwsem);
2162 }
2163
2164 /*
2165  * CP calls this function, which flushes SIT entries including sit_journal,
2166  * and moves prefree segs to free segs.
2167  */
2168 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2169 {
2170         struct sit_info *sit_i = SIT_I(sbi);
2171         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2172         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2173         struct f2fs_journal *journal = curseg->journal;
2174         struct sit_entry_set *ses, *tmp;
2175         struct list_head *head = &SM_I(sbi)->sit_entry_set;
2176         bool to_journal = true;
2177         struct seg_entry *se;
2178
2179         mutex_lock(&sit_i->sentry_lock);
2180
2181         if (!sit_i->dirty_sentries)
2182                 goto out;
2183
2184         /*
2185          * add and account sit entries of dirty bitmap in sit entry
2186          * set temporarily
2187          */
2188         add_sits_in_set(sbi);
2189
2190         /*
2191          * if there are no enough space in journal to store dirty sit
2192          * entries, remove all entries from journal and add and account
2193          * them in sit entry set.
2194          */
2195         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
2196                 remove_sits_in_journal(sbi);
2197
2198         /*
2199          * there are two steps to flush sit entries:
2200          * #1, flush sit entries to journal in current cold data summary block.
2201          * #2, flush sit entries to sit page.
2202          */
2203         list_for_each_entry_safe(ses, tmp, head, set_list) {
2204                 struct page *page = NULL;
2205                 struct f2fs_sit_block *raw_sit = NULL;
2206                 unsigned int start_segno = ses->start_segno;
2207                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
2208                                                 (unsigned long)MAIN_SEGS(sbi));
2209                 unsigned int segno = start_segno;
2210
2211                 if (to_journal &&
2212                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
2213                         to_journal = false;
2214
2215                 if (to_journal) {
2216                         down_write(&curseg->journal_rwsem);
2217                 } else {
2218                         page = get_next_sit_page(sbi, start_segno);
2219                         raw_sit = page_address(page);
2220                 }
2221
2222                 /* flush dirty sit entries in region of current sit set */
2223                 for_each_set_bit_from(segno, bitmap, end) {
2224                         int offset, sit_offset;
2225
2226                         se = get_seg_entry(sbi, segno);
2227
2228                         /* add discard candidates */
2229                         if (cpc->reason != CP_DISCARD) {
2230                                 cpc->trim_start = segno;
2231                                 add_discard_addrs(sbi, cpc);
2232                         }
2233
2234                         if (to_journal) {
2235                                 offset = lookup_journal_in_cursum(journal,
2236                                                         SIT_JOURNAL, segno, 1);
2237                                 f2fs_bug_on(sbi, offset < 0);
2238                                 segno_in_journal(journal, offset) =
2239                                                         cpu_to_le32(segno);
2240                                 seg_info_to_raw_sit(se,
2241                                         &sit_in_journal(journal, offset));
2242                         } else {
2243                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2244                                 seg_info_to_raw_sit(se,
2245                                                 &raw_sit->entries[sit_offset]);
2246                         }
2247
2248                         __clear_bit(segno, bitmap);
2249                         sit_i->dirty_sentries--;
2250                         ses->entry_cnt--;
2251                 }
2252
2253                 if (to_journal)
2254                         up_write(&curseg->journal_rwsem);
2255                 else
2256                         f2fs_put_page(page, 1);
2257
2258                 f2fs_bug_on(sbi, ses->entry_cnt);
2259                 release_sit_entry_set(ses);
2260         }
2261
2262         f2fs_bug_on(sbi, !list_empty(head));
2263         f2fs_bug_on(sbi, sit_i->dirty_sentries);
2264 out:
2265         if (cpc->reason == CP_DISCARD) {
2266                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2267                         add_discard_addrs(sbi, cpc);
2268         }
2269         mutex_unlock(&sit_i->sentry_lock);
2270
2271         set_prefree_as_free_segments(sbi);
2272 }
2273
2274 static int build_sit_info(struct f2fs_sb_info *sbi)
2275 {
2276         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2277         struct sit_info *sit_i;
2278         unsigned int sit_segs, start;
2279         char *src_bitmap, *dst_bitmap;
2280         unsigned int bitmap_size;
2281
2282         /* allocate memory for SIT information */
2283         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2284         if (!sit_i)
2285                 return -ENOMEM;
2286
2287         SM_I(sbi)->sit_info = sit_i;
2288
2289         sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2290                                         sizeof(struct seg_entry), GFP_KERNEL);
2291         if (!sit_i->sentries)
2292                 return -ENOMEM;
2293
2294         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2295         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2296         if (!sit_i->dirty_sentries_bitmap)
2297                 return -ENOMEM;
2298
2299         for (start = 0; start < MAIN_SEGS(sbi); start++) {
2300                 sit_i->sentries[start].cur_valid_map
2301                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2302                 sit_i->sentries[start].ckpt_valid_map
2303                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2304                 if (!sit_i->sentries[start].cur_valid_map ||
2305                                 !sit_i->sentries[start].ckpt_valid_map)
2306                         return -ENOMEM;
2307
2308                 if (f2fs_discard_en(sbi)) {
2309                         sit_i->sentries[start].discard_map
2310                                 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2311                         if (!sit_i->sentries[start].discard_map)
2312                                 return -ENOMEM;
2313                 }
2314         }
2315
2316         sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2317         if (!sit_i->tmp_map)
2318                 return -ENOMEM;
2319
2320         if (sbi->segs_per_sec > 1) {
2321                 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2322                                         sizeof(struct sec_entry), GFP_KERNEL);
2323                 if (!sit_i->sec_entries)
2324                         return -ENOMEM;
2325         }
2326
2327         /* get information related with SIT */
2328         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2329
2330         /* setup SIT bitmap from ckeckpoint pack */
2331         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2332         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2333
2334         dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2335         if (!dst_bitmap)
2336                 return -ENOMEM;
2337
2338         /* init SIT information */
2339         sit_i->s_ops = &default_salloc_ops;
2340
2341         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2342         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2343         sit_i->written_valid_blocks = 0;
2344         sit_i->sit_bitmap = dst_bitmap;
2345         sit_i->bitmap_size = bitmap_size;
2346         sit_i->dirty_sentries = 0;
2347         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2348         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2349         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2350         mutex_init(&sit_i->sentry_lock);
2351         return 0;
2352 }
2353
2354 static int build_free_segmap(struct f2fs_sb_info *sbi)
2355 {
2356         struct free_segmap_info *free_i;
2357         unsigned int bitmap_size, sec_bitmap_size;
2358
2359         /* allocate memory for free segmap information */
2360         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2361         if (!free_i)
2362                 return -ENOMEM;
2363
2364         SM_I(sbi)->free_info = free_i;
2365
2366         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2367         free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2368         if (!free_i->free_segmap)
2369                 return -ENOMEM;
2370
2371         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2372         free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2373         if (!free_i->free_secmap)
2374                 return -ENOMEM;
2375
2376         /* set all segments as dirty temporarily */
2377         memset(free_i->free_segmap, 0xff, bitmap_size);
2378         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2379
2380         /* init free segmap information */
2381         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2382         free_i->free_segments = 0;
2383         free_i->free_sections = 0;
2384         spin_lock_init(&free_i->segmap_lock);
2385         return 0;
2386 }
2387
2388 static int build_curseg(struct f2fs_sb_info *sbi)
2389 {
2390         struct curseg_info *array;
2391         int i;
2392
2393         array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2394         if (!array)
2395                 return -ENOMEM;
2396
2397         SM_I(sbi)->curseg_array = array;
2398
2399         for (i = 0; i < NR_CURSEG_TYPE; i++) {
2400                 mutex_init(&array[i].curseg_mutex);
2401                 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2402                 if (!array[i].sum_blk)
2403                         return -ENOMEM;
2404                 init_rwsem(&array[i].journal_rwsem);
2405                 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2406                                                         GFP_KERNEL);
2407                 if (!array[i].journal)
2408                         return -ENOMEM;
2409                 array[i].segno = NULL_SEGNO;
2410                 array[i].next_blkoff = 0;
2411         }
2412         return restore_curseg_summaries(sbi);
2413 }
2414
2415 static void build_sit_entries(struct f2fs_sb_info *sbi)
2416 {
2417         struct sit_info *sit_i = SIT_I(sbi);
2418         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2419         struct f2fs_journal *journal = curseg->journal;
2420         struct seg_entry *se;
2421         struct f2fs_sit_entry sit;
2422         int sit_blk_cnt = SIT_BLK_CNT(sbi);
2423         unsigned int i, start, end;
2424         unsigned int readed, start_blk = 0;
2425
2426         do {
2427                 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
2428                                                         META_SIT, true);
2429
2430                 start = start_blk * sit_i->sents_per_block;
2431                 end = (start_blk + readed) * sit_i->sents_per_block;
2432
2433                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2434                         struct f2fs_sit_block *sit_blk;
2435                         struct page *page;
2436
2437                         se = &sit_i->sentries[start];
2438                         page = get_current_sit_page(sbi, start);
2439                         sit_blk = (struct f2fs_sit_block *)page_address(page);
2440                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2441                         f2fs_put_page(page, 1);
2442
2443                         check_block_count(sbi, start, &sit);
2444                         seg_info_from_raw_sit(se, &sit);
2445
2446                         /* build discard map only one time */
2447                         if (f2fs_discard_en(sbi)) {
2448                                 memcpy(se->discard_map, se->cur_valid_map,
2449                                                         SIT_VBLOCK_MAP_SIZE);
2450                                 sbi->discard_blks += sbi->blocks_per_seg -
2451                                                         se->valid_blocks;
2452                         }
2453
2454                         if (sbi->segs_per_sec > 1)
2455                                 get_sec_entry(sbi, start)->valid_blocks +=
2456                                                         se->valid_blocks;
2457                 }
2458                 start_blk += readed;
2459         } while (start_blk < sit_blk_cnt);
2460
2461         down_read(&curseg->journal_rwsem);
2462         for (i = 0; i < sits_in_cursum(journal); i++) {
2463                 unsigned int old_valid_blocks;
2464
2465                 start = le32_to_cpu(segno_in_journal(journal, i));
2466                 se = &sit_i->sentries[start];
2467                 sit = sit_in_journal(journal, i);
2468
2469                 old_valid_blocks = se->valid_blocks;
2470
2471                 check_block_count(sbi, start, &sit);
2472                 seg_info_from_raw_sit(se, &sit);
2473
2474                 if (f2fs_discard_en(sbi)) {
2475                         memcpy(se->discard_map, se->cur_valid_map,
2476                                                 SIT_VBLOCK_MAP_SIZE);
2477                         sbi->discard_blks += old_valid_blocks -
2478                                                 se->valid_blocks;
2479                 }
2480
2481                 if (sbi->segs_per_sec > 1)
2482                         get_sec_entry(sbi, start)->valid_blocks +=
2483                                 se->valid_blocks - old_valid_blocks;
2484         }
2485         up_read(&curseg->journal_rwsem);
2486 }
2487
2488 static void init_free_segmap(struct f2fs_sb_info *sbi)
2489 {
2490         unsigned int start;
2491         int type;
2492
2493         for (start = 0; start < MAIN_SEGS(sbi); start++) {
2494                 struct seg_entry *sentry = get_seg_entry(sbi, start);
2495                 if (!sentry->valid_blocks)
2496                         __set_free(sbi, start);
2497                 else
2498                         SIT_I(sbi)->written_valid_blocks +=
2499                                                 sentry->valid_blocks;
2500         }
2501
2502         /* set use the current segments */
2503         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2504                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2505                 __set_test_and_inuse(sbi, curseg_t->segno);
2506         }
2507 }
2508
2509 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2510 {
2511         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2512         struct free_segmap_info *free_i = FREE_I(sbi);
2513         unsigned int segno = 0, offset = 0;
2514         unsigned short valid_blocks;
2515
2516         while (1) {
2517                 /* find dirty segment based on free segmap */
2518                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2519                 if (segno >= MAIN_SEGS(sbi))
2520                         break;
2521                 offset = segno + 1;
2522                 valid_blocks = get_valid_blocks(sbi, segno, 0);
2523                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2524                         continue;
2525                 if (valid_blocks > sbi->blocks_per_seg) {
2526                         f2fs_bug_on(sbi, 1);
2527                         continue;
2528                 }
2529                 mutex_lock(&dirty_i->seglist_lock);
2530                 __locate_dirty_segment(sbi, segno, DIRTY);
2531                 mutex_unlock(&dirty_i->seglist_lock);
2532         }
2533 }
2534
2535 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2536 {
2537         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2538         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2539
2540         dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2541         if (!dirty_i->victim_secmap)
2542                 return -ENOMEM;
2543         return 0;
2544 }
2545
2546 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2547 {
2548         struct dirty_seglist_info *dirty_i;
2549         unsigned int bitmap_size, i;
2550
2551         /* allocate memory for dirty segments list information */
2552         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2553         if (!dirty_i)
2554                 return -ENOMEM;
2555
2556         SM_I(sbi)->dirty_info = dirty_i;
2557         mutex_init(&dirty_i->seglist_lock);
2558
2559         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2560
2561         for (i = 0; i < NR_DIRTY_TYPE; i++) {
2562                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2563                 if (!dirty_i->dirty_segmap[i])
2564                         return -ENOMEM;
2565         }
2566
2567         init_dirty_segmap(sbi);
2568         return init_victim_secmap(sbi);
2569 }
2570
2571 /*
2572  * Update min, max modified time for cost-benefit GC algorithm
2573  */
2574 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2575 {
2576         struct sit_info *sit_i = SIT_I(sbi);
2577         unsigned int segno;
2578
2579         mutex_lock(&sit_i->sentry_lock);
2580
2581         sit_i->min_mtime = LLONG_MAX;
2582
2583         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2584                 unsigned int i;
2585                 unsigned long long mtime = 0;
2586
2587                 for (i = 0; i < sbi->segs_per_sec; i++)
2588                         mtime += get_seg_entry(sbi, segno + i)->mtime;
2589
2590                 mtime = div_u64(mtime, sbi->segs_per_sec);
2591
2592                 if (sit_i->min_mtime > mtime)
2593                         sit_i->min_mtime = mtime;
2594         }
2595         sit_i->max_mtime = get_mtime(sbi);
2596         mutex_unlock(&sit_i->sentry_lock);
2597 }
2598
2599 int build_segment_manager(struct f2fs_sb_info *sbi)
2600 {
2601         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2602         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2603         struct f2fs_sm_info *sm_info;
2604         int err;
2605
2606         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2607         if (!sm_info)
2608                 return -ENOMEM;
2609
2610         /* init sm info */
2611         sbi->sm_info = sm_info;
2612         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2613         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2614         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2615         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2616         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2617         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2618         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2619         sm_info->rec_prefree_segments = sm_info->main_segments *
2620                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2621         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2622                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2623
2624         if (!test_opt(sbi, LFS))
2625                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2626         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2627         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2628
2629         INIT_LIST_HEAD(&sm_info->discard_list);
2630         INIT_LIST_HEAD(&sm_info->wait_list);
2631         sm_info->nr_discards = 0;
2632         sm_info->max_discards = 0;
2633
2634         sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2635
2636         INIT_LIST_HEAD(&sm_info->sit_entry_set);
2637
2638         if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2639                 err = create_flush_cmd_control(sbi);
2640                 if (err)
2641                         return err;
2642         }
2643
2644         err = build_sit_info(sbi);
2645         if (err)
2646                 return err;
2647         err = build_free_segmap(sbi);
2648         if (err)
2649                 return err;
2650         err = build_curseg(sbi);
2651         if (err)
2652                 return err;
2653
2654         /* reinit free segmap based on SIT */
2655         build_sit_entries(sbi);
2656
2657         init_free_segmap(sbi);
2658         err = build_dirty_segmap(sbi);
2659         if (err)
2660                 return err;
2661
2662         init_min_max_mtime(sbi);
2663         return 0;
2664 }
2665
2666 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2667                 enum dirty_type dirty_type)
2668 {
2669         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2670
2671         mutex_lock(&dirty_i->seglist_lock);
2672         kvfree(dirty_i->dirty_segmap[dirty_type]);
2673         dirty_i->nr_dirty[dirty_type] = 0;
2674         mutex_unlock(&dirty_i->seglist_lock);
2675 }
2676
2677 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2678 {
2679         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2680         kvfree(dirty_i->victim_secmap);
2681 }
2682
2683 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2684 {
2685         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2686         int i;
2687
2688         if (!dirty_i)
2689                 return;
2690
2691         /* discard pre-free/dirty segments list */
2692         for (i = 0; i < NR_DIRTY_TYPE; i++)
2693                 discard_dirty_segmap(sbi, i);
2694
2695         destroy_victim_secmap(sbi);
2696         SM_I(sbi)->dirty_info = NULL;
2697         kfree(dirty_i);
2698 }
2699
2700 static void destroy_curseg(struct f2fs_sb_info *sbi)
2701 {
2702         struct curseg_info *array = SM_I(sbi)->curseg_array;
2703         int i;
2704
2705         if (!array)
2706                 return;
2707         SM_I(sbi)->curseg_array = NULL;
2708         for (i = 0; i < NR_CURSEG_TYPE; i++) {
2709                 kfree(array[i].sum_blk);
2710                 kfree(array[i].journal);
2711         }
2712         kfree(array);
2713 }
2714
2715 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2716 {
2717         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2718         if (!free_i)
2719                 return;
2720         SM_I(sbi)->free_info = NULL;
2721         kvfree(free_i->free_segmap);
2722         kvfree(free_i->free_secmap);
2723         kfree(free_i);
2724 }
2725
2726 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2727 {
2728         struct sit_info *sit_i = SIT_I(sbi);
2729         unsigned int start;
2730
2731         if (!sit_i)
2732                 return;
2733
2734         if (sit_i->sentries) {
2735                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2736                         kfree(sit_i->sentries[start].cur_valid_map);
2737                         kfree(sit_i->sentries[start].ckpt_valid_map);
2738                         kfree(sit_i->sentries[start].discard_map);
2739                 }
2740         }
2741         kfree(sit_i->tmp_map);
2742
2743         kvfree(sit_i->sentries);
2744         kvfree(sit_i->sec_entries);
2745         kvfree(sit_i->dirty_sentries_bitmap);
2746
2747         SM_I(sbi)->sit_info = NULL;
2748         kfree(sit_i->sit_bitmap);
2749         kfree(sit_i);
2750 }
2751
2752 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2753 {
2754         struct f2fs_sm_info *sm_info = SM_I(sbi);
2755
2756         if (!sm_info)
2757                 return;
2758         destroy_flush_cmd_control(sbi, true);
2759         destroy_dirty_segmap(sbi);
2760         destroy_curseg(sbi);
2761         destroy_free_segmap(sbi);
2762         destroy_sit_info(sbi);
2763         sbi->sm_info = NULL;
2764         kfree(sm_info);
2765 }
2766
2767 int __init create_segment_manager_caches(void)
2768 {
2769         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2770                         sizeof(struct discard_entry));
2771         if (!discard_entry_slab)
2772                 goto fail;
2773
2774         bio_entry_slab = f2fs_kmem_cache_create("bio_entry",
2775                         sizeof(struct bio_entry));
2776         if (!bio_entry_slab)
2777                 goto destroy_discard_entry;
2778
2779         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2780                         sizeof(struct sit_entry_set));
2781         if (!sit_entry_set_slab)
2782                 goto destroy_bio_entry;
2783
2784         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2785                         sizeof(struct inmem_pages));
2786         if (!inmem_entry_slab)
2787                 goto destroy_sit_entry_set;
2788         return 0;
2789
2790 destroy_sit_entry_set:
2791         kmem_cache_destroy(sit_entry_set_slab);
2792 destroy_bio_entry:
2793         kmem_cache_destroy(bio_entry_slab);
2794 destroy_discard_entry:
2795         kmem_cache_destroy(discard_entry_slab);
2796 fail:
2797         return -ENOMEM;
2798 }
2799
2800 void destroy_segment_manager_caches(void)
2801 {
2802         kmem_cache_destroy(sit_entry_set_slab);
2803         kmem_cache_destroy(bio_entry_slab);
2804         kmem_cache_destroy(discard_entry_slab);
2805         kmem_cache_destroy(inmem_entry_slab);
2806 }