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 *discard_cmd_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         mutex_lock(&fi->inmem_lock);
 246         __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 247         mutex_unlock(&fi->inmem_lock);
 248
 249         clear_inode_flag(inode, FI_ATOMIC_FILE);
 250         stat_dec_atomic_write(inode);
 251 }
 252
 253 void drop_inmem_page(struct inode *inode, struct page *page)
 254 {
 255         struct f2fs_inode_info *fi = F2FS_I(inode);
 256         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 257         struct list_head *head = &fi->inmem_pages;
 258         struct inmem_pages *cur = NULL;
 259
 260         f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
 261
 262         mutex_lock(&fi->inmem_lock);
 263         list_for_each_entry(cur, head, list) {
 264                 if (cur->page == page)
 265                         break;
 266         }
 267
 268         f2fs_bug_on(sbi, !cur || cur->page != page);
 269         list_del(&cur->list);
 270         mutex_unlock(&fi->inmem_lock);
 271
 272         dec_page_count(sbi, F2FS_INMEM_PAGES);
 273         kmem_cache_free(inmem_entry_slab, cur);
 274
 275         ClearPageUptodate(page);
 276         set_page_private(page, 0);
 277         ClearPagePrivate(page);
 278         f2fs_put_page(page, 0);
 279
 280         trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 281 }
 282
 283 static int __commit_inmem_pages(struct inode *inode,
 284                                         struct list_head *revoke_list)
 285 {
 286         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 287         struct f2fs_inode_info *fi = F2FS_I(inode);
 288         struct inmem_pages *cur, *tmp;
 289         struct f2fs_io_info fio = {
 290                 .sbi = sbi,
 291                 .type = DATA,
 292                 .op = REQ_OP_WRITE,
 293                 .op_flags = REQ_SYNC | REQ_PRIO,
 294         };
 295         pgoff_t last_idx = ULONG_MAX;
 296         int err = 0;
 297
 298         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 299                 struct page *page = cur->page;
 300
 301                 lock_page(page);
 302                 if (page->mapping == inode->i_mapping) {
 303                         trace_f2fs_commit_inmem_page(page, INMEM);
 304
 305                         set_page_dirty(page);
 306                         f2fs_wait_on_page_writeback(page, DATA, true);
 307                         if (clear_page_dirty_for_io(page)) {
 308                                 inode_dec_dirty_pages(inode);
 309                                 remove_dirty_inode(inode);
 310                         }
 311
 312                         fio.page = page;
 313                         fio.old_blkaddr = NULL_ADDR;
 314                         fio.encrypted_page = NULL;
 315                         fio.need_lock = false,
 316                         err = do_write_data_page(&fio);
 317                         if (err) {
 318                                 unlock_page(page);
 319                                 break;
 320                         }
 321
 322                         /* record old blkaddr for revoking */
 323                         cur->old_addr = fio.old_blkaddr;
 324                         last_idx = page->index;
 325                 }
 326                 unlock_page(page);
 327                 list_move_tail(&cur->list, revoke_list);
 328         }
 329
 330         if (last_idx != ULONG_MAX)
 331                 f2fs_submit_merged_bio_cond(sbi, inode, 0, last_idx,
 332                                                         DATA, WRITE);
 333
 334         if (!err)
 335                 __revoke_inmem_pages(inode, revoke_list, false, false);
 336
 337         return err;
 338 }
 339
 340 int commit_inmem_pages(struct inode *inode)
 341 {
 342         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 343         struct f2fs_inode_info *fi = F2FS_I(inode);
 344         struct list_head revoke_list;
 345         int err;
 346
 347         INIT_LIST_HEAD(&revoke_list);
 348         f2fs_balance_fs(sbi, true);
 349         f2fs_lock_op(sbi);
 350
 351         set_inode_flag(inode, FI_ATOMIC_COMMIT);
 352
 353         mutex_lock(&fi->inmem_lock);
 354         err = __commit_inmem_pages(inode, &revoke_list);
 355         if (err) {
 356                 int ret;
 357                 /*
 358                  * try to revoke all committed pages, but still we could fail
 359                  * due to no memory or other reason, if that happened, EAGAIN
 360                  * will be returned, which means in such case, transaction is
 361                  * already not integrity, caller should use journal to do the
 362                  * recovery or rewrite & commit last transaction. For other
 363                  * error number, revoking was done by filesystem itself.
 364                  */
 365                 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
 366                 if (ret)
 367                         err = ret;
 368
 369                 /* drop all uncommitted pages */
 370                 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 371         }
 372         mutex_unlock(&fi->inmem_lock);
 373
 374         clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 375
 376         f2fs_unlock_op(sbi);
 377         return err;
 378 }
 379
 380 /*
 381  * This function balances dirty node and dentry pages.
 382  * In addition, it controls garbage collection.
 383  */
 384 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 385 {
 386 #ifdef CONFIG_F2FS_FAULT_INJECTION
 387         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 388                 f2fs_show_injection_info(FAULT_CHECKPOINT);
 389                 f2fs_stop_checkpoint(sbi, false);
 390         }
 391 #endif
 392
 393         /* balance_fs_bg is able to be pending */
 394         if (need && excess_cached_nats(sbi))
 395                 f2fs_balance_fs_bg(sbi);
 396
 397         /*
 398          * We should do GC or end up with checkpoint, if there are so many dirty
 399          * dir/node pages without enough free segments.
 400          */
 401         if (has_not_enough_free_secs(sbi, 0, 0)) {
 402                 mutex_lock(&sbi->gc_mutex);
 403                 f2fs_gc(sbi, false, false, NULL_SEGNO);
 404         }
 405 }
 406
 407 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 408 {
 409         /* try to shrink extent cache when there is no enough memory */
 410         if (!available_free_memory(sbi, EXTENT_CACHE))
 411                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 412
 413         /* check the # of cached NAT entries */
 414         if (!available_free_memory(sbi, NAT_ENTRIES))
 415                 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 416
 417         if (!available_free_memory(sbi, FREE_NIDS))
 418                 try_to_free_nids(sbi, MAX_FREE_NIDS);
 419         else
 420                 build_free_nids(sbi, false, false);
 421
 422         if (!is_idle(sbi) && !excess_dirty_nats(sbi))
 423                 return;
 424
 425         /* checkpoint is the only way to shrink partial cached entries */
 426         if (!available_free_memory(sbi, NAT_ENTRIES) ||
 427                         !available_free_memory(sbi, INO_ENTRIES) ||
 428                         excess_prefree_segs(sbi) ||
 429                         excess_dirty_nats(sbi) ||
 430                         f2fs_time_over(sbi, CP_TIME)) {
 431                 if (test_opt(sbi, DATA_FLUSH)) {
 432                         struct blk_plug plug;
 433
 434                         blk_start_plug(&plug);
 435                         sync_dirty_inodes(sbi, FILE_INODE);
 436                         blk_finish_plug(&plug);
 437                 }
 438                 f2fs_sync_fs(sbi->sb, true);
 439                 stat_inc_bg_cp_count(sbi->stat_info);
 440         }
 441 }
 442
 443 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 444                                 struct block_device *bdev)
 445 {
 446         struct bio *bio = f2fs_bio_alloc(0);
 447         int ret;
 448
 449         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
 450         bio->bi_bdev = bdev;
 451         ret = submit_bio_wait(bio);
 452         bio_put(bio);
 453
 454         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 455                                 test_opt(sbi, FLUSH_MERGE), ret);
 456         return ret;
 457 }
 458
 459 static int submit_flush_wait(struct f2fs_sb_info *sbi)
 460 {
 461         int ret = __submit_flush_wait(sbi, sbi->sb->s_bdev);
 462         int i;
 463
 464         if (!sbi->s_ndevs || ret)
 465                 return ret;
 466
 467         for (i = 1; i < sbi->s_ndevs; i++) {
 468                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 469                 if (ret)
 470                         break;
 471         }
 472         return ret;
 473 }
 474
 475 static int issue_flush_thread(void *data)
 476 {
 477         struct f2fs_sb_info *sbi = data;
 478         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 479         wait_queue_head_t *q = &fcc->flush_wait_queue;
 480 repeat:
 481         if (kthread_should_stop())
 482                 return 0;
 483
 484         if (!llist_empty(&fcc->issue_list)) {
 485                 struct flush_cmd *cmd, *next;
 486                 int ret;
 487
 488                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 489                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 490
 491                 ret = submit_flush_wait(sbi);
 492                 atomic_inc(&fcc->issued_flush);
 493
 494                 llist_for_each_entry_safe(cmd, next,
 495                                           fcc->dispatch_list, llnode) {
 496                         cmd->ret = ret;
 497                         complete(&cmd->wait);
 498                 }
 499                 fcc->dispatch_list = NULL;
 500         }
 501
 502         wait_event_interruptible(*q,
 503                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 504         goto repeat;
 505 }
 506
 507 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
 508 {
 509         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 510         struct flush_cmd cmd;
 511         int ret;
 512
 513         if (test_opt(sbi, NOBARRIER))
 514                 return 0;
 515
 516         if (!test_opt(sbi, FLUSH_MERGE)) {
 517                 ret = submit_flush_wait(sbi);
 518                 atomic_inc(&fcc->issued_flush);
 519                 return ret;
 520         }
 521
 522         if (!atomic_read(&fcc->issing_flush)) {
 523                 atomic_inc(&fcc->issing_flush);
 524                 ret = submit_flush_wait(sbi);
 525                 atomic_dec(&fcc->issing_flush);
 526
 527                 atomic_inc(&fcc->issued_flush);
 528                 return ret;
 529         }
 530
 531         init_completion(&cmd.wait);
 532
 533         atomic_inc(&fcc->issing_flush);
 534         llist_add(&cmd.llnode, &fcc->issue_list);
 535
 536         if (!fcc->dispatch_list)
 537                 wake_up(&fcc->flush_wait_queue);
 538
 539         if (fcc->f2fs_issue_flush) {
 540                 wait_for_completion(&cmd.wait);
 541                 atomic_dec(&fcc->issing_flush);
 542         } else {
 543                 llist_del_all(&fcc->issue_list);
 544                 atomic_set(&fcc->issing_flush, 0);
 545         }
 546
 547         return cmd.ret;
 548 }
 549
 550 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
 551 {
 552         dev_t dev = sbi->sb->s_bdev->bd_dev;
 553         struct flush_cmd_control *fcc;
 554         int err = 0;
 555
 556         if (SM_I(sbi)->fcc_info) {
 557                 fcc = SM_I(sbi)->fcc_info;
 558                 goto init_thread;
 559         }
 560
 561         fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
 562         if (!fcc)
 563                 return -ENOMEM;
 564         atomic_set(&fcc->issued_flush, 0);
 565         atomic_set(&fcc->issing_flush, 0);
 566         init_waitqueue_head(&fcc->flush_wait_queue);
 567         init_llist_head(&fcc->issue_list);
 568         SM_I(sbi)->fcc_info = fcc;
 569 init_thread:
 570         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 571                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 572         if (IS_ERR(fcc->f2fs_issue_flush)) {
 573                 err = PTR_ERR(fcc->f2fs_issue_flush);
 574                 kfree(fcc);
 575                 SM_I(sbi)->fcc_info = NULL;
 576                 return err;
 577         }
 578
 579         return err;
 580 }
 581
 582 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 583 {
 584         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 585
 586         if (fcc && fcc->f2fs_issue_flush) {
 587                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 588
 589                 fcc->f2fs_issue_flush = NULL;
 590                 kthread_stop(flush_thread);
 591         }
 592         if (free) {
 593                 kfree(fcc);
 594                 SM_I(sbi)->fcc_info = NULL;
 595         }
 596 }
 597
 598 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 599                 enum dirty_type dirty_type)
 600 {
 601         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 602
 603         /* need not be added */
 604         if (IS_CURSEG(sbi, segno))
 605                 return;
 606
 607         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 608                 dirty_i->nr_dirty[dirty_type]++;
 609
 610         if (dirty_type == DIRTY) {
 611                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 612                 enum dirty_type t = sentry->type;
 613
 614                 if (unlikely(t >= DIRTY)) {
 615                         f2fs_bug_on(sbi, 1);
 616                         return;
 617                 }
 618                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 619                         dirty_i->nr_dirty[t]++;
 620         }
 621 }
 622
 623 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 624                 enum dirty_type dirty_type)
 625 {
 626         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 627
 628         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 629                 dirty_i->nr_dirty[dirty_type]--;
 630
 631         if (dirty_type == DIRTY) {
 632                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 633                 enum dirty_type t = sentry->type;
 634
 635                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 636                         dirty_i->nr_dirty[t]--;
 637
 638                 if (get_valid_blocks(sbi, segno, true) == 0)
 639                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 640                                                 dirty_i->victim_secmap);
 641         }
 642 }
 643
 644 /*
 645  * Should not occur error such as -ENOMEM.
 646  * Adding dirty entry into seglist is not critical operation.
 647  * If a given segment is one of current working segments, it won't be added.
 648  */
 649 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 650 {
 651         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 652         unsigned short valid_blocks;
 653
 654         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 655                 return;
 656
 657         mutex_lock(&dirty_i->seglist_lock);
 658
 659         valid_blocks = get_valid_blocks(sbi, segno, false);
 660
 661         if (valid_blocks == 0) {
 662                 __locate_dirty_segment(sbi, segno, PRE);
 663                 __remove_dirty_segment(sbi, segno, DIRTY);
 664         } else if (valid_blocks < sbi->blocks_per_seg) {
 665                 __locate_dirty_segment(sbi, segno, DIRTY);
 666         } else {
 667                 /* Recovery routine with SSR needs this */
 668                 __remove_dirty_segment(sbi, segno, DIRTY);
 669         }
 670
 671         mutex_unlock(&dirty_i->seglist_lock);
 672 }
 673
 674 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 675                 struct block_device *bdev, block_t lstart,
 676                 block_t start, block_t len)
 677 {
 678         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 679         struct list_head *pend_list;
 680         struct discard_cmd *dc;
 681
 682         f2fs_bug_on(sbi, !len);
 683
 684         pend_list = &dcc->pend_list[plist_idx(len)];
 685
 686         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 687         INIT_LIST_HEAD(&dc->list);
 688         dc->bdev = bdev;
 689         dc->lstart = lstart;
 690         dc->start = start;
 691         dc->len = len;
 692         dc->ref = 0;
 693         dc->state = D_PREP;
 694         dc->error = 0;
 695         init_completion(&dc->wait);
 696         list_add_tail(&dc->list, pend_list);
 697         atomic_inc(&dcc->discard_cmd_cnt);
 698         dcc->undiscard_blks += len;
 699
 700         return dc;
 701 }
 702
 703 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 704                                 struct block_device *bdev, block_t lstart,
 705                                 block_t start, block_t len,
 706                                 struct rb_node *parent, struct rb_node **p)
 707 {
 708         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 709         struct discard_cmd *dc;
 710
 711         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
 712
 713         rb_link_node(&dc->rb_node, parent, p);
 714         rb_insert_color(&dc->rb_node, &dcc->root);
 715
 716         return dc;
 717 }
 718
 719 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
 720                                                         struct discard_cmd *dc)
 721 {
 722         if (dc->state == D_DONE)
 723                 atomic_dec(&dcc->issing_discard);
 724
 725         list_del(&dc->list);
 726         rb_erase(&dc->rb_node, &dcc->root);
 727         dcc->undiscard_blks -= dc->len;
 728
 729         kmem_cache_free(discard_cmd_slab, dc);
 730
 731         atomic_dec(&dcc->discard_cmd_cnt);
 732 }
 733
 734 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
 735                                                         struct discard_cmd *dc)
 736 {
 737         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 738
 739         if (dc->error == -EOPNOTSUPP)
 740                 dc->error = 0;
 741
 742         if (dc->error)
 743                 f2fs_msg(sbi->sb, KERN_INFO,
 744                                 "Issue discard failed, ret: %d", dc->error);
 745         __detach_discard_cmd(dcc, dc);
 746 }
 747
 748 static void f2fs_submit_discard_endio(struct bio *bio)
 749 {
 750         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
 751
 752         dc->error = blk_status_to_errno(bio->bi_status);
 753         dc->state = D_DONE;
 754         complete(&dc->wait);
 755         bio_put(bio);
 756 }
 757
 758 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
 759 static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
 760                                 struct discard_cmd *dc)
 761 {
 762         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 763         struct bio *bio = NULL;
 764
 765         if (dc->state != D_PREP)
 766                 return;
 767
 768         trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
 769
 770         dc->error = __blkdev_issue_discard(dc->bdev,
 771                                 SECTOR_FROM_BLOCK(dc->start),
 772                                 SECTOR_FROM_BLOCK(dc->len),
 773                                 GFP_NOFS, 0, &bio);
 774         if (!dc->error) {
 775                 /* should keep before submission to avoid D_DONE right away */
 776                 dc->state = D_SUBMIT;
 777                 atomic_inc(&dcc->issued_discard);
 778                 atomic_inc(&dcc->issing_discard);
 779                 if (bio) {
 780                         bio->bi_private = dc;
 781                         bio->bi_end_io = f2fs_submit_discard_endio;
 782                         bio->bi_opf |= REQ_SYNC;
 783                         submit_bio(bio);
 784                         list_move_tail(&dc->list, &dcc->wait_list);
 785                 }
 786         } else {
 787                 __remove_discard_cmd(sbi, dc);
 788         }
 789 }
 790
 791 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
 792                                 struct block_device *bdev, block_t lstart,
 793                                 block_t start, block_t len,
 794                                 struct rb_node **insert_p,
 795                                 struct rb_node *insert_parent)
 796 {
 797         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 798         struct rb_node **p = &dcc->root.rb_node;
 799         struct rb_node *parent = NULL;
 800         struct discard_cmd *dc = NULL;
 801
 802         if (insert_p && insert_parent) {
 803                 parent = insert_parent;
 804                 p = insert_p;
 805                 goto do_insert;
 806         }
 807
 808         p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
 809 do_insert:
 810         dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
 811         if (!dc)
 812                 return NULL;
 813
 814         return dc;
 815 }
 816
 817 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
 818                                                 struct discard_cmd *dc)
 819 {
 820         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
 821 }
 822
 823 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
 824                                 struct discard_cmd *dc, block_t blkaddr)
 825 {
 826         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 827         struct discard_info di = dc->di;
 828         bool modified = false;
 829
 830         if (dc->state == D_DONE || dc->len == 1) {
 831                 __remove_discard_cmd(sbi, dc);
 832                 return;
 833         }
 834
 835         dcc->undiscard_blks -= di.len;
 836
 837         if (blkaddr > di.lstart) {
 838                 dc->len = blkaddr - dc->lstart;
 839                 dcc->undiscard_blks += dc->len;
 840                 __relocate_discard_cmd(dcc, dc);
 841                 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
 842                 modified = true;
 843         }
 844
 845         if (blkaddr < di.lstart + di.len - 1) {
 846                 if (modified) {
 847                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
 848                                         di.start + blkaddr + 1 - di.lstart,
 849                                         di.lstart + di.len - 1 - blkaddr,
 850                                         NULL, NULL);
 851                         f2fs_bug_on(sbi,
 852                                 !__check_rb_tree_consistence(sbi, &dcc->root));
 853                 } else {
 854                         dc->lstart++;
 855                         dc->len--;
 856                         dc->start++;
 857                         dcc->undiscard_blks += dc->len;
 858                         __relocate_discard_cmd(dcc, dc);
 859                         f2fs_bug_on(sbi,
 860                                 !__check_rb_tree_consistence(sbi, &dcc->root));
 861                 }
 862         }
 863 }
 864
 865 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
 866                                 struct block_device *bdev, block_t lstart,
 867                                 block_t start, block_t len)
 868 {
 869         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 870         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
 871         struct discard_cmd *dc;
 872         struct discard_info di = {0};
 873         struct rb_node **insert_p = NULL, *insert_parent = NULL;
 874         block_t end = lstart + len;
 875
 876         mutex_lock(&dcc->cmd_lock);
 877
 878         dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
 879                                         NULL, lstart,
 880                                         (struct rb_entry **)&prev_dc,
 881                                         (struct rb_entry **)&next_dc,
 882                                         &insert_p, &insert_parent, true);
 883         if (dc)
 884                 prev_dc = dc;
 885
 886         if (!prev_dc) {
 887                 di.lstart = lstart;
 888                 di.len = next_dc ? next_dc->lstart - lstart : len;
 889                 di.len = min(di.len, len);
 890                 di.start = start;
 891         }
 892
 893         while (1) {
 894                 struct rb_node *node;
 895                 bool merged = false;
 896                 struct discard_cmd *tdc = NULL;
 897
 898                 if (prev_dc) {
 899                         di.lstart = prev_dc->lstart + prev_dc->len;
 900                         if (di.lstart < lstart)
 901                                 di.lstart = lstart;
 902                         if (di.lstart >= end)
 903                                 break;
 904
 905                         if (!next_dc || next_dc->lstart > end)
 906                                 di.len = end - di.lstart;
 907                         else
 908                                 di.len = next_dc->lstart - di.lstart;
 909                         di.start = start + di.lstart - lstart;
 910                 }
 911
 912                 if (!di.len)
 913                         goto next;
 914
 915                 if (prev_dc && prev_dc->state == D_PREP &&
 916                         prev_dc->bdev == bdev &&
 917                         __is_discard_back_mergeable(&di, &prev_dc->di)) {
 918                         prev_dc->di.len += di.len;
 919                         dcc->undiscard_blks += di.len;
 920                         __relocate_discard_cmd(dcc, prev_dc);
 921                         f2fs_bug_on(sbi,
 922                                 !__check_rb_tree_consistence(sbi, &dcc->root));
 923                         di = prev_dc->di;
 924                         tdc = prev_dc;
 925                         merged = true;
 926                 }
 927
 928                 if (next_dc && next_dc->state == D_PREP &&
 929                         next_dc->bdev == bdev &&
 930                         __is_discard_front_mergeable(&di, &next_dc->di)) {
 931                         next_dc->di.lstart = di.lstart;
 932                         next_dc->di.len += di.len;
 933                         next_dc->di.start = di.start;
 934                         dcc->undiscard_blks += di.len;
 935                         __relocate_discard_cmd(dcc, next_dc);
 936                         if (tdc)
 937                                 __remove_discard_cmd(sbi, tdc);
 938                         f2fs_bug_on(sbi,
 939                                 !__check_rb_tree_consistence(sbi, &dcc->root));
 940                         merged = true;
 941                 }
 942
 943                 if (!merged) {
 944                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
 945                                                         di.len, NULL, NULL);
 946                         f2fs_bug_on(sbi,
 947                                 !__check_rb_tree_consistence(sbi, &dcc->root));
 948                 }
 949  next:
 950                 prev_dc = next_dc;
 951                 if (!prev_dc)
 952                         break;
 953
 954                 node = rb_next(&prev_dc->rb_node);
 955                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
 956         }
 957
 958         mutex_unlock(&dcc->cmd_lock);
 959 }
 960
 961 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
 962                 struct block_device *bdev, block_t blkstart, block_t blklen)
 963 {
 964         block_t lblkstart = blkstart;
 965
 966         trace_f2fs_queue_discard(bdev, blkstart, blklen);
 967
 968         if (sbi->s_ndevs) {
 969                 int devi = f2fs_target_device_index(sbi, blkstart);
 970
 971                 blkstart -= FDEV(devi).start_blk;
 972         }
 973         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
 974         return 0;
 975 }
 976
 977 static void __issue_discard_cmd(struct f2fs_sb_info *sbi, bool issue_cond)
 978 {
 979         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 980         struct list_head *pend_list;
 981         struct discard_cmd *dc, *tmp;
 982         struct blk_plug plug;
 983         int i, iter = 0;
 984
 985         mutex_lock(&dcc->cmd_lock);
 986         blk_start_plug(&plug);
 987         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
 988                 pend_list = &dcc->pend_list[i];
 989                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
 990                         f2fs_bug_on(sbi, dc->state != D_PREP);
 991
 992                         if (!issue_cond || is_idle(sbi))
 993                                 __submit_discard_cmd(sbi, dc);
 994                         if (issue_cond && iter++ > DISCARD_ISSUE_RATE)
 995                                 goto out;
 996                 }
 997         }
 998 out:
 999         blk_finish_plug(&plug);
1000         mutex_unlock(&dcc->cmd_lock);
1001 }
1002
1003 static void __wait_discard_cmd(struct f2fs_sb_info *sbi, bool wait_cond)
1004 {
1005         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1006         struct list_head *wait_list = &(dcc->wait_list);
1007         struct discard_cmd *dc, *tmp;
1008
1009         mutex_lock(&dcc->cmd_lock);
1010         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1011                 if (!wait_cond || dc->state == D_DONE) {
1012                         if (dc->ref)
1013                                 continue;
1014                         wait_for_completion_io(&dc->wait);
1015                         __remove_discard_cmd(sbi, dc);
1016                 }
1017         }
1018         mutex_unlock(&dcc->cmd_lock);
1019 }
1020
1021 /* This should be covered by global mutex, &sit_i->sentry_lock */
1022 void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1023 {
1024         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1025         struct discard_cmd *dc;
1026         bool need_wait = false;
1027
1028         mutex_lock(&dcc->cmd_lock);
1029         dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1030         if (dc) {
1031                 if (dc->state == D_PREP) {
1032                         __punch_discard_cmd(sbi, dc, blkaddr);
1033                 } else {
1034                         dc->ref++;
1035                         need_wait = true;
1036                 }
1037         }
1038         mutex_unlock(&dcc->cmd_lock);
1039
1040         if (need_wait) {
1041                 wait_for_completion_io(&dc->wait);
1042                 mutex_lock(&dcc->cmd_lock);
1043                 f2fs_bug_on(sbi, dc->state != D_DONE);
1044                 dc->ref--;
1045                 if (!dc->ref)
1046                         __remove_discard_cmd(sbi, dc);
1047                 mutex_unlock(&dcc->cmd_lock);
1048         }
1049 }
1050
1051 /* This comes from f2fs_put_super */
1052 void f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1053 {
1054         __issue_discard_cmd(sbi, false);
1055         __wait_discard_cmd(sbi, false);
1056 }
1057
1058 static int issue_discard_thread(void *data)
1059 {
1060         struct f2fs_sb_info *sbi = data;
1061         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1062         wait_queue_head_t *q = &dcc->discard_wait_queue;
1063 repeat:
1064         if (kthread_should_stop())
1065                 return 0;
1066
1067         __issue_discard_cmd(sbi, true);
1068         __wait_discard_cmd(sbi, true);
1069
1070         congestion_wait(BLK_RW_SYNC, HZ/50);
1071
1072         wait_event_interruptible(*q, kthread_should_stop() ||
1073                                 atomic_read(&dcc->discard_cmd_cnt));
1074         goto repeat;
1075 }
1076
1077 #ifdef CONFIG_BLK_DEV_ZONED
1078 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1079                 struct block_device *bdev, block_t blkstart, block_t blklen)
1080 {
1081         sector_t sector, nr_sects;
1082         block_t lblkstart = blkstart;
1083         int devi = 0;
1084
1085         if (sbi->s_ndevs) {
1086                 devi = f2fs_target_device_index(sbi, blkstart);
1087                 blkstart -= FDEV(devi).start_blk;
1088         }
1089
1090         /*
1091          * We need to know the type of the zone: for conventional zones,
1092          * use regular discard if the drive supports it. For sequential
1093          * zones, reset the zone write pointer.
1094          */
1095         switch (get_blkz_type(sbi, bdev, blkstart)) {
1096
1097         case BLK_ZONE_TYPE_CONVENTIONAL:
1098                 if (!blk_queue_discard(bdev_get_queue(bdev)))
1099                         return 0;
1100                 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1101         case BLK_ZONE_TYPE_SEQWRITE_REQ:
1102         case BLK_ZONE_TYPE_SEQWRITE_PREF:
1103                 sector = SECTOR_FROM_BLOCK(blkstart);
1104                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1105
1106                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1107                                 nr_sects != bdev_zone_sectors(bdev)) {
1108                         f2fs_msg(sbi->sb, KERN_INFO,
1109                                 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1110                                 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1111                                 blkstart, blklen);
1112                         return -EIO;
1113                 }
1114                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1115                 return blkdev_reset_zones(bdev, sector,
1116                                           nr_sects, GFP_NOFS);
1117         default:
1118                 /* Unknown zone type: broken device ? */
1119                 return -EIO;
1120         }
1121 }
1122 #endif
1123
1124 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1125                 struct block_device *bdev, block_t blkstart, block_t blklen)
1126 {
1127 #ifdef CONFIG_BLK_DEV_ZONED
1128         if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
1129                                 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1130                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1131 #endif
1132         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1133 }
1134
1135 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1136                                 block_t blkstart, block_t blklen)
1137 {
1138         sector_t start = blkstart, len = 0;
1139         struct block_device *bdev;
1140         struct seg_entry *se;
1141         unsigned int offset;
1142         block_t i;
1143         int err = 0;
1144
1145         bdev = f2fs_target_device(sbi, blkstart, NULL);
1146
1147         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1148                 if (i != start) {
1149                         struct block_device *bdev2 =
1150                                 f2fs_target_device(sbi, i, NULL);
1151
1152                         if (bdev2 != bdev) {
1153                                 err = __issue_discard_async(sbi, bdev,
1154                                                 start, len);
1155                                 if (err)
1156                                         return err;
1157                                 bdev = bdev2;
1158                                 start = i;
1159                                 len = 0;
1160                         }
1161                 }
1162
1163                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1164                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1165
1166                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1167                         sbi->discard_blks--;
1168         }
1169
1170         if (len)
1171                 err = __issue_discard_async(sbi, bdev, start, len);
1172         return err;
1173 }
1174
1175 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1176                                                         bool check_only)
1177 {
1178         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1179         int max_blocks = sbi->blocks_per_seg;
1180         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1181         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1182         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1183         unsigned long *discard_map = (unsigned long *)se->discard_map;
1184         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1185         unsigned int start = 0, end = -1;
1186         bool force = (cpc->reason & CP_DISCARD);
1187         struct discard_entry *de = NULL;
1188         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1189         int i;
1190
1191         if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
1192                 return false;
1193
1194         if (!force) {
1195                 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
1196                         SM_I(sbi)->dcc_info->nr_discards >=
1197                                 SM_I(sbi)->dcc_info->max_discards)
1198                         return false;
1199         }
1200
1201         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1202         for (i = 0; i < entries; i++)
1203                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1204                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1205
1206         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1207                                 SM_I(sbi)->dcc_info->max_discards) {
1208                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1209                 if (start >= max_blocks)
1210                         break;
1211
1212                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1213                 if (force && start && end != max_blocks
1214                                         && (end - start) < cpc->trim_minlen)
1215                         continue;
1216
1217                 if (check_only)
1218                         return true;
1219
1220                 if (!de) {
1221                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1222                                                                 GFP_F2FS_ZERO);
1223                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1224                         list_add_tail(&de->list, head);
1225                 }
1226
1227                 for (i = start; i < end; i++)
1228                         __set_bit_le(i, (void *)de->discard_map);
1229
1230                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1231         }
1232         return false;
1233 }
1234
1235 void release_discard_addrs(struct f2fs_sb_info *sbi)
1236 {
1237         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1238         struct discard_entry *entry, *this;
1239
1240         /* drop caches */
1241         list_for_each_entry_safe(entry, this, head, list) {
1242                 list_del(&entry->list);
1243                 kmem_cache_free(discard_entry_slab, entry);
1244         }
1245 }
1246
1247 /*
1248  * Should call clear_prefree_segments after checkpoint is done.
1249  */
1250 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1251 {
1252         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1253         unsigned int segno;
1254
1255         mutex_lock(&dirty_i->seglist_lock);
1256         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1257                 __set_test_and_free(sbi, segno);
1258         mutex_unlock(&dirty_i->seglist_lock);
1259 }
1260
1261 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1262 {
1263         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1264         struct discard_entry *entry, *this;
1265         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1266         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1267         unsigned int start = 0, end = -1;
1268         unsigned int secno, start_segno;
1269         bool force = (cpc->reason & CP_DISCARD);
1270
1271         mutex_lock(&dirty_i->seglist_lock);
1272
1273         while (1) {
1274                 int i;
1275                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1276                 if (start >= MAIN_SEGS(sbi))
1277                         break;
1278                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1279                                                                 start + 1);
1280
1281                 for (i = start; i < end; i++)
1282                         clear_bit(i, prefree_map);
1283
1284                 dirty_i->nr_dirty[PRE] -= end - start;
1285
1286                 if (!test_opt(sbi, DISCARD))
1287                         continue;
1288
1289                 if (force && start >= cpc->trim_start &&
1290                                         (end - 1) <= cpc->trim_end)
1291                                 continue;
1292
1293                 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1294                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1295                                 (end - start) << sbi->log_blocks_per_seg);
1296                         continue;
1297                 }
1298 next:
1299                 secno = GET_SEC_FROM_SEG(sbi, start);
1300                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1301                 if (!IS_CURSEC(sbi, secno) &&
1302                         !get_valid_blocks(sbi, start, true))
1303                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1304                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1305
1306                 start = start_segno + sbi->segs_per_sec;
1307                 if (start < end)
1308                         goto next;
1309                 else
1310                         end = start - 1;
1311         }
1312         mutex_unlock(&dirty_i->seglist_lock);
1313
1314         /* send small discards */
1315         list_for_each_entry_safe(entry, this, head, list) {
1316                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1317                 bool is_valid = test_bit_le(0, entry->discard_map);
1318
1319 find_next:
1320                 if (is_valid) {
1321                         next_pos = find_next_zero_bit_le(entry->discard_map,
1322                                         sbi->blocks_per_seg, cur_pos);
1323                         len = next_pos - cur_pos;
1324
1325                         if (force && len < cpc->trim_minlen)
1326                                 goto skip;
1327
1328                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1329                                                                         len);
1330                         cpc->trimmed += len;
1331                         total_len += len;
1332                 } else {
1333                         next_pos = find_next_bit_le(entry->discard_map,
1334                                         sbi->blocks_per_seg, cur_pos);
1335                 }
1336 skip:
1337                 cur_pos = next_pos;
1338                 is_valid = !is_valid;
1339
1340                 if (cur_pos < sbi->blocks_per_seg)
1341                         goto find_next;
1342
1343                 list_del(&entry->list);
1344                 SM_I(sbi)->dcc_info->nr_discards -= total_len;
1345                 kmem_cache_free(discard_entry_slab, entry);
1346         }
1347
1348         wake_up(&SM_I(sbi)->dcc_info->discard_wait_queue);
1349 }
1350
1351 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1352 {
1353         dev_t dev = sbi->sb->s_bdev->bd_dev;
1354         struct discard_cmd_control *dcc;
1355         int err = 0, i;
1356
1357         if (SM_I(sbi)->dcc_info) {
1358                 dcc = SM_I(sbi)->dcc_info;
1359                 goto init_thread;
1360         }
1361
1362         dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1363         if (!dcc)
1364                 return -ENOMEM;
1365
1366         INIT_LIST_HEAD(&dcc->entry_list);
1367         for (i = 0; i < MAX_PLIST_NUM; i++)
1368                 INIT_LIST_HEAD(&dcc->pend_list[i]);
1369         INIT_LIST_HEAD(&dcc->wait_list);
1370         mutex_init(&dcc->cmd_lock);
1371         atomic_set(&dcc->issued_discard, 0);
1372         atomic_set(&dcc->issing_discard, 0);
1373         atomic_set(&dcc->discard_cmd_cnt, 0);
1374         dcc->nr_discards = 0;
1375         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1376         dcc->undiscard_blks = 0;
1377         dcc->root = RB_ROOT;
1378
1379         init_waitqueue_head(&dcc->discard_wait_queue);
1380         SM_I(sbi)->dcc_info = dcc;
1381 init_thread:
1382         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1383                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1384         if (IS_ERR(dcc->f2fs_issue_discard)) {
1385                 err = PTR_ERR(dcc->f2fs_issue_discard);
1386                 kfree(dcc);
1387                 SM_I(sbi)->dcc_info = NULL;
1388                 return err;
1389         }
1390
1391         return err;
1392 }
1393
1394 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1395 {
1396         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1397
1398         if (!dcc)
1399                 return;
1400
1401         if (dcc->f2fs_issue_discard) {
1402                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1403
1404                 dcc->f2fs_issue_discard = NULL;
1405                 kthread_stop(discard_thread);
1406         }
1407
1408         kfree(dcc);
1409         SM_I(sbi)->dcc_info = NULL;
1410 }
1411
1412 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1413 {
1414         struct sit_info *sit_i = SIT_I(sbi);
1415
1416         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1417                 sit_i->dirty_sentries++;
1418                 return false;
1419         }
1420
1421         return true;
1422 }
1423
1424 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1425                                         unsigned int segno, int modified)
1426 {
1427         struct seg_entry *se = get_seg_entry(sbi, segno);
1428         se->type = type;
1429         if (modified)
1430                 __mark_sit_entry_dirty(sbi, segno);
1431 }
1432
1433 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1434 {
1435         struct seg_entry *se;
1436         unsigned int segno, offset;
1437         long int new_vblocks;
1438
1439         segno = GET_SEGNO(sbi, blkaddr);
1440
1441         se = get_seg_entry(sbi, segno);
1442         new_vblocks = se->valid_blocks + del;
1443         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1444
1445         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
1446                                 (new_vblocks > sbi->blocks_per_seg)));
1447
1448         se->valid_blocks = new_vblocks;
1449         se->mtime = get_mtime(sbi);
1450         SIT_I(sbi)->max_mtime = se->mtime;
1451
1452         /* Update valid block bitmap */
1453         if (del > 0) {
1454                 if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) {
1455 #ifdef CONFIG_F2FS_CHECK_FS
1456                         if (f2fs_test_and_set_bit(offset,
1457                                                 se->cur_valid_map_mir))
1458                                 f2fs_bug_on(sbi, 1);
1459                         else
1460                                 WARN_ON(1);
1461 #else
1462                         f2fs_bug_on(sbi, 1);
1463 #endif
1464                 }
1465                 if (f2fs_discard_en(sbi) &&
1466                         !f2fs_test_and_set_bit(offset, se->discard_map))
1467                         sbi->discard_blks--;
1468
1469                 /* don't overwrite by SSR to keep node chain */
1470                 if (se->type == CURSEG_WARM_NODE) {
1471                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1472                                 se->ckpt_valid_blocks++;
1473                 }
1474         } else {
1475                 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) {
1476 #ifdef CONFIG_F2FS_CHECK_FS
1477                         if (!f2fs_test_and_clear_bit(offset,
1478                                                 se->cur_valid_map_mir))
1479                                 f2fs_bug_on(sbi, 1);
1480                         else
1481                                 WARN_ON(1);
1482 #else
1483                         f2fs_bug_on(sbi, 1);
1484 #endif
1485                 }
1486                 if (f2fs_discard_en(sbi) &&
1487                         f2fs_test_and_clear_bit(offset, se->discard_map))
1488                         sbi->discard_blks++;
1489         }
1490         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1491                 se->ckpt_valid_blocks += del;
1492
1493         __mark_sit_entry_dirty(sbi, segno);
1494
1495         /* update total number of valid blocks to be written in ckpt area */
1496         SIT_I(sbi)->written_valid_blocks += del;
1497
1498         if (sbi->segs_per_sec > 1)
1499                 get_sec_entry(sbi, segno)->valid_blocks += del;
1500 }
1501
1502 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
1503 {
1504         update_sit_entry(sbi, new, 1);
1505         if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1506                 update_sit_entry(sbi, old, -1);
1507
1508         locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1509         locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
1510 }
1511
1512 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1513 {
1514         unsigned int segno = GET_SEGNO(sbi, addr);
1515         struct sit_info *sit_i = SIT_I(sbi);
1516
1517         f2fs_bug_on(sbi, addr == NULL_ADDR);
1518         if (addr == NEW_ADDR)
1519                 return;
1520
1521         /* add it into sit main buffer */
1522         mutex_lock(&sit_i->sentry_lock);
1523
1524         update_sit_entry(sbi, addr, -1);
1525
1526         /* add it into dirty seglist */
1527         locate_dirty_segment(sbi, segno);
1528
1529         mutex_unlock(&sit_i->sentry_lock);
1530 }
1531
1532 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1533 {
1534         struct sit_info *sit_i = SIT_I(sbi);
1535         unsigned int segno, offset;
1536         struct seg_entry *se;
1537         bool is_cp = false;
1538
1539         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1540                 return true;
1541
1542         mutex_lock(&sit_i->sentry_lock);
1543
1544         segno = GET_SEGNO(sbi, blkaddr);
1545         se = get_seg_entry(sbi, segno);
1546         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1547
1548         if (f2fs_test_bit(offset, se->ckpt_valid_map))
1549                 is_cp = true;
1550
1551         mutex_unlock(&sit_i->sentry_lock);
1552
1553         return is_cp;
1554 }
1555
1556 /*
1557  * This function should be resided under the curseg_mutex lock
1558  */
1559 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1560                                         struct f2fs_summary *sum)
1561 {
1562         struct curseg_info *curseg = CURSEG_I(sbi, type);
1563         void *addr = curseg->sum_blk;
1564         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1565         memcpy(addr, sum, sizeof(struct f2fs_summary));
1566 }
1567
1568 /*
1569  * Calculate the number of current summary pages for writing
1570  */
1571 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1572 {
1573         int valid_sum_count = 0;
1574         int i, sum_in_page;
1575
1576         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1577                 if (sbi->ckpt->alloc_type[i] == SSR)
1578                         valid_sum_count += sbi->blocks_per_seg;
1579                 else {
1580                         if (for_ra)
1581                                 valid_sum_count += le16_to_cpu(
1582                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1583                         else
1584                                 valid_sum_count += curseg_blkoff(sbi, i);
1585                 }
1586         }
1587
1588         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1589                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1590         if (valid_sum_count <= sum_in_page)
1591                 return 1;
1592         else if ((valid_sum_count - sum_in_page) <=
1593                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1594                 return 2;
1595         return 3;
1596 }
1597
1598 /*
1599  * Caller should put this summary page
1600  */
1601 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1602 {
1603         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1604 }
1605
1606 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1607 {
1608         struct page *page = grab_meta_page(sbi, blk_addr);
1609         void *dst = page_address(page);
1610
1611         if (src)
1612                 memcpy(dst, src, PAGE_SIZE);
1613         else
1614                 memset(dst, 0, PAGE_SIZE);
1615         set_page_dirty(page);
1616         f2fs_put_page(page, 1);
1617 }
1618
1619 static void write_sum_page(struct f2fs_sb_info *sbi,
1620                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
1621 {
1622         update_meta_page(sbi, (void *)sum_blk, blk_addr);
1623 }
1624
1625 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1626                                                 int type, block_t blk_addr)
1627 {
1628         struct curseg_info *curseg = CURSEG_I(sbi, type);
1629         struct page *page = grab_meta_page(sbi, blk_addr);
1630         struct f2fs_summary_block *src = curseg->sum_blk;
1631         struct f2fs_summary_block *dst;
1632
1633         dst = (struct f2fs_summary_block *)page_address(page);
1634
1635         mutex_lock(&curseg->curseg_mutex);
1636
1637         down_read(&curseg->journal_rwsem);
1638         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1639         up_read(&curseg->journal_rwsem);
1640
1641         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1642         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1643
1644         mutex_unlock(&curseg->curseg_mutex);
1645
1646         set_page_dirty(page);
1647         f2fs_put_page(page, 1);
1648 }
1649
1650 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1651 {
1652         struct curseg_info *curseg = CURSEG_I(sbi, type);
1653         unsigned int segno = curseg->segno + 1;
1654         struct free_segmap_info *free_i = FREE_I(sbi);
1655
1656         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
1657                 return !test_bit(segno, free_i->free_segmap);
1658         return 0;
1659 }
1660
1661 /*
1662  * Find a new segment from the free segments bitmap to right order
1663  * This function should be returned with success, otherwise BUG
1664  */
1665 static void get_new_segment(struct f2fs_sb_info *sbi,
1666                         unsigned int *newseg, bool new_sec, int dir)
1667 {
1668         struct free_segmap_info *free_i = FREE_I(sbi);
1669         unsigned int segno, secno, zoneno;
1670         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
1671         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
1672         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
1673         unsigned int left_start = hint;
1674         bool init = true;
1675         int go_left = 0;
1676         int i;
1677
1678         spin_lock(&free_i->segmap_lock);
1679
1680         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1681                 segno = find_next_zero_bit(free_i->free_segmap,
1682                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
1683                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
1684                         goto got_it;
1685         }
1686 find_other_zone:
1687         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1688         if (secno >= MAIN_SECS(sbi)) {
1689                 if (dir == ALLOC_RIGHT) {
1690                         secno = find_next_zero_bit(free_i->free_secmap,
1691                                                         MAIN_SECS(sbi), 0);
1692                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1693                 } else {
1694                         go_left = 1;
1695                         left_start = hint - 1;
1696                 }
1697         }
1698         if (go_left == 0)
1699                 goto skip_left;
1700
1701         while (test_bit(left_start, free_i->free_secmap)) {
1702                 if (left_start > 0) {
1703                         left_start--;
1704                         continue;
1705                 }
1706                 left_start = find_next_zero_bit(free_i->free_secmap,
1707                                                         MAIN_SECS(sbi), 0);
1708                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1709                 break;
1710         }
1711         secno = left_start;
1712 skip_left:
1713         hint = secno;
1714         segno = GET_SEG_FROM_SEC(sbi, secno);
1715         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
1716
1717         /* give up on finding another zone */
1718         if (!init)
1719                 goto got_it;
1720         if (sbi->secs_per_zone == 1)
1721                 goto got_it;
1722         if (zoneno == old_zoneno)
1723                 goto got_it;
1724         if (dir == ALLOC_LEFT) {
1725                 if (!go_left && zoneno + 1 >= total_zones)
1726                         goto got_it;
1727                 if (go_left && zoneno == 0)
1728                         goto got_it;
1729         }
1730         for (i = 0; i < NR_CURSEG_TYPE; i++)
1731                 if (CURSEG_I(sbi, i)->zone == zoneno)
1732                         break;
1733
1734         if (i < NR_CURSEG_TYPE) {
1735                 /* zone is in user, try another */
1736                 if (go_left)
1737                         hint = zoneno * sbi->secs_per_zone - 1;
1738                 else if (zoneno + 1 >= total_zones)
1739                         hint = 0;
1740                 else
1741                         hint = (zoneno + 1) * sbi->secs_per_zone;
1742                 init = false;
1743                 goto find_other_zone;
1744         }
1745 got_it:
1746         /* set it as dirty segment in free segmap */
1747         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1748         __set_inuse(sbi, segno);
1749         *newseg = segno;
1750         spin_unlock(&free_i->segmap_lock);
1751 }
1752
1753 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1754 {
1755         struct curseg_info *curseg = CURSEG_I(sbi, type);
1756         struct summary_footer *sum_footer;
1757
1758         curseg->segno = curseg->next_segno;
1759         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
1760         curseg->next_blkoff = 0;
1761         curseg->next_segno = NULL_SEGNO;
1762
1763         sum_footer = &(curseg->sum_blk->footer);
1764         memset(sum_footer, 0, sizeof(struct summary_footer));
1765         if (IS_DATASEG(type))
1766                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1767         if (IS_NODESEG(type))
1768                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1769         __set_sit_entry_type(sbi, type, curseg->segno, modified);
1770 }
1771
1772 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
1773 {
1774         /* if segs_per_sec is large than 1, we need to keep original policy. */
1775         if (sbi->segs_per_sec != 1)
1776                 return CURSEG_I(sbi, type)->segno;
1777
1778         if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
1779                 return 0;
1780
1781         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
1782                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
1783         return CURSEG_I(sbi, type)->segno;
1784 }
1785
1786 /*
1787  * Allocate a current working segment.
1788  * This function always allocates a free segment in LFS manner.
1789  */
1790 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1791 {
1792         struct curseg_info *curseg = CURSEG_I(sbi, type);
1793         unsigned int segno = curseg->segno;
1794         int dir = ALLOC_LEFT;
1795
1796         write_sum_page(sbi, curseg->sum_blk,
1797                                 GET_SUM_BLOCK(sbi, segno));
1798         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1799                 dir = ALLOC_RIGHT;
1800
1801         if (test_opt(sbi, NOHEAP))
1802                 dir = ALLOC_RIGHT;
1803
1804         segno = __get_next_segno(sbi, type);
1805         get_new_segment(sbi, &segno, new_sec, dir);
1806         curseg->next_segno = segno;
1807         reset_curseg(sbi, type, 1);
1808         curseg->alloc_type = LFS;
1809 }
1810
1811 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1812                         struct curseg_info *seg, block_t start)
1813 {
1814         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1815         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1816         unsigned long *target_map = SIT_I(sbi)->tmp_map;
1817         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1818         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1819         int i, pos;
1820
1821         for (i = 0; i < entries; i++)
1822                 target_map[i] = ckpt_map[i] | cur_map[i];
1823
1824         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1825
1826         seg->next_blkoff = pos;
1827 }
1828
1829 /*
1830  * If a segment is written by LFS manner, next block offset is just obtained
1831  * by increasing the current block offset. However, if a segment is written by
1832  * SSR manner, next block offset obtained by calling __next_free_blkoff
1833  */
1834 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1835                                 struct curseg_info *seg)
1836 {
1837         if (seg->alloc_type == SSR)
1838                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1839         else
1840                 seg->next_blkoff++;
1841 }
1842
1843 /*
1844  * This function always allocates a used segment(from dirty seglist) by SSR
1845  * manner, so it should recover the existing segment information of valid blocks
1846  */
1847 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1848 {
1849         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1850         struct curseg_info *curseg = CURSEG_I(sbi, type);
1851         unsigned int new_segno = curseg->next_segno;
1852         struct f2fs_summary_block *sum_node;
1853         struct page *sum_page;
1854
1855         write_sum_page(sbi, curseg->sum_blk,
1856                                 GET_SUM_BLOCK(sbi, curseg->segno));
1857         __set_test_and_inuse(sbi, new_segno);
1858
1859         mutex_lock(&dirty_i->seglist_lock);
1860         __remove_dirty_segment(sbi, new_segno, PRE);
1861         __remove_dirty_segment(sbi, new_segno, DIRTY);
1862         mutex_unlock(&dirty_i->seglist_lock);
1863
1864         reset_curseg(sbi, type, 1);
1865         curseg->alloc_type = SSR;
1866         __next_free_blkoff(sbi, curseg, 0);
1867
1868         if (reuse) {
1869                 sum_page = get_sum_page(sbi, new_segno);
1870                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1871                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1872                 f2fs_put_page(sum_page, 1);
1873         }
1874 }
1875
1876 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1877 {
1878         struct curseg_info *curseg = CURSEG_I(sbi, type);
1879         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1880         unsigned segno = NULL_SEGNO;
1881         int i, cnt;
1882         bool reversed = false;
1883
1884         /* need_SSR() already forces to do this */
1885         if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
1886                 curseg->next_segno = segno;
1887                 return 1;
1888         }
1889
1890         /* For node segments, let's do SSR more intensively */
1891         if (IS_NODESEG(type)) {
1892                 if (type >= CURSEG_WARM_NODE) {
1893                         reversed = true;
1894                         i = CURSEG_COLD_NODE;
1895                 } else {
1896                         i = CURSEG_HOT_NODE;
1897                 }
1898                 cnt = NR_CURSEG_NODE_TYPE;
1899         } else {
1900                 if (type >= CURSEG_WARM_DATA) {
1901                         reversed = true;
1902                         i = CURSEG_COLD_DATA;
1903                 } else {
1904                         i = CURSEG_HOT_DATA;
1905                 }
1906                 cnt = NR_CURSEG_DATA_TYPE;
1907         }
1908
1909         for (; cnt-- > 0; reversed ? i-- : i++) {
1910                 if (i == type)
1911                         continue;
1912                 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
1913                         curseg->next_segno = segno;
1914                         return 1;
1915                 }
1916         }
1917         return 0;
1918 }
1919
1920 /*
1921  * flush out current segment and replace it with new segment
1922  * This function should be returned with success, otherwise BUG
1923  */
1924 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1925                                                 int type, bool force)
1926 {
1927         struct curseg_info *curseg = CURSEG_I(sbi, type);
1928
1929         if (force)
1930                 new_curseg(sbi, type, true);
1931         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
1932                                         type == CURSEG_WARM_NODE)
1933                 new_curseg(sbi, type, false);
1934         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1935                 new_curseg(sbi, type, false);
1936         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1937                 change_curseg(sbi, type, true);
1938         else
1939                 new_curseg(sbi, type, false);
1940
1941         stat_inc_seg_type(sbi, curseg);
1942 }
1943
1944 void allocate_new_segments(struct f2fs_sb_info *sbi)
1945 {
1946         struct curseg_info *curseg;
1947         unsigned int old_segno;
1948         int i;
1949
1950         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1951                 curseg = CURSEG_I(sbi, i);
1952                 old_segno = curseg->segno;
1953                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
1954                 locate_dirty_segment(sbi, old_segno);
1955         }
1956 }
1957
1958 static const struct segment_allocation default_salloc_ops = {
1959         .allocate_segment = allocate_segment_by_default,
1960 };
1961
1962 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1963 {
1964         __u64 trim_start = cpc->trim_start;
1965         bool has_candidate = false;
1966
1967         mutex_lock(&SIT_I(sbi)->sentry_lock);
1968         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
1969                 if (add_discard_addrs(sbi, cpc, true)) {
1970                         has_candidate = true;
1971                         break;
1972                 }
1973         }
1974         mutex_unlock(&SIT_I(sbi)->sentry_lock);
1975
1976         cpc->trim_start = trim_start;
1977         return has_candidate;
1978 }
1979
1980 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1981 {
1982         __u64 start = F2FS_BYTES_TO_BLK(range->start);
1983         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1984         unsigned int start_segno, end_segno;
1985         struct cp_control cpc;
1986         int err = 0;
1987
1988         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1989                 return -EINVAL;
1990
1991         cpc.trimmed = 0;
1992         if (end <= MAIN_BLKADDR(sbi))
1993                 goto out;
1994
1995         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1996                 f2fs_msg(sbi->sb, KERN_WARNING,
1997                         "Found FS corruption, run fsck to fix.");
1998                 goto out;
1999         }
2000
2001         /* start/end segment number in main_area */
2002         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2003         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2004                                                 GET_SEGNO(sbi, end);
2005         cpc.reason = CP_DISCARD;
2006         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2007
2008         /* do checkpoint to issue discard commands safely */
2009         for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
2010                 cpc.trim_start = start_segno;
2011
2012                 if (sbi->discard_blks == 0)
2013                         break;
2014                 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
2015                         cpc.trim_end = end_segno;
2016                 else
2017                         cpc.trim_end = min_t(unsigned int,
2018                                 rounddown(start_segno +
2019                                 BATCHED_TRIM_SEGMENTS(sbi),
2020                                 sbi->segs_per_sec) - 1, end_segno);
2021
2022                 mutex_lock(&sbi->gc_mutex);
2023                 err = write_checkpoint(sbi, &cpc);
2024                 mutex_unlock(&sbi->gc_mutex);
2025                 if (err)
2026                         break;
2027
2028                 schedule();
2029         }
2030 out:
2031         range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
2032         return err;
2033 }
2034
2035 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2036 {
2037         struct curseg_info *curseg = CURSEG_I(sbi, type);
2038         if (curseg->next_blkoff < sbi->blocks_per_seg)
2039                 return true;
2040         return false;
2041 }
2042
2043 static int __get_segment_type_2(struct page *page, enum page_type p_type)
2044 {
2045         if (p_type == DATA)
2046                 return CURSEG_HOT_DATA;
2047         else
2048                 return CURSEG_HOT_NODE;
2049 }
2050
2051 static int __get_segment_type_4(struct page *page, enum page_type p_type)
2052 {
2053         if (p_type == DATA) {
2054                 struct inode *inode = page->mapping->host;
2055
2056                 if (S_ISDIR(inode->i_mode))
2057                         return CURSEG_HOT_DATA;
2058                 else
2059                         return CURSEG_COLD_DATA;
2060         } else {
2061                 if (IS_DNODE(page) && is_cold_node(page))
2062                         return CURSEG_WARM_NODE;
2063                 else
2064                         return CURSEG_COLD_NODE;
2065         }
2066 }
2067
2068 static int __get_segment_type_6(struct page *page, enum page_type p_type)
2069 {
2070         if (p_type == DATA) {
2071                 struct inode *inode = page->mapping->host;
2072
2073                 if (is_cold_data(page) || file_is_cold(inode))
2074                         return CURSEG_COLD_DATA;
2075                 if (is_inode_flag_set(inode, FI_HOT_DATA))
2076                         return CURSEG_HOT_DATA;
2077                 return CURSEG_WARM_DATA;
2078         } else {
2079                 if (IS_DNODE(page))
2080                         return is_cold_node(page) ? CURSEG_WARM_NODE :
2081                                                 CURSEG_HOT_NODE;
2082                 return CURSEG_COLD_NODE;
2083         }
2084 }
2085
2086 static int __get_segment_type(struct page *page, enum page_type p_type)
2087 {
2088         switch (F2FS_P_SB(page)->active_logs) {
2089         case 2:
2090                 return __get_segment_type_2(page, p_type);
2091         case 4:
2092                 return __get_segment_type_4(page, p_type);
2093         }
2094         /* NR_CURSEG_TYPE(6) logs by default */
2095         f2fs_bug_on(F2FS_P_SB(page),
2096                 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
2097         return __get_segment_type_6(page, p_type);
2098 }
2099
2100 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2101                 block_t old_blkaddr, block_t *new_blkaddr,
2102                 struct f2fs_summary *sum, int type)
2103 {
2104         struct sit_info *sit_i = SIT_I(sbi);
2105         struct curseg_info *curseg = CURSEG_I(sbi, type);
2106
2107         mutex_lock(&curseg->curseg_mutex);
2108         mutex_lock(&sit_i->sentry_lock);
2109
2110         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2111
2112         f2fs_wait_discard_bio(sbi, *new_blkaddr);
2113
2114         /*
2115          * __add_sum_entry should be resided under the curseg_mutex
2116          * because, this function updates a summary entry in the
2117          * current summary block.
2118          */
2119         __add_sum_entry(sbi, type, sum);
2120
2121         __refresh_next_blkoff(sbi, curseg);
2122
2123         stat_inc_block_count(sbi, curseg);
2124
2125         if (!__has_curseg_space(sbi, type))
2126                 sit_i->s_ops->allocate_segment(sbi, type, false);
2127         /*
2128          * SIT information should be updated after segment allocation,
2129          * since we need to keep dirty segments precisely under SSR.
2130          */
2131         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
2132
2133         mutex_unlock(&sit_i->sentry_lock);
2134
2135         if (page && IS_NODESEG(type))
2136                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2137
2138         mutex_unlock(&curseg->curseg_mutex);
2139 }
2140
2141 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2142 {
2143         int type = __get_segment_type(fio->page, fio->type);
2144         int err;
2145
2146         if (fio->type == NODE || fio->type == DATA)
2147                 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
2148 reallocate:
2149         allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
2150                                         &fio->new_blkaddr, sum, type);
2151
2152         /* writeout dirty page into bdev */
2153         err = f2fs_submit_page_mbio(fio);
2154         if (err == -EAGAIN) {
2155                 fio->old_blkaddr = fio->new_blkaddr;
2156                 goto reallocate;
2157         }
2158
2159         if (fio->type == NODE || fio->type == DATA)
2160                 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
2161 }
2162
2163 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
2164 {
2165         struct f2fs_io_info fio = {
2166                 .sbi = sbi,
2167                 .type = META,
2168                 .op = REQ_OP_WRITE,
2169                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
2170                 .old_blkaddr = page->index,
2171                 .new_blkaddr = page->index,
2172                 .page = page,
2173                 .encrypted_page = NULL,
2174         };
2175
2176         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
2177                 fio.op_flags &= ~REQ_META;
2178
2179         set_page_writeback(page);
2180         f2fs_submit_page_mbio(&fio);
2181 }
2182
2183 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
2184 {
2185         struct f2fs_summary sum;
2186
2187         set_summary(&sum, nid, 0, 0);
2188         do_write_page(&sum, fio);
2189 }
2190
2191 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
2192 {
2193         struct f2fs_sb_info *sbi = fio->sbi;
2194         struct f2fs_summary sum;
2195         struct node_info ni;
2196
2197         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
2198         get_node_info(sbi, dn->nid, &ni);
2199         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
2200         do_write_page(&sum, fio);
2201         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
2202 }
2203
2204 int rewrite_data_page(struct f2fs_io_info *fio)
2205 {
2206         fio->new_blkaddr = fio->old_blkaddr;
2207         stat_inc_inplace_blocks(fio->sbi);
2208         return f2fs_submit_page_bio(fio);
2209 }
2210
2211 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
2212                                 block_t old_blkaddr, block_t new_blkaddr,
2213                                 bool recover_curseg, bool recover_newaddr)
2214 {
2215         struct sit_info *sit_i = SIT_I(sbi);
2216         struct curseg_info *curseg;
2217         unsigned int segno, old_cursegno;
2218         struct seg_entry *se;
2219         int type;
2220         unsigned short old_blkoff;
2221
2222         segno = GET_SEGNO(sbi, new_blkaddr);
2223         se = get_seg_entry(sbi, segno);
2224         type = se->type;
2225
2226         if (!recover_curseg) {
2227                 /* for recovery flow */
2228                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2229                         if (old_blkaddr == NULL_ADDR)
2230                                 type = CURSEG_COLD_DATA;
2231                         else
2232                                 type = CURSEG_WARM_DATA;
2233                 }
2234         } else {
2235                 if (!IS_CURSEG(sbi, segno))
2236                         type = CURSEG_WARM_DATA;
2237         }
2238
2239         curseg = CURSEG_I(sbi, type);
2240
2241         mutex_lock(&curseg->curseg_mutex);
2242         mutex_lock(&sit_i->sentry_lock);
2243
2244         old_cursegno = curseg->segno;
2245         old_blkoff = curseg->next_blkoff;
2246
2247         /* change the current segment */
2248         if (segno != curseg->segno) {
2249                 curseg->next_segno = segno;
2250                 change_curseg(sbi, type, true);
2251         }
2252
2253         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
2254         __add_sum_entry(sbi, type, sum);
2255
2256         if (!recover_curseg || recover_newaddr)
2257                 update_sit_entry(sbi, new_blkaddr, 1);
2258         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2259                 update_sit_entry(sbi, old_blkaddr, -1);
2260
2261         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2262         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2263
2264         locate_dirty_segment(sbi, old_cursegno);
2265
2266         if (recover_curseg) {
2267                 if (old_cursegno != curseg->segno) {
2268                         curseg->next_segno = old_cursegno;
2269                         change_curseg(sbi, type, true);
2270                 }
2271                 curseg->next_blkoff = old_blkoff;
2272         }
2273
2274         mutex_unlock(&sit_i->sentry_lock);
2275         mutex_unlock(&curseg->curseg_mutex);
2276 }
2277
2278 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2279                                 block_t old_addr, block_t new_addr,
2280                                 unsigned char version, bool recover_curseg,
2281                                 bool recover_newaddr)
2282 {
2283         struct f2fs_summary sum;
2284
2285         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2286
2287         __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2288                                         recover_curseg, recover_newaddr);
2289
2290         f2fs_update_data_blkaddr(dn, new_addr);
2291 }
2292
2293 void f2fs_wait_on_page_writeback(struct page *page,
2294                                 enum page_type type, bool ordered)
2295 {
2296         if (PageWriteback(page)) {
2297                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2298
2299                 f2fs_submit_merged_bio_cond(sbi, page->mapping->host,
2300                                                 0, page->index, type, WRITE);
2301                 if (ordered)
2302                         wait_on_page_writeback(page);
2303                 else
2304                         wait_for_stable_page(page);
2305         }
2306 }
2307
2308 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
2309                                                         block_t blkaddr)
2310 {
2311         struct page *cpage;
2312
2313         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
2314                 return;
2315
2316         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2317         if (cpage) {
2318                 f2fs_wait_on_page_writeback(cpage, DATA, true);
2319                 f2fs_put_page(cpage, 1);
2320         }
2321 }
2322
2323 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
2324 {
2325         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2326         struct curseg_info *seg_i;
2327         unsigned char *kaddr;
2328         struct page *page;
2329         block_t start;
2330         int i, j, offset;
2331
2332         start = start_sum_block(sbi);
2333
2334         page = get_meta_page(sbi, start++);
2335         kaddr = (unsigned char *)page_address(page);
2336
2337         /* Step 1: restore nat cache */
2338         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2339         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
2340
2341         /* Step 2: restore sit cache */
2342         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2343         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
2344         offset = 2 * SUM_JOURNAL_SIZE;
2345
2346         /* Step 3: restore summary entries */
2347         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2348                 unsigned short blk_off;
2349                 unsigned int segno;
2350
2351                 seg_i = CURSEG_I(sbi, i);
2352                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2353                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2354                 seg_i->next_segno = segno;
2355                 reset_curseg(sbi, i, 0);
2356                 seg_i->alloc_type = ckpt->alloc_type[i];
2357                 seg_i->next_blkoff = blk_off;
2358
2359                 if (seg_i->alloc_type == SSR)
2360                         blk_off = sbi->blocks_per_seg;
2361
2362                 for (j = 0; j < blk_off; j++) {
2363                         struct f2fs_summary *s;
2364                         s = (struct f2fs_summary *)(kaddr + offset);
2365                         seg_i->sum_blk->entries[j] = *s;
2366                         offset += SUMMARY_SIZE;
2367                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
2368                                                 SUM_FOOTER_SIZE)
2369                                 continue;
2370
2371                         f2fs_put_page(page, 1);
2372                         page = NULL;
2373
2374                         page = get_meta_page(sbi, start++);
2375                         kaddr = (unsigned char *)page_address(page);
2376                         offset = 0;
2377                 }
2378         }
2379         f2fs_put_page(page, 1);
2380         return 0;
2381 }
2382
2383 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2384 {
2385         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2386         struct f2fs_summary_block *sum;
2387         struct curseg_info *curseg;
2388         struct page *new;
2389         unsigned short blk_off;
2390         unsigned int segno = 0;
2391         block_t blk_addr = 0;
2392
2393         /* get segment number and block addr */
2394         if (IS_DATASEG(type)) {
2395                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2396                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2397                                                         CURSEG_HOT_DATA]);
2398                 if (__exist_node_summaries(sbi))
2399                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2400                 else
2401                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2402         } else {
2403                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2404                                                         CURSEG_HOT_NODE]);
2405                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2406                                                         CURSEG_HOT_NODE]);
2407                 if (__exist_node_summaries(sbi))
2408                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2409                                                         type - CURSEG_HOT_NODE);
2410                 else
2411                         blk_addr = GET_SUM_BLOCK(sbi, segno);
2412         }
2413
2414         new = get_meta_page(sbi, blk_addr);
2415         sum = (struct f2fs_summary_block *)page_address(new);
2416
2417         if (IS_NODESEG(type)) {
2418                 if (__exist_node_summaries(sbi)) {
2419                         struct f2fs_summary *ns = &sum->entries[0];
2420                         int i;
2421                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2422                                 ns->version = 0;
2423                                 ns->ofs_in_node = 0;
2424                         }
2425                 } else {
2426                         int err;
2427
2428                         err = restore_node_summary(sbi, segno, sum);
2429                         if (err) {
2430                                 f2fs_put_page(new, 1);
2431                                 return err;
2432                         }
2433                 }
2434         }
2435
2436         /* set uncompleted segment to curseg */
2437         curseg = CURSEG_I(sbi, type);
2438         mutex_lock(&curseg->curseg_mutex);
2439
2440         /* update journal info */
2441         down_write(&curseg->journal_rwsem);
2442         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2443         up_write(&curseg->journal_rwsem);
2444
2445         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2446         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
2447         curseg->next_segno = segno;
2448         reset_curseg(sbi, type, 0);
2449         curseg->alloc_type = ckpt->alloc_type[type];
2450         curseg->next_blkoff = blk_off;
2451         mutex_unlock(&curseg->curseg_mutex);
2452         f2fs_put_page(new, 1);
2453         return 0;
2454 }
2455
2456 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2457 {
2458         int type = CURSEG_HOT_DATA;
2459         int err;
2460
2461         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
2462                 int npages = npages_for_summary_flush(sbi, true);
2463
2464                 if (npages >= 2)
2465                         ra_meta_pages(sbi, start_sum_block(sbi), npages,
2466                                                         META_CP, true);
2467
2468                 /* restore for compacted data summary */
2469                 if (read_compacted_summaries(sbi))
2470                         return -EINVAL;
2471                 type = CURSEG_HOT_NODE;
2472         }
2473
2474         if (__exist_node_summaries(sbi))
2475                 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
2476                                         NR_CURSEG_TYPE - type, META_CP, true);
2477
2478         for (; type <= CURSEG_COLD_NODE; type++) {
2479                 err = read_normal_summaries(sbi, type);
2480                 if (err)
2481                         return err;
2482         }
2483
2484         return 0;
2485 }
2486
2487 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2488 {
2489         struct page *page;
2490         unsigned char *kaddr;
2491         struct f2fs_summary *summary;
2492         struct curseg_info *seg_i;
2493         int written_size = 0;
2494         int i, j;
2495
2496         page = grab_meta_page(sbi, blkaddr++);
2497         kaddr = (unsigned char *)page_address(page);
2498
2499         /* Step 1: write nat cache */
2500         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2501         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
2502         written_size += SUM_JOURNAL_SIZE;
2503
2504         /* Step 2: write sit cache */
2505         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2506         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
2507         written_size += SUM_JOURNAL_SIZE;
2508
2509         /* Step 3: write summary entries */
2510         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2511                 unsigned short blkoff;
2512                 seg_i = CURSEG_I(sbi, i);
2513                 if (sbi->ckpt->alloc_type[i] == SSR)
2514                         blkoff = sbi->blocks_per_seg;
2515                 else
2516                         blkoff = curseg_blkoff(sbi, i);
2517
2518                 for (j = 0; j < blkoff; j++) {
2519                         if (!page) {
2520                                 page = grab_meta_page(sbi, blkaddr++);
2521                                 kaddr = (unsigned char *)page_address(page);
2522                                 written_size = 0;
2523                         }
2524                         summary = (struct f2fs_summary *)(kaddr + written_size);
2525                         *summary = seg_i->sum_blk->entries[j];
2526                         written_size += SUMMARY_SIZE;
2527
2528                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
2529                                                         SUM_FOOTER_SIZE)
2530                                 continue;
2531
2532                         set_page_dirty(page);
2533                         f2fs_put_page(page, 1);
2534                         page = NULL;
2535                 }
2536         }
2537         if (page) {
2538                 set_page_dirty(page);
2539                 f2fs_put_page(page, 1);
2540         }
2541 }
2542
2543 static void write_normal_summaries(struct f2fs_sb_info *sbi,
2544                                         block_t blkaddr, int type)
2545 {
2546         int i, end;
2547         if (IS_DATASEG(type))
2548                 end = type + NR_CURSEG_DATA_TYPE;
2549         else
2550                 end = type + NR_CURSEG_NODE_TYPE;
2551
2552         for (i = type; i < end; i++)
2553                 write_current_sum_page(sbi, i, blkaddr + (i - type));
2554 }
2555
2556 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2557 {
2558         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
2559                 write_compacted_summaries(sbi, start_blk);
2560         else
2561                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2562 }
2563
2564 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2565 {
2566         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
2567 }
2568
2569 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
2570                                         unsigned int val, int alloc)
2571 {
2572         int i;
2573
2574         if (type == NAT_JOURNAL) {
2575                 for (i = 0; i < nats_in_cursum(journal); i++) {
2576                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
2577                                 return i;
2578                 }
2579                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2580                         return update_nats_in_cursum(journal, 1);
2581         } else if (type == SIT_JOURNAL) {
2582                 for (i = 0; i < sits_in_cursum(journal); i++)
2583                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
2584                                 return i;
2585                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2586                         return update_sits_in_cursum(journal, 1);
2587         }
2588         return -1;
2589 }
2590
2591 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2592                                         unsigned int segno)
2593 {
2594         return get_meta_page(sbi, current_sit_addr(sbi, segno));
2595 }
2596
2597 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2598                                         unsigned int start)
2599 {
2600         struct sit_info *sit_i = SIT_I(sbi);
2601         struct page *src_page, *dst_page;
2602         pgoff_t src_off, dst_off;
2603         void *src_addr, *dst_addr;
2604
2605         src_off = current_sit_addr(sbi, start);
2606         dst_off = next_sit_addr(sbi, src_off);
2607
2608         /* get current sit block page without lock */
2609         src_page = get_meta_page(sbi, src_off);
2610         dst_page = grab_meta_page(sbi, dst_off);
2611         f2fs_bug_on(sbi, PageDirty(src_page));
2612
2613         src_addr = page_address(src_page);
2614         dst_addr = page_address(dst_page);
2615         memcpy(dst_addr, src_addr, PAGE_SIZE);
2616
2617         set_page_dirty(dst_page);
2618         f2fs_put_page(src_page, 1);
2619
2620         set_to_next_sit(sit_i, start);
2621
2622         return dst_page;
2623 }
2624
2625 static struct sit_entry_set *grab_sit_entry_set(void)
2626 {
2627         struct sit_entry_set *ses =
2628                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
2629
2630         ses->entry_cnt = 0;
2631         INIT_LIST_HEAD(&ses->set_list);
2632         return ses;
2633 }
2634
2635 static void release_sit_entry_set(struct sit_entry_set *ses)
2636 {
2637         list_del(&ses->set_list);
2638         kmem_cache_free(sit_entry_set_slab, ses);
2639 }
2640
2641 static void adjust_sit_entry_set(struct sit_entry_set *ses,
2642                                                 struct list_head *head)
2643 {
2644         struct sit_entry_set *next = ses;
2645
2646         if (list_is_last(&ses->set_list, head))
2647                 return;
2648
2649         list_for_each_entry_continue(next, head, set_list)
2650                 if (ses->entry_cnt <= next->entry_cnt)
2651                         break;
2652
2653         list_move_tail(&ses->set_list, &next->set_list);
2654 }
2655
2656 static void add_sit_entry(unsigned int segno, struct list_head *head)
2657 {
2658         struct sit_entry_set *ses;
2659         unsigned int start_segno = START_SEGNO(segno);
2660
2661         list_for_each_entry(ses, head, set_list) {
2662                 if (ses->start_segno == start_segno) {
2663                         ses->entry_cnt++;
2664                         adjust_sit_entry_set(ses, head);
2665                         return;
2666                 }
2667         }
2668
2669         ses = grab_sit_entry_set();
2670
2671         ses->start_segno = start_segno;
2672         ses->entry_cnt++;
2673         list_add(&ses->set_list, head);
2674 }
2675
2676 static void add_sits_in_set(struct f2fs_sb_info *sbi)
2677 {
2678         struct f2fs_sm_info *sm_info = SM_I(sbi);
2679         struct list_head *set_list = &sm_info->sit_entry_set;
2680         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
2681         unsigned int segno;
2682
2683         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
2684                 add_sit_entry(segno, set_list);
2685 }
2686
2687 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
2688 {
2689         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2690         struct f2fs_journal *journal = curseg->journal;
2691         int i;
2692
2693         down_write(&curseg->journal_rwsem);
2694         for (i = 0; i < sits_in_cursum(journal); i++) {
2695                 unsigned int segno;
2696                 bool dirtied;
2697
2698                 segno = le32_to_cpu(segno_in_journal(journal, i));
2699                 dirtied = __mark_sit_entry_dirty(sbi, segno);
2700
2701                 if (!dirtied)
2702                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
2703         }
2704         update_sits_in_cursum(journal, -i);
2705         up_write(&curseg->journal_rwsem);
2706 }
2707
2708 /*
2709  * CP calls this function, which flushes SIT entries including sit_journal,
2710  * and moves prefree segs to free segs.
2711  */
2712 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2713 {
2714         struct sit_info *sit_i = SIT_I(sbi);
2715         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2716         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2717         struct f2fs_journal *journal = curseg->journal;
2718         struct sit_entry_set *ses, *tmp;
2719         struct list_head *head = &SM_I(sbi)->sit_entry_set;
2720         bool to_journal = true;
2721         struct seg_entry *se;
2722
2723         mutex_lock(&sit_i->sentry_lock);
2724
2725         if (!sit_i->dirty_sentries)
2726                 goto out;
2727
2728         /*
2729          * add and account sit entries of dirty bitmap in sit entry
2730          * set temporarily
2731          */
2732         add_sits_in_set(sbi);
2733
2734         /*
2735          * if there are no enough space in journal to store dirty sit
2736          * entries, remove all entries from journal and add and account
2737          * them in sit entry set.
2738          */
2739         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
2740                 remove_sits_in_journal(sbi);
2741
2742         /*
2743          * there are two steps to flush sit entries:
2744          * #1, flush sit entries to journal in current cold data summary block.
2745          * #2, flush sit entries to sit page.
2746          */
2747         list_for_each_entry_safe(ses, tmp, head, set_list) {
2748                 struct page *page = NULL;
2749                 struct f2fs_sit_block *raw_sit = NULL;
2750                 unsigned int start_segno = ses->start_segno;
2751                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
2752                                                 (unsigned long)MAIN_SEGS(sbi));
2753                 unsigned int segno = start_segno;
2754
2755                 if (to_journal &&
2756                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
2757                         to_journal = false;
2758
2759                 if (to_journal) {
2760                         down_write(&curseg->journal_rwsem);
2761                 } else {
2762                         page = get_next_sit_page(sbi, start_segno);
2763                         raw_sit = page_address(page);
2764                 }
2765
2766                 /* flush dirty sit entries in region of current sit set */
2767                 for_each_set_bit_from(segno, bitmap, end) {
2768                         int offset, sit_offset;
2769
2770                         se = get_seg_entry(sbi, segno);
2771
2772                         /* add discard candidates */
2773                         if (!(cpc->reason & CP_DISCARD)) {
2774                                 cpc->trim_start = segno;
2775                                 add_discard_addrs(sbi, cpc, false);
2776                         }
2777
2778                         if (to_journal) {
2779                                 offset = lookup_journal_in_cursum(journal,
2780                                                         SIT_JOURNAL, segno, 1);
2781                                 f2fs_bug_on(sbi, offset < 0);
2782                                 segno_in_journal(journal, offset) =
2783                                                         cpu_to_le32(segno);
2784                                 seg_info_to_raw_sit(se,
2785                                         &sit_in_journal(journal, offset));
2786                         } else {
2787                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2788                                 seg_info_to_raw_sit(se,
2789                                                 &raw_sit->entries[sit_offset]);
2790                         }
2791
2792                         __clear_bit(segno, bitmap);
2793                         sit_i->dirty_sentries--;
2794                         ses->entry_cnt--;
2795                 }
2796
2797                 if (to_journal)
2798                         up_write(&curseg->journal_rwsem);
2799                 else
2800                         f2fs_put_page(page, 1);
2801
2802                 f2fs_bug_on(sbi, ses->entry_cnt);
2803                 release_sit_entry_set(ses);
2804         }
2805
2806         f2fs_bug_on(sbi, !list_empty(head));
2807         f2fs_bug_on(sbi, sit_i->dirty_sentries);
2808 out:
2809         if (cpc->reason & CP_DISCARD) {
2810                 __u64 trim_start = cpc->trim_start;
2811
2812                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2813                         add_discard_addrs(sbi, cpc, false);
2814
2815                 cpc->trim_start = trim_start;
2816         }
2817         mutex_unlock(&sit_i->sentry_lock);
2818
2819         set_prefree_as_free_segments(sbi);
2820 }
2821
2822 static int build_sit_info(struct f2fs_sb_info *sbi)
2823 {
2824         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2825         struct sit_info *sit_i;
2826         unsigned int sit_segs, start;
2827         char *src_bitmap;
2828         unsigned int bitmap_size;
2829
2830         /* allocate memory for SIT information */
2831         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2832         if (!sit_i)
2833                 return -ENOMEM;
2834
2835         SM_I(sbi)->sit_info = sit_i;
2836
2837         sit_i->sentries = kvzalloc(MAIN_SEGS(sbi) *
2838                                         sizeof(struct seg_entry), GFP_KERNEL);
2839         if (!sit_i->sentries)
2840                 return -ENOMEM;
2841
2842         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2843         sit_i->dirty_sentries_bitmap = kvzalloc(bitmap_size, GFP_KERNEL);
2844         if (!sit_i->dirty_sentries_bitmap)
2845                 return -ENOMEM;
2846
2847         for (start = 0; start < MAIN_SEGS(sbi); start++) {
2848                 sit_i->sentries[start].cur_valid_map
2849                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2850                 sit_i->sentries[start].ckpt_valid_map
2851                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2852                 if (!sit_i->sentries[start].cur_valid_map ||
2853                                 !sit_i->sentries[start].ckpt_valid_map)
2854                         return -ENOMEM;
2855
2856 #ifdef CONFIG_F2FS_CHECK_FS
2857                 sit_i->sentries[start].cur_valid_map_mir
2858                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2859                 if (!sit_i->sentries[start].cur_valid_map_mir)
2860                         return -ENOMEM;
2861 #endif
2862
2863                 if (f2fs_discard_en(sbi)) {
2864                         sit_i->sentries[start].discard_map
2865                                 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2866                         if (!sit_i->sentries[start].discard_map)
2867                                 return -ENOMEM;
2868                 }
2869         }
2870
2871         sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2872         if (!sit_i->tmp_map)
2873                 return -ENOMEM;
2874
2875         if (sbi->segs_per_sec > 1) {
2876                 sit_i->sec_entries = kvzalloc(MAIN_SECS(sbi) *
2877                                         sizeof(struct sec_entry), GFP_KERNEL);
2878                 if (!sit_i->sec_entries)
2879                         return -ENOMEM;
2880         }
2881
2882         /* get information related with SIT */
2883         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2884
2885         /* setup SIT bitmap from ckeckpoint pack */
2886         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2887         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2888
2889         sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2890         if (!sit_i->sit_bitmap)
2891                 return -ENOMEM;
2892
2893 #ifdef CONFIG_F2FS_CHECK_FS
2894         sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2895         if (!sit_i->sit_bitmap_mir)
2896                 return -ENOMEM;
2897 #endif
2898
2899         /* init SIT information */
2900         sit_i->s_ops = &default_salloc_ops;
2901
2902         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2903         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2904         sit_i->written_valid_blocks = 0;
2905         sit_i->bitmap_size = bitmap_size;
2906         sit_i->dirty_sentries = 0;
2907         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2908         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2909         sit_i->mounted_time = ktime_get_real_seconds();
2910         mutex_init(&sit_i->sentry_lock);
2911         return 0;
2912 }
2913
2914 static int build_free_segmap(struct f2fs_sb_info *sbi)
2915 {
2916         struct free_segmap_info *free_i;
2917         unsigned int bitmap_size, sec_bitmap_size;
2918
2919         /* allocate memory for free segmap information */
2920         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2921         if (!free_i)
2922                 return -ENOMEM;
2923
2924         SM_I(sbi)->free_info = free_i;
2925
2926         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2927         free_i->free_segmap = kvmalloc(bitmap_size, GFP_KERNEL);
2928         if (!free_i->free_segmap)
2929                 return -ENOMEM;
2930
2931         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2932         free_i->free_secmap = kvmalloc(sec_bitmap_size, GFP_KERNEL);
2933         if (!free_i->free_secmap)
2934                 return -ENOMEM;
2935
2936         /* set all segments as dirty temporarily */
2937         memset(free_i->free_segmap, 0xff, bitmap_size);
2938         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2939
2940         /* init free segmap information */
2941         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2942         free_i->free_segments = 0;
2943         free_i->free_sections = 0;
2944         spin_lock_init(&free_i->segmap_lock);
2945         return 0;
2946 }
2947
2948 static int build_curseg(struct f2fs_sb_info *sbi)
2949 {
2950         struct curseg_info *array;
2951         int i;
2952
2953         array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2954         if (!array)
2955                 return -ENOMEM;
2956
2957         SM_I(sbi)->curseg_array = array;
2958
2959         for (i = 0; i < NR_CURSEG_TYPE; i++) {
2960                 mutex_init(&array[i].curseg_mutex);
2961                 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2962                 if (!array[i].sum_blk)
2963                         return -ENOMEM;
2964                 init_rwsem(&array[i].journal_rwsem);
2965                 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2966                                                         GFP_KERNEL);
2967                 if (!array[i].journal)
2968                         return -ENOMEM;
2969                 array[i].segno = NULL_SEGNO;
2970                 array[i].next_blkoff = 0;
2971         }
2972         return restore_curseg_summaries(sbi);
2973 }
2974
2975 static void build_sit_entries(struct f2fs_sb_info *sbi)
2976 {
2977         struct sit_info *sit_i = SIT_I(sbi);
2978         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2979         struct f2fs_journal *journal = curseg->journal;
2980         struct seg_entry *se;
2981         struct f2fs_sit_entry sit;
2982         int sit_blk_cnt = SIT_BLK_CNT(sbi);
2983         unsigned int i, start, end;
2984         unsigned int readed, start_blk = 0;
2985
2986         do {
2987                 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
2988                                                         META_SIT, true);
2989
2990                 start = start_blk * sit_i->sents_per_block;
2991                 end = (start_blk + readed) * sit_i->sents_per_block;
2992
2993                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2994                         struct f2fs_sit_block *sit_blk;
2995                         struct page *page;
2996
2997                         se = &sit_i->sentries[start];
2998                         page = get_current_sit_page(sbi, start);
2999                         sit_blk = (struct f2fs_sit_block *)page_address(page);
3000                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3001                         f2fs_put_page(page, 1);
3002
3003                         check_block_count(sbi, start, &sit);
3004                         seg_info_from_raw_sit(se, &sit);
3005
3006                         /* build discard map only one time */
3007                         if (f2fs_discard_en(sbi)) {
3008                                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3009                                         memset(se->discard_map, 0xff,
3010                                                 SIT_VBLOCK_MAP_SIZE);
3011                                 } else {
3012                                         memcpy(se->discard_map,
3013                                                 se->cur_valid_map,
3014                                                 SIT_VBLOCK_MAP_SIZE);
3015                                         sbi->discard_blks +=
3016                                                 sbi->blocks_per_seg -
3017                                                 se->valid_blocks;
3018                                 }
3019                         }
3020
3021                         if (sbi->segs_per_sec > 1)
3022                                 get_sec_entry(sbi, start)->valid_blocks +=
3023                                                         se->valid_blocks;
3024                 }
3025                 start_blk += readed;
3026         } while (start_blk < sit_blk_cnt);
3027
3028         down_read(&curseg->journal_rwsem);
3029         for (i = 0; i < sits_in_cursum(journal); i++) {
3030                 unsigned int old_valid_blocks;
3031
3032                 start = le32_to_cpu(segno_in_journal(journal, i));
3033                 se = &sit_i->sentries[start];
3034                 sit = sit_in_journal(journal, i);
3035
3036                 old_valid_blocks = se->valid_blocks;
3037
3038                 check_block_count(sbi, start, &sit);
3039                 seg_info_from_raw_sit(se, &sit);
3040
3041                 if (f2fs_discard_en(sbi)) {
3042                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3043                                 memset(se->discard_map, 0xff,
3044                                                         SIT_VBLOCK_MAP_SIZE);
3045                         } else {
3046                                 memcpy(se->discard_map, se->cur_valid_map,
3047                                                         SIT_VBLOCK_MAP_SIZE);
3048                                 sbi->discard_blks += old_valid_blocks -
3049                                                         se->valid_blocks;
3050                         }
3051                 }
3052
3053                 if (sbi->segs_per_sec > 1)
3054                         get_sec_entry(sbi, start)->valid_blocks +=
3055                                 se->valid_blocks - old_valid_blocks;
3056         }
3057         up_read(&curseg->journal_rwsem);
3058 }
3059
3060 static void init_free_segmap(struct f2fs_sb_info *sbi)
3061 {
3062         unsigned int start;
3063         int type;
3064
3065         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3066                 struct seg_entry *sentry = get_seg_entry(sbi, start);
3067                 if (!sentry->valid_blocks)
3068                         __set_free(sbi, start);
3069                 else
3070                         SIT_I(sbi)->written_valid_blocks +=
3071                                                 sentry->valid_blocks;
3072         }
3073
3074         /* set use the current segments */
3075         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3076                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3077                 __set_test_and_inuse(sbi, curseg_t->segno);
3078         }
3079 }
3080
3081 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3082 {
3083         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3084         struct free_segmap_info *free_i = FREE_I(sbi);
3085         unsigned int segno = 0, offset = 0;
3086         unsigned short valid_blocks;
3087
3088         while (1) {
3089                 /* find dirty segment based on free segmap */
3090                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3091                 if (segno >= MAIN_SEGS(sbi))
3092                         break;
3093                 offset = segno + 1;
3094                 valid_blocks = get_valid_blocks(sbi, segno, false);
3095                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
3096                         continue;
3097                 if (valid_blocks > sbi->blocks_per_seg) {
3098                         f2fs_bug_on(sbi, 1);
3099                         continue;
3100                 }
3101                 mutex_lock(&dirty_i->seglist_lock);
3102                 __locate_dirty_segment(sbi, segno, DIRTY);
3103                 mutex_unlock(&dirty_i->seglist_lock);
3104         }
3105 }
3106
3107 static int init_victim_secmap(struct f2fs_sb_info *sbi)
3108 {
3109         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3110         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3111
3112         dirty_i->victim_secmap = kvzalloc(bitmap_size, GFP_KERNEL);
3113         if (!dirty_i->victim_secmap)
3114                 return -ENOMEM;
3115         return 0;
3116 }
3117
3118 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3119 {
3120         struct dirty_seglist_info *dirty_i;
3121         unsigned int bitmap_size, i;
3122
3123         /* allocate memory for dirty segments list information */
3124         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
3125         if (!dirty_i)
3126                 return -ENOMEM;
3127
3128         SM_I(sbi)->dirty_info = dirty_i;
3129         mutex_init(&dirty_i->seglist_lock);
3130
3131         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3132
3133         for (i = 0; i < NR_DIRTY_TYPE; i++) {
3134                 dirty_i->dirty_segmap[i] = kvzalloc(bitmap_size, GFP_KERNEL);
3135                 if (!dirty_i->dirty_segmap[i])
3136                         return -ENOMEM;
3137         }
3138
3139         init_dirty_segmap(sbi);
3140         return init_victim_secmap(sbi);
3141 }
3142
3143 /*
3144  * Update min, max modified time for cost-benefit GC algorithm
3145  */
3146 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3147 {
3148         struct sit_info *sit_i = SIT_I(sbi);
3149         unsigned int segno;
3150
3151         mutex_lock(&sit_i->sentry_lock);
3152
3153         sit_i->min_mtime = LLONG_MAX;
3154
3155         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
3156                 unsigned int i;
3157                 unsigned long long mtime = 0;
3158
3159                 for (i = 0; i < sbi->segs_per_sec; i++)
3160                         mtime += get_seg_entry(sbi, segno + i)->mtime;
3161
3162                 mtime = div_u64(mtime, sbi->segs_per_sec);
3163
3164                 if (sit_i->min_mtime > mtime)
3165                         sit_i->min_mtime = mtime;
3166         }
3167         sit_i->max_mtime = get_mtime(sbi);
3168         mutex_unlock(&sit_i->sentry_lock);
3169 }
3170
3171 int build_segment_manager(struct f2fs_sb_info *sbi)
3172 {
3173         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3174         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3175         struct f2fs_sm_info *sm_info;
3176         int err;
3177
3178         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
3179         if (!sm_info)
3180                 return -ENOMEM;
3181
3182         /* init sm info */
3183         sbi->sm_info = sm_info;
3184         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3185         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3186         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3187         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3188         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3189         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3190         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
3191         sm_info->rec_prefree_segments = sm_info->main_segments *
3192                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
3193         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3194                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3195
3196         if (!test_opt(sbi, LFS))
3197                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
3198         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
3199         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
3200         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
3201
3202         sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
3203
3204         INIT_LIST_HEAD(&sm_info->sit_entry_set);
3205
3206         if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
3207                 err = create_flush_cmd_control(sbi);
3208                 if (err)
3209                         return err;
3210         }
3211
3212         err = create_discard_cmd_control(sbi);
3213         if (err)
3214                 return err;
3215
3216         err = build_sit_info(sbi);
3217         if (err)
3218                 return err;
3219         err = build_free_segmap(sbi);
3220         if (err)
3221                 return err;
3222         err = build_curseg(sbi);
3223         if (err)
3224                 return err;
3225
3226         /* reinit free segmap based on SIT */
3227         build_sit_entries(sbi);
3228
3229         init_free_segmap(sbi);
3230         err = build_dirty_segmap(sbi);
3231         if (err)
3232                 return err;
3233
3234         init_min_max_mtime(sbi);
3235         return 0;
3236 }
3237
3238 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3239                 enum dirty_type dirty_type)
3240 {
3241         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3242
3243         mutex_lock(&dirty_i->seglist_lock);
3244         kvfree(dirty_i->dirty_segmap[dirty_type]);
3245         dirty_i->nr_dirty[dirty_type] = 0;
3246         mutex_unlock(&dirty_i->seglist_lock);
3247 }
3248
3249 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
3250 {
3251         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3252         kvfree(dirty_i->victim_secmap);
3253 }
3254
3255 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3256 {
3257         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3258         int i;
3259
3260         if (!dirty_i)
3261                 return;
3262
3263         /* discard pre-free/dirty segments list */
3264         for (i = 0; i < NR_DIRTY_TYPE; i++)
3265                 discard_dirty_segmap(sbi, i);
3266
3267         destroy_victim_secmap(sbi);
3268         SM_I(sbi)->dirty_info = NULL;
3269         kfree(dirty_i);
3270 }
3271
3272 static void destroy_curseg(struct f2fs_sb_info *sbi)
3273 {
3274         struct curseg_info *array = SM_I(sbi)->curseg_array;
3275         int i;
3276
3277         if (!array)
3278                 return;
3279         SM_I(sbi)->curseg_array = NULL;
3280         for (i = 0; i < NR_CURSEG_TYPE; i++) {
3281                 kfree(array[i].sum_blk);
3282                 kfree(array[i].journal);
3283         }
3284         kfree(array);
3285 }
3286
3287 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3288 {
3289         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3290         if (!free_i)
3291                 return;
3292         SM_I(sbi)->free_info = NULL;
3293         kvfree(free_i->free_segmap);
3294         kvfree(free_i->free_secmap);
3295         kfree(free_i);
3296 }
3297
3298 static void destroy_sit_info(struct f2fs_sb_info *sbi)
3299 {
3300         struct sit_info *sit_i = SIT_I(sbi);
3301         unsigned int start;
3302
3303         if (!sit_i)
3304                 return;
3305
3306         if (sit_i->sentries) {
3307                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3308                         kfree(sit_i->sentries[start].cur_valid_map);
3309 #ifdef CONFIG_F2FS_CHECK_FS
3310                         kfree(sit_i->sentries[start].cur_valid_map_mir);
3311 #endif
3312                         kfree(sit_i->sentries[start].ckpt_valid_map);
3313                         kfree(sit_i->sentries[start].discard_map);
3314                 }
3315         }
3316         kfree(sit_i->tmp_map);
3317
3318         kvfree(sit_i->sentries);
3319         kvfree(sit_i->sec_entries);
3320         kvfree(sit_i->dirty_sentries_bitmap);
3321
3322         SM_I(sbi)->sit_info = NULL;
3323         kfree(sit_i->sit_bitmap);
3324 #ifdef CONFIG_F2FS_CHECK_FS
3325         kfree(sit_i->sit_bitmap_mir);
3326 #endif
3327         kfree(sit_i);
3328 }
3329
3330 void destroy_segment_manager(struct f2fs_sb_info *sbi)
3331 {
3332         struct f2fs_sm_info *sm_info = SM_I(sbi);
3333
3334         if (!sm_info)
3335                 return;
3336         destroy_flush_cmd_control(sbi, true);
3337         destroy_discard_cmd_control(sbi);
3338         destroy_dirty_segmap(sbi);
3339         destroy_curseg(sbi);
3340         destroy_free_segmap(sbi);
3341         destroy_sit_info(sbi);
3342         sbi->sm_info = NULL;
3343         kfree(sm_info);
3344 }
3345
3346 int __init create_segment_manager_caches(void)
3347 {
3348         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
3349                         sizeof(struct discard_entry));
3350         if (!discard_entry_slab)
3351                 goto fail;
3352
3353         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3354                         sizeof(struct discard_cmd));
3355         if (!discard_cmd_slab)
3356                 goto destroy_discard_entry;
3357
3358         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
3359                         sizeof(struct sit_entry_set));
3360         if (!sit_entry_set_slab)
3361                 goto destroy_discard_cmd;
3362
3363         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3364                         sizeof(struct inmem_pages));
3365         if (!inmem_entry_slab)
3366                 goto destroy_sit_entry_set;
3367         return 0;
3368
3369 destroy_sit_entry_set:
3370         kmem_cache_destroy(sit_entry_set_slab);
3371 destroy_discard_cmd:
3372         kmem_cache_destroy(discard_cmd_slab);
3373 destroy_discard_entry:
3374         kmem_cache_destroy(discard_entry_slab);
3375 fail:
3376         return -ENOMEM;
3377 }
3378
3379 void destroy_segment_manager_caches(void)
3380 {
3381         kmem_cache_destroy(sit_entry_set_slab);
3382         kmem_cache_destroy(discard_cmd_slab);
3383         kmem_cache_destroy(discard_entry_slab);
3384         kmem_cache_destroy(inmem_entry_slab);
3385 }