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 static int __commit_inmem_pages(struct inode *inode,
 254                                         struct list_head *revoke_list)
 255 {
 256         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 257         struct f2fs_inode_info *fi = F2FS_I(inode);
 258         struct inmem_pages *cur, *tmp;
 259         struct f2fs_io_info fio = {
 260                 .sbi = sbi,
 261                 .type = DATA,
 262                 .op = REQ_OP_WRITE,
 263                 .op_flags = REQ_SYNC | REQ_PRIO,
 264                 .encrypted_page = NULL,
 265         };
 266         pgoff_t last_idx = ULONG_MAX;
 267         int err = 0;
 268
 269         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 270                 struct page *page = cur->page;
 271
 272                 lock_page(page);
 273                 if (page->mapping == inode->i_mapping) {
 274                         trace_f2fs_commit_inmem_page(page, INMEM);
 275
 276                         set_page_dirty(page);
 277                         f2fs_wait_on_page_writeback(page, DATA, true);
 278                         if (clear_page_dirty_for_io(page)) {
 279                                 inode_dec_dirty_pages(inode);
 280                                 remove_dirty_inode(inode);
 281                         }
 282
 283                         fio.page = page;
 284                         err = do_write_data_page(&fio);
 285                         if (err) {
 286                                 unlock_page(page);
 287                                 break;
 288                         }
 289
 290                         /* record old blkaddr for revoking */
 291                         cur->old_addr = fio.old_blkaddr;
 292                         last_idx = page->index;
 293                 }
 294                 unlock_page(page);
 295                 list_move_tail(&cur->list, revoke_list);
 296         }
 297
 298         if (last_idx != ULONG_MAX)
 299                 f2fs_submit_merged_bio_cond(sbi, inode, 0, last_idx,
 300                                                         DATA, WRITE);
 301
 302         if (!err)
 303                 __revoke_inmem_pages(inode, revoke_list, false, false);
 304
 305         return err;
 306 }
 307
 308 int commit_inmem_pages(struct inode *inode)
 309 {
 310         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 311         struct f2fs_inode_info *fi = F2FS_I(inode);
 312         struct list_head revoke_list;
 313         int err;
 314
 315         INIT_LIST_HEAD(&revoke_list);
 316         f2fs_balance_fs(sbi, true);
 317         f2fs_lock_op(sbi);
 318
 319         set_inode_flag(inode, FI_ATOMIC_COMMIT);
 320
 321         mutex_lock(&fi->inmem_lock);
 322         err = __commit_inmem_pages(inode, &revoke_list);
 323         if (err) {
 324                 int ret;
 325                 /*
 326                  * try to revoke all committed pages, but still we could fail
 327                  * due to no memory or other reason, if that happened, EAGAIN
 328                  * will be returned, which means in such case, transaction is
 329                  * already not integrity, caller should use journal to do the
 330                  * recovery or rewrite & commit last transaction. For other
 331                  * error number, revoking was done by filesystem itself.
 332                  */
 333                 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
 334                 if (ret)
 335                         err = ret;
 336
 337                 /* drop all uncommitted pages */
 338                 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 339         }
 340         mutex_unlock(&fi->inmem_lock);
 341
 342         clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 343
 344         f2fs_unlock_op(sbi);
 345         return err;
 346 }
 347
 348 /*
 349  * This function balances dirty node and dentry pages.
 350  * In addition, it controls garbage collection.
 351  */
 352 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 353 {
 354 #ifdef CONFIG_F2FS_FAULT_INJECTION
 355         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 356                 f2fs_show_injection_info(FAULT_CHECKPOINT);
 357                 f2fs_stop_checkpoint(sbi, false);
 358         }
 359 #endif
 360
 361         if (!need)
 362                 return;
 363
 364         /* balance_fs_bg is able to be pending */
 365         if (excess_cached_nats(sbi))
 366                 f2fs_balance_fs_bg(sbi);
 367
 368         /*
 369          * We should do GC or end up with checkpoint, if there are so many dirty
 370          * dir/node pages without enough free segments.
 371          */
 372         if (has_not_enough_free_secs(sbi, 0, 0)) {
 373                 mutex_lock(&sbi->gc_mutex);
 374                 f2fs_gc(sbi, false, false);
 375         }
 376 }
 377
 378 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 379 {
 380         /* try to shrink extent cache when there is no enough memory */
 381         if (!available_free_memory(sbi, EXTENT_CACHE))
 382                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 383
 384         /* check the # of cached NAT entries */
 385         if (!available_free_memory(sbi, NAT_ENTRIES))
 386                 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 387
 388         if (!available_free_memory(sbi, FREE_NIDS))
 389                 try_to_free_nids(sbi, MAX_FREE_NIDS);
 390         else
 391                 build_free_nids(sbi, false, false);
 392
 393         if (!is_idle(sbi))
 394                 return;
 395
 396         /* checkpoint is the only way to shrink partial cached entries */
 397         if (!available_free_memory(sbi, NAT_ENTRIES) ||
 398                         !available_free_memory(sbi, INO_ENTRIES) ||
 399                         excess_prefree_segs(sbi) ||
 400                         excess_dirty_nats(sbi) ||
 401                         f2fs_time_over(sbi, CP_TIME)) {
 402                 if (test_opt(sbi, DATA_FLUSH)) {
 403                         struct blk_plug plug;
 404
 405                         blk_start_plug(&plug);
 406                         sync_dirty_inodes(sbi, FILE_INODE);
 407                         blk_finish_plug(&plug);
 408                 }
 409                 f2fs_sync_fs(sbi->sb, true);
 410                 stat_inc_bg_cp_count(sbi->stat_info);
 411         }
 412 }
 413
 414 static int __submit_flush_wait(struct block_device *bdev)
 415 {
 416         struct bio *bio = f2fs_bio_alloc(0);
 417         int ret;
 418
 419         bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
 420         bio->bi_bdev = bdev;
 421         ret = submit_bio_wait(bio);
 422         bio_put(bio);
 423         return ret;
 424 }
 425
 426 static int submit_flush_wait(struct f2fs_sb_info *sbi)
 427 {
 428         int ret = __submit_flush_wait(sbi->sb->s_bdev);
 429         int i;
 430
 431         if (sbi->s_ndevs && !ret) {
 432                 for (i = 1; i < sbi->s_ndevs; i++) {
 433                         trace_f2fs_issue_flush(FDEV(i).bdev,
 434                                         test_opt(sbi, NOBARRIER),
 435                                         test_opt(sbi, FLUSH_MERGE));
 436                         ret = __submit_flush_wait(FDEV(i).bdev);
 437                         if (ret)
 438                                 break;
 439                 }
 440         }
 441         return ret;
 442 }
 443
 444 static int issue_flush_thread(void *data)
 445 {
 446         struct f2fs_sb_info *sbi = data;
 447         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 448         wait_queue_head_t *q = &fcc->flush_wait_queue;
 449 repeat:
 450         if (kthread_should_stop())
 451                 return 0;
 452
 453         if (!llist_empty(&fcc->issue_list)) {
 454                 struct flush_cmd *cmd, *next;
 455                 int ret;
 456
 457                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 458                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 459
 460                 ret = submit_flush_wait(sbi);
 461                 llist_for_each_entry_safe(cmd, next,
 462                                           fcc->dispatch_list, llnode) {
 463                         cmd->ret = ret;
 464                         complete(&cmd->wait);
 465                 }
 466                 fcc->dispatch_list = NULL;
 467         }
 468
 469         wait_event_interruptible(*q,
 470                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 471         goto repeat;
 472 }
 473
 474 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
 475 {
 476         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 477         struct flush_cmd cmd;
 478
 479         if (test_opt(sbi, NOBARRIER))
 480                 return 0;
 481
 482         if (!test_opt(sbi, FLUSH_MERGE))
 483                 return submit_flush_wait(sbi);
 484
 485         if (!atomic_read(&fcc->submit_flush)) {
 486                 int ret;
 487
 488                 atomic_inc(&fcc->submit_flush);
 489                 ret = submit_flush_wait(sbi);
 490                 atomic_dec(&fcc->submit_flush);
 491                 return ret;
 492         }
 493
 494         init_completion(&cmd.wait);
 495
 496         atomic_inc(&fcc->submit_flush);
 497         llist_add(&cmd.llnode, &fcc->issue_list);
 498
 499         if (!fcc->dispatch_list)
 500                 wake_up(&fcc->flush_wait_queue);
 501
 502         if (fcc->f2fs_issue_flush) {
 503                 wait_for_completion(&cmd.wait);
 504                 atomic_dec(&fcc->submit_flush);
 505         } else {
 506                 llist_del_all(&fcc->issue_list);
 507                 atomic_set(&fcc->submit_flush, 0);
 508         }
 509
 510         return cmd.ret;
 511 }
 512
 513 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
 514 {
 515         dev_t dev = sbi->sb->s_bdev->bd_dev;
 516         struct flush_cmd_control *fcc;
 517         int err = 0;
 518
 519         if (SM_I(sbi)->fcc_info) {
 520                 fcc = SM_I(sbi)->fcc_info;
 521                 goto init_thread;
 522         }
 523
 524         fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
 525         if (!fcc)
 526                 return -ENOMEM;
 527         atomic_set(&fcc->submit_flush, 0);
 528         init_waitqueue_head(&fcc->flush_wait_queue);
 529         init_llist_head(&fcc->issue_list);
 530         SM_I(sbi)->fcc_info = fcc;
 531 init_thread:
 532         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 533                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 534         if (IS_ERR(fcc->f2fs_issue_flush)) {
 535                 err = PTR_ERR(fcc->f2fs_issue_flush);
 536                 kfree(fcc);
 537                 SM_I(sbi)->fcc_info = NULL;
 538                 return err;
 539         }
 540
 541         return err;
 542 }
 543
 544 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 545 {
 546         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 547
 548         if (fcc && fcc->f2fs_issue_flush) {
 549                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 550
 551                 fcc->f2fs_issue_flush = NULL;
 552                 kthread_stop(flush_thread);
 553         }
 554         if (free) {
 555                 kfree(fcc);
 556                 SM_I(sbi)->fcc_info = NULL;
 557         }
 558 }
 559
 560 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 561                 enum dirty_type dirty_type)
 562 {
 563         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 564
 565         /* need not be added */
 566         if (IS_CURSEG(sbi, segno))
 567                 return;
 568
 569         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 570                 dirty_i->nr_dirty[dirty_type]++;
 571
 572         if (dirty_type == DIRTY) {
 573                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 574                 enum dirty_type t = sentry->type;
 575
 576                 if (unlikely(t >= DIRTY)) {
 577                         f2fs_bug_on(sbi, 1);
 578                         return;
 579                 }
 580                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 581                         dirty_i->nr_dirty[t]++;
 582         }
 583 }
 584
 585 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 586                 enum dirty_type dirty_type)
 587 {
 588         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 589
 590         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 591                 dirty_i->nr_dirty[dirty_type]--;
 592
 593         if (dirty_type == DIRTY) {
 594                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 595                 enum dirty_type t = sentry->type;
 596
 597                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 598                         dirty_i->nr_dirty[t]--;
 599
 600                 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
 601                         clear_bit(GET_SECNO(sbi, segno),
 602                                                 dirty_i->victim_secmap);
 603         }
 604 }
 605
 606 /*
 607  * Should not occur error such as -ENOMEM.
 608  * Adding dirty entry into seglist is not critical operation.
 609  * If a given segment is one of current working segments, it won't be added.
 610  */
 611 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 612 {
 613         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 614         unsigned short valid_blocks;
 615
 616         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 617                 return;
 618
 619         mutex_lock(&dirty_i->seglist_lock);
 620
 621         valid_blocks = get_valid_blocks(sbi, segno, 0);
 622
 623         if (valid_blocks == 0) {
 624                 __locate_dirty_segment(sbi, segno, PRE);
 625                 __remove_dirty_segment(sbi, segno, DIRTY);
 626         } else if (valid_blocks < sbi->blocks_per_seg) {
 627                 __locate_dirty_segment(sbi, segno, DIRTY);
 628         } else {
 629                 /* Recovery routine with SSR needs this */
 630                 __remove_dirty_segment(sbi, segno, DIRTY);
 631         }
 632
 633         mutex_unlock(&dirty_i->seglist_lock);
 634 }
 635
 636 static void __add_discard_cmd(struct f2fs_sb_info *sbi,
 637                         struct bio *bio, block_t lstart, block_t len)
 638 {
 639         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 640         struct list_head *cmd_list = &(dcc->discard_cmd_list);
 641         struct discard_cmd *dc;
 642
 643         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 644         INIT_LIST_HEAD(&dc->list);
 645         dc->bio = bio;
 646         bio->bi_private = dc;
 647         dc->lstart = lstart;
 648         dc->len = len;
 649         dc->state = D_PREP;
 650         init_completion(&dc->wait);
 651
 652         mutex_lock(&dcc->cmd_lock);
 653         list_add_tail(&dc->list, cmd_list);
 654         mutex_unlock(&dcc->cmd_lock);
 655 }
 656
 657 static void __remove_discard_cmd(struct f2fs_sb_info *sbi, struct discard_cmd *dc)
 658 {
 659         int err = dc->bio->bi_error;
 660
 661         if (dc->state == D_DONE)
 662                 atomic_dec(&(SM_I(sbi)->dcc_info->submit_discard));
 663
 664         if (err == -EOPNOTSUPP)
 665                 err = 0;
 666
 667         if (err)
 668                 f2fs_msg(sbi->sb, KERN_INFO,
 669                                 "Issue discard failed, ret: %d", err);
 670         bio_put(dc->bio);
 671         list_del(&dc->list);
 672         kmem_cache_free(discard_cmd_slab, dc);
 673 }
 674
 675 /* This should be covered by global mutex, &sit_i->sentry_lock */
 676 void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
 677 {
 678         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 679         struct list_head *wait_list = &(dcc->discard_cmd_list);
 680         struct discard_cmd *dc, *tmp;
 681         struct blk_plug plug;
 682
 683         mutex_lock(&dcc->cmd_lock);
 684
 685         blk_start_plug(&plug);
 686
 687         list_for_each_entry_safe(dc, tmp, wait_list, list) {
 688
 689                 if (blkaddr == NULL_ADDR) {
 690                         if (dc->state == D_PREP) {
 691                                 dc->state = D_SUBMIT;
 692                                 submit_bio(dc->bio);
 693                                 atomic_inc(&dcc->submit_discard);
 694                         }
 695                         continue;
 696                 }
 697
 698                 if (dc->lstart <= blkaddr && blkaddr < dc->lstart + dc->len) {
 699                         if (dc->state == D_SUBMIT)
 700                                 wait_for_completion_io(&dc->wait);
 701                         else
 702                                 __remove_discard_cmd(sbi, dc);
 703                 }
 704         }
 705         blk_finish_plug(&plug);
 706
 707         /* this comes from f2fs_put_super */
 708         if (blkaddr == NULL_ADDR) {
 709                 list_for_each_entry_safe(dc, tmp, wait_list, list) {
 710                         wait_for_completion_io(&dc->wait);
 711                         __remove_discard_cmd(sbi, dc);
 712                 }
 713         }
 714         mutex_unlock(&dcc->cmd_lock);
 715 }
 716
 717 static void f2fs_submit_discard_endio(struct bio *bio)
 718 {
 719         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
 720
 721         complete(&dc->wait);
 722         dc->state = D_DONE;
 723 }
 724
 725 static int issue_discard_thread(void *data)
 726 {
 727         struct f2fs_sb_info *sbi = data;
 728         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 729         wait_queue_head_t *q = &dcc->discard_wait_queue;
 730         struct list_head *cmd_list = &dcc->discard_cmd_list;
 731         struct discard_cmd *dc, *tmp;
 732         struct blk_plug plug;
 733         int iter = 0;
 734 repeat:
 735         if (kthread_should_stop())
 736                 return 0;
 737
 738         blk_start_plug(&plug);
 739
 740         mutex_lock(&dcc->cmd_lock);
 741         list_for_each_entry_safe(dc, tmp, cmd_list, list) {
 742                 if (dc->state == D_PREP) {
 743                         dc->state = D_SUBMIT;
 744                         submit_bio(dc->bio);
 745                         atomic_inc(&dcc->submit_discard);
 746                         if (iter++ > DISCARD_ISSUE_RATE)
 747                                 break;
 748                 } else if (dc->state == D_DONE) {
 749                         __remove_discard_cmd(sbi, dc);
 750                 }
 751         }
 752         mutex_unlock(&dcc->cmd_lock);
 753
 754         blk_finish_plug(&plug);
 755
 756         iter = 0;
 757         congestion_wait(BLK_RW_SYNC, HZ/50);
 758
 759         wait_event_interruptible(*q,
 760                 kthread_should_stop() || !list_empty(&dcc->discard_cmd_list));
 761         goto repeat;
 762 }
 763
 764
 765 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
 766 static int __f2fs_issue_discard_async(struct f2fs_sb_info *sbi,
 767                 struct block_device *bdev, block_t blkstart, block_t blklen)
 768 {
 769         struct bio *bio = NULL;
 770         block_t lblkstart = blkstart;
 771         int err;
 772
 773         trace_f2fs_issue_discard(bdev, blkstart, blklen);
 774
 775         if (sbi->s_ndevs) {
 776                 int devi = f2fs_target_device_index(sbi, blkstart);
 777
 778                 blkstart -= FDEV(devi).start_blk;
 779         }
 780         err = __blkdev_issue_discard(bdev,
 781                                 SECTOR_FROM_BLOCK(blkstart),
 782                                 SECTOR_FROM_BLOCK(blklen),
 783                                 GFP_NOFS, 0, &bio);
 784         if (!err && bio) {
 785                 bio->bi_end_io = f2fs_submit_discard_endio;
 786                 bio->bi_opf |= REQ_SYNC;
 787
 788                 __add_discard_cmd(sbi, bio, lblkstart, blklen);
 789                 wake_up(&SM_I(sbi)->dcc_info->discard_wait_queue);
 790         }
 791         return err;
 792 }
 793
 794 #ifdef CONFIG_BLK_DEV_ZONED
 795 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
 796                 struct block_device *bdev, block_t blkstart, block_t blklen)
 797 {
 798         sector_t sector, nr_sects;
 799         int devi = 0;
 800
 801         if (sbi->s_ndevs) {
 802                 devi = f2fs_target_device_index(sbi, blkstart);
 803                 blkstart -= FDEV(devi).start_blk;
 804         }
 805
 806         /*
 807          * We need to know the type of the zone: for conventional zones,
 808          * use regular discard if the drive supports it. For sequential
 809          * zones, reset the zone write pointer.
 810          */
 811         switch (get_blkz_type(sbi, bdev, blkstart)) {
 812
 813         case BLK_ZONE_TYPE_CONVENTIONAL:
 814                 if (!blk_queue_discard(bdev_get_queue(bdev)))
 815                         return 0;
 816                 return __f2fs_issue_discard_async(sbi, bdev, blkstart, blklen);
 817         case BLK_ZONE_TYPE_SEQWRITE_REQ:
 818         case BLK_ZONE_TYPE_SEQWRITE_PREF:
 819                 sector = SECTOR_FROM_BLOCK(blkstart);
 820                 nr_sects = SECTOR_FROM_BLOCK(blklen);
 821
 822                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
 823                                 nr_sects != bdev_zone_sectors(bdev)) {
 824                         f2fs_msg(sbi->sb, KERN_INFO,
 825                                 "(%d) %s: Unaligned discard attempted (block %x + %x)",
 826                                 devi, sbi->s_ndevs ? FDEV(devi).path: "",
 827                                 blkstart, blklen);
 828                         return -EIO;
 829                 }
 830                 trace_f2fs_issue_reset_zone(bdev, blkstart);
 831                 return blkdev_reset_zones(bdev, sector,
 832                                           nr_sects, GFP_NOFS);
 833         default:
 834                 /* Unknown zone type: broken device ? */
 835                 return -EIO;
 836         }
 837 }
 838 #endif
 839
 840 static int __issue_discard_async(struct f2fs_sb_info *sbi,
 841                 struct block_device *bdev, block_t blkstart, block_t blklen)
 842 {
 843 #ifdef CONFIG_BLK_DEV_ZONED
 844         if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
 845                                 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
 846                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
 847 #endif
 848         return __f2fs_issue_discard_async(sbi, bdev, blkstart, blklen);
 849 }
 850
 851 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
 852                                 block_t blkstart, block_t blklen)
 853 {
 854         sector_t start = blkstart, len = 0;
 855         struct block_device *bdev;
 856         struct seg_entry *se;
 857         unsigned int offset;
 858         block_t i;
 859         int err = 0;
 860
 861         bdev = f2fs_target_device(sbi, blkstart, NULL);
 862
 863         for (i = blkstart; i < blkstart + blklen; i++, len++) {
 864                 if (i != start) {
 865                         struct block_device *bdev2 =
 866                                 f2fs_target_device(sbi, i, NULL);
 867
 868                         if (bdev2 != bdev) {
 869                                 err = __issue_discard_async(sbi, bdev,
 870                                                 start, len);
 871                                 if (err)
 872                                         return err;
 873                                 bdev = bdev2;
 874                                 start = i;
 875                                 len = 0;
 876                         }
 877                 }
 878
 879                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
 880                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
 881
 882                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
 883                         sbi->discard_blks--;
 884         }
 885
 886         if (len)
 887                 err = __issue_discard_async(sbi, bdev, start, len);
 888         return err;
 889 }
 890
 891 static void __add_discard_entry(struct f2fs_sb_info *sbi,
 892                 struct cp_control *cpc, struct seg_entry *se,
 893                 unsigned int start, unsigned int end)
 894 {
 895         struct list_head *head = &SM_I(sbi)->dcc_info->discard_entry_list;
 896         struct discard_entry *new, *last;
 897
 898         if (!list_empty(head)) {
 899                 last = list_last_entry(head, struct discard_entry, list);
 900                 if (START_BLOCK(sbi, cpc->trim_start) + start ==
 901                                 last->blkaddr + last->len &&
 902                                 last->len < MAX_DISCARD_BLOCKS(sbi)) {
 903                         last->len += end - start;
 904                         goto done;
 905                 }
 906         }
 907
 908         new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
 909         INIT_LIST_HEAD(&new->list);
 910         new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
 911         new->len = end - start;
 912         list_add_tail(&new->list, head);
 913 done:
 914         SM_I(sbi)->dcc_info->nr_discards += end - start;
 915 }
 916
 917 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
 918                                                         bool check_only)
 919 {
 920         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
 921         int max_blocks = sbi->blocks_per_seg;
 922         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
 923         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
 924         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
 925         unsigned long *discard_map = (unsigned long *)se->discard_map;
 926         unsigned long *dmap = SIT_I(sbi)->tmp_map;
 927         unsigned int start = 0, end = -1;
 928         bool force = (cpc->reason == CP_DISCARD);
 929         int i;
 930
 931         if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
 932                 return false;
 933
 934         if (!force) {
 935                 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
 936                         SM_I(sbi)->dcc_info->nr_discards >=
 937                                 SM_I(sbi)->dcc_info->max_discards)
 938                         return false;
 939         }
 940
 941         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
 942         for (i = 0; i < entries; i++)
 943                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
 944                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
 945
 946         while (force || SM_I(sbi)->dcc_info->nr_discards <=
 947                                 SM_I(sbi)->dcc_info->max_discards) {
 948                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
 949                 if (start >= max_blocks)
 950                         break;
 951
 952                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
 953                 if (force && start && end != max_blocks
 954                                         && (end - start) < cpc->trim_minlen)
 955                         continue;
 956
 957                 if (check_only)
 958                         return true;
 959
 960                 __add_discard_entry(sbi, cpc, se, start, end);
 961         }
 962         return false;
 963 }
 964
 965 void release_discard_addrs(struct f2fs_sb_info *sbi)
 966 {
 967         struct list_head *head = &(SM_I(sbi)->dcc_info->discard_entry_list);
 968         struct discard_entry *entry, *this;
 969
 970         /* drop caches */
 971         list_for_each_entry_safe(entry, this, head, list) {
 972                 list_del(&entry->list);
 973                 kmem_cache_free(discard_entry_slab, entry);
 974         }
 975 }
 976
 977 /*
 978  * Should call clear_prefree_segments after checkpoint is done.
 979  */
 980 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
 981 {
 982         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 983         unsigned int segno;
 984
 985         mutex_lock(&dirty_i->seglist_lock);
 986         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
 987                 __set_test_and_free(sbi, segno);
 988         mutex_unlock(&dirty_i->seglist_lock);
 989 }
 990
 991 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 992 {
 993         struct list_head *head = &(SM_I(sbi)->dcc_info->discard_entry_list);
 994         struct discard_entry *entry, *this;
 995         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 996         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
 997         unsigned int start = 0, end = -1;
 998         unsigned int secno, start_segno;
 999         bool force = (cpc->reason == CP_DISCARD);
1000
1001         mutex_lock(&dirty_i->seglist_lock);
1002
1003         while (1) {
1004                 int i;
1005                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1006                 if (start >= MAIN_SEGS(sbi))
1007                         break;
1008                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1009                                                                 start + 1);
1010
1011                 for (i = start; i < end; i++)
1012                         clear_bit(i, prefree_map);
1013
1014                 dirty_i->nr_dirty[PRE] -= end - start;
1015
1016                 if (!test_opt(sbi, DISCARD))
1017                         continue;
1018
1019                 if (force && start >= cpc->trim_start &&
1020                                         (end - 1) <= cpc->trim_end)
1021                                 continue;
1022
1023                 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1024                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1025                                 (end - start) << sbi->log_blocks_per_seg);
1026                         continue;
1027                 }
1028 next:
1029                 secno = GET_SECNO(sbi, start);
1030                 start_segno = secno * sbi->segs_per_sec;
1031                 if (!IS_CURSEC(sbi, secno) &&
1032                         !get_valid_blocks(sbi, start, sbi->segs_per_sec))
1033                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1034                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1035
1036                 start = start_segno + sbi->segs_per_sec;
1037                 if (start < end)
1038                         goto next;
1039                 else
1040                         end = start - 1;
1041         }
1042         mutex_unlock(&dirty_i->seglist_lock);
1043
1044         /* send small discards */
1045         list_for_each_entry_safe(entry, this, head, list) {
1046                 if (force && entry->len < cpc->trim_minlen)
1047                         goto skip;
1048                 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
1049                 cpc->trimmed += entry->len;
1050 skip:
1051                 list_del(&entry->list);
1052                 SM_I(sbi)->dcc_info->nr_discards -= entry->len;
1053                 kmem_cache_free(discard_entry_slab, entry);
1054         }
1055 }
1056
1057 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1058 {
1059         dev_t dev = sbi->sb->s_bdev->bd_dev;
1060         struct discard_cmd_control *dcc;
1061         int err = 0;
1062
1063         if (SM_I(sbi)->dcc_info) {
1064                 dcc = SM_I(sbi)->dcc_info;
1065                 goto init_thread;
1066         }
1067
1068         dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1069         if (!dcc)
1070                 return -ENOMEM;
1071
1072         INIT_LIST_HEAD(&dcc->discard_entry_list);
1073         INIT_LIST_HEAD(&dcc->discard_cmd_list);
1074         mutex_init(&dcc->cmd_lock);
1075         atomic_set(&dcc->submit_discard, 0);
1076         dcc->nr_discards = 0;
1077         dcc->max_discards = 0;
1078
1079         init_waitqueue_head(&dcc->discard_wait_queue);
1080         SM_I(sbi)->dcc_info = dcc;
1081 init_thread:
1082         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1083                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1084         if (IS_ERR(dcc->f2fs_issue_discard)) {
1085                 err = PTR_ERR(dcc->f2fs_issue_discard);
1086                 kfree(dcc);
1087                 SM_I(sbi)->dcc_info = NULL;
1088                 return err;
1089         }
1090
1091         return err;
1092 }
1093
1094 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi, bool free)
1095 {
1096         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1097
1098         if (dcc && dcc->f2fs_issue_discard) {
1099                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1100
1101                 dcc->f2fs_issue_discard = NULL;
1102                 kthread_stop(discard_thread);
1103         }
1104         if (free) {
1105                 kfree(dcc);
1106                 SM_I(sbi)->dcc_info = NULL;
1107         }
1108 }
1109
1110 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1111 {
1112         struct sit_info *sit_i = SIT_I(sbi);
1113
1114         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1115                 sit_i->dirty_sentries++;
1116                 return false;
1117         }
1118
1119         return true;
1120 }
1121
1122 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1123                                         unsigned int segno, int modified)
1124 {
1125         struct seg_entry *se = get_seg_entry(sbi, segno);
1126         se->type = type;
1127         if (modified)
1128                 __mark_sit_entry_dirty(sbi, segno);
1129 }
1130
1131 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1132 {
1133         struct seg_entry *se;
1134         unsigned int segno, offset;
1135         long int new_vblocks;
1136
1137         segno = GET_SEGNO(sbi, blkaddr);
1138
1139         se = get_seg_entry(sbi, segno);
1140         new_vblocks = se->valid_blocks + del;
1141         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1142
1143         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
1144                                 (new_vblocks > sbi->blocks_per_seg)));
1145
1146         se->valid_blocks = new_vblocks;
1147         se->mtime = get_mtime(sbi);
1148         SIT_I(sbi)->max_mtime = se->mtime;
1149
1150         /* Update valid block bitmap */
1151         if (del > 0) {
1152                 if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) {
1153 #ifdef CONFIG_F2FS_CHECK_FS
1154                         if (f2fs_test_and_set_bit(offset,
1155                                                 se->cur_valid_map_mir))
1156                                 f2fs_bug_on(sbi, 1);
1157                         else
1158                                 WARN_ON(1);
1159 #else
1160                         f2fs_bug_on(sbi, 1);
1161 #endif
1162                 }
1163                 if (f2fs_discard_en(sbi) &&
1164                         !f2fs_test_and_set_bit(offset, se->discard_map))
1165                         sbi->discard_blks--;
1166         } else {
1167                 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) {
1168 #ifdef CONFIG_F2FS_CHECK_FS
1169                         if (!f2fs_test_and_clear_bit(offset,
1170                                                 se->cur_valid_map_mir))
1171                                 f2fs_bug_on(sbi, 1);
1172                         else
1173                                 WARN_ON(1);
1174 #else
1175                         f2fs_bug_on(sbi, 1);
1176 #endif
1177                 }
1178                 if (f2fs_discard_en(sbi) &&
1179                         f2fs_test_and_clear_bit(offset, se->discard_map))
1180                         sbi->discard_blks++;
1181         }
1182         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1183                 se->ckpt_valid_blocks += del;
1184
1185         __mark_sit_entry_dirty(sbi, segno);
1186
1187         /* update total number of valid blocks to be written in ckpt area */
1188         SIT_I(sbi)->written_valid_blocks += del;
1189
1190         if (sbi->segs_per_sec > 1)
1191                 get_sec_entry(sbi, segno)->valid_blocks += del;
1192 }
1193
1194 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
1195 {
1196         update_sit_entry(sbi, new, 1);
1197         if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1198                 update_sit_entry(sbi, old, -1);
1199
1200         locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1201         locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
1202 }
1203
1204 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1205 {
1206         unsigned int segno = GET_SEGNO(sbi, addr);
1207         struct sit_info *sit_i = SIT_I(sbi);
1208
1209         f2fs_bug_on(sbi, addr == NULL_ADDR);
1210         if (addr == NEW_ADDR)
1211                 return;
1212
1213         /* add it into sit main buffer */
1214         mutex_lock(&sit_i->sentry_lock);
1215
1216         update_sit_entry(sbi, addr, -1);
1217
1218         /* add it into dirty seglist */
1219         locate_dirty_segment(sbi, segno);
1220
1221         mutex_unlock(&sit_i->sentry_lock);
1222 }
1223
1224 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1225 {
1226         struct sit_info *sit_i = SIT_I(sbi);
1227         unsigned int segno, offset;
1228         struct seg_entry *se;
1229         bool is_cp = false;
1230
1231         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1232                 return true;
1233
1234         mutex_lock(&sit_i->sentry_lock);
1235
1236         segno = GET_SEGNO(sbi, blkaddr);
1237         se = get_seg_entry(sbi, segno);
1238         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1239
1240         if (f2fs_test_bit(offset, se->ckpt_valid_map))
1241                 is_cp = true;
1242
1243         mutex_unlock(&sit_i->sentry_lock);
1244
1245         return is_cp;
1246 }
1247
1248 /*
1249  * This function should be resided under the curseg_mutex lock
1250  */
1251 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1252                                         struct f2fs_summary *sum)
1253 {
1254         struct curseg_info *curseg = CURSEG_I(sbi, type);
1255         void *addr = curseg->sum_blk;
1256         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1257         memcpy(addr, sum, sizeof(struct f2fs_summary));
1258 }
1259
1260 /*
1261  * Calculate the number of current summary pages for writing
1262  */
1263 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1264 {
1265         int valid_sum_count = 0;
1266         int i, sum_in_page;
1267
1268         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1269                 if (sbi->ckpt->alloc_type[i] == SSR)
1270                         valid_sum_count += sbi->blocks_per_seg;
1271                 else {
1272                         if (for_ra)
1273                                 valid_sum_count += le16_to_cpu(
1274                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1275                         else
1276                                 valid_sum_count += curseg_blkoff(sbi, i);
1277                 }
1278         }
1279
1280         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1281                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1282         if (valid_sum_count <= sum_in_page)
1283                 return 1;
1284         else if ((valid_sum_count - sum_in_page) <=
1285                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1286                 return 2;
1287         return 3;
1288 }
1289
1290 /*
1291  * Caller should put this summary page
1292  */
1293 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1294 {
1295         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1296 }
1297
1298 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1299 {
1300         struct page *page = grab_meta_page(sbi, blk_addr);
1301         void *dst = page_address(page);
1302
1303         if (src)
1304                 memcpy(dst, src, PAGE_SIZE);
1305         else
1306                 memset(dst, 0, PAGE_SIZE);
1307         set_page_dirty(page);
1308         f2fs_put_page(page, 1);
1309 }
1310
1311 static void write_sum_page(struct f2fs_sb_info *sbi,
1312                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
1313 {
1314         update_meta_page(sbi, (void *)sum_blk, blk_addr);
1315 }
1316
1317 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1318                                                 int type, block_t blk_addr)
1319 {
1320         struct curseg_info *curseg = CURSEG_I(sbi, type);
1321         struct page *page = grab_meta_page(sbi, blk_addr);
1322         struct f2fs_summary_block *src = curseg->sum_blk;
1323         struct f2fs_summary_block *dst;
1324
1325         dst = (struct f2fs_summary_block *)page_address(page);
1326
1327         mutex_lock(&curseg->curseg_mutex);
1328
1329         down_read(&curseg->journal_rwsem);
1330         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1331         up_read(&curseg->journal_rwsem);
1332
1333         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1334         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1335
1336         mutex_unlock(&curseg->curseg_mutex);
1337
1338         set_page_dirty(page);
1339         f2fs_put_page(page, 1);
1340 }
1341
1342 /*
1343  * Find a new segment from the free segments bitmap to right order
1344  * This function should be returned with success, otherwise BUG
1345  */
1346 static void get_new_segment(struct f2fs_sb_info *sbi,
1347                         unsigned int *newseg, bool new_sec, int dir)
1348 {
1349         struct free_segmap_info *free_i = FREE_I(sbi);
1350         unsigned int segno, secno, zoneno;
1351         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
1352         unsigned int hint = *newseg / sbi->segs_per_sec;
1353         unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1354         unsigned int left_start = hint;
1355         bool init = true;
1356         int go_left = 0;
1357         int i;
1358
1359         spin_lock(&free_i->segmap_lock);
1360
1361         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1362                 segno = find_next_zero_bit(free_i->free_segmap,
1363                                 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1364                 if (segno < (hint + 1) * sbi->segs_per_sec)
1365                         goto got_it;
1366         }
1367 find_other_zone:
1368         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1369         if (secno >= MAIN_SECS(sbi)) {
1370                 if (dir == ALLOC_RIGHT) {
1371                         secno = find_next_zero_bit(free_i->free_secmap,
1372                                                         MAIN_SECS(sbi), 0);
1373                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1374                 } else {
1375                         go_left = 1;
1376                         left_start = hint - 1;
1377                 }
1378         }
1379         if (go_left == 0)
1380                 goto skip_left;
1381
1382         while (test_bit(left_start, free_i->free_secmap)) {
1383                 if (left_start > 0) {
1384                         left_start--;
1385                         continue;
1386                 }
1387                 left_start = find_next_zero_bit(free_i->free_secmap,
1388                                                         MAIN_SECS(sbi), 0);
1389                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1390                 break;
1391         }
1392         secno = left_start;
1393 skip_left:
1394         hint = secno;
1395         segno = secno * sbi->segs_per_sec;
1396         zoneno = secno / sbi->secs_per_zone;
1397
1398         /* give up on finding another zone */
1399         if (!init)
1400                 goto got_it;
1401         if (sbi->secs_per_zone == 1)
1402                 goto got_it;
1403         if (zoneno == old_zoneno)
1404                 goto got_it;
1405         if (dir == ALLOC_LEFT) {
1406                 if (!go_left && zoneno + 1 >= total_zones)
1407                         goto got_it;
1408                 if (go_left && zoneno == 0)
1409                         goto got_it;
1410         }
1411         for (i = 0; i < NR_CURSEG_TYPE; i++)
1412                 if (CURSEG_I(sbi, i)->zone == zoneno)
1413                         break;
1414
1415         if (i < NR_CURSEG_TYPE) {
1416                 /* zone is in user, try another */
1417                 if (go_left)
1418                         hint = zoneno * sbi->secs_per_zone - 1;
1419                 else if (zoneno + 1 >= total_zones)
1420                         hint = 0;
1421                 else
1422                         hint = (zoneno + 1) * sbi->secs_per_zone;
1423                 init = false;
1424                 goto find_other_zone;
1425         }
1426 got_it:
1427         /* set it as dirty segment in free segmap */
1428         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1429         __set_inuse(sbi, segno);
1430         *newseg = segno;
1431         spin_unlock(&free_i->segmap_lock);
1432 }
1433
1434 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1435 {
1436         struct curseg_info *curseg = CURSEG_I(sbi, type);
1437         struct summary_footer *sum_footer;
1438
1439         curseg->segno = curseg->next_segno;
1440         curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1441         curseg->next_blkoff = 0;
1442         curseg->next_segno = NULL_SEGNO;
1443
1444         sum_footer = &(curseg->sum_blk->footer);
1445         memset(sum_footer, 0, sizeof(struct summary_footer));
1446         if (IS_DATASEG(type))
1447                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1448         if (IS_NODESEG(type))
1449                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1450         __set_sit_entry_type(sbi, type, curseg->segno, modified);
1451 }
1452
1453 /*
1454  * Allocate a current working segment.
1455  * This function always allocates a free segment in LFS manner.
1456  */
1457 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1458 {
1459         struct curseg_info *curseg = CURSEG_I(sbi, type);
1460         unsigned int segno = curseg->segno;
1461         int dir = ALLOC_LEFT;
1462
1463         write_sum_page(sbi, curseg->sum_blk,
1464                                 GET_SUM_BLOCK(sbi, segno));
1465         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1466                 dir = ALLOC_RIGHT;
1467
1468         if (test_opt(sbi, NOHEAP))
1469                 dir = ALLOC_RIGHT;
1470
1471         get_new_segment(sbi, &segno, new_sec, dir);
1472         curseg->next_segno = segno;
1473         reset_curseg(sbi, type, 1);
1474         curseg->alloc_type = LFS;
1475 }
1476
1477 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1478                         struct curseg_info *seg, block_t start)
1479 {
1480         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1481         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1482         unsigned long *target_map = SIT_I(sbi)->tmp_map;
1483         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1484         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1485         int i, pos;
1486
1487         for (i = 0; i < entries; i++)
1488                 target_map[i] = ckpt_map[i] | cur_map[i];
1489
1490         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1491
1492         seg->next_blkoff = pos;
1493 }
1494
1495 /*
1496  * If a segment is written by LFS manner, next block offset is just obtained
1497  * by increasing the current block offset. However, if a segment is written by
1498  * SSR manner, next block offset obtained by calling __next_free_blkoff
1499  */
1500 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1501                                 struct curseg_info *seg)
1502 {
1503         if (seg->alloc_type == SSR)
1504                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1505         else
1506                 seg->next_blkoff++;
1507 }
1508
1509 /*
1510  * This function always allocates a used segment(from dirty seglist) by SSR
1511  * manner, so it should recover the existing segment information of valid blocks
1512  */
1513 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1514 {
1515         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1516         struct curseg_info *curseg = CURSEG_I(sbi, type);
1517         unsigned int new_segno = curseg->next_segno;
1518         struct f2fs_summary_block *sum_node;
1519         struct page *sum_page;
1520
1521         write_sum_page(sbi, curseg->sum_blk,
1522                                 GET_SUM_BLOCK(sbi, curseg->segno));
1523         __set_test_and_inuse(sbi, new_segno);
1524
1525         mutex_lock(&dirty_i->seglist_lock);
1526         __remove_dirty_segment(sbi, new_segno, PRE);
1527         __remove_dirty_segment(sbi, new_segno, DIRTY);
1528         mutex_unlock(&dirty_i->seglist_lock);
1529
1530         reset_curseg(sbi, type, 1);
1531         curseg->alloc_type = SSR;
1532         __next_free_blkoff(sbi, curseg, 0);
1533
1534         if (reuse) {
1535                 sum_page = get_sum_page(sbi, new_segno);
1536                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1537                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1538                 f2fs_put_page(sum_page, 1);
1539         }
1540 }
1541
1542 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1543 {
1544         struct curseg_info *curseg = CURSEG_I(sbi, type);
1545         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1546         int i, cnt;
1547         bool reversed = false;
1548
1549         /* need_SSR() already forces to do this */
1550         if (v_ops->get_victim(sbi, &(curseg)->next_segno, BG_GC, type, SSR))
1551                 return 1;
1552
1553         /* For node segments, let's do SSR more intensively */
1554         if (IS_NODESEG(type)) {
1555                 if (type >= CURSEG_WARM_NODE) {
1556                         reversed = true;
1557                         i = CURSEG_COLD_NODE;
1558                 } else {
1559                         i = CURSEG_HOT_NODE;
1560                 }
1561                 cnt = NR_CURSEG_NODE_TYPE;
1562         } else {
1563                 if (type >= CURSEG_WARM_DATA) {
1564                         reversed = true;
1565                         i = CURSEG_COLD_DATA;
1566                 } else {
1567                         i = CURSEG_HOT_DATA;
1568                 }
1569                 cnt = NR_CURSEG_DATA_TYPE;
1570         }
1571
1572         for (; cnt-- > 0; reversed ? i-- : i++) {
1573                 if (i == type)
1574                         continue;
1575                 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1576                                                 BG_GC, i, SSR))
1577                         return 1;
1578         }
1579         return 0;
1580 }
1581
1582 /*
1583  * flush out current segment and replace it with new segment
1584  * This function should be returned with success, otherwise BUG
1585  */
1586 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1587                                                 int type, bool force)
1588 {
1589         if (force)
1590                 new_curseg(sbi, type, true);
1591         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
1592                                         type == CURSEG_WARM_NODE)
1593                 new_curseg(sbi, type, false);
1594         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1595                 change_curseg(sbi, type, true);
1596         else
1597                 new_curseg(sbi, type, false);
1598
1599         stat_inc_seg_type(sbi, CURSEG_I(sbi, type));
1600 }
1601
1602 void allocate_new_segments(struct f2fs_sb_info *sbi)
1603 {
1604         struct curseg_info *curseg;
1605         unsigned int old_segno;
1606         int i;
1607
1608         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1609                 curseg = CURSEG_I(sbi, i);
1610                 old_segno = curseg->segno;
1611                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
1612                 locate_dirty_segment(sbi, old_segno);
1613         }
1614 }
1615
1616 static const struct segment_allocation default_salloc_ops = {
1617         .allocate_segment = allocate_segment_by_default,
1618 };
1619
1620 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1621 {
1622         __u64 trim_start = cpc->trim_start;
1623         bool has_candidate = false;
1624
1625         mutex_lock(&SIT_I(sbi)->sentry_lock);
1626         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
1627                 if (add_discard_addrs(sbi, cpc, true)) {
1628                         has_candidate = true;
1629                         break;
1630                 }
1631         }
1632         mutex_unlock(&SIT_I(sbi)->sentry_lock);
1633
1634         cpc->trim_start = trim_start;
1635         return has_candidate;
1636 }
1637
1638 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1639 {
1640         __u64 start = F2FS_BYTES_TO_BLK(range->start);
1641         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1642         unsigned int start_segno, end_segno;
1643         struct cp_control cpc;
1644         int err = 0;
1645
1646         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1647                 return -EINVAL;
1648
1649         cpc.trimmed = 0;
1650         if (end <= MAIN_BLKADDR(sbi))
1651                 goto out;
1652
1653         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1654                 f2fs_msg(sbi->sb, KERN_WARNING,
1655                         "Found FS corruption, run fsck to fix.");
1656                 goto out;
1657         }
1658
1659         /* start/end segment number in main_area */
1660         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1661         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1662                                                 GET_SEGNO(sbi, end);
1663         cpc.reason = CP_DISCARD;
1664         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1665
1666         /* do checkpoint to issue discard commands safely */
1667         for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1668                 cpc.trim_start = start_segno;
1669
1670                 if (sbi->discard_blks == 0)
1671                         break;
1672                 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1673                         cpc.trim_end = end_segno;
1674                 else
1675                         cpc.trim_end = min_t(unsigned int,
1676                                 rounddown(start_segno +
1677                                 BATCHED_TRIM_SEGMENTS(sbi),
1678                                 sbi->segs_per_sec) - 1, end_segno);
1679
1680                 mutex_lock(&sbi->gc_mutex);
1681                 err = write_checkpoint(sbi, &cpc);
1682                 mutex_unlock(&sbi->gc_mutex);
1683                 if (err)
1684                         break;
1685
1686                 schedule();
1687         }
1688 out:
1689         range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1690         return err;
1691 }
1692
1693 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1694 {
1695         struct curseg_info *curseg = CURSEG_I(sbi, type);
1696         if (curseg->next_blkoff < sbi->blocks_per_seg)
1697                 return true;
1698         return false;
1699 }
1700
1701 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1702 {
1703         if (p_type == DATA)
1704                 return CURSEG_HOT_DATA;
1705         else
1706                 return CURSEG_HOT_NODE;
1707 }
1708
1709 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1710 {
1711         if (p_type == DATA) {
1712                 struct inode *inode = page->mapping->host;
1713
1714                 if (S_ISDIR(inode->i_mode))
1715                         return CURSEG_HOT_DATA;
1716                 else
1717                         return CURSEG_COLD_DATA;
1718         } else {
1719                 if (IS_DNODE(page) && is_cold_node(page))
1720                         return CURSEG_WARM_NODE;
1721                 else
1722                         return CURSEG_COLD_NODE;
1723         }
1724 }
1725
1726 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1727 {
1728         if (p_type == DATA) {
1729                 struct inode *inode = page->mapping->host;
1730
1731                 if (S_ISDIR(inode->i_mode))
1732                         return CURSEG_HOT_DATA;
1733                 else if (is_cold_data(page) || file_is_cold(inode))
1734                         return CURSEG_COLD_DATA;
1735                 else
1736                         return CURSEG_WARM_DATA;
1737         } else {
1738                 if (IS_DNODE(page))
1739                         return is_cold_node(page) ? CURSEG_WARM_NODE :
1740                                                 CURSEG_HOT_NODE;
1741                 else
1742                         return CURSEG_COLD_NODE;
1743         }
1744 }
1745
1746 static int __get_segment_type(struct page *page, enum page_type p_type)
1747 {
1748         switch (F2FS_P_SB(page)->active_logs) {
1749         case 2:
1750                 return __get_segment_type_2(page, p_type);
1751         case 4:
1752                 return __get_segment_type_4(page, p_type);
1753         }
1754         /* NR_CURSEG_TYPE(6) logs by default */
1755         f2fs_bug_on(F2FS_P_SB(page),
1756                 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1757         return __get_segment_type_6(page, p_type);
1758 }
1759
1760 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1761                 block_t old_blkaddr, block_t *new_blkaddr,
1762                 struct f2fs_summary *sum, int type)
1763 {
1764         struct sit_info *sit_i = SIT_I(sbi);
1765         struct curseg_info *curseg = CURSEG_I(sbi, type);
1766
1767         mutex_lock(&curseg->curseg_mutex);
1768         mutex_lock(&sit_i->sentry_lock);
1769
1770         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1771
1772         f2fs_wait_discard_bio(sbi, *new_blkaddr);
1773
1774         /*
1775          * __add_sum_entry should be resided under the curseg_mutex
1776          * because, this function updates a summary entry in the
1777          * current summary block.
1778          */
1779         __add_sum_entry(sbi, type, sum);
1780
1781         __refresh_next_blkoff(sbi, curseg);
1782
1783         stat_inc_block_count(sbi, curseg);
1784
1785         /*
1786          * SIT information should be updated before segment allocation,
1787          * since SSR needs latest valid block information.
1788          */
1789         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1790
1791         if (!__has_curseg_space(sbi, type))
1792                 sit_i->s_ops->allocate_segment(sbi, type, false);
1793
1794         mutex_unlock(&sit_i->sentry_lock);
1795
1796         if (page && IS_NODESEG(type))
1797                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1798
1799         mutex_unlock(&curseg->curseg_mutex);
1800 }
1801
1802 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1803 {
1804         int type = __get_segment_type(fio->page, fio->type);
1805         int err;
1806
1807         if (fio->type == NODE || fio->type == DATA)
1808                 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1809 reallocate:
1810         allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1811                                         &fio->new_blkaddr, sum, type);
1812
1813         /* writeout dirty page into bdev */
1814         err = f2fs_submit_page_mbio(fio);
1815         if (err == -EAGAIN) {
1816                 fio->old_blkaddr = fio->new_blkaddr;
1817                 goto reallocate;
1818         }
1819
1820         if (fio->type == NODE || fio->type == DATA)
1821                 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
1822 }
1823
1824 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1825 {
1826         struct f2fs_io_info fio = {
1827                 .sbi = sbi,
1828                 .type = META,
1829                 .op = REQ_OP_WRITE,
1830                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
1831                 .old_blkaddr = page->index,
1832                 .new_blkaddr = page->index,
1833                 .page = page,
1834                 .encrypted_page = NULL,
1835         };
1836
1837         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1838                 fio.op_flags &= ~REQ_META;
1839
1840         set_page_writeback(page);
1841         f2fs_submit_page_mbio(&fio);
1842 }
1843
1844 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1845 {
1846         struct f2fs_summary sum;
1847
1848         set_summary(&sum, nid, 0, 0);
1849         do_write_page(&sum, fio);
1850 }
1851
1852 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1853 {
1854         struct f2fs_sb_info *sbi = fio->sbi;
1855         struct f2fs_summary sum;
1856         struct node_info ni;
1857
1858         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1859         get_node_info(sbi, dn->nid, &ni);
1860         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1861         do_write_page(&sum, fio);
1862         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
1863 }
1864
1865 void rewrite_data_page(struct f2fs_io_info *fio)
1866 {
1867         fio->new_blkaddr = fio->old_blkaddr;
1868         stat_inc_inplace_blocks(fio->sbi);
1869         f2fs_submit_page_mbio(fio);
1870 }
1871
1872 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1873                                 block_t old_blkaddr, block_t new_blkaddr,
1874                                 bool recover_curseg, bool recover_newaddr)
1875 {
1876         struct sit_info *sit_i = SIT_I(sbi);
1877         struct curseg_info *curseg;
1878         unsigned int segno, old_cursegno;
1879         struct seg_entry *se;
1880         int type;
1881         unsigned short old_blkoff;
1882
1883         segno = GET_SEGNO(sbi, new_blkaddr);
1884         se = get_seg_entry(sbi, segno);
1885         type = se->type;
1886
1887         if (!recover_curseg) {
1888                 /* for recovery flow */
1889                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1890                         if (old_blkaddr == NULL_ADDR)
1891                                 type = CURSEG_COLD_DATA;
1892                         else
1893                                 type = CURSEG_WARM_DATA;
1894                 }
1895         } else {
1896                 if (!IS_CURSEG(sbi, segno))
1897                         type = CURSEG_WARM_DATA;
1898         }
1899
1900         curseg = CURSEG_I(sbi, type);
1901
1902         mutex_lock(&curseg->curseg_mutex);
1903         mutex_lock(&sit_i->sentry_lock);
1904
1905         old_cursegno = curseg->segno;
1906         old_blkoff = curseg->next_blkoff;
1907
1908         /* change the current segment */
1909         if (segno != curseg->segno) {
1910                 curseg->next_segno = segno;
1911                 change_curseg(sbi, type, true);
1912         }
1913
1914         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1915         __add_sum_entry(sbi, type, sum);
1916
1917         if (!recover_curseg || recover_newaddr)
1918                 update_sit_entry(sbi, new_blkaddr, 1);
1919         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1920                 update_sit_entry(sbi, old_blkaddr, -1);
1921
1922         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1923         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1924
1925         locate_dirty_segment(sbi, old_cursegno);
1926
1927         if (recover_curseg) {
1928                 if (old_cursegno != curseg->segno) {
1929                         curseg->next_segno = old_cursegno;
1930                         change_curseg(sbi, type, true);
1931                 }
1932                 curseg->next_blkoff = old_blkoff;
1933         }
1934
1935         mutex_unlock(&sit_i->sentry_lock);
1936         mutex_unlock(&curseg->curseg_mutex);
1937 }
1938
1939 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1940                                 block_t old_addr, block_t new_addr,
1941                                 unsigned char version, bool recover_curseg,
1942                                 bool recover_newaddr)
1943 {
1944         struct f2fs_summary sum;
1945
1946         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1947
1948         __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1949                                         recover_curseg, recover_newaddr);
1950
1951         f2fs_update_data_blkaddr(dn, new_addr);
1952 }
1953
1954 void f2fs_wait_on_page_writeback(struct page *page,
1955                                 enum page_type type, bool ordered)
1956 {
1957         if (PageWriteback(page)) {
1958                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1959
1960                 f2fs_submit_merged_bio_cond(sbi, page->mapping->host,
1961                                                 0, page->index, type, WRITE);
1962                 if (ordered)
1963                         wait_on_page_writeback(page);
1964                 else
1965                         wait_for_stable_page(page);
1966         }
1967 }
1968
1969 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1970                                                         block_t blkaddr)
1971 {
1972         struct page *cpage;
1973
1974         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1975                 return;
1976
1977         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1978         if (cpage) {
1979                 f2fs_wait_on_page_writeback(cpage, DATA, true);
1980                 f2fs_put_page(cpage, 1);
1981         }
1982 }
1983
1984 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1985 {
1986         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1987         struct curseg_info *seg_i;
1988         unsigned char *kaddr;
1989         struct page *page;
1990         block_t start;
1991         int i, j, offset;
1992
1993         start = start_sum_block(sbi);
1994
1995         page = get_meta_page(sbi, start++);
1996         kaddr = (unsigned char *)page_address(page);
1997
1998         /* Step 1: restore nat cache */
1999         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2000         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
2001
2002         /* Step 2: restore sit cache */
2003         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2004         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
2005         offset = 2 * SUM_JOURNAL_SIZE;
2006
2007         /* Step 3: restore summary entries */
2008         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2009                 unsigned short blk_off;
2010                 unsigned int segno;
2011
2012                 seg_i = CURSEG_I(sbi, i);
2013                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2014                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2015                 seg_i->next_segno = segno;
2016                 reset_curseg(sbi, i, 0);
2017                 seg_i->alloc_type = ckpt->alloc_type[i];
2018                 seg_i->next_blkoff = blk_off;
2019
2020                 if (seg_i->alloc_type == SSR)
2021                         blk_off = sbi->blocks_per_seg;
2022
2023                 for (j = 0; j < blk_off; j++) {
2024                         struct f2fs_summary *s;
2025                         s = (struct f2fs_summary *)(kaddr + offset);
2026                         seg_i->sum_blk->entries[j] = *s;
2027                         offset += SUMMARY_SIZE;
2028                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
2029                                                 SUM_FOOTER_SIZE)
2030                                 continue;
2031
2032                         f2fs_put_page(page, 1);
2033                         page = NULL;
2034
2035                         page = get_meta_page(sbi, start++);
2036                         kaddr = (unsigned char *)page_address(page);
2037                         offset = 0;
2038                 }
2039         }
2040         f2fs_put_page(page, 1);
2041         return 0;
2042 }
2043
2044 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2045 {
2046         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2047         struct f2fs_summary_block *sum;
2048         struct curseg_info *curseg;
2049         struct page *new;
2050         unsigned short blk_off;
2051         unsigned int segno = 0;
2052         block_t blk_addr = 0;
2053
2054         /* get segment number and block addr */
2055         if (IS_DATASEG(type)) {
2056                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2057                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2058                                                         CURSEG_HOT_DATA]);
2059                 if (__exist_node_summaries(sbi))
2060                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2061                 else
2062                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2063         } else {
2064                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2065                                                         CURSEG_HOT_NODE]);
2066                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2067                                                         CURSEG_HOT_NODE]);
2068                 if (__exist_node_summaries(sbi))
2069                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2070                                                         type - CURSEG_HOT_NODE);
2071                 else
2072                         blk_addr = GET_SUM_BLOCK(sbi, segno);
2073         }
2074
2075         new = get_meta_page(sbi, blk_addr);
2076         sum = (struct f2fs_summary_block *)page_address(new);
2077
2078         if (IS_NODESEG(type)) {
2079                 if (__exist_node_summaries(sbi)) {
2080                         struct f2fs_summary *ns = &sum->entries[0];
2081                         int i;
2082                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2083                                 ns->version = 0;
2084                                 ns->ofs_in_node = 0;
2085                         }
2086                 } else {
2087                         int err;
2088
2089                         err = restore_node_summary(sbi, segno, sum);
2090                         if (err) {
2091                                 f2fs_put_page(new, 1);
2092                                 return err;
2093                         }
2094                 }
2095         }
2096
2097         /* set uncompleted segment to curseg */
2098         curseg = CURSEG_I(sbi, type);
2099         mutex_lock(&curseg->curseg_mutex);
2100
2101         /* update journal info */
2102         down_write(&curseg->journal_rwsem);
2103         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2104         up_write(&curseg->journal_rwsem);
2105
2106         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2107         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
2108         curseg->next_segno = segno;
2109         reset_curseg(sbi, type, 0);
2110         curseg->alloc_type = ckpt->alloc_type[type];
2111         curseg->next_blkoff = blk_off;
2112         mutex_unlock(&curseg->curseg_mutex);
2113         f2fs_put_page(new, 1);
2114         return 0;
2115 }
2116
2117 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2118 {
2119         int type = CURSEG_HOT_DATA;
2120         int err;
2121
2122         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
2123                 int npages = npages_for_summary_flush(sbi, true);
2124
2125                 if (npages >= 2)
2126                         ra_meta_pages(sbi, start_sum_block(sbi), npages,
2127                                                         META_CP, true);
2128
2129                 /* restore for compacted data summary */
2130                 if (read_compacted_summaries(sbi))
2131                         return -EINVAL;
2132                 type = CURSEG_HOT_NODE;
2133         }
2134
2135         if (__exist_node_summaries(sbi))
2136                 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
2137                                         NR_CURSEG_TYPE - type, META_CP, true);
2138
2139         for (; type <= CURSEG_COLD_NODE; type++) {
2140                 err = read_normal_summaries(sbi, type);
2141                 if (err)
2142                         return err;
2143         }
2144
2145         return 0;
2146 }
2147
2148 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2149 {
2150         struct page *page;
2151         unsigned char *kaddr;
2152         struct f2fs_summary *summary;
2153         struct curseg_info *seg_i;
2154         int written_size = 0;
2155         int i, j;
2156
2157         page = grab_meta_page(sbi, blkaddr++);
2158         kaddr = (unsigned char *)page_address(page);
2159
2160         /* Step 1: write nat cache */
2161         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2162         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
2163         written_size += SUM_JOURNAL_SIZE;
2164
2165         /* Step 2: write sit cache */
2166         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2167         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
2168         written_size += SUM_JOURNAL_SIZE;
2169
2170         /* Step 3: write summary entries */
2171         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2172                 unsigned short blkoff;
2173                 seg_i = CURSEG_I(sbi, i);
2174                 if (sbi->ckpt->alloc_type[i] == SSR)
2175                         blkoff = sbi->blocks_per_seg;
2176                 else
2177                         blkoff = curseg_blkoff(sbi, i);
2178
2179                 for (j = 0; j < blkoff; j++) {
2180                         if (!page) {
2181                                 page = grab_meta_page(sbi, blkaddr++);
2182                                 kaddr = (unsigned char *)page_address(page);
2183                                 written_size = 0;
2184                         }
2185                         summary = (struct f2fs_summary *)(kaddr + written_size);
2186                         *summary = seg_i->sum_blk->entries[j];
2187                         written_size += SUMMARY_SIZE;
2188
2189                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
2190                                                         SUM_FOOTER_SIZE)
2191                                 continue;
2192
2193                         set_page_dirty(page);
2194                         f2fs_put_page(page, 1);
2195                         page = NULL;
2196                 }
2197         }
2198         if (page) {
2199                 set_page_dirty(page);
2200                 f2fs_put_page(page, 1);
2201         }
2202 }
2203
2204 static void write_normal_summaries(struct f2fs_sb_info *sbi,
2205                                         block_t blkaddr, int type)
2206 {
2207         int i, end;
2208         if (IS_DATASEG(type))
2209                 end = type + NR_CURSEG_DATA_TYPE;
2210         else
2211                 end = type + NR_CURSEG_NODE_TYPE;
2212
2213         for (i = type; i < end; i++)
2214                 write_current_sum_page(sbi, i, blkaddr + (i - type));
2215 }
2216
2217 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2218 {
2219         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
2220                 write_compacted_summaries(sbi, start_blk);
2221         else
2222                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2223 }
2224
2225 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2226 {
2227         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
2228 }
2229
2230 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
2231                                         unsigned int val, int alloc)
2232 {
2233         int i;
2234
2235         if (type == NAT_JOURNAL) {
2236                 for (i = 0; i < nats_in_cursum(journal); i++) {
2237                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
2238                                 return i;
2239                 }
2240                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2241                         return update_nats_in_cursum(journal, 1);
2242         } else if (type == SIT_JOURNAL) {
2243                 for (i = 0; i < sits_in_cursum(journal); i++)
2244                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
2245                                 return i;
2246                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2247                         return update_sits_in_cursum(journal, 1);
2248         }
2249         return -1;
2250 }
2251
2252 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2253                                         unsigned int segno)
2254 {
2255         return get_meta_page(sbi, current_sit_addr(sbi, segno));
2256 }
2257
2258 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2259                                         unsigned int start)
2260 {
2261         struct sit_info *sit_i = SIT_I(sbi);
2262         struct page *src_page, *dst_page;
2263         pgoff_t src_off, dst_off;
2264         void *src_addr, *dst_addr;
2265
2266         src_off = current_sit_addr(sbi, start);
2267         dst_off = next_sit_addr(sbi, src_off);
2268
2269         /* get current sit block page without lock */
2270         src_page = get_meta_page(sbi, src_off);
2271         dst_page = grab_meta_page(sbi, dst_off);
2272         f2fs_bug_on(sbi, PageDirty(src_page));
2273
2274         src_addr = page_address(src_page);
2275         dst_addr = page_address(dst_page);
2276         memcpy(dst_addr, src_addr, PAGE_SIZE);
2277
2278         set_page_dirty(dst_page);
2279         f2fs_put_page(src_page, 1);
2280
2281         set_to_next_sit(sit_i, start);
2282
2283         return dst_page;
2284 }
2285
2286 static struct sit_entry_set *grab_sit_entry_set(void)
2287 {
2288         struct sit_entry_set *ses =
2289                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
2290
2291         ses->entry_cnt = 0;
2292         INIT_LIST_HEAD(&ses->set_list);
2293         return ses;
2294 }
2295
2296 static void release_sit_entry_set(struct sit_entry_set *ses)
2297 {
2298         list_del(&ses->set_list);
2299         kmem_cache_free(sit_entry_set_slab, ses);
2300 }
2301
2302 static void adjust_sit_entry_set(struct sit_entry_set *ses,
2303                                                 struct list_head *head)
2304 {
2305         struct sit_entry_set *next = ses;
2306
2307         if (list_is_last(&ses->set_list, head))
2308                 return;
2309
2310         list_for_each_entry_continue(next, head, set_list)
2311                 if (ses->entry_cnt <= next->entry_cnt)
2312                         break;
2313
2314         list_move_tail(&ses->set_list, &next->set_list);
2315 }
2316
2317 static void add_sit_entry(unsigned int segno, struct list_head *head)
2318 {
2319         struct sit_entry_set *ses;
2320         unsigned int start_segno = START_SEGNO(segno);
2321
2322         list_for_each_entry(ses, head, set_list) {
2323                 if (ses->start_segno == start_segno) {
2324                         ses->entry_cnt++;
2325                         adjust_sit_entry_set(ses, head);
2326                         return;
2327                 }
2328         }
2329
2330         ses = grab_sit_entry_set();
2331
2332         ses->start_segno = start_segno;
2333         ses->entry_cnt++;
2334         list_add(&ses->set_list, head);
2335 }
2336
2337 static void add_sits_in_set(struct f2fs_sb_info *sbi)
2338 {
2339         struct f2fs_sm_info *sm_info = SM_I(sbi);
2340         struct list_head *set_list = &sm_info->sit_entry_set;
2341         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
2342         unsigned int segno;
2343
2344         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
2345                 add_sit_entry(segno, set_list);
2346 }
2347
2348 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
2349 {
2350         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2351         struct f2fs_journal *journal = curseg->journal;
2352         int i;
2353
2354         down_write(&curseg->journal_rwsem);
2355         for (i = 0; i < sits_in_cursum(journal); i++) {
2356                 unsigned int segno;
2357                 bool dirtied;
2358
2359                 segno = le32_to_cpu(segno_in_journal(journal, i));
2360                 dirtied = __mark_sit_entry_dirty(sbi, segno);
2361
2362                 if (!dirtied)
2363                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
2364         }
2365         update_sits_in_cursum(journal, -i);
2366         up_write(&curseg->journal_rwsem);
2367 }
2368
2369 /*
2370  * CP calls this function, which flushes SIT entries including sit_journal,
2371  * and moves prefree segs to free segs.
2372  */
2373 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2374 {
2375         struct sit_info *sit_i = SIT_I(sbi);
2376         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2377         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2378         struct f2fs_journal *journal = curseg->journal;
2379         struct sit_entry_set *ses, *tmp;
2380         struct list_head *head = &SM_I(sbi)->sit_entry_set;
2381         bool to_journal = true;
2382         struct seg_entry *se;
2383
2384         mutex_lock(&sit_i->sentry_lock);
2385
2386         if (!sit_i->dirty_sentries)
2387                 goto out;
2388
2389         /*
2390          * add and account sit entries of dirty bitmap in sit entry
2391          * set temporarily
2392          */
2393         add_sits_in_set(sbi);
2394
2395         /*
2396          * if there are no enough space in journal to store dirty sit
2397          * entries, remove all entries from journal and add and account
2398          * them in sit entry set.
2399          */
2400         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
2401                 remove_sits_in_journal(sbi);
2402
2403         /*
2404          * there are two steps to flush sit entries:
2405          * #1, flush sit entries to journal in current cold data summary block.
2406          * #2, flush sit entries to sit page.
2407          */
2408         list_for_each_entry_safe(ses, tmp, head, set_list) {
2409                 struct page *page = NULL;
2410                 struct f2fs_sit_block *raw_sit = NULL;
2411                 unsigned int start_segno = ses->start_segno;
2412                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
2413                                                 (unsigned long)MAIN_SEGS(sbi));
2414                 unsigned int segno = start_segno;
2415
2416                 if (to_journal &&
2417                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
2418                         to_journal = false;
2419
2420                 if (to_journal) {
2421                         down_write(&curseg->journal_rwsem);
2422                 } else {
2423                         page = get_next_sit_page(sbi, start_segno);
2424                         raw_sit = page_address(page);
2425                 }
2426
2427                 /* flush dirty sit entries in region of current sit set */
2428                 for_each_set_bit_from(segno, bitmap, end) {
2429                         int offset, sit_offset;
2430
2431                         se = get_seg_entry(sbi, segno);
2432
2433                         /* add discard candidates */
2434                         if (cpc->reason != CP_DISCARD) {
2435                                 cpc->trim_start = segno;
2436                                 add_discard_addrs(sbi, cpc, false);
2437                         }
2438
2439                         if (to_journal) {
2440                                 offset = lookup_journal_in_cursum(journal,
2441                                                         SIT_JOURNAL, segno, 1);
2442                                 f2fs_bug_on(sbi, offset < 0);
2443                                 segno_in_journal(journal, offset) =
2444                                                         cpu_to_le32(segno);
2445                                 seg_info_to_raw_sit(se,
2446                                         &sit_in_journal(journal, offset));
2447                         } else {
2448                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2449                                 seg_info_to_raw_sit(se,
2450                                                 &raw_sit->entries[sit_offset]);
2451                         }
2452
2453                         __clear_bit(segno, bitmap);
2454                         sit_i->dirty_sentries--;
2455                         ses->entry_cnt--;
2456                 }
2457
2458                 if (to_journal)
2459                         up_write(&curseg->journal_rwsem);
2460                 else
2461                         f2fs_put_page(page, 1);
2462
2463                 f2fs_bug_on(sbi, ses->entry_cnt);
2464                 release_sit_entry_set(ses);
2465         }
2466
2467         f2fs_bug_on(sbi, !list_empty(head));
2468         f2fs_bug_on(sbi, sit_i->dirty_sentries);
2469 out:
2470         if (cpc->reason == CP_DISCARD) {
2471                 __u64 trim_start = cpc->trim_start;
2472
2473                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2474                         add_discard_addrs(sbi, cpc, false);
2475
2476                 cpc->trim_start = trim_start;
2477         }
2478         mutex_unlock(&sit_i->sentry_lock);
2479
2480         set_prefree_as_free_segments(sbi);
2481 }
2482
2483 static int build_sit_info(struct f2fs_sb_info *sbi)
2484 {
2485         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2486         struct sit_info *sit_i;
2487         unsigned int sit_segs, start;
2488         char *src_bitmap;
2489         unsigned int bitmap_size;
2490
2491         /* allocate memory for SIT information */
2492         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2493         if (!sit_i)
2494                 return -ENOMEM;
2495
2496         SM_I(sbi)->sit_info = sit_i;
2497
2498         sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2499                                         sizeof(struct seg_entry), GFP_KERNEL);
2500         if (!sit_i->sentries)
2501                 return -ENOMEM;
2502
2503         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2504         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2505         if (!sit_i->dirty_sentries_bitmap)
2506                 return -ENOMEM;
2507
2508         for (start = 0; start < MAIN_SEGS(sbi); start++) {
2509                 sit_i->sentries[start].cur_valid_map
2510                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2511                 sit_i->sentries[start].ckpt_valid_map
2512                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2513                 if (!sit_i->sentries[start].cur_valid_map ||
2514                                 !sit_i->sentries[start].ckpt_valid_map)
2515                         return -ENOMEM;
2516
2517 #ifdef CONFIG_F2FS_CHECK_FS
2518                 sit_i->sentries[start].cur_valid_map_mir
2519                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2520                 if (!sit_i->sentries[start].cur_valid_map_mir)
2521                         return -ENOMEM;
2522 #endif
2523
2524                 if (f2fs_discard_en(sbi)) {
2525                         sit_i->sentries[start].discard_map
2526                                 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2527                         if (!sit_i->sentries[start].discard_map)
2528                                 return -ENOMEM;
2529                 }
2530         }
2531
2532         sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2533         if (!sit_i->tmp_map)
2534                 return -ENOMEM;
2535
2536         if (sbi->segs_per_sec > 1) {
2537                 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2538                                         sizeof(struct sec_entry), GFP_KERNEL);
2539                 if (!sit_i->sec_entries)
2540                         return -ENOMEM;
2541         }
2542
2543         /* get information related with SIT */
2544         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2545
2546         /* setup SIT bitmap from ckeckpoint pack */
2547         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2548         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2549
2550         sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2551         if (!sit_i->sit_bitmap)
2552                 return -ENOMEM;
2553
2554 #ifdef CONFIG_F2FS_CHECK_FS
2555         sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2556         if (!sit_i->sit_bitmap_mir)
2557                 return -ENOMEM;
2558 #endif
2559
2560         /* init SIT information */
2561         sit_i->s_ops = &default_salloc_ops;
2562
2563         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2564         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2565         sit_i->written_valid_blocks = 0;
2566         sit_i->bitmap_size = bitmap_size;
2567         sit_i->dirty_sentries = 0;
2568         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2569         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2570         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2571         mutex_init(&sit_i->sentry_lock);
2572         return 0;
2573 }
2574
2575 static int build_free_segmap(struct f2fs_sb_info *sbi)
2576 {
2577         struct free_segmap_info *free_i;
2578         unsigned int bitmap_size, sec_bitmap_size;
2579
2580         /* allocate memory for free segmap information */
2581         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2582         if (!free_i)
2583                 return -ENOMEM;
2584
2585         SM_I(sbi)->free_info = free_i;
2586
2587         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2588         free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2589         if (!free_i->free_segmap)
2590                 return -ENOMEM;
2591
2592         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2593         free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2594         if (!free_i->free_secmap)
2595                 return -ENOMEM;
2596
2597         /* set all segments as dirty temporarily */
2598         memset(free_i->free_segmap, 0xff, bitmap_size);
2599         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2600
2601         /* init free segmap information */
2602         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2603         free_i->free_segments = 0;
2604         free_i->free_sections = 0;
2605         spin_lock_init(&free_i->segmap_lock);
2606         return 0;
2607 }
2608
2609 static int build_curseg(struct f2fs_sb_info *sbi)
2610 {
2611         struct curseg_info *array;
2612         int i;
2613
2614         array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2615         if (!array)
2616                 return -ENOMEM;
2617
2618         SM_I(sbi)->curseg_array = array;
2619
2620         for (i = 0; i < NR_CURSEG_TYPE; i++) {
2621                 mutex_init(&array[i].curseg_mutex);
2622                 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2623                 if (!array[i].sum_blk)
2624                         return -ENOMEM;
2625                 init_rwsem(&array[i].journal_rwsem);
2626                 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2627                                                         GFP_KERNEL);
2628                 if (!array[i].journal)
2629                         return -ENOMEM;
2630                 array[i].segno = NULL_SEGNO;
2631                 array[i].next_blkoff = 0;
2632         }
2633         return restore_curseg_summaries(sbi);
2634 }
2635
2636 static void build_sit_entries(struct f2fs_sb_info *sbi)
2637 {
2638         struct sit_info *sit_i = SIT_I(sbi);
2639         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2640         struct f2fs_journal *journal = curseg->journal;
2641         struct seg_entry *se;
2642         struct f2fs_sit_entry sit;
2643         int sit_blk_cnt = SIT_BLK_CNT(sbi);
2644         unsigned int i, start, end;
2645         unsigned int readed, start_blk = 0;
2646
2647         do {
2648                 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
2649                                                         META_SIT, true);
2650
2651                 start = start_blk * sit_i->sents_per_block;
2652                 end = (start_blk + readed) * sit_i->sents_per_block;
2653
2654                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2655                         struct f2fs_sit_block *sit_blk;
2656                         struct page *page;
2657
2658                         se = &sit_i->sentries[start];
2659                         page = get_current_sit_page(sbi, start);
2660                         sit_blk = (struct f2fs_sit_block *)page_address(page);
2661                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2662                         f2fs_put_page(page, 1);
2663
2664                         check_block_count(sbi, start, &sit);
2665                         seg_info_from_raw_sit(se, &sit);
2666
2667                         /* build discard map only one time */
2668                         if (f2fs_discard_en(sbi)) {
2669                                 memcpy(se->discard_map, se->cur_valid_map,
2670                                                         SIT_VBLOCK_MAP_SIZE);
2671                                 sbi->discard_blks += sbi->blocks_per_seg -
2672                                                         se->valid_blocks;
2673                         }
2674
2675                         if (sbi->segs_per_sec > 1)
2676                                 get_sec_entry(sbi, start)->valid_blocks +=
2677                                                         se->valid_blocks;
2678                 }
2679                 start_blk += readed;
2680         } while (start_blk < sit_blk_cnt);
2681
2682         down_read(&curseg->journal_rwsem);
2683         for (i = 0; i < sits_in_cursum(journal); i++) {
2684                 unsigned int old_valid_blocks;
2685
2686                 start = le32_to_cpu(segno_in_journal(journal, i));
2687                 se = &sit_i->sentries[start];
2688                 sit = sit_in_journal(journal, i);
2689
2690                 old_valid_blocks = se->valid_blocks;
2691
2692                 check_block_count(sbi, start, &sit);
2693                 seg_info_from_raw_sit(se, &sit);
2694
2695                 if (f2fs_discard_en(sbi)) {
2696                         memcpy(se->discard_map, se->cur_valid_map,
2697                                                 SIT_VBLOCK_MAP_SIZE);
2698                         sbi->discard_blks += old_valid_blocks -
2699                                                 se->valid_blocks;
2700                 }
2701
2702                 if (sbi->segs_per_sec > 1)
2703                         get_sec_entry(sbi, start)->valid_blocks +=
2704                                 se->valid_blocks - old_valid_blocks;
2705         }
2706         up_read(&curseg->journal_rwsem);
2707 }
2708
2709 static void init_free_segmap(struct f2fs_sb_info *sbi)
2710 {
2711         unsigned int start;
2712         int type;
2713
2714         for (start = 0; start < MAIN_SEGS(sbi); start++) {
2715                 struct seg_entry *sentry = get_seg_entry(sbi, start);
2716                 if (!sentry->valid_blocks)
2717                         __set_free(sbi, start);
2718                 else
2719                         SIT_I(sbi)->written_valid_blocks +=
2720                                                 sentry->valid_blocks;
2721         }
2722
2723         /* set use the current segments */
2724         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2725                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2726                 __set_test_and_inuse(sbi, curseg_t->segno);
2727         }
2728 }
2729
2730 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2731 {
2732         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2733         struct free_segmap_info *free_i = FREE_I(sbi);
2734         unsigned int segno = 0, offset = 0;
2735         unsigned short valid_blocks;
2736
2737         while (1) {
2738                 /* find dirty segment based on free segmap */
2739                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2740                 if (segno >= MAIN_SEGS(sbi))
2741                         break;
2742                 offset = segno + 1;
2743                 valid_blocks = get_valid_blocks(sbi, segno, 0);
2744                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2745                         continue;
2746                 if (valid_blocks > sbi->blocks_per_seg) {
2747                         f2fs_bug_on(sbi, 1);
2748                         continue;
2749                 }
2750                 mutex_lock(&dirty_i->seglist_lock);
2751                 __locate_dirty_segment(sbi, segno, DIRTY);
2752                 mutex_unlock(&dirty_i->seglist_lock);
2753         }
2754 }
2755
2756 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2757 {
2758         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2759         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2760
2761         dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2762         if (!dirty_i->victim_secmap)
2763                 return -ENOMEM;
2764         return 0;
2765 }
2766
2767 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2768 {
2769         struct dirty_seglist_info *dirty_i;
2770         unsigned int bitmap_size, i;
2771
2772         /* allocate memory for dirty segments list information */
2773         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2774         if (!dirty_i)
2775                 return -ENOMEM;
2776
2777         SM_I(sbi)->dirty_info = dirty_i;
2778         mutex_init(&dirty_i->seglist_lock);
2779
2780         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2781
2782         for (i = 0; i < NR_DIRTY_TYPE; i++) {
2783                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2784                 if (!dirty_i->dirty_segmap[i])
2785                         return -ENOMEM;
2786         }
2787
2788         init_dirty_segmap(sbi);
2789         return init_victim_secmap(sbi);
2790 }
2791
2792 /*
2793  * Update min, max modified time for cost-benefit GC algorithm
2794  */
2795 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2796 {
2797         struct sit_info *sit_i = SIT_I(sbi);
2798         unsigned int segno;
2799
2800         mutex_lock(&sit_i->sentry_lock);
2801
2802         sit_i->min_mtime = LLONG_MAX;
2803
2804         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2805                 unsigned int i;
2806                 unsigned long long mtime = 0;
2807
2808                 for (i = 0; i < sbi->segs_per_sec; i++)
2809                         mtime += get_seg_entry(sbi, segno + i)->mtime;
2810
2811                 mtime = div_u64(mtime, sbi->segs_per_sec);
2812
2813                 if (sit_i->min_mtime > mtime)
2814                         sit_i->min_mtime = mtime;
2815         }
2816         sit_i->max_mtime = get_mtime(sbi);
2817         mutex_unlock(&sit_i->sentry_lock);
2818 }
2819
2820 int build_segment_manager(struct f2fs_sb_info *sbi)
2821 {
2822         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2823         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2824         struct f2fs_sm_info *sm_info;
2825         int err;
2826
2827         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2828         if (!sm_info)
2829                 return -ENOMEM;
2830
2831         /* init sm info */
2832         sbi->sm_info = sm_info;
2833         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2834         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2835         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2836         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2837         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2838         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2839         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2840         sm_info->rec_prefree_segments = sm_info->main_segments *
2841                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2842         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2843                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2844
2845         if (!test_opt(sbi, LFS))
2846                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2847         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2848         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2849
2850         sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2851
2852         INIT_LIST_HEAD(&sm_info->sit_entry_set);
2853
2854         if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2855                 err = create_flush_cmd_control(sbi);
2856                 if (err)
2857                         return err;
2858         }
2859
2860         err = create_discard_cmd_control(sbi);
2861         if (err)
2862                 return err;
2863
2864         err = build_sit_info(sbi);
2865         if (err)
2866                 return err;
2867         err = build_free_segmap(sbi);
2868         if (err)
2869                 return err;
2870         err = build_curseg(sbi);
2871         if (err)
2872                 return err;
2873
2874         /* reinit free segmap based on SIT */
2875         build_sit_entries(sbi);
2876
2877         init_free_segmap(sbi);
2878         err = build_dirty_segmap(sbi);
2879         if (err)
2880                 return err;
2881
2882         init_min_max_mtime(sbi);
2883         return 0;
2884 }
2885
2886 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2887                 enum dirty_type dirty_type)
2888 {
2889         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2890
2891         mutex_lock(&dirty_i->seglist_lock);
2892         kvfree(dirty_i->dirty_segmap[dirty_type]);
2893         dirty_i->nr_dirty[dirty_type] = 0;
2894         mutex_unlock(&dirty_i->seglist_lock);
2895 }
2896
2897 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2898 {
2899         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2900         kvfree(dirty_i->victim_secmap);
2901 }
2902
2903 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2904 {
2905         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2906         int i;
2907
2908         if (!dirty_i)
2909                 return;
2910
2911         /* discard pre-free/dirty segments list */
2912         for (i = 0; i < NR_DIRTY_TYPE; i++)
2913                 discard_dirty_segmap(sbi, i);
2914
2915         destroy_victim_secmap(sbi);
2916         SM_I(sbi)->dirty_info = NULL;
2917         kfree(dirty_i);
2918 }
2919
2920 static void destroy_curseg(struct f2fs_sb_info *sbi)
2921 {
2922         struct curseg_info *array = SM_I(sbi)->curseg_array;
2923         int i;
2924
2925         if (!array)
2926                 return;
2927         SM_I(sbi)->curseg_array = NULL;
2928         for (i = 0; i < NR_CURSEG_TYPE; i++) {
2929                 kfree(array[i].sum_blk);
2930                 kfree(array[i].journal);
2931         }
2932         kfree(array);
2933 }
2934
2935 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2936 {
2937         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2938         if (!free_i)
2939                 return;
2940         SM_I(sbi)->free_info = NULL;
2941         kvfree(free_i->free_segmap);
2942         kvfree(free_i->free_secmap);
2943         kfree(free_i);
2944 }
2945
2946 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2947 {
2948         struct sit_info *sit_i = SIT_I(sbi);
2949         unsigned int start;
2950
2951         if (!sit_i)
2952                 return;
2953
2954         if (sit_i->sentries) {
2955                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2956                         kfree(sit_i->sentries[start].cur_valid_map);
2957 #ifdef CONFIG_F2FS_CHECK_FS
2958                         kfree(sit_i->sentries[start].cur_valid_map_mir);
2959 #endif
2960                         kfree(sit_i->sentries[start].ckpt_valid_map);
2961                         kfree(sit_i->sentries[start].discard_map);
2962                 }
2963         }
2964         kfree(sit_i->tmp_map);
2965
2966         kvfree(sit_i->sentries);
2967         kvfree(sit_i->sec_entries);
2968         kvfree(sit_i->dirty_sentries_bitmap);
2969
2970         SM_I(sbi)->sit_info = NULL;
2971         kfree(sit_i->sit_bitmap);
2972 #ifdef CONFIG_F2FS_CHECK_FS
2973         kfree(sit_i->sit_bitmap_mir);
2974 #endif
2975         kfree(sit_i);
2976 }
2977
2978 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2979 {
2980         struct f2fs_sm_info *sm_info = SM_I(sbi);
2981
2982         if (!sm_info)
2983                 return;
2984         destroy_flush_cmd_control(sbi, true);
2985         destroy_discard_cmd_control(sbi, true);
2986         destroy_dirty_segmap(sbi);
2987         destroy_curseg(sbi);
2988         destroy_free_segmap(sbi);
2989         destroy_sit_info(sbi);
2990         sbi->sm_info = NULL;
2991         kfree(sm_info);
2992 }
2993
2994 int __init create_segment_manager_caches(void)
2995 {
2996         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2997                         sizeof(struct discard_entry));
2998         if (!discard_entry_slab)
2999                 goto fail;
3000
3001         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3002                         sizeof(struct discard_cmd));
3003         if (!discard_cmd_slab)
3004                 goto destroy_discard_entry;
3005
3006         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
3007                         sizeof(struct sit_entry_set));
3008         if (!sit_entry_set_slab)
3009                 goto destroy_discard_cmd;
3010
3011         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3012                         sizeof(struct inmem_pages));
3013         if (!inmem_entry_slab)
3014                 goto destroy_sit_entry_set;
3015         return 0;
3016
3017 destroy_sit_entry_set:
3018         kmem_cache_destroy(sit_entry_set_slab);
3019 destroy_discard_cmd:
3020         kmem_cache_destroy(discard_cmd_slab);
3021 destroy_discard_entry:
3022         kmem_cache_destroy(discard_entry_slab);
3023 fail:
3024         return -ENOMEM;
3025 }
3026
3027 void destroy_segment_manager_caches(void)
3028 {
3029         kmem_cache_destroy(sit_entry_set_slab);
3030         kmem_cache_destroy(discard_cmd_slab);
3031         kmem_cache_destroy(discard_entry_slab);
3032         kmem_cache_destroy(inmem_entry_slab);
3033 }