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         }
1040         mutex_unlock(&dirty_i->seglist_lock);
1041
1042         /* send small discards */
1043         list_for_each_entry_safe(entry, this, head, list) {
1044                 if (force && entry->len < cpc->trim_minlen)
1045                         goto skip;
1046                 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
1047                 cpc->trimmed += entry->len;
1048 skip:
1049                 list_del(&entry->list);
1050                 SM_I(sbi)->dcc_info->nr_discards -= entry->len;
1051                 kmem_cache_free(discard_entry_slab, entry);
1052         }
1053 }
1054
1055 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1056 {
1057         dev_t dev = sbi->sb->s_bdev->bd_dev;
1058         struct discard_cmd_control *dcc;
1059         int err = 0;
1060
1061         if (SM_I(sbi)->dcc_info) {
1062                 dcc = SM_I(sbi)->dcc_info;
1063                 goto init_thread;
1064         }
1065
1066         dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1067         if (!dcc)
1068                 return -ENOMEM;
1069
1070         INIT_LIST_HEAD(&dcc->discard_entry_list);
1071         INIT_LIST_HEAD(&dcc->discard_cmd_list);
1072         mutex_init(&dcc->cmd_lock);
1073         atomic_set(&dcc->submit_discard, 0);
1074         dcc->nr_discards = 0;
1075         dcc->max_discards = 0;
1076
1077         init_waitqueue_head(&dcc->discard_wait_queue);
1078         SM_I(sbi)->dcc_info = dcc;
1079 init_thread:
1080         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1081                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1082         if (IS_ERR(dcc->f2fs_issue_discard)) {
1083                 err = PTR_ERR(dcc->f2fs_issue_discard);
1084                 kfree(dcc);
1085                 SM_I(sbi)->dcc_info = NULL;
1086                 return err;
1087         }
1088
1089         return err;
1090 }
1091
1092 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi, bool free)
1093 {
1094         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1095
1096         if (dcc && dcc->f2fs_issue_discard) {
1097                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1098
1099                 dcc->f2fs_issue_discard = NULL;
1100                 kthread_stop(discard_thread);
1101         }
1102         if (free) {
1103                 kfree(dcc);
1104                 SM_I(sbi)->dcc_info = NULL;
1105         }
1106 }
1107
1108 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1109 {
1110         struct sit_info *sit_i = SIT_I(sbi);
1111
1112         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1113                 sit_i->dirty_sentries++;
1114                 return false;
1115         }
1116
1117         return true;
1118 }
1119
1120 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1121                                         unsigned int segno, int modified)
1122 {
1123         struct seg_entry *se = get_seg_entry(sbi, segno);
1124         se->type = type;
1125         if (modified)
1126                 __mark_sit_entry_dirty(sbi, segno);
1127 }
1128
1129 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1130 {
1131         struct seg_entry *se;
1132         unsigned int segno, offset;
1133         long int new_vblocks;
1134
1135         segno = GET_SEGNO(sbi, blkaddr);
1136
1137         se = get_seg_entry(sbi, segno);
1138         new_vblocks = se->valid_blocks + del;
1139         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1140
1141         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
1142                                 (new_vblocks > sbi->blocks_per_seg)));
1143
1144         se->valid_blocks = new_vblocks;
1145         se->mtime = get_mtime(sbi);
1146         SIT_I(sbi)->max_mtime = se->mtime;
1147
1148         /* Update valid block bitmap */
1149         if (del > 0) {
1150                 if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) {
1151 #ifdef CONFIG_F2FS_CHECK_FS
1152                         if (f2fs_test_and_set_bit(offset,
1153                                                 se->cur_valid_map_mir))
1154                                 f2fs_bug_on(sbi, 1);
1155                         else
1156                                 WARN_ON(1);
1157 #else
1158                         f2fs_bug_on(sbi, 1);
1159 #endif
1160                 }
1161                 if (f2fs_discard_en(sbi) &&
1162                         !f2fs_test_and_set_bit(offset, se->discard_map))
1163                         sbi->discard_blks--;
1164         } else {
1165                 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) {
1166 #ifdef CONFIG_F2FS_CHECK_FS
1167                         if (!f2fs_test_and_clear_bit(offset,
1168                                                 se->cur_valid_map_mir))
1169                                 f2fs_bug_on(sbi, 1);
1170                         else
1171                                 WARN_ON(1);
1172 #else
1173                         f2fs_bug_on(sbi, 1);
1174 #endif
1175                 }
1176                 if (f2fs_discard_en(sbi) &&
1177                         f2fs_test_and_clear_bit(offset, se->discard_map))
1178                         sbi->discard_blks++;
1179         }
1180         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1181                 se->ckpt_valid_blocks += del;
1182
1183         __mark_sit_entry_dirty(sbi, segno);
1184
1185         /* update total number of valid blocks to be written in ckpt area */
1186         SIT_I(sbi)->written_valid_blocks += del;
1187
1188         if (sbi->segs_per_sec > 1)
1189                 get_sec_entry(sbi, segno)->valid_blocks += del;
1190 }
1191
1192 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
1193 {
1194         update_sit_entry(sbi, new, 1);
1195         if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1196                 update_sit_entry(sbi, old, -1);
1197
1198         locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1199         locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
1200 }
1201
1202 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1203 {
1204         unsigned int segno = GET_SEGNO(sbi, addr);
1205         struct sit_info *sit_i = SIT_I(sbi);
1206
1207         f2fs_bug_on(sbi, addr == NULL_ADDR);
1208         if (addr == NEW_ADDR)
1209                 return;
1210
1211         /* add it into sit main buffer */
1212         mutex_lock(&sit_i->sentry_lock);
1213
1214         update_sit_entry(sbi, addr, -1);
1215
1216         /* add it into dirty seglist */
1217         locate_dirty_segment(sbi, segno);
1218
1219         mutex_unlock(&sit_i->sentry_lock);
1220 }
1221
1222 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1223 {
1224         struct sit_info *sit_i = SIT_I(sbi);
1225         unsigned int segno, offset;
1226         struct seg_entry *se;
1227         bool is_cp = false;
1228
1229         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1230                 return true;
1231
1232         mutex_lock(&sit_i->sentry_lock);
1233
1234         segno = GET_SEGNO(sbi, blkaddr);
1235         se = get_seg_entry(sbi, segno);
1236         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1237
1238         if (f2fs_test_bit(offset, se->ckpt_valid_map))
1239                 is_cp = true;
1240
1241         mutex_unlock(&sit_i->sentry_lock);
1242
1243         return is_cp;
1244 }
1245
1246 /*
1247  * This function should be resided under the curseg_mutex lock
1248  */
1249 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1250                                         struct f2fs_summary *sum)
1251 {
1252         struct curseg_info *curseg = CURSEG_I(sbi, type);
1253         void *addr = curseg->sum_blk;
1254         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1255         memcpy(addr, sum, sizeof(struct f2fs_summary));
1256 }
1257
1258 /*
1259  * Calculate the number of current summary pages for writing
1260  */
1261 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1262 {
1263         int valid_sum_count = 0;
1264         int i, sum_in_page;
1265
1266         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1267                 if (sbi->ckpt->alloc_type[i] == SSR)
1268                         valid_sum_count += sbi->blocks_per_seg;
1269                 else {
1270                         if (for_ra)
1271                                 valid_sum_count += le16_to_cpu(
1272                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1273                         else
1274                                 valid_sum_count += curseg_blkoff(sbi, i);
1275                 }
1276         }
1277
1278         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1279                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1280         if (valid_sum_count <= sum_in_page)
1281                 return 1;
1282         else if ((valid_sum_count - sum_in_page) <=
1283                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1284                 return 2;
1285         return 3;
1286 }
1287
1288 /*
1289  * Caller should put this summary page
1290  */
1291 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1292 {
1293         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1294 }
1295
1296 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1297 {
1298         struct page *page = grab_meta_page(sbi, blk_addr);
1299         void *dst = page_address(page);
1300
1301         if (src)
1302                 memcpy(dst, src, PAGE_SIZE);
1303         else
1304                 memset(dst, 0, PAGE_SIZE);
1305         set_page_dirty(page);
1306         f2fs_put_page(page, 1);
1307 }
1308
1309 static void write_sum_page(struct f2fs_sb_info *sbi,
1310                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
1311 {
1312         update_meta_page(sbi, (void *)sum_blk, blk_addr);
1313 }
1314
1315 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1316                                                 int type, block_t blk_addr)
1317 {
1318         struct curseg_info *curseg = CURSEG_I(sbi, type);
1319         struct page *page = grab_meta_page(sbi, blk_addr);
1320         struct f2fs_summary_block *src = curseg->sum_blk;
1321         struct f2fs_summary_block *dst;
1322
1323         dst = (struct f2fs_summary_block *)page_address(page);
1324
1325         mutex_lock(&curseg->curseg_mutex);
1326
1327         down_read(&curseg->journal_rwsem);
1328         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1329         up_read(&curseg->journal_rwsem);
1330
1331         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1332         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1333
1334         mutex_unlock(&curseg->curseg_mutex);
1335
1336         set_page_dirty(page);
1337         f2fs_put_page(page, 1);
1338 }
1339
1340 /*
1341  * Find a new segment from the free segments bitmap to right order
1342  * This function should be returned with success, otherwise BUG
1343  */
1344 static void get_new_segment(struct f2fs_sb_info *sbi,
1345                         unsigned int *newseg, bool new_sec, int dir)
1346 {
1347         struct free_segmap_info *free_i = FREE_I(sbi);
1348         unsigned int segno, secno, zoneno;
1349         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
1350         unsigned int hint = *newseg / sbi->segs_per_sec;
1351         unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1352         unsigned int left_start = hint;
1353         bool init = true;
1354         int go_left = 0;
1355         int i;
1356
1357         spin_lock(&free_i->segmap_lock);
1358
1359         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1360                 segno = find_next_zero_bit(free_i->free_segmap,
1361                                 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1362                 if (segno < (hint + 1) * sbi->segs_per_sec)
1363                         goto got_it;
1364         }
1365 find_other_zone:
1366         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1367         if (secno >= MAIN_SECS(sbi)) {
1368                 if (dir == ALLOC_RIGHT) {
1369                         secno = find_next_zero_bit(free_i->free_secmap,
1370                                                         MAIN_SECS(sbi), 0);
1371                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1372                 } else {
1373                         go_left = 1;
1374                         left_start = hint - 1;
1375                 }
1376         }
1377         if (go_left == 0)
1378                 goto skip_left;
1379
1380         while (test_bit(left_start, free_i->free_secmap)) {
1381                 if (left_start > 0) {
1382                         left_start--;
1383                         continue;
1384                 }
1385                 left_start = find_next_zero_bit(free_i->free_secmap,
1386                                                         MAIN_SECS(sbi), 0);
1387                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1388                 break;
1389         }
1390         secno = left_start;
1391 skip_left:
1392         hint = secno;
1393         segno = secno * sbi->segs_per_sec;
1394         zoneno = secno / sbi->secs_per_zone;
1395
1396         /* give up on finding another zone */
1397         if (!init)
1398                 goto got_it;
1399         if (sbi->secs_per_zone == 1)
1400                 goto got_it;
1401         if (zoneno == old_zoneno)
1402                 goto got_it;
1403         if (dir == ALLOC_LEFT) {
1404                 if (!go_left && zoneno + 1 >= total_zones)
1405                         goto got_it;
1406                 if (go_left && zoneno == 0)
1407                         goto got_it;
1408         }
1409         for (i = 0; i < NR_CURSEG_TYPE; i++)
1410                 if (CURSEG_I(sbi, i)->zone == zoneno)
1411                         break;
1412
1413         if (i < NR_CURSEG_TYPE) {
1414                 /* zone is in user, try another */
1415                 if (go_left)
1416                         hint = zoneno * sbi->secs_per_zone - 1;
1417                 else if (zoneno + 1 >= total_zones)
1418                         hint = 0;
1419                 else
1420                         hint = (zoneno + 1) * sbi->secs_per_zone;
1421                 init = false;
1422                 goto find_other_zone;
1423         }
1424 got_it:
1425         /* set it as dirty segment in free segmap */
1426         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1427         __set_inuse(sbi, segno);
1428         *newseg = segno;
1429         spin_unlock(&free_i->segmap_lock);
1430 }
1431
1432 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1433 {
1434         struct curseg_info *curseg = CURSEG_I(sbi, type);
1435         struct summary_footer *sum_footer;
1436
1437         curseg->segno = curseg->next_segno;
1438         curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1439         curseg->next_blkoff = 0;
1440         curseg->next_segno = NULL_SEGNO;
1441
1442         sum_footer = &(curseg->sum_blk->footer);
1443         memset(sum_footer, 0, sizeof(struct summary_footer));
1444         if (IS_DATASEG(type))
1445                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1446         if (IS_NODESEG(type))
1447                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1448         __set_sit_entry_type(sbi, type, curseg->segno, modified);
1449 }
1450
1451 /*
1452  * Allocate a current working segment.
1453  * This function always allocates a free segment in LFS manner.
1454  */
1455 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1456 {
1457         struct curseg_info *curseg = CURSEG_I(sbi, type);
1458         unsigned int segno = curseg->segno;
1459         int dir = ALLOC_LEFT;
1460
1461         write_sum_page(sbi, curseg->sum_blk,
1462                                 GET_SUM_BLOCK(sbi, segno));
1463         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1464                 dir = ALLOC_RIGHT;
1465
1466         if (test_opt(sbi, NOHEAP))
1467                 dir = ALLOC_RIGHT;
1468
1469         get_new_segment(sbi, &segno, new_sec, dir);
1470         curseg->next_segno = segno;
1471         reset_curseg(sbi, type, 1);
1472         curseg->alloc_type = LFS;
1473 }
1474
1475 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1476                         struct curseg_info *seg, block_t start)
1477 {
1478         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1479         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1480         unsigned long *target_map = SIT_I(sbi)->tmp_map;
1481         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1482         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1483         int i, pos;
1484
1485         for (i = 0; i < entries; i++)
1486                 target_map[i] = ckpt_map[i] | cur_map[i];
1487
1488         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1489
1490         seg->next_blkoff = pos;
1491 }
1492
1493 /*
1494  * If a segment is written by LFS manner, next block offset is just obtained
1495  * by increasing the current block offset. However, if a segment is written by
1496  * SSR manner, next block offset obtained by calling __next_free_blkoff
1497  */
1498 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1499                                 struct curseg_info *seg)
1500 {
1501         if (seg->alloc_type == SSR)
1502                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1503         else
1504                 seg->next_blkoff++;
1505 }
1506
1507 /*
1508  * This function always allocates a used segment(from dirty seglist) by SSR
1509  * manner, so it should recover the existing segment information of valid blocks
1510  */
1511 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1512 {
1513         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1514         struct curseg_info *curseg = CURSEG_I(sbi, type);
1515         unsigned int new_segno = curseg->next_segno;
1516         struct f2fs_summary_block *sum_node;
1517         struct page *sum_page;
1518
1519         write_sum_page(sbi, curseg->sum_blk,
1520                                 GET_SUM_BLOCK(sbi, curseg->segno));
1521         __set_test_and_inuse(sbi, new_segno);
1522
1523         mutex_lock(&dirty_i->seglist_lock);
1524         __remove_dirty_segment(sbi, new_segno, PRE);
1525         __remove_dirty_segment(sbi, new_segno, DIRTY);
1526         mutex_unlock(&dirty_i->seglist_lock);
1527
1528         reset_curseg(sbi, type, 1);
1529         curseg->alloc_type = SSR;
1530         __next_free_blkoff(sbi, curseg, 0);
1531
1532         if (reuse) {
1533                 sum_page = get_sum_page(sbi, new_segno);
1534                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1535                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1536                 f2fs_put_page(sum_page, 1);
1537         }
1538 }
1539
1540 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1541 {
1542         struct curseg_info *curseg = CURSEG_I(sbi, type);
1543         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1544         int i, n;
1545
1546         /* need_SSR() already forces to do this */
1547         if (v_ops->get_victim(sbi, &(curseg)->next_segno, BG_GC, type, SSR))
1548                 return 1;
1549
1550         /* For node segments, let's do SSR more intensively */
1551         if (IS_NODESEG(type)) {
1552                 i = CURSEG_HOT_NODE;
1553                 n = CURSEG_COLD_NODE;
1554         } else {
1555                 i = CURSEG_HOT_DATA;
1556                 n = CURSEG_COLD_DATA;
1557         }
1558
1559         for (; i <= n; i++) {
1560                 if (i == type)
1561                         continue;
1562                 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1563                                                 BG_GC, i, SSR))
1564                         return 1;
1565         }
1566         return 0;
1567 }
1568
1569 /*
1570  * flush out current segment and replace it with new segment
1571  * This function should be returned with success, otherwise BUG
1572  */
1573 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1574                                                 int type, bool force)
1575 {
1576         if (force)
1577                 new_curseg(sbi, type, true);
1578         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
1579                                         type == CURSEG_WARM_NODE)
1580                 new_curseg(sbi, type, false);
1581         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1582                 change_curseg(sbi, type, true);
1583         else
1584                 new_curseg(sbi, type, false);
1585
1586         stat_inc_seg_type(sbi, CURSEG_I(sbi, type));
1587 }
1588
1589 void allocate_new_segments(struct f2fs_sb_info *sbi)
1590 {
1591         struct curseg_info *curseg;
1592         unsigned int old_segno;
1593         int i;
1594
1595         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1596                 curseg = CURSEG_I(sbi, i);
1597                 old_segno = curseg->segno;
1598                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
1599                 locate_dirty_segment(sbi, old_segno);
1600         }
1601 }
1602
1603 static const struct segment_allocation default_salloc_ops = {
1604         .allocate_segment = allocate_segment_by_default,
1605 };
1606
1607 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1608 {
1609         __u64 trim_start = cpc->trim_start;
1610         bool has_candidate = false;
1611
1612         mutex_lock(&SIT_I(sbi)->sentry_lock);
1613         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
1614                 if (add_discard_addrs(sbi, cpc, true)) {
1615                         has_candidate = true;
1616                         break;
1617                 }
1618         }
1619         mutex_unlock(&SIT_I(sbi)->sentry_lock);
1620
1621         cpc->trim_start = trim_start;
1622         return has_candidate;
1623 }
1624
1625 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1626 {
1627         __u64 start = F2FS_BYTES_TO_BLK(range->start);
1628         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1629         unsigned int start_segno, end_segno;
1630         struct cp_control cpc;
1631         int err = 0;
1632
1633         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1634                 return -EINVAL;
1635
1636         cpc.trimmed = 0;
1637         if (end <= MAIN_BLKADDR(sbi))
1638                 goto out;
1639
1640         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1641                 f2fs_msg(sbi->sb, KERN_WARNING,
1642                         "Found FS corruption, run fsck to fix.");
1643                 goto out;
1644         }
1645
1646         /* start/end segment number in main_area */
1647         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1648         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1649                                                 GET_SEGNO(sbi, end);
1650         cpc.reason = CP_DISCARD;
1651         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1652
1653         /* do checkpoint to issue discard commands safely */
1654         for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1655                 cpc.trim_start = start_segno;
1656
1657                 if (sbi->discard_blks == 0)
1658                         break;
1659                 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1660                         cpc.trim_end = end_segno;
1661                 else
1662                         cpc.trim_end = min_t(unsigned int,
1663                                 rounddown(start_segno +
1664                                 BATCHED_TRIM_SEGMENTS(sbi),
1665                                 sbi->segs_per_sec) - 1, end_segno);
1666
1667                 mutex_lock(&sbi->gc_mutex);
1668                 err = write_checkpoint(sbi, &cpc);
1669                 mutex_unlock(&sbi->gc_mutex);
1670                 if (err)
1671                         break;
1672
1673                 schedule();
1674         }
1675 out:
1676         range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1677         return err;
1678 }
1679
1680 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1681 {
1682         struct curseg_info *curseg = CURSEG_I(sbi, type);
1683         if (curseg->next_blkoff < sbi->blocks_per_seg)
1684                 return true;
1685         return false;
1686 }
1687
1688 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1689 {
1690         if (p_type == DATA)
1691                 return CURSEG_HOT_DATA;
1692         else
1693                 return CURSEG_HOT_NODE;
1694 }
1695
1696 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1697 {
1698         if (p_type == DATA) {
1699                 struct inode *inode = page->mapping->host;
1700
1701                 if (S_ISDIR(inode->i_mode))
1702                         return CURSEG_HOT_DATA;
1703                 else
1704                         return CURSEG_COLD_DATA;
1705         } else {
1706                 if (IS_DNODE(page) && is_cold_node(page))
1707                         return CURSEG_WARM_NODE;
1708                 else
1709                         return CURSEG_COLD_NODE;
1710         }
1711 }
1712
1713 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1714 {
1715         if (p_type == DATA) {
1716                 struct inode *inode = page->mapping->host;
1717
1718                 if (S_ISDIR(inode->i_mode))
1719                         return CURSEG_HOT_DATA;
1720                 else if (is_cold_data(page) || file_is_cold(inode))
1721                         return CURSEG_COLD_DATA;
1722                 else
1723                         return CURSEG_WARM_DATA;
1724         } else {
1725                 if (IS_DNODE(page))
1726                         return is_cold_node(page) ? CURSEG_WARM_NODE :
1727                                                 CURSEG_HOT_NODE;
1728                 else
1729                         return CURSEG_COLD_NODE;
1730         }
1731 }
1732
1733 static int __get_segment_type(struct page *page, enum page_type p_type)
1734 {
1735         switch (F2FS_P_SB(page)->active_logs) {
1736         case 2:
1737                 return __get_segment_type_2(page, p_type);
1738         case 4:
1739                 return __get_segment_type_4(page, p_type);
1740         }
1741         /* NR_CURSEG_TYPE(6) logs by default */
1742         f2fs_bug_on(F2FS_P_SB(page),
1743                 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1744         return __get_segment_type_6(page, p_type);
1745 }
1746
1747 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1748                 block_t old_blkaddr, block_t *new_blkaddr,
1749                 struct f2fs_summary *sum, int type)
1750 {
1751         struct sit_info *sit_i = SIT_I(sbi);
1752         struct curseg_info *curseg = CURSEG_I(sbi, type);
1753
1754         mutex_lock(&curseg->curseg_mutex);
1755         mutex_lock(&sit_i->sentry_lock);
1756
1757         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1758
1759         f2fs_wait_discard_bio(sbi, *new_blkaddr);
1760
1761         /*
1762          * __add_sum_entry should be resided under the curseg_mutex
1763          * because, this function updates a summary entry in the
1764          * current summary block.
1765          */
1766         __add_sum_entry(sbi, type, sum);
1767
1768         __refresh_next_blkoff(sbi, curseg);
1769
1770         stat_inc_block_count(sbi, curseg);
1771
1772         /*
1773          * SIT information should be updated before segment allocation,
1774          * since SSR needs latest valid block information.
1775          */
1776         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1777
1778         if (!__has_curseg_space(sbi, type))
1779                 sit_i->s_ops->allocate_segment(sbi, type, false);
1780
1781         mutex_unlock(&sit_i->sentry_lock);
1782
1783         if (page && IS_NODESEG(type))
1784                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1785
1786         mutex_unlock(&curseg->curseg_mutex);
1787 }
1788
1789 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1790 {
1791         int type = __get_segment_type(fio->page, fio->type);
1792         int err;
1793
1794         if (fio->type == NODE || fio->type == DATA)
1795                 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1796 reallocate:
1797         allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1798                                         &fio->new_blkaddr, sum, type);
1799
1800         /* writeout dirty page into bdev */
1801         err = f2fs_submit_page_mbio(fio);
1802         if (err == -EAGAIN) {
1803                 fio->old_blkaddr = fio->new_blkaddr;
1804                 goto reallocate;
1805         }
1806
1807         if (fio->type == NODE || fio->type == DATA)
1808                 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
1809 }
1810
1811 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1812 {
1813         struct f2fs_io_info fio = {
1814                 .sbi = sbi,
1815                 .type = META,
1816                 .op = REQ_OP_WRITE,
1817                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
1818                 .old_blkaddr = page->index,
1819                 .new_blkaddr = page->index,
1820                 .page = page,
1821                 .encrypted_page = NULL,
1822         };
1823
1824         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1825                 fio.op_flags &= ~REQ_META;
1826
1827         set_page_writeback(page);
1828         f2fs_submit_page_mbio(&fio);
1829 }
1830
1831 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1832 {
1833         struct f2fs_summary sum;
1834
1835         set_summary(&sum, nid, 0, 0);
1836         do_write_page(&sum, fio);
1837 }
1838
1839 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1840 {
1841         struct f2fs_sb_info *sbi = fio->sbi;
1842         struct f2fs_summary sum;
1843         struct node_info ni;
1844
1845         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1846         get_node_info(sbi, dn->nid, &ni);
1847         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1848         do_write_page(&sum, fio);
1849         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
1850 }
1851
1852 void rewrite_data_page(struct f2fs_io_info *fio)
1853 {
1854         fio->new_blkaddr = fio->old_blkaddr;
1855         stat_inc_inplace_blocks(fio->sbi);
1856         f2fs_submit_page_mbio(fio);
1857 }
1858
1859 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1860                                 block_t old_blkaddr, block_t new_blkaddr,
1861                                 bool recover_curseg, bool recover_newaddr)
1862 {
1863         struct sit_info *sit_i = SIT_I(sbi);
1864         struct curseg_info *curseg;
1865         unsigned int segno, old_cursegno;
1866         struct seg_entry *se;
1867         int type;
1868         unsigned short old_blkoff;
1869
1870         segno = GET_SEGNO(sbi, new_blkaddr);
1871         se = get_seg_entry(sbi, segno);
1872         type = se->type;
1873
1874         if (!recover_curseg) {
1875                 /* for recovery flow */
1876                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1877                         if (old_blkaddr == NULL_ADDR)
1878                                 type = CURSEG_COLD_DATA;
1879                         else
1880                                 type = CURSEG_WARM_DATA;
1881                 }
1882         } else {
1883                 if (!IS_CURSEG(sbi, segno))
1884                         type = CURSEG_WARM_DATA;
1885         }
1886
1887         curseg = CURSEG_I(sbi, type);
1888
1889         mutex_lock(&curseg->curseg_mutex);
1890         mutex_lock(&sit_i->sentry_lock);
1891
1892         old_cursegno = curseg->segno;
1893         old_blkoff = curseg->next_blkoff;
1894
1895         /* change the current segment */
1896         if (segno != curseg->segno) {
1897                 curseg->next_segno = segno;
1898                 change_curseg(sbi, type, true);
1899         }
1900
1901         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1902         __add_sum_entry(sbi, type, sum);
1903
1904         if (!recover_curseg || recover_newaddr)
1905                 update_sit_entry(sbi, new_blkaddr, 1);
1906         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1907                 update_sit_entry(sbi, old_blkaddr, -1);
1908
1909         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1910         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1911
1912         locate_dirty_segment(sbi, old_cursegno);
1913
1914         if (recover_curseg) {
1915                 if (old_cursegno != curseg->segno) {
1916                         curseg->next_segno = old_cursegno;
1917                         change_curseg(sbi, type, true);
1918                 }
1919                 curseg->next_blkoff = old_blkoff;
1920         }
1921
1922         mutex_unlock(&sit_i->sentry_lock);
1923         mutex_unlock(&curseg->curseg_mutex);
1924 }
1925
1926 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1927                                 block_t old_addr, block_t new_addr,
1928                                 unsigned char version, bool recover_curseg,
1929                                 bool recover_newaddr)
1930 {
1931         struct f2fs_summary sum;
1932
1933         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1934
1935         __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1936                                         recover_curseg, recover_newaddr);
1937
1938         f2fs_update_data_blkaddr(dn, new_addr);
1939 }
1940
1941 void f2fs_wait_on_page_writeback(struct page *page,
1942                                 enum page_type type, bool ordered)
1943 {
1944         if (PageWriteback(page)) {
1945                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1946
1947                 f2fs_submit_merged_bio_cond(sbi, page->mapping->host,
1948                                                 0, page->index, type, WRITE);
1949                 if (ordered)
1950                         wait_on_page_writeback(page);
1951                 else
1952                         wait_for_stable_page(page);
1953         }
1954 }
1955
1956 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1957                                                         block_t blkaddr)
1958 {
1959         struct page *cpage;
1960
1961         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1962                 return;
1963
1964         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1965         if (cpage) {
1966                 f2fs_wait_on_page_writeback(cpage, DATA, true);
1967                 f2fs_put_page(cpage, 1);
1968         }
1969 }
1970
1971 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1972 {
1973         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1974         struct curseg_info *seg_i;
1975         unsigned char *kaddr;
1976         struct page *page;
1977         block_t start;
1978         int i, j, offset;
1979
1980         start = start_sum_block(sbi);
1981
1982         page = get_meta_page(sbi, start++);
1983         kaddr = (unsigned char *)page_address(page);
1984
1985         /* Step 1: restore nat cache */
1986         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1987         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
1988
1989         /* Step 2: restore sit cache */
1990         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1991         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
1992         offset = 2 * SUM_JOURNAL_SIZE;
1993
1994         /* Step 3: restore summary entries */
1995         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1996                 unsigned short blk_off;
1997                 unsigned int segno;
1998
1999                 seg_i = CURSEG_I(sbi, i);
2000                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2001                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2002                 seg_i->next_segno = segno;
2003                 reset_curseg(sbi, i, 0);
2004                 seg_i->alloc_type = ckpt->alloc_type[i];
2005                 seg_i->next_blkoff = blk_off;
2006
2007                 if (seg_i->alloc_type == SSR)
2008                         blk_off = sbi->blocks_per_seg;
2009
2010                 for (j = 0; j < blk_off; j++) {
2011                         struct f2fs_summary *s;
2012                         s = (struct f2fs_summary *)(kaddr + offset);
2013                         seg_i->sum_blk->entries[j] = *s;
2014                         offset += SUMMARY_SIZE;
2015                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
2016                                                 SUM_FOOTER_SIZE)
2017                                 continue;
2018
2019                         f2fs_put_page(page, 1);
2020                         page = NULL;
2021
2022                         page = get_meta_page(sbi, start++);
2023                         kaddr = (unsigned char *)page_address(page);
2024                         offset = 0;
2025                 }
2026         }
2027         f2fs_put_page(page, 1);
2028         return 0;
2029 }
2030
2031 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2032 {
2033         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2034         struct f2fs_summary_block *sum;
2035         struct curseg_info *curseg;
2036         struct page *new;
2037         unsigned short blk_off;
2038         unsigned int segno = 0;
2039         block_t blk_addr = 0;
2040
2041         /* get segment number and block addr */
2042         if (IS_DATASEG(type)) {
2043                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2044                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2045                                                         CURSEG_HOT_DATA]);
2046                 if (__exist_node_summaries(sbi))
2047                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2048                 else
2049                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2050         } else {
2051                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2052                                                         CURSEG_HOT_NODE]);
2053                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2054                                                         CURSEG_HOT_NODE]);
2055                 if (__exist_node_summaries(sbi))
2056                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2057                                                         type - CURSEG_HOT_NODE);
2058                 else
2059                         blk_addr = GET_SUM_BLOCK(sbi, segno);
2060         }
2061
2062         new = get_meta_page(sbi, blk_addr);
2063         sum = (struct f2fs_summary_block *)page_address(new);
2064
2065         if (IS_NODESEG(type)) {
2066                 if (__exist_node_summaries(sbi)) {
2067                         struct f2fs_summary *ns = &sum->entries[0];
2068                         int i;
2069                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2070                                 ns->version = 0;
2071                                 ns->ofs_in_node = 0;
2072                         }
2073                 } else {
2074                         int err;
2075
2076                         err = restore_node_summary(sbi, segno, sum);
2077                         if (err) {
2078                                 f2fs_put_page(new, 1);
2079                                 return err;
2080                         }
2081                 }
2082         }
2083
2084         /* set uncompleted segment to curseg */
2085         curseg = CURSEG_I(sbi, type);
2086         mutex_lock(&curseg->curseg_mutex);
2087
2088         /* update journal info */
2089         down_write(&curseg->journal_rwsem);
2090         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2091         up_write(&curseg->journal_rwsem);
2092
2093         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2094         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
2095         curseg->next_segno = segno;
2096         reset_curseg(sbi, type, 0);
2097         curseg->alloc_type = ckpt->alloc_type[type];
2098         curseg->next_blkoff = blk_off;
2099         mutex_unlock(&curseg->curseg_mutex);
2100         f2fs_put_page(new, 1);
2101         return 0;
2102 }
2103
2104 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2105 {
2106         int type = CURSEG_HOT_DATA;
2107         int err;
2108
2109         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
2110                 int npages = npages_for_summary_flush(sbi, true);
2111
2112                 if (npages >= 2)
2113                         ra_meta_pages(sbi, start_sum_block(sbi), npages,
2114                                                         META_CP, true);
2115
2116                 /* restore for compacted data summary */
2117                 if (read_compacted_summaries(sbi))
2118                         return -EINVAL;
2119                 type = CURSEG_HOT_NODE;
2120         }
2121
2122         if (__exist_node_summaries(sbi))
2123                 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
2124                                         NR_CURSEG_TYPE - type, META_CP, true);
2125
2126         for (; type <= CURSEG_COLD_NODE; type++) {
2127                 err = read_normal_summaries(sbi, type);
2128                 if (err)
2129                         return err;
2130         }
2131
2132         return 0;
2133 }
2134
2135 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2136 {
2137         struct page *page;
2138         unsigned char *kaddr;
2139         struct f2fs_summary *summary;
2140         struct curseg_info *seg_i;
2141         int written_size = 0;
2142         int i, j;
2143
2144         page = grab_meta_page(sbi, blkaddr++);
2145         kaddr = (unsigned char *)page_address(page);
2146
2147         /* Step 1: write nat cache */
2148         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2149         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
2150         written_size += SUM_JOURNAL_SIZE;
2151
2152         /* Step 2: write sit cache */
2153         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2154         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
2155         written_size += SUM_JOURNAL_SIZE;
2156
2157         /* Step 3: write summary entries */
2158         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2159                 unsigned short blkoff;
2160                 seg_i = CURSEG_I(sbi, i);
2161                 if (sbi->ckpt->alloc_type[i] == SSR)
2162                         blkoff = sbi->blocks_per_seg;
2163                 else
2164                         blkoff = curseg_blkoff(sbi, i);
2165
2166                 for (j = 0; j < blkoff; j++) {
2167                         if (!page) {
2168                                 page = grab_meta_page(sbi, blkaddr++);
2169                                 kaddr = (unsigned char *)page_address(page);
2170                                 written_size = 0;
2171                         }
2172                         summary = (struct f2fs_summary *)(kaddr + written_size);
2173                         *summary = seg_i->sum_blk->entries[j];
2174                         written_size += SUMMARY_SIZE;
2175
2176                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
2177                                                         SUM_FOOTER_SIZE)
2178                                 continue;
2179
2180                         set_page_dirty(page);
2181                         f2fs_put_page(page, 1);
2182                         page = NULL;
2183                 }
2184         }
2185         if (page) {
2186                 set_page_dirty(page);
2187                 f2fs_put_page(page, 1);
2188         }
2189 }
2190
2191 static void write_normal_summaries(struct f2fs_sb_info *sbi,
2192                                         block_t blkaddr, int type)
2193 {
2194         int i, end;
2195         if (IS_DATASEG(type))
2196                 end = type + NR_CURSEG_DATA_TYPE;
2197         else
2198                 end = type + NR_CURSEG_NODE_TYPE;
2199
2200         for (i = type; i < end; i++)
2201                 write_current_sum_page(sbi, i, blkaddr + (i - type));
2202 }
2203
2204 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2205 {
2206         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
2207                 write_compacted_summaries(sbi, start_blk);
2208         else
2209                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2210 }
2211
2212 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2213 {
2214         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
2215 }
2216
2217 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
2218                                         unsigned int val, int alloc)
2219 {
2220         int i;
2221
2222         if (type == NAT_JOURNAL) {
2223                 for (i = 0; i < nats_in_cursum(journal); i++) {
2224                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
2225                                 return i;
2226                 }
2227                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2228                         return update_nats_in_cursum(journal, 1);
2229         } else if (type == SIT_JOURNAL) {
2230                 for (i = 0; i < sits_in_cursum(journal); i++)
2231                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
2232                                 return i;
2233                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2234                         return update_sits_in_cursum(journal, 1);
2235         }
2236         return -1;
2237 }
2238
2239 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2240                                         unsigned int segno)
2241 {
2242         return get_meta_page(sbi, current_sit_addr(sbi, segno));
2243 }
2244
2245 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2246                                         unsigned int start)
2247 {
2248         struct sit_info *sit_i = SIT_I(sbi);
2249         struct page *src_page, *dst_page;
2250         pgoff_t src_off, dst_off;
2251         void *src_addr, *dst_addr;
2252
2253         src_off = current_sit_addr(sbi, start);
2254         dst_off = next_sit_addr(sbi, src_off);
2255
2256         /* get current sit block page without lock */
2257         src_page = get_meta_page(sbi, src_off);
2258         dst_page = grab_meta_page(sbi, dst_off);
2259         f2fs_bug_on(sbi, PageDirty(src_page));
2260
2261         src_addr = page_address(src_page);
2262         dst_addr = page_address(dst_page);
2263         memcpy(dst_addr, src_addr, PAGE_SIZE);
2264
2265         set_page_dirty(dst_page);
2266         f2fs_put_page(src_page, 1);
2267
2268         set_to_next_sit(sit_i, start);
2269
2270         return dst_page;
2271 }
2272
2273 static struct sit_entry_set *grab_sit_entry_set(void)
2274 {
2275         struct sit_entry_set *ses =
2276                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
2277
2278         ses->entry_cnt = 0;
2279         INIT_LIST_HEAD(&ses->set_list);
2280         return ses;
2281 }
2282
2283 static void release_sit_entry_set(struct sit_entry_set *ses)
2284 {
2285         list_del(&ses->set_list);
2286         kmem_cache_free(sit_entry_set_slab, ses);
2287 }
2288
2289 static void adjust_sit_entry_set(struct sit_entry_set *ses,
2290                                                 struct list_head *head)
2291 {
2292         struct sit_entry_set *next = ses;
2293
2294         if (list_is_last(&ses->set_list, head))
2295                 return;
2296
2297         list_for_each_entry_continue(next, head, set_list)
2298                 if (ses->entry_cnt <= next->entry_cnt)
2299                         break;
2300
2301         list_move_tail(&ses->set_list, &next->set_list);
2302 }
2303
2304 static void add_sit_entry(unsigned int segno, struct list_head *head)
2305 {
2306         struct sit_entry_set *ses;
2307         unsigned int start_segno = START_SEGNO(segno);
2308
2309         list_for_each_entry(ses, head, set_list) {
2310                 if (ses->start_segno == start_segno) {
2311                         ses->entry_cnt++;
2312                         adjust_sit_entry_set(ses, head);
2313                         return;
2314                 }
2315         }
2316
2317         ses = grab_sit_entry_set();
2318
2319         ses->start_segno = start_segno;
2320         ses->entry_cnt++;
2321         list_add(&ses->set_list, head);
2322 }
2323
2324 static void add_sits_in_set(struct f2fs_sb_info *sbi)
2325 {
2326         struct f2fs_sm_info *sm_info = SM_I(sbi);
2327         struct list_head *set_list = &sm_info->sit_entry_set;
2328         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
2329         unsigned int segno;
2330
2331         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
2332                 add_sit_entry(segno, set_list);
2333 }
2334
2335 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
2336 {
2337         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2338         struct f2fs_journal *journal = curseg->journal;
2339         int i;
2340
2341         down_write(&curseg->journal_rwsem);
2342         for (i = 0; i < sits_in_cursum(journal); i++) {
2343                 unsigned int segno;
2344                 bool dirtied;
2345
2346                 segno = le32_to_cpu(segno_in_journal(journal, i));
2347                 dirtied = __mark_sit_entry_dirty(sbi, segno);
2348
2349                 if (!dirtied)
2350                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
2351         }
2352         update_sits_in_cursum(journal, -i);
2353         up_write(&curseg->journal_rwsem);
2354 }
2355
2356 /*
2357  * CP calls this function, which flushes SIT entries including sit_journal,
2358  * and moves prefree segs to free segs.
2359  */
2360 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2361 {
2362         struct sit_info *sit_i = SIT_I(sbi);
2363         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2364         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2365         struct f2fs_journal *journal = curseg->journal;
2366         struct sit_entry_set *ses, *tmp;
2367         struct list_head *head = &SM_I(sbi)->sit_entry_set;
2368         bool to_journal = true;
2369         struct seg_entry *se;
2370
2371         mutex_lock(&sit_i->sentry_lock);
2372
2373         if (!sit_i->dirty_sentries)
2374                 goto out;
2375
2376         /*
2377          * add and account sit entries of dirty bitmap in sit entry
2378          * set temporarily
2379          */
2380         add_sits_in_set(sbi);
2381
2382         /*
2383          * if there are no enough space in journal to store dirty sit
2384          * entries, remove all entries from journal and add and account
2385          * them in sit entry set.
2386          */
2387         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
2388                 remove_sits_in_journal(sbi);
2389
2390         /*
2391          * there are two steps to flush sit entries:
2392          * #1, flush sit entries to journal in current cold data summary block.
2393          * #2, flush sit entries to sit page.
2394          */
2395         list_for_each_entry_safe(ses, tmp, head, set_list) {
2396                 struct page *page = NULL;
2397                 struct f2fs_sit_block *raw_sit = NULL;
2398                 unsigned int start_segno = ses->start_segno;
2399                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
2400                                                 (unsigned long)MAIN_SEGS(sbi));
2401                 unsigned int segno = start_segno;
2402
2403                 if (to_journal &&
2404                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
2405                         to_journal = false;
2406
2407                 if (to_journal) {
2408                         down_write(&curseg->journal_rwsem);
2409                 } else {
2410                         page = get_next_sit_page(sbi, start_segno);
2411                         raw_sit = page_address(page);
2412                 }
2413
2414                 /* flush dirty sit entries in region of current sit set */
2415                 for_each_set_bit_from(segno, bitmap, end) {
2416                         int offset, sit_offset;
2417
2418                         se = get_seg_entry(sbi, segno);
2419
2420                         /* add discard candidates */
2421                         if (cpc->reason != CP_DISCARD) {
2422                                 cpc->trim_start = segno;
2423                                 add_discard_addrs(sbi, cpc, false);
2424                         }
2425
2426                         if (to_journal) {
2427                                 offset = lookup_journal_in_cursum(journal,
2428                                                         SIT_JOURNAL, segno, 1);
2429                                 f2fs_bug_on(sbi, offset < 0);
2430                                 segno_in_journal(journal, offset) =
2431                                                         cpu_to_le32(segno);
2432                                 seg_info_to_raw_sit(se,
2433                                         &sit_in_journal(journal, offset));
2434                         } else {
2435                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2436                                 seg_info_to_raw_sit(se,
2437                                                 &raw_sit->entries[sit_offset]);
2438                         }
2439
2440                         __clear_bit(segno, bitmap);
2441                         sit_i->dirty_sentries--;
2442                         ses->entry_cnt--;
2443                 }
2444
2445                 if (to_journal)
2446                         up_write(&curseg->journal_rwsem);
2447                 else
2448                         f2fs_put_page(page, 1);
2449
2450                 f2fs_bug_on(sbi, ses->entry_cnt);
2451                 release_sit_entry_set(ses);
2452         }
2453
2454         f2fs_bug_on(sbi, !list_empty(head));
2455         f2fs_bug_on(sbi, sit_i->dirty_sentries);
2456 out:
2457         if (cpc->reason == CP_DISCARD) {
2458                 __u64 trim_start = cpc->trim_start;
2459
2460                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2461                         add_discard_addrs(sbi, cpc, false);
2462
2463                 cpc->trim_start = trim_start;
2464         }
2465         mutex_unlock(&sit_i->sentry_lock);
2466
2467         set_prefree_as_free_segments(sbi);
2468 }
2469
2470 static int build_sit_info(struct f2fs_sb_info *sbi)
2471 {
2472         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2473         struct sit_info *sit_i;
2474         unsigned int sit_segs, start;
2475         char *src_bitmap;
2476         unsigned int bitmap_size;
2477
2478         /* allocate memory for SIT information */
2479         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2480         if (!sit_i)
2481                 return -ENOMEM;
2482
2483         SM_I(sbi)->sit_info = sit_i;
2484
2485         sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2486                                         sizeof(struct seg_entry), GFP_KERNEL);
2487         if (!sit_i->sentries)
2488                 return -ENOMEM;
2489
2490         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2491         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2492         if (!sit_i->dirty_sentries_bitmap)
2493                 return -ENOMEM;
2494
2495         for (start = 0; start < MAIN_SEGS(sbi); start++) {
2496                 sit_i->sentries[start].cur_valid_map
2497                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2498                 sit_i->sentries[start].ckpt_valid_map
2499                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2500                 if (!sit_i->sentries[start].cur_valid_map ||
2501                                 !sit_i->sentries[start].ckpt_valid_map)
2502                         return -ENOMEM;
2503
2504 #ifdef CONFIG_F2FS_CHECK_FS
2505                 sit_i->sentries[start].cur_valid_map_mir
2506                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2507                 if (!sit_i->sentries[start].cur_valid_map_mir)
2508                         return -ENOMEM;
2509 #endif
2510
2511                 if (f2fs_discard_en(sbi)) {
2512                         sit_i->sentries[start].discard_map
2513                                 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2514                         if (!sit_i->sentries[start].discard_map)
2515                                 return -ENOMEM;
2516                 }
2517         }
2518
2519         sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2520         if (!sit_i->tmp_map)
2521                 return -ENOMEM;
2522
2523         if (sbi->segs_per_sec > 1) {
2524                 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2525                                         sizeof(struct sec_entry), GFP_KERNEL);
2526                 if (!sit_i->sec_entries)
2527                         return -ENOMEM;
2528         }
2529
2530         /* get information related with SIT */
2531         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2532
2533         /* setup SIT bitmap from ckeckpoint pack */
2534         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2535         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2536
2537         sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2538         if (!sit_i->sit_bitmap)
2539                 return -ENOMEM;
2540
2541 #ifdef CONFIG_F2FS_CHECK_FS
2542         sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2543         if (!sit_i->sit_bitmap_mir)
2544                 return -ENOMEM;
2545 #endif
2546
2547         /* init SIT information */
2548         sit_i->s_ops = &default_salloc_ops;
2549
2550         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2551         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2552         sit_i->written_valid_blocks = 0;
2553         sit_i->bitmap_size = bitmap_size;
2554         sit_i->dirty_sentries = 0;
2555         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2556         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2557         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2558         mutex_init(&sit_i->sentry_lock);
2559         return 0;
2560 }
2561
2562 static int build_free_segmap(struct f2fs_sb_info *sbi)
2563 {
2564         struct free_segmap_info *free_i;
2565         unsigned int bitmap_size, sec_bitmap_size;
2566
2567         /* allocate memory for free segmap information */
2568         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2569         if (!free_i)
2570                 return -ENOMEM;
2571
2572         SM_I(sbi)->free_info = free_i;
2573
2574         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2575         free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2576         if (!free_i->free_segmap)
2577                 return -ENOMEM;
2578
2579         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2580         free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2581         if (!free_i->free_secmap)
2582                 return -ENOMEM;
2583
2584         /* set all segments as dirty temporarily */
2585         memset(free_i->free_segmap, 0xff, bitmap_size);
2586         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2587
2588         /* init free segmap information */
2589         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2590         free_i->free_segments = 0;
2591         free_i->free_sections = 0;
2592         spin_lock_init(&free_i->segmap_lock);
2593         return 0;
2594 }
2595
2596 static int build_curseg(struct f2fs_sb_info *sbi)
2597 {
2598         struct curseg_info *array;
2599         int i;
2600
2601         array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2602         if (!array)
2603                 return -ENOMEM;
2604
2605         SM_I(sbi)->curseg_array = array;
2606
2607         for (i = 0; i < NR_CURSEG_TYPE; i++) {
2608                 mutex_init(&array[i].curseg_mutex);
2609                 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2610                 if (!array[i].sum_blk)
2611                         return -ENOMEM;
2612                 init_rwsem(&array[i].journal_rwsem);
2613                 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2614                                                         GFP_KERNEL);
2615                 if (!array[i].journal)
2616                         return -ENOMEM;
2617                 array[i].segno = NULL_SEGNO;
2618                 array[i].next_blkoff = 0;
2619         }
2620         return restore_curseg_summaries(sbi);
2621 }
2622
2623 static void build_sit_entries(struct f2fs_sb_info *sbi)
2624 {
2625         struct sit_info *sit_i = SIT_I(sbi);
2626         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2627         struct f2fs_journal *journal = curseg->journal;
2628         struct seg_entry *se;
2629         struct f2fs_sit_entry sit;
2630         int sit_blk_cnt = SIT_BLK_CNT(sbi);
2631         unsigned int i, start, end;
2632         unsigned int readed, start_blk = 0;
2633
2634         do {
2635                 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
2636                                                         META_SIT, true);
2637
2638                 start = start_blk * sit_i->sents_per_block;
2639                 end = (start_blk + readed) * sit_i->sents_per_block;
2640
2641                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2642                         struct f2fs_sit_block *sit_blk;
2643                         struct page *page;
2644
2645                         se = &sit_i->sentries[start];
2646                         page = get_current_sit_page(sbi, start);
2647                         sit_blk = (struct f2fs_sit_block *)page_address(page);
2648                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2649                         f2fs_put_page(page, 1);
2650
2651                         check_block_count(sbi, start, &sit);
2652                         seg_info_from_raw_sit(se, &sit);
2653
2654                         /* build discard map only one time */
2655                         if (f2fs_discard_en(sbi)) {
2656                                 memcpy(se->discard_map, se->cur_valid_map,
2657                                                         SIT_VBLOCK_MAP_SIZE);
2658                                 sbi->discard_blks += sbi->blocks_per_seg -
2659                                                         se->valid_blocks;
2660                         }
2661
2662                         if (sbi->segs_per_sec > 1)
2663                                 get_sec_entry(sbi, start)->valid_blocks +=
2664                                                         se->valid_blocks;
2665                 }
2666                 start_blk += readed;
2667         } while (start_blk < sit_blk_cnt);
2668
2669         down_read(&curseg->journal_rwsem);
2670         for (i = 0; i < sits_in_cursum(journal); i++) {
2671                 unsigned int old_valid_blocks;
2672
2673                 start = le32_to_cpu(segno_in_journal(journal, i));
2674                 se = &sit_i->sentries[start];
2675                 sit = sit_in_journal(journal, i);
2676
2677                 old_valid_blocks = se->valid_blocks;
2678
2679                 check_block_count(sbi, start, &sit);
2680                 seg_info_from_raw_sit(se, &sit);
2681
2682                 if (f2fs_discard_en(sbi)) {
2683                         memcpy(se->discard_map, se->cur_valid_map,
2684                                                 SIT_VBLOCK_MAP_SIZE);
2685                         sbi->discard_blks += old_valid_blocks -
2686                                                 se->valid_blocks;
2687                 }
2688
2689                 if (sbi->segs_per_sec > 1)
2690                         get_sec_entry(sbi, start)->valid_blocks +=
2691                                 se->valid_blocks - old_valid_blocks;
2692         }
2693         up_read(&curseg->journal_rwsem);
2694 }
2695
2696 static void init_free_segmap(struct f2fs_sb_info *sbi)
2697 {
2698         unsigned int start;
2699         int type;
2700
2701         for (start = 0; start < MAIN_SEGS(sbi); start++) {
2702                 struct seg_entry *sentry = get_seg_entry(sbi, start);
2703                 if (!sentry->valid_blocks)
2704                         __set_free(sbi, start);
2705                 else
2706                         SIT_I(sbi)->written_valid_blocks +=
2707                                                 sentry->valid_blocks;
2708         }
2709
2710         /* set use the current segments */
2711         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2712                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2713                 __set_test_and_inuse(sbi, curseg_t->segno);
2714         }
2715 }
2716
2717 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2718 {
2719         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2720         struct free_segmap_info *free_i = FREE_I(sbi);
2721         unsigned int segno = 0, offset = 0;
2722         unsigned short valid_blocks;
2723
2724         while (1) {
2725                 /* find dirty segment based on free segmap */
2726                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2727                 if (segno >= MAIN_SEGS(sbi))
2728                         break;
2729                 offset = segno + 1;
2730                 valid_blocks = get_valid_blocks(sbi, segno, 0);
2731                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2732                         continue;
2733                 if (valid_blocks > sbi->blocks_per_seg) {
2734                         f2fs_bug_on(sbi, 1);
2735                         continue;
2736                 }
2737                 mutex_lock(&dirty_i->seglist_lock);
2738                 __locate_dirty_segment(sbi, segno, DIRTY);
2739                 mutex_unlock(&dirty_i->seglist_lock);
2740         }
2741 }
2742
2743 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2744 {
2745         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2746         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2747
2748         dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2749         if (!dirty_i->victim_secmap)
2750                 return -ENOMEM;
2751         return 0;
2752 }
2753
2754 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2755 {
2756         struct dirty_seglist_info *dirty_i;
2757         unsigned int bitmap_size, i;
2758
2759         /* allocate memory for dirty segments list information */
2760         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2761         if (!dirty_i)
2762                 return -ENOMEM;
2763
2764         SM_I(sbi)->dirty_info = dirty_i;
2765         mutex_init(&dirty_i->seglist_lock);
2766
2767         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2768
2769         for (i = 0; i < NR_DIRTY_TYPE; i++) {
2770                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2771                 if (!dirty_i->dirty_segmap[i])
2772                         return -ENOMEM;
2773         }
2774
2775         init_dirty_segmap(sbi);
2776         return init_victim_secmap(sbi);
2777 }
2778
2779 /*
2780  * Update min, max modified time for cost-benefit GC algorithm
2781  */
2782 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2783 {
2784         struct sit_info *sit_i = SIT_I(sbi);
2785         unsigned int segno;
2786
2787         mutex_lock(&sit_i->sentry_lock);
2788
2789         sit_i->min_mtime = LLONG_MAX;
2790
2791         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2792                 unsigned int i;
2793                 unsigned long long mtime = 0;
2794
2795                 for (i = 0; i < sbi->segs_per_sec; i++)
2796                         mtime += get_seg_entry(sbi, segno + i)->mtime;
2797
2798                 mtime = div_u64(mtime, sbi->segs_per_sec);
2799
2800                 if (sit_i->min_mtime > mtime)
2801                         sit_i->min_mtime = mtime;
2802         }
2803         sit_i->max_mtime = get_mtime(sbi);
2804         mutex_unlock(&sit_i->sentry_lock);
2805 }
2806
2807 int build_segment_manager(struct f2fs_sb_info *sbi)
2808 {
2809         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2810         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2811         struct f2fs_sm_info *sm_info;
2812         int err;
2813
2814         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2815         if (!sm_info)
2816                 return -ENOMEM;
2817
2818         /* init sm info */
2819         sbi->sm_info = sm_info;
2820         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2821         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2822         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2823         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2824         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2825         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2826         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2827         sm_info->rec_prefree_segments = sm_info->main_segments *
2828                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2829         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2830                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2831
2832         if (!test_opt(sbi, LFS))
2833                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2834         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2835         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2836
2837         sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2838
2839         INIT_LIST_HEAD(&sm_info->sit_entry_set);
2840
2841         if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2842                 err = create_flush_cmd_control(sbi);
2843                 if (err)
2844                         return err;
2845         }
2846
2847         err = create_discard_cmd_control(sbi);
2848         if (err)
2849                 return err;
2850
2851         err = build_sit_info(sbi);
2852         if (err)
2853                 return err;
2854         err = build_free_segmap(sbi);
2855         if (err)
2856                 return err;
2857         err = build_curseg(sbi);
2858         if (err)
2859                 return err;
2860
2861         /* reinit free segmap based on SIT */
2862         build_sit_entries(sbi);
2863
2864         init_free_segmap(sbi);
2865         err = build_dirty_segmap(sbi);
2866         if (err)
2867                 return err;
2868
2869         init_min_max_mtime(sbi);
2870         return 0;
2871 }
2872
2873 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2874                 enum dirty_type dirty_type)
2875 {
2876         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2877
2878         mutex_lock(&dirty_i->seglist_lock);
2879         kvfree(dirty_i->dirty_segmap[dirty_type]);
2880         dirty_i->nr_dirty[dirty_type] = 0;
2881         mutex_unlock(&dirty_i->seglist_lock);
2882 }
2883
2884 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2885 {
2886         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2887         kvfree(dirty_i->victim_secmap);
2888 }
2889
2890 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2891 {
2892         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2893         int i;
2894
2895         if (!dirty_i)
2896                 return;
2897
2898         /* discard pre-free/dirty segments list */
2899         for (i = 0; i < NR_DIRTY_TYPE; i++)
2900                 discard_dirty_segmap(sbi, i);
2901
2902         destroy_victim_secmap(sbi);
2903         SM_I(sbi)->dirty_info = NULL;
2904         kfree(dirty_i);
2905 }
2906
2907 static void destroy_curseg(struct f2fs_sb_info *sbi)
2908 {
2909         struct curseg_info *array = SM_I(sbi)->curseg_array;
2910         int i;
2911
2912         if (!array)
2913                 return;
2914         SM_I(sbi)->curseg_array = NULL;
2915         for (i = 0; i < NR_CURSEG_TYPE; i++) {
2916                 kfree(array[i].sum_blk);
2917                 kfree(array[i].journal);
2918         }
2919         kfree(array);
2920 }
2921
2922 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2923 {
2924         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2925         if (!free_i)
2926                 return;
2927         SM_I(sbi)->free_info = NULL;
2928         kvfree(free_i->free_segmap);
2929         kvfree(free_i->free_secmap);
2930         kfree(free_i);
2931 }
2932
2933 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2934 {
2935         struct sit_info *sit_i = SIT_I(sbi);
2936         unsigned int start;
2937
2938         if (!sit_i)
2939                 return;
2940
2941         if (sit_i->sentries) {
2942                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2943                         kfree(sit_i->sentries[start].cur_valid_map);
2944 #ifdef CONFIG_F2FS_CHECK_FS
2945                         kfree(sit_i->sentries[start].cur_valid_map_mir);
2946 #endif
2947                         kfree(sit_i->sentries[start].ckpt_valid_map);
2948                         kfree(sit_i->sentries[start].discard_map);
2949                 }
2950         }
2951         kfree(sit_i->tmp_map);
2952
2953         kvfree(sit_i->sentries);
2954         kvfree(sit_i->sec_entries);
2955         kvfree(sit_i->dirty_sentries_bitmap);
2956
2957         SM_I(sbi)->sit_info = NULL;
2958         kfree(sit_i->sit_bitmap);
2959 #ifdef CONFIG_F2FS_CHECK_FS
2960         kfree(sit_i->sit_bitmap_mir);
2961 #endif
2962         kfree(sit_i);
2963 }
2964
2965 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2966 {
2967         struct f2fs_sm_info *sm_info = SM_I(sbi);
2968
2969         if (!sm_info)
2970                 return;
2971         destroy_flush_cmd_control(sbi, true);
2972         destroy_discard_cmd_control(sbi, true);
2973         destroy_dirty_segmap(sbi);
2974         destroy_curseg(sbi);
2975         destroy_free_segmap(sbi);
2976         destroy_sit_info(sbi);
2977         sbi->sm_info = NULL;
2978         kfree(sm_info);
2979 }
2980
2981 int __init create_segment_manager_caches(void)
2982 {
2983         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2984                         sizeof(struct discard_entry));
2985         if (!discard_entry_slab)
2986                 goto fail;
2987
2988         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
2989                         sizeof(struct discard_cmd));
2990         if (!discard_cmd_slab)
2991                 goto destroy_discard_entry;
2992
2993         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2994                         sizeof(struct sit_entry_set));
2995         if (!sit_entry_set_slab)
2996                 goto destroy_discard_cmd;
2997
2998         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2999                         sizeof(struct inmem_pages));
3000         if (!inmem_entry_slab)
3001                 goto destroy_sit_entry_set;
3002         return 0;
3003
3004 destroy_sit_entry_set:
3005         kmem_cache_destroy(sit_entry_set_slab);
3006 destroy_discard_cmd:
3007         kmem_cache_destroy(discard_cmd_slab);
3008 destroy_discard_entry:
3009         kmem_cache_destroy(discard_entry_slab);
3010 fail:
3011         return -ENOMEM;
3012 }
3013
3014 void destroy_segment_manager_caches(void)
3015 {
3016         kmem_cache_destroy(sit_entry_set_slab);
3017         kmem_cache_destroy(discard_cmd_slab);
3018         kmem_cache_destroy(discard_entry_slab);
3019         kmem_cache_destroy(inmem_entry_slab);
3020 }