fs/f2fs/segment.c

   1 /*
   2  * fs/f2fs/segment.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/bio.h>
  14 #include <linux/blkdev.h>
  15 #include <linux/prefetch.h>
  16 #include <linux/kthread.h>
  17 #include <linux/swap.h>
  18 #include <linux/timer.h>
  19
  20 #include "f2fs.h"
  21 #include "segment.h"
  22 #include "node.h"
  23 #include "trace.h"
  24 #include <trace/events/f2fs.h>
  25
  26 #define __reverse_ffz(x) __reverse_ffs(~(x))
  27
  28 static struct kmem_cache *discard_entry_slab;
  29 static struct kmem_cache *bio_entry_slab;
  30 static struct kmem_cache *sit_entry_set_slab;
  31 static struct kmem_cache *inmem_entry_slab;
  32
  33 static unsigned long __reverse_ulong(unsigned char *str)
  34 {
  35         unsigned long tmp = 0;
  36         int shift = 24, idx = 0;
  37
  38 #if BITS_PER_LONG == 64
  39         shift = 56;
  40 #endif
  41         while (shift >= 0) {
  42                 tmp |= (unsigned long)str[idx++] << shift;
  43                 shift -= BITS_PER_BYTE;
  44         }
  45         return tmp;
  46 }
  47
  48 /*
  49  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  50  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  51  */
  52 static inline unsigned long __reverse_ffs(unsigned long word)
  53 {
  54         int num = 0;
  55
  56 #if BITS_PER_LONG == 64
  57         if ((word & 0xffffffff00000000UL) == 0)
  58                 num += 32;
  59         else
  60                 word >>= 32;
  61 #endif
  62         if ((word & 0xffff0000) == 0)
  63                 num += 16;
  64         else
  65                 word >>= 16;
  66
  67         if ((word & 0xff00) == 0)
  68                 num += 8;
  69         else
  70                 word >>= 8;
  71
  72         if ((word & 0xf0) == 0)
  73                 num += 4;
  74         else
  75                 word >>= 4;
  76
  77         if ((word & 0xc) == 0)
  78                 num += 2;
  79         else
  80                 word >>= 2;
  81
  82         if ((word & 0x2) == 0)
  83                 num += 1;
  84         return num;
  85 }
  86
  87 /*
  88  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  89  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  90  * @size must be integral times of unsigned long.
  91  * Example:
  92  *                             MSB <--> LSB
  93  *   f2fs_set_bit(0, bitmap) => 1000 0000
  94  *   f2fs_set_bit(7, bitmap) => 0000 0001
  95  */
  96 static unsigned long __find_rev_next_bit(const unsigned long *addr,
  97                         unsigned long size, unsigned long offset)
  98 {
  99         const unsigned long *p = addr + BIT_WORD(offset);
 100         unsigned long result = size;
 101         unsigned long tmp;
 102
 103         if (offset >= size)
 104                 return size;
 105
 106         size -= (offset & ~(BITS_PER_LONG - 1));
 107         offset %= BITS_PER_LONG;
 108
 109         while (1) {
 110                 if (*p == 0)
 111                         goto pass;
 112
 113                 tmp = __reverse_ulong((unsigned char *)p);
 114
 115                 tmp &= ~0UL >> offset;
 116                 if (size < BITS_PER_LONG)
 117                         tmp &= (~0UL << (BITS_PER_LONG - size));
 118                 if (tmp)
 119                         goto found;
 120 pass:
 121                 if (size <= BITS_PER_LONG)
 122                         break;
 123                 size -= BITS_PER_LONG;
 124                 offset = 0;
 125                 p++;
 126         }
 127         return result;
 128 found:
 129         return result - size + __reverse_ffs(tmp);
 130 }
 131
 132 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 133                         unsigned long size, unsigned long offset)
 134 {
 135         const unsigned long *p = addr + BIT_WORD(offset);
 136         unsigned long result = size;
 137         unsigned long tmp;
 138
 139         if (offset >= size)
 140                 return size;
 141
 142         size -= (offset & ~(BITS_PER_LONG - 1));
 143         offset %= BITS_PER_LONG;
 144
 145         while (1) {
 146                 if (*p == ~0UL)
 147                         goto pass;
 148
 149                 tmp = __reverse_ulong((unsigned char *)p);
 150
 151                 if (offset)
 152                         tmp |= ~0UL << (BITS_PER_LONG - offset);
 153                 if (size < BITS_PER_LONG)
 154                         tmp |= ~0UL >> size;
 155                 if (tmp != ~0UL)
 156                         goto found;
 157 pass:
 158                 if (size <= BITS_PER_LONG)
 159                         break;
 160                 size -= BITS_PER_LONG;
 161                 offset = 0;
 162                 p++;
 163         }
 164         return result;
 165 found:
 166         return result - size + __reverse_ffz(tmp);
 167 }
 168
 169 void register_inmem_page(struct inode *inode, struct page *page)
 170 {
 171         struct f2fs_inode_info *fi = F2FS_I(inode);
 172         struct inmem_pages *new;
 173
 174         f2fs_trace_pid(page);
 175
 176         set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
 177         SetPagePrivate(page);
 178
 179         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 180
 181         /* add atomic page indices to the list */
 182         new->page = page;
 183         INIT_LIST_HEAD(&new->list);
 184
 185         /* increase reference count with clean state */
 186         mutex_lock(&fi->inmem_lock);
 187         get_page(page);
 188         list_add_tail(&new->list, &fi->inmem_pages);
 189         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 190         mutex_unlock(&fi->inmem_lock);
 191
 192         trace_f2fs_register_inmem_page(page, INMEM);
 193 }
 194
 195 static int __revoke_inmem_pages(struct inode *inode,
 196                                 struct list_head *head, bool drop, bool recover)
 197 {
 198         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 199         struct inmem_pages *cur, *tmp;
 200         int err = 0;
 201
 202         list_for_each_entry_safe(cur, tmp, head, list) {
 203                 struct page *page = cur->page;
 204
 205                 if (drop)
 206                         trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 207
 208                 lock_page(page);
 209
 210                 f2fs_wait_on_page_writeback(page, DATA, true);
 211
 212                 if (recover) {
 213                         struct dnode_of_data dn;
 214                         struct node_info ni;
 215
 216                         trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 217
 218                         set_new_dnode(&dn, inode, NULL, NULL, 0);
 219                         if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) {
 220                                 err = -EAGAIN;
 221                                 goto next;
 222                         }
 223                         get_node_info(sbi, dn.nid, &ni);
 224                         f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 225                                         cur->old_addr, ni.version, true, true);
 226                         f2fs_put_dnode(&dn);
 227                 }
 228 next:
 229                 /* we don't need to invalidate this in the sccessful status */
 230                 if (drop || recover) {
 231                         ClearPageUptodate(page);
 232                         clear_cold_data(page);
 233                 }
 234                 set_page_private(page, 0);
 235                 ClearPagePrivate(page);
 236                 f2fs_put_page(page, 1);
 237
 238                 list_del(&cur->list);
 239                 kmem_cache_free(inmem_entry_slab, cur);
 240                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 241         }
 242         return err;
 243 }
 244
 245 void drop_inmem_pages(struct inode *inode)
 246 {
 247         struct f2fs_inode_info *fi = F2FS_I(inode);
 248
 249         clear_inode_flag(inode, FI_ATOMIC_FILE);
 250
 251         mutex_lock(&fi->inmem_lock);
 252         __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 253         mutex_unlock(&fi->inmem_lock);
 254 }
 255
 256 static int __commit_inmem_pages(struct inode *inode,
 257                                         struct list_head *revoke_list)
 258 {
 259         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 260         struct f2fs_inode_info *fi = F2FS_I(inode);
 261         struct inmem_pages *cur, *tmp;
 262         struct f2fs_io_info fio = {
 263                 .sbi = sbi,
 264                 .type = DATA,
 265                 .op = REQ_OP_WRITE,
 266                 .op_flags = WRITE_SYNC | REQ_PRIO,
 267                 .encrypted_page = NULL,
 268         };
 269         bool submit_bio = false;
 270         int err = 0;
 271
 272         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 273                 struct page *page = cur->page;
 274
 275                 lock_page(page);
 276                 if (page->mapping == inode->i_mapping) {
 277                         trace_f2fs_commit_inmem_page(page, INMEM);
 278
 279                         set_page_dirty(page);
 280                         f2fs_wait_on_page_writeback(page, DATA, true);
 281                         if (clear_page_dirty_for_io(page))
 282                                 inode_dec_dirty_pages(inode);
 283
 284                         fio.page = page;
 285                         err = do_write_data_page(&fio);
 286                         if (err) {
 287                                 unlock_page(page);
 288                                 break;
 289                         }
 290
 291                         /* record old blkaddr for revoking */
 292                         cur->old_addr = fio.old_blkaddr;
 293
 294                         clear_cold_data(page);
 295                         submit_bio = true;
 296                 }
 297                 unlock_page(page);
 298                 list_move_tail(&cur->list, revoke_list);
 299         }
 300
 301         if (submit_bio)
 302                 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
 303
 304         if (!err)
 305                 __revoke_inmem_pages(inode, revoke_list, false, false);
 306
 307         return err;
 308 }
 309
 310 int commit_inmem_pages(struct inode *inode)
 311 {
 312         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 313         struct f2fs_inode_info *fi = F2FS_I(inode);
 314         struct list_head revoke_list;
 315         int err;
 316
 317         INIT_LIST_HEAD(&revoke_list);
 318         f2fs_balance_fs(sbi, true);
 319         f2fs_lock_op(sbi);
 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         f2fs_unlock_op(sbi);
 343         return err;
 344 }
 345
 346 /*
 347  * This function balances dirty node and dentry pages.
 348  * In addition, it controls garbage collection.
 349  */
 350 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 351 {
 352 #ifdef CONFIG_F2FS_FAULT_INJECTION
 353         if (time_to_inject(sbi, FAULT_CHECKPOINT))
 354                 f2fs_stop_checkpoint(sbi, false);
 355 #endif
 356
 357         if (!need)
 358                 return;
 359
 360         /* balance_fs_bg is able to be pending */
 361         if (excess_cached_nats(sbi))
 362                 f2fs_balance_fs_bg(sbi);
 363
 364         /*
 365          * We should do GC or end up with checkpoint, if there are so many dirty
 366          * dir/node pages without enough free segments.
 367          */
 368         if (has_not_enough_free_secs(sbi, 0, 0)) {
 369                 mutex_lock(&sbi->gc_mutex);
 370                 f2fs_gc(sbi, false);
 371         }
 372 }
 373
 374 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 375 {
 376         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 377                 return;
 378
 379         /* try to shrink extent cache when there is no enough memory */
 380         if (!available_free_memory(sbi, EXTENT_CACHE))
 381                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 382
 383         /* check the # of cached NAT entries */
 384         if (!available_free_memory(sbi, NAT_ENTRIES))
 385                 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 386
 387         if (!available_free_memory(sbi, FREE_NIDS))
 388                 try_to_free_nids(sbi, MAX_FREE_NIDS);
 389         else
 390                 build_free_nids(sbi);
 391
 392         /* checkpoint is the only way to shrink partial cached entries */
 393         if (!available_free_memory(sbi, NAT_ENTRIES) ||
 394                         !available_free_memory(sbi, INO_ENTRIES) ||
 395                         excess_prefree_segs(sbi) ||
 396                         excess_dirty_nats(sbi) ||
 397                         (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
 398                 if (test_opt(sbi, DATA_FLUSH)) {
 399                         struct blk_plug plug;
 400
 401                         blk_start_plug(&plug);
 402                         sync_dirty_inodes(sbi, FILE_INODE);
 403                         blk_finish_plug(&plug);
 404                 }
 405                 f2fs_sync_fs(sbi->sb, true);
 406                 stat_inc_bg_cp_count(sbi->stat_info);
 407         }
 408 }
 409
 410 static int issue_flush_thread(void *data)
 411 {
 412         struct f2fs_sb_info *sbi = data;
 413         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 414         wait_queue_head_t *q = &fcc->flush_wait_queue;
 415 repeat:
 416         if (kthread_should_stop())
 417                 return 0;
 418
 419         if (!llist_empty(&fcc->issue_list)) {
 420                 struct bio *bio;
 421                 struct flush_cmd *cmd, *next;
 422                 int ret;
 423
 424                 bio = f2fs_bio_alloc(0);
 425
 426                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 427                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 428
 429                 bio->bi_bdev = sbi->sb->s_bdev;
 430                 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
 431                 ret = submit_bio_wait(bio);
 432
 433                 llist_for_each_entry_safe(cmd, next,
 434                                           fcc->dispatch_list, llnode) {
 435                         cmd->ret = ret;
 436                         complete(&cmd->wait);
 437                 }
 438                 bio_put(bio);
 439                 fcc->dispatch_list = NULL;
 440         }
 441
 442         wait_event_interruptible(*q,
 443                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 444         goto repeat;
 445 }
 446
 447 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
 448 {
 449         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 450         struct flush_cmd cmd;
 451
 452         trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
 453                                         test_opt(sbi, FLUSH_MERGE));
 454
 455         if (test_opt(sbi, NOBARRIER))
 456                 return 0;
 457
 458         if (!test_opt(sbi, FLUSH_MERGE) || !atomic_read(&fcc->submit_flush)) {
 459                 struct bio *bio = f2fs_bio_alloc(0);
 460                 int ret;
 461
 462                 atomic_inc(&fcc->submit_flush);
 463                 bio->bi_bdev = sbi->sb->s_bdev;
 464                 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
 465                 ret = submit_bio_wait(bio);
 466                 atomic_dec(&fcc->submit_flush);
 467                 bio_put(bio);
 468                 return ret;
 469         }
 470
 471         init_completion(&cmd.wait);
 472
 473         atomic_inc(&fcc->submit_flush);
 474         llist_add(&cmd.llnode, &fcc->issue_list);
 475
 476         if (!fcc->dispatch_list)
 477                 wake_up(&fcc->flush_wait_queue);
 478
 479         wait_for_completion(&cmd.wait);
 480         atomic_dec(&fcc->submit_flush);
 481
 482         return cmd.ret;
 483 }
 484
 485 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
 486 {
 487         dev_t dev = sbi->sb->s_bdev->bd_dev;
 488         struct flush_cmd_control *fcc;
 489         int err = 0;
 490
 491         fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
 492         if (!fcc)
 493                 return -ENOMEM;
 494         atomic_set(&fcc->submit_flush, 0);
 495         init_waitqueue_head(&fcc->flush_wait_queue);
 496         init_llist_head(&fcc->issue_list);
 497         SM_I(sbi)->cmd_control_info = fcc;
 498         if (!test_opt(sbi, FLUSH_MERGE))
 499                 return err;
 500
 501         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 502                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 503         if (IS_ERR(fcc->f2fs_issue_flush)) {
 504                 err = PTR_ERR(fcc->f2fs_issue_flush);
 505                 kfree(fcc);
 506                 SM_I(sbi)->cmd_control_info = NULL;
 507                 return err;
 508         }
 509
 510         return err;
 511 }
 512
 513 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
 514 {
 515         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 516
 517         if (fcc && fcc->f2fs_issue_flush)
 518                 kthread_stop(fcc->f2fs_issue_flush);
 519         kfree(fcc);
 520         SM_I(sbi)->cmd_control_info = NULL;
 521 }
 522
 523 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 524                 enum dirty_type dirty_type)
 525 {
 526         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 527
 528         /* need not be added */
 529         if (IS_CURSEG(sbi, segno))
 530                 return;
 531
 532         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 533                 dirty_i->nr_dirty[dirty_type]++;
 534
 535         if (dirty_type == DIRTY) {
 536                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 537                 enum dirty_type t = sentry->type;
 538
 539                 if (unlikely(t >= DIRTY)) {
 540                         f2fs_bug_on(sbi, 1);
 541                         return;
 542                 }
 543                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 544                         dirty_i->nr_dirty[t]++;
 545         }
 546 }
 547
 548 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 549                 enum dirty_type dirty_type)
 550 {
 551         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 552
 553         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 554                 dirty_i->nr_dirty[dirty_type]--;
 555
 556         if (dirty_type == DIRTY) {
 557                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 558                 enum dirty_type t = sentry->type;
 559
 560                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 561                         dirty_i->nr_dirty[t]--;
 562
 563                 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
 564                         clear_bit(GET_SECNO(sbi, segno),
 565                                                 dirty_i->victim_secmap);
 566         }
 567 }
 568
 569 /*
 570  * Should not occur error such as -ENOMEM.
 571  * Adding dirty entry into seglist is not critical operation.
 572  * If a given segment is one of current working segments, it won't be added.
 573  */
 574 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 575 {
 576         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 577         unsigned short valid_blocks;
 578
 579         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 580                 return;
 581
 582         mutex_lock(&dirty_i->seglist_lock);
 583
 584         valid_blocks = get_valid_blocks(sbi, segno, 0);
 585
 586         if (valid_blocks == 0) {
 587                 __locate_dirty_segment(sbi, segno, PRE);
 588                 __remove_dirty_segment(sbi, segno, DIRTY);
 589         } else if (valid_blocks < sbi->blocks_per_seg) {
 590                 __locate_dirty_segment(sbi, segno, DIRTY);
 591         } else {
 592                 /* Recovery routine with SSR needs this */
 593                 __remove_dirty_segment(sbi, segno, DIRTY);
 594         }
 595
 596         mutex_unlock(&dirty_i->seglist_lock);
 597 }
 598
 599 static struct bio_entry *__add_bio_entry(struct f2fs_sb_info *sbi,
 600                                                         struct bio *bio)
 601 {
 602         struct list_head *wait_list = &(SM_I(sbi)->wait_list);
 603         struct bio_entry *be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS);
 604
 605         INIT_LIST_HEAD(&be->list);
 606         be->bio = bio;
 607         init_completion(&be->event);
 608         list_add_tail(&be->list, wait_list);
 609
 610         return be;
 611 }
 612
 613 void f2fs_wait_all_discard_bio(struct f2fs_sb_info *sbi)
 614 {
 615         struct list_head *wait_list = &(SM_I(sbi)->wait_list);
 616         struct bio_entry *be, *tmp;
 617
 618         list_for_each_entry_safe(be, tmp, wait_list, list) {
 619                 struct bio *bio = be->bio;
 620                 int err;
 621
 622                 wait_for_completion_io(&be->event);
 623                 err = be->error;
 624                 if (err == -EOPNOTSUPP)
 625                         err = 0;
 626
 627                 if (err)
 628                         f2fs_msg(sbi->sb, KERN_INFO,
 629                                 "Issue discard failed, ret: %d", err);
 630
 631                 bio_put(bio);
 632                 list_del(&be->list);
 633                 kmem_cache_free(bio_entry_slab, be);
 634         }
 635 }
 636
 637 static void f2fs_submit_bio_wait_endio(struct bio *bio)
 638 {
 639         struct bio_entry *be = (struct bio_entry *)bio->bi_private;
 640
 641         be->error = bio->bi_error;
 642         complete(&be->event);
 643 }
 644
 645 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
 646 int __f2fs_issue_discard_async(struct f2fs_sb_info *sbi, sector_t sector,
 647                 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
 648 {
 649         struct block_device *bdev = sbi->sb->s_bdev;
 650         struct bio *bio = NULL;
 651         int err;
 652
 653         err = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, flags,
 654                         &bio);
 655         if (!err && bio) {
 656                 struct bio_entry *be = __add_bio_entry(sbi, bio);
 657
 658                 bio->bi_private = be;
 659                 bio->bi_end_io = f2fs_submit_bio_wait_endio;
 660                 bio->bi_opf |= REQ_SYNC;
 661                 submit_bio(bio);
 662         }
 663
 664         return err;
 665 }
 666
 667 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
 668                                 block_t blkstart, block_t blklen)
 669 {
 670         sector_t start = SECTOR_FROM_BLOCK(blkstart);
 671         sector_t len = SECTOR_FROM_BLOCK(blklen);
 672         struct seg_entry *se;
 673         unsigned int offset;
 674         block_t i;
 675
 676         for (i = blkstart; i < blkstart + blklen; i++) {
 677                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
 678                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
 679
 680                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
 681                         sbi->discard_blks--;
 682         }
 683         trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
 684         return __f2fs_issue_discard_async(sbi, start, len, GFP_NOFS, 0);
 685 }
 686
 687 static void __add_discard_entry(struct f2fs_sb_info *sbi,
 688                 struct cp_control *cpc, struct seg_entry *se,
 689                 unsigned int start, unsigned int end)
 690 {
 691         struct list_head *head = &SM_I(sbi)->discard_list;
 692         struct discard_entry *new, *last;
 693
 694         if (!list_empty(head)) {
 695                 last = list_last_entry(head, struct discard_entry, list);
 696                 if (START_BLOCK(sbi, cpc->trim_start) + start ==
 697                                                 last->blkaddr + last->len) {
 698                         last->len += end - start;
 699                         goto done;
 700                 }
 701         }
 702
 703         new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
 704         INIT_LIST_HEAD(&new->list);
 705         new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
 706         new->len = end - start;
 707         list_add_tail(&new->list, head);
 708 done:
 709         SM_I(sbi)->nr_discards += end - start;
 710 }
 711
 712 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 713 {
 714         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
 715         int max_blocks = sbi->blocks_per_seg;
 716         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
 717         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
 718         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
 719         unsigned long *discard_map = (unsigned long *)se->discard_map;
 720         unsigned long *dmap = SIT_I(sbi)->tmp_map;
 721         unsigned int start = 0, end = -1;
 722         bool force = (cpc->reason == CP_DISCARD);
 723         int i;
 724
 725         if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
 726                 return;
 727
 728         if (!force) {
 729                 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
 730                     SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
 731                         return;
 732         }
 733
 734         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
 735         for (i = 0; i < entries; i++)
 736                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
 737                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
 738
 739         while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
 740                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
 741                 if (start >= max_blocks)
 742                         break;
 743
 744                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
 745                 if (force && start && end != max_blocks
 746                                         && (end - start) < cpc->trim_minlen)
 747                         continue;
 748
 749                 __add_discard_entry(sbi, cpc, se, start, end);
 750         }
 751 }
 752
 753 void release_discard_addrs(struct f2fs_sb_info *sbi)
 754 {
 755         struct list_head *head = &(SM_I(sbi)->discard_list);
 756         struct discard_entry *entry, *this;
 757
 758         /* drop caches */
 759         list_for_each_entry_safe(entry, this, head, list) {
 760                 list_del(&entry->list);
 761                 kmem_cache_free(discard_entry_slab, entry);
 762         }
 763 }
 764
 765 /*
 766  * Should call clear_prefree_segments after checkpoint is done.
 767  */
 768 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
 769 {
 770         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 771         unsigned int segno;
 772
 773         mutex_lock(&dirty_i->seglist_lock);
 774         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
 775                 __set_test_and_free(sbi, segno);
 776         mutex_unlock(&dirty_i->seglist_lock);
 777 }
 778
 779 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 780 {
 781         struct list_head *head = &(SM_I(sbi)->discard_list);
 782         struct discard_entry *entry, *this;
 783         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 784         struct blk_plug plug;
 785         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
 786         unsigned int start = 0, end = -1;
 787         unsigned int secno, start_segno;
 788         bool force = (cpc->reason == CP_DISCARD);
 789
 790         blk_start_plug(&plug);
 791
 792         mutex_lock(&dirty_i->seglist_lock);
 793
 794         while (1) {
 795                 int i;
 796                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
 797                 if (start >= MAIN_SEGS(sbi))
 798                         break;
 799                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
 800                                                                 start + 1);
 801
 802                 for (i = start; i < end; i++)
 803                         clear_bit(i, prefree_map);
 804
 805                 dirty_i->nr_dirty[PRE] -= end - start;
 806
 807                 if (force || !test_opt(sbi, DISCARD))
 808                         continue;
 809
 810                 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
 811                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
 812                                 (end - start) << sbi->log_blocks_per_seg);
 813                         continue;
 814                 }
 815 next:
 816                 secno = GET_SECNO(sbi, start);
 817                 start_segno = secno * sbi->segs_per_sec;
 818                 if (!IS_CURSEC(sbi, secno) &&
 819                         !get_valid_blocks(sbi, start, sbi->segs_per_sec))
 820                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
 821                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
 822
 823                 start = start_segno + sbi->segs_per_sec;
 824                 if (start < end)
 825                         goto next;
 826                 else
 827                         end = start - 1;
 828         }
 829         mutex_unlock(&dirty_i->seglist_lock);
 830
 831         /* send small discards */
 832         list_for_each_entry_safe(entry, this, head, list) {
 833                 if (force && entry->len < cpc->trim_minlen)
 834                         goto skip;
 835                 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
 836                 cpc->trimmed += entry->len;
 837 skip:
 838                 list_del(&entry->list);
 839                 SM_I(sbi)->nr_discards -= entry->len;
 840                 kmem_cache_free(discard_entry_slab, entry);
 841         }
 842
 843         blk_finish_plug(&plug);
 844 }
 845
 846 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
 847 {
 848         struct sit_info *sit_i = SIT_I(sbi);
 849
 850         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
 851                 sit_i->dirty_sentries++;
 852                 return false;
 853         }
 854
 855         return true;
 856 }
 857
 858 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
 859                                         unsigned int segno, int modified)
 860 {
 861         struct seg_entry *se = get_seg_entry(sbi, segno);
 862         se->type = type;
 863         if (modified)
 864                 __mark_sit_entry_dirty(sbi, segno);
 865 }
 866
 867 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
 868 {
 869         struct seg_entry *se;
 870         unsigned int segno, offset;
 871         long int new_vblocks;
 872
 873         segno = GET_SEGNO(sbi, blkaddr);
 874
 875         se = get_seg_entry(sbi, segno);
 876         new_vblocks = se->valid_blocks + del;
 877         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 878
 879         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
 880                                 (new_vblocks > sbi->blocks_per_seg)));
 881
 882         se->valid_blocks = new_vblocks;
 883         se->mtime = get_mtime(sbi);
 884         SIT_I(sbi)->max_mtime = se->mtime;
 885
 886         /* Update valid block bitmap */
 887         if (del > 0) {
 888                 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
 889                         f2fs_bug_on(sbi, 1);
 890                 if (f2fs_discard_en(sbi) &&
 891                         !f2fs_test_and_set_bit(offset, se->discard_map))
 892                         sbi->discard_blks--;
 893         } else {
 894                 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
 895                         f2fs_bug_on(sbi, 1);
 896                 if (f2fs_discard_en(sbi) &&
 897                         f2fs_test_and_clear_bit(offset, se->discard_map))
 898                         sbi->discard_blks++;
 899         }
 900         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
 901                 se->ckpt_valid_blocks += del;
 902
 903         __mark_sit_entry_dirty(sbi, segno);
 904
 905         /* update total number of valid blocks to be written in ckpt area */
 906         SIT_I(sbi)->written_valid_blocks += del;
 907
 908         if (sbi->segs_per_sec > 1)
 909                 get_sec_entry(sbi, segno)->valid_blocks += del;
 910 }
 911
 912 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
 913 {
 914         update_sit_entry(sbi, new, 1);
 915         if (GET_SEGNO(sbi, old) != NULL_SEGNO)
 916                 update_sit_entry(sbi, old, -1);
 917
 918         locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
 919         locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
 920 }
 921
 922 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
 923 {
 924         unsigned int segno = GET_SEGNO(sbi, addr);
 925         struct sit_info *sit_i = SIT_I(sbi);
 926
 927         f2fs_bug_on(sbi, addr == NULL_ADDR);
 928         if (addr == NEW_ADDR)
 929                 return;
 930
 931         /* add it into sit main buffer */
 932         mutex_lock(&sit_i->sentry_lock);
 933
 934         update_sit_entry(sbi, addr, -1);
 935
 936         /* add it into dirty seglist */
 937         locate_dirty_segment(sbi, segno);
 938
 939         mutex_unlock(&sit_i->sentry_lock);
 940 }
 941
 942 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
 943 {
 944         struct sit_info *sit_i = SIT_I(sbi);
 945         unsigned int segno, offset;
 946         struct seg_entry *se;
 947         bool is_cp = false;
 948
 949         if (!is_valid_data_blkaddr(sbi, blkaddr))
 950                 return true;
 951
 952         mutex_lock(&sit_i->sentry_lock);
 953
 954         segno = GET_SEGNO(sbi, blkaddr);
 955         se = get_seg_entry(sbi, segno);
 956         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 957
 958         if (f2fs_test_bit(offset, se->ckpt_valid_map))
 959                 is_cp = true;
 960
 961         mutex_unlock(&sit_i->sentry_lock);
 962
 963         return is_cp;
 964 }
 965
 966 /*
 967  * This function should be resided under the curseg_mutex lock
 968  */
 969 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
 970                                         struct f2fs_summary *sum)
 971 {
 972         struct curseg_info *curseg = CURSEG_I(sbi, type);
 973         void *addr = curseg->sum_blk;
 974         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
 975         memcpy(addr, sum, sizeof(struct f2fs_summary));
 976 }
 977
 978 /*
 979  * Calculate the number of current summary pages for writing
 980  */
 981 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
 982 {
 983         int valid_sum_count = 0;
 984         int i, sum_in_page;
 985
 986         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 987                 if (sbi->ckpt->alloc_type[i] == SSR)
 988                         valid_sum_count += sbi->blocks_per_seg;
 989                 else {
 990                         if (for_ra)
 991                                 valid_sum_count += le16_to_cpu(
 992                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
 993                         else
 994                                 valid_sum_count += curseg_blkoff(sbi, i);
 995                 }
 996         }
 997
 998         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
 999                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1000         if (valid_sum_count <= sum_in_page)
1001                 return 1;
1002         else if ((valid_sum_count - sum_in_page) <=
1003                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1004                 return 2;
1005         return 3;
1006 }
1007
1008 /*
1009  * Caller should put this summary page
1010  */
1011 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1012 {
1013         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1014 }
1015
1016 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1017 {
1018         struct page *page = grab_meta_page(sbi, blk_addr);
1019         void *dst = page_address(page);
1020
1021         if (src)
1022                 memcpy(dst, src, PAGE_SIZE);
1023         else
1024                 memset(dst, 0, PAGE_SIZE);
1025         set_page_dirty(page);
1026         f2fs_put_page(page, 1);
1027 }
1028
1029 static void write_sum_page(struct f2fs_sb_info *sbi,
1030                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
1031 {
1032         update_meta_page(sbi, (void *)sum_blk, blk_addr);
1033 }
1034
1035 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1036                                                 int type, block_t blk_addr)
1037 {
1038         struct curseg_info *curseg = CURSEG_I(sbi, type);
1039         struct page *page = grab_meta_page(sbi, blk_addr);
1040         struct f2fs_summary_block *src = curseg->sum_blk;
1041         struct f2fs_summary_block *dst;
1042
1043         dst = (struct f2fs_summary_block *)page_address(page);
1044
1045         mutex_lock(&curseg->curseg_mutex);
1046
1047         down_read(&curseg->journal_rwsem);
1048         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1049         up_read(&curseg->journal_rwsem);
1050
1051         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1052         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1053
1054         mutex_unlock(&curseg->curseg_mutex);
1055
1056         set_page_dirty(page);
1057         f2fs_put_page(page, 1);
1058 }
1059
1060 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1061 {
1062         struct curseg_info *curseg = CURSEG_I(sbi, type);
1063         unsigned int segno = curseg->segno + 1;
1064         struct free_segmap_info *free_i = FREE_I(sbi);
1065
1066         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
1067                 return !test_bit(segno, free_i->free_segmap);
1068         return 0;
1069 }
1070
1071 /*
1072  * Find a new segment from the free segments bitmap to right order
1073  * This function should be returned with success, otherwise BUG
1074  */
1075 static void get_new_segment(struct f2fs_sb_info *sbi,
1076                         unsigned int *newseg, bool new_sec, int dir)
1077 {
1078         struct free_segmap_info *free_i = FREE_I(sbi);
1079         unsigned int segno, secno, zoneno;
1080         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
1081         unsigned int hint = *newseg / sbi->segs_per_sec;
1082         unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1083         unsigned int left_start = hint;
1084         bool init = true;
1085         int go_left = 0;
1086         int i;
1087
1088         spin_lock(&free_i->segmap_lock);
1089
1090         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1091                 segno = find_next_zero_bit(free_i->free_segmap,
1092                                 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1093                 if (segno < (hint + 1) * sbi->segs_per_sec)
1094                         goto got_it;
1095         }
1096 find_other_zone:
1097         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1098         if (secno >= MAIN_SECS(sbi)) {
1099                 if (dir == ALLOC_RIGHT) {
1100                         secno = find_next_zero_bit(free_i->free_secmap,
1101                                                         MAIN_SECS(sbi), 0);
1102                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1103                 } else {
1104                         go_left = 1;
1105                         left_start = hint - 1;
1106                 }
1107         }
1108         if (go_left == 0)
1109                 goto skip_left;
1110
1111         while (test_bit(left_start, free_i->free_secmap)) {
1112                 if (left_start > 0) {
1113                         left_start--;
1114                         continue;
1115                 }
1116                 left_start = find_next_zero_bit(free_i->free_secmap,
1117                                                         MAIN_SECS(sbi), 0);
1118                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1119                 break;
1120         }
1121         secno = left_start;
1122 skip_left:
1123         hint = secno;
1124         segno = secno * sbi->segs_per_sec;
1125         zoneno = secno / sbi->secs_per_zone;
1126
1127         /* give up on finding another zone */
1128         if (!init)
1129                 goto got_it;
1130         if (sbi->secs_per_zone == 1)
1131                 goto got_it;
1132         if (zoneno == old_zoneno)
1133                 goto got_it;
1134         if (dir == ALLOC_LEFT) {
1135                 if (!go_left && zoneno + 1 >= total_zones)
1136                         goto got_it;
1137                 if (go_left && zoneno == 0)
1138                         goto got_it;
1139         }
1140         for (i = 0; i < NR_CURSEG_TYPE; i++)
1141                 if (CURSEG_I(sbi, i)->zone == zoneno)
1142                         break;
1143
1144         if (i < NR_CURSEG_TYPE) {
1145                 /* zone is in user, try another */
1146                 if (go_left)
1147                         hint = zoneno * sbi->secs_per_zone - 1;
1148                 else if (zoneno + 1 >= total_zones)
1149                         hint = 0;
1150                 else
1151                         hint = (zoneno + 1) * sbi->secs_per_zone;
1152                 init = false;
1153                 goto find_other_zone;
1154         }
1155 got_it:
1156         /* set it as dirty segment in free segmap */
1157         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1158         __set_inuse(sbi, segno);
1159         *newseg = segno;
1160         spin_unlock(&free_i->segmap_lock);
1161 }
1162
1163 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1164 {
1165         struct curseg_info *curseg = CURSEG_I(sbi, type);
1166         struct summary_footer *sum_footer;
1167
1168         curseg->segno = curseg->next_segno;
1169         curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1170         curseg->next_blkoff = 0;
1171         curseg->next_segno = NULL_SEGNO;
1172
1173         sum_footer = &(curseg->sum_blk->footer);
1174         memset(sum_footer, 0, sizeof(struct summary_footer));
1175         if (IS_DATASEG(type))
1176                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1177         if (IS_NODESEG(type))
1178                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1179         __set_sit_entry_type(sbi, type, curseg->segno, modified);
1180 }
1181
1182 /*
1183  * Allocate a current working segment.
1184  * This function always allocates a free segment in LFS manner.
1185  */
1186 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1187 {
1188         struct curseg_info *curseg = CURSEG_I(sbi, type);
1189         unsigned int segno = curseg->segno;
1190         int dir = ALLOC_LEFT;
1191
1192         write_sum_page(sbi, curseg->sum_blk,
1193                                 GET_SUM_BLOCK(sbi, segno));
1194         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1195                 dir = ALLOC_RIGHT;
1196
1197         if (test_opt(sbi, NOHEAP))
1198                 dir = ALLOC_RIGHT;
1199
1200         get_new_segment(sbi, &segno, new_sec, dir);
1201         curseg->next_segno = segno;
1202         reset_curseg(sbi, type, 1);
1203         curseg->alloc_type = LFS;
1204 }
1205
1206 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1207                         struct curseg_info *seg, block_t start)
1208 {
1209         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1210         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1211         unsigned long *target_map = SIT_I(sbi)->tmp_map;
1212         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1213         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1214         int i, pos;
1215
1216         for (i = 0; i < entries; i++)
1217                 target_map[i] = ckpt_map[i] | cur_map[i];
1218
1219         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1220
1221         seg->next_blkoff = pos;
1222 }
1223
1224 /*
1225  * If a segment is written by LFS manner, next block offset is just obtained
1226  * by increasing the current block offset. However, if a segment is written by
1227  * SSR manner, next block offset obtained by calling __next_free_blkoff
1228  */
1229 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1230                                 struct curseg_info *seg)
1231 {
1232         if (seg->alloc_type == SSR)
1233                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1234         else
1235                 seg->next_blkoff++;
1236 }
1237
1238 /*
1239  * This function always allocates a used segment(from dirty seglist) by SSR
1240  * manner, so it should recover the existing segment information of valid blocks
1241  */
1242 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1243 {
1244         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1245         struct curseg_info *curseg = CURSEG_I(sbi, type);
1246         unsigned int new_segno = curseg->next_segno;
1247         struct f2fs_summary_block *sum_node;
1248         struct page *sum_page;
1249
1250         write_sum_page(sbi, curseg->sum_blk,
1251                                 GET_SUM_BLOCK(sbi, curseg->segno));
1252         __set_test_and_inuse(sbi, new_segno);
1253
1254         mutex_lock(&dirty_i->seglist_lock);
1255         __remove_dirty_segment(sbi, new_segno, PRE);
1256         __remove_dirty_segment(sbi, new_segno, DIRTY);
1257         mutex_unlock(&dirty_i->seglist_lock);
1258
1259         reset_curseg(sbi, type, 1);
1260         curseg->alloc_type = SSR;
1261         __next_free_blkoff(sbi, curseg, 0);
1262
1263         if (reuse) {
1264                 sum_page = get_sum_page(sbi, new_segno);
1265                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1266                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1267                 f2fs_put_page(sum_page, 1);
1268         }
1269 }
1270
1271 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1272 {
1273         struct curseg_info *curseg = CURSEG_I(sbi, type);
1274         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1275
1276         if (IS_NODESEG(type))
1277                 return v_ops->get_victim(sbi,
1278                                 &(curseg)->next_segno, BG_GC, type, SSR);
1279
1280         /* For data segments, let's do SSR more intensively */
1281         for (; type >= CURSEG_HOT_DATA; type--)
1282                 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1283                                                 BG_GC, type, SSR))
1284                         return 1;
1285         return 0;
1286 }
1287
1288 /*
1289  * flush out current segment and replace it with new segment
1290  * This function should be returned with success, otherwise BUG
1291  */
1292 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1293                                                 int type, bool force)
1294 {
1295         struct curseg_info *curseg = CURSEG_I(sbi, type);
1296
1297         if (force)
1298                 new_curseg(sbi, type, true);
1299         else if (type == CURSEG_WARM_NODE)
1300                 new_curseg(sbi, type, false);
1301         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1302                 new_curseg(sbi, type, false);
1303         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1304                 change_curseg(sbi, type, true);
1305         else
1306                 new_curseg(sbi, type, false);
1307
1308         stat_inc_seg_type(sbi, curseg);
1309 }
1310
1311 static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1312 {
1313         struct curseg_info *curseg = CURSEG_I(sbi, type);
1314         unsigned int old_segno;
1315
1316         old_segno = curseg->segno;
1317         SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1318         locate_dirty_segment(sbi, old_segno);
1319 }
1320
1321 void allocate_new_segments(struct f2fs_sb_info *sbi)
1322 {
1323         int i;
1324
1325         if (test_opt(sbi, LFS))
1326                 return;
1327
1328         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1329                 __allocate_new_segments(sbi, i);
1330 }
1331
1332 static const struct segment_allocation default_salloc_ops = {
1333         .allocate_segment = allocate_segment_by_default,
1334 };
1335
1336 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1337 {
1338         __u64 start = F2FS_BYTES_TO_BLK(range->start);
1339         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1340         unsigned int start_segno, end_segno;
1341         struct cp_control cpc;
1342         int err = 0;
1343
1344         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1345                 return -EINVAL;
1346
1347         cpc.trimmed = 0;
1348         if (end <= MAIN_BLKADDR(sbi))
1349                 goto out;
1350
1351         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1352                 f2fs_msg(sbi->sb, KERN_WARNING,
1353                         "Found FS corruption, run fsck to fix.");
1354                 goto out;
1355         }
1356
1357         /* start/end segment number in main_area */
1358         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1359         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1360                                                 GET_SEGNO(sbi, end);
1361         cpc.reason = CP_DISCARD;
1362         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1363
1364         /* do checkpoint to issue discard commands safely */
1365         for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1366                 cpc.trim_start = start_segno;
1367
1368                 if (sbi->discard_blks == 0)
1369                         break;
1370                 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1371                         cpc.trim_end = end_segno;
1372                 else
1373                         cpc.trim_end = min_t(unsigned int,
1374                                 rounddown(start_segno +
1375                                 BATCHED_TRIM_SEGMENTS(sbi),
1376                                 sbi->segs_per_sec) - 1, end_segno);
1377
1378                 mutex_lock(&sbi->gc_mutex);
1379                 err = write_checkpoint(sbi, &cpc);
1380                 mutex_unlock(&sbi->gc_mutex);
1381                 if (err)
1382                         break;
1383
1384                 schedule();
1385         }
1386 out:
1387         range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1388         return err;
1389 }
1390
1391 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1392 {
1393         struct curseg_info *curseg = CURSEG_I(sbi, type);
1394         if (curseg->next_blkoff < sbi->blocks_per_seg)
1395                 return true;
1396         return false;
1397 }
1398
1399 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1400 {
1401         if (p_type == DATA)
1402                 return CURSEG_HOT_DATA;
1403         else
1404                 return CURSEG_HOT_NODE;
1405 }
1406
1407 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1408 {
1409         if (p_type == DATA) {
1410                 struct inode *inode = page->mapping->host;
1411
1412                 if (S_ISDIR(inode->i_mode))
1413                         return CURSEG_HOT_DATA;
1414                 else
1415                         return CURSEG_COLD_DATA;
1416         } else {
1417                 if (IS_DNODE(page) && is_cold_node(page))
1418                         return CURSEG_WARM_NODE;
1419                 else
1420                         return CURSEG_COLD_NODE;
1421         }
1422 }
1423
1424 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1425 {
1426         if (p_type == DATA) {
1427                 struct inode *inode = page->mapping->host;
1428
1429                 if (S_ISDIR(inode->i_mode))
1430                         return CURSEG_HOT_DATA;
1431                 else if (is_cold_data(page) || file_is_cold(inode))
1432                         return CURSEG_COLD_DATA;
1433                 else
1434                         return CURSEG_WARM_DATA;
1435         } else {
1436                 if (IS_DNODE(page))
1437                         return is_cold_node(page) ? CURSEG_WARM_NODE :
1438                                                 CURSEG_HOT_NODE;
1439                 else
1440                         return CURSEG_COLD_NODE;
1441         }
1442 }
1443
1444 static int __get_segment_type(struct page *page, enum page_type p_type)
1445 {
1446         switch (F2FS_P_SB(page)->active_logs) {
1447         case 2:
1448                 return __get_segment_type_2(page, p_type);
1449         case 4:
1450                 return __get_segment_type_4(page, p_type);
1451         }
1452         /* NR_CURSEG_TYPE(6) logs by default */
1453         f2fs_bug_on(F2FS_P_SB(page),
1454                 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1455         return __get_segment_type_6(page, p_type);
1456 }
1457
1458 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1459                 block_t old_blkaddr, block_t *new_blkaddr,
1460                 struct f2fs_summary *sum, int type)
1461 {
1462         struct sit_info *sit_i = SIT_I(sbi);
1463         struct curseg_info *curseg;
1464         bool direct_io = (type == CURSEG_DIRECT_IO);
1465
1466         type = direct_io ? CURSEG_WARM_DATA : type;
1467
1468         curseg = CURSEG_I(sbi, type);
1469
1470         mutex_lock(&curseg->curseg_mutex);
1471         mutex_lock(&sit_i->sentry_lock);
1472
1473         /* direct_io'ed data is aligned to the segment for better performance */
1474         if (direct_io && curseg->next_blkoff &&
1475                                 !has_not_enough_free_secs(sbi, 0, 0))
1476                 __allocate_new_segments(sbi, type);
1477
1478         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1479
1480         /*
1481          * __add_sum_entry should be resided under the curseg_mutex
1482          * because, this function updates a summary entry in the
1483          * current summary block.
1484          */
1485         __add_sum_entry(sbi, type, sum);
1486
1487         __refresh_next_blkoff(sbi, curseg);
1488
1489         stat_inc_block_count(sbi, curseg);
1490
1491         if (!__has_curseg_space(sbi, type))
1492                 sit_i->s_ops->allocate_segment(sbi, type, false);
1493         /*
1494          * SIT information should be updated before segment allocation,
1495          * since SSR needs latest valid block information.
1496          */
1497         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1498
1499         mutex_unlock(&sit_i->sentry_lock);
1500
1501         if (page && IS_NODESEG(type))
1502                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1503
1504         mutex_unlock(&curseg->curseg_mutex);
1505 }
1506
1507 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1508 {
1509         int type = __get_segment_type(fio->page, fio->type);
1510
1511         if (fio->type == NODE || fio->type == DATA)
1512                 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1513
1514         allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1515                                         &fio->new_blkaddr, sum, type);
1516
1517         /* writeout dirty page into bdev */
1518         f2fs_submit_page_mbio(fio);
1519
1520         if (fio->type == NODE || fio->type == DATA)
1521                 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
1522 }
1523
1524 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1525 {
1526         struct f2fs_io_info fio = {
1527                 .sbi = sbi,
1528                 .type = META,
1529                 .op = REQ_OP_WRITE,
1530                 .op_flags = WRITE_SYNC | REQ_META | REQ_PRIO,
1531                 .old_blkaddr = page->index,
1532                 .new_blkaddr = page->index,
1533                 .page = page,
1534                 .encrypted_page = NULL,
1535         };
1536
1537         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1538                 fio.op_flags &= ~REQ_META;
1539
1540         set_page_writeback(page);
1541         f2fs_submit_page_mbio(&fio);
1542 }
1543
1544 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1545 {
1546         struct f2fs_summary sum;
1547
1548         set_summary(&sum, nid, 0, 0);
1549         do_write_page(&sum, fio);
1550 }
1551
1552 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1553 {
1554         struct f2fs_sb_info *sbi = fio->sbi;
1555         struct f2fs_summary sum;
1556         struct node_info ni;
1557
1558         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1559         get_node_info(sbi, dn->nid, &ni);
1560         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1561         do_write_page(&sum, fio);
1562         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
1563 }
1564
1565 void rewrite_data_page(struct f2fs_io_info *fio)
1566 {
1567         fio->new_blkaddr = fio->old_blkaddr;
1568         stat_inc_inplace_blocks(fio->sbi);
1569         f2fs_submit_page_mbio(fio);
1570 }
1571
1572 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1573                                 block_t old_blkaddr, block_t new_blkaddr,
1574                                 bool recover_curseg, bool recover_newaddr)
1575 {
1576         struct sit_info *sit_i = SIT_I(sbi);
1577         struct curseg_info *curseg;
1578         unsigned int segno, old_cursegno;
1579         struct seg_entry *se;
1580         int type;
1581         unsigned short old_blkoff;
1582
1583         segno = GET_SEGNO(sbi, new_blkaddr);
1584         se = get_seg_entry(sbi, segno);
1585         type = se->type;
1586
1587         if (!recover_curseg) {
1588                 /* for recovery flow */
1589                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1590                         if (old_blkaddr == NULL_ADDR)
1591                                 type = CURSEG_COLD_DATA;
1592                         else
1593                                 type = CURSEG_WARM_DATA;
1594                 }
1595         } else {
1596                 if (!IS_CURSEG(sbi, segno))
1597                         type = CURSEG_WARM_DATA;
1598         }
1599
1600         curseg = CURSEG_I(sbi, type);
1601
1602         mutex_lock(&curseg->curseg_mutex);
1603         mutex_lock(&sit_i->sentry_lock);
1604
1605         old_cursegno = curseg->segno;
1606         old_blkoff = curseg->next_blkoff;
1607
1608         /* change the current segment */
1609         if (segno != curseg->segno) {
1610                 curseg->next_segno = segno;
1611                 change_curseg(sbi, type, true);
1612         }
1613
1614         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1615         __add_sum_entry(sbi, type, sum);
1616
1617         if (!recover_curseg || recover_newaddr)
1618                 update_sit_entry(sbi, new_blkaddr, 1);
1619         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1620                 update_sit_entry(sbi, old_blkaddr, -1);
1621
1622         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1623         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1624
1625         locate_dirty_segment(sbi, old_cursegno);
1626
1627         if (recover_curseg) {
1628                 if (old_cursegno != curseg->segno) {
1629                         curseg->next_segno = old_cursegno;
1630                         change_curseg(sbi, type, true);
1631                 }
1632                 curseg->next_blkoff = old_blkoff;
1633         }
1634
1635         mutex_unlock(&sit_i->sentry_lock);
1636         mutex_unlock(&curseg->curseg_mutex);
1637 }
1638
1639 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1640                                 block_t old_addr, block_t new_addr,
1641                                 unsigned char version, bool recover_curseg,
1642                                 bool recover_newaddr)
1643 {
1644         struct f2fs_summary sum;
1645
1646         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1647
1648         __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1649                                         recover_curseg, recover_newaddr);
1650
1651         f2fs_update_data_blkaddr(dn, new_addr);
1652 }
1653
1654 void f2fs_wait_on_page_writeback(struct page *page,
1655                                 enum page_type type, bool ordered)
1656 {
1657         if (PageWriteback(page)) {
1658                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1659
1660                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
1661                 if (ordered)
1662                         wait_on_page_writeback(page);
1663                 else
1664                         wait_for_stable_page(page);
1665         }
1666 }
1667
1668 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1669                                                         block_t blkaddr)
1670 {
1671         struct page *cpage;
1672
1673         if (!is_valid_data_blkaddr(sbi, blkaddr))
1674                 return;
1675
1676         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1677         if (cpage) {
1678                 f2fs_wait_on_page_writeback(cpage, DATA, true);
1679                 f2fs_put_page(cpage, 1);
1680         }
1681 }
1682
1683 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1684 {
1685         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1686         struct curseg_info *seg_i;
1687         unsigned char *kaddr;
1688         struct page *page;
1689         block_t start;
1690         int i, j, offset;
1691
1692         start = start_sum_block(sbi);
1693
1694         page = get_meta_page(sbi, start++);
1695         kaddr = (unsigned char *)page_address(page);
1696
1697         /* Step 1: restore nat cache */
1698         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1699         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
1700
1701         /* Step 2: restore sit cache */
1702         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1703         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
1704         offset = 2 * SUM_JOURNAL_SIZE;
1705
1706         /* Step 3: restore summary entries */
1707         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1708                 unsigned short blk_off;
1709                 unsigned int segno;
1710
1711                 seg_i = CURSEG_I(sbi, i);
1712                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1713                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1714                 seg_i->next_segno = segno;
1715                 reset_curseg(sbi, i, 0);
1716                 seg_i->alloc_type = ckpt->alloc_type[i];
1717                 seg_i->next_blkoff = blk_off;
1718
1719                 if (seg_i->alloc_type == SSR)
1720                         blk_off = sbi->blocks_per_seg;
1721
1722                 for (j = 0; j < blk_off; j++) {
1723                         struct f2fs_summary *s;
1724                         s = (struct f2fs_summary *)(kaddr + offset);
1725                         seg_i->sum_blk->entries[j] = *s;
1726                         offset += SUMMARY_SIZE;
1727                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
1728                                                 SUM_FOOTER_SIZE)
1729                                 continue;
1730
1731                         f2fs_put_page(page, 1);
1732                         page = NULL;
1733
1734                         page = get_meta_page(sbi, start++);
1735                         kaddr = (unsigned char *)page_address(page);
1736                         offset = 0;
1737                 }
1738         }
1739         f2fs_put_page(page, 1);
1740         return 0;
1741 }
1742
1743 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1744 {
1745         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1746         struct f2fs_summary_block *sum;
1747         struct curseg_info *curseg;
1748         struct page *new;
1749         unsigned short blk_off;
1750         unsigned int segno = 0;
1751         block_t blk_addr = 0;
1752
1753         /* get segment number and block addr */
1754         if (IS_DATASEG(type)) {
1755                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1756                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1757                                                         CURSEG_HOT_DATA]);
1758                 if (__exist_node_summaries(sbi))
1759                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1760                 else
1761                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1762         } else {
1763                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1764                                                         CURSEG_HOT_NODE]);
1765                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1766                                                         CURSEG_HOT_NODE]);
1767                 if (__exist_node_summaries(sbi))
1768                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1769                                                         type - CURSEG_HOT_NODE);
1770                 else
1771                         blk_addr = GET_SUM_BLOCK(sbi, segno);
1772         }
1773
1774         new = get_meta_page(sbi, blk_addr);
1775         sum = (struct f2fs_summary_block *)page_address(new);
1776
1777         if (IS_NODESEG(type)) {
1778                 if (__exist_node_summaries(sbi)) {
1779                         struct f2fs_summary *ns = &sum->entries[0];
1780                         int i;
1781                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1782                                 ns->version = 0;
1783                                 ns->ofs_in_node = 0;
1784                         }
1785                 } else {
1786                         int err;
1787
1788                         err = restore_node_summary(sbi, segno, sum);
1789                         if (err) {
1790                                 f2fs_put_page(new, 1);
1791                                 return err;
1792                         }
1793                 }
1794         }
1795
1796         /* set uncompleted segment to curseg */
1797         curseg = CURSEG_I(sbi, type);
1798         mutex_lock(&curseg->curseg_mutex);
1799
1800         /* update journal info */
1801         down_write(&curseg->journal_rwsem);
1802         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1803         up_write(&curseg->journal_rwsem);
1804
1805         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1806         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
1807         curseg->next_segno = segno;
1808         reset_curseg(sbi, type, 0);
1809         curseg->alloc_type = ckpt->alloc_type[type];
1810         curseg->next_blkoff = blk_off;
1811         mutex_unlock(&curseg->curseg_mutex);
1812         f2fs_put_page(new, 1);
1813         return 0;
1814 }
1815
1816 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1817 {
1818         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
1819         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
1820         int type = CURSEG_HOT_DATA;
1821         int err;
1822
1823         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
1824                 int npages = npages_for_summary_flush(sbi, true);
1825
1826                 if (npages >= 2)
1827                         ra_meta_pages(sbi, start_sum_block(sbi), npages,
1828                                                         META_CP, true);
1829
1830                 /* restore for compacted data summary */
1831                 if (read_compacted_summaries(sbi))
1832                         return -EINVAL;
1833                 type = CURSEG_HOT_NODE;
1834         }
1835
1836         if (__exist_node_summaries(sbi))
1837                 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1838                                         NR_CURSEG_TYPE - type, META_CP, true);
1839
1840         for (; type <= CURSEG_COLD_NODE; type++) {
1841                 err = read_normal_summaries(sbi, type);
1842                 if (err)
1843                         return err;
1844         }
1845
1846         /* sanity check for summary blocks */
1847         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
1848                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
1849                 return -EINVAL;
1850
1851         return 0;
1852 }
1853
1854 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1855 {
1856         struct page *page;
1857         unsigned char *kaddr;
1858         struct f2fs_summary *summary;
1859         struct curseg_info *seg_i;
1860         int written_size = 0;
1861         int i, j;
1862
1863         page = grab_meta_page(sbi, blkaddr++);
1864         kaddr = (unsigned char *)page_address(page);
1865
1866         /* Step 1: write nat cache */
1867         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1868         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
1869         written_size += SUM_JOURNAL_SIZE;
1870
1871         /* Step 2: write sit cache */
1872         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1873         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
1874         written_size += SUM_JOURNAL_SIZE;
1875
1876         /* Step 3: write summary entries */
1877         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1878                 unsigned short blkoff;
1879                 seg_i = CURSEG_I(sbi, i);
1880                 if (sbi->ckpt->alloc_type[i] == SSR)
1881                         blkoff = sbi->blocks_per_seg;
1882                 else
1883                         blkoff = curseg_blkoff(sbi, i);
1884
1885                 for (j = 0; j < blkoff; j++) {
1886                         if (!page) {
1887                                 page = grab_meta_page(sbi, blkaddr++);
1888                                 kaddr = (unsigned char *)page_address(page);
1889                                 written_size = 0;
1890                         }
1891                         summary = (struct f2fs_summary *)(kaddr + written_size);
1892                         *summary = seg_i->sum_blk->entries[j];
1893                         written_size += SUMMARY_SIZE;
1894
1895                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
1896                                                         SUM_FOOTER_SIZE)
1897                                 continue;
1898
1899                         set_page_dirty(page);
1900                         f2fs_put_page(page, 1);
1901                         page = NULL;
1902                 }
1903         }
1904         if (page) {
1905                 set_page_dirty(page);
1906                 f2fs_put_page(page, 1);
1907         }
1908 }
1909
1910 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1911                                         block_t blkaddr, int type)
1912 {
1913         int i, end;
1914         if (IS_DATASEG(type))
1915                 end = type + NR_CURSEG_DATA_TYPE;
1916         else
1917                 end = type + NR_CURSEG_NODE_TYPE;
1918
1919         for (i = type; i < end; i++)
1920                 write_current_sum_page(sbi, i, blkaddr + (i - type));
1921 }
1922
1923 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1924 {
1925         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
1926                 write_compacted_summaries(sbi, start_blk);
1927         else
1928                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1929 }
1930
1931 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1932 {
1933         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1934 }
1935
1936 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
1937                                         unsigned int val, int alloc)
1938 {
1939         int i;
1940
1941         if (type == NAT_JOURNAL) {
1942                 for (i = 0; i < nats_in_cursum(journal); i++) {
1943                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
1944                                 return i;
1945                 }
1946                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
1947                         return update_nats_in_cursum(journal, 1);
1948         } else if (type == SIT_JOURNAL) {
1949                 for (i = 0; i < sits_in_cursum(journal); i++)
1950                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
1951                                 return i;
1952                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
1953                         return update_sits_in_cursum(journal, 1);
1954         }
1955         return -1;
1956 }
1957
1958 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1959                                         unsigned int segno)
1960 {
1961         return get_meta_page(sbi, current_sit_addr(sbi, segno));
1962 }
1963
1964 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1965                                         unsigned int start)
1966 {
1967         struct sit_info *sit_i = SIT_I(sbi);
1968         struct page *src_page, *dst_page;
1969         pgoff_t src_off, dst_off;
1970         void *src_addr, *dst_addr;
1971
1972         src_off = current_sit_addr(sbi, start);
1973         dst_off = next_sit_addr(sbi, src_off);
1974
1975         /* get current sit block page without lock */
1976         src_page = get_meta_page(sbi, src_off);
1977         dst_page = grab_meta_page(sbi, dst_off);
1978         f2fs_bug_on(sbi, PageDirty(src_page));
1979
1980         src_addr = page_address(src_page);
1981         dst_addr = page_address(dst_page);
1982         memcpy(dst_addr, src_addr, PAGE_SIZE);
1983
1984         set_page_dirty(dst_page);
1985         f2fs_put_page(src_page, 1);
1986
1987         set_to_next_sit(sit_i, start);
1988
1989         return dst_page;
1990 }
1991
1992 static struct sit_entry_set *grab_sit_entry_set(void)
1993 {
1994         struct sit_entry_set *ses =
1995                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
1996
1997         ses->entry_cnt = 0;
1998         INIT_LIST_HEAD(&ses->set_list);
1999         return ses;
2000 }
2001
2002 static void release_sit_entry_set(struct sit_entry_set *ses)
2003 {
2004         list_del(&ses->set_list);
2005         kmem_cache_free(sit_entry_set_slab, ses);
2006 }
2007
2008 static void adjust_sit_entry_set(struct sit_entry_set *ses,
2009                                                 struct list_head *head)
2010 {
2011         struct sit_entry_set *next = ses;
2012
2013         if (list_is_last(&ses->set_list, head))
2014                 return;
2015
2016         list_for_each_entry_continue(next, head, set_list)
2017                 if (ses->entry_cnt <= next->entry_cnt)
2018                         break;
2019
2020         list_move_tail(&ses->set_list, &next->set_list);
2021 }
2022
2023 static void add_sit_entry(unsigned int segno, struct list_head *head)
2024 {
2025         struct sit_entry_set *ses;
2026         unsigned int start_segno = START_SEGNO(segno);
2027
2028         list_for_each_entry(ses, head, set_list) {
2029                 if (ses->start_segno == start_segno) {
2030                         ses->entry_cnt++;
2031                         adjust_sit_entry_set(ses, head);
2032                         return;
2033                 }
2034         }
2035
2036         ses = grab_sit_entry_set();
2037
2038         ses->start_segno = start_segno;
2039         ses->entry_cnt++;
2040         list_add(&ses->set_list, head);
2041 }
2042
2043 static void add_sits_in_set(struct f2fs_sb_info *sbi)
2044 {
2045         struct f2fs_sm_info *sm_info = SM_I(sbi);
2046         struct list_head *set_list = &sm_info->sit_entry_set;
2047         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
2048         unsigned int segno;
2049
2050         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
2051                 add_sit_entry(segno, set_list);
2052 }
2053
2054 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
2055 {
2056         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2057         struct f2fs_journal *journal = curseg->journal;
2058         int i;
2059
2060         down_write(&curseg->journal_rwsem);
2061         for (i = 0; i < sits_in_cursum(journal); i++) {
2062                 unsigned int segno;
2063                 bool dirtied;
2064
2065                 segno = le32_to_cpu(segno_in_journal(journal, i));
2066                 dirtied = __mark_sit_entry_dirty(sbi, segno);
2067
2068                 if (!dirtied)
2069                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
2070         }
2071         update_sits_in_cursum(journal, -i);
2072         up_write(&curseg->journal_rwsem);
2073 }
2074
2075 /*
2076  * CP calls this function, which flushes SIT entries including sit_journal,
2077  * and moves prefree segs to free segs.
2078  */
2079 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2080 {
2081         struct sit_info *sit_i = SIT_I(sbi);
2082         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2083         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2084         struct f2fs_journal *journal = curseg->journal;
2085         struct sit_entry_set *ses, *tmp;
2086         struct list_head *head = &SM_I(sbi)->sit_entry_set;
2087         bool to_journal = true;
2088         struct seg_entry *se;
2089
2090         mutex_lock(&sit_i->sentry_lock);
2091
2092         if (!sit_i->dirty_sentries)
2093                 goto out;
2094
2095         /*
2096          * add and account sit entries of dirty bitmap in sit entry
2097          * set temporarily
2098          */
2099         add_sits_in_set(sbi);
2100
2101         /*
2102          * if there are no enough space in journal to store dirty sit
2103          * entries, remove all entries from journal and add and account
2104          * them in sit entry set.
2105          */
2106         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
2107                 remove_sits_in_journal(sbi);
2108
2109         /*
2110          * there are two steps to flush sit entries:
2111          * #1, flush sit entries to journal in current cold data summary block.
2112          * #2, flush sit entries to sit page.
2113          */
2114         list_for_each_entry_safe(ses, tmp, head, set_list) {
2115                 struct page *page = NULL;
2116                 struct f2fs_sit_block *raw_sit = NULL;
2117                 unsigned int start_segno = ses->start_segno;
2118                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
2119                                                 (unsigned long)MAIN_SEGS(sbi));
2120                 unsigned int segno = start_segno;
2121
2122                 if (to_journal &&
2123                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
2124                         to_journal = false;
2125
2126                 if (to_journal) {
2127                         down_write(&curseg->journal_rwsem);
2128                 } else {
2129                         page = get_next_sit_page(sbi, start_segno);
2130                         raw_sit = page_address(page);
2131                 }
2132
2133                 /* flush dirty sit entries in region of current sit set */
2134                 for_each_set_bit_from(segno, bitmap, end) {
2135                         int offset, sit_offset;
2136
2137                         se = get_seg_entry(sbi, segno);
2138
2139                         /* add discard candidates */
2140                         if (cpc->reason != CP_DISCARD) {
2141                                 cpc->trim_start = segno;
2142                                 add_discard_addrs(sbi, cpc);
2143                         }
2144
2145                         if (to_journal) {
2146                                 offset = lookup_journal_in_cursum(journal,
2147                                                         SIT_JOURNAL, segno, 1);
2148                                 f2fs_bug_on(sbi, offset < 0);
2149                                 segno_in_journal(journal, offset) =
2150                                                         cpu_to_le32(segno);
2151                                 seg_info_to_raw_sit(se,
2152                                         &sit_in_journal(journal, offset));
2153                         } else {
2154                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2155                                 seg_info_to_raw_sit(se,
2156                                                 &raw_sit->entries[sit_offset]);
2157                         }
2158
2159                         __clear_bit(segno, bitmap);
2160                         sit_i->dirty_sentries--;
2161                         ses->entry_cnt--;
2162                 }
2163
2164                 if (to_journal)
2165                         up_write(&curseg->journal_rwsem);
2166                 else
2167                         f2fs_put_page(page, 1);
2168
2169                 f2fs_bug_on(sbi, ses->entry_cnt);
2170                 release_sit_entry_set(ses);
2171         }
2172
2173         f2fs_bug_on(sbi, !list_empty(head));
2174         f2fs_bug_on(sbi, sit_i->dirty_sentries);
2175 out:
2176         if (cpc->reason == CP_DISCARD) {
2177                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2178                         add_discard_addrs(sbi, cpc);
2179         }
2180         mutex_unlock(&sit_i->sentry_lock);
2181
2182         set_prefree_as_free_segments(sbi);
2183 }
2184
2185 static int build_sit_info(struct f2fs_sb_info *sbi)
2186 {
2187         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2188         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2189         struct sit_info *sit_i;
2190         unsigned int sit_segs, start;
2191         char *src_bitmap, *dst_bitmap;
2192         unsigned int bitmap_size;
2193
2194         /* allocate memory for SIT information */
2195         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2196         if (!sit_i)
2197                 return -ENOMEM;
2198
2199         SM_I(sbi)->sit_info = sit_i;
2200
2201         sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2202                                         sizeof(struct seg_entry), GFP_KERNEL);
2203         if (!sit_i->sentries)
2204                 return -ENOMEM;
2205
2206         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2207         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2208         if (!sit_i->dirty_sentries_bitmap)
2209                 return -ENOMEM;
2210
2211         for (start = 0; start < MAIN_SEGS(sbi); start++) {
2212                 sit_i->sentries[start].cur_valid_map
2213                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2214                 sit_i->sentries[start].ckpt_valid_map
2215                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2216                 if (!sit_i->sentries[start].cur_valid_map ||
2217                                 !sit_i->sentries[start].ckpt_valid_map)
2218                         return -ENOMEM;
2219
2220                 if (f2fs_discard_en(sbi)) {
2221                         sit_i->sentries[start].discard_map
2222                                 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2223                         if (!sit_i->sentries[start].discard_map)
2224                                 return -ENOMEM;
2225                 }
2226         }
2227
2228         sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2229         if (!sit_i->tmp_map)
2230                 return -ENOMEM;
2231
2232         if (sbi->segs_per_sec > 1) {
2233                 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2234                                         sizeof(struct sec_entry), GFP_KERNEL);
2235                 if (!sit_i->sec_entries)
2236                         return -ENOMEM;
2237         }
2238
2239         /* get information related with SIT */
2240         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2241
2242         /* setup SIT bitmap from ckeckpoint pack */
2243         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2244         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2245
2246         dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2247         if (!dst_bitmap)
2248                 return -ENOMEM;
2249
2250         /* init SIT information */
2251         sit_i->s_ops = &default_salloc_ops;
2252
2253         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2254         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2255         sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2256         sit_i->sit_bitmap = dst_bitmap;
2257         sit_i->bitmap_size = bitmap_size;
2258         sit_i->dirty_sentries = 0;
2259         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2260         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2261         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2262         mutex_init(&sit_i->sentry_lock);
2263         return 0;
2264 }
2265
2266 static int build_free_segmap(struct f2fs_sb_info *sbi)
2267 {
2268         struct free_segmap_info *free_i;
2269         unsigned int bitmap_size, sec_bitmap_size;
2270
2271         /* allocate memory for free segmap information */
2272         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2273         if (!free_i)
2274                 return -ENOMEM;
2275
2276         SM_I(sbi)->free_info = free_i;
2277
2278         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2279         free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2280         if (!free_i->free_segmap)
2281                 return -ENOMEM;
2282
2283         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2284         free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2285         if (!free_i->free_secmap)
2286                 return -ENOMEM;
2287
2288         /* set all segments as dirty temporarily */
2289         memset(free_i->free_segmap, 0xff, bitmap_size);
2290         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2291
2292         /* init free segmap information */
2293         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2294         free_i->free_segments = 0;
2295         free_i->free_sections = 0;
2296         spin_lock_init(&free_i->segmap_lock);
2297         return 0;
2298 }
2299
2300 static int build_curseg(struct f2fs_sb_info *sbi)
2301 {
2302         struct curseg_info *array;
2303         int i;
2304
2305         array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2306         if (!array)
2307                 return -ENOMEM;
2308
2309         SM_I(sbi)->curseg_array = array;
2310
2311         for (i = 0; i < NR_CURSEG_TYPE; i++) {
2312                 mutex_init(&array[i].curseg_mutex);
2313                 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2314                 if (!array[i].sum_blk)
2315                         return -ENOMEM;
2316                 init_rwsem(&array[i].journal_rwsem);
2317                 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2318                                                         GFP_KERNEL);
2319                 if (!array[i].journal)
2320                         return -ENOMEM;
2321                 array[i].segno = NULL_SEGNO;
2322                 array[i].next_blkoff = 0;
2323         }
2324         return restore_curseg_summaries(sbi);
2325 }
2326
2327 static int build_sit_entries(struct f2fs_sb_info *sbi)
2328 {
2329         struct sit_info *sit_i = SIT_I(sbi);
2330         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2331         struct f2fs_journal *journal = curseg->journal;
2332         struct seg_entry *se;
2333         struct f2fs_sit_entry sit;
2334         int sit_blk_cnt = SIT_BLK_CNT(sbi);
2335         unsigned int i, start, end;
2336         unsigned int readed, start_blk = 0;
2337         int nrpages = MAX_BIO_BLOCKS(sbi) * 8;
2338         int err = 0;
2339
2340         do {
2341                 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
2342
2343                 start = start_blk * sit_i->sents_per_block;
2344                 end = (start_blk + readed) * sit_i->sents_per_block;
2345
2346                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2347                         struct f2fs_sit_block *sit_blk;
2348                         struct page *page;
2349
2350                         se = &sit_i->sentries[start];
2351                         page = get_current_sit_page(sbi, start);
2352                         sit_blk = (struct f2fs_sit_block *)page_address(page);
2353                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2354                         f2fs_put_page(page, 1);
2355
2356                         err = check_block_count(sbi, start, &sit);
2357                         if (err)
2358                                 return err;
2359                         seg_info_from_raw_sit(se, &sit);
2360
2361                         /* build discard map only one time */
2362                         if (f2fs_discard_en(sbi)) {
2363                                 memcpy(se->discard_map, se->cur_valid_map,
2364                                                         SIT_VBLOCK_MAP_SIZE);
2365                                 sbi->discard_blks += sbi->blocks_per_seg -
2366                                                         se->valid_blocks;
2367                         }
2368
2369                         if (sbi->segs_per_sec > 1)
2370                                 get_sec_entry(sbi, start)->valid_blocks +=
2371                                                         se->valid_blocks;
2372                 }
2373                 start_blk += readed;
2374         } while (start_blk < sit_blk_cnt);
2375
2376         down_read(&curseg->journal_rwsem);
2377         for (i = 0; i < sits_in_cursum(journal); i++) {
2378                 unsigned int old_valid_blocks;
2379
2380                 start = le32_to_cpu(segno_in_journal(journal, i));
2381                 if (start >= MAIN_SEGS(sbi)) {
2382                         f2fs_msg(sbi->sb, KERN_ERR,
2383                                         "Wrong journal entry on segno %u",
2384                                         start);
2385                         set_sbi_flag(sbi, SBI_NEED_FSCK);
2386                         err = -EINVAL;
2387                         break;
2388                 }
2389
2390                 se = &sit_i->sentries[start];
2391                 sit = sit_in_journal(journal, i);
2392
2393                 old_valid_blocks = se->valid_blocks;
2394
2395                 err = check_block_count(sbi, start, &sit);
2396                 if (err)
2397                         break;
2398                 seg_info_from_raw_sit(se, &sit);
2399
2400                 if (f2fs_discard_en(sbi)) {
2401                         memcpy(se->discard_map, se->cur_valid_map,
2402                                                 SIT_VBLOCK_MAP_SIZE);
2403                         sbi->discard_blks += old_valid_blocks -
2404                                                 se->valid_blocks;
2405                 }
2406
2407                 if (sbi->segs_per_sec > 1)
2408                         get_sec_entry(sbi, start)->valid_blocks +=
2409                                 se->valid_blocks - old_valid_blocks;
2410         }
2411         up_read(&curseg->journal_rwsem);
2412         return err;
2413 }
2414
2415 static void init_free_segmap(struct f2fs_sb_info *sbi)
2416 {
2417         unsigned int start;
2418         int type;
2419
2420         for (start = 0; start < MAIN_SEGS(sbi); start++) {
2421                 struct seg_entry *sentry = get_seg_entry(sbi, start);
2422                 if (!sentry->valid_blocks)
2423                         __set_free(sbi, start);
2424         }
2425
2426         /* set use the current segments */
2427         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2428                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2429                 __set_test_and_inuse(sbi, curseg_t->segno);
2430         }
2431 }
2432
2433 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2434 {
2435         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2436         struct free_segmap_info *free_i = FREE_I(sbi);
2437         unsigned int segno = 0, offset = 0;
2438         unsigned short valid_blocks;
2439
2440         while (1) {
2441                 /* find dirty segment based on free segmap */
2442                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2443                 if (segno >= MAIN_SEGS(sbi))
2444                         break;
2445                 offset = segno + 1;
2446                 valid_blocks = get_valid_blocks(sbi, segno, 0);
2447                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2448                         continue;
2449                 if (valid_blocks > sbi->blocks_per_seg) {
2450                         f2fs_bug_on(sbi, 1);
2451                         continue;
2452                 }
2453                 mutex_lock(&dirty_i->seglist_lock);
2454                 __locate_dirty_segment(sbi, segno, DIRTY);
2455                 mutex_unlock(&dirty_i->seglist_lock);
2456         }
2457 }
2458
2459 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2460 {
2461         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2462         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2463
2464         dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2465         if (!dirty_i->victim_secmap)
2466                 return -ENOMEM;
2467         return 0;
2468 }
2469
2470 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2471 {
2472         struct dirty_seglist_info *dirty_i;
2473         unsigned int bitmap_size, i;
2474
2475         /* allocate memory for dirty segments list information */
2476         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2477         if (!dirty_i)
2478                 return -ENOMEM;
2479
2480         SM_I(sbi)->dirty_info = dirty_i;
2481         mutex_init(&dirty_i->seglist_lock);
2482
2483         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2484
2485         for (i = 0; i < NR_DIRTY_TYPE; i++) {
2486                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2487                 if (!dirty_i->dirty_segmap[i])
2488                         return -ENOMEM;
2489         }
2490
2491         init_dirty_segmap(sbi);
2492         return init_victim_secmap(sbi);
2493 }
2494
2495 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
2496 {
2497         int i;
2498
2499         /*
2500          * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
2501          * In LFS curseg, all blkaddr after .next_blkoff should be unused.
2502          */
2503         for (i = 0; i < NO_CHECK_TYPE; i++) {
2504                 struct curseg_info *curseg = CURSEG_I(sbi, i);
2505                 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
2506                 unsigned int blkofs = curseg->next_blkoff;
2507
2508                 if (f2fs_test_bit(blkofs, se->cur_valid_map))
2509                         goto out;
2510
2511                 if (curseg->alloc_type == SSR)
2512                         continue;
2513
2514                 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
2515                         if (!f2fs_test_bit(blkofs, se->cur_valid_map))
2516                                 continue;
2517 out:
2518                         f2fs_msg(sbi->sb, KERN_ERR,
2519                                 "Current segment's next free block offset is "
2520                                 "inconsistent with bitmap, logtype:%u, "
2521                                 "segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
2522                                 i, curseg->segno, curseg->alloc_type,
2523                                 curseg->next_blkoff, blkofs);
2524                         return -EINVAL;
2525                 }
2526         }
2527         return 0;
2528 }
2529
2530 /*
2531  * Update min, max modified time for cost-benefit GC algorithm
2532  */
2533 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2534 {
2535         struct sit_info *sit_i = SIT_I(sbi);
2536         unsigned int segno;
2537
2538         mutex_lock(&sit_i->sentry_lock);
2539
2540         sit_i->min_mtime = LLONG_MAX;
2541
2542         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2543                 unsigned int i;
2544                 unsigned long long mtime = 0;
2545
2546                 for (i = 0; i < sbi->segs_per_sec; i++)
2547                         mtime += get_seg_entry(sbi, segno + i)->mtime;
2548
2549                 mtime = div_u64(mtime, sbi->segs_per_sec);
2550
2551                 if (sit_i->min_mtime > mtime)
2552                         sit_i->min_mtime = mtime;
2553         }
2554         sit_i->max_mtime = get_mtime(sbi);
2555         mutex_unlock(&sit_i->sentry_lock);
2556 }
2557
2558 int build_segment_manager(struct f2fs_sb_info *sbi)
2559 {
2560         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2561         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2562         struct f2fs_sm_info *sm_info;
2563         int err;
2564
2565         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2566         if (!sm_info)
2567                 return -ENOMEM;
2568
2569         /* init sm info */
2570         sbi->sm_info = sm_info;
2571         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2572         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2573         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2574         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2575         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2576         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2577         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2578         sm_info->rec_prefree_segments = sm_info->main_segments *
2579                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2580         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2581                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2582
2583         if (!test_opt(sbi, LFS))
2584                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2585         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2586         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2587
2588         INIT_LIST_HEAD(&sm_info->discard_list);
2589         INIT_LIST_HEAD(&sm_info->wait_list);
2590         sm_info->nr_discards = 0;
2591         sm_info->max_discards = 0;
2592
2593         sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2594
2595         INIT_LIST_HEAD(&sm_info->sit_entry_set);
2596
2597         if (!f2fs_readonly(sbi->sb)) {
2598                 err = create_flush_cmd_control(sbi);
2599                 if (err)
2600                         return err;
2601         }
2602
2603         err = build_sit_info(sbi);
2604         if (err)
2605                 return err;
2606         err = build_free_segmap(sbi);
2607         if (err)
2608                 return err;
2609         err = build_curseg(sbi);
2610         if (err)
2611                 return err;
2612
2613         /* reinit free segmap based on SIT */
2614         err = build_sit_entries(sbi);
2615         if (err)
2616                 return err;
2617
2618         init_free_segmap(sbi);
2619         err = build_dirty_segmap(sbi);
2620         if (err)
2621                 return err;
2622
2623         err = sanity_check_curseg(sbi);
2624         if (err)
2625                 return err;
2626
2627         init_min_max_mtime(sbi);
2628         return 0;
2629 }
2630
2631 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2632                 enum dirty_type dirty_type)
2633 {
2634         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2635
2636         mutex_lock(&dirty_i->seglist_lock);
2637         kvfree(dirty_i->dirty_segmap[dirty_type]);
2638         dirty_i->nr_dirty[dirty_type] = 0;
2639         mutex_unlock(&dirty_i->seglist_lock);
2640 }
2641
2642 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2643 {
2644         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2645         kvfree(dirty_i->victim_secmap);
2646 }
2647
2648 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2649 {
2650         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2651         int i;
2652
2653         if (!dirty_i)
2654                 return;
2655
2656         /* discard pre-free/dirty segments list */
2657         for (i = 0; i < NR_DIRTY_TYPE; i++)
2658                 discard_dirty_segmap(sbi, i);
2659
2660         destroy_victim_secmap(sbi);
2661         SM_I(sbi)->dirty_info = NULL;
2662         kfree(dirty_i);
2663 }
2664
2665 static void destroy_curseg(struct f2fs_sb_info *sbi)
2666 {
2667         struct curseg_info *array = SM_I(sbi)->curseg_array;
2668         int i;
2669
2670         if (!array)
2671                 return;
2672         SM_I(sbi)->curseg_array = NULL;
2673         for (i = 0; i < NR_CURSEG_TYPE; i++) {
2674                 kfree(array[i].sum_blk);
2675                 kfree(array[i].journal);
2676         }
2677         kfree(array);
2678 }
2679
2680 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2681 {
2682         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2683         if (!free_i)
2684                 return;
2685         SM_I(sbi)->free_info = NULL;
2686         kvfree(free_i->free_segmap);
2687         kvfree(free_i->free_secmap);
2688         kfree(free_i);
2689 }
2690
2691 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2692 {
2693         struct sit_info *sit_i = SIT_I(sbi);
2694         unsigned int start;
2695
2696         if (!sit_i)
2697                 return;
2698
2699         if (sit_i->sentries) {
2700                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2701                         kfree(sit_i->sentries[start].cur_valid_map);
2702                         kfree(sit_i->sentries[start].ckpt_valid_map);
2703                         kfree(sit_i->sentries[start].discard_map);
2704                 }
2705         }
2706         kfree(sit_i->tmp_map);
2707
2708         kvfree(sit_i->sentries);
2709         kvfree(sit_i->sec_entries);
2710         kvfree(sit_i->dirty_sentries_bitmap);
2711
2712         SM_I(sbi)->sit_info = NULL;
2713         kfree(sit_i->sit_bitmap);
2714         kfree(sit_i);
2715 }
2716
2717 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2718 {
2719         struct f2fs_sm_info *sm_info = SM_I(sbi);
2720
2721         if (!sm_info)
2722                 return;
2723         destroy_flush_cmd_control(sbi);
2724         destroy_dirty_segmap(sbi);
2725         destroy_curseg(sbi);
2726         destroy_free_segmap(sbi);
2727         destroy_sit_info(sbi);
2728         sbi->sm_info = NULL;
2729         kfree(sm_info);
2730 }
2731
2732 int __init create_segment_manager_caches(void)
2733 {
2734         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2735                         sizeof(struct discard_entry));
2736         if (!discard_entry_slab)
2737                 goto fail;
2738
2739         bio_entry_slab = f2fs_kmem_cache_create("bio_entry",
2740                         sizeof(struct bio_entry));
2741         if (!bio_entry_slab)
2742                 goto destroy_discard_entry;
2743
2744         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2745                         sizeof(struct sit_entry_set));
2746         if (!sit_entry_set_slab)
2747                 goto destroy_bio_entry;
2748
2749         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2750                         sizeof(struct inmem_pages));
2751         if (!inmem_entry_slab)
2752                 goto destroy_sit_entry_set;
2753         return 0;
2754
2755 destroy_sit_entry_set:
2756         kmem_cache_destroy(sit_entry_set_slab);
2757 destroy_bio_entry:
2758         kmem_cache_destroy(bio_entry_slab);
2759 destroy_discard_entry:
2760         kmem_cache_destroy(discard_entry_slab);
2761 fail:
2762         return -ENOMEM;
2763 }
2764
2765 void destroy_segment_manager_caches(void)
2766 {
2767         kmem_cache_destroy(sit_entry_set_slab);
2768         kmem_cache_destroy(bio_entry_slab);
2769         kmem_cache_destroy(discard_entry_slab);
2770         kmem_cache_destroy(inmem_entry_slab);
2771 }