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 #include <linux/freezer.h>
  20 #include <linux/sched/signal.h>
  21
  22 #include "f2fs.h"
  23 #include "segment.h"
  24 #include "node.h"
  25 #include "gc.h"
  26 #include "trace.h"
  27 #include <trace/events/f2fs.h>
  28
  29 #define __reverse_ffz(x) __reverse_ffs(~(x))
  30
  31 static struct kmem_cache *discard_entry_slab;
  32 static struct kmem_cache *discard_cmd_slab;
  33 static struct kmem_cache *sit_entry_set_slab;
  34 static struct kmem_cache *inmem_entry_slab;
  35
  36 static unsigned long __reverse_ulong(unsigned char *str)
  37 {
  38         unsigned long tmp = 0;
  39         int shift = 24, idx = 0;
  40
  41 #if BITS_PER_LONG == 64
  42         shift = 56;
  43 #endif
  44         while (shift >= 0) {
  45                 tmp |= (unsigned long)str[idx++] << shift;
  46                 shift -= BITS_PER_BYTE;
  47         }
  48         return tmp;
  49 }
  50
  51 /*
  52  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  53  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  54  */
  55 static inline unsigned long __reverse_ffs(unsigned long word)
  56 {
  57         int num = 0;
  58
  59 #if BITS_PER_LONG == 64
  60         if ((word & 0xffffffff00000000UL) == 0)
  61                 num += 32;
  62         else
  63                 word >>= 32;
  64 #endif
  65         if ((word & 0xffff0000) == 0)
  66                 num += 16;
  67         else
  68                 word >>= 16;
  69
  70         if ((word & 0xff00) == 0)
  71                 num += 8;
  72         else
  73                 word >>= 8;
  74
  75         if ((word & 0xf0) == 0)
  76                 num += 4;
  77         else
  78                 word >>= 4;
  79
  80         if ((word & 0xc) == 0)
  81                 num += 2;
  82         else
  83                 word >>= 2;
  84
  85         if ((word & 0x2) == 0)
  86                 num += 1;
  87         return num;
  88 }
  89
  90 /*
  91  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  92  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  93  * @size must be integral times of unsigned long.
  94  * Example:
  95  *                             MSB <--> LSB
  96  *   f2fs_set_bit(0, bitmap) => 1000 0000
  97  *   f2fs_set_bit(7, bitmap) => 0000 0001
  98  */
  99 static unsigned long __find_rev_next_bit(const unsigned long *addr,
 100                         unsigned long size, unsigned long offset)
 101 {
 102         const unsigned long *p = addr + BIT_WORD(offset);
 103         unsigned long result = size;
 104         unsigned long tmp;
 105
 106         if (offset >= size)
 107                 return size;
 108
 109         size -= (offset & ~(BITS_PER_LONG - 1));
 110         offset %= BITS_PER_LONG;
 111
 112         while (1) {
 113                 if (*p == 0)
 114                         goto pass;
 115
 116                 tmp = __reverse_ulong((unsigned char *)p);
 117
 118                 tmp &= ~0UL >> offset;
 119                 if (size < BITS_PER_LONG)
 120                         tmp &= (~0UL << (BITS_PER_LONG - size));
 121                 if (tmp)
 122                         goto found;
 123 pass:
 124                 if (size <= BITS_PER_LONG)
 125                         break;
 126                 size -= BITS_PER_LONG;
 127                 offset = 0;
 128                 p++;
 129         }
 130         return result;
 131 found:
 132         return result - size + __reverse_ffs(tmp);
 133 }
 134
 135 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 136                         unsigned long size, unsigned long offset)
 137 {
 138         const unsigned long *p = addr + BIT_WORD(offset);
 139         unsigned long result = size;
 140         unsigned long tmp;
 141
 142         if (offset >= size)
 143                 return size;
 144
 145         size -= (offset & ~(BITS_PER_LONG - 1));
 146         offset %= BITS_PER_LONG;
 147
 148         while (1) {
 149                 if (*p == ~0UL)
 150                         goto pass;
 151
 152                 tmp = __reverse_ulong((unsigned char *)p);
 153
 154                 if (offset)
 155                         tmp |= ~0UL << (BITS_PER_LONG - offset);
 156                 if (size < BITS_PER_LONG)
 157                         tmp |= ~0UL >> size;
 158                 if (tmp != ~0UL)
 159                         goto found;
 160 pass:
 161                 if (size <= BITS_PER_LONG)
 162                         break;
 163                 size -= BITS_PER_LONG;
 164                 offset = 0;
 165                 p++;
 166         }
 167         return result;
 168 found:
 169         return result - size + __reverse_ffz(tmp);
 170 }
 171
 172 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
 173 {
 174         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 175         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 176         int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
 177
 178         if (test_opt(sbi, LFS))
 179                 return false;
 180         if (sbi->gc_mode == GC_URGENT)
 181                 return true;
 182
 183         return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
 184                         SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
 185 }
 186
 187 void f2fs_register_inmem_page(struct inode *inode, struct page *page)
 188 {
 189         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 190         struct f2fs_inode_info *fi = F2FS_I(inode);
 191         struct inmem_pages *new;
 192
 193         f2fs_trace_pid(page);
 194
 195         set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
 196         SetPagePrivate(page);
 197
 198         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 199
 200         /* add atomic page indices to the list */
 201         new->page = page;
 202         INIT_LIST_HEAD(&new->list);
 203
 204         /* increase reference count with clean state */
 205         mutex_lock(&fi->inmem_lock);
 206         get_page(page);
 207         list_add_tail(&new->list, &fi->inmem_pages);
 208         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 209         if (list_empty(&fi->inmem_ilist))
 210                 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
 211         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 212         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 213         mutex_unlock(&fi->inmem_lock);
 214
 215         trace_f2fs_register_inmem_page(page, INMEM);
 216 }
 217
 218 static int __revoke_inmem_pages(struct inode *inode,
 219                                 struct list_head *head, bool drop, bool recover)
 220 {
 221         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 222         struct inmem_pages *cur, *tmp;
 223         int err = 0;
 224
 225         list_for_each_entry_safe(cur, tmp, head, list) {
 226                 struct page *page = cur->page;
 227
 228                 if (drop)
 229                         trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 230
 231                 lock_page(page);
 232
 233                 f2fs_wait_on_page_writeback(page, DATA, true);
 234
 235                 if (recover) {
 236                         struct dnode_of_data dn;
 237                         struct node_info ni;
 238
 239                         trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 240 retry:
 241                         set_new_dnode(&dn, inode, NULL, NULL, 0);
 242                         err = f2fs_get_dnode_of_data(&dn, page->index,
 243                                                                 LOOKUP_NODE);
 244                         if (err) {
 245                                 if (err == -ENOMEM) {
 246                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 247                                         cond_resched();
 248                                         goto retry;
 249                                 }
 250                                 err = -EAGAIN;
 251                                 goto next;
 252                         }
 253                         f2fs_get_node_info(sbi, dn.nid, &ni);
 254                         if (cur->old_addr == NEW_ADDR) {
 255                                 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
 256                                 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
 257                         } else
 258                                 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 259                                         cur->old_addr, ni.version, true, true);
 260                         f2fs_put_dnode(&dn);
 261                 }
 262 next:
 263                 /* we don't need to invalidate this in the sccessful status */
 264                 if (drop || recover)
 265                         ClearPageUptodate(page);
 266                 set_page_private(page, 0);
 267                 ClearPagePrivate(page);
 268                 f2fs_put_page(page, 1);
 269
 270                 list_del(&cur->list);
 271                 kmem_cache_free(inmem_entry_slab, cur);
 272                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 273         }
 274         return err;
 275 }
 276
 277 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
 278 {
 279         struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
 280         struct inode *inode;
 281         struct f2fs_inode_info *fi;
 282 next:
 283         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 284         if (list_empty(head)) {
 285                 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 286                 return;
 287         }
 288         fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
 289         inode = igrab(&fi->vfs_inode);
 290         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 291
 292         if (inode) {
 293                 if (gc_failure) {
 294                         if (fi->i_gc_failures[GC_FAILURE_ATOMIC])
 295                                 goto drop;
 296                         goto skip;
 297                 }
 298 drop:
 299                 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
 300                 f2fs_drop_inmem_pages(inode);
 301                 iput(inode);
 302         }
 303 skip:
 304         congestion_wait(BLK_RW_ASYNC, HZ/50);
 305         cond_resched();
 306         goto next;
 307 }
 308
 309 void f2fs_drop_inmem_pages(struct inode *inode)
 310 {
 311         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 312         struct f2fs_inode_info *fi = F2FS_I(inode);
 313
 314         mutex_lock(&fi->inmem_lock);
 315         __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 316         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 317         if (!list_empty(&fi->inmem_ilist))
 318                 list_del_init(&fi->inmem_ilist);
 319         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 320         mutex_unlock(&fi->inmem_lock);
 321
 322         clear_inode_flag(inode, FI_ATOMIC_FILE);
 323         fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
 324         stat_dec_atomic_write(inode);
 325 }
 326
 327 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
 328 {
 329         struct f2fs_inode_info *fi = F2FS_I(inode);
 330         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 331         struct list_head *head = &fi->inmem_pages;
 332         struct inmem_pages *cur = NULL;
 333
 334         f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
 335
 336         mutex_lock(&fi->inmem_lock);
 337         list_for_each_entry(cur, head, list) {
 338                 if (cur->page == page)
 339                         break;
 340         }
 341
 342         f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
 343         list_del(&cur->list);
 344         mutex_unlock(&fi->inmem_lock);
 345
 346         dec_page_count(sbi, F2FS_INMEM_PAGES);
 347         kmem_cache_free(inmem_entry_slab, cur);
 348
 349         ClearPageUptodate(page);
 350         set_page_private(page, 0);
 351         ClearPagePrivate(page);
 352         f2fs_put_page(page, 0);
 353
 354         trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 355 }
 356
 357 static int __f2fs_commit_inmem_pages(struct inode *inode)
 358 {
 359         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 360         struct f2fs_inode_info *fi = F2FS_I(inode);
 361         struct inmem_pages *cur, *tmp;
 362         struct f2fs_io_info fio = {
 363                 .sbi = sbi,
 364                 .ino = inode->i_ino,
 365                 .type = DATA,
 366                 .op = REQ_OP_WRITE,
 367                 .op_flags = REQ_SYNC | REQ_PRIO,
 368                 .io_type = FS_DATA_IO,
 369         };
 370         struct list_head revoke_list;
 371         pgoff_t last_idx = ULONG_MAX;
 372         int err = 0;
 373
 374         INIT_LIST_HEAD(&revoke_list);
 375
 376         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 377                 struct page *page = cur->page;
 378
 379                 lock_page(page);
 380                 if (page->mapping == inode->i_mapping) {
 381                         trace_f2fs_commit_inmem_page(page, INMEM);
 382
 383                         set_page_dirty(page);
 384                         f2fs_wait_on_page_writeback(page, DATA, true);
 385                         if (clear_page_dirty_for_io(page)) {
 386                                 inode_dec_dirty_pages(inode);
 387                                 f2fs_remove_dirty_inode(inode);
 388                         }
 389 retry:
 390                         fio.page = page;
 391                         fio.old_blkaddr = NULL_ADDR;
 392                         fio.encrypted_page = NULL;
 393                         fio.need_lock = LOCK_DONE;
 394                         err = f2fs_do_write_data_page(&fio);
 395                         if (err) {
 396                                 if (err == -ENOMEM) {
 397                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 398                                         cond_resched();
 399                                         goto retry;
 400                                 }
 401                                 unlock_page(page);
 402                                 break;
 403                         }
 404                         /* record old blkaddr for revoking */
 405                         cur->old_addr = fio.old_blkaddr;
 406                         last_idx = page->index;
 407                 }
 408                 unlock_page(page);
 409                 list_move_tail(&cur->list, &revoke_list);
 410         }
 411
 412         if (last_idx != ULONG_MAX)
 413                 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
 414
 415         if (err) {
 416                 /*
 417                  * try to revoke all committed pages, but still we could fail
 418                  * due to no memory or other reason, if that happened, EAGAIN
 419                  * will be returned, which means in such case, transaction is
 420                  * already not integrity, caller should use journal to do the
 421                  * recovery or rewrite & commit last transaction. For other
 422                  * error number, revoking was done by filesystem itself.
 423                  */
 424                 err = __revoke_inmem_pages(inode, &revoke_list, false, true);
 425
 426                 /* drop all uncommitted pages */
 427                 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 428         } else {
 429                 __revoke_inmem_pages(inode, &revoke_list, false, false);
 430         }
 431
 432         return err;
 433 }
 434
 435 int f2fs_commit_inmem_pages(struct inode *inode)
 436 {
 437         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 438         struct f2fs_inode_info *fi = F2FS_I(inode);
 439         int err;
 440
 441         f2fs_balance_fs(sbi, true);
 442         f2fs_lock_op(sbi);
 443
 444         set_inode_flag(inode, FI_ATOMIC_COMMIT);
 445
 446         mutex_lock(&fi->inmem_lock);
 447         err = __f2fs_commit_inmem_pages(inode);
 448
 449         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 450         if (!list_empty(&fi->inmem_ilist))
 451                 list_del_init(&fi->inmem_ilist);
 452         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 453         mutex_unlock(&fi->inmem_lock);
 454
 455         clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 456
 457         f2fs_unlock_op(sbi);
 458         return err;
 459 }
 460
 461 /*
 462  * This function balances dirty node and dentry pages.
 463  * In addition, it controls garbage collection.
 464  */
 465 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 466 {
 467 #ifdef CONFIG_F2FS_FAULT_INJECTION
 468         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 469                 f2fs_show_injection_info(FAULT_CHECKPOINT);
 470                 f2fs_stop_checkpoint(sbi, false);
 471         }
 472 #endif
 473
 474         /* balance_fs_bg is able to be pending */
 475         if (need && excess_cached_nats(sbi))
 476                 f2fs_balance_fs_bg(sbi);
 477
 478         /*
 479          * We should do GC or end up with checkpoint, if there are so many dirty
 480          * dir/node pages without enough free segments.
 481          */
 482         if (has_not_enough_free_secs(sbi, 0, 0)) {
 483                 mutex_lock(&sbi->gc_mutex);
 484                 f2fs_gc(sbi, false, false, NULL_SEGNO);
 485         }
 486 }
 487
 488 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 489 {
 490         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 491                 return;
 492
 493         /* try to shrink extent cache when there is no enough memory */
 494         if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
 495                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 496
 497         /* check the # of cached NAT entries */
 498         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
 499                 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 500
 501         if (!f2fs_available_free_memory(sbi, FREE_NIDS))
 502                 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
 503         else
 504                 f2fs_build_free_nids(sbi, false, false);
 505
 506         if (!is_idle(sbi) && !excess_dirty_nats(sbi))
 507                 return;
 508
 509         /* checkpoint is the only way to shrink partial cached entries */
 510         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
 511                         !f2fs_available_free_memory(sbi, INO_ENTRIES) ||
 512                         excess_prefree_segs(sbi) ||
 513                         excess_dirty_nats(sbi) ||
 514                         f2fs_time_over(sbi, CP_TIME)) {
 515                 if (test_opt(sbi, DATA_FLUSH)) {
 516                         struct blk_plug plug;
 517
 518                         blk_start_plug(&plug);
 519                         f2fs_sync_dirty_inodes(sbi, FILE_INODE);
 520                         blk_finish_plug(&plug);
 521                 }
 522                 f2fs_sync_fs(sbi->sb, true);
 523                 stat_inc_bg_cp_count(sbi->stat_info);
 524         }
 525 }
 526
 527 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 528                                 struct block_device *bdev)
 529 {
 530         struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
 531         int ret;
 532
 533         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
 534         bio_set_dev(bio, bdev);
 535         ret = submit_bio_wait(bio);
 536         bio_put(bio);
 537
 538         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 539                                 test_opt(sbi, FLUSH_MERGE), ret);
 540         return ret;
 541 }
 542
 543 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 544 {
 545         int ret = 0;
 546         int i;
 547
 548         if (!sbi->s_ndevs)
 549                 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 550
 551         for (i = 0; i < sbi->s_ndevs; i++) {
 552                 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
 553                         continue;
 554                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 555                 if (ret)
 556                         break;
 557         }
 558         return ret;
 559 }
 560
 561 static int issue_flush_thread(void *data)
 562 {
 563         struct f2fs_sb_info *sbi = data;
 564         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 565         wait_queue_head_t *q = &fcc->flush_wait_queue;
 566 repeat:
 567         if (kthread_should_stop())
 568                 return 0;
 569
 570         sb_start_intwrite(sbi->sb);
 571
 572         if (!llist_empty(&fcc->issue_list)) {
 573                 struct flush_cmd *cmd, *next;
 574                 int ret;
 575
 576                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 577                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 578
 579                 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 580
 581                 ret = submit_flush_wait(sbi, cmd->ino);
 582                 atomic_inc(&fcc->issued_flush);
 583
 584                 llist_for_each_entry_safe(cmd, next,
 585                                           fcc->dispatch_list, llnode) {
 586                         cmd->ret = ret;
 587                         complete(&cmd->wait);
 588                 }
 589                 fcc->dispatch_list = NULL;
 590         }
 591
 592         sb_end_intwrite(sbi->sb);
 593
 594         wait_event_interruptible(*q,
 595                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 596         goto repeat;
 597 }
 598
 599 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 600 {
 601         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 602         struct flush_cmd cmd;
 603         int ret;
 604
 605         if (test_opt(sbi, NOBARRIER))
 606                 return 0;
 607
 608         if (!test_opt(sbi, FLUSH_MERGE)) {
 609                 ret = submit_flush_wait(sbi, ino);
 610                 atomic_inc(&fcc->issued_flush);
 611                 return ret;
 612         }
 613
 614         if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
 615                 ret = submit_flush_wait(sbi, ino);
 616                 atomic_dec(&fcc->issing_flush);
 617
 618                 atomic_inc(&fcc->issued_flush);
 619                 return ret;
 620         }
 621
 622         cmd.ino = ino;
 623         init_completion(&cmd.wait);
 624
 625         llist_add(&cmd.llnode, &fcc->issue_list);
 626
 627         /* update issue_list before we wake up issue_flush thread */
 628         smp_mb();
 629
 630         if (waitqueue_active(&fcc->flush_wait_queue))
 631                 wake_up(&fcc->flush_wait_queue);
 632
 633         if (fcc->f2fs_issue_flush) {
 634                 wait_for_completion(&cmd.wait);
 635                 atomic_dec(&fcc->issing_flush);
 636         } else {
 637                 struct llist_node *list;
 638
 639                 list = llist_del_all(&fcc->issue_list);
 640                 if (!list) {
 641                         wait_for_completion(&cmd.wait);
 642                         atomic_dec(&fcc->issing_flush);
 643                 } else {
 644                         struct flush_cmd *tmp, *next;
 645
 646                         ret = submit_flush_wait(sbi, ino);
 647
 648                         llist_for_each_entry_safe(tmp, next, list, llnode) {
 649                                 if (tmp == &cmd) {
 650                                         cmd.ret = ret;
 651                                         atomic_dec(&fcc->issing_flush);
 652                                         continue;
 653                                 }
 654                                 tmp->ret = ret;
 655                                 complete(&tmp->wait);
 656                         }
 657                 }
 658         }
 659
 660         return cmd.ret;
 661 }
 662
 663 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
 664 {
 665         dev_t dev = sbi->sb->s_bdev->bd_dev;
 666         struct flush_cmd_control *fcc;
 667         int err = 0;
 668
 669         if (SM_I(sbi)->fcc_info) {
 670                 fcc = SM_I(sbi)->fcc_info;
 671                 if (fcc->f2fs_issue_flush)
 672                         return err;
 673                 goto init_thread;
 674         }
 675
 676         fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
 677         if (!fcc)
 678                 return -ENOMEM;
 679         atomic_set(&fcc->issued_flush, 0);
 680         atomic_set(&fcc->issing_flush, 0);
 681         init_waitqueue_head(&fcc->flush_wait_queue);
 682         init_llist_head(&fcc->issue_list);
 683         SM_I(sbi)->fcc_info = fcc;
 684         if (!test_opt(sbi, FLUSH_MERGE))
 685                 return err;
 686
 687 init_thread:
 688         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 689                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 690         if (IS_ERR(fcc->f2fs_issue_flush)) {
 691                 err = PTR_ERR(fcc->f2fs_issue_flush);
 692                 kfree(fcc);
 693                 SM_I(sbi)->fcc_info = NULL;
 694                 return err;
 695         }
 696
 697         return err;
 698 }
 699
 700 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 701 {
 702         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 703
 704         if (fcc && fcc->f2fs_issue_flush) {
 705                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 706
 707                 fcc->f2fs_issue_flush = NULL;
 708                 kthread_stop(flush_thread);
 709         }
 710         if (free) {
 711                 kfree(fcc);
 712                 SM_I(sbi)->fcc_info = NULL;
 713         }
 714 }
 715
 716 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
 717 {
 718         int ret = 0, i;
 719
 720         if (!sbi->s_ndevs)
 721                 return 0;
 722
 723         for (i = 1; i < sbi->s_ndevs; i++) {
 724                 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
 725                         continue;
 726                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 727                 if (ret)
 728                         break;
 729
 730                 spin_lock(&sbi->dev_lock);
 731                 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
 732                 spin_unlock(&sbi->dev_lock);
 733         }
 734
 735         return ret;
 736 }
 737
 738 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 739                 enum dirty_type dirty_type)
 740 {
 741         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 742
 743         /* need not be added */
 744         if (IS_CURSEG(sbi, segno))
 745                 return;
 746
 747         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 748                 dirty_i->nr_dirty[dirty_type]++;
 749
 750         if (dirty_type == DIRTY) {
 751                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 752                 enum dirty_type t = sentry->type;
 753
 754                 if (unlikely(t >= DIRTY)) {
 755                         f2fs_bug_on(sbi, 1);
 756                         return;
 757                 }
 758                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 759                         dirty_i->nr_dirty[t]++;
 760         }
 761 }
 762
 763 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 764                 enum dirty_type dirty_type)
 765 {
 766         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 767
 768         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 769                 dirty_i->nr_dirty[dirty_type]--;
 770
 771         if (dirty_type == DIRTY) {
 772                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 773                 enum dirty_type t = sentry->type;
 774
 775                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 776                         dirty_i->nr_dirty[t]--;
 777
 778                 if (get_valid_blocks(sbi, segno, true) == 0)
 779                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 780                                                 dirty_i->victim_secmap);
 781         }
 782 }
 783
 784 /*
 785  * Should not occur error such as -ENOMEM.
 786  * Adding dirty entry into seglist is not critical operation.
 787  * If a given segment is one of current working segments, it won't be added.
 788  */
 789 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 790 {
 791         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 792         unsigned short valid_blocks;
 793
 794         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 795                 return;
 796
 797         mutex_lock(&dirty_i->seglist_lock);
 798
 799         valid_blocks = get_valid_blocks(sbi, segno, false);
 800
 801         if (valid_blocks == 0) {
 802                 __locate_dirty_segment(sbi, segno, PRE);
 803                 __remove_dirty_segment(sbi, segno, DIRTY);
 804         } else if (valid_blocks < sbi->blocks_per_seg) {
 805                 __locate_dirty_segment(sbi, segno, DIRTY);
 806         } else {
 807                 /* Recovery routine with SSR needs this */
 808                 __remove_dirty_segment(sbi, segno, DIRTY);
 809         }
 810
 811         mutex_unlock(&dirty_i->seglist_lock);
 812 }
 813
 814 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 815                 struct block_device *bdev, block_t lstart,
 816                 block_t start, block_t len)
 817 {
 818         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 819         struct list_head *pend_list;
 820         struct discard_cmd *dc;
 821
 822         f2fs_bug_on(sbi, !len);
 823
 824         pend_list = &dcc->pend_list[plist_idx(len)];
 825
 826         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 827         INIT_LIST_HEAD(&dc->list);
 828         dc->bdev = bdev;
 829         dc->lstart = lstart;
 830         dc->start = start;
 831         dc->len = len;
 832         dc->ref = 0;
 833         dc->state = D_PREP;
 834         dc->error = 0;
 835         init_completion(&dc->wait);
 836         list_add_tail(&dc->list, pend_list);
 837         atomic_inc(&dcc->discard_cmd_cnt);
 838         dcc->undiscard_blks += len;
 839
 840         return dc;
 841 }
 842
 843 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 844                                 struct block_device *bdev, block_t lstart,
 845                                 block_t start, block_t len,
 846                                 struct rb_node *parent, struct rb_node **p)
 847 {
 848         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 849         struct discard_cmd *dc;
 850
 851         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
 852
 853         rb_link_node(&dc->rb_node, parent, p);
 854         rb_insert_color(&dc->rb_node, &dcc->root);
 855
 856         return dc;
 857 }
 858
 859 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
 860                                                         struct discard_cmd *dc)
 861 {
 862         if (dc->state == D_DONE)
 863                 atomic_dec(&dcc->issing_discard);
 864
 865         list_del(&dc->list);
 866         rb_erase(&dc->rb_node, &dcc->root);
 867         dcc->undiscard_blks -= dc->len;
 868
 869         kmem_cache_free(discard_cmd_slab, dc);
 870
 871         atomic_dec(&dcc->discard_cmd_cnt);
 872 }
 873
 874 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
 875                                                         struct discard_cmd *dc)
 876 {
 877         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 878
 879         trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
 880
 881         f2fs_bug_on(sbi, dc->ref);
 882
 883         if (dc->error == -EOPNOTSUPP)
 884                 dc->error = 0;
 885
 886         if (dc->error)
 887                 f2fs_msg(sbi->sb, KERN_INFO,
 888                         "Issue discard(%u, %u, %u) failed, ret: %d",
 889                         dc->lstart, dc->start, dc->len, dc->error);
 890         __detach_discard_cmd(dcc, dc);
 891 }
 892
 893 static void f2fs_submit_discard_endio(struct bio *bio)
 894 {
 895         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
 896
 897         dc->error = blk_status_to_errno(bio->bi_status);
 898         dc->state = D_DONE;
 899         complete_all(&dc->wait);
 900         bio_put(bio);
 901 }
 902
 903 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
 904                                 block_t start, block_t end)
 905 {
 906 #ifdef CONFIG_F2FS_CHECK_FS
 907         struct seg_entry *sentry;
 908         unsigned int segno;
 909         block_t blk = start;
 910         unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
 911         unsigned long *map;
 912
 913         while (blk < end) {
 914                 segno = GET_SEGNO(sbi, blk);
 915                 sentry = get_seg_entry(sbi, segno);
 916                 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
 917
 918                 if (end < START_BLOCK(sbi, segno + 1))
 919                         size = GET_BLKOFF_FROM_SEG0(sbi, end);
 920                 else
 921                         size = max_blocks;
 922                 map = (unsigned long *)(sentry->cur_valid_map);
 923                 offset = __find_rev_next_bit(map, size, offset);
 924                 f2fs_bug_on(sbi, offset != size);
 925                 blk = START_BLOCK(sbi, segno + 1);
 926         }
 927 #endif
 928 }
 929
 930 static void __init_discard_policy(struct f2fs_sb_info *sbi,
 931                                 struct discard_policy *dpolicy,
 932                                 int discard_type, unsigned int granularity)
 933 {
 934         /* common policy */
 935         dpolicy->type = discard_type;
 936         dpolicy->sync = true;
 937         dpolicy->granularity = granularity;
 938
 939         dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
 940         dpolicy->io_aware_gran = MAX_PLIST_NUM;
 941
 942         if (discard_type == DPOLICY_BG) {
 943                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
 944                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
 945                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
 946                 dpolicy->io_aware = true;
 947                 dpolicy->sync = false;
 948                 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
 949                         dpolicy->granularity = 1;
 950                         dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
 951                 }
 952         } else if (discard_type == DPOLICY_FORCE) {
 953                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
 954                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
 955                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
 956                 dpolicy->io_aware = false;
 957         } else if (discard_type == DPOLICY_FSTRIM) {
 958                 dpolicy->io_aware = false;
 959         } else if (discard_type == DPOLICY_UMOUNT) {
 960                 dpolicy->max_requests = UINT_MAX;
 961                 dpolicy->io_aware = false;
 962         }
 963 }
 964
 965
 966 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
 967 static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
 968                                                 struct discard_policy *dpolicy,
 969                                                 struct discard_cmd *dc)
 970 {
 971         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 972         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
 973                                         &(dcc->fstrim_list) : &(dcc->wait_list);
 974         struct bio *bio = NULL;
 975         int flag = dpolicy->sync ? REQ_SYNC : 0;
 976
 977         if (dc->state != D_PREP)
 978                 return;
 979
 980         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
 981                 return;
 982
 983         trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
 984
 985         dc->error = __blkdev_issue_discard(dc->bdev,
 986                                 SECTOR_FROM_BLOCK(dc->start),
 987                                 SECTOR_FROM_BLOCK(dc->len),
 988                                 GFP_NOFS, 0, &bio);
 989         if (!dc->error) {
 990                 /* should keep before submission to avoid D_DONE right away */
 991                 dc->state = D_SUBMIT;
 992                 atomic_inc(&dcc->issued_discard);
 993                 atomic_inc(&dcc->issing_discard);
 994                 if (bio) {
 995                         bio->bi_private = dc;
 996                         bio->bi_end_io = f2fs_submit_discard_endio;
 997                         bio->bi_opf |= flag;
 998                         submit_bio(bio);
 999                         list_move_tail(&dc->list, wait_list);
1000                         __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
1001
1002                         f2fs_update_iostat(sbi, FS_DISCARD, 1);
1003                 }
1004         } else {
1005                 __remove_discard_cmd(sbi, dc);
1006         }
1007 }
1008
1009 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1010                                 struct block_device *bdev, block_t lstart,
1011                                 block_t start, block_t len,
1012                                 struct rb_node **insert_p,
1013                                 struct rb_node *insert_parent)
1014 {
1015         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1016         struct rb_node **p;
1017         struct rb_node *parent = NULL;
1018         struct discard_cmd *dc = NULL;
1019
1020         if (insert_p && insert_parent) {
1021                 parent = insert_parent;
1022                 p = insert_p;
1023                 goto do_insert;
1024         }
1025
1026         p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
1027 do_insert:
1028         dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
1029         if (!dc)
1030                 return NULL;
1031
1032         return dc;
1033 }
1034
1035 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1036                                                 struct discard_cmd *dc)
1037 {
1038         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1039 }
1040
1041 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1042                                 struct discard_cmd *dc, block_t blkaddr)
1043 {
1044         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1045         struct discard_info di = dc->di;
1046         bool modified = false;
1047
1048         if (dc->state == D_DONE || dc->len == 1) {
1049                 __remove_discard_cmd(sbi, dc);
1050                 return;
1051         }
1052
1053         dcc->undiscard_blks -= di.len;
1054
1055         if (blkaddr > di.lstart) {
1056                 dc->len = blkaddr - dc->lstart;
1057                 dcc->undiscard_blks += dc->len;
1058                 __relocate_discard_cmd(dcc, dc);
1059                 modified = true;
1060         }
1061
1062         if (blkaddr < di.lstart + di.len - 1) {
1063                 if (modified) {
1064                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1065                                         di.start + blkaddr + 1 - di.lstart,
1066                                         di.lstart + di.len - 1 - blkaddr,
1067                                         NULL, NULL);
1068                 } else {
1069                         dc->lstart++;
1070                         dc->len--;
1071                         dc->start++;
1072                         dcc->undiscard_blks += dc->len;
1073                         __relocate_discard_cmd(dcc, dc);
1074                 }
1075         }
1076 }
1077
1078 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1079                                 struct block_device *bdev, block_t lstart,
1080                                 block_t start, block_t len)
1081 {
1082         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1083         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1084         struct discard_cmd *dc;
1085         struct discard_info di = {0};
1086         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1087         block_t end = lstart + len;
1088
1089         mutex_lock(&dcc->cmd_lock);
1090
1091         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1092                                         NULL, lstart,
1093                                         (struct rb_entry **)&prev_dc,
1094                                         (struct rb_entry **)&next_dc,
1095                                         &insert_p, &insert_parent, true);
1096         if (dc)
1097                 prev_dc = dc;
1098
1099         if (!prev_dc) {
1100                 di.lstart = lstart;
1101                 di.len = next_dc ? next_dc->lstart - lstart : len;
1102                 di.len = min(di.len, len);
1103                 di.start = start;
1104         }
1105
1106         while (1) {
1107                 struct rb_node *node;
1108                 bool merged = false;
1109                 struct discard_cmd *tdc = NULL;
1110
1111                 if (prev_dc) {
1112                         di.lstart = prev_dc->lstart + prev_dc->len;
1113                         if (di.lstart < lstart)
1114                                 di.lstart = lstart;
1115                         if (di.lstart >= end)
1116                                 break;
1117
1118                         if (!next_dc || next_dc->lstart > end)
1119                                 di.len = end - di.lstart;
1120                         else
1121                                 di.len = next_dc->lstart - di.lstart;
1122                         di.start = start + di.lstart - lstart;
1123                 }
1124
1125                 if (!di.len)
1126                         goto next;
1127
1128                 if (prev_dc && prev_dc->state == D_PREP &&
1129                         prev_dc->bdev == bdev &&
1130                         __is_discard_back_mergeable(&di, &prev_dc->di)) {
1131                         prev_dc->di.len += di.len;
1132                         dcc->undiscard_blks += di.len;
1133                         __relocate_discard_cmd(dcc, prev_dc);
1134                         di = prev_dc->di;
1135                         tdc = prev_dc;
1136                         merged = true;
1137                 }
1138
1139                 if (next_dc && next_dc->state == D_PREP &&
1140                         next_dc->bdev == bdev &&
1141                         __is_discard_front_mergeable(&di, &next_dc->di)) {
1142                         next_dc->di.lstart = di.lstart;
1143                         next_dc->di.len += di.len;
1144                         next_dc->di.start = di.start;
1145                         dcc->undiscard_blks += di.len;
1146                         __relocate_discard_cmd(dcc, next_dc);
1147                         if (tdc)
1148                                 __remove_discard_cmd(sbi, tdc);
1149                         merged = true;
1150                 }
1151
1152                 if (!merged) {
1153                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1154                                                         di.len, NULL, NULL);
1155                 }
1156  next:
1157                 prev_dc = next_dc;
1158                 if (!prev_dc)
1159                         break;
1160
1161                 node = rb_next(&prev_dc->rb_node);
1162                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1163         }
1164
1165         mutex_unlock(&dcc->cmd_lock);
1166 }
1167
1168 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1169                 struct block_device *bdev, block_t blkstart, block_t blklen)
1170 {
1171         block_t lblkstart = blkstart;
1172
1173         trace_f2fs_queue_discard(bdev, blkstart, blklen);
1174
1175         if (sbi->s_ndevs) {
1176                 int devi = f2fs_target_device_index(sbi, blkstart);
1177
1178                 blkstart -= FDEV(devi).start_blk;
1179         }
1180         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1181         return 0;
1182 }
1183
1184 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1185                                         struct discard_policy *dpolicy)
1186 {
1187         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1188         struct list_head *pend_list;
1189         struct discard_cmd *dc, *tmp;
1190         struct blk_plug plug;
1191         int i, iter = 0, issued = 0;
1192         bool io_interrupted = false;
1193
1194         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1195                 if (i + 1 < dpolicy->granularity)
1196                         break;
1197                 pend_list = &dcc->pend_list[i];
1198
1199                 mutex_lock(&dcc->cmd_lock);
1200                 if (list_empty(pend_list))
1201                         goto next;
1202                 f2fs_bug_on(sbi,
1203                         !f2fs_check_rb_tree_consistence(sbi, &dcc->root));
1204                 blk_start_plug(&plug);
1205                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1206                         f2fs_bug_on(sbi, dc->state != D_PREP);
1207
1208                         if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1209                                                                 !is_idle(sbi)) {
1210                                 io_interrupted = true;
1211                                 goto skip;
1212                         }
1213
1214                         __submit_discard_cmd(sbi, dpolicy, dc);
1215                         issued++;
1216 skip:
1217                         if (++iter >= dpolicy->max_requests)
1218                                 break;
1219                 }
1220                 blk_finish_plug(&plug);
1221 next:
1222                 mutex_unlock(&dcc->cmd_lock);
1223
1224                 if (iter >= dpolicy->max_requests)
1225                         break;
1226         }
1227
1228         if (!issued && io_interrupted)
1229                 issued = -1;
1230
1231         return issued;
1232 }
1233
1234 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1235 {
1236         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1237         struct list_head *pend_list;
1238         struct discard_cmd *dc, *tmp;
1239         int i;
1240         bool dropped = false;
1241
1242         mutex_lock(&dcc->cmd_lock);
1243         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1244                 pend_list = &dcc->pend_list[i];
1245                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1246                         f2fs_bug_on(sbi, dc->state != D_PREP);
1247                         __remove_discard_cmd(sbi, dc);
1248                         dropped = true;
1249                 }
1250         }
1251         mutex_unlock(&dcc->cmd_lock);
1252
1253         return dropped;
1254 }
1255
1256 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1257 {
1258         __drop_discard_cmd(sbi);
1259 }
1260
1261 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1262                                                         struct discard_cmd *dc)
1263 {
1264         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1265         unsigned int len = 0;
1266
1267         wait_for_completion_io(&dc->wait);
1268         mutex_lock(&dcc->cmd_lock);
1269         f2fs_bug_on(sbi, dc->state != D_DONE);
1270         dc->ref--;
1271         if (!dc->ref) {
1272                 if (!dc->error)
1273                         len = dc->len;
1274                 __remove_discard_cmd(sbi, dc);
1275         }
1276         mutex_unlock(&dcc->cmd_lock);
1277
1278         return len;
1279 }
1280
1281 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1282                                                 struct discard_policy *dpolicy,
1283                                                 block_t start, block_t end)
1284 {
1285         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1286         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1287                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1288         struct discard_cmd *dc, *tmp;
1289         bool need_wait;
1290         unsigned int trimmed = 0;
1291
1292 next:
1293         need_wait = false;
1294
1295         mutex_lock(&dcc->cmd_lock);
1296         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1297                 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1298                         continue;
1299                 if (dc->len < dpolicy->granularity)
1300                         continue;
1301                 if (dc->state == D_DONE && !dc->ref) {
1302                         wait_for_completion_io(&dc->wait);
1303                         if (!dc->error)
1304                                 trimmed += dc->len;
1305                         __remove_discard_cmd(sbi, dc);
1306                 } else {
1307                         dc->ref++;
1308                         need_wait = true;
1309                         break;
1310                 }
1311         }
1312         mutex_unlock(&dcc->cmd_lock);
1313
1314         if (need_wait) {
1315                 trimmed += __wait_one_discard_bio(sbi, dc);
1316                 goto next;
1317         }
1318
1319         return trimmed;
1320 }
1321
1322 static void __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1323                                                 struct discard_policy *dpolicy)
1324 {
1325         struct discard_policy dp;
1326
1327         if (dpolicy) {
1328                 __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1329                 return;
1330         }
1331
1332         /* wait all */
1333         __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1334         __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1335         __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1336         __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1337 }
1338
1339 /* This should be covered by global mutex, &sit_i->sentry_lock */
1340 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1341 {
1342         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1343         struct discard_cmd *dc;
1344         bool need_wait = false;
1345
1346         mutex_lock(&dcc->cmd_lock);
1347         dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1348                                                         NULL, blkaddr);
1349         if (dc) {
1350                 if (dc->state == D_PREP) {
1351                         __punch_discard_cmd(sbi, dc, blkaddr);
1352                 } else {
1353                         dc->ref++;
1354                         need_wait = true;
1355                 }
1356         }
1357         mutex_unlock(&dcc->cmd_lock);
1358
1359         if (need_wait)
1360                 __wait_one_discard_bio(sbi, dc);
1361 }
1362
1363 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1364 {
1365         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1366
1367         if (dcc && dcc->f2fs_issue_discard) {
1368                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1369
1370                 dcc->f2fs_issue_discard = NULL;
1371                 kthread_stop(discard_thread);
1372         }
1373 }
1374
1375 /* This comes from f2fs_put_super */
1376 bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1377 {
1378         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1379         struct discard_policy dpolicy;
1380         bool dropped;
1381
1382         __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1383                                         dcc->discard_granularity);
1384         __issue_discard_cmd(sbi, &dpolicy);
1385         dropped = __drop_discard_cmd(sbi);
1386
1387         /* just to make sure there is no pending discard commands */
1388         __wait_all_discard_cmd(sbi, NULL);
1389         return dropped;
1390 }
1391
1392 static int issue_discard_thread(void *data)
1393 {
1394         struct f2fs_sb_info *sbi = data;
1395         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1396         wait_queue_head_t *q = &dcc->discard_wait_queue;
1397         struct discard_policy dpolicy;
1398         unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1399         int issued;
1400
1401         set_freezable();
1402
1403         do {
1404                 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1405                                         dcc->discard_granularity);
1406
1407                 wait_event_interruptible_timeout(*q,
1408                                 kthread_should_stop() || freezing(current) ||
1409                                 dcc->discard_wake,
1410                                 msecs_to_jiffies(wait_ms));
1411
1412                 if (dcc->discard_wake)
1413                         dcc->discard_wake = 0;
1414
1415                 if (try_to_freeze())
1416                         continue;
1417                 if (f2fs_readonly(sbi->sb))
1418                         continue;
1419                 if (kthread_should_stop())
1420                         return 0;
1421                 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1422                         wait_ms = dpolicy.max_interval;
1423                         continue;
1424                 }
1425
1426                 if (sbi->gc_mode == GC_URGENT)
1427                         __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1428
1429                 sb_start_intwrite(sbi->sb);
1430
1431                 issued = __issue_discard_cmd(sbi, &dpolicy);
1432                 if (issued > 0) {
1433                         __wait_all_discard_cmd(sbi, &dpolicy);
1434                         wait_ms = dpolicy.min_interval;
1435                 } else if (issued == -1){
1436                         wait_ms = dpolicy.mid_interval;
1437                 } else {
1438                         wait_ms = dpolicy.max_interval;
1439                 }
1440
1441                 sb_end_intwrite(sbi->sb);
1442
1443         } while (!kthread_should_stop());
1444         return 0;
1445 }
1446
1447 #ifdef CONFIG_BLK_DEV_ZONED
1448 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1449                 struct block_device *bdev, block_t blkstart, block_t blklen)
1450 {
1451         sector_t sector, nr_sects;
1452         block_t lblkstart = blkstart;
1453         int devi = 0;
1454
1455         if (sbi->s_ndevs) {
1456                 devi = f2fs_target_device_index(sbi, blkstart);
1457                 blkstart -= FDEV(devi).start_blk;
1458         }
1459
1460         /*
1461          * We need to know the type of the zone: for conventional zones,
1462          * use regular discard if the drive supports it. For sequential
1463          * zones, reset the zone write pointer.
1464          */
1465         switch (get_blkz_type(sbi, bdev, blkstart)) {
1466
1467         case BLK_ZONE_TYPE_CONVENTIONAL:
1468                 if (!blk_queue_discard(bdev_get_queue(bdev)))
1469                         return 0;
1470                 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1471         case BLK_ZONE_TYPE_SEQWRITE_REQ:
1472         case BLK_ZONE_TYPE_SEQWRITE_PREF:
1473                 sector = SECTOR_FROM_BLOCK(blkstart);
1474                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1475
1476                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1477                                 nr_sects != bdev_zone_sectors(bdev)) {
1478                         f2fs_msg(sbi->sb, KERN_INFO,
1479                                 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1480                                 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1481                                 blkstart, blklen);
1482                         return -EIO;
1483                 }
1484                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1485                 return blkdev_reset_zones(bdev, sector,
1486                                           nr_sects, GFP_NOFS);
1487         default:
1488                 /* Unknown zone type: broken device ? */
1489                 return -EIO;
1490         }
1491 }
1492 #endif
1493
1494 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1495                 struct block_device *bdev, block_t blkstart, block_t blklen)
1496 {
1497 #ifdef CONFIG_BLK_DEV_ZONED
1498         if (f2fs_sb_has_blkzoned(sbi->sb) &&
1499                                 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1500                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1501 #endif
1502         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1503 }
1504
1505 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1506                                 block_t blkstart, block_t blklen)
1507 {
1508         sector_t start = blkstart, len = 0;
1509         struct block_device *bdev;
1510         struct seg_entry *se;
1511         unsigned int offset;
1512         block_t i;
1513         int err = 0;
1514
1515         bdev = f2fs_target_device(sbi, blkstart, NULL);
1516
1517         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1518                 if (i != start) {
1519                         struct block_device *bdev2 =
1520                                 f2fs_target_device(sbi, i, NULL);
1521
1522                         if (bdev2 != bdev) {
1523                                 err = __issue_discard_async(sbi, bdev,
1524                                                 start, len);
1525                                 if (err)
1526                                         return err;
1527                                 bdev = bdev2;
1528                                 start = i;
1529                                 len = 0;
1530                         }
1531                 }
1532
1533                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1534                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1535
1536                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1537                         sbi->discard_blks--;
1538         }
1539
1540         if (len)
1541                 err = __issue_discard_async(sbi, bdev, start, len);
1542         return err;
1543 }
1544
1545 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1546                                                         bool check_only)
1547 {
1548         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1549         int max_blocks = sbi->blocks_per_seg;
1550         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1551         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1552         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1553         unsigned long *discard_map = (unsigned long *)se->discard_map;
1554         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1555         unsigned int start = 0, end = -1;
1556         bool force = (cpc->reason & CP_DISCARD);
1557         struct discard_entry *de = NULL;
1558         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1559         int i;
1560
1561         if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
1562                 return false;
1563
1564         if (!force) {
1565                 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
1566                         SM_I(sbi)->dcc_info->nr_discards >=
1567                                 SM_I(sbi)->dcc_info->max_discards)
1568                         return false;
1569         }
1570
1571         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1572         for (i = 0; i < entries; i++)
1573                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1574                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1575
1576         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1577                                 SM_I(sbi)->dcc_info->max_discards) {
1578                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1579                 if (start >= max_blocks)
1580                         break;
1581
1582                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1583                 if (force && start && end != max_blocks
1584                                         && (end - start) < cpc->trim_minlen)
1585                         continue;
1586
1587                 if (check_only)
1588                         return true;
1589
1590                 if (!de) {
1591                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1592                                                                 GFP_F2FS_ZERO);
1593                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1594                         list_add_tail(&de->list, head);
1595                 }
1596
1597                 for (i = start; i < end; i++)
1598                         __set_bit_le(i, (void *)de->discard_map);
1599
1600                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1601         }
1602         return false;
1603 }
1604
1605 static void release_discard_addr(struct discard_entry *entry)
1606 {
1607         list_del(&entry->list);
1608         kmem_cache_free(discard_entry_slab, entry);
1609 }
1610
1611 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1612 {
1613         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1614         struct discard_entry *entry, *this;
1615
1616         /* drop caches */
1617         list_for_each_entry_safe(entry, this, head, list)
1618                 release_discard_addr(entry);
1619 }
1620
1621 /*
1622  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1623  */
1624 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1625 {
1626         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1627         unsigned int segno;
1628
1629         mutex_lock(&dirty_i->seglist_lock);
1630         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1631                 __set_test_and_free(sbi, segno);
1632         mutex_unlock(&dirty_i->seglist_lock);
1633 }
1634
1635 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1636                                                 struct cp_control *cpc)
1637 {
1638         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1639         struct list_head *head = &dcc->entry_list;
1640         struct discard_entry *entry, *this;
1641         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1642         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1643         unsigned int start = 0, end = -1;
1644         unsigned int secno, start_segno;
1645         bool force = (cpc->reason & CP_DISCARD);
1646         bool need_align = test_opt(sbi, LFS) && sbi->segs_per_sec > 1;
1647
1648         mutex_lock(&dirty_i->seglist_lock);
1649
1650         while (1) {
1651                 int i;
1652
1653                 if (need_align && end != -1)
1654                         end--;
1655                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1656                 if (start >= MAIN_SEGS(sbi))
1657                         break;
1658                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1659                                                                 start + 1);
1660
1661                 if (need_align) {
1662                         start = rounddown(start, sbi->segs_per_sec);
1663                         end = roundup(end, sbi->segs_per_sec);
1664                 }
1665
1666                 for (i = start; i < end; i++) {
1667                         if (test_and_clear_bit(i, prefree_map))
1668                                 dirty_i->nr_dirty[PRE]--;
1669                 }
1670
1671                 if (!test_opt(sbi, DISCARD))
1672                         continue;
1673
1674                 if (force && start >= cpc->trim_start &&
1675                                         (end - 1) <= cpc->trim_end)
1676                                 continue;
1677
1678                 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1679                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1680                                 (end - start) << sbi->log_blocks_per_seg);
1681                         continue;
1682                 }
1683 next:
1684                 secno = GET_SEC_FROM_SEG(sbi, start);
1685                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1686                 if (!IS_CURSEC(sbi, secno) &&
1687                         !get_valid_blocks(sbi, start, true))
1688                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1689                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1690
1691                 start = start_segno + sbi->segs_per_sec;
1692                 if (start < end)
1693                         goto next;
1694                 else
1695                         end = start - 1;
1696         }
1697         mutex_unlock(&dirty_i->seglist_lock);
1698
1699         /* send small discards */
1700         list_for_each_entry_safe(entry, this, head, list) {
1701                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1702                 bool is_valid = test_bit_le(0, entry->discard_map);
1703
1704 find_next:
1705                 if (is_valid) {
1706                         next_pos = find_next_zero_bit_le(entry->discard_map,
1707                                         sbi->blocks_per_seg, cur_pos);
1708                         len = next_pos - cur_pos;
1709
1710                         if (f2fs_sb_has_blkzoned(sbi->sb) ||
1711                             (force && len < cpc->trim_minlen))
1712                                 goto skip;
1713
1714                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1715                                                                         len);
1716                         total_len += len;
1717                 } else {
1718                         next_pos = find_next_bit_le(entry->discard_map,
1719                                         sbi->blocks_per_seg, cur_pos);
1720                 }
1721 skip:
1722                 cur_pos = next_pos;
1723                 is_valid = !is_valid;
1724
1725                 if (cur_pos < sbi->blocks_per_seg)
1726                         goto find_next;
1727
1728                 release_discard_addr(entry);
1729                 dcc->nr_discards -= total_len;
1730         }
1731
1732         wake_up_discard_thread(sbi, false);
1733 }
1734
1735 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1736 {
1737         dev_t dev = sbi->sb->s_bdev->bd_dev;
1738         struct discard_cmd_control *dcc;
1739         int err = 0, i;
1740
1741         if (SM_I(sbi)->dcc_info) {
1742                 dcc = SM_I(sbi)->dcc_info;
1743                 goto init_thread;
1744         }
1745
1746         dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
1747         if (!dcc)
1748                 return -ENOMEM;
1749
1750         dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
1751         INIT_LIST_HEAD(&dcc->entry_list);
1752         for (i = 0; i < MAX_PLIST_NUM; i++)
1753                 INIT_LIST_HEAD(&dcc->pend_list[i]);
1754         INIT_LIST_HEAD(&dcc->wait_list);
1755         INIT_LIST_HEAD(&dcc->fstrim_list);
1756         mutex_init(&dcc->cmd_lock);
1757         atomic_set(&dcc->issued_discard, 0);
1758         atomic_set(&dcc->issing_discard, 0);
1759         atomic_set(&dcc->discard_cmd_cnt, 0);
1760         dcc->nr_discards = 0;
1761         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1762         dcc->undiscard_blks = 0;
1763         dcc->root = RB_ROOT;
1764
1765         init_waitqueue_head(&dcc->discard_wait_queue);
1766         SM_I(sbi)->dcc_info = dcc;
1767 init_thread:
1768         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1769                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1770         if (IS_ERR(dcc->f2fs_issue_discard)) {
1771                 err = PTR_ERR(dcc->f2fs_issue_discard);
1772                 kfree(dcc);
1773                 SM_I(sbi)->dcc_info = NULL;
1774                 return err;
1775         }
1776
1777         return err;
1778 }
1779
1780 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1781 {
1782         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1783
1784         if (!dcc)
1785                 return;
1786
1787         f2fs_stop_discard_thread(sbi);
1788
1789         kfree(dcc);
1790         SM_I(sbi)->dcc_info = NULL;
1791 }
1792
1793 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1794 {
1795         struct sit_info *sit_i = SIT_I(sbi);
1796
1797         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1798                 sit_i->dirty_sentries++;
1799                 return false;
1800         }
1801
1802         return true;
1803 }
1804
1805 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1806                                         unsigned int segno, int modified)
1807 {
1808         struct seg_entry *se = get_seg_entry(sbi, segno);
1809         se->type = type;
1810         if (modified)
1811                 __mark_sit_entry_dirty(sbi, segno);
1812 }
1813
1814 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1815 {
1816         struct seg_entry *se;
1817         unsigned int segno, offset;
1818         long int new_vblocks;
1819         bool exist;
1820 #ifdef CONFIG_F2FS_CHECK_FS
1821         bool mir_exist;
1822 #endif
1823
1824         segno = GET_SEGNO(sbi, blkaddr);
1825
1826         se = get_seg_entry(sbi, segno);
1827         new_vblocks = se->valid_blocks + del;
1828         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1829
1830         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
1831                                 (new_vblocks > sbi->blocks_per_seg)));
1832
1833         se->valid_blocks = new_vblocks;
1834         se->mtime = get_mtime(sbi, false);
1835         if (se->mtime > SIT_I(sbi)->max_mtime)
1836                 SIT_I(sbi)->max_mtime = se->mtime;
1837
1838         /* Update valid block bitmap */
1839         if (del > 0) {
1840                 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
1841 #ifdef CONFIG_F2FS_CHECK_FS
1842                 mir_exist = f2fs_test_and_set_bit(offset,
1843                                                 se->cur_valid_map_mir);
1844                 if (unlikely(exist != mir_exist)) {
1845                         f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1846                                 "when setting bitmap, blk:%u, old bit:%d",
1847                                 blkaddr, exist);
1848                         f2fs_bug_on(sbi, 1);
1849                 }
1850 #endif
1851                 if (unlikely(exist)) {
1852                         f2fs_msg(sbi->sb, KERN_ERR,
1853                                 "Bitmap was wrongly set, blk:%u", blkaddr);
1854                         f2fs_bug_on(sbi, 1);
1855                         se->valid_blocks--;
1856                         del = 0;
1857                 }
1858
1859                 if (f2fs_discard_en(sbi) &&
1860                         !f2fs_test_and_set_bit(offset, se->discard_map))
1861                         sbi->discard_blks--;
1862
1863                 /* don't overwrite by SSR to keep node chain */
1864                 if (IS_NODESEG(se->type)) {
1865                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1866                                 se->ckpt_valid_blocks++;
1867                 }
1868         } else {
1869                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
1870 #ifdef CONFIG_F2FS_CHECK_FS
1871                 mir_exist = f2fs_test_and_clear_bit(offset,
1872                                                 se->cur_valid_map_mir);
1873                 if (unlikely(exist != mir_exist)) {
1874                         f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1875                                 "when clearing bitmap, blk:%u, old bit:%d",
1876                                 blkaddr, exist);
1877                         f2fs_bug_on(sbi, 1);
1878                 }
1879 #endif
1880                 if (unlikely(!exist)) {
1881                         f2fs_msg(sbi->sb, KERN_ERR,
1882                                 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1883                         f2fs_bug_on(sbi, 1);
1884                         se->valid_blocks++;
1885                         del = 0;
1886                 }
1887
1888                 if (f2fs_discard_en(sbi) &&
1889                         f2fs_test_and_clear_bit(offset, se->discard_map))
1890                         sbi->discard_blks++;
1891         }
1892         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1893                 se->ckpt_valid_blocks += del;
1894
1895         __mark_sit_entry_dirty(sbi, segno);
1896
1897         /* update total number of valid blocks to be written in ckpt area */
1898         SIT_I(sbi)->written_valid_blocks += del;
1899
1900         if (sbi->segs_per_sec > 1)
1901                 get_sec_entry(sbi, segno)->valid_blocks += del;
1902 }
1903
1904 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1905 {
1906         unsigned int segno = GET_SEGNO(sbi, addr);
1907         struct sit_info *sit_i = SIT_I(sbi);
1908
1909         f2fs_bug_on(sbi, addr == NULL_ADDR);
1910         if (addr == NEW_ADDR)
1911                 return;
1912
1913         /* add it into sit main buffer */
1914         down_write(&sit_i->sentry_lock);
1915
1916         update_sit_entry(sbi, addr, -1);
1917
1918         /* add it into dirty seglist */
1919         locate_dirty_segment(sbi, segno);
1920
1921         up_write(&sit_i->sentry_lock);
1922 }
1923
1924 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1925 {
1926         struct sit_info *sit_i = SIT_I(sbi);
1927         unsigned int segno, offset;
1928         struct seg_entry *se;
1929         bool is_cp = false;
1930
1931         if (!is_valid_blkaddr(blkaddr))
1932                 return true;
1933
1934         down_read(&sit_i->sentry_lock);
1935
1936         segno = GET_SEGNO(sbi, blkaddr);
1937         se = get_seg_entry(sbi, segno);
1938         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1939
1940         if (f2fs_test_bit(offset, se->ckpt_valid_map))
1941                 is_cp = true;
1942
1943         up_read(&sit_i->sentry_lock);
1944
1945         return is_cp;
1946 }
1947
1948 /*
1949  * This function should be resided under the curseg_mutex lock
1950  */
1951 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1952                                         struct f2fs_summary *sum)
1953 {
1954         struct curseg_info *curseg = CURSEG_I(sbi, type);
1955         void *addr = curseg->sum_blk;
1956         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1957         memcpy(addr, sum, sizeof(struct f2fs_summary));
1958 }
1959
1960 /*
1961  * Calculate the number of current summary pages for writing
1962  */
1963 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1964 {
1965         int valid_sum_count = 0;
1966         int i, sum_in_page;
1967
1968         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1969                 if (sbi->ckpt->alloc_type[i] == SSR)
1970                         valid_sum_count += sbi->blocks_per_seg;
1971                 else {
1972                         if (for_ra)
1973                                 valid_sum_count += le16_to_cpu(
1974                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1975                         else
1976                                 valid_sum_count += curseg_blkoff(sbi, i);
1977                 }
1978         }
1979
1980         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1981                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1982         if (valid_sum_count <= sum_in_page)
1983                 return 1;
1984         else if ((valid_sum_count - sum_in_page) <=
1985                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1986                 return 2;
1987         return 3;
1988 }
1989
1990 /*
1991  * Caller should put this summary page
1992  */
1993 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1994 {
1995         return f2fs_get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1996 }
1997
1998 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
1999                                         void *src, block_t blk_addr)
2000 {
2001         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2002
2003         memcpy(page_address(page), src, PAGE_SIZE);
2004         set_page_dirty(page);
2005         f2fs_put_page(page, 1);
2006 }
2007
2008 static void write_sum_page(struct f2fs_sb_info *sbi,
2009                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
2010 {
2011         f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2012 }
2013
2014 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2015                                                 int type, block_t blk_addr)
2016 {
2017         struct curseg_info *curseg = CURSEG_I(sbi, type);
2018         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2019         struct f2fs_summary_block *src = curseg->sum_blk;
2020         struct f2fs_summary_block *dst;
2021
2022         dst = (struct f2fs_summary_block *)page_address(page);
2023         memset(dst, 0, PAGE_SIZE);
2024
2025         mutex_lock(&curseg->curseg_mutex);
2026
2027         down_read(&curseg->journal_rwsem);
2028         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2029         up_read(&curseg->journal_rwsem);
2030
2031         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2032         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2033
2034         mutex_unlock(&curseg->curseg_mutex);
2035
2036         set_page_dirty(page);
2037         f2fs_put_page(page, 1);
2038 }
2039
2040 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2041 {
2042         struct curseg_info *curseg = CURSEG_I(sbi, type);
2043         unsigned int segno = curseg->segno + 1;
2044         struct free_segmap_info *free_i = FREE_I(sbi);
2045
2046         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2047                 return !test_bit(segno, free_i->free_segmap);
2048         return 0;
2049 }
2050
2051 /*
2052  * Find a new segment from the free segments bitmap to right order
2053  * This function should be returned with success, otherwise BUG
2054  */
2055 static void get_new_segment(struct f2fs_sb_info *sbi,
2056                         unsigned int *newseg, bool new_sec, int dir)
2057 {
2058         struct free_segmap_info *free_i = FREE_I(sbi);
2059         unsigned int segno, secno, zoneno;
2060         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2061         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2062         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2063         unsigned int left_start = hint;
2064         bool init = true;
2065         int go_left = 0;
2066         int i;
2067
2068         spin_lock(&free_i->segmap_lock);
2069
2070         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2071                 segno = find_next_zero_bit(free_i->free_segmap,
2072                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2073                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2074                         goto got_it;
2075         }
2076 find_other_zone:
2077         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2078         if (secno >= MAIN_SECS(sbi)) {
2079                 if (dir == ALLOC_RIGHT) {
2080                         secno = find_next_zero_bit(free_i->free_secmap,
2081                                                         MAIN_SECS(sbi), 0);
2082                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2083                 } else {
2084                         go_left = 1;
2085                         left_start = hint - 1;
2086                 }
2087         }
2088         if (go_left == 0)
2089                 goto skip_left;
2090
2091         while (test_bit(left_start, free_i->free_secmap)) {
2092                 if (left_start > 0) {
2093                         left_start--;
2094                         continue;
2095                 }
2096                 left_start = find_next_zero_bit(free_i->free_secmap,
2097                                                         MAIN_SECS(sbi), 0);
2098                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2099                 break;
2100         }
2101         secno = left_start;
2102 skip_left:
2103         segno = GET_SEG_FROM_SEC(sbi, secno);
2104         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2105
2106         /* give up on finding another zone */
2107         if (!init)
2108                 goto got_it;
2109         if (sbi->secs_per_zone == 1)
2110                 goto got_it;
2111         if (zoneno == old_zoneno)
2112                 goto got_it;
2113         if (dir == ALLOC_LEFT) {
2114                 if (!go_left && zoneno + 1 >= total_zones)
2115                         goto got_it;
2116                 if (go_left && zoneno == 0)
2117                         goto got_it;
2118         }
2119         for (i = 0; i < NR_CURSEG_TYPE; i++)
2120                 if (CURSEG_I(sbi, i)->zone == zoneno)
2121                         break;
2122
2123         if (i < NR_CURSEG_TYPE) {
2124                 /* zone is in user, try another */
2125                 if (go_left)
2126                         hint = zoneno * sbi->secs_per_zone - 1;
2127                 else if (zoneno + 1 >= total_zones)
2128                         hint = 0;
2129                 else
2130                         hint = (zoneno + 1) * sbi->secs_per_zone;
2131                 init = false;
2132                 goto find_other_zone;
2133         }
2134 got_it:
2135         /* set it as dirty segment in free segmap */
2136         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2137         __set_inuse(sbi, segno);
2138         *newseg = segno;
2139         spin_unlock(&free_i->segmap_lock);
2140 }
2141
2142 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2143 {
2144         struct curseg_info *curseg = CURSEG_I(sbi, type);
2145         struct summary_footer *sum_footer;
2146
2147         curseg->segno = curseg->next_segno;
2148         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2149         curseg->next_blkoff = 0;
2150         curseg->next_segno = NULL_SEGNO;
2151
2152         sum_footer = &(curseg->sum_blk->footer);
2153         memset(sum_footer, 0, sizeof(struct summary_footer));
2154         if (IS_DATASEG(type))
2155                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2156         if (IS_NODESEG(type))
2157                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2158         __set_sit_entry_type(sbi, type, curseg->segno, modified);
2159 }
2160
2161 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2162 {
2163         /* if segs_per_sec is large than 1, we need to keep original policy. */
2164         if (sbi->segs_per_sec != 1)
2165                 return CURSEG_I(sbi, type)->segno;
2166
2167         if (test_opt(sbi, NOHEAP) &&
2168                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2169                 return 0;
2170
2171         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2172                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2173
2174         /* find segments from 0 to reuse freed segments */
2175         if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2176                 return 0;
2177
2178         return CURSEG_I(sbi, type)->segno;
2179 }
2180
2181 /*
2182  * Allocate a current working segment.
2183  * This function always allocates a free segment in LFS manner.
2184  */
2185 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2186 {
2187         struct curseg_info *curseg = CURSEG_I(sbi, type);
2188         unsigned int segno = curseg->segno;
2189         int dir = ALLOC_LEFT;
2190
2191         write_sum_page(sbi, curseg->sum_blk,
2192                                 GET_SUM_BLOCK(sbi, segno));
2193         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2194                 dir = ALLOC_RIGHT;
2195
2196         if (test_opt(sbi, NOHEAP))
2197                 dir = ALLOC_RIGHT;
2198
2199         segno = __get_next_segno(sbi, type);
2200         get_new_segment(sbi, &segno, new_sec, dir);
2201         curseg->next_segno = segno;
2202         reset_curseg(sbi, type, 1);
2203         curseg->alloc_type = LFS;
2204 }
2205
2206 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2207                         struct curseg_info *seg, block_t start)
2208 {
2209         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2210         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2211         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2212         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2213         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2214         int i, pos;
2215
2216         for (i = 0; i < entries; i++)
2217                 target_map[i] = ckpt_map[i] | cur_map[i];
2218
2219         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2220
2221         seg->next_blkoff = pos;
2222 }
2223
2224 /*
2225  * If a segment is written by LFS manner, next block offset is just obtained
2226  * by increasing the current block offset. However, if a segment is written by
2227  * SSR manner, next block offset obtained by calling __next_free_blkoff
2228  */
2229 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2230                                 struct curseg_info *seg)
2231 {
2232         if (seg->alloc_type == SSR)
2233                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2234         else
2235                 seg->next_blkoff++;
2236 }
2237
2238 /*
2239  * This function always allocates a used segment(from dirty seglist) by SSR
2240  * manner, so it should recover the existing segment information of valid blocks
2241  */
2242 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2243 {
2244         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2245         struct curseg_info *curseg = CURSEG_I(sbi, type);
2246         unsigned int new_segno = curseg->next_segno;
2247         struct f2fs_summary_block *sum_node;
2248         struct page *sum_page;
2249
2250         write_sum_page(sbi, curseg->sum_blk,
2251                                 GET_SUM_BLOCK(sbi, curseg->segno));
2252         __set_test_and_inuse(sbi, new_segno);
2253
2254         mutex_lock(&dirty_i->seglist_lock);
2255         __remove_dirty_segment(sbi, new_segno, PRE);
2256         __remove_dirty_segment(sbi, new_segno, DIRTY);
2257         mutex_unlock(&dirty_i->seglist_lock);
2258
2259         reset_curseg(sbi, type, 1);
2260         curseg->alloc_type = SSR;
2261         __next_free_blkoff(sbi, curseg, 0);
2262
2263         sum_page = f2fs_get_sum_page(sbi, new_segno);
2264         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2265         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2266         f2fs_put_page(sum_page, 1);
2267 }
2268
2269 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2270 {
2271         struct curseg_info *curseg = CURSEG_I(sbi, type);
2272         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2273         unsigned segno = NULL_SEGNO;
2274         int i, cnt;
2275         bool reversed = false;
2276
2277         /* f2fs_need_SSR() already forces to do this */
2278         if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2279                 curseg->next_segno = segno;
2280                 return 1;
2281         }
2282
2283         /* For node segments, let's do SSR more intensively */
2284         if (IS_NODESEG(type)) {
2285                 if (type >= CURSEG_WARM_NODE) {
2286                         reversed = true;
2287                         i = CURSEG_COLD_NODE;
2288                 } else {
2289                         i = CURSEG_HOT_NODE;
2290                 }
2291                 cnt = NR_CURSEG_NODE_TYPE;
2292         } else {
2293                 if (type >= CURSEG_WARM_DATA) {
2294                         reversed = true;
2295                         i = CURSEG_COLD_DATA;
2296                 } else {
2297                         i = CURSEG_HOT_DATA;
2298                 }
2299                 cnt = NR_CURSEG_DATA_TYPE;
2300         }
2301
2302         for (; cnt-- > 0; reversed ? i-- : i++) {
2303                 if (i == type)
2304                         continue;
2305                 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2306                         curseg->next_segno = segno;
2307                         return 1;
2308                 }
2309         }
2310         return 0;
2311 }
2312
2313 /*
2314  * flush out current segment and replace it with new segment
2315  * This function should be returned with success, otherwise BUG
2316  */
2317 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2318                                                 int type, bool force)
2319 {
2320         struct curseg_info *curseg = CURSEG_I(sbi, type);
2321
2322         if (force)
2323                 new_curseg(sbi, type, true);
2324         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2325                                         type == CURSEG_WARM_NODE)
2326                 new_curseg(sbi, type, false);
2327         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2328                 new_curseg(sbi, type, false);
2329         else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2330                 change_curseg(sbi, type);
2331         else
2332                 new_curseg(sbi, type, false);
2333
2334         stat_inc_seg_type(sbi, curseg);
2335 }
2336
2337 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2338 {
2339         struct curseg_info *curseg;
2340         unsigned int old_segno;
2341         int i;
2342
2343         down_write(&SIT_I(sbi)->sentry_lock);
2344
2345         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2346                 curseg = CURSEG_I(sbi, i);
2347                 old_segno = curseg->segno;
2348                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2349                 locate_dirty_segment(sbi, old_segno);
2350         }
2351
2352         up_write(&SIT_I(sbi)->sentry_lock);
2353 }
2354
2355 static const struct segment_allocation default_salloc_ops = {
2356         .allocate_segment = allocate_segment_by_default,
2357 };
2358
2359 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2360                                                 struct cp_control *cpc)
2361 {
2362         __u64 trim_start = cpc->trim_start;
2363         bool has_candidate = false;
2364
2365         down_write(&SIT_I(sbi)->sentry_lock);
2366         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2367                 if (add_discard_addrs(sbi, cpc, true)) {
2368                         has_candidate = true;
2369                         break;
2370                 }
2371         }
2372         up_write(&SIT_I(sbi)->sentry_lock);
2373
2374         cpc->trim_start = trim_start;
2375         return has_candidate;
2376 }
2377
2378 static void __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2379                                         struct discard_policy *dpolicy,
2380                                         unsigned int start, unsigned int end)
2381 {
2382         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2383         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2384         struct rb_node **insert_p = NULL, *insert_parent = NULL;
2385         struct discard_cmd *dc;
2386         struct blk_plug plug;
2387         int issued;
2388
2389 next:
2390         issued = 0;
2391
2392         mutex_lock(&dcc->cmd_lock);
2393         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi, &dcc->root));
2394
2395         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2396                                         NULL, start,
2397                                         (struct rb_entry **)&prev_dc,
2398                                         (struct rb_entry **)&next_dc,
2399                                         &insert_p, &insert_parent, true);
2400         if (!dc)
2401                 dc = next_dc;
2402
2403         blk_start_plug(&plug);
2404
2405         while (dc && dc->lstart <= end) {
2406                 struct rb_node *node;
2407
2408                 if (dc->len < dpolicy->granularity)
2409                         goto skip;
2410
2411                 if (dc->state != D_PREP) {
2412                         list_move_tail(&dc->list, &dcc->fstrim_list);
2413                         goto skip;
2414                 }
2415
2416                 __submit_discard_cmd(sbi, dpolicy, dc);
2417
2418                 if (++issued >= dpolicy->max_requests) {
2419                         start = dc->lstart + dc->len;
2420
2421                         blk_finish_plug(&plug);
2422                         mutex_unlock(&dcc->cmd_lock);
2423                         __wait_all_discard_cmd(sbi, NULL);
2424                         congestion_wait(BLK_RW_ASYNC, HZ/50);
2425                         goto next;
2426                 }
2427 skip:
2428                 node = rb_next(&dc->rb_node);
2429                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2430
2431                 if (fatal_signal_pending(current))
2432                         break;
2433         }
2434
2435         blk_finish_plug(&plug);
2436         mutex_unlock(&dcc->cmd_lock);
2437 }
2438
2439 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2440 {
2441         __u64 start = F2FS_BYTES_TO_BLK(range->start);
2442         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2443         unsigned int start_segno, end_segno;
2444         block_t start_block, end_block;
2445         struct cp_control cpc;
2446         struct discard_policy dpolicy;
2447         unsigned long long trimmed = 0;
2448         int err = 0;
2449         bool need_align = test_opt(sbi, LFS) && sbi->segs_per_sec > 1;
2450
2451         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2452                 return -EINVAL;
2453
2454         if (end <= MAIN_BLKADDR(sbi))
2455                 return -EINVAL;
2456
2457         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2458                 f2fs_msg(sbi->sb, KERN_WARNING,
2459                         "Found FS corruption, run fsck to fix.");
2460                 return -EIO;
2461         }
2462
2463         /* start/end segment number in main_area */
2464         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2465         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2466                                                 GET_SEGNO(sbi, end);
2467         if (need_align) {
2468                 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2469                 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2470         }
2471
2472         cpc.reason = CP_DISCARD;
2473         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2474         cpc.trim_start = start_segno;
2475         cpc.trim_end = end_segno;
2476
2477         if (sbi->discard_blks == 0)
2478                 goto out;
2479
2480         mutex_lock(&sbi->gc_mutex);
2481         err = f2fs_write_checkpoint(sbi, &cpc);
2482         mutex_unlock(&sbi->gc_mutex);
2483         if (err)
2484                 goto out;
2485
2486         start_block = START_BLOCK(sbi, start_segno);
2487         end_block = START_BLOCK(sbi, end_segno + 1);
2488
2489         __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2490         __issue_discard_cmd_range(sbi, &dpolicy, start_block, end_block);
2491
2492         /*
2493          * We filed discard candidates, but actually we don't need to wait for
2494          * all of them, since they'll be issued in idle time along with runtime
2495          * discard option. User configuration looks like using runtime discard
2496          * or periodic fstrim instead of it.
2497          */
2498         if (!test_opt(sbi, DISCARD)) {
2499                 trimmed = __wait_discard_cmd_range(sbi, &dpolicy,
2500                                         start_block, end_block);
2501                 range->len = F2FS_BLK_TO_BYTES(trimmed);
2502         }
2503 out:
2504         return err;
2505 }
2506
2507 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2508 {
2509         struct curseg_info *curseg = CURSEG_I(sbi, type);
2510         if (curseg->next_blkoff < sbi->blocks_per_seg)
2511                 return true;
2512         return false;
2513 }
2514
2515 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
2516 {
2517         switch (hint) {
2518         case WRITE_LIFE_SHORT:
2519                 return CURSEG_HOT_DATA;
2520         case WRITE_LIFE_EXTREME:
2521                 return CURSEG_COLD_DATA;
2522         default:
2523                 return CURSEG_WARM_DATA;
2524         }
2525 }
2526
2527 /* This returns write hints for each segment type. This hints will be
2528  * passed down to block layer. There are mapping tables which depend on
2529  * the mount option 'whint_mode'.
2530  *
2531  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2532  *
2533  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2534  *
2535  * User                  F2FS                     Block
2536  * ----                  ----                     -----
2537  *                       META                     WRITE_LIFE_NOT_SET
2538  *                       HOT_NODE                 "
2539  *                       WARM_NODE                "
2540  *                       COLD_NODE                "
2541  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2542  * extension list        "                        "
2543  *
2544  * -- buffered io
2545  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2546  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2547  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2548  * WRITE_LIFE_NONE       "                        "
2549  * WRITE_LIFE_MEDIUM     "                        "
2550  * WRITE_LIFE_LONG       "                        "
2551  *
2552  * -- direct io
2553  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2554  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2555  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2556  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2557  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2558  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2559  *
2560  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2561  *
2562  * User                  F2FS                     Block
2563  * ----                  ----                     -----
2564  *                       META                     WRITE_LIFE_MEDIUM;
2565  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
2566  *                       WARM_NODE                "
2567  *                       COLD_NODE                WRITE_LIFE_NONE
2568  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2569  * extension list        "                        "
2570  *
2571  * -- buffered io
2572  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2573  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2574  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
2575  * WRITE_LIFE_NONE       "                        "
2576  * WRITE_LIFE_MEDIUM     "                        "
2577  * WRITE_LIFE_LONG       "                        "
2578  *
2579  * -- direct io
2580  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2581  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2582  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2583  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2584  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2585  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2586  */
2587
2588 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2589                                 enum page_type type, enum temp_type temp)
2590 {
2591         if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2592                 if (type == DATA) {
2593                         if (temp == WARM)
2594                                 return WRITE_LIFE_NOT_SET;
2595                         else if (temp == HOT)
2596                                 return WRITE_LIFE_SHORT;
2597                         else if (temp == COLD)
2598                                 return WRITE_LIFE_EXTREME;
2599                 } else {
2600                         return WRITE_LIFE_NOT_SET;
2601                 }
2602         } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2603                 if (type == DATA) {
2604                         if (temp == WARM)
2605                                 return WRITE_LIFE_LONG;
2606                         else if (temp == HOT)
2607                                 return WRITE_LIFE_SHORT;
2608                         else if (temp == COLD)
2609                                 return WRITE_LIFE_EXTREME;
2610                 } else if (type == NODE) {
2611                         if (temp == WARM || temp == HOT)
2612                                 return WRITE_LIFE_NOT_SET;
2613                         else if (temp == COLD)
2614                                 return WRITE_LIFE_NONE;
2615                 } else if (type == META) {
2616                         return WRITE_LIFE_MEDIUM;
2617                 }
2618         }
2619         return WRITE_LIFE_NOT_SET;
2620 }
2621
2622 static int __get_segment_type_2(struct f2fs_io_info *fio)
2623 {
2624         if (fio->type == DATA)
2625                 return CURSEG_HOT_DATA;
2626         else
2627                 return CURSEG_HOT_NODE;
2628 }
2629
2630 static int __get_segment_type_4(struct f2fs_io_info *fio)
2631 {
2632         if (fio->type == DATA) {
2633                 struct inode *inode = fio->page->mapping->host;
2634
2635                 if (S_ISDIR(inode->i_mode))
2636                         return CURSEG_HOT_DATA;
2637                 else
2638                         return CURSEG_COLD_DATA;
2639         } else {
2640                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2641                         return CURSEG_WARM_NODE;
2642                 else
2643                         return CURSEG_COLD_NODE;
2644         }
2645 }
2646
2647 static int __get_segment_type_6(struct f2fs_io_info *fio)
2648 {
2649         if (fio->type == DATA) {
2650                 struct inode *inode = fio->page->mapping->host;
2651
2652                 if (is_cold_data(fio->page) || file_is_cold(inode))
2653                         return CURSEG_COLD_DATA;
2654                 if (file_is_hot(inode) ||
2655                                 is_inode_flag_set(inode, FI_HOT_DATA) ||
2656                                 is_inode_flag_set(inode, FI_ATOMIC_FILE) ||
2657                                 is_inode_flag_set(inode, FI_VOLATILE_FILE))
2658                         return CURSEG_HOT_DATA;
2659                 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
2660         } else {
2661                 if (IS_DNODE(fio->page))
2662                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2663                                                 CURSEG_HOT_NODE;
2664                 return CURSEG_COLD_NODE;
2665         }
2666 }
2667
2668 static int __get_segment_type(struct f2fs_io_info *fio)
2669 {
2670         int type = 0;
2671
2672         switch (F2FS_OPTION(fio->sbi).active_logs) {
2673         case 2:
2674                 type = __get_segment_type_2(fio);
2675                 break;
2676         case 4:
2677                 type = __get_segment_type_4(fio);
2678                 break;
2679         case 6:
2680                 type = __get_segment_type_6(fio);
2681                 break;
2682         default:
2683                 f2fs_bug_on(fio->sbi, true);
2684         }
2685
2686         if (IS_HOT(type))
2687                 fio->temp = HOT;
2688         else if (IS_WARM(type))
2689                 fio->temp = WARM;
2690         else
2691                 fio->temp = COLD;
2692         return type;
2693 }
2694
2695 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2696                 block_t old_blkaddr, block_t *new_blkaddr,
2697                 struct f2fs_summary *sum, int type,
2698                 struct f2fs_io_info *fio, bool add_list)
2699 {
2700         struct sit_info *sit_i = SIT_I(sbi);
2701         struct curseg_info *curseg = CURSEG_I(sbi, type);
2702
2703         down_read(&SM_I(sbi)->curseg_lock);
2704
2705         mutex_lock(&curseg->curseg_mutex);
2706         down_write(&sit_i->sentry_lock);
2707
2708         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2709
2710         f2fs_wait_discard_bio(sbi, *new_blkaddr);
2711
2712         /*
2713          * __add_sum_entry should be resided under the curseg_mutex
2714          * because, this function updates a summary entry in the
2715          * current summary block.
2716          */
2717         __add_sum_entry(sbi, type, sum);
2718
2719         __refresh_next_blkoff(sbi, curseg);
2720
2721         stat_inc_block_count(sbi, curseg);
2722
2723         /*
2724          * SIT information should be updated before segment allocation,
2725          * since SSR needs latest valid block information.
2726          */
2727         update_sit_entry(sbi, *new_blkaddr, 1);
2728         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2729                 update_sit_entry(sbi, old_blkaddr, -1);
2730
2731         if (!__has_curseg_space(sbi, type))
2732                 sit_i->s_ops->allocate_segment(sbi, type, false);
2733
2734         /*
2735          * segment dirty status should be updated after segment allocation,
2736          * so we just need to update status only one time after previous
2737          * segment being closed.
2738          */
2739         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2740         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
2741
2742         up_write(&sit_i->sentry_lock);
2743
2744         if (page && IS_NODESEG(type)) {
2745                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2746
2747                 f2fs_inode_chksum_set(sbi, page);
2748         }
2749
2750         if (add_list) {
2751                 struct f2fs_bio_info *io;
2752
2753                 INIT_LIST_HEAD(&fio->list);
2754                 fio->in_list = true;
2755                 fio->retry = false;
2756                 io = sbi->write_io[fio->type] + fio->temp;
2757                 spin_lock(&io->io_lock);
2758                 list_add_tail(&fio->list, &io->io_list);
2759                 spin_unlock(&io->io_lock);
2760         }
2761
2762         mutex_unlock(&curseg->curseg_mutex);
2763
2764         up_read(&SM_I(sbi)->curseg_lock);
2765 }
2766
2767 static void update_device_state(struct f2fs_io_info *fio)
2768 {
2769         struct f2fs_sb_info *sbi = fio->sbi;
2770         unsigned int devidx;
2771
2772         if (!sbi->s_ndevs)
2773                 return;
2774
2775         devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2776
2777         /* update device state for fsync */
2778         f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
2779
2780         /* update device state for checkpoint */
2781         if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2782                 spin_lock(&sbi->dev_lock);
2783                 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2784                 spin_unlock(&sbi->dev_lock);
2785         }
2786 }
2787
2788 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2789 {
2790         int type = __get_segment_type(fio);
2791         bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
2792
2793         if (keep_order)
2794                 down_read(&fio->sbi->io_order_lock);
2795 reallocate:
2796         f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
2797                         &fio->new_blkaddr, sum, type, fio, true);
2798
2799         /* writeout dirty page into bdev */
2800         f2fs_submit_page_write(fio);
2801         if (fio->retry) {
2802                 fio->old_blkaddr = fio->new_blkaddr;
2803                 goto reallocate;
2804         }
2805
2806         update_device_state(fio);
2807
2808         if (keep_order)
2809                 up_read(&fio->sbi->io_order_lock);
2810 }
2811
2812 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2813                                         enum iostat_type io_type)
2814 {
2815         struct f2fs_io_info fio = {
2816                 .sbi = sbi,
2817                 .type = META,
2818                 .temp = HOT,
2819                 .op = REQ_OP_WRITE,
2820                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
2821                 .old_blkaddr = page->index,
2822                 .new_blkaddr = page->index,
2823                 .page = page,
2824                 .encrypted_page = NULL,
2825                 .in_list = false,
2826         };
2827
2828         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
2829                 fio.op_flags &= ~REQ_META;
2830
2831         set_page_writeback(page);
2832         ClearPageError(page);
2833         f2fs_submit_page_write(&fio);
2834
2835         f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
2836 }
2837
2838 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
2839 {
2840         struct f2fs_summary sum;
2841
2842         set_summary(&sum, nid, 0, 0);
2843         do_write_page(&sum, fio);
2844
2845         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2846 }
2847
2848 void f2fs_outplace_write_data(struct dnode_of_data *dn,
2849                                         struct f2fs_io_info *fio)
2850 {
2851         struct f2fs_sb_info *sbi = fio->sbi;
2852         struct f2fs_summary sum;
2853         struct node_info ni;
2854
2855         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
2856         f2fs_get_node_info(sbi, dn->nid, &ni);
2857         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
2858         do_write_page(&sum, fio);
2859         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
2860
2861         f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
2862 }
2863
2864 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
2865 {
2866         int err;
2867         struct f2fs_sb_info *sbi = fio->sbi;
2868
2869         fio->new_blkaddr = fio->old_blkaddr;
2870         /* i/o temperature is needed for passing down write hints */
2871         __get_segment_type(fio);
2872
2873         f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
2874                         GET_SEGNO(sbi, fio->new_blkaddr))->type));
2875
2876         stat_inc_inplace_blocks(fio->sbi);
2877
2878         err = f2fs_submit_page_bio(fio);
2879         if (!err)
2880                 update_device_state(fio);
2881
2882         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2883
2884         return err;
2885 }
2886
2887 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
2888                                                 unsigned int segno)
2889 {
2890         int i;
2891
2892         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
2893                 if (CURSEG_I(sbi, i)->segno == segno)
2894                         break;
2895         }
2896         return i;
2897 }
2898
2899 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
2900                                 block_t old_blkaddr, block_t new_blkaddr,
2901                                 bool recover_curseg, bool recover_newaddr)
2902 {
2903         struct sit_info *sit_i = SIT_I(sbi);
2904         struct curseg_info *curseg;
2905         unsigned int segno, old_cursegno;
2906         struct seg_entry *se;
2907         int type;
2908         unsigned short old_blkoff;
2909
2910         segno = GET_SEGNO(sbi, new_blkaddr);
2911         se = get_seg_entry(sbi, segno);
2912         type = se->type;
2913
2914         down_write(&SM_I(sbi)->curseg_lock);
2915
2916         if (!recover_curseg) {
2917                 /* for recovery flow */
2918                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2919                         if (old_blkaddr == NULL_ADDR)
2920                                 type = CURSEG_COLD_DATA;
2921                         else
2922                                 type = CURSEG_WARM_DATA;
2923                 }
2924         } else {
2925                 if (IS_CURSEG(sbi, segno)) {
2926                         /* se->type is volatile as SSR allocation */
2927                         type = __f2fs_get_curseg(sbi, segno);
2928                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
2929                 } else {
2930                         type = CURSEG_WARM_DATA;
2931                 }
2932         }
2933
2934         f2fs_bug_on(sbi, !IS_DATASEG(type));
2935         curseg = CURSEG_I(sbi, type);
2936
2937         mutex_lock(&curseg->curseg_mutex);
2938         down_write(&sit_i->sentry_lock);
2939
2940         old_cursegno = curseg->segno;
2941         old_blkoff = curseg->next_blkoff;
2942
2943         /* change the current segment */
2944         if (segno != curseg->segno) {
2945                 curseg->next_segno = segno;
2946                 change_curseg(sbi, type);
2947         }
2948
2949         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
2950         __add_sum_entry(sbi, type, sum);
2951
2952         if (!recover_curseg || recover_newaddr)
2953                 update_sit_entry(sbi, new_blkaddr, 1);
2954         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2955                 update_sit_entry(sbi, old_blkaddr, -1);
2956
2957         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2958         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2959
2960         locate_dirty_segment(sbi, old_cursegno);
2961
2962         if (recover_curseg) {
2963                 if (old_cursegno != curseg->segno) {
2964                         curseg->next_segno = old_cursegno;
2965                         change_curseg(sbi, type);
2966                 }
2967                 curseg->next_blkoff = old_blkoff;
2968         }
2969
2970         up_write(&sit_i->sentry_lock);
2971         mutex_unlock(&curseg->curseg_mutex);
2972         up_write(&SM_I(sbi)->curseg_lock);
2973 }
2974
2975 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2976                                 block_t old_addr, block_t new_addr,
2977                                 unsigned char version, bool recover_curseg,
2978                                 bool recover_newaddr)
2979 {
2980         struct f2fs_summary sum;
2981
2982         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2983
2984         f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
2985                                         recover_curseg, recover_newaddr);
2986
2987         f2fs_update_data_blkaddr(dn, new_addr);
2988 }
2989
2990 void f2fs_wait_on_page_writeback(struct page *page,
2991                                 enum page_type type, bool ordered)
2992 {
2993         if (PageWriteback(page)) {
2994                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2995
2996                 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2997                                                 0, page->index, type);
2998                 if (ordered)
2999                         wait_on_page_writeback(page);
3000                 else
3001                         wait_for_stable_page(page);
3002         }
3003 }
3004
3005 void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
3006 {
3007         struct page *cpage;
3008
3009         if (!is_valid_blkaddr(blkaddr))
3010                 return;
3011
3012         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3013         if (cpage) {
3014                 f2fs_wait_on_page_writeback(cpage, DATA, true);
3015                 f2fs_put_page(cpage, 1);
3016         }
3017 }
3018
3019 static void read_compacted_summaries(struct f2fs_sb_info *sbi)
3020 {
3021         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3022         struct curseg_info *seg_i;
3023         unsigned char *kaddr;
3024         struct page *page;
3025         block_t start;
3026         int i, j, offset;
3027
3028         start = start_sum_block(sbi);
3029
3030         page = f2fs_get_meta_page(sbi, start++);
3031         kaddr = (unsigned char *)page_address(page);
3032
3033         /* Step 1: restore nat cache */
3034         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3035         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3036
3037         /* Step 2: restore sit cache */
3038         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3039         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3040         offset = 2 * SUM_JOURNAL_SIZE;
3041
3042         /* Step 3: restore summary entries */
3043         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3044                 unsigned short blk_off;
3045                 unsigned int segno;
3046
3047                 seg_i = CURSEG_I(sbi, i);
3048                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3049                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3050                 seg_i->next_segno = segno;
3051                 reset_curseg(sbi, i, 0);
3052                 seg_i->alloc_type = ckpt->alloc_type[i];
3053                 seg_i->next_blkoff = blk_off;
3054
3055                 if (seg_i->alloc_type == SSR)
3056                         blk_off = sbi->blocks_per_seg;
3057
3058                 for (j = 0; j < blk_off; j++) {
3059                         struct f2fs_summary *s;
3060                         s = (struct f2fs_summary *)(kaddr + offset);
3061                         seg_i->sum_blk->entries[j] = *s;
3062                         offset += SUMMARY_SIZE;
3063                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3064                                                 SUM_FOOTER_SIZE)
3065                                 continue;
3066
3067                         f2fs_put_page(page, 1);
3068                         page = NULL;
3069
3070                         page = f2fs_get_meta_page(sbi, start++);
3071                         kaddr = (unsigned char *)page_address(page);
3072                         offset = 0;
3073                 }
3074         }
3075         f2fs_put_page(page, 1);
3076 }
3077
3078 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3079 {
3080         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3081         struct f2fs_summary_block *sum;
3082         struct curseg_info *curseg;
3083         struct page *new;
3084         unsigned short blk_off;
3085         unsigned int segno = 0;
3086         block_t blk_addr = 0;
3087
3088         /* get segment number and block addr */
3089         if (IS_DATASEG(type)) {
3090                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3091                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3092                                                         CURSEG_HOT_DATA]);
3093                 if (__exist_node_summaries(sbi))
3094                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3095                 else
3096                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3097         } else {
3098                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3099                                                         CURSEG_HOT_NODE]);
3100                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3101                                                         CURSEG_HOT_NODE]);
3102                 if (__exist_node_summaries(sbi))
3103                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3104                                                         type - CURSEG_HOT_NODE);
3105                 else
3106                         blk_addr = GET_SUM_BLOCK(sbi, segno);
3107         }
3108
3109         new = f2fs_get_meta_page(sbi, blk_addr);
3110         sum = (struct f2fs_summary_block *)page_address(new);
3111
3112         if (IS_NODESEG(type)) {
3113                 if (__exist_node_summaries(sbi)) {
3114                         struct f2fs_summary *ns = &sum->entries[0];
3115                         int i;
3116                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3117                                 ns->version = 0;
3118                                 ns->ofs_in_node = 0;
3119                         }
3120                 } else {
3121                         f2fs_restore_node_summary(sbi, segno, sum);
3122                 }
3123         }
3124
3125         /* set uncompleted segment to curseg */
3126         curseg = CURSEG_I(sbi, type);
3127         mutex_lock(&curseg->curseg_mutex);
3128
3129         /* update journal info */
3130         down_write(&curseg->journal_rwsem);
3131         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3132         up_write(&curseg->journal_rwsem);
3133
3134         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3135         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3136         curseg->next_segno = segno;
3137         reset_curseg(sbi, type, 0);
3138         curseg->alloc_type = ckpt->alloc_type[type];
3139         curseg->next_blkoff = blk_off;
3140         mutex_unlock(&curseg->curseg_mutex);
3141         f2fs_put_page(new, 1);
3142         return 0;
3143 }
3144
3145 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3146 {
3147         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3148         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3149         int type = CURSEG_HOT_DATA;
3150         int err;
3151
3152         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3153                 int npages = f2fs_npages_for_summary_flush(sbi, true);
3154
3155                 if (npages >= 2)
3156                         f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3157                                                         META_CP, true);
3158
3159                 /* restore for compacted data summary */
3160                 read_compacted_summaries(sbi);
3161                 type = CURSEG_HOT_NODE;
3162         }
3163
3164         if (__exist_node_summaries(sbi))
3165                 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3166                                         NR_CURSEG_TYPE - type, META_CP, true);
3167
3168         for (; type <= CURSEG_COLD_NODE; type++) {
3169                 err = read_normal_summaries(sbi, type);
3170                 if (err)
3171                         return err;
3172         }
3173
3174         /* sanity check for summary blocks */
3175         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3176                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3177                 return -EINVAL;
3178
3179         return 0;
3180 }
3181
3182 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3183 {
3184         struct page *page;
3185         unsigned char *kaddr;
3186         struct f2fs_summary *summary;
3187         struct curseg_info *seg_i;
3188         int written_size = 0;
3189         int i, j;
3190
3191         page = f2fs_grab_meta_page(sbi, blkaddr++);
3192         kaddr = (unsigned char *)page_address(page);
3193         memset(kaddr, 0, PAGE_SIZE);
3194
3195         /* Step 1: write nat cache */
3196         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3197         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3198         written_size += SUM_JOURNAL_SIZE;
3199
3200         /* Step 2: write sit cache */
3201         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3202         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3203         written_size += SUM_JOURNAL_SIZE;
3204
3205         /* Step 3: write summary entries */
3206         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3207                 unsigned short blkoff;
3208                 seg_i = CURSEG_I(sbi, i);
3209                 if (sbi->ckpt->alloc_type[i] == SSR)
3210                         blkoff = sbi->blocks_per_seg;
3211                 else
3212                         blkoff = curseg_blkoff(sbi, i);
3213
3214                 for (j = 0; j < blkoff; j++) {
3215                         if (!page) {
3216                                 page = f2fs_grab_meta_page(sbi, blkaddr++);
3217                                 kaddr = (unsigned char *)page_address(page);
3218                                 memset(kaddr, 0, PAGE_SIZE);
3219                                 written_size = 0;
3220                         }
3221                         summary = (struct f2fs_summary *)(kaddr + written_size);
3222                         *summary = seg_i->sum_blk->entries[j];
3223                         written_size += SUMMARY_SIZE;
3224
3225                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3226                                                         SUM_FOOTER_SIZE)
3227                                 continue;
3228
3229                         set_page_dirty(page);
3230                         f2fs_put_page(page, 1);
3231                         page = NULL;
3232                 }
3233         }
3234         if (page) {
3235                 set_page_dirty(page);
3236                 f2fs_put_page(page, 1);
3237         }
3238 }
3239
3240 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3241                                         block_t blkaddr, int type)
3242 {
3243         int i, end;
3244         if (IS_DATASEG(type))
3245                 end = type + NR_CURSEG_DATA_TYPE;
3246         else
3247                 end = type + NR_CURSEG_NODE_TYPE;
3248
3249         for (i = type; i < end; i++)
3250                 write_current_sum_page(sbi, i, blkaddr + (i - type));
3251 }
3252
3253 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3254 {
3255         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3256                 write_compacted_summaries(sbi, start_blk);
3257         else
3258                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3259 }
3260
3261 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3262 {
3263         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3264 }
3265
3266 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3267                                         unsigned int val, int alloc)
3268 {
3269         int i;
3270
3271         if (type == NAT_JOURNAL) {
3272                 for (i = 0; i < nats_in_cursum(journal); i++) {
3273                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3274                                 return i;
3275                 }
3276                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3277                         return update_nats_in_cursum(journal, 1);
3278         } else if (type == SIT_JOURNAL) {
3279                 for (i = 0; i < sits_in_cursum(journal); i++)
3280                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3281                                 return i;
3282                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3283                         return update_sits_in_cursum(journal, 1);
3284         }
3285         return -1;
3286 }
3287
3288 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3289                                         unsigned int segno)
3290 {
3291         return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
3292 }
3293
3294 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3295                                         unsigned int start)
3296 {
3297         struct sit_info *sit_i = SIT_I(sbi);
3298         struct page *page;
3299         pgoff_t src_off, dst_off;
3300
3301         src_off = current_sit_addr(sbi, start);
3302         dst_off = next_sit_addr(sbi, src_off);
3303
3304         page = f2fs_grab_meta_page(sbi, dst_off);
3305         seg_info_to_sit_page(sbi, page, start);
3306
3307         set_page_dirty(page);
3308         set_to_next_sit(sit_i, start);
3309
3310         return page;
3311 }
3312
3313 static struct sit_entry_set *grab_sit_entry_set(void)
3314 {
3315         struct sit_entry_set *ses =
3316                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3317
3318         ses->entry_cnt = 0;
3319         INIT_LIST_HEAD(&ses->set_list);
3320         return ses;
3321 }
3322
3323 static void release_sit_entry_set(struct sit_entry_set *ses)
3324 {
3325         list_del(&ses->set_list);
3326         kmem_cache_free(sit_entry_set_slab, ses);
3327 }
3328
3329 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3330                                                 struct list_head *head)
3331 {
3332         struct sit_entry_set *next = ses;
3333
3334         if (list_is_last(&ses->set_list, head))
3335                 return;
3336
3337         list_for_each_entry_continue(next, head, set_list)
3338                 if (ses->entry_cnt <= next->entry_cnt)
3339                         break;
3340
3341         list_move_tail(&ses->set_list, &next->set_list);
3342 }
3343
3344 static void add_sit_entry(unsigned int segno, struct list_head *head)
3345 {
3346         struct sit_entry_set *ses;
3347         unsigned int start_segno = START_SEGNO(segno);
3348
3349         list_for_each_entry(ses, head, set_list) {
3350                 if (ses->start_segno == start_segno) {
3351                         ses->entry_cnt++;
3352                         adjust_sit_entry_set(ses, head);
3353                         return;
3354                 }
3355         }
3356
3357         ses = grab_sit_entry_set();
3358
3359         ses->start_segno = start_segno;
3360         ses->entry_cnt++;
3361         list_add(&ses->set_list, head);
3362 }
3363
3364 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3365 {
3366         struct f2fs_sm_info *sm_info = SM_I(sbi);
3367         struct list_head *set_list = &sm_info->sit_entry_set;
3368         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3369         unsigned int segno;
3370
3371         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3372                 add_sit_entry(segno, set_list);
3373 }
3374
3375 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3376 {
3377         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3378         struct f2fs_journal *journal = curseg->journal;
3379         int i;
3380
3381         down_write(&curseg->journal_rwsem);
3382         for (i = 0; i < sits_in_cursum(journal); i++) {
3383                 unsigned int segno;
3384                 bool dirtied;
3385
3386                 segno = le32_to_cpu(segno_in_journal(journal, i));
3387                 dirtied = __mark_sit_entry_dirty(sbi, segno);
3388
3389                 if (!dirtied)
3390                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3391         }
3392         update_sits_in_cursum(journal, -i);
3393         up_write(&curseg->journal_rwsem);
3394 }
3395
3396 /*
3397  * CP calls this function, which flushes SIT entries including sit_journal,
3398  * and moves prefree segs to free segs.
3399  */
3400 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3401 {
3402         struct sit_info *sit_i = SIT_I(sbi);
3403         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3404         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3405         struct f2fs_journal *journal = curseg->journal;
3406         struct sit_entry_set *ses, *tmp;
3407         struct list_head *head = &SM_I(sbi)->sit_entry_set;
3408         bool to_journal = true;
3409         struct seg_entry *se;
3410
3411         down_write(&sit_i->sentry_lock);
3412
3413         if (!sit_i->dirty_sentries)
3414                 goto out;
3415
3416         /*
3417          * add and account sit entries of dirty bitmap in sit entry
3418          * set temporarily
3419          */
3420         add_sits_in_set(sbi);
3421
3422         /*
3423          * if there are no enough space in journal to store dirty sit
3424          * entries, remove all entries from journal and add and account
3425          * them in sit entry set.
3426          */
3427         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3428                 remove_sits_in_journal(sbi);
3429
3430         /*
3431          * there are two steps to flush sit entries:
3432          * #1, flush sit entries to journal in current cold data summary block.
3433          * #2, flush sit entries to sit page.
3434          */
3435         list_for_each_entry_safe(ses, tmp, head, set_list) {
3436                 struct page *page = NULL;
3437                 struct f2fs_sit_block *raw_sit = NULL;
3438                 unsigned int start_segno = ses->start_segno;
3439                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3440                                                 (unsigned long)MAIN_SEGS(sbi));
3441                 unsigned int segno = start_segno;
3442
3443                 if (to_journal &&
3444                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3445                         to_journal = false;
3446
3447                 if (to_journal) {
3448                         down_write(&curseg->journal_rwsem);
3449                 } else {
3450                         page = get_next_sit_page(sbi, start_segno);
3451                         raw_sit = page_address(page);
3452                 }
3453
3454                 /* flush dirty sit entries in region of current sit set */
3455                 for_each_set_bit_from(segno, bitmap, end) {
3456                         int offset, sit_offset;
3457
3458                         se = get_seg_entry(sbi, segno);
3459 #ifdef CONFIG_F2FS_CHECK_FS
3460                         if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3461                                                 SIT_VBLOCK_MAP_SIZE))
3462                                 f2fs_bug_on(sbi, 1);
3463 #endif
3464
3465                         /* add discard candidates */
3466                         if (!(cpc->reason & CP_DISCARD)) {
3467                                 cpc->trim_start = segno;
3468                                 add_discard_addrs(sbi, cpc, false);
3469                         }
3470
3471                         if (to_journal) {
3472                                 offset = f2fs_lookup_journal_in_cursum(journal,
3473                                                         SIT_JOURNAL, segno, 1);
3474                                 f2fs_bug_on(sbi, offset < 0);
3475                                 segno_in_journal(journal, offset) =
3476                                                         cpu_to_le32(segno);
3477                                 seg_info_to_raw_sit(se,
3478                                         &sit_in_journal(journal, offset));
3479                                 check_block_count(sbi, segno,
3480                                         &sit_in_journal(journal, offset));
3481                         } else {
3482                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3483                                 seg_info_to_raw_sit(se,
3484                                                 &raw_sit->entries[sit_offset]);
3485                                 check_block_count(sbi, segno,
3486                                                 &raw_sit->entries[sit_offset]);
3487                         }
3488
3489                         __clear_bit(segno, bitmap);
3490                         sit_i->dirty_sentries--;
3491                         ses->entry_cnt--;
3492                 }
3493
3494                 if (to_journal)
3495                         up_write(&curseg->journal_rwsem);
3496                 else
3497                         f2fs_put_page(page, 1);
3498
3499                 f2fs_bug_on(sbi, ses->entry_cnt);
3500                 release_sit_entry_set(ses);
3501         }
3502
3503         f2fs_bug_on(sbi, !list_empty(head));
3504         f2fs_bug_on(sbi, sit_i->dirty_sentries);
3505 out:
3506         if (cpc->reason & CP_DISCARD) {
3507                 __u64 trim_start = cpc->trim_start;
3508
3509                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3510                         add_discard_addrs(sbi, cpc, false);
3511
3512                 cpc->trim_start = trim_start;
3513         }
3514         up_write(&sit_i->sentry_lock);
3515
3516         set_prefree_as_free_segments(sbi);
3517 }
3518
3519 static int build_sit_info(struct f2fs_sb_info *sbi)
3520 {
3521         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3522         struct sit_info *sit_i;
3523         unsigned int sit_segs, start;
3524         char *src_bitmap;
3525         unsigned int bitmap_size;
3526
3527         /* allocate memory for SIT information */
3528         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3529         if (!sit_i)
3530                 return -ENOMEM;
3531
3532         SM_I(sbi)->sit_info = sit_i;
3533
3534         sit_i->sentries =
3535                 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3536                                               MAIN_SEGS(sbi)),
3537                               GFP_KERNEL);
3538         if (!sit_i->sentries)
3539                 return -ENOMEM;
3540
3541         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3542         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3543                                                                 GFP_KERNEL);
3544         if (!sit_i->dirty_sentries_bitmap)
3545                 return -ENOMEM;
3546
3547         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3548                 sit_i->sentries[start].cur_valid_map
3549                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3550                 sit_i->sentries[start].ckpt_valid_map
3551                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3552                 if (!sit_i->sentries[start].cur_valid_map ||
3553                                 !sit_i->sentries[start].ckpt_valid_map)
3554                         return -ENOMEM;
3555
3556 #ifdef CONFIG_F2FS_CHECK_FS
3557                 sit_i->sentries[start].cur_valid_map_mir
3558                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3559                 if (!sit_i->sentries[start].cur_valid_map_mir)
3560                         return -ENOMEM;
3561 #endif
3562
3563                 if (f2fs_discard_en(sbi)) {
3564                         sit_i->sentries[start].discard_map
3565                                 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3566                                                                 GFP_KERNEL);
3567                         if (!sit_i->sentries[start].discard_map)
3568                                 return -ENOMEM;
3569                 }
3570         }
3571
3572         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3573         if (!sit_i->tmp_map)
3574                 return -ENOMEM;
3575
3576         if (sbi->segs_per_sec > 1) {
3577                 sit_i->sec_entries =
3578                         f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
3579                                                       MAIN_SECS(sbi)),
3580                                       GFP_KERNEL);
3581                 if (!sit_i->sec_entries)
3582                         return -ENOMEM;
3583         }
3584
3585         /* get information related with SIT */
3586         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3587
3588         /* setup SIT bitmap from ckeckpoint pack */
3589         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3590         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3591
3592         sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3593         if (!sit_i->sit_bitmap)
3594                 return -ENOMEM;
3595
3596 #ifdef CONFIG_F2FS_CHECK_FS
3597         sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3598         if (!sit_i->sit_bitmap_mir)
3599                 return -ENOMEM;
3600 #endif
3601
3602         /* init SIT information */
3603         sit_i->s_ops = &default_salloc_ops;
3604
3605         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3606         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
3607         sit_i->written_valid_blocks = 0;
3608         sit_i->bitmap_size = bitmap_size;
3609         sit_i->dirty_sentries = 0;
3610         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3611         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3612         sit_i->mounted_time = ktime_get_real_seconds();
3613         init_rwsem(&sit_i->sentry_lock);
3614         return 0;
3615 }
3616
3617 static int build_free_segmap(struct f2fs_sb_info *sbi)
3618 {
3619         struct free_segmap_info *free_i;
3620         unsigned int bitmap_size, sec_bitmap_size;
3621
3622         /* allocate memory for free segmap information */
3623         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
3624         if (!free_i)
3625                 return -ENOMEM;
3626
3627         SM_I(sbi)->free_info = free_i;
3628
3629         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3630         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
3631         if (!free_i->free_segmap)
3632                 return -ENOMEM;
3633
3634         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3635         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
3636         if (!free_i->free_secmap)
3637                 return -ENOMEM;
3638
3639         /* set all segments as dirty temporarily */
3640         memset(free_i->free_segmap, 0xff, bitmap_size);
3641         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3642
3643         /* init free segmap information */
3644         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
3645         free_i->free_segments = 0;
3646         free_i->free_sections = 0;
3647         spin_lock_init(&free_i->segmap_lock);
3648         return 0;
3649 }
3650
3651 static int build_curseg(struct f2fs_sb_info *sbi)
3652 {
3653         struct curseg_info *array;
3654         int i;
3655
3656         array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
3657                              GFP_KERNEL);
3658         if (!array)
3659                 return -ENOMEM;
3660
3661         SM_I(sbi)->curseg_array = array;
3662
3663         for (i = 0; i < NR_CURSEG_TYPE; i++) {
3664                 mutex_init(&array[i].curseg_mutex);
3665                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
3666                 if (!array[i].sum_blk)
3667                         return -ENOMEM;
3668                 init_rwsem(&array[i].journal_rwsem);
3669                 array[i].journal = f2fs_kzalloc(sbi,
3670                                 sizeof(struct f2fs_journal), GFP_KERNEL);
3671                 if (!array[i].journal)
3672                         return -ENOMEM;
3673                 array[i].segno = NULL_SEGNO;
3674                 array[i].next_blkoff = 0;
3675         }
3676         return restore_curseg_summaries(sbi);
3677 }
3678
3679 static int build_sit_entries(struct f2fs_sb_info *sbi)
3680 {
3681         struct sit_info *sit_i = SIT_I(sbi);
3682         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3683         struct f2fs_journal *journal = curseg->journal;
3684         struct seg_entry *se;
3685         struct f2fs_sit_entry sit;
3686         int sit_blk_cnt = SIT_BLK_CNT(sbi);
3687         unsigned int i, start, end;
3688         unsigned int readed, start_blk = 0;
3689         int err = 0;
3690         block_t total_node_blocks = 0;
3691
3692         do {
3693                 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3694                                                         META_SIT, true);
3695
3696                 start = start_blk * sit_i->sents_per_block;
3697                 end = (start_blk + readed) * sit_i->sents_per_block;
3698
3699                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
3700                         struct f2fs_sit_block *sit_blk;
3701                         struct page *page;
3702
3703                         se = &sit_i->sentries[start];
3704                         page = get_current_sit_page(sbi, start);
3705                         sit_blk = (struct f2fs_sit_block *)page_address(page);
3706                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3707                         f2fs_put_page(page, 1);
3708
3709                         err = check_block_count(sbi, start, &sit);
3710                         if (err)
3711                                 return err;
3712                         seg_info_from_raw_sit(se, &sit);
3713                         if (IS_NODESEG(se->type))
3714                                 total_node_blocks += se->valid_blocks;
3715
3716                         /* build discard map only one time */
3717                         if (f2fs_discard_en(sbi)) {
3718                                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3719                                         memset(se->discard_map, 0xff,
3720                                                 SIT_VBLOCK_MAP_SIZE);
3721                                 } else {
3722                                         memcpy(se->discard_map,
3723                                                 se->cur_valid_map,
3724                                                 SIT_VBLOCK_MAP_SIZE);
3725                                         sbi->discard_blks +=
3726                                                 sbi->blocks_per_seg -
3727                                                 se->valid_blocks;
3728                                 }
3729                         }
3730
3731                         if (sbi->segs_per_sec > 1)
3732                                 get_sec_entry(sbi, start)->valid_blocks +=
3733                                                         se->valid_blocks;
3734                 }
3735                 start_blk += readed;
3736         } while (start_blk < sit_blk_cnt);
3737
3738         down_read(&curseg->journal_rwsem);
3739         for (i = 0; i < sits_in_cursum(journal); i++) {
3740                 unsigned int old_valid_blocks;
3741
3742                 start = le32_to_cpu(segno_in_journal(journal, i));
3743                 if (start >= MAIN_SEGS(sbi)) {
3744                         f2fs_msg(sbi->sb, KERN_ERR,
3745                                         "Wrong journal entry on segno %u",
3746                                         start);
3747                         set_sbi_flag(sbi, SBI_NEED_FSCK);
3748                         err = -EINVAL;
3749                         break;
3750                 }
3751
3752                 se = &sit_i->sentries[start];
3753                 sit = sit_in_journal(journal, i);
3754
3755                 old_valid_blocks = se->valid_blocks;
3756                 if (IS_NODESEG(se->type))
3757                         total_node_blocks -= old_valid_blocks;
3758
3759                 err = check_block_count(sbi, start, &sit);
3760                 if (err)
3761                         break;
3762                 seg_info_from_raw_sit(se, &sit);
3763                 if (IS_NODESEG(se->type))
3764                         total_node_blocks += se->valid_blocks;
3765
3766                 if (f2fs_discard_en(sbi)) {
3767                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3768                                 memset(se->discard_map, 0xff,
3769                                                         SIT_VBLOCK_MAP_SIZE);
3770                         } else {
3771                                 memcpy(se->discard_map, se->cur_valid_map,
3772                                                         SIT_VBLOCK_MAP_SIZE);
3773                                 sbi->discard_blks += old_valid_blocks;
3774                                 sbi->discard_blks -= se->valid_blocks;
3775                         }
3776                 }
3777
3778                 if (sbi->segs_per_sec > 1) {
3779                         get_sec_entry(sbi, start)->valid_blocks +=
3780                                                         se->valid_blocks;
3781                         get_sec_entry(sbi, start)->valid_blocks -=
3782                                                         old_valid_blocks;
3783                 }
3784         }
3785         up_read(&curseg->journal_rwsem);
3786
3787         if (!err && total_node_blocks != valid_node_count(sbi)) {
3788                 f2fs_msg(sbi->sb, KERN_ERR,
3789                         "SIT is corrupted node# %u vs %u",
3790                         total_node_blocks, valid_node_count(sbi));
3791                 set_sbi_flag(sbi, SBI_NEED_FSCK);
3792                 err = -EINVAL;
3793         }
3794
3795         return err;
3796 }
3797
3798 static void init_free_segmap(struct f2fs_sb_info *sbi)
3799 {
3800         unsigned int start;
3801         int type;
3802
3803         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3804                 struct seg_entry *sentry = get_seg_entry(sbi, start);
3805                 if (!sentry->valid_blocks)
3806                         __set_free(sbi, start);
3807                 else
3808                         SIT_I(sbi)->written_valid_blocks +=
3809                                                 sentry->valid_blocks;
3810         }
3811
3812         /* set use the current segments */
3813         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3814                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3815                 __set_test_and_inuse(sbi, curseg_t->segno);
3816         }
3817 }
3818
3819 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3820 {
3821         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3822         struct free_segmap_info *free_i = FREE_I(sbi);
3823         unsigned int segno = 0, offset = 0;
3824         unsigned short valid_blocks;
3825
3826         while (1) {
3827                 /* find dirty segment based on free segmap */
3828                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3829                 if (segno >= MAIN_SEGS(sbi))
3830                         break;
3831                 offset = segno + 1;
3832                 valid_blocks = get_valid_blocks(sbi, segno, false);
3833                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
3834                         continue;
3835                 if (valid_blocks > sbi->blocks_per_seg) {
3836                         f2fs_bug_on(sbi, 1);
3837                         continue;
3838                 }
3839                 mutex_lock(&dirty_i->seglist_lock);
3840                 __locate_dirty_segment(sbi, segno, DIRTY);
3841                 mutex_unlock(&dirty_i->seglist_lock);
3842         }
3843 }
3844
3845 static int init_victim_secmap(struct f2fs_sb_info *sbi)
3846 {
3847         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3848         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3849
3850         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
3851         if (!dirty_i->victim_secmap)
3852                 return -ENOMEM;
3853         return 0;
3854 }
3855
3856 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3857 {
3858         struct dirty_seglist_info *dirty_i;
3859         unsigned int bitmap_size, i;
3860
3861         /* allocate memory for dirty segments list information */
3862         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
3863                                                                 GFP_KERNEL);
3864         if (!dirty_i)
3865                 return -ENOMEM;
3866
3867         SM_I(sbi)->dirty_info = dirty_i;
3868         mutex_init(&dirty_i->seglist_lock);
3869
3870         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3871
3872         for (i = 0; i < NR_DIRTY_TYPE; i++) {
3873                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
3874                                                                 GFP_KERNEL);
3875                 if (!dirty_i->dirty_segmap[i])
3876                         return -ENOMEM;
3877         }
3878
3879         init_dirty_segmap(sbi);
3880         return init_victim_secmap(sbi);
3881 }
3882
3883 /*
3884  * Update min, max modified time for cost-benefit GC algorithm
3885  */
3886 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3887 {
3888         struct sit_info *sit_i = SIT_I(sbi);
3889         unsigned int segno;
3890
3891         down_write(&sit_i->sentry_lock);
3892
3893         sit_i->min_mtime = ULLONG_MAX;
3894
3895         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
3896                 unsigned int i;
3897                 unsigned long long mtime = 0;
3898
3899                 for (i = 0; i < sbi->segs_per_sec; i++)
3900                         mtime += get_seg_entry(sbi, segno + i)->mtime;
3901
3902                 mtime = div_u64(mtime, sbi->segs_per_sec);
3903
3904                 if (sit_i->min_mtime > mtime)
3905                         sit_i->min_mtime = mtime;
3906         }
3907         sit_i->max_mtime = get_mtime(sbi, false);
3908         up_write(&sit_i->sentry_lock);
3909 }
3910
3911 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
3912 {
3913         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3914         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3915         struct f2fs_sm_info *sm_info;
3916         int err;
3917
3918         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
3919         if (!sm_info)
3920                 return -ENOMEM;
3921
3922         /* init sm info */
3923         sbi->sm_info = sm_info;
3924         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3925         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3926         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3927         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3928         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3929         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3930         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
3931         sm_info->rec_prefree_segments = sm_info->main_segments *
3932                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
3933         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3934                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3935
3936         if (!test_opt(sbi, LFS))
3937                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
3938         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
3939         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
3940         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
3941         sm_info->min_ssr_sections = reserved_sections(sbi);
3942
3943         INIT_LIST_HEAD(&sm_info->sit_entry_set);
3944
3945         init_rwsem(&sm_info->curseg_lock);
3946
3947         if (!f2fs_readonly(sbi->sb)) {
3948                 err = f2fs_create_flush_cmd_control(sbi);
3949                 if (err)
3950                         return err;
3951         }
3952
3953         err = create_discard_cmd_control(sbi);
3954         if (err)
3955                 return err;
3956
3957         err = build_sit_info(sbi);
3958         if (err)
3959                 return err;
3960         err = build_free_segmap(sbi);
3961         if (err)
3962                 return err;
3963         err = build_curseg(sbi);
3964         if (err)
3965                 return err;
3966
3967         /* reinit free segmap based on SIT */
3968         err = build_sit_entries(sbi);
3969         if (err)
3970                 return err;
3971
3972         init_free_segmap(sbi);
3973         err = build_dirty_segmap(sbi);
3974         if (err)
3975                 return err;
3976
3977         init_min_max_mtime(sbi);
3978         return 0;
3979 }
3980
3981 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3982                 enum dirty_type dirty_type)
3983 {
3984         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3985
3986         mutex_lock(&dirty_i->seglist_lock);
3987         kvfree(dirty_i->dirty_segmap[dirty_type]);
3988         dirty_i->nr_dirty[dirty_type] = 0;
3989         mutex_unlock(&dirty_i->seglist_lock);
3990 }
3991
3992 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
3993 {
3994         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3995         kvfree(dirty_i->victim_secmap);
3996 }
3997
3998 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3999 {
4000         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4001         int i;
4002
4003         if (!dirty_i)
4004                 return;
4005
4006         /* discard pre-free/dirty segments list */
4007         for (i = 0; i < NR_DIRTY_TYPE; i++)
4008                 discard_dirty_segmap(sbi, i);
4009
4010         destroy_victim_secmap(sbi);
4011         SM_I(sbi)->dirty_info = NULL;
4012         kfree(dirty_i);
4013 }
4014
4015 static void destroy_curseg(struct f2fs_sb_info *sbi)
4016 {
4017         struct curseg_info *array = SM_I(sbi)->curseg_array;
4018         int i;
4019
4020         if (!array)
4021                 return;
4022         SM_I(sbi)->curseg_array = NULL;
4023         for (i = 0; i < NR_CURSEG_TYPE; i++) {
4024                 kfree(array[i].sum_blk);
4025                 kfree(array[i].journal);
4026         }
4027         kfree(array);
4028 }
4029
4030 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4031 {
4032         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4033         if (!free_i)
4034                 return;
4035         SM_I(sbi)->free_info = NULL;
4036         kvfree(free_i->free_segmap);
4037         kvfree(free_i->free_secmap);
4038         kfree(free_i);
4039 }
4040
4041 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4042 {
4043         struct sit_info *sit_i = SIT_I(sbi);
4044         unsigned int start;
4045
4046         if (!sit_i)
4047                 return;
4048
4049         if (sit_i->sentries) {
4050                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4051                         kfree(sit_i->sentries[start].cur_valid_map);
4052 #ifdef CONFIG_F2FS_CHECK_FS
4053                         kfree(sit_i->sentries[start].cur_valid_map_mir);
4054 #endif
4055                         kfree(sit_i->sentries[start].ckpt_valid_map);
4056                         kfree(sit_i->sentries[start].discard_map);
4057                 }
4058         }
4059         kfree(sit_i->tmp_map);
4060
4061         kvfree(sit_i->sentries);
4062         kvfree(sit_i->sec_entries);
4063         kvfree(sit_i->dirty_sentries_bitmap);
4064
4065         SM_I(sbi)->sit_info = NULL;
4066         kfree(sit_i->sit_bitmap);
4067 #ifdef CONFIG_F2FS_CHECK_FS
4068         kfree(sit_i->sit_bitmap_mir);
4069 #endif
4070         kfree(sit_i);
4071 }
4072
4073 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
4074 {
4075         struct f2fs_sm_info *sm_info = SM_I(sbi);
4076
4077         if (!sm_info)
4078                 return;
4079         f2fs_destroy_flush_cmd_control(sbi, true);
4080         destroy_discard_cmd_control(sbi);
4081         destroy_dirty_segmap(sbi);
4082         destroy_curseg(sbi);
4083         destroy_free_segmap(sbi);
4084         destroy_sit_info(sbi);
4085         sbi->sm_info = NULL;
4086         kfree(sm_info);
4087 }
4088
4089 int __init f2fs_create_segment_manager_caches(void)
4090 {
4091         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4092                         sizeof(struct discard_entry));
4093         if (!discard_entry_slab)
4094                 goto fail;
4095
4096         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4097                         sizeof(struct discard_cmd));
4098         if (!discard_cmd_slab)
4099                 goto destroy_discard_entry;
4100
4101         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4102                         sizeof(struct sit_entry_set));
4103         if (!sit_entry_set_slab)
4104                 goto destroy_discard_cmd;
4105
4106         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4107                         sizeof(struct inmem_pages));
4108         if (!inmem_entry_slab)
4109                 goto destroy_sit_entry_set;
4110         return 0;
4111
4112 destroy_sit_entry_set:
4113         kmem_cache_destroy(sit_entry_set_slab);
4114 destroy_discard_cmd:
4115         kmem_cache_destroy(discard_cmd_slab);
4116 destroy_discard_entry:
4117         kmem_cache_destroy(discard_entry_slab);
4118 fail:
4119         return -ENOMEM;
4120 }
4121
4122 void f2fs_destroy_segment_manager_caches(void)
4123 {
4124         kmem_cache_destroy(sit_entry_set_slab);
4125         kmem_cache_destroy(discard_cmd_slab);
4126         kmem_cache_destroy(discard_entry_slab);
4127         kmem_cache_destroy(inmem_entry_slab);
4128 }