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 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_thread && sbi->gc_thread->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 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                 if (recover) {
 234                         struct dnode_of_data dn;
 235                         struct node_info ni;
 236
 237                         trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 238 retry:
 239                         set_new_dnode(&dn, inode, NULL, NULL, 0);
 240                         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
 241                         if (err) {
 242                                 if (err == -ENOMEM) {
 243                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 244                                         cond_resched();
 245                                         goto retry;
 246                                 }
 247                                 err = -EAGAIN;
 248                                 goto next;
 249                         }
 250                         get_node_info(sbi, dn.nid, &ni);
 251                         if (cur->old_addr == NEW_ADDR) {
 252                                 invalidate_blocks(sbi, dn.data_blkaddr);
 253                                 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
 254                         } else
 255                                 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 256                                         cur->old_addr, ni.version, true, true);
 257                         f2fs_put_dnode(&dn);
 258                 }
 259 next:
 260                 /* we don't need to invalidate this in the sccessful status */
 261                 if (drop || recover)
 262                         ClearPageUptodate(page);
 263                 set_page_private(page, 0);
 264                 ClearPagePrivate(page);
 265                 f2fs_put_page(page, 1);
 266
 267                 list_del(&cur->list);
 268                 kmem_cache_free(inmem_entry_slab, cur);
 269                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 270         }
 271         return err;
 272 }
 273
 274 void drop_inmem_pages_all(struct f2fs_sb_info *sbi)
 275 {
 276         struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
 277         struct inode *inode;
 278         struct f2fs_inode_info *fi;
 279 next:
 280         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 281         if (list_empty(head)) {
 282                 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 283                 return;
 284         }
 285         fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
 286         inode = igrab(&fi->vfs_inode);
 287         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 288
 289         if (inode) {
 290                 drop_inmem_pages(inode);
 291                 iput(inode);
 292         }
 293         congestion_wait(BLK_RW_ASYNC, HZ/50);
 294         cond_resched();
 295         goto next;
 296 }
 297
 298 void drop_inmem_pages(struct inode *inode)
 299 {
 300         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 301         struct f2fs_inode_info *fi = F2FS_I(inode);
 302
 303         mutex_lock(&fi->inmem_lock);
 304         __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 305         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 306         if (!list_empty(&fi->inmem_ilist))
 307                 list_del_init(&fi->inmem_ilist);
 308         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 309         mutex_unlock(&fi->inmem_lock);
 310
 311         clear_inode_flag(inode, FI_ATOMIC_FILE);
 312         clear_inode_flag(inode, FI_HOT_DATA);
 313         stat_dec_atomic_write(inode);
 314 }
 315
 316 void drop_inmem_page(struct inode *inode, struct page *page)
 317 {
 318         struct f2fs_inode_info *fi = F2FS_I(inode);
 319         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 320         struct list_head *head = &fi->inmem_pages;
 321         struct inmem_pages *cur = NULL;
 322
 323         f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
 324
 325         mutex_lock(&fi->inmem_lock);
 326         list_for_each_entry(cur, head, list) {
 327                 if (cur->page == page)
 328                         break;
 329         }
 330
 331         f2fs_bug_on(sbi, !cur || cur->page != page);
 332         list_del(&cur->list);
 333         mutex_unlock(&fi->inmem_lock);
 334
 335         dec_page_count(sbi, F2FS_INMEM_PAGES);
 336         kmem_cache_free(inmem_entry_slab, cur);
 337
 338         ClearPageUptodate(page);
 339         set_page_private(page, 0);
 340         ClearPagePrivate(page);
 341         f2fs_put_page(page, 0);
 342
 343         trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 344 }
 345
 346 static int __commit_inmem_pages(struct inode *inode,
 347                                         struct list_head *revoke_list)
 348 {
 349         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 350         struct f2fs_inode_info *fi = F2FS_I(inode);
 351         struct inmem_pages *cur, *tmp;
 352         struct f2fs_io_info fio = {
 353                 .sbi = sbi,
 354                 .ino = inode->i_ino,
 355                 .type = DATA,
 356                 .op = REQ_OP_WRITE,
 357                 .op_flags = REQ_SYNC | REQ_PRIO,
 358                 .io_type = FS_DATA_IO,
 359         };
 360         pgoff_t last_idx = ULONG_MAX;
 361         int err = 0;
 362
 363         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 364                 struct page *page = cur->page;
 365
 366                 lock_page(page);
 367                 if (page->mapping == inode->i_mapping) {
 368                         trace_f2fs_commit_inmem_page(page, INMEM);
 369
 370                         set_page_dirty(page);
 371                         f2fs_wait_on_page_writeback(page, DATA, true);
 372                         if (clear_page_dirty_for_io(page)) {
 373                                 inode_dec_dirty_pages(inode);
 374                                 remove_dirty_inode(inode);
 375                         }
 376 retry:
 377                         fio.page = page;
 378                         fio.old_blkaddr = NULL_ADDR;
 379                         fio.encrypted_page = NULL;
 380                         fio.need_lock = LOCK_DONE;
 381                         err = do_write_data_page(&fio);
 382                         if (err) {
 383                                 if (err == -ENOMEM) {
 384                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 385                                         cond_resched();
 386                                         goto retry;
 387                                 }
 388                                 unlock_page(page);
 389                                 break;
 390                         }
 391                         /* record old blkaddr for revoking */
 392                         cur->old_addr = fio.old_blkaddr;
 393                         last_idx = page->index;
 394                 }
 395                 unlock_page(page);
 396                 list_move_tail(&cur->list, revoke_list);
 397         }
 398
 399         if (last_idx != ULONG_MAX)
 400                 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
 401
 402         if (!err)
 403                 __revoke_inmem_pages(inode, revoke_list, false, false);
 404
 405         return err;
 406 }
 407
 408 int commit_inmem_pages(struct inode *inode)
 409 {
 410         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 411         struct f2fs_inode_info *fi = F2FS_I(inode);
 412         struct list_head revoke_list;
 413         int err;
 414
 415         INIT_LIST_HEAD(&revoke_list);
 416         f2fs_balance_fs(sbi, true);
 417         f2fs_lock_op(sbi);
 418
 419         set_inode_flag(inode, FI_ATOMIC_COMMIT);
 420
 421         mutex_lock(&fi->inmem_lock);
 422         err = __commit_inmem_pages(inode, &revoke_list);
 423         if (err) {
 424                 int ret;
 425                 /*
 426                  * try to revoke all committed pages, but still we could fail
 427                  * due to no memory or other reason, if that happened, EAGAIN
 428                  * will be returned, which means in such case, transaction is
 429                  * already not integrity, caller should use journal to do the
 430                  * recovery or rewrite & commit last transaction. For other
 431                  * error number, revoking was done by filesystem itself.
 432                  */
 433                 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
 434                 if (ret)
 435                         err = ret;
 436
 437                 /* drop all uncommitted pages */
 438                 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 439         }
 440         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 441         if (!list_empty(&fi->inmem_ilist))
 442                 list_del_init(&fi->inmem_ilist);
 443         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 444         mutex_unlock(&fi->inmem_lock);
 445
 446         clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 447
 448         f2fs_unlock_op(sbi);
 449         return err;
 450 }
 451
 452 /*
 453  * This function balances dirty node and dentry pages.
 454  * In addition, it controls garbage collection.
 455  */
 456 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 457 {
 458 #ifdef CONFIG_F2FS_FAULT_INJECTION
 459         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 460                 f2fs_show_injection_info(FAULT_CHECKPOINT);
 461                 f2fs_stop_checkpoint(sbi, false);
 462         }
 463 #endif
 464
 465         /* balance_fs_bg is able to be pending */
 466         if (need && excess_cached_nats(sbi))
 467                 f2fs_balance_fs_bg(sbi);
 468
 469         /*
 470          * We should do GC or end up with checkpoint, if there are so many dirty
 471          * dir/node pages without enough free segments.
 472          */
 473         if (has_not_enough_free_secs(sbi, 0, 0)) {
 474                 mutex_lock(&sbi->gc_mutex);
 475                 f2fs_gc(sbi, false, false, NULL_SEGNO);
 476         }
 477 }
 478
 479 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 480 {
 481         /* try to shrink extent cache when there is no enough memory */
 482         if (!available_free_memory(sbi, EXTENT_CACHE))
 483                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 484
 485         /* check the # of cached NAT entries */
 486         if (!available_free_memory(sbi, NAT_ENTRIES))
 487                 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 488
 489         if (!available_free_memory(sbi, FREE_NIDS))
 490                 try_to_free_nids(sbi, MAX_FREE_NIDS);
 491         else
 492                 build_free_nids(sbi, false, false);
 493
 494         if (!is_idle(sbi) && !excess_dirty_nats(sbi))
 495                 return;
 496
 497         /* checkpoint is the only way to shrink partial cached entries */
 498         if (!available_free_memory(sbi, NAT_ENTRIES) ||
 499                         !available_free_memory(sbi, INO_ENTRIES) ||
 500                         excess_prefree_segs(sbi) ||
 501                         excess_dirty_nats(sbi) ||
 502                         f2fs_time_over(sbi, CP_TIME)) {
 503                 if (test_opt(sbi, DATA_FLUSH)) {
 504                         struct blk_plug plug;
 505
 506                         blk_start_plug(&plug);
 507                         sync_dirty_inodes(sbi, FILE_INODE);
 508                         blk_finish_plug(&plug);
 509                 }
 510                 f2fs_sync_fs(sbi->sb, true);
 511                 stat_inc_bg_cp_count(sbi->stat_info);
 512         }
 513 }
 514
 515 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 516                                 struct block_device *bdev)
 517 {
 518         struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
 519         int ret;
 520
 521         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
 522         bio_set_dev(bio, bdev);
 523         ret = submit_bio_wait(bio);
 524         bio_put(bio);
 525
 526         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 527                                 test_opt(sbi, FLUSH_MERGE), ret);
 528         return ret;
 529 }
 530
 531 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 532 {
 533         int ret = 0;
 534         int i;
 535
 536         if (!sbi->s_ndevs)
 537                 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 538
 539         for (i = 0; i < sbi->s_ndevs; i++) {
 540                 if (!is_dirty_device(sbi, ino, i, FLUSH_INO))
 541                         continue;
 542                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 543                 if (ret)
 544                         break;
 545         }
 546         return ret;
 547 }
 548
 549 static int issue_flush_thread(void *data)
 550 {
 551         struct f2fs_sb_info *sbi = data;
 552         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 553         wait_queue_head_t *q = &fcc->flush_wait_queue;
 554 repeat:
 555         if (kthread_should_stop())
 556                 return 0;
 557
 558         sb_start_intwrite(sbi->sb);
 559
 560         if (!llist_empty(&fcc->issue_list)) {
 561                 struct flush_cmd *cmd, *next;
 562                 int ret;
 563
 564                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 565                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 566
 567                 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 568
 569                 ret = submit_flush_wait(sbi, cmd->ino);
 570                 atomic_inc(&fcc->issued_flush);
 571
 572                 llist_for_each_entry_safe(cmd, next,
 573                                           fcc->dispatch_list, llnode) {
 574                         cmd->ret = ret;
 575                         complete(&cmd->wait);
 576                 }
 577                 fcc->dispatch_list = NULL;
 578         }
 579
 580         sb_end_intwrite(sbi->sb);
 581
 582         wait_event_interruptible(*q,
 583                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 584         goto repeat;
 585 }
 586
 587 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 588 {
 589         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 590         struct flush_cmd cmd;
 591         int ret;
 592
 593         if (test_opt(sbi, NOBARRIER))
 594                 return 0;
 595
 596         if (!test_opt(sbi, FLUSH_MERGE)) {
 597                 ret = submit_flush_wait(sbi, ino);
 598                 atomic_inc(&fcc->issued_flush);
 599                 return ret;
 600         }
 601
 602         if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
 603                 ret = submit_flush_wait(sbi, ino);
 604                 atomic_dec(&fcc->issing_flush);
 605
 606                 atomic_inc(&fcc->issued_flush);
 607                 return ret;
 608         }
 609
 610         cmd.ino = ino;
 611         init_completion(&cmd.wait);
 612
 613         llist_add(&cmd.llnode, &fcc->issue_list);
 614
 615         /* update issue_list before we wake up issue_flush thread */
 616         smp_mb();
 617
 618         if (waitqueue_active(&fcc->flush_wait_queue))
 619                 wake_up(&fcc->flush_wait_queue);
 620
 621         if (fcc->f2fs_issue_flush) {
 622                 wait_for_completion(&cmd.wait);
 623                 atomic_dec(&fcc->issing_flush);
 624         } else {
 625                 struct llist_node *list;
 626
 627                 list = llist_del_all(&fcc->issue_list);
 628                 if (!list) {
 629                         wait_for_completion(&cmd.wait);
 630                         atomic_dec(&fcc->issing_flush);
 631                 } else {
 632                         struct flush_cmd *tmp, *next;
 633
 634                         ret = submit_flush_wait(sbi, ino);
 635
 636                         llist_for_each_entry_safe(tmp, next, list, llnode) {
 637                                 if (tmp == &cmd) {
 638                                         cmd.ret = ret;
 639                                         atomic_dec(&fcc->issing_flush);
 640                                         continue;
 641                                 }
 642                                 tmp->ret = ret;
 643                                 complete(&tmp->wait);
 644                         }
 645                 }
 646         }
 647
 648         return cmd.ret;
 649 }
 650
 651 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
 652 {
 653         dev_t dev = sbi->sb->s_bdev->bd_dev;
 654         struct flush_cmd_control *fcc;
 655         int err = 0;
 656
 657         if (SM_I(sbi)->fcc_info) {
 658                 fcc = SM_I(sbi)->fcc_info;
 659                 if (fcc->f2fs_issue_flush)
 660                         return err;
 661                 goto init_thread;
 662         }
 663
 664         fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
 665         if (!fcc)
 666                 return -ENOMEM;
 667         atomic_set(&fcc->issued_flush, 0);
 668         atomic_set(&fcc->issing_flush, 0);
 669         init_waitqueue_head(&fcc->flush_wait_queue);
 670         init_llist_head(&fcc->issue_list);
 671         SM_I(sbi)->fcc_info = fcc;
 672         if (!test_opt(sbi, FLUSH_MERGE))
 673                 return err;
 674
 675 init_thread:
 676         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 677                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 678         if (IS_ERR(fcc->f2fs_issue_flush)) {
 679                 err = PTR_ERR(fcc->f2fs_issue_flush);
 680                 kfree(fcc);
 681                 SM_I(sbi)->fcc_info = NULL;
 682                 return err;
 683         }
 684
 685         return err;
 686 }
 687
 688 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 689 {
 690         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 691
 692         if (fcc && fcc->f2fs_issue_flush) {
 693                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 694
 695                 fcc->f2fs_issue_flush = NULL;
 696                 kthread_stop(flush_thread);
 697         }
 698         if (free) {
 699                 kfree(fcc);
 700                 SM_I(sbi)->fcc_info = NULL;
 701         }
 702 }
 703
 704 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
 705 {
 706         int ret = 0, i;
 707
 708         if (!sbi->s_ndevs)
 709                 return 0;
 710
 711         for (i = 1; i < sbi->s_ndevs; i++) {
 712                 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
 713                         continue;
 714                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 715                 if (ret)
 716                         break;
 717
 718                 spin_lock(&sbi->dev_lock);
 719                 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
 720                 spin_unlock(&sbi->dev_lock);
 721         }
 722
 723         return ret;
 724 }
 725
 726 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 727                 enum dirty_type dirty_type)
 728 {
 729         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 730
 731         /* need not be added */
 732         if (IS_CURSEG(sbi, segno))
 733                 return;
 734
 735         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 736                 dirty_i->nr_dirty[dirty_type]++;
 737
 738         if (dirty_type == DIRTY) {
 739                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 740                 enum dirty_type t = sentry->type;
 741
 742                 if (unlikely(t >= DIRTY)) {
 743                         f2fs_bug_on(sbi, 1);
 744                         return;
 745                 }
 746                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 747                         dirty_i->nr_dirty[t]++;
 748         }
 749 }
 750
 751 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 752                 enum dirty_type dirty_type)
 753 {
 754         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 755
 756         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 757                 dirty_i->nr_dirty[dirty_type]--;
 758
 759         if (dirty_type == DIRTY) {
 760                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 761                 enum dirty_type t = sentry->type;
 762
 763                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 764                         dirty_i->nr_dirty[t]--;
 765
 766                 if (get_valid_blocks(sbi, segno, true) == 0)
 767                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 768                                                 dirty_i->victim_secmap);
 769         }
 770 }
 771
 772 /*
 773  * Should not occur error such as -ENOMEM.
 774  * Adding dirty entry into seglist is not critical operation.
 775  * If a given segment is one of current working segments, it won't be added.
 776  */
 777 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 778 {
 779         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 780         unsigned short valid_blocks;
 781
 782         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 783                 return;
 784
 785         mutex_lock(&dirty_i->seglist_lock);
 786
 787         valid_blocks = get_valid_blocks(sbi, segno, false);
 788
 789         if (valid_blocks == 0) {
 790                 __locate_dirty_segment(sbi, segno, PRE);
 791                 __remove_dirty_segment(sbi, segno, DIRTY);
 792         } else if (valid_blocks < sbi->blocks_per_seg) {
 793                 __locate_dirty_segment(sbi, segno, DIRTY);
 794         } else {
 795                 /* Recovery routine with SSR needs this */
 796                 __remove_dirty_segment(sbi, segno, DIRTY);
 797         }
 798
 799         mutex_unlock(&dirty_i->seglist_lock);
 800 }
 801
 802 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 803                 struct block_device *bdev, block_t lstart,
 804                 block_t start, block_t len)
 805 {
 806         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 807         struct list_head *pend_list;
 808         struct discard_cmd *dc;
 809
 810         f2fs_bug_on(sbi, !len);
 811
 812         pend_list = &dcc->pend_list[plist_idx(len)];
 813
 814         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 815         INIT_LIST_HEAD(&dc->list);
 816         dc->bdev = bdev;
 817         dc->lstart = lstart;
 818         dc->start = start;
 819         dc->len = len;
 820         dc->ref = 0;
 821         dc->state = D_PREP;
 822         dc->error = 0;
 823         init_completion(&dc->wait);
 824         list_add_tail(&dc->list, pend_list);
 825         atomic_inc(&dcc->discard_cmd_cnt);
 826         dcc->undiscard_blks += len;
 827
 828         return dc;
 829 }
 830
 831 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 832                                 struct block_device *bdev, block_t lstart,
 833                                 block_t start, block_t len,
 834                                 struct rb_node *parent, struct rb_node **p)
 835 {
 836         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 837         struct discard_cmd *dc;
 838
 839         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
 840
 841         rb_link_node(&dc->rb_node, parent, p);
 842         rb_insert_color(&dc->rb_node, &dcc->root);
 843
 844         return dc;
 845 }
 846
 847 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
 848                                                         struct discard_cmd *dc)
 849 {
 850         if (dc->state == D_DONE)
 851                 atomic_dec(&dcc->issing_discard);
 852
 853         list_del(&dc->list);
 854         rb_erase(&dc->rb_node, &dcc->root);
 855         dcc->undiscard_blks -= dc->len;
 856
 857         kmem_cache_free(discard_cmd_slab, dc);
 858
 859         atomic_dec(&dcc->discard_cmd_cnt);
 860 }
 861
 862 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
 863                                                         struct discard_cmd *dc)
 864 {
 865         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 866
 867         trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
 868
 869         f2fs_bug_on(sbi, dc->ref);
 870
 871         if (dc->error == -EOPNOTSUPP)
 872                 dc->error = 0;
 873
 874         if (dc->error)
 875                 f2fs_msg(sbi->sb, KERN_INFO,
 876                         "Issue discard(%u, %u, %u) failed, ret: %d",
 877                         dc->lstart, dc->start, dc->len, dc->error);
 878         __detach_discard_cmd(dcc, dc);
 879 }
 880
 881 static void f2fs_submit_discard_endio(struct bio *bio)
 882 {
 883         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
 884
 885         dc->error = blk_status_to_errno(bio->bi_status);
 886         dc->state = D_DONE;
 887         complete_all(&dc->wait);
 888         bio_put(bio);
 889 }
 890
 891 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
 892                                 block_t start, block_t end)
 893 {
 894 #ifdef CONFIG_F2FS_CHECK_FS
 895         struct seg_entry *sentry;
 896         unsigned int segno;
 897         block_t blk = start;
 898         unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
 899         unsigned long *map;
 900
 901         while (blk < end) {
 902                 segno = GET_SEGNO(sbi, blk);
 903                 sentry = get_seg_entry(sbi, segno);
 904                 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
 905
 906                 if (end < START_BLOCK(sbi, segno + 1))
 907                         size = GET_BLKOFF_FROM_SEG0(sbi, end);
 908                 else
 909                         size = max_blocks;
 910                 map = (unsigned long *)(sentry->cur_valid_map);
 911                 offset = __find_rev_next_bit(map, size, offset);
 912                 f2fs_bug_on(sbi, offset != size);
 913                 blk = START_BLOCK(sbi, segno + 1);
 914         }
 915 #endif
 916 }
 917
 918 static void __init_discard_policy(struct f2fs_sb_info *sbi,
 919                                 struct discard_policy *dpolicy,
 920                                 int discard_type, unsigned int granularity)
 921 {
 922         /* common policy */
 923         dpolicy->type = discard_type;
 924         dpolicy->sync = true;
 925         dpolicy->granularity = granularity;
 926
 927         dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
 928         dpolicy->io_aware_gran = MAX_PLIST_NUM;
 929
 930         if (discard_type == DPOLICY_BG) {
 931                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
 932                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
 933                 dpolicy->io_aware = true;
 934                 dpolicy->sync = false;
 935                 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
 936                         dpolicy->granularity = 1;
 937                         dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
 938                 }
 939         } else if (discard_type == DPOLICY_FORCE) {
 940                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
 941                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
 942                 dpolicy->io_aware = false;
 943         } else if (discard_type == DPOLICY_FSTRIM) {
 944                 dpolicy->io_aware = false;
 945         } else if (discard_type == DPOLICY_UMOUNT) {
 946                 dpolicy->max_requests = UINT_MAX;
 947                 dpolicy->io_aware = false;
 948         }
 949 }
 950
 951
 952 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
 953 static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
 954                                                 struct discard_policy *dpolicy,
 955                                                 struct discard_cmd *dc)
 956 {
 957         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 958         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
 959                                         &(dcc->fstrim_list) : &(dcc->wait_list);
 960         struct bio *bio = NULL;
 961         int flag = dpolicy->sync ? REQ_SYNC : 0;
 962
 963         if (dc->state != D_PREP)
 964                 return;
 965
 966         trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
 967
 968         dc->error = __blkdev_issue_discard(dc->bdev,
 969                                 SECTOR_FROM_BLOCK(dc->start),
 970                                 SECTOR_FROM_BLOCK(dc->len),
 971                                 GFP_NOFS, 0, &bio);
 972         if (!dc->error) {
 973                 /* should keep before submission to avoid D_DONE right away */
 974                 dc->state = D_SUBMIT;
 975                 atomic_inc(&dcc->issued_discard);
 976                 atomic_inc(&dcc->issing_discard);
 977                 if (bio) {
 978                         bio->bi_private = dc;
 979                         bio->bi_end_io = f2fs_submit_discard_endio;
 980                         bio->bi_opf |= flag;
 981                         submit_bio(bio);
 982                         list_move_tail(&dc->list, wait_list);
 983                         __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
 984
 985                         f2fs_update_iostat(sbi, FS_DISCARD, 1);
 986                 }
 987         } else {
 988                 __remove_discard_cmd(sbi, dc);
 989         }
 990 }
 991
 992 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
 993                                 struct block_device *bdev, block_t lstart,
 994                                 block_t start, block_t len,
 995                                 struct rb_node **insert_p,
 996                                 struct rb_node *insert_parent)
 997 {
 998         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 999         struct rb_node **p;
1000         struct rb_node *parent = NULL;
1001         struct discard_cmd *dc = NULL;
1002
1003         if (insert_p && insert_parent) {
1004                 parent = insert_parent;
1005                 p = insert_p;
1006                 goto do_insert;
1007         }
1008
1009         p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
1010 do_insert:
1011         dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
1012         if (!dc)
1013                 return NULL;
1014
1015         return dc;
1016 }
1017
1018 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1019                                                 struct discard_cmd *dc)
1020 {
1021         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1022 }
1023
1024 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1025                                 struct discard_cmd *dc, block_t blkaddr)
1026 {
1027         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1028         struct discard_info di = dc->di;
1029         bool modified = false;
1030
1031         if (dc->state == D_DONE || dc->len == 1) {
1032                 __remove_discard_cmd(sbi, dc);
1033                 return;
1034         }
1035
1036         dcc->undiscard_blks -= di.len;
1037
1038         if (blkaddr > di.lstart) {
1039                 dc->len = blkaddr - dc->lstart;
1040                 dcc->undiscard_blks += dc->len;
1041                 __relocate_discard_cmd(dcc, dc);
1042                 modified = true;
1043         }
1044
1045         if (blkaddr < di.lstart + di.len - 1) {
1046                 if (modified) {
1047                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1048                                         di.start + blkaddr + 1 - di.lstart,
1049                                         di.lstart + di.len - 1 - blkaddr,
1050                                         NULL, NULL);
1051                 } else {
1052                         dc->lstart++;
1053                         dc->len--;
1054                         dc->start++;
1055                         dcc->undiscard_blks += dc->len;
1056                         __relocate_discard_cmd(dcc, dc);
1057                 }
1058         }
1059 }
1060
1061 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1062                                 struct block_device *bdev, block_t lstart,
1063                                 block_t start, block_t len)
1064 {
1065         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1066         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1067         struct discard_cmd *dc;
1068         struct discard_info di = {0};
1069         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1070         block_t end = lstart + len;
1071
1072         mutex_lock(&dcc->cmd_lock);
1073
1074         dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1075                                         NULL, lstart,
1076                                         (struct rb_entry **)&prev_dc,
1077                                         (struct rb_entry **)&next_dc,
1078                                         &insert_p, &insert_parent, true);
1079         if (dc)
1080                 prev_dc = dc;
1081
1082         if (!prev_dc) {
1083                 di.lstart = lstart;
1084                 di.len = next_dc ? next_dc->lstart - lstart : len;
1085                 di.len = min(di.len, len);
1086                 di.start = start;
1087         }
1088
1089         while (1) {
1090                 struct rb_node *node;
1091                 bool merged = false;
1092                 struct discard_cmd *tdc = NULL;
1093
1094                 if (prev_dc) {
1095                         di.lstart = prev_dc->lstart + prev_dc->len;
1096                         if (di.lstart < lstart)
1097                                 di.lstart = lstart;
1098                         if (di.lstart >= end)
1099                                 break;
1100
1101                         if (!next_dc || next_dc->lstart > end)
1102                                 di.len = end - di.lstart;
1103                         else
1104                                 di.len = next_dc->lstart - di.lstart;
1105                         di.start = start + di.lstart - lstart;
1106                 }
1107
1108                 if (!di.len)
1109                         goto next;
1110
1111                 if (prev_dc && prev_dc->state == D_PREP &&
1112                         prev_dc->bdev == bdev &&
1113                         __is_discard_back_mergeable(&di, &prev_dc->di)) {
1114                         prev_dc->di.len += di.len;
1115                         dcc->undiscard_blks += di.len;
1116                         __relocate_discard_cmd(dcc, prev_dc);
1117                         di = prev_dc->di;
1118                         tdc = prev_dc;
1119                         merged = true;
1120                 }
1121
1122                 if (next_dc && next_dc->state == D_PREP &&
1123                         next_dc->bdev == bdev &&
1124                         __is_discard_front_mergeable(&di, &next_dc->di)) {
1125                         next_dc->di.lstart = di.lstart;
1126                         next_dc->di.len += di.len;
1127                         next_dc->di.start = di.start;
1128                         dcc->undiscard_blks += di.len;
1129                         __relocate_discard_cmd(dcc, next_dc);
1130                         if (tdc)
1131                                 __remove_discard_cmd(sbi, tdc);
1132                         merged = true;
1133                 }
1134
1135                 if (!merged) {
1136                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1137                                                         di.len, NULL, NULL);
1138                 }
1139  next:
1140                 prev_dc = next_dc;
1141                 if (!prev_dc)
1142                         break;
1143
1144                 node = rb_next(&prev_dc->rb_node);
1145                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1146         }
1147
1148         mutex_unlock(&dcc->cmd_lock);
1149 }
1150
1151 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1152                 struct block_device *bdev, block_t blkstart, block_t blklen)
1153 {
1154         block_t lblkstart = blkstart;
1155
1156         trace_f2fs_queue_discard(bdev, blkstart, blklen);
1157
1158         if (sbi->s_ndevs) {
1159                 int devi = f2fs_target_device_index(sbi, blkstart);
1160
1161                 blkstart -= FDEV(devi).start_blk;
1162         }
1163         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1164         return 0;
1165 }
1166
1167 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1168                                         struct discard_policy *dpolicy)
1169 {
1170         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1171         struct list_head *pend_list;
1172         struct discard_cmd *dc, *tmp;
1173         struct blk_plug plug;
1174         int i, iter = 0, issued = 0;
1175         bool io_interrupted = false;
1176
1177         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1178                 if (i + 1 < dpolicy->granularity)
1179                         break;
1180                 pend_list = &dcc->pend_list[i];
1181
1182                 mutex_lock(&dcc->cmd_lock);
1183                 if (list_empty(pend_list))
1184                         goto next;
1185                 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1186                 blk_start_plug(&plug);
1187                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1188                         f2fs_bug_on(sbi, dc->state != D_PREP);
1189
1190                         if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1191                                                                 !is_idle(sbi)) {
1192                                 io_interrupted = true;
1193                                 goto skip;
1194                         }
1195
1196                         __submit_discard_cmd(sbi, dpolicy, dc);
1197                         issued++;
1198 skip:
1199                         if (++iter >= dpolicy->max_requests)
1200                                 break;
1201                 }
1202                 blk_finish_plug(&plug);
1203 next:
1204                 mutex_unlock(&dcc->cmd_lock);
1205
1206                 if (iter >= dpolicy->max_requests)
1207                         break;
1208         }
1209
1210         if (!issued && io_interrupted)
1211                 issued = -1;
1212
1213         return issued;
1214 }
1215
1216 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1217 {
1218         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1219         struct list_head *pend_list;
1220         struct discard_cmd *dc, *tmp;
1221         int i;
1222         bool dropped = false;
1223
1224         mutex_lock(&dcc->cmd_lock);
1225         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1226                 pend_list = &dcc->pend_list[i];
1227                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1228                         f2fs_bug_on(sbi, dc->state != D_PREP);
1229                         __remove_discard_cmd(sbi, dc);
1230                         dropped = true;
1231                 }
1232         }
1233         mutex_unlock(&dcc->cmd_lock);
1234
1235         return dropped;
1236 }
1237
1238 void drop_discard_cmd(struct f2fs_sb_info *sbi)
1239 {
1240         __drop_discard_cmd(sbi);
1241 }
1242
1243 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1244                                                         struct discard_cmd *dc)
1245 {
1246         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1247         unsigned int len = 0;
1248
1249         wait_for_completion_io(&dc->wait);
1250         mutex_lock(&dcc->cmd_lock);
1251         f2fs_bug_on(sbi, dc->state != D_DONE);
1252         dc->ref--;
1253         if (!dc->ref) {
1254                 if (!dc->error)
1255                         len = dc->len;
1256                 __remove_discard_cmd(sbi, dc);
1257         }
1258         mutex_unlock(&dcc->cmd_lock);
1259
1260         return len;
1261 }
1262
1263 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1264                                                 struct discard_policy *dpolicy,
1265                                                 block_t start, block_t end)
1266 {
1267         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1268         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1269                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1270         struct discard_cmd *dc, *tmp;
1271         bool need_wait;
1272         unsigned int trimmed = 0;
1273
1274 next:
1275         need_wait = false;
1276
1277         mutex_lock(&dcc->cmd_lock);
1278         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1279                 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1280                         continue;
1281                 if (dc->len < dpolicy->granularity)
1282                         continue;
1283                 if (dc->state == D_DONE && !dc->ref) {
1284                         wait_for_completion_io(&dc->wait);
1285                         if (!dc->error)
1286                                 trimmed += dc->len;
1287                         __remove_discard_cmd(sbi, dc);
1288                 } else {
1289                         dc->ref++;
1290                         need_wait = true;
1291                         break;
1292                 }
1293         }
1294         mutex_unlock(&dcc->cmd_lock);
1295
1296         if (need_wait) {
1297                 trimmed += __wait_one_discard_bio(sbi, dc);
1298                 goto next;
1299         }
1300
1301         return trimmed;
1302 }
1303
1304 static void __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1305                                                 struct discard_policy *dpolicy)
1306 {
1307         struct discard_policy dp;
1308
1309         if (dpolicy) {
1310                 __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1311                 return;
1312         }
1313
1314         /* wait all */
1315         __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1316         __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1317         __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1318         __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1319 }
1320
1321 /* This should be covered by global mutex, &sit_i->sentry_lock */
1322 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1323 {
1324         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1325         struct discard_cmd *dc;
1326         bool need_wait = false;
1327
1328         mutex_lock(&dcc->cmd_lock);
1329         dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1330         if (dc) {
1331                 if (dc->state == D_PREP) {
1332                         __punch_discard_cmd(sbi, dc, blkaddr);
1333                 } else {
1334                         dc->ref++;
1335                         need_wait = true;
1336                 }
1337         }
1338         mutex_unlock(&dcc->cmd_lock);
1339
1340         if (need_wait)
1341                 __wait_one_discard_bio(sbi, dc);
1342 }
1343
1344 void stop_discard_thread(struct f2fs_sb_info *sbi)
1345 {
1346         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1347
1348         if (dcc && dcc->f2fs_issue_discard) {
1349                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1350
1351                 dcc->f2fs_issue_discard = NULL;
1352                 kthread_stop(discard_thread);
1353         }
1354 }
1355
1356 /* This comes from f2fs_put_super */
1357 bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1358 {
1359         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1360         struct discard_policy dpolicy;
1361         bool dropped;
1362
1363         __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1364                                         dcc->discard_granularity);
1365         __issue_discard_cmd(sbi, &dpolicy);
1366         dropped = __drop_discard_cmd(sbi);
1367
1368         /* just to make sure there is no pending discard commands */
1369         __wait_all_discard_cmd(sbi, NULL);
1370         return dropped;
1371 }
1372
1373 static int issue_discard_thread(void *data)
1374 {
1375         struct f2fs_sb_info *sbi = data;
1376         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1377         wait_queue_head_t *q = &dcc->discard_wait_queue;
1378         struct discard_policy dpolicy;
1379         unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1380         int issued;
1381
1382         set_freezable();
1383
1384         do {
1385                 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1386                                         dcc->discard_granularity);
1387
1388                 wait_event_interruptible_timeout(*q,
1389                                 kthread_should_stop() || freezing(current) ||
1390                                 dcc->discard_wake,
1391                                 msecs_to_jiffies(wait_ms));
1392                 if (try_to_freeze())
1393                         continue;
1394                 if (f2fs_readonly(sbi->sb))
1395                         continue;
1396                 if (kthread_should_stop())
1397                         return 0;
1398
1399                 if (dcc->discard_wake)
1400                         dcc->discard_wake = 0;
1401
1402                 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
1403                         __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1404
1405                 sb_start_intwrite(sbi->sb);
1406
1407                 issued = __issue_discard_cmd(sbi, &dpolicy);
1408                 if (issued) {
1409                         __wait_all_discard_cmd(sbi, &dpolicy);
1410                         wait_ms = dpolicy.min_interval;
1411                 } else {
1412                         wait_ms = dpolicy.max_interval;
1413                 }
1414
1415                 sb_end_intwrite(sbi->sb);
1416
1417         } while (!kthread_should_stop());
1418         return 0;
1419 }
1420
1421 #ifdef CONFIG_BLK_DEV_ZONED
1422 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1423                 struct block_device *bdev, block_t blkstart, block_t blklen)
1424 {
1425         sector_t sector, nr_sects;
1426         block_t lblkstart = blkstart;
1427         int devi = 0;
1428
1429         if (sbi->s_ndevs) {
1430                 devi = f2fs_target_device_index(sbi, blkstart);
1431                 blkstart -= FDEV(devi).start_blk;
1432         }
1433
1434         /*
1435          * We need to know the type of the zone: for conventional zones,
1436          * use regular discard if the drive supports it. For sequential
1437          * zones, reset the zone write pointer.
1438          */
1439         switch (get_blkz_type(sbi, bdev, blkstart)) {
1440
1441         case BLK_ZONE_TYPE_CONVENTIONAL:
1442                 if (!blk_queue_discard(bdev_get_queue(bdev)))
1443                         return 0;
1444                 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1445         case BLK_ZONE_TYPE_SEQWRITE_REQ:
1446         case BLK_ZONE_TYPE_SEQWRITE_PREF:
1447                 sector = SECTOR_FROM_BLOCK(blkstart);
1448                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1449
1450                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1451                                 nr_sects != bdev_zone_sectors(bdev)) {
1452                         f2fs_msg(sbi->sb, KERN_INFO,
1453                                 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1454                                 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1455                                 blkstart, blklen);
1456                         return -EIO;
1457                 }
1458                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1459                 return blkdev_reset_zones(bdev, sector,
1460                                           nr_sects, GFP_NOFS);
1461         default:
1462                 /* Unknown zone type: broken device ? */
1463                 return -EIO;
1464         }
1465 }
1466 #endif
1467
1468 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1469                 struct block_device *bdev, block_t blkstart, block_t blklen)
1470 {
1471 #ifdef CONFIG_BLK_DEV_ZONED
1472         if (f2fs_sb_has_blkzoned(sbi->sb) &&
1473                                 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1474                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1475 #endif
1476         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1477 }
1478
1479 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1480                                 block_t blkstart, block_t blklen)
1481 {
1482         sector_t start = blkstart, len = 0;
1483         struct block_device *bdev;
1484         struct seg_entry *se;
1485         unsigned int offset;
1486         block_t i;
1487         int err = 0;
1488
1489         bdev = f2fs_target_device(sbi, blkstart, NULL);
1490
1491         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1492                 if (i != start) {
1493                         struct block_device *bdev2 =
1494                                 f2fs_target_device(sbi, i, NULL);
1495
1496                         if (bdev2 != bdev) {
1497                                 err = __issue_discard_async(sbi, bdev,
1498                                                 start, len);
1499                                 if (err)
1500                                         return err;
1501                                 bdev = bdev2;
1502                                 start = i;
1503                                 len = 0;
1504                         }
1505                 }
1506
1507                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1508                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1509
1510                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1511                         sbi->discard_blks--;
1512         }
1513
1514         if (len)
1515                 err = __issue_discard_async(sbi, bdev, start, len);
1516         return err;
1517 }
1518
1519 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1520                                                         bool check_only)
1521 {
1522         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1523         int max_blocks = sbi->blocks_per_seg;
1524         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1525         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1526         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1527         unsigned long *discard_map = (unsigned long *)se->discard_map;
1528         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1529         unsigned int start = 0, end = -1;
1530         bool force = (cpc->reason & CP_DISCARD);
1531         struct discard_entry *de = NULL;
1532         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1533         int i;
1534
1535         if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
1536                 return false;
1537
1538         if (!force) {
1539                 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
1540                         SM_I(sbi)->dcc_info->nr_discards >=
1541                                 SM_I(sbi)->dcc_info->max_discards)
1542                         return false;
1543         }
1544
1545         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1546         for (i = 0; i < entries; i++)
1547                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1548                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1549
1550         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1551                                 SM_I(sbi)->dcc_info->max_discards) {
1552                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1553                 if (start >= max_blocks)
1554                         break;
1555
1556                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1557                 if (force && start && end != max_blocks
1558                                         && (end - start) < cpc->trim_minlen)
1559                         continue;
1560
1561                 if (check_only)
1562                         return true;
1563
1564                 if (!de) {
1565                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1566                                                                 GFP_F2FS_ZERO);
1567                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1568                         list_add_tail(&de->list, head);
1569                 }
1570
1571                 for (i = start; i < end; i++)
1572                         __set_bit_le(i, (void *)de->discard_map);
1573
1574                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1575         }
1576         return false;
1577 }
1578
1579 void release_discard_addrs(struct f2fs_sb_info *sbi)
1580 {
1581         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1582         struct discard_entry *entry, *this;
1583
1584         /* drop caches */
1585         list_for_each_entry_safe(entry, this, head, list) {
1586                 list_del(&entry->list);
1587                 kmem_cache_free(discard_entry_slab, entry);
1588         }
1589 }
1590
1591 /*
1592  * Should call clear_prefree_segments after checkpoint is done.
1593  */
1594 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1595 {
1596         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1597         unsigned int segno;
1598
1599         mutex_lock(&dirty_i->seglist_lock);
1600         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1601                 __set_test_and_free(sbi, segno);
1602         mutex_unlock(&dirty_i->seglist_lock);
1603 }
1604
1605 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1606 {
1607         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1608         struct list_head *head = &dcc->entry_list;
1609         struct discard_entry *entry, *this;
1610         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1611         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1612         unsigned int start = 0, end = -1;
1613         unsigned int secno, start_segno;
1614         bool force = (cpc->reason & CP_DISCARD);
1615
1616         mutex_lock(&dirty_i->seglist_lock);
1617
1618         while (1) {
1619                 int i;
1620                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1621                 if (start >= MAIN_SEGS(sbi))
1622                         break;
1623                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1624                                                                 start + 1);
1625
1626                 for (i = start; i < end; i++)
1627                         clear_bit(i, prefree_map);
1628
1629                 dirty_i->nr_dirty[PRE] -= end - start;
1630
1631                 if (!test_opt(sbi, DISCARD))
1632                         continue;
1633
1634                 if (force && start >= cpc->trim_start &&
1635                                         (end - 1) <= cpc->trim_end)
1636                                 continue;
1637
1638                 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1639                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1640                                 (end - start) << sbi->log_blocks_per_seg);
1641                         continue;
1642                 }
1643 next:
1644                 secno = GET_SEC_FROM_SEG(sbi, start);
1645                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1646                 if (!IS_CURSEC(sbi, secno) &&
1647                         !get_valid_blocks(sbi, start, true))
1648                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1649                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1650
1651                 start = start_segno + sbi->segs_per_sec;
1652                 if (start < end)
1653                         goto next;
1654                 else
1655                         end = start - 1;
1656         }
1657         mutex_unlock(&dirty_i->seglist_lock);
1658
1659         /* send small discards */
1660         list_for_each_entry_safe(entry, this, head, list) {
1661                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1662                 bool is_valid = test_bit_le(0, entry->discard_map);
1663
1664 find_next:
1665                 if (is_valid) {
1666                         next_pos = find_next_zero_bit_le(entry->discard_map,
1667                                         sbi->blocks_per_seg, cur_pos);
1668                         len = next_pos - cur_pos;
1669
1670                         if (f2fs_sb_has_blkzoned(sbi->sb) ||
1671                             (force && len < cpc->trim_minlen))
1672                                 goto skip;
1673
1674                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1675                                                                         len);
1676                         total_len += len;
1677                 } else {
1678                         next_pos = find_next_bit_le(entry->discard_map,
1679                                         sbi->blocks_per_seg, cur_pos);
1680                 }
1681 skip:
1682                 cur_pos = next_pos;
1683                 is_valid = !is_valid;
1684
1685                 if (cur_pos < sbi->blocks_per_seg)
1686                         goto find_next;
1687
1688                 list_del(&entry->list);
1689                 dcc->nr_discards -= total_len;
1690                 kmem_cache_free(discard_entry_slab, entry);
1691         }
1692
1693         wake_up_discard_thread(sbi, false);
1694 }
1695
1696 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1697 {
1698         dev_t dev = sbi->sb->s_bdev->bd_dev;
1699         struct discard_cmd_control *dcc;
1700         int err = 0, i;
1701
1702         if (SM_I(sbi)->dcc_info) {
1703                 dcc = SM_I(sbi)->dcc_info;
1704                 goto init_thread;
1705         }
1706
1707         dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
1708         if (!dcc)
1709                 return -ENOMEM;
1710
1711         dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
1712         INIT_LIST_HEAD(&dcc->entry_list);
1713         for (i = 0; i < MAX_PLIST_NUM; i++)
1714                 INIT_LIST_HEAD(&dcc->pend_list[i]);
1715         INIT_LIST_HEAD(&dcc->wait_list);
1716         INIT_LIST_HEAD(&dcc->fstrim_list);
1717         mutex_init(&dcc->cmd_lock);
1718         atomic_set(&dcc->issued_discard, 0);
1719         atomic_set(&dcc->issing_discard, 0);
1720         atomic_set(&dcc->discard_cmd_cnt, 0);
1721         dcc->nr_discards = 0;
1722         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1723         dcc->undiscard_blks = 0;
1724         dcc->root = RB_ROOT;
1725
1726         init_waitqueue_head(&dcc->discard_wait_queue);
1727         SM_I(sbi)->dcc_info = dcc;
1728 init_thread:
1729         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1730                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1731         if (IS_ERR(dcc->f2fs_issue_discard)) {
1732                 err = PTR_ERR(dcc->f2fs_issue_discard);
1733                 kfree(dcc);
1734                 SM_I(sbi)->dcc_info = NULL;
1735                 return err;
1736         }
1737
1738         return err;
1739 }
1740
1741 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1742 {
1743         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1744
1745         if (!dcc)
1746                 return;
1747
1748         stop_discard_thread(sbi);
1749
1750         kfree(dcc);
1751         SM_I(sbi)->dcc_info = NULL;
1752 }
1753
1754 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1755 {
1756         struct sit_info *sit_i = SIT_I(sbi);
1757
1758         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1759                 sit_i->dirty_sentries++;
1760                 return false;
1761         }
1762
1763         return true;
1764 }
1765
1766 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1767                                         unsigned int segno, int modified)
1768 {
1769         struct seg_entry *se = get_seg_entry(sbi, segno);
1770         se->type = type;
1771         if (modified)
1772                 __mark_sit_entry_dirty(sbi, segno);
1773 }
1774
1775 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1776 {
1777         struct seg_entry *se;
1778         unsigned int segno, offset;
1779         long int new_vblocks;
1780         bool exist;
1781 #ifdef CONFIG_F2FS_CHECK_FS
1782         bool mir_exist;
1783 #endif
1784
1785         segno = GET_SEGNO(sbi, blkaddr);
1786
1787         se = get_seg_entry(sbi, segno);
1788         new_vblocks = se->valid_blocks + del;
1789         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1790
1791         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
1792                                 (new_vblocks > sbi->blocks_per_seg)));
1793
1794         se->valid_blocks = new_vblocks;
1795         se->mtime = get_mtime(sbi);
1796         SIT_I(sbi)->max_mtime = se->mtime;
1797
1798         /* Update valid block bitmap */
1799         if (del > 0) {
1800                 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
1801 #ifdef CONFIG_F2FS_CHECK_FS
1802                 mir_exist = f2fs_test_and_set_bit(offset,
1803                                                 se->cur_valid_map_mir);
1804                 if (unlikely(exist != mir_exist)) {
1805                         f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1806                                 "when setting bitmap, blk:%u, old bit:%d",
1807                                 blkaddr, exist);
1808                         f2fs_bug_on(sbi, 1);
1809                 }
1810 #endif
1811                 if (unlikely(exist)) {
1812                         f2fs_msg(sbi->sb, KERN_ERR,
1813                                 "Bitmap was wrongly set, blk:%u", blkaddr);
1814                         f2fs_bug_on(sbi, 1);
1815                         se->valid_blocks--;
1816                         del = 0;
1817                 }
1818
1819                 if (f2fs_discard_en(sbi) &&
1820                         !f2fs_test_and_set_bit(offset, se->discard_map))
1821                         sbi->discard_blks--;
1822
1823                 /* don't overwrite by SSR to keep node chain */
1824                 if (IS_NODESEG(se->type)) {
1825                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1826                                 se->ckpt_valid_blocks++;
1827                 }
1828         } else {
1829                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
1830 #ifdef CONFIG_F2FS_CHECK_FS
1831                 mir_exist = f2fs_test_and_clear_bit(offset,
1832                                                 se->cur_valid_map_mir);
1833                 if (unlikely(exist != mir_exist)) {
1834                         f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1835                                 "when clearing bitmap, blk:%u, old bit:%d",
1836                                 blkaddr, exist);
1837                         f2fs_bug_on(sbi, 1);
1838                 }
1839 #endif
1840                 if (unlikely(!exist)) {
1841                         f2fs_msg(sbi->sb, KERN_ERR,
1842                                 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1843                         f2fs_bug_on(sbi, 1);
1844                         se->valid_blocks++;
1845                         del = 0;
1846                 }
1847
1848                 if (f2fs_discard_en(sbi) &&
1849                         f2fs_test_and_clear_bit(offset, se->discard_map))
1850                         sbi->discard_blks++;
1851         }
1852         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1853                 se->ckpt_valid_blocks += del;
1854
1855         __mark_sit_entry_dirty(sbi, segno);
1856
1857         /* update total number of valid blocks to be written in ckpt area */
1858         SIT_I(sbi)->written_valid_blocks += del;
1859
1860         if (sbi->segs_per_sec > 1)
1861                 get_sec_entry(sbi, segno)->valid_blocks += del;
1862 }
1863
1864 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1865 {
1866         unsigned int segno = GET_SEGNO(sbi, addr);
1867         struct sit_info *sit_i = SIT_I(sbi);
1868
1869         f2fs_bug_on(sbi, addr == NULL_ADDR);
1870         if (addr == NEW_ADDR)
1871                 return;
1872
1873         /* add it into sit main buffer */
1874         down_write(&sit_i->sentry_lock);
1875
1876         update_sit_entry(sbi, addr, -1);
1877
1878         /* add it into dirty seglist */
1879         locate_dirty_segment(sbi, segno);
1880
1881         up_write(&sit_i->sentry_lock);
1882 }
1883
1884 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1885 {
1886         struct sit_info *sit_i = SIT_I(sbi);
1887         unsigned int segno, offset;
1888         struct seg_entry *se;
1889         bool is_cp = false;
1890
1891         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1892                 return true;
1893
1894         down_read(&sit_i->sentry_lock);
1895
1896         segno = GET_SEGNO(sbi, blkaddr);
1897         se = get_seg_entry(sbi, segno);
1898         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1899
1900         if (f2fs_test_bit(offset, se->ckpt_valid_map))
1901                 is_cp = true;
1902
1903         up_read(&sit_i->sentry_lock);
1904
1905         return is_cp;
1906 }
1907
1908 /*
1909  * This function should be resided under the curseg_mutex lock
1910  */
1911 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1912                                         struct f2fs_summary *sum)
1913 {
1914         struct curseg_info *curseg = CURSEG_I(sbi, type);
1915         void *addr = curseg->sum_blk;
1916         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1917         memcpy(addr, sum, sizeof(struct f2fs_summary));
1918 }
1919
1920 /*
1921  * Calculate the number of current summary pages for writing
1922  */
1923 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1924 {
1925         int valid_sum_count = 0;
1926         int i, sum_in_page;
1927
1928         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1929                 if (sbi->ckpt->alloc_type[i] == SSR)
1930                         valid_sum_count += sbi->blocks_per_seg;
1931                 else {
1932                         if (for_ra)
1933                                 valid_sum_count += le16_to_cpu(
1934                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1935                         else
1936                                 valid_sum_count += curseg_blkoff(sbi, i);
1937                 }
1938         }
1939
1940         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1941                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1942         if (valid_sum_count <= sum_in_page)
1943                 return 1;
1944         else if ((valid_sum_count - sum_in_page) <=
1945                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1946                 return 2;
1947         return 3;
1948 }
1949
1950 /*
1951  * Caller should put this summary page
1952  */
1953 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1954 {
1955         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1956 }
1957
1958 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1959 {
1960         struct page *page = grab_meta_page(sbi, blk_addr);
1961
1962         memcpy(page_address(page), src, PAGE_SIZE);
1963         set_page_dirty(page);
1964         f2fs_put_page(page, 1);
1965 }
1966
1967 static void write_sum_page(struct f2fs_sb_info *sbi,
1968                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
1969 {
1970         update_meta_page(sbi, (void *)sum_blk, blk_addr);
1971 }
1972
1973 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1974                                                 int type, block_t blk_addr)
1975 {
1976         struct curseg_info *curseg = CURSEG_I(sbi, type);
1977         struct page *page = grab_meta_page(sbi, blk_addr);
1978         struct f2fs_summary_block *src = curseg->sum_blk;
1979         struct f2fs_summary_block *dst;
1980
1981         dst = (struct f2fs_summary_block *)page_address(page);
1982         memset(dst, 0, PAGE_SIZE);
1983
1984         mutex_lock(&curseg->curseg_mutex);
1985
1986         down_read(&curseg->journal_rwsem);
1987         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1988         up_read(&curseg->journal_rwsem);
1989
1990         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1991         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1992
1993         mutex_unlock(&curseg->curseg_mutex);
1994
1995         set_page_dirty(page);
1996         f2fs_put_page(page, 1);
1997 }
1998
1999 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2000 {
2001         struct curseg_info *curseg = CURSEG_I(sbi, type);
2002         unsigned int segno = curseg->segno + 1;
2003         struct free_segmap_info *free_i = FREE_I(sbi);
2004
2005         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2006                 return !test_bit(segno, free_i->free_segmap);
2007         return 0;
2008 }
2009
2010 /*
2011  * Find a new segment from the free segments bitmap to right order
2012  * This function should be returned with success, otherwise BUG
2013  */
2014 static void get_new_segment(struct f2fs_sb_info *sbi,
2015                         unsigned int *newseg, bool new_sec, int dir)
2016 {
2017         struct free_segmap_info *free_i = FREE_I(sbi);
2018         unsigned int segno, secno, zoneno;
2019         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2020         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2021         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2022         unsigned int left_start = hint;
2023         bool init = true;
2024         int go_left = 0;
2025         int i;
2026
2027         spin_lock(&free_i->segmap_lock);
2028
2029         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2030                 segno = find_next_zero_bit(free_i->free_segmap,
2031                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2032                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2033                         goto got_it;
2034         }
2035 find_other_zone:
2036         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2037         if (secno >= MAIN_SECS(sbi)) {
2038                 if (dir == ALLOC_RIGHT) {
2039                         secno = find_next_zero_bit(free_i->free_secmap,
2040                                                         MAIN_SECS(sbi), 0);
2041                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2042                 } else {
2043                         go_left = 1;
2044                         left_start = hint - 1;
2045                 }
2046         }
2047         if (go_left == 0)
2048                 goto skip_left;
2049
2050         while (test_bit(left_start, free_i->free_secmap)) {
2051                 if (left_start > 0) {
2052                         left_start--;
2053                         continue;
2054                 }
2055                 left_start = find_next_zero_bit(free_i->free_secmap,
2056                                                         MAIN_SECS(sbi), 0);
2057                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2058                 break;
2059         }
2060         secno = left_start;
2061 skip_left:
2062         segno = GET_SEG_FROM_SEC(sbi, secno);
2063         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2064
2065         /* give up on finding another zone */
2066         if (!init)
2067                 goto got_it;
2068         if (sbi->secs_per_zone == 1)
2069                 goto got_it;
2070         if (zoneno == old_zoneno)
2071                 goto got_it;
2072         if (dir == ALLOC_LEFT) {
2073                 if (!go_left && zoneno + 1 >= total_zones)
2074                         goto got_it;
2075                 if (go_left && zoneno == 0)
2076                         goto got_it;
2077         }
2078         for (i = 0; i < NR_CURSEG_TYPE; i++)
2079                 if (CURSEG_I(sbi, i)->zone == zoneno)
2080                         break;
2081
2082         if (i < NR_CURSEG_TYPE) {
2083                 /* zone is in user, try another */
2084                 if (go_left)
2085                         hint = zoneno * sbi->secs_per_zone - 1;
2086                 else if (zoneno + 1 >= total_zones)
2087                         hint = 0;
2088                 else
2089                         hint = (zoneno + 1) * sbi->secs_per_zone;
2090                 init = false;
2091                 goto find_other_zone;
2092         }
2093 got_it:
2094         /* set it as dirty segment in free segmap */
2095         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2096         __set_inuse(sbi, segno);
2097         *newseg = segno;
2098         spin_unlock(&free_i->segmap_lock);
2099 }
2100
2101 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2102 {
2103         struct curseg_info *curseg = CURSEG_I(sbi, type);
2104         struct summary_footer *sum_footer;
2105
2106         curseg->segno = curseg->next_segno;
2107         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2108         curseg->next_blkoff = 0;
2109         curseg->next_segno = NULL_SEGNO;
2110
2111         sum_footer = &(curseg->sum_blk->footer);
2112         memset(sum_footer, 0, sizeof(struct summary_footer));
2113         if (IS_DATASEG(type))
2114                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2115         if (IS_NODESEG(type))
2116                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2117         __set_sit_entry_type(sbi, type, curseg->segno, modified);
2118 }
2119
2120 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2121 {
2122         /* if segs_per_sec is large than 1, we need to keep original policy. */
2123         if (sbi->segs_per_sec != 1)
2124                 return CURSEG_I(sbi, type)->segno;
2125
2126         if (test_opt(sbi, NOHEAP) &&
2127                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2128                 return 0;
2129
2130         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2131                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2132
2133         /* find segments from 0 to reuse freed segments */
2134         if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2135                 return 0;
2136
2137         return CURSEG_I(sbi, type)->segno;
2138 }
2139
2140 /*
2141  * Allocate a current working segment.
2142  * This function always allocates a free segment in LFS manner.
2143  */
2144 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2145 {
2146         struct curseg_info *curseg = CURSEG_I(sbi, type);
2147         unsigned int segno = curseg->segno;
2148         int dir = ALLOC_LEFT;
2149
2150         write_sum_page(sbi, curseg->sum_blk,
2151                                 GET_SUM_BLOCK(sbi, segno));
2152         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2153                 dir = ALLOC_RIGHT;
2154
2155         if (test_opt(sbi, NOHEAP))
2156                 dir = ALLOC_RIGHT;
2157
2158         segno = __get_next_segno(sbi, type);
2159         get_new_segment(sbi, &segno, new_sec, dir);
2160         curseg->next_segno = segno;
2161         reset_curseg(sbi, type, 1);
2162         curseg->alloc_type = LFS;
2163 }
2164
2165 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2166                         struct curseg_info *seg, block_t start)
2167 {
2168         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2169         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2170         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2171         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2172         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2173         int i, pos;
2174
2175         for (i = 0; i < entries; i++)
2176                 target_map[i] = ckpt_map[i] | cur_map[i];
2177
2178         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2179
2180         seg->next_blkoff = pos;
2181 }
2182
2183 /*
2184  * If a segment is written by LFS manner, next block offset is just obtained
2185  * by increasing the current block offset. However, if a segment is written by
2186  * SSR manner, next block offset obtained by calling __next_free_blkoff
2187  */
2188 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2189                                 struct curseg_info *seg)
2190 {
2191         if (seg->alloc_type == SSR)
2192                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2193         else
2194                 seg->next_blkoff++;
2195 }
2196
2197 /*
2198  * This function always allocates a used segment(from dirty seglist) by SSR
2199  * manner, so it should recover the existing segment information of valid blocks
2200  */
2201 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2202 {
2203         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2204         struct curseg_info *curseg = CURSEG_I(sbi, type);
2205         unsigned int new_segno = curseg->next_segno;
2206         struct f2fs_summary_block *sum_node;
2207         struct page *sum_page;
2208
2209         write_sum_page(sbi, curseg->sum_blk,
2210                                 GET_SUM_BLOCK(sbi, curseg->segno));
2211         __set_test_and_inuse(sbi, new_segno);
2212
2213         mutex_lock(&dirty_i->seglist_lock);
2214         __remove_dirty_segment(sbi, new_segno, PRE);
2215         __remove_dirty_segment(sbi, new_segno, DIRTY);
2216         mutex_unlock(&dirty_i->seglist_lock);
2217
2218         reset_curseg(sbi, type, 1);
2219         curseg->alloc_type = SSR;
2220         __next_free_blkoff(sbi, curseg, 0);
2221
2222         sum_page = get_sum_page(sbi, new_segno);
2223         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2224         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2225         f2fs_put_page(sum_page, 1);
2226 }
2227
2228 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2229 {
2230         struct curseg_info *curseg = CURSEG_I(sbi, type);
2231         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2232         unsigned segno = NULL_SEGNO;
2233         int i, cnt;
2234         bool reversed = false;
2235
2236         /* need_SSR() already forces to do this */
2237         if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2238                 curseg->next_segno = segno;
2239                 return 1;
2240         }
2241
2242         /* For node segments, let's do SSR more intensively */
2243         if (IS_NODESEG(type)) {
2244                 if (type >= CURSEG_WARM_NODE) {
2245                         reversed = true;
2246                         i = CURSEG_COLD_NODE;
2247                 } else {
2248                         i = CURSEG_HOT_NODE;
2249                 }
2250                 cnt = NR_CURSEG_NODE_TYPE;
2251         } else {
2252                 if (type >= CURSEG_WARM_DATA) {
2253                         reversed = true;
2254                         i = CURSEG_COLD_DATA;
2255                 } else {
2256                         i = CURSEG_HOT_DATA;
2257                 }
2258                 cnt = NR_CURSEG_DATA_TYPE;
2259         }
2260
2261         for (; cnt-- > 0; reversed ? i-- : i++) {
2262                 if (i == type)
2263                         continue;
2264                 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2265                         curseg->next_segno = segno;
2266                         return 1;
2267                 }
2268         }
2269         return 0;
2270 }
2271
2272 /*
2273  * flush out current segment and replace it with new segment
2274  * This function should be returned with success, otherwise BUG
2275  */
2276 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2277                                                 int type, bool force)
2278 {
2279         struct curseg_info *curseg = CURSEG_I(sbi, type);
2280
2281         if (force)
2282                 new_curseg(sbi, type, true);
2283         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2284                                         type == CURSEG_WARM_NODE)
2285                 new_curseg(sbi, type, false);
2286         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2287                 new_curseg(sbi, type, false);
2288         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
2289                 change_curseg(sbi, type);
2290         else
2291                 new_curseg(sbi, type, false);
2292
2293         stat_inc_seg_type(sbi, curseg);
2294 }
2295
2296 void allocate_new_segments(struct f2fs_sb_info *sbi)
2297 {
2298         struct curseg_info *curseg;
2299         unsigned int old_segno;
2300         int i;
2301
2302         down_write(&SIT_I(sbi)->sentry_lock);
2303
2304         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2305                 curseg = CURSEG_I(sbi, i);
2306                 old_segno = curseg->segno;
2307                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2308                 locate_dirty_segment(sbi, old_segno);
2309         }
2310
2311         up_write(&SIT_I(sbi)->sentry_lock);
2312 }
2313
2314 static const struct segment_allocation default_salloc_ops = {
2315         .allocate_segment = allocate_segment_by_default,
2316 };
2317
2318 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2319 {
2320         __u64 trim_start = cpc->trim_start;
2321         bool has_candidate = false;
2322
2323         down_write(&SIT_I(sbi)->sentry_lock);
2324         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2325                 if (add_discard_addrs(sbi, cpc, true)) {
2326                         has_candidate = true;
2327                         break;
2328                 }
2329         }
2330         up_write(&SIT_I(sbi)->sentry_lock);
2331
2332         cpc->trim_start = trim_start;
2333         return has_candidate;
2334 }
2335
2336 static void __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2337                                         struct discard_policy *dpolicy,
2338                                         unsigned int start, unsigned int end)
2339 {
2340         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2341         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2342         struct rb_node **insert_p = NULL, *insert_parent = NULL;
2343         struct discard_cmd *dc;
2344         struct blk_plug plug;
2345         int issued;
2346
2347 next:
2348         issued = 0;
2349
2350         mutex_lock(&dcc->cmd_lock);
2351         f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
2352
2353         dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
2354                                         NULL, start,
2355                                         (struct rb_entry **)&prev_dc,
2356                                         (struct rb_entry **)&next_dc,
2357                                         &insert_p, &insert_parent, true);
2358         if (!dc)
2359                 dc = next_dc;
2360
2361         blk_start_plug(&plug);
2362
2363         while (dc && dc->lstart <= end) {
2364                 struct rb_node *node;
2365
2366                 if (dc->len < dpolicy->granularity)
2367                         goto skip;
2368
2369                 if (dc->state != D_PREP) {
2370                         list_move_tail(&dc->list, &dcc->fstrim_list);
2371                         goto skip;
2372                 }
2373
2374                 __submit_discard_cmd(sbi, dpolicy, dc);
2375
2376                 if (++issued >= dpolicy->max_requests) {
2377                         start = dc->lstart + dc->len;
2378
2379                         blk_finish_plug(&plug);
2380                         mutex_unlock(&dcc->cmd_lock);
2381                         __wait_all_discard_cmd(sbi, NULL);
2382                         congestion_wait(BLK_RW_ASYNC, HZ/50);
2383                         goto next;
2384                 }
2385 skip:
2386                 node = rb_next(&dc->rb_node);
2387                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2388
2389                 if (fatal_signal_pending(current))
2390                         break;
2391         }
2392
2393         blk_finish_plug(&plug);
2394         mutex_unlock(&dcc->cmd_lock);
2395 }
2396
2397 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2398 {
2399         __u64 start = F2FS_BYTES_TO_BLK(range->start);
2400         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2401         unsigned int start_segno, end_segno;
2402         block_t start_block, end_block;
2403         struct cp_control cpc;
2404         struct discard_policy dpolicy;
2405         unsigned long long trimmed = 0;
2406         int err = 0;
2407
2408         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2409                 return -EINVAL;
2410
2411         if (end <= MAIN_BLKADDR(sbi))
2412                 return -EINVAL;
2413
2414         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2415                 f2fs_msg(sbi->sb, KERN_WARNING,
2416                         "Found FS corruption, run fsck to fix.");
2417                 return -EIO;
2418         }
2419
2420         /* start/end segment number in main_area */
2421         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2422         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2423                                                 GET_SEGNO(sbi, end);
2424
2425         cpc.reason = CP_DISCARD;
2426         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2427         cpc.trim_start = start_segno;
2428         cpc.trim_end = end_segno;
2429
2430         if (sbi->discard_blks == 0)
2431                 goto out;
2432
2433         mutex_lock(&sbi->gc_mutex);
2434         err = write_checkpoint(sbi, &cpc);
2435         mutex_unlock(&sbi->gc_mutex);
2436         if (err)
2437                 goto out;
2438
2439         start_block = START_BLOCK(sbi, start_segno);
2440         end_block = START_BLOCK(sbi, end_segno + 1);
2441
2442         __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2443         __issue_discard_cmd_range(sbi, &dpolicy, start_block, end_block);
2444
2445         /*
2446          * We filed discard candidates, but actually we don't need to wait for
2447          * all of them, since they'll be issued in idle time along with runtime
2448          * discard option. User configuration looks like using runtime discard
2449          * or periodic fstrim instead of it.
2450          */
2451         if (!test_opt(sbi, DISCARD)) {
2452                 trimmed = __wait_discard_cmd_range(sbi, &dpolicy,
2453                                         start_block, end_block);
2454                 range->len = F2FS_BLK_TO_BYTES(trimmed);
2455         }
2456 out:
2457         return err;
2458 }
2459
2460 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2461 {
2462         struct curseg_info *curseg = CURSEG_I(sbi, type);
2463         if (curseg->next_blkoff < sbi->blocks_per_seg)
2464                 return true;
2465         return false;
2466 }
2467
2468 int rw_hint_to_seg_type(enum rw_hint hint)
2469 {
2470         switch (hint) {
2471         case WRITE_LIFE_SHORT:
2472                 return CURSEG_HOT_DATA;
2473         case WRITE_LIFE_EXTREME:
2474                 return CURSEG_COLD_DATA;
2475         default:
2476                 return CURSEG_WARM_DATA;
2477         }
2478 }
2479
2480 /* This returns write hints for each segment type. This hints will be
2481  * passed down to block layer. There are mapping tables which depend on
2482  * the mount option 'whint_mode'.
2483  *
2484  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2485  *
2486  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2487  *
2488  * User                  F2FS                     Block
2489  * ----                  ----                     -----
2490  *                       META                     WRITE_LIFE_NOT_SET
2491  *                       HOT_NODE                 "
2492  *                       WARM_NODE                "
2493  *                       COLD_NODE                "
2494  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2495  * extension list        "                        "
2496  *
2497  * -- buffered io
2498  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2499  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2500  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2501  * WRITE_LIFE_NONE       "                        "
2502  * WRITE_LIFE_MEDIUM     "                        "
2503  * WRITE_LIFE_LONG       "                        "
2504  *
2505  * -- direct io
2506  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2507  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2508  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2509  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2510  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2511  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2512  *
2513  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2514  *
2515  * User                  F2FS                     Block
2516  * ----                  ----                     -----
2517  *                       META                     WRITE_LIFE_MEDIUM;
2518  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
2519  *                       WARM_NODE                "
2520  *                       COLD_NODE                WRITE_LIFE_NONE
2521  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2522  * extension list        "                        "
2523  *
2524  * -- buffered io
2525  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2526  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2527  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
2528  * WRITE_LIFE_NONE       "                        "
2529  * WRITE_LIFE_MEDIUM     "                        "
2530  * WRITE_LIFE_LONG       "                        "
2531  *
2532  * -- direct io
2533  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2534  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2535  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2536  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2537  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2538  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2539  */
2540
2541 enum rw_hint io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2542                                 enum page_type type, enum temp_type temp)
2543 {
2544         if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2545                 if (type == DATA) {
2546                         if (temp == WARM)
2547                                 return WRITE_LIFE_NOT_SET;
2548                         else if (temp == HOT)
2549                                 return WRITE_LIFE_SHORT;
2550                         else if (temp == COLD)
2551                                 return WRITE_LIFE_EXTREME;
2552                 } else {
2553                         return WRITE_LIFE_NOT_SET;
2554                 }
2555         } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2556                 if (type == DATA) {
2557                         if (temp == WARM)
2558                                 return WRITE_LIFE_LONG;
2559                         else if (temp == HOT)
2560                                 return WRITE_LIFE_SHORT;
2561                         else if (temp == COLD)
2562                                 return WRITE_LIFE_EXTREME;
2563                 } else if (type == NODE) {
2564                         if (temp == WARM || temp == HOT)
2565                                 return WRITE_LIFE_NOT_SET;
2566                         else if (temp == COLD)
2567                                 return WRITE_LIFE_NONE;
2568                 } else if (type == META) {
2569                         return WRITE_LIFE_MEDIUM;
2570                 }
2571         }
2572         return WRITE_LIFE_NOT_SET;
2573 }
2574
2575 static int __get_segment_type_2(struct f2fs_io_info *fio)
2576 {
2577         if (fio->type == DATA)
2578                 return CURSEG_HOT_DATA;
2579         else
2580                 return CURSEG_HOT_NODE;
2581 }
2582
2583 static int __get_segment_type_4(struct f2fs_io_info *fio)
2584 {
2585         if (fio->type == DATA) {
2586                 struct inode *inode = fio->page->mapping->host;
2587
2588                 if (S_ISDIR(inode->i_mode))
2589                         return CURSEG_HOT_DATA;
2590                 else
2591                         return CURSEG_COLD_DATA;
2592         } else {
2593                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2594                         return CURSEG_WARM_NODE;
2595                 else
2596                         return CURSEG_COLD_NODE;
2597         }
2598 }
2599
2600 static int __get_segment_type_6(struct f2fs_io_info *fio)
2601 {
2602         if (fio->type == DATA) {
2603                 struct inode *inode = fio->page->mapping->host;
2604
2605                 if (is_cold_data(fio->page) || file_is_cold(inode))
2606                         return CURSEG_COLD_DATA;
2607                 if (file_is_hot(inode) ||
2608                                 is_inode_flag_set(inode, FI_HOT_DATA))
2609                         return CURSEG_HOT_DATA;
2610                 return rw_hint_to_seg_type(inode->i_write_hint);
2611         } else {
2612                 if (IS_DNODE(fio->page))
2613                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2614                                                 CURSEG_HOT_NODE;
2615                 return CURSEG_COLD_NODE;
2616         }
2617 }
2618
2619 static int __get_segment_type(struct f2fs_io_info *fio)
2620 {
2621         int type = 0;
2622
2623         switch (F2FS_OPTION(fio->sbi).active_logs) {
2624         case 2:
2625                 type = __get_segment_type_2(fio);
2626                 break;
2627         case 4:
2628                 type = __get_segment_type_4(fio);
2629                 break;
2630         case 6:
2631                 type = __get_segment_type_6(fio);
2632                 break;
2633         default:
2634                 f2fs_bug_on(fio->sbi, true);
2635         }
2636
2637         if (IS_HOT(type))
2638                 fio->temp = HOT;
2639         else if (IS_WARM(type))
2640                 fio->temp = WARM;
2641         else
2642                 fio->temp = COLD;
2643         return type;
2644 }
2645
2646 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2647                 block_t old_blkaddr, block_t *new_blkaddr,
2648                 struct f2fs_summary *sum, int type,
2649                 struct f2fs_io_info *fio, bool add_list)
2650 {
2651         struct sit_info *sit_i = SIT_I(sbi);
2652         struct curseg_info *curseg = CURSEG_I(sbi, type);
2653
2654         down_read(&SM_I(sbi)->curseg_lock);
2655
2656         mutex_lock(&curseg->curseg_mutex);
2657         down_write(&sit_i->sentry_lock);
2658
2659         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2660
2661         f2fs_wait_discard_bio(sbi, *new_blkaddr);
2662
2663         /*
2664          * __add_sum_entry should be resided under the curseg_mutex
2665          * because, this function updates a summary entry in the
2666          * current summary block.
2667          */
2668         __add_sum_entry(sbi, type, sum);
2669
2670         __refresh_next_blkoff(sbi, curseg);
2671
2672         stat_inc_block_count(sbi, curseg);
2673
2674         /*
2675          * SIT information should be updated before segment allocation,
2676          * since SSR needs latest valid block information.
2677          */
2678         update_sit_entry(sbi, *new_blkaddr, 1);
2679         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2680                 update_sit_entry(sbi, old_blkaddr, -1);
2681
2682         if (!__has_curseg_space(sbi, type))
2683                 sit_i->s_ops->allocate_segment(sbi, type, false);
2684
2685         /*
2686          * segment dirty status should be updated after segment allocation,
2687          * so we just need to update status only one time after previous
2688          * segment being closed.
2689          */
2690         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2691         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
2692
2693         up_write(&sit_i->sentry_lock);
2694
2695         if (page && IS_NODESEG(type)) {
2696                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2697
2698                 f2fs_inode_chksum_set(sbi, page);
2699         }
2700
2701         if (add_list) {
2702                 struct f2fs_bio_info *io;
2703
2704                 INIT_LIST_HEAD(&fio->list);
2705                 fio->in_list = true;
2706                 io = sbi->write_io[fio->type] + fio->temp;
2707                 spin_lock(&io->io_lock);
2708                 list_add_tail(&fio->list, &io->io_list);
2709                 spin_unlock(&io->io_lock);
2710         }
2711
2712         mutex_unlock(&curseg->curseg_mutex);
2713
2714         up_read(&SM_I(sbi)->curseg_lock);
2715 }
2716
2717 static void update_device_state(struct f2fs_io_info *fio)
2718 {
2719         struct f2fs_sb_info *sbi = fio->sbi;
2720         unsigned int devidx;
2721
2722         if (!sbi->s_ndevs)
2723                 return;
2724
2725         devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2726
2727         /* update device state for fsync */
2728         set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
2729
2730         /* update device state for checkpoint */
2731         if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2732                 spin_lock(&sbi->dev_lock);
2733                 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2734                 spin_unlock(&sbi->dev_lock);
2735         }
2736 }
2737
2738 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2739 {
2740         int type = __get_segment_type(fio);
2741         int err;
2742
2743 reallocate:
2744         allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
2745                         &fio->new_blkaddr, sum, type, fio, true);
2746
2747         /* writeout dirty page into bdev */
2748         err = f2fs_submit_page_write(fio);
2749         if (err == -EAGAIN) {
2750                 fio->old_blkaddr = fio->new_blkaddr;
2751                 goto reallocate;
2752         } else if (!err) {
2753                 update_device_state(fio);
2754         }
2755 }
2756
2757 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2758                                         enum iostat_type io_type)
2759 {
2760         struct f2fs_io_info fio = {
2761                 .sbi = sbi,
2762                 .type = META,
2763                 .temp = HOT,
2764                 .op = REQ_OP_WRITE,
2765                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
2766                 .old_blkaddr = page->index,
2767                 .new_blkaddr = page->index,
2768                 .page = page,
2769                 .encrypted_page = NULL,
2770                 .in_list = false,
2771         };
2772
2773         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
2774                 fio.op_flags &= ~REQ_META;
2775
2776         set_page_writeback(page);
2777         ClearPageError(page);
2778         f2fs_submit_page_write(&fio);
2779
2780         f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
2781 }
2782
2783 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
2784 {
2785         struct f2fs_summary sum;
2786
2787         set_summary(&sum, nid, 0, 0);
2788         do_write_page(&sum, fio);
2789
2790         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2791 }
2792
2793 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
2794 {
2795         struct f2fs_sb_info *sbi = fio->sbi;
2796         struct f2fs_summary sum;
2797         struct node_info ni;
2798
2799         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
2800         get_node_info(sbi, dn->nid, &ni);
2801         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
2802         do_write_page(&sum, fio);
2803         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
2804
2805         f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
2806 }
2807
2808 int rewrite_data_page(struct f2fs_io_info *fio)
2809 {
2810         int err;
2811         struct f2fs_sb_info *sbi = fio->sbi;
2812
2813         fio->new_blkaddr = fio->old_blkaddr;
2814         /* i/o temperature is needed for passing down write hints */
2815         __get_segment_type(fio);
2816
2817         f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
2818                         GET_SEGNO(sbi, fio->new_blkaddr))->type));
2819
2820         stat_inc_inplace_blocks(fio->sbi);
2821
2822         err = f2fs_submit_page_bio(fio);
2823         if (!err)
2824                 update_device_state(fio);
2825
2826         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2827
2828         return err;
2829 }
2830
2831 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
2832                                                 unsigned int segno)
2833 {
2834         int i;
2835
2836         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
2837                 if (CURSEG_I(sbi, i)->segno == segno)
2838                         break;
2839         }
2840         return i;
2841 }
2842
2843 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
2844                                 block_t old_blkaddr, block_t new_blkaddr,
2845                                 bool recover_curseg, bool recover_newaddr)
2846 {
2847         struct sit_info *sit_i = SIT_I(sbi);
2848         struct curseg_info *curseg;
2849         unsigned int segno, old_cursegno;
2850         struct seg_entry *se;
2851         int type;
2852         unsigned short old_blkoff;
2853
2854         segno = GET_SEGNO(sbi, new_blkaddr);
2855         se = get_seg_entry(sbi, segno);
2856         type = se->type;
2857
2858         down_write(&SM_I(sbi)->curseg_lock);
2859
2860         if (!recover_curseg) {
2861                 /* for recovery flow */
2862                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2863                         if (old_blkaddr == NULL_ADDR)
2864                                 type = CURSEG_COLD_DATA;
2865                         else
2866                                 type = CURSEG_WARM_DATA;
2867                 }
2868         } else {
2869                 if (IS_CURSEG(sbi, segno)) {
2870                         /* se->type is volatile as SSR allocation */
2871                         type = __f2fs_get_curseg(sbi, segno);
2872                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
2873                 } else {
2874                         type = CURSEG_WARM_DATA;
2875                 }
2876         }
2877
2878         f2fs_bug_on(sbi, !IS_DATASEG(type));
2879         curseg = CURSEG_I(sbi, type);
2880
2881         mutex_lock(&curseg->curseg_mutex);
2882         down_write(&sit_i->sentry_lock);
2883
2884         old_cursegno = curseg->segno;
2885         old_blkoff = curseg->next_blkoff;
2886
2887         /* change the current segment */
2888         if (segno != curseg->segno) {
2889                 curseg->next_segno = segno;
2890                 change_curseg(sbi, type);
2891         }
2892
2893         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
2894         __add_sum_entry(sbi, type, sum);
2895
2896         if (!recover_curseg || recover_newaddr)
2897                 update_sit_entry(sbi, new_blkaddr, 1);
2898         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2899                 update_sit_entry(sbi, old_blkaddr, -1);
2900
2901         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2902         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2903
2904         locate_dirty_segment(sbi, old_cursegno);
2905
2906         if (recover_curseg) {
2907                 if (old_cursegno != curseg->segno) {
2908                         curseg->next_segno = old_cursegno;
2909                         change_curseg(sbi, type);
2910                 }
2911                 curseg->next_blkoff = old_blkoff;
2912         }
2913
2914         up_write(&sit_i->sentry_lock);
2915         mutex_unlock(&curseg->curseg_mutex);
2916         up_write(&SM_I(sbi)->curseg_lock);
2917 }
2918
2919 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2920                                 block_t old_addr, block_t new_addr,
2921                                 unsigned char version, bool recover_curseg,
2922                                 bool recover_newaddr)
2923 {
2924         struct f2fs_summary sum;
2925
2926         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2927
2928         __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2929                                         recover_curseg, recover_newaddr);
2930
2931         f2fs_update_data_blkaddr(dn, new_addr);
2932 }
2933
2934 void f2fs_wait_on_page_writeback(struct page *page,
2935                                 enum page_type type, bool ordered)
2936 {
2937         if (PageWriteback(page)) {
2938                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2939
2940                 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2941                                                 0, page->index, type);
2942                 if (ordered)
2943                         wait_on_page_writeback(page);
2944                 else
2945                         wait_for_stable_page(page);
2946         }
2947 }
2948
2949 void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
2950 {
2951         struct page *cpage;
2952
2953         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
2954                 return;
2955
2956         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2957         if (cpage) {
2958                 f2fs_wait_on_page_writeback(cpage, DATA, true);
2959                 f2fs_put_page(cpage, 1);
2960         }
2961 }
2962
2963 static void read_compacted_summaries(struct f2fs_sb_info *sbi)
2964 {
2965         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2966         struct curseg_info *seg_i;
2967         unsigned char *kaddr;
2968         struct page *page;
2969         block_t start;
2970         int i, j, offset;
2971
2972         start = start_sum_block(sbi);
2973
2974         page = get_meta_page(sbi, start++);
2975         kaddr = (unsigned char *)page_address(page);
2976
2977         /* Step 1: restore nat cache */
2978         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2979         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
2980
2981         /* Step 2: restore sit cache */
2982         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2983         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
2984         offset = 2 * SUM_JOURNAL_SIZE;
2985
2986         /* Step 3: restore summary entries */
2987         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2988                 unsigned short blk_off;
2989                 unsigned int segno;
2990
2991                 seg_i = CURSEG_I(sbi, i);
2992                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2993                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2994                 seg_i->next_segno = segno;
2995                 reset_curseg(sbi, i, 0);
2996                 seg_i->alloc_type = ckpt->alloc_type[i];
2997                 seg_i->next_blkoff = blk_off;
2998
2999                 if (seg_i->alloc_type == SSR)
3000                         blk_off = sbi->blocks_per_seg;
3001
3002                 for (j = 0; j < blk_off; j++) {
3003                         struct f2fs_summary *s;
3004                         s = (struct f2fs_summary *)(kaddr + offset);
3005                         seg_i->sum_blk->entries[j] = *s;
3006                         offset += SUMMARY_SIZE;
3007                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3008                                                 SUM_FOOTER_SIZE)
3009                                 continue;
3010
3011                         f2fs_put_page(page, 1);
3012                         page = NULL;
3013
3014                         page = get_meta_page(sbi, start++);
3015                         kaddr = (unsigned char *)page_address(page);
3016                         offset = 0;
3017                 }
3018         }
3019         f2fs_put_page(page, 1);
3020 }
3021
3022 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3023 {
3024         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3025         struct f2fs_summary_block *sum;
3026         struct curseg_info *curseg;
3027         struct page *new;
3028         unsigned short blk_off;
3029         unsigned int segno = 0;
3030         block_t blk_addr = 0;
3031
3032         /* get segment number and block addr */
3033         if (IS_DATASEG(type)) {
3034                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3035                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3036                                                         CURSEG_HOT_DATA]);
3037                 if (__exist_node_summaries(sbi))
3038                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3039                 else
3040                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3041         } else {
3042                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3043                                                         CURSEG_HOT_NODE]);
3044                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3045                                                         CURSEG_HOT_NODE]);
3046                 if (__exist_node_summaries(sbi))
3047                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3048                                                         type - CURSEG_HOT_NODE);
3049                 else
3050                         blk_addr = GET_SUM_BLOCK(sbi, segno);
3051         }
3052
3053         new = get_meta_page(sbi, blk_addr);
3054         sum = (struct f2fs_summary_block *)page_address(new);
3055
3056         if (IS_NODESEG(type)) {
3057                 if (__exist_node_summaries(sbi)) {
3058                         struct f2fs_summary *ns = &sum->entries[0];
3059                         int i;
3060                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3061                                 ns->version = 0;
3062                                 ns->ofs_in_node = 0;
3063                         }
3064                 } else {
3065                         restore_node_summary(sbi, segno, sum);
3066                 }
3067         }
3068
3069         /* set uncompleted segment to curseg */
3070         curseg = CURSEG_I(sbi, type);
3071         mutex_lock(&curseg->curseg_mutex);
3072
3073         /* update journal info */
3074         down_write(&curseg->journal_rwsem);
3075         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3076         up_write(&curseg->journal_rwsem);
3077
3078         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3079         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3080         curseg->next_segno = segno;
3081         reset_curseg(sbi, type, 0);
3082         curseg->alloc_type = ckpt->alloc_type[type];
3083         curseg->next_blkoff = blk_off;
3084         mutex_unlock(&curseg->curseg_mutex);
3085         f2fs_put_page(new, 1);
3086         return 0;
3087 }
3088
3089 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3090 {
3091         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3092         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3093         int type = CURSEG_HOT_DATA;
3094         int err;
3095
3096         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3097                 int npages = npages_for_summary_flush(sbi, true);
3098
3099                 if (npages >= 2)
3100                         ra_meta_pages(sbi, start_sum_block(sbi), npages,
3101                                                         META_CP, true);
3102
3103                 /* restore for compacted data summary */
3104                 read_compacted_summaries(sbi);
3105                 type = CURSEG_HOT_NODE;
3106         }
3107
3108         if (__exist_node_summaries(sbi))
3109                 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3110                                         NR_CURSEG_TYPE - type, META_CP, true);
3111
3112         for (; type <= CURSEG_COLD_NODE; type++) {
3113                 err = read_normal_summaries(sbi, type);
3114                 if (err)
3115                         return err;
3116         }
3117
3118         /* sanity check for summary blocks */
3119         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3120                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3121                 return -EINVAL;
3122
3123         return 0;
3124 }
3125
3126 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3127 {
3128         struct page *page;
3129         unsigned char *kaddr;
3130         struct f2fs_summary *summary;
3131         struct curseg_info *seg_i;
3132         int written_size = 0;
3133         int i, j;
3134
3135         page = grab_meta_page(sbi, blkaddr++);
3136         kaddr = (unsigned char *)page_address(page);
3137         memset(kaddr, 0, PAGE_SIZE);
3138
3139         /* Step 1: write nat cache */
3140         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3141         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3142         written_size += SUM_JOURNAL_SIZE;
3143
3144         /* Step 2: write sit cache */
3145         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3146         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3147         written_size += SUM_JOURNAL_SIZE;
3148
3149         /* Step 3: write summary entries */
3150         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3151                 unsigned short blkoff;
3152                 seg_i = CURSEG_I(sbi, i);
3153                 if (sbi->ckpt->alloc_type[i] == SSR)
3154                         blkoff = sbi->blocks_per_seg;
3155                 else
3156                         blkoff = curseg_blkoff(sbi, i);
3157
3158                 for (j = 0; j < blkoff; j++) {
3159                         if (!page) {
3160                                 page = grab_meta_page(sbi, blkaddr++);
3161                                 kaddr = (unsigned char *)page_address(page);
3162                                 memset(kaddr, 0, PAGE_SIZE);
3163                                 written_size = 0;
3164                         }
3165                         summary = (struct f2fs_summary *)(kaddr + written_size);
3166                         *summary = seg_i->sum_blk->entries[j];
3167                         written_size += SUMMARY_SIZE;
3168
3169                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3170                                                         SUM_FOOTER_SIZE)
3171                                 continue;
3172
3173                         set_page_dirty(page);
3174                         f2fs_put_page(page, 1);
3175                         page = NULL;
3176                 }
3177         }
3178         if (page) {
3179                 set_page_dirty(page);
3180                 f2fs_put_page(page, 1);
3181         }
3182 }
3183
3184 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3185                                         block_t blkaddr, int type)
3186 {
3187         int i, end;
3188         if (IS_DATASEG(type))
3189                 end = type + NR_CURSEG_DATA_TYPE;
3190         else
3191                 end = type + NR_CURSEG_NODE_TYPE;
3192
3193         for (i = type; i < end; i++)
3194                 write_current_sum_page(sbi, i, blkaddr + (i - type));
3195 }
3196
3197 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3198 {
3199         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3200                 write_compacted_summaries(sbi, start_blk);
3201         else
3202                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3203 }
3204
3205 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3206 {
3207         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3208 }
3209
3210 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3211                                         unsigned int val, int alloc)
3212 {
3213         int i;
3214
3215         if (type == NAT_JOURNAL) {
3216                 for (i = 0; i < nats_in_cursum(journal); i++) {
3217                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3218                                 return i;
3219                 }
3220                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3221                         return update_nats_in_cursum(journal, 1);
3222         } else if (type == SIT_JOURNAL) {
3223                 for (i = 0; i < sits_in_cursum(journal); i++)
3224                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3225                                 return i;
3226                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3227                         return update_sits_in_cursum(journal, 1);
3228         }
3229         return -1;
3230 }
3231
3232 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3233                                         unsigned int segno)
3234 {
3235         return get_meta_page(sbi, current_sit_addr(sbi, segno));
3236 }
3237
3238 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3239                                         unsigned int start)
3240 {
3241         struct sit_info *sit_i = SIT_I(sbi);
3242         struct page *page;
3243         pgoff_t src_off, dst_off;
3244
3245         src_off = current_sit_addr(sbi, start);
3246         dst_off = next_sit_addr(sbi, src_off);
3247
3248         page = grab_meta_page(sbi, dst_off);
3249         seg_info_to_sit_page(sbi, page, start);
3250
3251         set_page_dirty(page);
3252         set_to_next_sit(sit_i, start);
3253
3254         return page;
3255 }
3256
3257 static struct sit_entry_set *grab_sit_entry_set(void)
3258 {
3259         struct sit_entry_set *ses =
3260                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3261
3262         ses->entry_cnt = 0;
3263         INIT_LIST_HEAD(&ses->set_list);
3264         return ses;
3265 }
3266
3267 static void release_sit_entry_set(struct sit_entry_set *ses)
3268 {
3269         list_del(&ses->set_list);
3270         kmem_cache_free(sit_entry_set_slab, ses);
3271 }
3272
3273 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3274                                                 struct list_head *head)
3275 {
3276         struct sit_entry_set *next = ses;
3277
3278         if (list_is_last(&ses->set_list, head))
3279                 return;
3280
3281         list_for_each_entry_continue(next, head, set_list)
3282                 if (ses->entry_cnt <= next->entry_cnt)
3283                         break;
3284
3285         list_move_tail(&ses->set_list, &next->set_list);
3286 }
3287
3288 static void add_sit_entry(unsigned int segno, struct list_head *head)
3289 {
3290         struct sit_entry_set *ses;
3291         unsigned int start_segno = START_SEGNO(segno);
3292
3293         list_for_each_entry(ses, head, set_list) {
3294                 if (ses->start_segno == start_segno) {
3295                         ses->entry_cnt++;
3296                         adjust_sit_entry_set(ses, head);
3297                         return;
3298                 }
3299         }
3300
3301         ses = grab_sit_entry_set();
3302
3303         ses->start_segno = start_segno;
3304         ses->entry_cnt++;
3305         list_add(&ses->set_list, head);
3306 }
3307
3308 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3309 {
3310         struct f2fs_sm_info *sm_info = SM_I(sbi);
3311         struct list_head *set_list = &sm_info->sit_entry_set;
3312         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3313         unsigned int segno;
3314
3315         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3316                 add_sit_entry(segno, set_list);
3317 }
3318
3319 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3320 {
3321         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3322         struct f2fs_journal *journal = curseg->journal;
3323         int i;
3324
3325         down_write(&curseg->journal_rwsem);
3326         for (i = 0; i < sits_in_cursum(journal); i++) {
3327                 unsigned int segno;
3328                 bool dirtied;
3329
3330                 segno = le32_to_cpu(segno_in_journal(journal, i));
3331                 dirtied = __mark_sit_entry_dirty(sbi, segno);
3332
3333                 if (!dirtied)
3334                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3335         }
3336         update_sits_in_cursum(journal, -i);
3337         up_write(&curseg->journal_rwsem);
3338 }
3339
3340 /*
3341  * CP calls this function, which flushes SIT entries including sit_journal,
3342  * and moves prefree segs to free segs.
3343  */
3344 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3345 {
3346         struct sit_info *sit_i = SIT_I(sbi);
3347         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3348         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3349         struct f2fs_journal *journal = curseg->journal;
3350         struct sit_entry_set *ses, *tmp;
3351         struct list_head *head = &SM_I(sbi)->sit_entry_set;
3352         bool to_journal = true;
3353         struct seg_entry *se;
3354
3355         down_write(&sit_i->sentry_lock);
3356
3357         if (!sit_i->dirty_sentries)
3358                 goto out;
3359
3360         /*
3361          * add and account sit entries of dirty bitmap in sit entry
3362          * set temporarily
3363          */
3364         add_sits_in_set(sbi);
3365
3366         /*
3367          * if there are no enough space in journal to store dirty sit
3368          * entries, remove all entries from journal and add and account
3369          * them in sit entry set.
3370          */
3371         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3372                 remove_sits_in_journal(sbi);
3373
3374         /*
3375          * there are two steps to flush sit entries:
3376          * #1, flush sit entries to journal in current cold data summary block.
3377          * #2, flush sit entries to sit page.
3378          */
3379         list_for_each_entry_safe(ses, tmp, head, set_list) {
3380                 struct page *page = NULL;
3381                 struct f2fs_sit_block *raw_sit = NULL;
3382                 unsigned int start_segno = ses->start_segno;
3383                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3384                                                 (unsigned long)MAIN_SEGS(sbi));
3385                 unsigned int segno = start_segno;
3386
3387                 if (to_journal &&
3388                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3389                         to_journal = false;
3390
3391                 if (to_journal) {
3392                         down_write(&curseg->journal_rwsem);
3393                 } else {
3394                         page = get_next_sit_page(sbi, start_segno);
3395                         raw_sit = page_address(page);
3396                 }
3397
3398                 /* flush dirty sit entries in region of current sit set */
3399                 for_each_set_bit_from(segno, bitmap, end) {
3400                         int offset, sit_offset;
3401
3402                         se = get_seg_entry(sbi, segno);
3403 #ifdef CONFIG_F2FS_CHECK_FS
3404                         if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3405                                                 SIT_VBLOCK_MAP_SIZE))
3406                                 f2fs_bug_on(sbi, 1);
3407 #endif
3408
3409                         /* add discard candidates */
3410                         if (!(cpc->reason & CP_DISCARD)) {
3411                                 cpc->trim_start = segno;
3412                                 add_discard_addrs(sbi, cpc, false);
3413                         }
3414
3415                         if (to_journal) {
3416                                 offset = lookup_journal_in_cursum(journal,
3417                                                         SIT_JOURNAL, segno, 1);
3418                                 f2fs_bug_on(sbi, offset < 0);
3419                                 segno_in_journal(journal, offset) =
3420                                                         cpu_to_le32(segno);
3421                                 seg_info_to_raw_sit(se,
3422                                         &sit_in_journal(journal, offset));
3423                                 check_block_count(sbi, segno,
3424                                         &sit_in_journal(journal, offset));
3425                         } else {
3426                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3427                                 seg_info_to_raw_sit(se,
3428                                                 &raw_sit->entries[sit_offset]);
3429                                 check_block_count(sbi, segno,
3430                                                 &raw_sit->entries[sit_offset]);
3431                         }
3432
3433                         __clear_bit(segno, bitmap);
3434                         sit_i->dirty_sentries--;
3435                         ses->entry_cnt--;
3436                 }
3437
3438                 if (to_journal)
3439                         up_write(&curseg->journal_rwsem);
3440                 else
3441                         f2fs_put_page(page, 1);
3442
3443                 f2fs_bug_on(sbi, ses->entry_cnt);
3444                 release_sit_entry_set(ses);
3445         }
3446
3447         f2fs_bug_on(sbi, !list_empty(head));
3448         f2fs_bug_on(sbi, sit_i->dirty_sentries);
3449 out:
3450         if (cpc->reason & CP_DISCARD) {
3451                 __u64 trim_start = cpc->trim_start;
3452
3453                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3454                         add_discard_addrs(sbi, cpc, false);
3455
3456                 cpc->trim_start = trim_start;
3457         }
3458         up_write(&sit_i->sentry_lock);
3459
3460         set_prefree_as_free_segments(sbi);
3461 }
3462
3463 static int build_sit_info(struct f2fs_sb_info *sbi)
3464 {
3465         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3466         struct sit_info *sit_i;
3467         unsigned int sit_segs, start;
3468         char *src_bitmap;
3469         unsigned int bitmap_size;
3470
3471         /* allocate memory for SIT information */
3472         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3473         if (!sit_i)
3474                 return -ENOMEM;
3475
3476         SM_I(sbi)->sit_info = sit_i;
3477
3478         sit_i->sentries = f2fs_kvzalloc(sbi, MAIN_SEGS(sbi) *
3479                                         sizeof(struct seg_entry), GFP_KERNEL);
3480         if (!sit_i->sentries)
3481                 return -ENOMEM;
3482
3483         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3484         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3485                                                                 GFP_KERNEL);
3486         if (!sit_i->dirty_sentries_bitmap)
3487                 return -ENOMEM;
3488
3489         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3490                 sit_i->sentries[start].cur_valid_map
3491                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3492                 sit_i->sentries[start].ckpt_valid_map
3493                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3494                 if (!sit_i->sentries[start].cur_valid_map ||
3495                                 !sit_i->sentries[start].ckpt_valid_map)
3496                         return -ENOMEM;
3497
3498 #ifdef CONFIG_F2FS_CHECK_FS
3499                 sit_i->sentries[start].cur_valid_map_mir
3500                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3501                 if (!sit_i->sentries[start].cur_valid_map_mir)
3502                         return -ENOMEM;
3503 #endif
3504
3505                 if (f2fs_discard_en(sbi)) {
3506                         sit_i->sentries[start].discard_map
3507                                 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3508                                                                 GFP_KERNEL);
3509                         if (!sit_i->sentries[start].discard_map)
3510                                 return -ENOMEM;
3511                 }
3512         }
3513
3514         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3515         if (!sit_i->tmp_map)
3516                 return -ENOMEM;
3517
3518         if (sbi->segs_per_sec > 1) {
3519                 sit_i->sec_entries = f2fs_kvzalloc(sbi, MAIN_SECS(sbi) *
3520                                         sizeof(struct sec_entry), GFP_KERNEL);
3521                 if (!sit_i->sec_entries)
3522                         return -ENOMEM;
3523         }
3524
3525         /* get information related with SIT */
3526         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3527
3528         /* setup SIT bitmap from ckeckpoint pack */
3529         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3530         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3531
3532         sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3533         if (!sit_i->sit_bitmap)
3534                 return -ENOMEM;
3535
3536 #ifdef CONFIG_F2FS_CHECK_FS
3537         sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3538         if (!sit_i->sit_bitmap_mir)
3539                 return -ENOMEM;
3540 #endif
3541
3542         /* init SIT information */
3543         sit_i->s_ops = &default_salloc_ops;
3544
3545         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3546         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
3547         sit_i->written_valid_blocks = 0;
3548         sit_i->bitmap_size = bitmap_size;
3549         sit_i->dirty_sentries = 0;
3550         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3551         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3552         sit_i->mounted_time = ktime_get_real_seconds();
3553         init_rwsem(&sit_i->sentry_lock);
3554         return 0;
3555 }
3556
3557 static int build_free_segmap(struct f2fs_sb_info *sbi)
3558 {
3559         struct free_segmap_info *free_i;
3560         unsigned int bitmap_size, sec_bitmap_size;
3561
3562         /* allocate memory for free segmap information */
3563         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
3564         if (!free_i)
3565                 return -ENOMEM;
3566
3567         SM_I(sbi)->free_info = free_i;
3568
3569         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3570         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
3571         if (!free_i->free_segmap)
3572                 return -ENOMEM;
3573
3574         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3575         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
3576         if (!free_i->free_secmap)
3577                 return -ENOMEM;
3578
3579         /* set all segments as dirty temporarily */
3580         memset(free_i->free_segmap, 0xff, bitmap_size);
3581         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3582
3583         /* init free segmap information */
3584         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
3585         free_i->free_segments = 0;
3586         free_i->free_sections = 0;
3587         spin_lock_init(&free_i->segmap_lock);
3588         return 0;
3589 }
3590
3591 static int build_curseg(struct f2fs_sb_info *sbi)
3592 {
3593         struct curseg_info *array;
3594         int i;
3595
3596         array = f2fs_kzalloc(sbi, sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
3597         if (!array)
3598                 return -ENOMEM;
3599
3600         SM_I(sbi)->curseg_array = array;
3601
3602         for (i = 0; i < NR_CURSEG_TYPE; i++) {
3603                 mutex_init(&array[i].curseg_mutex);
3604                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
3605                 if (!array[i].sum_blk)
3606                         return -ENOMEM;
3607                 init_rwsem(&array[i].journal_rwsem);
3608                 array[i].journal = f2fs_kzalloc(sbi,
3609                                 sizeof(struct f2fs_journal), GFP_KERNEL);
3610                 if (!array[i].journal)
3611                         return -ENOMEM;
3612                 array[i].segno = NULL_SEGNO;
3613                 array[i].next_blkoff = 0;
3614         }
3615         return restore_curseg_summaries(sbi);
3616 }
3617
3618 static int build_sit_entries(struct f2fs_sb_info *sbi)
3619 {
3620         struct sit_info *sit_i = SIT_I(sbi);
3621         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3622         struct f2fs_journal *journal = curseg->journal;
3623         struct seg_entry *se;
3624         struct f2fs_sit_entry sit;
3625         int sit_blk_cnt = SIT_BLK_CNT(sbi);
3626         unsigned int i, start, end;
3627         unsigned int readed, start_blk = 0;
3628         int err = 0;
3629
3630         do {
3631                 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3632                                                         META_SIT, true);
3633
3634                 start = start_blk * sit_i->sents_per_block;
3635                 end = (start_blk + readed) * sit_i->sents_per_block;
3636
3637                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
3638                         struct f2fs_sit_block *sit_blk;
3639                         struct page *page;
3640
3641                         se = &sit_i->sentries[start];
3642                         page = get_current_sit_page(sbi, start);
3643                         sit_blk = (struct f2fs_sit_block *)page_address(page);
3644                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3645                         f2fs_put_page(page, 1);
3646
3647                         err = check_block_count(sbi, start, &sit);
3648                         if (err)
3649                                 return err;
3650                         seg_info_from_raw_sit(se, &sit);
3651
3652                         /* build discard map only one time */
3653                         if (f2fs_discard_en(sbi)) {
3654                                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3655                                         memset(se->discard_map, 0xff,
3656                                                 SIT_VBLOCK_MAP_SIZE);
3657                                 } else {
3658                                         memcpy(se->discard_map,
3659                                                 se->cur_valid_map,
3660                                                 SIT_VBLOCK_MAP_SIZE);
3661                                         sbi->discard_blks +=
3662                                                 sbi->blocks_per_seg -
3663                                                 se->valid_blocks;
3664                                 }
3665                         }
3666
3667                         if (sbi->segs_per_sec > 1)
3668                                 get_sec_entry(sbi, start)->valid_blocks +=
3669                                                         se->valid_blocks;
3670                 }
3671                 start_blk += readed;
3672         } while (start_blk < sit_blk_cnt);
3673
3674         down_read(&curseg->journal_rwsem);
3675         for (i = 0; i < sits_in_cursum(journal); i++) {
3676                 unsigned int old_valid_blocks;
3677
3678                 start = le32_to_cpu(segno_in_journal(journal, i));
3679                 se = &sit_i->sentries[start];
3680                 sit = sit_in_journal(journal, i);
3681
3682                 old_valid_blocks = se->valid_blocks;
3683
3684                 err = check_block_count(sbi, start, &sit);
3685                 if (err)
3686                         break;
3687                 seg_info_from_raw_sit(se, &sit);
3688
3689                 if (f2fs_discard_en(sbi)) {
3690                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3691                                 memset(se->discard_map, 0xff,
3692                                                         SIT_VBLOCK_MAP_SIZE);
3693                         } else {
3694                                 memcpy(se->discard_map, se->cur_valid_map,
3695                                                         SIT_VBLOCK_MAP_SIZE);
3696                                 sbi->discard_blks += old_valid_blocks -
3697                                                         se->valid_blocks;
3698                         }
3699                 }
3700
3701                 if (sbi->segs_per_sec > 1)
3702                         get_sec_entry(sbi, start)->valid_blocks +=
3703                                 se->valid_blocks - old_valid_blocks;
3704         }
3705         up_read(&curseg->journal_rwsem);
3706         return err;
3707 }
3708
3709 static void init_free_segmap(struct f2fs_sb_info *sbi)
3710 {
3711         unsigned int start;
3712         int type;
3713
3714         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3715                 struct seg_entry *sentry = get_seg_entry(sbi, start);
3716                 if (!sentry->valid_blocks)
3717                         __set_free(sbi, start);
3718                 else
3719                         SIT_I(sbi)->written_valid_blocks +=
3720                                                 sentry->valid_blocks;
3721         }
3722
3723         /* set use the current segments */
3724         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3725                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3726                 __set_test_and_inuse(sbi, curseg_t->segno);
3727         }
3728 }
3729
3730 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3731 {
3732         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3733         struct free_segmap_info *free_i = FREE_I(sbi);
3734         unsigned int segno = 0, offset = 0;
3735         unsigned short valid_blocks;
3736
3737         while (1) {
3738                 /* find dirty segment based on free segmap */
3739                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3740                 if (segno >= MAIN_SEGS(sbi))
3741                         break;
3742                 offset = segno + 1;
3743                 valid_blocks = get_valid_blocks(sbi, segno, false);
3744                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
3745                         continue;
3746                 if (valid_blocks > sbi->blocks_per_seg) {
3747                         f2fs_bug_on(sbi, 1);
3748                         continue;
3749                 }
3750                 mutex_lock(&dirty_i->seglist_lock);
3751                 __locate_dirty_segment(sbi, segno, DIRTY);
3752                 mutex_unlock(&dirty_i->seglist_lock);
3753         }
3754 }
3755
3756 static int init_victim_secmap(struct f2fs_sb_info *sbi)
3757 {
3758         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3759         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3760
3761         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
3762         if (!dirty_i->victim_secmap)
3763                 return -ENOMEM;
3764         return 0;
3765 }
3766
3767 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3768 {
3769         struct dirty_seglist_info *dirty_i;
3770         unsigned int bitmap_size, i;
3771
3772         /* allocate memory for dirty segments list information */
3773         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
3774                                                                 GFP_KERNEL);
3775         if (!dirty_i)
3776                 return -ENOMEM;
3777
3778         SM_I(sbi)->dirty_info = dirty_i;
3779         mutex_init(&dirty_i->seglist_lock);
3780
3781         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3782
3783         for (i = 0; i < NR_DIRTY_TYPE; i++) {
3784                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
3785                                                                 GFP_KERNEL);
3786                 if (!dirty_i->dirty_segmap[i])
3787                         return -ENOMEM;
3788         }
3789
3790         init_dirty_segmap(sbi);
3791         return init_victim_secmap(sbi);
3792 }
3793
3794 /*
3795  * Update min, max modified time for cost-benefit GC algorithm
3796  */
3797 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3798 {
3799         struct sit_info *sit_i = SIT_I(sbi);
3800         unsigned int segno;
3801
3802         down_write(&sit_i->sentry_lock);
3803
3804         sit_i->min_mtime = LLONG_MAX;
3805
3806         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
3807                 unsigned int i;
3808                 unsigned long long mtime = 0;
3809
3810                 for (i = 0; i < sbi->segs_per_sec; i++)
3811                         mtime += get_seg_entry(sbi, segno + i)->mtime;
3812
3813                 mtime = div_u64(mtime, sbi->segs_per_sec);
3814
3815                 if (sit_i->min_mtime > mtime)
3816                         sit_i->min_mtime = mtime;
3817         }
3818         sit_i->max_mtime = get_mtime(sbi);
3819         up_write(&sit_i->sentry_lock);
3820 }
3821
3822 int build_segment_manager(struct f2fs_sb_info *sbi)
3823 {
3824         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3825         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3826         struct f2fs_sm_info *sm_info;
3827         int err;
3828
3829         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
3830         if (!sm_info)
3831                 return -ENOMEM;
3832
3833         /* init sm info */
3834         sbi->sm_info = sm_info;
3835         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3836         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3837         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3838         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3839         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3840         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3841         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
3842         sm_info->rec_prefree_segments = sm_info->main_segments *
3843                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
3844         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3845                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3846
3847         if (!test_opt(sbi, LFS))
3848                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
3849         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
3850         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
3851         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
3852         sm_info->min_ssr_sections = reserved_sections(sbi);
3853
3854         INIT_LIST_HEAD(&sm_info->sit_entry_set);
3855
3856         init_rwsem(&sm_info->curseg_lock);
3857
3858         if (!f2fs_readonly(sbi->sb)) {
3859                 err = create_flush_cmd_control(sbi);
3860                 if (err)
3861                         return err;
3862         }
3863
3864         err = create_discard_cmd_control(sbi);
3865         if (err)
3866                 return err;
3867
3868         err = build_sit_info(sbi);
3869         if (err)
3870                 return err;
3871         err = build_free_segmap(sbi);
3872         if (err)
3873                 return err;
3874         err = build_curseg(sbi);
3875         if (err)
3876                 return err;
3877
3878         /* reinit free segmap based on SIT */
3879         err = build_sit_entries(sbi);
3880         if (err)
3881                 return err;
3882
3883         init_free_segmap(sbi);
3884         err = build_dirty_segmap(sbi);
3885         if (err)
3886                 return err;
3887
3888         init_min_max_mtime(sbi);
3889         return 0;
3890 }
3891
3892 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3893                 enum dirty_type dirty_type)
3894 {
3895         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3896
3897         mutex_lock(&dirty_i->seglist_lock);
3898         kvfree(dirty_i->dirty_segmap[dirty_type]);
3899         dirty_i->nr_dirty[dirty_type] = 0;
3900         mutex_unlock(&dirty_i->seglist_lock);
3901 }
3902
3903 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
3904 {
3905         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3906         kvfree(dirty_i->victim_secmap);
3907 }
3908
3909 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3910 {
3911         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3912         int i;
3913
3914         if (!dirty_i)
3915                 return;
3916
3917         /* discard pre-free/dirty segments list */
3918         for (i = 0; i < NR_DIRTY_TYPE; i++)
3919                 discard_dirty_segmap(sbi, i);
3920
3921         destroy_victim_secmap(sbi);
3922         SM_I(sbi)->dirty_info = NULL;
3923         kfree(dirty_i);
3924 }
3925
3926 static void destroy_curseg(struct f2fs_sb_info *sbi)
3927 {
3928         struct curseg_info *array = SM_I(sbi)->curseg_array;
3929         int i;
3930
3931         if (!array)
3932                 return;
3933         SM_I(sbi)->curseg_array = NULL;
3934         for (i = 0; i < NR_CURSEG_TYPE; i++) {
3935                 kfree(array[i].sum_blk);
3936                 kfree(array[i].journal);
3937         }
3938         kfree(array);
3939 }
3940
3941 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3942 {
3943         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3944         if (!free_i)
3945                 return;
3946         SM_I(sbi)->free_info = NULL;
3947         kvfree(free_i->free_segmap);
3948         kvfree(free_i->free_secmap);
3949         kfree(free_i);
3950 }
3951
3952 static void destroy_sit_info(struct f2fs_sb_info *sbi)
3953 {
3954         struct sit_info *sit_i = SIT_I(sbi);
3955         unsigned int start;
3956
3957         if (!sit_i)
3958                 return;
3959
3960         if (sit_i->sentries) {
3961                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3962                         kfree(sit_i->sentries[start].cur_valid_map);
3963 #ifdef CONFIG_F2FS_CHECK_FS
3964                         kfree(sit_i->sentries[start].cur_valid_map_mir);
3965 #endif
3966                         kfree(sit_i->sentries[start].ckpt_valid_map);
3967                         kfree(sit_i->sentries[start].discard_map);
3968                 }
3969         }
3970         kfree(sit_i->tmp_map);
3971
3972         kvfree(sit_i->sentries);
3973         kvfree(sit_i->sec_entries);
3974         kvfree(sit_i->dirty_sentries_bitmap);
3975
3976         SM_I(sbi)->sit_info = NULL;
3977         kfree(sit_i->sit_bitmap);
3978 #ifdef CONFIG_F2FS_CHECK_FS
3979         kfree(sit_i->sit_bitmap_mir);
3980 #endif
3981         kfree(sit_i);
3982 }
3983
3984 void destroy_segment_manager(struct f2fs_sb_info *sbi)
3985 {
3986         struct f2fs_sm_info *sm_info = SM_I(sbi);
3987
3988         if (!sm_info)
3989                 return;
3990         destroy_flush_cmd_control(sbi, true);
3991         destroy_discard_cmd_control(sbi);
3992         destroy_dirty_segmap(sbi);
3993         destroy_curseg(sbi);
3994         destroy_free_segmap(sbi);
3995         destroy_sit_info(sbi);
3996         sbi->sm_info = NULL;
3997         kfree(sm_info);
3998 }
3999
4000 int __init create_segment_manager_caches(void)
4001 {
4002         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4003                         sizeof(struct discard_entry));
4004         if (!discard_entry_slab)
4005                 goto fail;
4006
4007         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4008                         sizeof(struct discard_cmd));
4009         if (!discard_cmd_slab)
4010                 goto destroy_discard_entry;
4011
4012         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4013                         sizeof(struct sit_entry_set));
4014         if (!sit_entry_set_slab)
4015                 goto destroy_discard_cmd;
4016
4017         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4018                         sizeof(struct inmem_pages));
4019         if (!inmem_entry_slab)
4020                 goto destroy_sit_entry_set;
4021         return 0;
4022
4023 destroy_sit_entry_set:
4024         kmem_cache_destroy(sit_entry_set_slab);
4025 destroy_discard_cmd:
4026         kmem_cache_destroy(discard_cmd_slab);
4027 destroy_discard_entry:
4028         kmem_cache_destroy(discard_entry_slab);
4029 fail:
4030         return -ENOMEM;
4031 }
4032
4033 void destroy_segment_manager_caches(void)
4034 {
4035         kmem_cache_destroy(sit_entry_set_slab);
4036         kmem_cache_destroy(discard_cmd_slab);
4037         kmem_cache_destroy(discard_entry_slab);
4038         kmem_cache_destroy(inmem_entry_slab);
4039 }