fs/f2fs/segment.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/f2fs/segment.c
   4  *
   5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6  *             http://www.samsung.com/
   7  */
   8 #include <linux/fs.h>
   9 #include <linux/f2fs_fs.h>
  10 #include <linux/bio.h>
  11 #include <linux/blkdev.h>
  12 #include <linux/prefetch.h>
  13 #include <linux/kthread.h>
  14 #include <linux/swap.h>
  15 #include <linux/timer.h>
  16 #include <linux/freezer.h>
  17 #include <linux/sched/signal.h>
  18
  19 #include "f2fs.h"
  20 #include "segment.h"
  21 #include "node.h"
  22 #include "gc.h"
  23 #include "trace.h"
  24 #include <trace/events/f2fs.h>
  25
  26 #define __reverse_ffz(x) __reverse_ffs(~(x))
  27
  28 static struct kmem_cache *discard_entry_slab;
  29 static struct kmem_cache *discard_cmd_slab;
  30 static struct kmem_cache *sit_entry_set_slab;
  31 static struct kmem_cache *inmem_entry_slab;
  32
  33 static unsigned long __reverse_ulong(unsigned char *str)
  34 {
  35         unsigned long tmp = 0;
  36         int shift = 24, idx = 0;
  37
  38 #if BITS_PER_LONG == 64
  39         shift = 56;
  40 #endif
  41         while (shift >= 0) {
  42                 tmp |= (unsigned long)str[idx++] << shift;
  43                 shift -= BITS_PER_BYTE;
  44         }
  45         return tmp;
  46 }
  47
  48 /*
  49  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  50  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  51  */
  52 static inline unsigned long __reverse_ffs(unsigned long word)
  53 {
  54         int num = 0;
  55
  56 #if BITS_PER_LONG == 64
  57         if ((word & 0xffffffff00000000UL) == 0)
  58                 num += 32;
  59         else
  60                 word >>= 32;
  61 #endif
  62         if ((word & 0xffff0000) == 0)
  63                 num += 16;
  64         else
  65                 word >>= 16;
  66
  67         if ((word & 0xff00) == 0)
  68                 num += 8;
  69         else
  70                 word >>= 8;
  71
  72         if ((word & 0xf0) == 0)
  73                 num += 4;
  74         else
  75                 word >>= 4;
  76
  77         if ((word & 0xc) == 0)
  78                 num += 2;
  79         else
  80                 word >>= 2;
  81
  82         if ((word & 0x2) == 0)
  83                 num += 1;
  84         return num;
  85 }
  86
  87 /*
  88  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  89  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  90  * @size must be integral times of unsigned long.
  91  * Example:
  92  *                             MSB <--> LSB
  93  *   f2fs_set_bit(0, bitmap) => 1000 0000
  94  *   f2fs_set_bit(7, bitmap) => 0000 0001
  95  */
  96 static unsigned long __find_rev_next_bit(const unsigned long *addr,
  97                         unsigned long size, unsigned long offset)
  98 {
  99         const unsigned long *p = addr + BIT_WORD(offset);
 100         unsigned long result = size;
 101         unsigned long tmp;
 102
 103         if (offset >= size)
 104                 return size;
 105
 106         size -= (offset & ~(BITS_PER_LONG - 1));
 107         offset %= BITS_PER_LONG;
 108
 109         while (1) {
 110                 if (*p == 0)
 111                         goto pass;
 112
 113                 tmp = __reverse_ulong((unsigned char *)p);
 114
 115                 tmp &= ~0UL >> offset;
 116                 if (size < BITS_PER_LONG)
 117                         tmp &= (~0UL << (BITS_PER_LONG - size));
 118                 if (tmp)
 119                         goto found;
 120 pass:
 121                 if (size <= BITS_PER_LONG)
 122                         break;
 123                 size -= BITS_PER_LONG;
 124                 offset = 0;
 125                 p++;
 126         }
 127         return result;
 128 found:
 129         return result - size + __reverse_ffs(tmp);
 130 }
 131
 132 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 133                         unsigned long size, unsigned long offset)
 134 {
 135         const unsigned long *p = addr + BIT_WORD(offset);
 136         unsigned long result = size;
 137         unsigned long tmp;
 138
 139         if (offset >= size)
 140                 return size;
 141
 142         size -= (offset & ~(BITS_PER_LONG - 1));
 143         offset %= BITS_PER_LONG;
 144
 145         while (1) {
 146                 if (*p == ~0UL)
 147                         goto pass;
 148
 149                 tmp = __reverse_ulong((unsigned char *)p);
 150
 151                 if (offset)
 152                         tmp |= ~0UL << (BITS_PER_LONG - offset);
 153                 if (size < BITS_PER_LONG)
 154                         tmp |= ~0UL >> size;
 155                 if (tmp != ~0UL)
 156                         goto found;
 157 pass:
 158                 if (size <= BITS_PER_LONG)
 159                         break;
 160                 size -= BITS_PER_LONG;
 161                 offset = 0;
 162                 p++;
 163         }
 164         return result;
 165 found:
 166         return result - size + __reverse_ffz(tmp);
 167 }
 168
 169 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
 170 {
 171         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 172         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 173         int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
 174
 175         if (test_opt(sbi, LFS))
 176                 return false;
 177         if (sbi->gc_mode == GC_URGENT)
 178                 return true;
 179         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
 180                 return true;
 181
 182         return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
 183                         SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
 184 }
 185
 186 void f2fs_register_inmem_page(struct inode *inode, struct page *page)
 187 {
 188         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 189         struct f2fs_inode_info *fi = F2FS_I(inode);
 190         struct inmem_pages *new;
 191
 192         f2fs_trace_pid(page);
 193
 194         set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
 195         SetPagePrivate(page);
 196
 197         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 198
 199         /* add atomic page indices to the list */
 200         new->page = page;
 201         INIT_LIST_HEAD(&new->list);
 202
 203         /* increase reference count with clean state */
 204         mutex_lock(&fi->inmem_lock);
 205         get_page(page);
 206         list_add_tail(&new->list, &fi->inmem_pages);
 207         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 208         if (list_empty(&fi->inmem_ilist))
 209                 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
 210         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 211         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 212         mutex_unlock(&fi->inmem_lock);
 213
 214         trace_f2fs_register_inmem_page(page, INMEM);
 215 }
 216
 217 static int __revoke_inmem_pages(struct inode *inode,
 218                                 struct list_head *head, bool drop, bool recover)
 219 {
 220         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 221         struct inmem_pages *cur, *tmp;
 222         int err = 0;
 223
 224         list_for_each_entry_safe(cur, tmp, head, list) {
 225                 struct page *page = cur->page;
 226
 227                 if (drop)
 228                         trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 229
 230                 lock_page(page);
 231
 232                 f2fs_wait_on_page_writeback(page, DATA, true, true);
 233
 234                 if (recover) {
 235                         struct dnode_of_data dn;
 236                         struct node_info ni;
 237
 238                         trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 239 retry:
 240                         set_new_dnode(&dn, inode, NULL, NULL, 0);
 241                         err = f2fs_get_dnode_of_data(&dn, page->index,
 242                                                                 LOOKUP_NODE);
 243                         if (err) {
 244                                 if (err == -ENOMEM) {
 245                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 246                                         cond_resched();
 247                                         goto retry;
 248                                 }
 249                                 err = -EAGAIN;
 250                                 goto next;
 251                         }
 252
 253                         err = f2fs_get_node_info(sbi, dn.nid, &ni);
 254                         if (err) {
 255                                 f2fs_put_dnode(&dn);
 256                                 return err;
 257                         }
 258
 259                         if (cur->old_addr == NEW_ADDR) {
 260                                 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
 261                                 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
 262                         } else
 263                                 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 264                                         cur->old_addr, ni.version, true, true);
 265                         f2fs_put_dnode(&dn);
 266                 }
 267 next:
 268                 /* we don't need to invalidate this in the sccessful status */
 269                 if (drop || recover) {
 270                         ClearPageUptodate(page);
 271                         clear_cold_data(page);
 272                 }
 273                 set_page_private(page, 0);
 274                 ClearPagePrivate(page);
 275                 f2fs_put_page(page, 1);
 276
 277                 list_del(&cur->list);
 278                 kmem_cache_free(inmem_entry_slab, cur);
 279                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 280         }
 281         return err;
 282 }
 283
 284 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
 285 {
 286         struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
 287         struct inode *inode;
 288         struct f2fs_inode_info *fi;
 289 next:
 290         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 291         if (list_empty(head)) {
 292                 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 293                 return;
 294         }
 295         fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
 296         inode = igrab(&fi->vfs_inode);
 297         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 298
 299         if (inode) {
 300                 if (gc_failure) {
 301                         if (fi->i_gc_failures[GC_FAILURE_ATOMIC])
 302                                 goto drop;
 303                         goto skip;
 304                 }
 305 drop:
 306                 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
 307                 f2fs_drop_inmem_pages(inode);
 308                 iput(inode);
 309         }
 310 skip:
 311         congestion_wait(BLK_RW_ASYNC, HZ/50);
 312         cond_resched();
 313         goto next;
 314 }
 315
 316 void f2fs_drop_inmem_pages(struct inode *inode)
 317 {
 318         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 319         struct f2fs_inode_info *fi = F2FS_I(inode);
 320
 321         mutex_lock(&fi->inmem_lock);
 322         __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 323         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 324         if (!list_empty(&fi->inmem_ilist))
 325                 list_del_init(&fi->inmem_ilist);
 326         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 327         mutex_unlock(&fi->inmem_lock);
 328
 329         clear_inode_flag(inode, FI_ATOMIC_FILE);
 330         fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
 331         stat_dec_atomic_write(inode);
 332 }
 333
 334 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
 335 {
 336         struct f2fs_inode_info *fi = F2FS_I(inode);
 337         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 338         struct list_head *head = &fi->inmem_pages;
 339         struct inmem_pages *cur = NULL;
 340
 341         f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
 342
 343         mutex_lock(&fi->inmem_lock);
 344         list_for_each_entry(cur, head, list) {
 345                 if (cur->page == page)
 346                         break;
 347         }
 348
 349         f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
 350         list_del(&cur->list);
 351         mutex_unlock(&fi->inmem_lock);
 352
 353         dec_page_count(sbi, F2FS_INMEM_PAGES);
 354         kmem_cache_free(inmem_entry_slab, cur);
 355
 356         ClearPageUptodate(page);
 357         set_page_private(page, 0);
 358         ClearPagePrivate(page);
 359         f2fs_put_page(page, 0);
 360
 361         trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 362 }
 363
 364 static int __f2fs_commit_inmem_pages(struct inode *inode)
 365 {
 366         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 367         struct f2fs_inode_info *fi = F2FS_I(inode);
 368         struct inmem_pages *cur, *tmp;
 369         struct f2fs_io_info fio = {
 370                 .sbi = sbi,
 371                 .ino = inode->i_ino,
 372                 .type = DATA,
 373                 .op = REQ_OP_WRITE,
 374                 .op_flags = REQ_SYNC | REQ_PRIO,
 375                 .io_type = FS_DATA_IO,
 376         };
 377         struct list_head revoke_list;
 378         bool submit_bio = false;
 379         int err = 0;
 380
 381         INIT_LIST_HEAD(&revoke_list);
 382
 383         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 384                 struct page *page = cur->page;
 385
 386                 lock_page(page);
 387                 if (page->mapping == inode->i_mapping) {
 388                         trace_f2fs_commit_inmem_page(page, INMEM);
 389
 390                         f2fs_wait_on_page_writeback(page, DATA, true, true);
 391
 392                         set_page_dirty(page);
 393                         if (clear_page_dirty_for_io(page)) {
 394                                 inode_dec_dirty_pages(inode);
 395                                 f2fs_remove_dirty_inode(inode);
 396                         }
 397 retry:
 398                         fio.page = page;
 399                         fio.old_blkaddr = NULL_ADDR;
 400                         fio.encrypted_page = NULL;
 401                         fio.need_lock = LOCK_DONE;
 402                         err = f2fs_do_write_data_page(&fio);
 403                         if (err) {
 404                                 if (err == -ENOMEM) {
 405                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 406                                         cond_resched();
 407                                         goto retry;
 408                                 }
 409                                 unlock_page(page);
 410                                 break;
 411                         }
 412                         /* record old blkaddr for revoking */
 413                         cur->old_addr = fio.old_blkaddr;
 414                         submit_bio = true;
 415                 }
 416                 unlock_page(page);
 417                 list_move_tail(&cur->list, &revoke_list);
 418         }
 419
 420         if (submit_bio)
 421                 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
 422
 423         if (err) {
 424                 /*
 425                  * try to revoke all committed pages, but still we could fail
 426                  * due to no memory or other reason, if that happened, EAGAIN
 427                  * will be returned, which means in such case, transaction is
 428                  * already not integrity, caller should use journal to do the
 429                  * recovery or rewrite & commit last transaction. For other
 430                  * error number, revoking was done by filesystem itself.
 431                  */
 432                 err = __revoke_inmem_pages(inode, &revoke_list, false, true);
 433
 434                 /* drop all uncommitted pages */
 435                 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 436         } else {
 437                 __revoke_inmem_pages(inode, &revoke_list, false, false);
 438         }
 439
 440         return err;
 441 }
 442
 443 int f2fs_commit_inmem_pages(struct inode *inode)
 444 {
 445         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 446         struct f2fs_inode_info *fi = F2FS_I(inode);
 447         int err;
 448
 449         f2fs_balance_fs(sbi, true);
 450
 451         down_write(&fi->i_gc_rwsem[WRITE]);
 452
 453         f2fs_lock_op(sbi);
 454         set_inode_flag(inode, FI_ATOMIC_COMMIT);
 455
 456         mutex_lock(&fi->inmem_lock);
 457         err = __f2fs_commit_inmem_pages(inode);
 458
 459         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 460         if (!list_empty(&fi->inmem_ilist))
 461                 list_del_init(&fi->inmem_ilist);
 462         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 463         mutex_unlock(&fi->inmem_lock);
 464
 465         clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 466
 467         f2fs_unlock_op(sbi);
 468         up_write(&fi->i_gc_rwsem[WRITE]);
 469
 470         return err;
 471 }
 472
 473 /*
 474  * This function balances dirty node and dentry pages.
 475  * In addition, it controls garbage collection.
 476  */
 477 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 478 {
 479         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 480                 f2fs_show_injection_info(FAULT_CHECKPOINT);
 481                 f2fs_stop_checkpoint(sbi, false);
 482         }
 483
 484         /* balance_fs_bg is able to be pending */
 485         if (need && excess_cached_nats(sbi))
 486                 f2fs_balance_fs_bg(sbi);
 487
 488         if (f2fs_is_checkpoint_ready(sbi))
 489                 return;
 490
 491         /*
 492          * We should do GC or end up with checkpoint, if there are so many dirty
 493          * dir/node pages without enough free segments.
 494          */
 495         if (has_not_enough_free_secs(sbi, 0, 0)) {
 496                 mutex_lock(&sbi->gc_mutex);
 497                 f2fs_gc(sbi, false, false, NULL_SEGNO);
 498         }
 499 }
 500
 501 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 502 {
 503         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 504                 return;
 505
 506         /* try to shrink extent cache when there is no enough memory */
 507         if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
 508                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 509
 510         /* check the # of cached NAT entries */
 511         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
 512                 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 513
 514         if (!f2fs_available_free_memory(sbi, FREE_NIDS))
 515                 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
 516         else
 517                 f2fs_build_free_nids(sbi, false, false);
 518
 519         if (!is_idle(sbi, REQ_TIME) &&
 520                 (!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
 521                 return;
 522
 523         /* checkpoint is the only way to shrink partial cached entries */
 524         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
 525                         !f2fs_available_free_memory(sbi, INO_ENTRIES) ||
 526                         excess_prefree_segs(sbi) ||
 527                         excess_dirty_nats(sbi) ||
 528                         excess_dirty_nodes(sbi) ||
 529                         f2fs_time_over(sbi, CP_TIME)) {
 530                 if (test_opt(sbi, DATA_FLUSH)) {
 531                         struct blk_plug plug;
 532
 533                         blk_start_plug(&plug);
 534                         f2fs_sync_dirty_inodes(sbi, FILE_INODE);
 535                         blk_finish_plug(&plug);
 536                 }
 537                 f2fs_sync_fs(sbi->sb, true);
 538                 stat_inc_bg_cp_count(sbi->stat_info);
 539         }
 540 }
 541
 542 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 543                                 struct block_device *bdev)
 544 {
 545         struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
 546         int ret;
 547
 548         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
 549         bio_set_dev(bio, bdev);
 550         ret = submit_bio_wait(bio);
 551         bio_put(bio);
 552
 553         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 554                                 test_opt(sbi, FLUSH_MERGE), ret);
 555         return ret;
 556 }
 557
 558 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 559 {
 560         int ret = 0;
 561         int i;
 562
 563         if (!sbi->s_ndevs)
 564                 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 565
 566         for (i = 0; i < sbi->s_ndevs; i++) {
 567                 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
 568                         continue;
 569                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 570                 if (ret)
 571                         break;
 572         }
 573         return ret;
 574 }
 575
 576 static int issue_flush_thread(void *data)
 577 {
 578         struct f2fs_sb_info *sbi = data;
 579         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 580         wait_queue_head_t *q = &fcc->flush_wait_queue;
 581 repeat:
 582         if (kthread_should_stop())
 583                 return 0;
 584
 585         sb_start_intwrite(sbi->sb);
 586
 587         if (!llist_empty(&fcc->issue_list)) {
 588                 struct flush_cmd *cmd, *next;
 589                 int ret;
 590
 591                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 592                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 593
 594                 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 595
 596                 ret = submit_flush_wait(sbi, cmd->ino);
 597                 atomic_inc(&fcc->issued_flush);
 598
 599                 llist_for_each_entry_safe(cmd, next,
 600                                           fcc->dispatch_list, llnode) {
 601                         cmd->ret = ret;
 602                         complete(&cmd->wait);
 603                 }
 604                 fcc->dispatch_list = NULL;
 605         }
 606
 607         sb_end_intwrite(sbi->sb);
 608
 609         wait_event_interruptible(*q,
 610                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 611         goto repeat;
 612 }
 613
 614 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 615 {
 616         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 617         struct flush_cmd cmd;
 618         int ret;
 619
 620         if (test_opt(sbi, NOBARRIER))
 621                 return 0;
 622
 623         if (!test_opt(sbi, FLUSH_MERGE)) {
 624                 atomic_inc(&fcc->queued_flush);
 625                 ret = submit_flush_wait(sbi, ino);
 626                 atomic_dec(&fcc->queued_flush);
 627                 atomic_inc(&fcc->issued_flush);
 628                 return ret;
 629         }
 630
 631         if (atomic_inc_return(&fcc->queued_flush) == 1 || sbi->s_ndevs > 1) {
 632                 ret = submit_flush_wait(sbi, ino);
 633                 atomic_dec(&fcc->queued_flush);
 634
 635                 atomic_inc(&fcc->issued_flush);
 636                 return ret;
 637         }
 638
 639         cmd.ino = ino;
 640         init_completion(&cmd.wait);
 641
 642         llist_add(&cmd.llnode, &fcc->issue_list);
 643
 644         /* update issue_list before we wake up issue_flush thread */
 645         smp_mb();
 646
 647         if (waitqueue_active(&fcc->flush_wait_queue))
 648                 wake_up(&fcc->flush_wait_queue);
 649
 650         if (fcc->f2fs_issue_flush) {
 651                 wait_for_completion(&cmd.wait);
 652                 atomic_dec(&fcc->queued_flush);
 653         } else {
 654                 struct llist_node *list;
 655
 656                 list = llist_del_all(&fcc->issue_list);
 657                 if (!list) {
 658                         wait_for_completion(&cmd.wait);
 659                         atomic_dec(&fcc->queued_flush);
 660                 } else {
 661                         struct flush_cmd *tmp, *next;
 662
 663                         ret = submit_flush_wait(sbi, ino);
 664
 665                         llist_for_each_entry_safe(tmp, next, list, llnode) {
 666                                 if (tmp == &cmd) {
 667                                         cmd.ret = ret;
 668                                         atomic_dec(&fcc->queued_flush);
 669                                         continue;
 670                                 }
 671                                 tmp->ret = ret;
 672                                 complete(&tmp->wait);
 673                         }
 674                 }
 675         }
 676
 677         return cmd.ret;
 678 }
 679
 680 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
 681 {
 682         dev_t dev = sbi->sb->s_bdev->bd_dev;
 683         struct flush_cmd_control *fcc;
 684         int err = 0;
 685
 686         if (SM_I(sbi)->fcc_info) {
 687                 fcc = SM_I(sbi)->fcc_info;
 688                 if (fcc->f2fs_issue_flush)
 689                         return err;
 690                 goto init_thread;
 691         }
 692
 693         fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
 694         if (!fcc)
 695                 return -ENOMEM;
 696         atomic_set(&fcc->issued_flush, 0);
 697         atomic_set(&fcc->queued_flush, 0);
 698         init_waitqueue_head(&fcc->flush_wait_queue);
 699         init_llist_head(&fcc->issue_list);
 700         SM_I(sbi)->fcc_info = fcc;
 701         if (!test_opt(sbi, FLUSH_MERGE))
 702                 return err;
 703
 704 init_thread:
 705         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 706                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 707         if (IS_ERR(fcc->f2fs_issue_flush)) {
 708                 err = PTR_ERR(fcc->f2fs_issue_flush);
 709                 kvfree(fcc);
 710                 SM_I(sbi)->fcc_info = NULL;
 711                 return err;
 712         }
 713
 714         return err;
 715 }
 716
 717 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 718 {
 719         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 720
 721         if (fcc && fcc->f2fs_issue_flush) {
 722                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 723
 724                 fcc->f2fs_issue_flush = NULL;
 725                 kthread_stop(flush_thread);
 726         }
 727         if (free) {
 728                 kvfree(fcc);
 729                 SM_I(sbi)->fcc_info = NULL;
 730         }
 731 }
 732
 733 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
 734 {
 735         int ret = 0, i;
 736
 737         if (!sbi->s_ndevs)
 738                 return 0;
 739
 740         for (i = 1; i < sbi->s_ndevs; i++) {
 741                 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
 742                         continue;
 743                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 744                 if (ret)
 745                         break;
 746
 747                 spin_lock(&sbi->dev_lock);
 748                 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
 749                 spin_unlock(&sbi->dev_lock);
 750         }
 751
 752         return ret;
 753 }
 754
 755 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 756                 enum dirty_type dirty_type)
 757 {
 758         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 759
 760         /* need not be added */
 761         if (IS_CURSEG(sbi, segno))
 762                 return;
 763
 764         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 765                 dirty_i->nr_dirty[dirty_type]++;
 766
 767         if (dirty_type == DIRTY) {
 768                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 769                 enum dirty_type t = sentry->type;
 770
 771                 if (unlikely(t >= DIRTY)) {
 772                         f2fs_bug_on(sbi, 1);
 773                         return;
 774                 }
 775                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 776                         dirty_i->nr_dirty[t]++;
 777         }
 778 }
 779
 780 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 781                 enum dirty_type dirty_type)
 782 {
 783         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 784
 785         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 786                 dirty_i->nr_dirty[dirty_type]--;
 787
 788         if (dirty_type == DIRTY) {
 789                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 790                 enum dirty_type t = sentry->type;
 791
 792                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 793                         dirty_i->nr_dirty[t]--;
 794
 795                 if (get_valid_blocks(sbi, segno, true) == 0)
 796                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 797                                                 dirty_i->victim_secmap);
 798         }
 799 }
 800
 801 /*
 802  * Should not occur error such as -ENOMEM.
 803  * Adding dirty entry into seglist is not critical operation.
 804  * If a given segment is one of current working segments, it won't be added.
 805  */
 806 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 807 {
 808         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 809         unsigned short valid_blocks, ckpt_valid_blocks;
 810
 811         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 812                 return;
 813
 814         mutex_lock(&dirty_i->seglist_lock);
 815
 816         valid_blocks = get_valid_blocks(sbi, segno, false);
 817         ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno);
 818
 819         if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
 820                                 ckpt_valid_blocks == sbi->blocks_per_seg)) {
 821                 __locate_dirty_segment(sbi, segno, PRE);
 822                 __remove_dirty_segment(sbi, segno, DIRTY);
 823         } else if (valid_blocks < sbi->blocks_per_seg) {
 824                 __locate_dirty_segment(sbi, segno, DIRTY);
 825         } else {
 826                 /* Recovery routine with SSR needs this */
 827                 __remove_dirty_segment(sbi, segno, DIRTY);
 828         }
 829
 830         mutex_unlock(&dirty_i->seglist_lock);
 831 }
 832
 833 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
 834 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
 835 {
 836         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 837         unsigned int segno;
 838
 839         mutex_lock(&dirty_i->seglist_lock);
 840         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 841                 if (get_valid_blocks(sbi, segno, false))
 842                         continue;
 843                 if (IS_CURSEG(sbi, segno))
 844                         continue;
 845                 __locate_dirty_segment(sbi, segno, PRE);
 846                 __remove_dirty_segment(sbi, segno, DIRTY);
 847         }
 848         mutex_unlock(&dirty_i->seglist_lock);
 849 }
 850
 851 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi)
 852 {
 853         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 854         block_t ovp = overprovision_segments(sbi) << sbi->log_blocks_per_seg;
 855         block_t holes[2] = {0, 0};      /* DATA and NODE */
 856         struct seg_entry *se;
 857         unsigned int segno;
 858
 859         mutex_lock(&dirty_i->seglist_lock);
 860         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 861                 se = get_seg_entry(sbi, segno);
 862                 if (IS_NODESEG(se->type))
 863                         holes[NODE] += sbi->blocks_per_seg - se->valid_blocks;
 864                 else
 865                         holes[DATA] += sbi->blocks_per_seg - se->valid_blocks;
 866         }
 867         mutex_unlock(&dirty_i->seglist_lock);
 868
 869         if (holes[DATA] > ovp || holes[NODE] > ovp)
 870                 return -EAGAIN;
 871         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
 872                 dirty_segments(sbi) > overprovision_segments(sbi))
 873                 return -EAGAIN;
 874         return 0;
 875 }
 876
 877 /* This is only used by SBI_CP_DISABLED */
 878 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
 879 {
 880         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 881         unsigned int segno = 0;
 882
 883         mutex_lock(&dirty_i->seglist_lock);
 884         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 885                 if (get_valid_blocks(sbi, segno, false))
 886                         continue;
 887                 if (get_ckpt_valid_blocks(sbi, segno))
 888                         continue;
 889                 mutex_unlock(&dirty_i->seglist_lock);
 890                 return segno;
 891         }
 892         mutex_unlock(&dirty_i->seglist_lock);
 893         return NULL_SEGNO;
 894 }
 895
 896 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 897                 struct block_device *bdev, block_t lstart,
 898                 block_t start, block_t len)
 899 {
 900         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 901         struct list_head *pend_list;
 902         struct discard_cmd *dc;
 903
 904         f2fs_bug_on(sbi, !len);
 905
 906         pend_list = &dcc->pend_list[plist_idx(len)];
 907
 908         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 909         INIT_LIST_HEAD(&dc->list);
 910         dc->bdev = bdev;
 911         dc->lstart = lstart;
 912         dc->start = start;
 913         dc->len = len;
 914         dc->ref = 0;
 915         dc->state = D_PREP;
 916         dc->queued = 0;
 917         dc->error = 0;
 918         init_completion(&dc->wait);
 919         list_add_tail(&dc->list, pend_list);
 920         spin_lock_init(&dc->lock);
 921         dc->bio_ref = 0;
 922         atomic_inc(&dcc->discard_cmd_cnt);
 923         dcc->undiscard_blks += len;
 924
 925         return dc;
 926 }
 927
 928 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 929                                 struct block_device *bdev, block_t lstart,
 930                                 block_t start, block_t len,
 931                                 struct rb_node *parent, struct rb_node **p,
 932                                 bool leftmost)
 933 {
 934         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 935         struct discard_cmd *dc;
 936
 937         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
 938
 939         rb_link_node(&dc->rb_node, parent, p);
 940         rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
 941
 942         return dc;
 943 }
 944
 945 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
 946                                                         struct discard_cmd *dc)
 947 {
 948         if (dc->state == D_DONE)
 949                 atomic_sub(dc->queued, &dcc->queued_discard);
 950
 951         list_del(&dc->list);
 952         rb_erase_cached(&dc->rb_node, &dcc->root);
 953         dcc->undiscard_blks -= dc->len;
 954
 955         kmem_cache_free(discard_cmd_slab, dc);
 956
 957         atomic_dec(&dcc->discard_cmd_cnt);
 958 }
 959
 960 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
 961                                                         struct discard_cmd *dc)
 962 {
 963         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 964         unsigned long flags;
 965
 966         trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
 967
 968         spin_lock_irqsave(&dc->lock, flags);
 969         if (dc->bio_ref) {
 970                 spin_unlock_irqrestore(&dc->lock, flags);
 971                 return;
 972         }
 973         spin_unlock_irqrestore(&dc->lock, flags);
 974
 975         f2fs_bug_on(sbi, dc->ref);
 976
 977         if (dc->error == -EOPNOTSUPP)
 978                 dc->error = 0;
 979
 980         if (dc->error)
 981                 printk_ratelimited(
 982                         "%sF2FS-fs: Issue discard(%u, %u, %u) failed, ret: %d",
 983                         KERN_INFO, dc->lstart, dc->start, dc->len, dc->error);
 984         __detach_discard_cmd(dcc, dc);
 985 }
 986
 987 static void f2fs_submit_discard_endio(struct bio *bio)
 988 {
 989         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
 990         unsigned long flags;
 991
 992         dc->error = blk_status_to_errno(bio->bi_status);
 993
 994         spin_lock_irqsave(&dc->lock, flags);
 995         dc->bio_ref--;
 996         if (!dc->bio_ref && dc->state == D_SUBMIT) {
 997                 dc->state = D_DONE;
 998                 complete_all(&dc->wait);
 999         }
1000         spin_unlock_irqrestore(&dc->lock, flags);
1001         bio_put(bio);
1002 }
1003
1004 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1005                                 block_t start, block_t end)
1006 {
1007 #ifdef CONFIG_F2FS_CHECK_FS
1008         struct seg_entry *sentry;
1009         unsigned int segno;
1010         block_t blk = start;
1011         unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1012         unsigned long *map;
1013
1014         while (blk < end) {
1015                 segno = GET_SEGNO(sbi, blk);
1016                 sentry = get_seg_entry(sbi, segno);
1017                 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1018
1019                 if (end < START_BLOCK(sbi, segno + 1))
1020                         size = GET_BLKOFF_FROM_SEG0(sbi, end);
1021                 else
1022                         size = max_blocks;
1023                 map = (unsigned long *)(sentry->cur_valid_map);
1024                 offset = __find_rev_next_bit(map, size, offset);
1025                 f2fs_bug_on(sbi, offset != size);
1026                 blk = START_BLOCK(sbi, segno + 1);
1027         }
1028 #endif
1029 }
1030
1031 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1032                                 struct discard_policy *dpolicy,
1033                                 int discard_type, unsigned int granularity)
1034 {
1035         /* common policy */
1036         dpolicy->type = discard_type;
1037         dpolicy->sync = true;
1038         dpolicy->ordered = false;
1039         dpolicy->granularity = granularity;
1040
1041         dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1042         dpolicy->io_aware_gran = MAX_PLIST_NUM;
1043         dpolicy->timeout = 0;
1044
1045         if (discard_type == DPOLICY_BG) {
1046                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1047                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1048                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1049                 dpolicy->io_aware = true;
1050                 dpolicy->sync = false;
1051                 dpolicy->ordered = true;
1052                 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1053                         dpolicy->granularity = 1;
1054                         dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1055                 }
1056         } else if (discard_type == DPOLICY_FORCE) {
1057                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1058                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1059                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1060                 dpolicy->io_aware = false;
1061         } else if (discard_type == DPOLICY_FSTRIM) {
1062                 dpolicy->io_aware = false;
1063         } else if (discard_type == DPOLICY_UMOUNT) {
1064                 dpolicy->max_requests = UINT_MAX;
1065                 dpolicy->io_aware = false;
1066         }
1067 }
1068
1069 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1070                                 struct block_device *bdev, block_t lstart,
1071                                 block_t start, block_t len);
1072 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1073 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1074                                                 struct discard_policy *dpolicy,
1075                                                 struct discard_cmd *dc,
1076                                                 unsigned int *issued)
1077 {
1078         struct block_device *bdev = dc->bdev;
1079         struct request_queue *q = bdev_get_queue(bdev);
1080         unsigned int max_discard_blocks =
1081                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1082         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1083         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1084                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1085         int flag = dpolicy->sync ? REQ_SYNC : 0;
1086         block_t lstart, start, len, total_len;
1087         int err = 0;
1088
1089         if (dc->state != D_PREP)
1090                 return 0;
1091
1092         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1093                 return 0;
1094
1095         trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1096
1097         lstart = dc->lstart;
1098         start = dc->start;
1099         len = dc->len;
1100         total_len = len;
1101
1102         dc->len = 0;
1103
1104         while (total_len && *issued < dpolicy->max_requests && !err) {
1105                 struct bio *bio = NULL;
1106                 unsigned long flags;
1107                 bool last = true;
1108
1109                 if (len > max_discard_blocks) {
1110                         len = max_discard_blocks;
1111                         last = false;
1112                 }
1113
1114                 (*issued)++;
1115                 if (*issued == dpolicy->max_requests)
1116                         last = true;
1117
1118                 dc->len += len;
1119
1120                 if (time_to_inject(sbi, FAULT_DISCARD)) {
1121                         f2fs_show_injection_info(FAULT_DISCARD);
1122                         err = -EIO;
1123                         goto submit;
1124                 }
1125                 err = __blkdev_issue_discard(bdev,
1126                                         SECTOR_FROM_BLOCK(start),
1127                                         SECTOR_FROM_BLOCK(len),
1128                                         GFP_NOFS, 0, &bio);
1129 submit:
1130                 if (err) {
1131                         spin_lock_irqsave(&dc->lock, flags);
1132                         if (dc->state == D_PARTIAL)
1133                                 dc->state = D_SUBMIT;
1134                         spin_unlock_irqrestore(&dc->lock, flags);
1135
1136                         break;
1137                 }
1138
1139                 f2fs_bug_on(sbi, !bio);
1140
1141                 /*
1142                  * should keep before submission to avoid D_DONE
1143                  * right away
1144                  */
1145                 spin_lock_irqsave(&dc->lock, flags);
1146                 if (last)
1147                         dc->state = D_SUBMIT;
1148                 else
1149                         dc->state = D_PARTIAL;
1150                 dc->bio_ref++;
1151                 spin_unlock_irqrestore(&dc->lock, flags);
1152
1153                 atomic_inc(&dcc->queued_discard);
1154                 dc->queued++;
1155                 list_move_tail(&dc->list, wait_list);
1156
1157                 /* sanity check on discard range */
1158                 __check_sit_bitmap(sbi, lstart, lstart + len);
1159
1160                 bio->bi_private = dc;
1161                 bio->bi_end_io = f2fs_submit_discard_endio;
1162                 bio->bi_opf |= flag;
1163                 submit_bio(bio);
1164
1165                 atomic_inc(&dcc->issued_discard);
1166
1167                 f2fs_update_iostat(sbi, FS_DISCARD, 1);
1168
1169                 lstart += len;
1170                 start += len;
1171                 total_len -= len;
1172                 len = total_len;
1173         }
1174
1175         if (!err && len)
1176                 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1177         return err;
1178 }
1179
1180 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1181                                 struct block_device *bdev, block_t lstart,
1182                                 block_t start, block_t len,
1183                                 struct rb_node **insert_p,
1184                                 struct rb_node *insert_parent)
1185 {
1186         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1187         struct rb_node **p;
1188         struct rb_node *parent = NULL;
1189         struct discard_cmd *dc = NULL;
1190         bool leftmost = true;
1191
1192         if (insert_p && insert_parent) {
1193                 parent = insert_parent;
1194                 p = insert_p;
1195                 goto do_insert;
1196         }
1197
1198         p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1199                                                         lstart, &leftmost);
1200 do_insert:
1201         dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1202                                                                 p, leftmost);
1203         if (!dc)
1204                 return NULL;
1205
1206         return dc;
1207 }
1208
1209 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1210                                                 struct discard_cmd *dc)
1211 {
1212         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1213 }
1214
1215 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1216                                 struct discard_cmd *dc, block_t blkaddr)
1217 {
1218         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1219         struct discard_info di = dc->di;
1220         bool modified = false;
1221
1222         if (dc->state == D_DONE || dc->len == 1) {
1223                 __remove_discard_cmd(sbi, dc);
1224                 return;
1225         }
1226
1227         dcc->undiscard_blks -= di.len;
1228
1229         if (blkaddr > di.lstart) {
1230                 dc->len = blkaddr - dc->lstart;
1231                 dcc->undiscard_blks += dc->len;
1232                 __relocate_discard_cmd(dcc, dc);
1233                 modified = true;
1234         }
1235
1236         if (blkaddr < di.lstart + di.len - 1) {
1237                 if (modified) {
1238                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1239                                         di.start + blkaddr + 1 - di.lstart,
1240                                         di.lstart + di.len - 1 - blkaddr,
1241                                         NULL, NULL);
1242                 } else {
1243                         dc->lstart++;
1244                         dc->len--;
1245                         dc->start++;
1246                         dcc->undiscard_blks += dc->len;
1247                         __relocate_discard_cmd(dcc, dc);
1248                 }
1249         }
1250 }
1251
1252 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1253                                 struct block_device *bdev, block_t lstart,
1254                                 block_t start, block_t len)
1255 {
1256         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1257         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1258         struct discard_cmd *dc;
1259         struct discard_info di = {0};
1260         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1261         struct request_queue *q = bdev_get_queue(bdev);
1262         unsigned int max_discard_blocks =
1263                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1264         block_t end = lstart + len;
1265
1266         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1267                                         NULL, lstart,
1268                                         (struct rb_entry **)&prev_dc,
1269                                         (struct rb_entry **)&next_dc,
1270                                         &insert_p, &insert_parent, true, NULL);
1271         if (dc)
1272                 prev_dc = dc;
1273
1274         if (!prev_dc) {
1275                 di.lstart = lstart;
1276                 di.len = next_dc ? next_dc->lstart - lstart : len;
1277                 di.len = min(di.len, len);
1278                 di.start = start;
1279         }
1280
1281         while (1) {
1282                 struct rb_node *node;
1283                 bool merged = false;
1284                 struct discard_cmd *tdc = NULL;
1285
1286                 if (prev_dc) {
1287                         di.lstart = prev_dc->lstart + prev_dc->len;
1288                         if (di.lstart < lstart)
1289                                 di.lstart = lstart;
1290                         if (di.lstart >= end)
1291                                 break;
1292
1293                         if (!next_dc || next_dc->lstart > end)
1294                                 di.len = end - di.lstart;
1295                         else
1296                                 di.len = next_dc->lstart - di.lstart;
1297                         di.start = start + di.lstart - lstart;
1298                 }
1299
1300                 if (!di.len)
1301                         goto next;
1302
1303                 if (prev_dc && prev_dc->state == D_PREP &&
1304                         prev_dc->bdev == bdev &&
1305                         __is_discard_back_mergeable(&di, &prev_dc->di,
1306                                                         max_discard_blocks)) {
1307                         prev_dc->di.len += di.len;
1308                         dcc->undiscard_blks += di.len;
1309                         __relocate_discard_cmd(dcc, prev_dc);
1310                         di = prev_dc->di;
1311                         tdc = prev_dc;
1312                         merged = true;
1313                 }
1314
1315                 if (next_dc && next_dc->state == D_PREP &&
1316                         next_dc->bdev == bdev &&
1317                         __is_discard_front_mergeable(&di, &next_dc->di,
1318                                                         max_discard_blocks)) {
1319                         next_dc->di.lstart = di.lstart;
1320                         next_dc->di.len += di.len;
1321                         next_dc->di.start = di.start;
1322                         dcc->undiscard_blks += di.len;
1323                         __relocate_discard_cmd(dcc, next_dc);
1324                         if (tdc)
1325                                 __remove_discard_cmd(sbi, tdc);
1326                         merged = true;
1327                 }
1328
1329                 if (!merged) {
1330                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1331                                                         di.len, NULL, NULL);
1332                 }
1333  next:
1334                 prev_dc = next_dc;
1335                 if (!prev_dc)
1336                         break;
1337
1338                 node = rb_next(&prev_dc->rb_node);
1339                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1340         }
1341 }
1342
1343 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1344                 struct block_device *bdev, block_t blkstart, block_t blklen)
1345 {
1346         block_t lblkstart = blkstart;
1347
1348         trace_f2fs_queue_discard(bdev, blkstart, blklen);
1349
1350         if (sbi->s_ndevs) {
1351                 int devi = f2fs_target_device_index(sbi, blkstart);
1352
1353                 blkstart -= FDEV(devi).start_blk;
1354         }
1355         mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1356         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1357         mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1358         return 0;
1359 }
1360
1361 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1362                                         struct discard_policy *dpolicy)
1363 {
1364         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1365         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1366         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1367         struct discard_cmd *dc;
1368         struct blk_plug plug;
1369         unsigned int pos = dcc->next_pos;
1370         unsigned int issued = 0;
1371         bool io_interrupted = false;
1372
1373         mutex_lock(&dcc->cmd_lock);
1374         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1375                                         NULL, pos,
1376                                         (struct rb_entry **)&prev_dc,
1377                                         (struct rb_entry **)&next_dc,
1378                                         &insert_p, &insert_parent, true, NULL);
1379         if (!dc)
1380                 dc = next_dc;
1381
1382         blk_start_plug(&plug);
1383
1384         while (dc) {
1385                 struct rb_node *node;
1386                 int err = 0;
1387
1388                 if (dc->state != D_PREP)
1389                         goto next;
1390
1391                 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1392                         io_interrupted = true;
1393                         break;
1394                 }
1395
1396                 dcc->next_pos = dc->lstart + dc->len;
1397                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1398
1399                 if (issued >= dpolicy->max_requests)
1400                         break;
1401 next:
1402                 node = rb_next(&dc->rb_node);
1403                 if (err)
1404                         __remove_discard_cmd(sbi, dc);
1405                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1406         }
1407
1408         blk_finish_plug(&plug);
1409
1410         if (!dc)
1411                 dcc->next_pos = 0;
1412
1413         mutex_unlock(&dcc->cmd_lock);
1414
1415         if (!issued && io_interrupted)
1416                 issued = -1;
1417
1418         return issued;
1419 }
1420
1421 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1422                                         struct discard_policy *dpolicy)
1423 {
1424         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1425         struct list_head *pend_list;
1426         struct discard_cmd *dc, *tmp;
1427         struct blk_plug plug;
1428         int i, issued = 0;
1429         bool io_interrupted = false;
1430
1431         if (dpolicy->timeout != 0)
1432                 f2fs_update_time(sbi, dpolicy->timeout);
1433
1434         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1435                 if (dpolicy->timeout != 0 &&
1436                                 f2fs_time_over(sbi, dpolicy->timeout))
1437                         break;
1438
1439                 if (i + 1 < dpolicy->granularity)
1440                         break;
1441
1442                 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1443                         return __issue_discard_cmd_orderly(sbi, dpolicy);
1444
1445                 pend_list = &dcc->pend_list[i];
1446
1447                 mutex_lock(&dcc->cmd_lock);
1448                 if (list_empty(pend_list))
1449                         goto next;
1450                 if (unlikely(dcc->rbtree_check))
1451                         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1452                                                                 &dcc->root));
1453                 blk_start_plug(&plug);
1454                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1455                         f2fs_bug_on(sbi, dc->state != D_PREP);
1456
1457                         if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1458                                                 !is_idle(sbi, DISCARD_TIME)) {
1459                                 io_interrupted = true;
1460                                 break;
1461                         }
1462
1463                         __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1464
1465                         if (issued >= dpolicy->max_requests)
1466                                 break;
1467                 }
1468                 blk_finish_plug(&plug);
1469 next:
1470                 mutex_unlock(&dcc->cmd_lock);
1471
1472                 if (issued >= dpolicy->max_requests || io_interrupted)
1473                         break;
1474         }
1475
1476         if (!issued && io_interrupted)
1477                 issued = -1;
1478
1479         return issued;
1480 }
1481
1482 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1483 {
1484         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1485         struct list_head *pend_list;
1486         struct discard_cmd *dc, *tmp;
1487         int i;
1488         bool dropped = false;
1489
1490         mutex_lock(&dcc->cmd_lock);
1491         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1492                 pend_list = &dcc->pend_list[i];
1493                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1494                         f2fs_bug_on(sbi, dc->state != D_PREP);
1495                         __remove_discard_cmd(sbi, dc);
1496                         dropped = true;
1497                 }
1498         }
1499         mutex_unlock(&dcc->cmd_lock);
1500
1501         return dropped;
1502 }
1503
1504 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1505 {
1506         __drop_discard_cmd(sbi);
1507 }
1508
1509 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1510                                                         struct discard_cmd *dc)
1511 {
1512         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1513         unsigned int len = 0;
1514
1515         wait_for_completion_io(&dc->wait);
1516         mutex_lock(&dcc->cmd_lock);
1517         f2fs_bug_on(sbi, dc->state != D_DONE);
1518         dc->ref--;
1519         if (!dc->ref) {
1520                 if (!dc->error)
1521                         len = dc->len;
1522                 __remove_discard_cmd(sbi, dc);
1523         }
1524         mutex_unlock(&dcc->cmd_lock);
1525
1526         return len;
1527 }
1528
1529 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1530                                                 struct discard_policy *dpolicy,
1531                                                 block_t start, block_t end)
1532 {
1533         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1534         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1535                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1536         struct discard_cmd *dc, *tmp;
1537         bool need_wait;
1538         unsigned int trimmed = 0;
1539
1540 next:
1541         need_wait = false;
1542
1543         mutex_lock(&dcc->cmd_lock);
1544         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1545                 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1546                         continue;
1547                 if (dc->len < dpolicy->granularity)
1548                         continue;
1549                 if (dc->state == D_DONE && !dc->ref) {
1550                         wait_for_completion_io(&dc->wait);
1551                         if (!dc->error)
1552                                 trimmed += dc->len;
1553                         __remove_discard_cmd(sbi, dc);
1554                 } else {
1555                         dc->ref++;
1556                         need_wait = true;
1557                         break;
1558                 }
1559         }
1560         mutex_unlock(&dcc->cmd_lock);
1561
1562         if (need_wait) {
1563                 trimmed += __wait_one_discard_bio(sbi, dc);
1564                 goto next;
1565         }
1566
1567         return trimmed;
1568 }
1569
1570 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1571                                                 struct discard_policy *dpolicy)
1572 {
1573         struct discard_policy dp;
1574         unsigned int discard_blks;
1575
1576         if (dpolicy)
1577                 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1578
1579         /* wait all */
1580         __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1581         discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1582         __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1583         discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1584
1585         return discard_blks;
1586 }
1587
1588 /* This should be covered by global mutex, &sit_i->sentry_lock */
1589 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1590 {
1591         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1592         struct discard_cmd *dc;
1593         bool need_wait = false;
1594
1595         mutex_lock(&dcc->cmd_lock);
1596         dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1597                                                         NULL, blkaddr);
1598         if (dc) {
1599                 if (dc->state == D_PREP) {
1600                         __punch_discard_cmd(sbi, dc, blkaddr);
1601                 } else {
1602                         dc->ref++;
1603                         need_wait = true;
1604                 }
1605         }
1606         mutex_unlock(&dcc->cmd_lock);
1607
1608         if (need_wait)
1609                 __wait_one_discard_bio(sbi, dc);
1610 }
1611
1612 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1613 {
1614         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1615
1616         if (dcc && dcc->f2fs_issue_discard) {
1617                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1618
1619                 dcc->f2fs_issue_discard = NULL;
1620                 kthread_stop(discard_thread);
1621         }
1622 }
1623
1624 /* This comes from f2fs_put_super */
1625 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1626 {
1627         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1628         struct discard_policy dpolicy;
1629         bool dropped;
1630
1631         __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1632                                         dcc->discard_granularity);
1633         dpolicy.timeout = UMOUNT_DISCARD_TIMEOUT;
1634         __issue_discard_cmd(sbi, &dpolicy);
1635         dropped = __drop_discard_cmd(sbi);
1636
1637         /* just to make sure there is no pending discard commands */
1638         __wait_all_discard_cmd(sbi, NULL);
1639
1640         f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1641         return dropped;
1642 }
1643
1644 static int issue_discard_thread(void *data)
1645 {
1646         struct f2fs_sb_info *sbi = data;
1647         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1648         wait_queue_head_t *q = &dcc->discard_wait_queue;
1649         struct discard_policy dpolicy;
1650         unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1651         int issued;
1652
1653         set_freezable();
1654
1655         do {
1656                 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1657                                         dcc->discard_granularity);
1658
1659                 wait_event_interruptible_timeout(*q,
1660                                 kthread_should_stop() || freezing(current) ||
1661                                 dcc->discard_wake,
1662                                 msecs_to_jiffies(wait_ms));
1663
1664                 if (dcc->discard_wake)
1665                         dcc->discard_wake = 0;
1666
1667                 /* clean up pending candidates before going to sleep */
1668                 if (atomic_read(&dcc->queued_discard))
1669                         __wait_all_discard_cmd(sbi, NULL);
1670
1671                 if (try_to_freeze())
1672                         continue;
1673                 if (f2fs_readonly(sbi->sb))
1674                         continue;
1675                 if (kthread_should_stop())
1676                         return 0;
1677                 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1678                         wait_ms = dpolicy.max_interval;
1679                         continue;
1680                 }
1681
1682                 if (sbi->gc_mode == GC_URGENT)
1683                         __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1684
1685                 sb_start_intwrite(sbi->sb);
1686
1687                 issued = __issue_discard_cmd(sbi, &dpolicy);
1688                 if (issued > 0) {
1689                         __wait_all_discard_cmd(sbi, &dpolicy);
1690                         wait_ms = dpolicy.min_interval;
1691                 } else if (issued == -1){
1692                         wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1693                         if (!wait_ms)
1694                                 wait_ms = dpolicy.mid_interval;
1695                 } else {
1696                         wait_ms = dpolicy.max_interval;
1697                 }
1698
1699                 sb_end_intwrite(sbi->sb);
1700
1701         } while (!kthread_should_stop());
1702         return 0;
1703 }
1704
1705 #ifdef CONFIG_BLK_DEV_ZONED
1706 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1707                 struct block_device *bdev, block_t blkstart, block_t blklen)
1708 {
1709         sector_t sector, nr_sects;
1710         block_t lblkstart = blkstart;
1711         int devi = 0;
1712
1713         if (sbi->s_ndevs) {
1714                 devi = f2fs_target_device_index(sbi, blkstart);
1715                 blkstart -= FDEV(devi).start_blk;
1716         }
1717
1718         /*
1719          * We need to know the type of the zone: for conventional zones,
1720          * use regular discard if the drive supports it. For sequential
1721          * zones, reset the zone write pointer.
1722          */
1723         switch (get_blkz_type(sbi, bdev, blkstart)) {
1724
1725         case BLK_ZONE_TYPE_CONVENTIONAL:
1726                 if (!blk_queue_discard(bdev_get_queue(bdev)))
1727                         return 0;
1728                 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1729         case BLK_ZONE_TYPE_SEQWRITE_REQ:
1730         case BLK_ZONE_TYPE_SEQWRITE_PREF:
1731                 sector = SECTOR_FROM_BLOCK(blkstart);
1732                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1733
1734                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1735                                 nr_sects != bdev_zone_sectors(bdev)) {
1736                         f2fs_msg(sbi->sb, KERN_INFO,
1737                                 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1738                                 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1739                                 blkstart, blklen);
1740                         return -EIO;
1741                 }
1742                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1743                 return blkdev_reset_zones(bdev, sector,
1744                                           nr_sects, GFP_NOFS);
1745         default:
1746                 /* Unknown zone type: broken device ? */
1747                 return -EIO;
1748         }
1749 }
1750 #endif
1751
1752 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1753                 struct block_device *bdev, block_t blkstart, block_t blklen)
1754 {
1755 #ifdef CONFIG_BLK_DEV_ZONED
1756         if (f2fs_sb_has_blkzoned(sbi) &&
1757                                 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1758                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1759 #endif
1760         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1761 }
1762
1763 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1764                                 block_t blkstart, block_t blklen)
1765 {
1766         sector_t start = blkstart, len = 0;
1767         struct block_device *bdev;
1768         struct seg_entry *se;
1769         unsigned int offset;
1770         block_t i;
1771         int err = 0;
1772
1773         bdev = f2fs_target_device(sbi, blkstart, NULL);
1774
1775         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1776                 if (i != start) {
1777                         struct block_device *bdev2 =
1778                                 f2fs_target_device(sbi, i, NULL);
1779
1780                         if (bdev2 != bdev) {
1781                                 err = __issue_discard_async(sbi, bdev,
1782                                                 start, len);
1783                                 if (err)
1784                                         return err;
1785                                 bdev = bdev2;
1786                                 start = i;
1787                                 len = 0;
1788                         }
1789                 }
1790
1791                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1792                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1793
1794                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1795                         sbi->discard_blks--;
1796         }
1797
1798         if (len)
1799                 err = __issue_discard_async(sbi, bdev, start, len);
1800         return err;
1801 }
1802
1803 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1804                                                         bool check_only)
1805 {
1806         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1807         int max_blocks = sbi->blocks_per_seg;
1808         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1809         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1810         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1811         unsigned long *discard_map = (unsigned long *)se->discard_map;
1812         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1813         unsigned int start = 0, end = -1;
1814         bool force = (cpc->reason & CP_DISCARD);
1815         struct discard_entry *de = NULL;
1816         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1817         int i;
1818
1819         if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
1820                 return false;
1821
1822         if (!force) {
1823                 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1824                         SM_I(sbi)->dcc_info->nr_discards >=
1825                                 SM_I(sbi)->dcc_info->max_discards)
1826                         return false;
1827         }
1828
1829         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1830         for (i = 0; i < entries; i++)
1831                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1832                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1833
1834         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1835                                 SM_I(sbi)->dcc_info->max_discards) {
1836                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1837                 if (start >= max_blocks)
1838                         break;
1839
1840                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1841                 if (force && start && end != max_blocks
1842                                         && (end - start) < cpc->trim_minlen)
1843                         continue;
1844
1845                 if (check_only)
1846                         return true;
1847
1848                 if (!de) {
1849                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1850                                                                 GFP_F2FS_ZERO);
1851                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1852                         list_add_tail(&de->list, head);
1853                 }
1854
1855                 for (i = start; i < end; i++)
1856                         __set_bit_le(i, (void *)de->discard_map);
1857
1858                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1859         }
1860         return false;
1861 }
1862
1863 static void release_discard_addr(struct discard_entry *entry)
1864 {
1865         list_del(&entry->list);
1866         kmem_cache_free(discard_entry_slab, entry);
1867 }
1868
1869 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1870 {
1871         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1872         struct discard_entry *entry, *this;
1873
1874         /* drop caches */
1875         list_for_each_entry_safe(entry, this, head, list)
1876                 release_discard_addr(entry);
1877 }
1878
1879 /*
1880  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1881  */
1882 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1883 {
1884         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1885         unsigned int segno;
1886
1887         mutex_lock(&dirty_i->seglist_lock);
1888         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1889                 __set_test_and_free(sbi, segno);
1890         mutex_unlock(&dirty_i->seglist_lock);
1891 }
1892
1893 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1894                                                 struct cp_control *cpc)
1895 {
1896         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1897         struct list_head *head = &dcc->entry_list;
1898         struct discard_entry *entry, *this;
1899         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1900         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1901         unsigned int start = 0, end = -1;
1902         unsigned int secno, start_segno;
1903         bool force = (cpc->reason & CP_DISCARD);
1904         bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
1905
1906         mutex_lock(&dirty_i->seglist_lock);
1907
1908         while (1) {
1909                 int i;
1910
1911                 if (need_align && end != -1)
1912                         end--;
1913                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1914                 if (start >= MAIN_SEGS(sbi))
1915                         break;
1916                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1917                                                                 start + 1);
1918
1919                 if (need_align) {
1920                         start = rounddown(start, sbi->segs_per_sec);
1921                         end = roundup(end, sbi->segs_per_sec);
1922                 }
1923
1924                 for (i = start; i < end; i++) {
1925                         if (test_and_clear_bit(i, prefree_map))
1926                                 dirty_i->nr_dirty[PRE]--;
1927                 }
1928
1929                 if (!f2fs_realtime_discard_enable(sbi))
1930                         continue;
1931
1932                 if (force && start >= cpc->trim_start &&
1933                                         (end - 1) <= cpc->trim_end)
1934                                 continue;
1935
1936                 if (!test_opt(sbi, LFS) || !__is_large_section(sbi)) {
1937                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1938                                 (end - start) << sbi->log_blocks_per_seg);
1939                         continue;
1940                 }
1941 next:
1942                 secno = GET_SEC_FROM_SEG(sbi, start);
1943                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1944                 if (!IS_CURSEC(sbi, secno) &&
1945                         !get_valid_blocks(sbi, start, true))
1946                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1947                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1948
1949                 start = start_segno + sbi->segs_per_sec;
1950                 if (start < end)
1951                         goto next;
1952                 else
1953                         end = start - 1;
1954         }
1955         mutex_unlock(&dirty_i->seglist_lock);
1956
1957         /* send small discards */
1958         list_for_each_entry_safe(entry, this, head, list) {
1959                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1960                 bool is_valid = test_bit_le(0, entry->discard_map);
1961
1962 find_next:
1963                 if (is_valid) {
1964                         next_pos = find_next_zero_bit_le(entry->discard_map,
1965                                         sbi->blocks_per_seg, cur_pos);
1966                         len = next_pos - cur_pos;
1967
1968                         if (f2fs_sb_has_blkzoned(sbi) ||
1969                             (force && len < cpc->trim_minlen))
1970                                 goto skip;
1971
1972                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1973                                                                         len);
1974                         total_len += len;
1975                 } else {
1976                         next_pos = find_next_bit_le(entry->discard_map,
1977                                         sbi->blocks_per_seg, cur_pos);
1978                 }
1979 skip:
1980                 cur_pos = next_pos;
1981                 is_valid = !is_valid;
1982
1983                 if (cur_pos < sbi->blocks_per_seg)
1984                         goto find_next;
1985
1986                 release_discard_addr(entry);
1987                 dcc->nr_discards -= total_len;
1988         }
1989
1990         wake_up_discard_thread(sbi, false);
1991 }
1992
1993 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1994 {
1995         dev_t dev = sbi->sb->s_bdev->bd_dev;
1996         struct discard_cmd_control *dcc;
1997         int err = 0, i;
1998
1999         if (SM_I(sbi)->dcc_info) {
2000                 dcc = SM_I(sbi)->dcc_info;
2001                 goto init_thread;
2002         }
2003
2004         dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2005         if (!dcc)
2006                 return -ENOMEM;
2007
2008         dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2009         INIT_LIST_HEAD(&dcc->entry_list);
2010         for (i = 0; i < MAX_PLIST_NUM; i++)
2011                 INIT_LIST_HEAD(&dcc->pend_list[i]);
2012         INIT_LIST_HEAD(&dcc->wait_list);
2013         INIT_LIST_HEAD(&dcc->fstrim_list);
2014         mutex_init(&dcc->cmd_lock);
2015         atomic_set(&dcc->issued_discard, 0);
2016         atomic_set(&dcc->queued_discard, 0);
2017         atomic_set(&dcc->discard_cmd_cnt, 0);
2018         dcc->nr_discards = 0;
2019         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2020         dcc->undiscard_blks = 0;
2021         dcc->next_pos = 0;
2022         dcc->root = RB_ROOT_CACHED;
2023         dcc->rbtree_check = false;
2024
2025         init_waitqueue_head(&dcc->discard_wait_queue);
2026         SM_I(sbi)->dcc_info = dcc;
2027 init_thread:
2028         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2029                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2030         if (IS_ERR(dcc->f2fs_issue_discard)) {
2031                 err = PTR_ERR(dcc->f2fs_issue_discard);
2032                 kvfree(dcc);
2033                 SM_I(sbi)->dcc_info = NULL;
2034                 return err;
2035         }
2036
2037         return err;
2038 }
2039
2040 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2041 {
2042         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2043
2044         if (!dcc)
2045                 return;
2046
2047         f2fs_stop_discard_thread(sbi);
2048
2049         kvfree(dcc);
2050         SM_I(sbi)->dcc_info = NULL;
2051 }
2052
2053 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2054 {
2055         struct sit_info *sit_i = SIT_I(sbi);
2056
2057         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2058                 sit_i->dirty_sentries++;
2059                 return false;
2060         }
2061
2062         return true;
2063 }
2064
2065 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2066                                         unsigned int segno, int modified)
2067 {
2068         struct seg_entry *se = get_seg_entry(sbi, segno);
2069         se->type = type;
2070         if (modified)
2071                 __mark_sit_entry_dirty(sbi, segno);
2072 }
2073
2074 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2075 {
2076         struct seg_entry *se;
2077         unsigned int segno, offset;
2078         long int new_vblocks;
2079         bool exist;
2080 #ifdef CONFIG_F2FS_CHECK_FS
2081         bool mir_exist;
2082 #endif
2083
2084         segno = GET_SEGNO(sbi, blkaddr);
2085
2086         se = get_seg_entry(sbi, segno);
2087         new_vblocks = se->valid_blocks + del;
2088         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2089
2090         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
2091                                 (new_vblocks > sbi->blocks_per_seg)));
2092
2093         se->valid_blocks = new_vblocks;
2094         se->mtime = get_mtime(sbi, false);
2095         if (se->mtime > SIT_I(sbi)->max_mtime)
2096                 SIT_I(sbi)->max_mtime = se->mtime;
2097
2098         /* Update valid block bitmap */
2099         if (del > 0) {
2100                 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2101 #ifdef CONFIG_F2FS_CHECK_FS
2102                 mir_exist = f2fs_test_and_set_bit(offset,
2103                                                 se->cur_valid_map_mir);
2104                 if (unlikely(exist != mir_exist)) {
2105                         f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2106                                 "when setting bitmap, blk:%u, old bit:%d",
2107                                 blkaddr, exist);
2108                         f2fs_bug_on(sbi, 1);
2109                 }
2110 #endif
2111                 if (unlikely(exist)) {
2112                         f2fs_msg(sbi->sb, KERN_ERR,
2113                                 "Bitmap was wrongly set, blk:%u", blkaddr);
2114                         f2fs_bug_on(sbi, 1);
2115                         se->valid_blocks--;
2116                         del = 0;
2117                 }
2118
2119                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
2120                         sbi->discard_blks--;
2121
2122                 /* don't overwrite by SSR to keep node chain */
2123                 if (IS_NODESEG(se->type) &&
2124                                 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2125                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2126                                 se->ckpt_valid_blocks++;
2127                 }
2128         } else {
2129                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2130 #ifdef CONFIG_F2FS_CHECK_FS
2131                 mir_exist = f2fs_test_and_clear_bit(offset,
2132                                                 se->cur_valid_map_mir);
2133                 if (unlikely(exist != mir_exist)) {
2134                         f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2135                                 "when clearing bitmap, blk:%u, old bit:%d",
2136                                 blkaddr, exist);
2137                         f2fs_bug_on(sbi, 1);
2138                 }
2139 #endif
2140                 if (unlikely(!exist)) {
2141                         f2fs_msg(sbi->sb, KERN_ERR,
2142                                 "Bitmap was wrongly cleared, blk:%u", blkaddr);
2143                         f2fs_bug_on(sbi, 1);
2144                         se->valid_blocks++;
2145                         del = 0;
2146                 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2147                         /*
2148                          * If checkpoints are off, we must not reuse data that
2149                          * was used in the previous checkpoint. If it was used
2150                          * before, we must track that to know how much space we
2151                          * really have.
2152                          */
2153                         if (f2fs_test_bit(offset, se->ckpt_valid_map))
2154                                 sbi->unusable_block_count++;
2155                 }
2156
2157                 if (f2fs_test_and_clear_bit(offset, se->discard_map))
2158                         sbi->discard_blks++;
2159         }
2160         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2161                 se->ckpt_valid_blocks += del;
2162
2163         __mark_sit_entry_dirty(sbi, segno);
2164
2165         /* update total number of valid blocks to be written in ckpt area */
2166         SIT_I(sbi)->written_valid_blocks += del;
2167
2168         if (__is_large_section(sbi))
2169                 get_sec_entry(sbi, segno)->valid_blocks += del;
2170 }
2171
2172 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2173 {
2174         unsigned int segno = GET_SEGNO(sbi, addr);
2175         struct sit_info *sit_i = SIT_I(sbi);
2176
2177         f2fs_bug_on(sbi, addr == NULL_ADDR);
2178         if (addr == NEW_ADDR)
2179                 return;
2180
2181         invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2182
2183         /* add it into sit main buffer */
2184         down_write(&sit_i->sentry_lock);
2185
2186         update_sit_entry(sbi, addr, -1);
2187
2188         /* add it into dirty seglist */
2189         locate_dirty_segment(sbi, segno);
2190
2191         up_write(&sit_i->sentry_lock);
2192 }
2193
2194 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2195 {
2196         struct sit_info *sit_i = SIT_I(sbi);
2197         unsigned int segno, offset;
2198         struct seg_entry *se;
2199         bool is_cp = false;
2200
2201         if (!is_valid_data_blkaddr(sbi, blkaddr))
2202                 return true;
2203
2204         down_read(&sit_i->sentry_lock);
2205
2206         segno = GET_SEGNO(sbi, blkaddr);
2207         se = get_seg_entry(sbi, segno);
2208         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2209
2210         if (f2fs_test_bit(offset, se->ckpt_valid_map))
2211                 is_cp = true;
2212
2213         up_read(&sit_i->sentry_lock);
2214
2215         return is_cp;
2216 }
2217
2218 /*
2219  * This function should be resided under the curseg_mutex lock
2220  */
2221 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2222                                         struct f2fs_summary *sum)
2223 {
2224         struct curseg_info *curseg = CURSEG_I(sbi, type);
2225         void *addr = curseg->sum_blk;
2226         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2227         memcpy(addr, sum, sizeof(struct f2fs_summary));
2228 }
2229
2230 /*
2231  * Calculate the number of current summary pages for writing
2232  */
2233 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2234 {
2235         int valid_sum_count = 0;
2236         int i, sum_in_page;
2237
2238         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2239                 if (sbi->ckpt->alloc_type[i] == SSR)
2240                         valid_sum_count += sbi->blocks_per_seg;
2241                 else {
2242                         if (for_ra)
2243                                 valid_sum_count += le16_to_cpu(
2244                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2245                         else
2246                                 valid_sum_count += curseg_blkoff(sbi, i);
2247                 }
2248         }
2249
2250         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2251                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2252         if (valid_sum_count <= sum_in_page)
2253                 return 1;
2254         else if ((valid_sum_count - sum_in_page) <=
2255                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2256                 return 2;
2257         return 3;
2258 }
2259
2260 /*
2261  * Caller should put this summary page
2262  */
2263 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2264 {
2265         return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
2266 }
2267
2268 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2269                                         void *src, block_t blk_addr)
2270 {
2271         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2272
2273         memcpy(page_address(page), src, PAGE_SIZE);
2274         set_page_dirty(page);
2275         f2fs_put_page(page, 1);
2276 }
2277
2278 static void write_sum_page(struct f2fs_sb_info *sbi,
2279                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
2280 {
2281         f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2282 }
2283
2284 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2285                                                 int type, block_t blk_addr)
2286 {
2287         struct curseg_info *curseg = CURSEG_I(sbi, type);
2288         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2289         struct f2fs_summary_block *src = curseg->sum_blk;
2290         struct f2fs_summary_block *dst;
2291
2292         dst = (struct f2fs_summary_block *)page_address(page);
2293         memset(dst, 0, PAGE_SIZE);
2294
2295         mutex_lock(&curseg->curseg_mutex);
2296
2297         down_read(&curseg->journal_rwsem);
2298         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2299         up_read(&curseg->journal_rwsem);
2300
2301         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2302         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2303
2304         mutex_unlock(&curseg->curseg_mutex);
2305
2306         set_page_dirty(page);
2307         f2fs_put_page(page, 1);
2308 }
2309
2310 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2311 {
2312         struct curseg_info *curseg = CURSEG_I(sbi, type);
2313         unsigned int segno = curseg->segno + 1;
2314         struct free_segmap_info *free_i = FREE_I(sbi);
2315
2316         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2317                 return !test_bit(segno, free_i->free_segmap);
2318         return 0;
2319 }
2320
2321 /*
2322  * Find a new segment from the free segments bitmap to right order
2323  * This function should be returned with success, otherwise BUG
2324  */
2325 static void get_new_segment(struct f2fs_sb_info *sbi,
2326                         unsigned int *newseg, bool new_sec, int dir)
2327 {
2328         struct free_segmap_info *free_i = FREE_I(sbi);
2329         unsigned int segno, secno, zoneno;
2330         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2331         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2332         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2333         unsigned int left_start = hint;
2334         bool init = true;
2335         int go_left = 0;
2336         int i;
2337
2338         spin_lock(&free_i->segmap_lock);
2339
2340         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2341                 segno = find_next_zero_bit(free_i->free_segmap,
2342                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2343                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2344                         goto got_it;
2345         }
2346 find_other_zone:
2347         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2348         if (secno >= MAIN_SECS(sbi)) {
2349                 if (dir == ALLOC_RIGHT) {
2350                         secno = find_next_zero_bit(free_i->free_secmap,
2351                                                         MAIN_SECS(sbi), 0);
2352                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2353                 } else {
2354                         go_left = 1;
2355                         left_start = hint - 1;
2356                 }
2357         }
2358         if (go_left == 0)
2359                 goto skip_left;
2360
2361         while (test_bit(left_start, free_i->free_secmap)) {
2362                 if (left_start > 0) {
2363                         left_start--;
2364                         continue;
2365                 }
2366                 left_start = find_next_zero_bit(free_i->free_secmap,
2367                                                         MAIN_SECS(sbi), 0);
2368                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2369                 break;
2370         }
2371         secno = left_start;
2372 skip_left:
2373         segno = GET_SEG_FROM_SEC(sbi, secno);
2374         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2375
2376         /* give up on finding another zone */
2377         if (!init)
2378                 goto got_it;
2379         if (sbi->secs_per_zone == 1)
2380                 goto got_it;
2381         if (zoneno == old_zoneno)
2382                 goto got_it;
2383         if (dir == ALLOC_LEFT) {
2384                 if (!go_left && zoneno + 1 >= total_zones)
2385                         goto got_it;
2386                 if (go_left && zoneno == 0)
2387                         goto got_it;
2388         }
2389         for (i = 0; i < NR_CURSEG_TYPE; i++)
2390                 if (CURSEG_I(sbi, i)->zone == zoneno)
2391                         break;
2392
2393         if (i < NR_CURSEG_TYPE) {
2394                 /* zone is in user, try another */
2395                 if (go_left)
2396                         hint = zoneno * sbi->secs_per_zone - 1;
2397                 else if (zoneno + 1 >= total_zones)
2398                         hint = 0;
2399                 else
2400                         hint = (zoneno + 1) * sbi->secs_per_zone;
2401                 init = false;
2402                 goto find_other_zone;
2403         }
2404 got_it:
2405         /* set it as dirty segment in free segmap */
2406         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2407         __set_inuse(sbi, segno);
2408         *newseg = segno;
2409         spin_unlock(&free_i->segmap_lock);
2410 }
2411
2412 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2413 {
2414         struct curseg_info *curseg = CURSEG_I(sbi, type);
2415         struct summary_footer *sum_footer;
2416
2417         curseg->segno = curseg->next_segno;
2418         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2419         curseg->next_blkoff = 0;
2420         curseg->next_segno = NULL_SEGNO;
2421
2422         sum_footer = &(curseg->sum_blk->footer);
2423         memset(sum_footer, 0, sizeof(struct summary_footer));
2424         if (IS_DATASEG(type))
2425                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2426         if (IS_NODESEG(type))
2427                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2428         __set_sit_entry_type(sbi, type, curseg->segno, modified);
2429 }
2430
2431 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2432 {
2433         /* if segs_per_sec is large than 1, we need to keep original policy. */
2434         if (__is_large_section(sbi))
2435                 return CURSEG_I(sbi, type)->segno;
2436
2437         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2438                 return 0;
2439
2440         if (test_opt(sbi, NOHEAP) &&
2441                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2442                 return 0;
2443
2444         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2445                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2446
2447         /* find segments from 0 to reuse freed segments */
2448         if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2449                 return 0;
2450
2451         return CURSEG_I(sbi, type)->segno;
2452 }
2453
2454 /*
2455  * Allocate a current working segment.
2456  * This function always allocates a free segment in LFS manner.
2457  */
2458 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2459 {
2460         struct curseg_info *curseg = CURSEG_I(sbi, type);
2461         unsigned int segno = curseg->segno;
2462         int dir = ALLOC_LEFT;
2463
2464         write_sum_page(sbi, curseg->sum_blk,
2465                                 GET_SUM_BLOCK(sbi, segno));
2466         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2467                 dir = ALLOC_RIGHT;
2468
2469         if (test_opt(sbi, NOHEAP))
2470                 dir = ALLOC_RIGHT;
2471
2472         segno = __get_next_segno(sbi, type);
2473         get_new_segment(sbi, &segno, new_sec, dir);
2474         curseg->next_segno = segno;
2475         reset_curseg(sbi, type, 1);
2476         curseg->alloc_type = LFS;
2477 }
2478
2479 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2480                         struct curseg_info *seg, block_t start)
2481 {
2482         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2483         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2484         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2485         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2486         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2487         int i, pos;
2488
2489         for (i = 0; i < entries; i++)
2490                 target_map[i] = ckpt_map[i] | cur_map[i];
2491
2492         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2493
2494         seg->next_blkoff = pos;
2495 }
2496
2497 /*
2498  * If a segment is written by LFS manner, next block offset is just obtained
2499  * by increasing the current block offset. However, if a segment is written by
2500  * SSR manner, next block offset obtained by calling __next_free_blkoff
2501  */
2502 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2503                                 struct curseg_info *seg)
2504 {
2505         if (seg->alloc_type == SSR)
2506                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2507         else
2508                 seg->next_blkoff++;
2509 }
2510
2511 /*
2512  * This function always allocates a used segment(from dirty seglist) by SSR
2513  * manner, so it should recover the existing segment information of valid blocks
2514  */
2515 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2516 {
2517         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2518         struct curseg_info *curseg = CURSEG_I(sbi, type);
2519         unsigned int new_segno = curseg->next_segno;
2520         struct f2fs_summary_block *sum_node;
2521         struct page *sum_page;
2522
2523         write_sum_page(sbi, curseg->sum_blk,
2524                                 GET_SUM_BLOCK(sbi, curseg->segno));
2525         __set_test_and_inuse(sbi, new_segno);
2526
2527         mutex_lock(&dirty_i->seglist_lock);
2528         __remove_dirty_segment(sbi, new_segno, PRE);
2529         __remove_dirty_segment(sbi, new_segno, DIRTY);
2530         mutex_unlock(&dirty_i->seglist_lock);
2531
2532         reset_curseg(sbi, type, 1);
2533         curseg->alloc_type = SSR;
2534         __next_free_blkoff(sbi, curseg, 0);
2535
2536         sum_page = f2fs_get_sum_page(sbi, new_segno);
2537         f2fs_bug_on(sbi, IS_ERR(sum_page));
2538         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2539         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2540         f2fs_put_page(sum_page, 1);
2541 }
2542
2543 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2544 {
2545         struct curseg_info *curseg = CURSEG_I(sbi, type);
2546         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2547         unsigned segno = NULL_SEGNO;
2548         int i, cnt;
2549         bool reversed = false;
2550
2551         /* f2fs_need_SSR() already forces to do this */
2552         if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2553                 curseg->next_segno = segno;
2554                 return 1;
2555         }
2556
2557         /* For node segments, let's do SSR more intensively */
2558         if (IS_NODESEG(type)) {
2559                 if (type >= CURSEG_WARM_NODE) {
2560                         reversed = true;
2561                         i = CURSEG_COLD_NODE;
2562                 } else {
2563                         i = CURSEG_HOT_NODE;
2564                 }
2565                 cnt = NR_CURSEG_NODE_TYPE;
2566         } else {
2567                 if (type >= CURSEG_WARM_DATA) {
2568                         reversed = true;
2569                         i = CURSEG_COLD_DATA;
2570                 } else {
2571                         i = CURSEG_HOT_DATA;
2572                 }
2573                 cnt = NR_CURSEG_DATA_TYPE;
2574         }
2575
2576         for (; cnt-- > 0; reversed ? i-- : i++) {
2577                 if (i == type)
2578                         continue;
2579                 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2580                         curseg->next_segno = segno;
2581                         return 1;
2582                 }
2583         }
2584
2585         /* find valid_blocks=0 in dirty list */
2586         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2587                 segno = get_free_segment(sbi);
2588                 if (segno != NULL_SEGNO) {
2589                         curseg->next_segno = segno;
2590                         return 1;
2591                 }
2592         }
2593         return 0;
2594 }
2595
2596 /*
2597  * flush out current segment and replace it with new segment
2598  * This function should be returned with success, otherwise BUG
2599  */
2600 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2601                                                 int type, bool force)
2602 {
2603         struct curseg_info *curseg = CURSEG_I(sbi, type);
2604
2605         if (force)
2606                 new_curseg(sbi, type, true);
2607         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2608                                         type == CURSEG_WARM_NODE)
2609                 new_curseg(sbi, type, false);
2610         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2611                         likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2612                 new_curseg(sbi, type, false);
2613         else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2614                 change_curseg(sbi, type);
2615         else
2616                 new_curseg(sbi, type, false);
2617
2618         stat_inc_seg_type(sbi, curseg);
2619 }
2620
2621 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2622 {
2623         struct curseg_info *curseg;
2624         unsigned int old_segno;
2625         int i;
2626
2627         down_write(&SIT_I(sbi)->sentry_lock);
2628
2629         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2630                 curseg = CURSEG_I(sbi, i);
2631                 old_segno = curseg->segno;
2632                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2633                 locate_dirty_segment(sbi, old_segno);
2634         }
2635
2636         up_write(&SIT_I(sbi)->sentry_lock);
2637 }
2638
2639 static const struct segment_allocation default_salloc_ops = {
2640         .allocate_segment = allocate_segment_by_default,
2641 };
2642
2643 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2644                                                 struct cp_control *cpc)
2645 {
2646         __u64 trim_start = cpc->trim_start;
2647         bool has_candidate = false;
2648
2649         down_write(&SIT_I(sbi)->sentry_lock);
2650         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2651                 if (add_discard_addrs(sbi, cpc, true)) {
2652                         has_candidate = true;
2653                         break;
2654                 }
2655         }
2656         up_write(&SIT_I(sbi)->sentry_lock);
2657
2658         cpc->trim_start = trim_start;
2659         return has_candidate;
2660 }
2661
2662 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2663                                         struct discard_policy *dpolicy,
2664                                         unsigned int start, unsigned int end)
2665 {
2666         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2667         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2668         struct rb_node **insert_p = NULL, *insert_parent = NULL;
2669         struct discard_cmd *dc;
2670         struct blk_plug plug;
2671         int issued;
2672         unsigned int trimmed = 0;
2673
2674 next:
2675         issued = 0;
2676
2677         mutex_lock(&dcc->cmd_lock);
2678         if (unlikely(dcc->rbtree_check))
2679                 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2680                                                                 &dcc->root));
2681
2682         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2683                                         NULL, start,
2684                                         (struct rb_entry **)&prev_dc,
2685                                         (struct rb_entry **)&next_dc,
2686                                         &insert_p, &insert_parent, true, NULL);
2687         if (!dc)
2688                 dc = next_dc;
2689
2690         blk_start_plug(&plug);
2691
2692         while (dc && dc->lstart <= end) {
2693                 struct rb_node *node;
2694                 int err = 0;
2695
2696                 if (dc->len < dpolicy->granularity)
2697                         goto skip;
2698
2699                 if (dc->state != D_PREP) {
2700                         list_move_tail(&dc->list, &dcc->fstrim_list);
2701                         goto skip;
2702                 }
2703
2704                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2705
2706                 if (issued >= dpolicy->max_requests) {
2707                         start = dc->lstart + dc->len;
2708
2709                         if (err)
2710                                 __remove_discard_cmd(sbi, dc);
2711
2712                         blk_finish_plug(&plug);
2713                         mutex_unlock(&dcc->cmd_lock);
2714                         trimmed += __wait_all_discard_cmd(sbi, NULL);
2715                         congestion_wait(BLK_RW_ASYNC, HZ/50);
2716                         goto next;
2717                 }
2718 skip:
2719                 node = rb_next(&dc->rb_node);
2720                 if (err)
2721                         __remove_discard_cmd(sbi, dc);
2722                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2723
2724                 if (fatal_signal_pending(current))
2725                         break;
2726         }
2727
2728         blk_finish_plug(&plug);
2729         mutex_unlock(&dcc->cmd_lock);
2730
2731         return trimmed;
2732 }
2733
2734 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2735 {
2736         __u64 start = F2FS_BYTES_TO_BLK(range->start);
2737         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2738         unsigned int start_segno, end_segno;
2739         block_t start_block, end_block;
2740         struct cp_control cpc;
2741         struct discard_policy dpolicy;
2742         unsigned long long trimmed = 0;
2743         int err = 0;
2744         bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
2745
2746         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2747                 return -EINVAL;
2748
2749         if (end < MAIN_BLKADDR(sbi))
2750                 goto out;
2751
2752         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2753                 f2fs_msg(sbi->sb, KERN_WARNING,
2754                         "Found FS corruption, run fsck to fix.");
2755                 return -EIO;
2756         }
2757
2758         /* start/end segment number in main_area */
2759         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2760         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2761                                                 GET_SEGNO(sbi, end);
2762         if (need_align) {
2763                 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2764                 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2765         }
2766
2767         cpc.reason = CP_DISCARD;
2768         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2769         cpc.trim_start = start_segno;
2770         cpc.trim_end = end_segno;
2771
2772         if (sbi->discard_blks == 0)
2773                 goto out;
2774
2775         mutex_lock(&sbi->gc_mutex);
2776         err = f2fs_write_checkpoint(sbi, &cpc);
2777         mutex_unlock(&sbi->gc_mutex);
2778         if (err)
2779                 goto out;
2780
2781         /*
2782          * We filed discard candidates, but actually we don't need to wait for
2783          * all of them, since they'll be issued in idle time along with runtime
2784          * discard option. User configuration looks like using runtime discard
2785          * or periodic fstrim instead of it.
2786          */
2787         if (f2fs_realtime_discard_enable(sbi))
2788                 goto out;
2789
2790         start_block = START_BLOCK(sbi, start_segno);
2791         end_block = START_BLOCK(sbi, end_segno + 1);
2792
2793         __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2794         trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2795                                         start_block, end_block);
2796
2797         trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
2798                                         start_block, end_block);
2799 out:
2800         if (!err)
2801                 range->len = F2FS_BLK_TO_BYTES(trimmed);
2802         return err;
2803 }
2804
2805 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2806 {
2807         struct curseg_info *curseg = CURSEG_I(sbi, type);
2808         if (curseg->next_blkoff < sbi->blocks_per_seg)
2809                 return true;
2810         return false;
2811 }
2812
2813 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
2814 {
2815         switch (hint) {
2816         case WRITE_LIFE_SHORT:
2817                 return CURSEG_HOT_DATA;
2818         case WRITE_LIFE_EXTREME:
2819                 return CURSEG_COLD_DATA;
2820         default:
2821                 return CURSEG_WARM_DATA;
2822         }
2823 }
2824
2825 /* This returns write hints for each segment type. This hints will be
2826  * passed down to block layer. There are mapping tables which depend on
2827  * the mount option 'whint_mode'.
2828  *
2829  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2830  *
2831  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2832  *
2833  * User                  F2FS                     Block
2834  * ----                  ----                     -----
2835  *                       META                     WRITE_LIFE_NOT_SET
2836  *                       HOT_NODE                 "
2837  *                       WARM_NODE                "
2838  *                       COLD_NODE                "
2839  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2840  * extension list        "                        "
2841  *
2842  * -- buffered io
2843  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2844  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2845  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2846  * WRITE_LIFE_NONE       "                        "
2847  * WRITE_LIFE_MEDIUM     "                        "
2848  * WRITE_LIFE_LONG       "                        "
2849  *
2850  * -- direct io
2851  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2852  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2853  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2854  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2855  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2856  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2857  *
2858  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2859  *
2860  * User                  F2FS                     Block
2861  * ----                  ----                     -----
2862  *                       META                     WRITE_LIFE_MEDIUM;
2863  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
2864  *                       WARM_NODE                "
2865  *                       COLD_NODE                WRITE_LIFE_NONE
2866  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2867  * extension list        "                        "
2868  *
2869  * -- buffered io
2870  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2871  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2872  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
2873  * WRITE_LIFE_NONE       "                        "
2874  * WRITE_LIFE_MEDIUM     "                        "
2875  * WRITE_LIFE_LONG       "                        "
2876  *
2877  * -- direct io
2878  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2879  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2880  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2881  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2882  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2883  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2884  */
2885
2886 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2887                                 enum page_type type, enum temp_type temp)
2888 {
2889         if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2890                 if (type == DATA) {
2891                         if (temp == WARM)
2892                                 return WRITE_LIFE_NOT_SET;
2893                         else if (temp == HOT)
2894                                 return WRITE_LIFE_SHORT;
2895                         else if (temp == COLD)
2896                                 return WRITE_LIFE_EXTREME;
2897                 } else {
2898                         return WRITE_LIFE_NOT_SET;
2899                 }
2900         } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2901                 if (type == DATA) {
2902                         if (temp == WARM)
2903                                 return WRITE_LIFE_LONG;
2904                         else if (temp == HOT)
2905                                 return WRITE_LIFE_SHORT;
2906                         else if (temp == COLD)
2907                                 return WRITE_LIFE_EXTREME;
2908                 } else if (type == NODE) {
2909                         if (temp == WARM || temp == HOT)
2910                                 return WRITE_LIFE_NOT_SET;
2911                         else if (temp == COLD)
2912                                 return WRITE_LIFE_NONE;
2913                 } else if (type == META) {
2914                         return WRITE_LIFE_MEDIUM;
2915                 }
2916         }
2917         return WRITE_LIFE_NOT_SET;
2918 }
2919
2920 static int __get_segment_type_2(struct f2fs_io_info *fio)
2921 {
2922         if (fio->type == DATA)
2923                 return CURSEG_HOT_DATA;
2924         else
2925                 return CURSEG_HOT_NODE;
2926 }
2927
2928 static int __get_segment_type_4(struct f2fs_io_info *fio)
2929 {
2930         if (fio->type == DATA) {
2931                 struct inode *inode = fio->page->mapping->host;
2932
2933                 if (S_ISDIR(inode->i_mode))
2934                         return CURSEG_HOT_DATA;
2935                 else
2936                         return CURSEG_COLD_DATA;
2937         } else {
2938                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2939                         return CURSEG_WARM_NODE;
2940                 else
2941                         return CURSEG_COLD_NODE;
2942         }
2943 }
2944
2945 static int __get_segment_type_6(struct f2fs_io_info *fio)
2946 {
2947         if (fio->type == DATA) {
2948                 struct inode *inode = fio->page->mapping->host;
2949
2950                 if (is_cold_data(fio->page) || file_is_cold(inode))
2951                         return CURSEG_COLD_DATA;
2952                 if (file_is_hot(inode) ||
2953                                 is_inode_flag_set(inode, FI_HOT_DATA) ||
2954                                 f2fs_is_atomic_file(inode) ||
2955                                 f2fs_is_volatile_file(inode))
2956                         return CURSEG_HOT_DATA;
2957                 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
2958         } else {
2959                 if (IS_DNODE(fio->page))
2960                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2961                                                 CURSEG_HOT_NODE;
2962                 return CURSEG_COLD_NODE;
2963         }
2964 }
2965
2966 static int __get_segment_type(struct f2fs_io_info *fio)
2967 {
2968         int type = 0;
2969
2970         switch (F2FS_OPTION(fio->sbi).active_logs) {
2971         case 2:
2972                 type = __get_segment_type_2(fio);
2973                 break;
2974         case 4:
2975                 type = __get_segment_type_4(fio);
2976                 break;
2977         case 6:
2978                 type = __get_segment_type_6(fio);
2979                 break;
2980         default:
2981                 f2fs_bug_on(fio->sbi, true);
2982         }
2983
2984         if (IS_HOT(type))
2985                 fio->temp = HOT;
2986         else if (IS_WARM(type))
2987                 fio->temp = WARM;
2988         else
2989                 fio->temp = COLD;
2990         return type;
2991 }
2992
2993 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2994                 block_t old_blkaddr, block_t *new_blkaddr,
2995                 struct f2fs_summary *sum, int type,
2996                 struct f2fs_io_info *fio, bool add_list)
2997 {
2998         struct sit_info *sit_i = SIT_I(sbi);
2999         struct curseg_info *curseg = CURSEG_I(sbi, type);
3000
3001         down_read(&SM_I(sbi)->curseg_lock);
3002
3003         mutex_lock(&curseg->curseg_mutex);
3004         down_write(&sit_i->sentry_lock);
3005
3006         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3007
3008         f2fs_wait_discard_bio(sbi, *new_blkaddr);
3009
3010         /*
3011          * __add_sum_entry should be resided under the curseg_mutex
3012          * because, this function updates a summary entry in the
3013          * current summary block.
3014          */
3015         __add_sum_entry(sbi, type, sum);
3016
3017         __refresh_next_blkoff(sbi, curseg);
3018
3019         stat_inc_block_count(sbi, curseg);
3020
3021         /*
3022          * SIT information should be updated before segment allocation,
3023          * since SSR needs latest valid block information.
3024          */
3025         update_sit_entry(sbi, *new_blkaddr, 1);
3026         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3027                 update_sit_entry(sbi, old_blkaddr, -1);
3028
3029         if (!__has_curseg_space(sbi, type))
3030                 sit_i->s_ops->allocate_segment(sbi, type, false);
3031
3032         /*
3033          * segment dirty status should be updated after segment allocation,
3034          * so we just need to update status only one time after previous
3035          * segment being closed.
3036          */
3037         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3038         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3039
3040         up_write(&sit_i->sentry_lock);
3041
3042         if (page && IS_NODESEG(type)) {
3043                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3044
3045                 f2fs_inode_chksum_set(sbi, page);
3046         }
3047
3048         if (add_list) {
3049                 struct f2fs_bio_info *io;
3050
3051                 INIT_LIST_HEAD(&fio->list);
3052                 fio->in_list = true;
3053                 fio->retry = false;
3054                 io = sbi->write_io[fio->type] + fio->temp;
3055                 spin_lock(&io->io_lock);
3056                 list_add_tail(&fio->list, &io->io_list);
3057                 spin_unlock(&io->io_lock);
3058         }
3059
3060         mutex_unlock(&curseg->curseg_mutex);
3061
3062         up_read(&SM_I(sbi)->curseg_lock);
3063 }
3064
3065 static void update_device_state(struct f2fs_io_info *fio)
3066 {
3067         struct f2fs_sb_info *sbi = fio->sbi;
3068         unsigned int devidx;
3069
3070         if (!sbi->s_ndevs)
3071                 return;
3072
3073         devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3074
3075         /* update device state for fsync */
3076         f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
3077
3078         /* update device state for checkpoint */
3079         if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3080                 spin_lock(&sbi->dev_lock);
3081                 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3082                 spin_unlock(&sbi->dev_lock);
3083         }
3084 }
3085
3086 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3087 {
3088         int type = __get_segment_type(fio);
3089         bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
3090
3091         if (keep_order)
3092                 down_read(&fio->sbi->io_order_lock);
3093 reallocate:
3094         f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3095                         &fio->new_blkaddr, sum, type, fio, true);
3096         if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3097                 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3098                                         fio->old_blkaddr, fio->old_blkaddr);
3099
3100         /* writeout dirty page into bdev */
3101         f2fs_submit_page_write(fio);
3102         if (fio->retry) {
3103                 fio->old_blkaddr = fio->new_blkaddr;
3104                 goto reallocate;
3105         }
3106
3107         update_device_state(fio);
3108
3109         if (keep_order)
3110                 up_read(&fio->sbi->io_order_lock);
3111 }
3112
3113 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3114                                         enum iostat_type io_type)
3115 {
3116         struct f2fs_io_info fio = {
3117                 .sbi = sbi,
3118                 .type = META,
3119                 .temp = HOT,
3120                 .op = REQ_OP_WRITE,
3121                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3122                 .old_blkaddr = page->index,
3123                 .new_blkaddr = page->index,
3124                 .page = page,
3125                 .encrypted_page = NULL,
3126                 .in_list = false,
3127         };
3128
3129         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3130                 fio.op_flags &= ~REQ_META;
3131
3132         set_page_writeback(page);
3133         ClearPageError(page);
3134         f2fs_submit_page_write(&fio);
3135
3136         stat_inc_meta_count(sbi, page->index);
3137         f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3138 }
3139
3140 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3141 {
3142         struct f2fs_summary sum;
3143
3144         set_summary(&sum, nid, 0, 0);
3145         do_write_page(&sum, fio);
3146
3147         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3148 }
3149
3150 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3151                                         struct f2fs_io_info *fio)
3152 {
3153         struct f2fs_sb_info *sbi = fio->sbi;
3154         struct f2fs_summary sum;
3155
3156         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3157         set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3158         do_write_page(&sum, fio);
3159         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3160
3161         f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3162 }
3163
3164 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3165 {
3166         int err;
3167         struct f2fs_sb_info *sbi = fio->sbi;
3168
3169         fio->new_blkaddr = fio->old_blkaddr;
3170         /* i/o temperature is needed for passing down write hints */
3171         __get_segment_type(fio);
3172
3173         f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
3174                         GET_SEGNO(sbi, fio->new_blkaddr))->type));
3175
3176         stat_inc_inplace_blocks(fio->sbi);
3177
3178         err = f2fs_submit_page_bio(fio);
3179         if (!err)
3180                 update_device_state(fio);
3181
3182         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3183
3184         return err;
3185 }
3186
3187 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3188                                                 unsigned int segno)
3189 {
3190         int i;
3191
3192         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3193                 if (CURSEG_I(sbi, i)->segno == segno)
3194                         break;
3195         }
3196         return i;
3197 }
3198
3199 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3200                                 block_t old_blkaddr, block_t new_blkaddr,
3201                                 bool recover_curseg, bool recover_newaddr)
3202 {
3203         struct sit_info *sit_i = SIT_I(sbi);
3204         struct curseg_info *curseg;
3205         unsigned int segno, old_cursegno;
3206         struct seg_entry *se;
3207         int type;
3208         unsigned short old_blkoff;
3209
3210         segno = GET_SEGNO(sbi, new_blkaddr);
3211         se = get_seg_entry(sbi, segno);
3212         type = se->type;
3213
3214         down_write(&SM_I(sbi)->curseg_lock);
3215
3216         if (!recover_curseg) {
3217                 /* for recovery flow */
3218                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3219                         if (old_blkaddr == NULL_ADDR)
3220                                 type = CURSEG_COLD_DATA;
3221                         else
3222                                 type = CURSEG_WARM_DATA;
3223                 }
3224         } else {
3225                 if (IS_CURSEG(sbi, segno)) {
3226                         /* se->type is volatile as SSR allocation */
3227                         type = __f2fs_get_curseg(sbi, segno);
3228                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3229                 } else {
3230                         type = CURSEG_WARM_DATA;
3231                 }
3232         }
3233
3234         f2fs_bug_on(sbi, !IS_DATASEG(type));
3235         curseg = CURSEG_I(sbi, type);
3236
3237         mutex_lock(&curseg->curseg_mutex);
3238         down_write(&sit_i->sentry_lock);
3239
3240         old_cursegno = curseg->segno;
3241         old_blkoff = curseg->next_blkoff;
3242
3243         /* change the current segment */
3244         if (segno != curseg->segno) {
3245                 curseg->next_segno = segno;
3246                 change_curseg(sbi, type);
3247         }
3248
3249         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3250         __add_sum_entry(sbi, type, sum);
3251
3252         if (!recover_curseg || recover_newaddr)
3253                 update_sit_entry(sbi, new_blkaddr, 1);
3254         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3255                 invalidate_mapping_pages(META_MAPPING(sbi),
3256                                         old_blkaddr, old_blkaddr);
3257                 update_sit_entry(sbi, old_blkaddr, -1);
3258         }
3259
3260         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3261         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3262
3263         locate_dirty_segment(sbi, old_cursegno);
3264
3265         if (recover_curseg) {
3266                 if (old_cursegno != curseg->segno) {
3267                         curseg->next_segno = old_cursegno;
3268                         change_curseg(sbi, type);
3269                 }
3270                 curseg->next_blkoff = old_blkoff;
3271         }
3272
3273         up_write(&sit_i->sentry_lock);
3274         mutex_unlock(&curseg->curseg_mutex);
3275         up_write(&SM_I(sbi)->curseg_lock);
3276 }
3277
3278 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3279                                 block_t old_addr, block_t new_addr,
3280                                 unsigned char version, bool recover_curseg,
3281                                 bool recover_newaddr)
3282 {
3283         struct f2fs_summary sum;
3284
3285         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3286
3287         f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3288                                         recover_curseg, recover_newaddr);
3289
3290         f2fs_update_data_blkaddr(dn, new_addr);
3291 }
3292
3293 void f2fs_wait_on_page_writeback(struct page *page,
3294                                 enum page_type type, bool ordered, bool locked)
3295 {
3296         if (PageWriteback(page)) {
3297                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3298
3299                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3300                 if (ordered) {
3301                         wait_on_page_writeback(page);
3302                         f2fs_bug_on(sbi, locked && PageWriteback(page));
3303                 } else {
3304                         wait_for_stable_page(page);
3305                 }
3306         }
3307 }
3308
3309 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3310 {
3311         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3312         struct page *cpage;
3313
3314         if (!f2fs_post_read_required(inode))
3315                 return;
3316
3317         if (!is_valid_data_blkaddr(sbi, blkaddr))
3318                 return;
3319
3320         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3321         if (cpage) {
3322                 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3323                 f2fs_put_page(cpage, 1);
3324         }
3325 }
3326
3327 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3328                                                                 block_t len)
3329 {
3330         block_t i;
3331
3332         for (i = 0; i < len; i++)
3333                 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3334 }
3335
3336 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3337 {
3338         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3339         struct curseg_info *seg_i;
3340         unsigned char *kaddr;
3341         struct page *page;
3342         block_t start;
3343         int i, j, offset;
3344
3345         start = start_sum_block(sbi);
3346
3347         page = f2fs_get_meta_page(sbi, start++);
3348         if (IS_ERR(page))
3349                 return PTR_ERR(page);
3350         kaddr = (unsigned char *)page_address(page);
3351
3352         /* Step 1: restore nat cache */
3353         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3354         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3355
3356         /* Step 2: restore sit cache */
3357         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3358         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3359         offset = 2 * SUM_JOURNAL_SIZE;
3360
3361         /* Step 3: restore summary entries */
3362         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3363                 unsigned short blk_off;
3364                 unsigned int segno;
3365
3366                 seg_i = CURSEG_I(sbi, i);
3367                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3368                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3369                 seg_i->next_segno = segno;
3370                 reset_curseg(sbi, i, 0);
3371                 seg_i->alloc_type = ckpt->alloc_type[i];
3372                 seg_i->next_blkoff = blk_off;
3373
3374                 if (seg_i->alloc_type == SSR)
3375                         blk_off = sbi->blocks_per_seg;
3376
3377                 for (j = 0; j < blk_off; j++) {
3378                         struct f2fs_summary *s;
3379                         s = (struct f2fs_summary *)(kaddr + offset);
3380                         seg_i->sum_blk->entries[j] = *s;
3381                         offset += SUMMARY_SIZE;
3382                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3383                                                 SUM_FOOTER_SIZE)
3384                                 continue;
3385
3386                         f2fs_put_page(page, 1);
3387                         page = NULL;
3388
3389                         page = f2fs_get_meta_page(sbi, start++);
3390                         if (IS_ERR(page))
3391                                 return PTR_ERR(page);
3392                         kaddr = (unsigned char *)page_address(page);
3393                         offset = 0;
3394                 }
3395         }
3396         f2fs_put_page(page, 1);
3397         return 0;
3398 }
3399
3400 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3401 {
3402         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3403         struct f2fs_summary_block *sum;
3404         struct curseg_info *curseg;
3405         struct page *new;
3406         unsigned short blk_off;
3407         unsigned int segno = 0;
3408         block_t blk_addr = 0;
3409         int err = 0;
3410
3411         /* get segment number and block addr */
3412         if (IS_DATASEG(type)) {
3413                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3414                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3415                                                         CURSEG_HOT_DATA]);
3416                 if (__exist_node_summaries(sbi))
3417                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3418                 else
3419                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3420         } else {
3421                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3422                                                         CURSEG_HOT_NODE]);
3423                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3424                                                         CURSEG_HOT_NODE]);
3425                 if (__exist_node_summaries(sbi))
3426                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3427                                                         type - CURSEG_HOT_NODE);
3428                 else
3429                         blk_addr = GET_SUM_BLOCK(sbi, segno);
3430         }
3431
3432         new = f2fs_get_meta_page(sbi, blk_addr);
3433         if (IS_ERR(new))
3434                 return PTR_ERR(new);
3435         sum = (struct f2fs_summary_block *)page_address(new);
3436
3437         if (IS_NODESEG(type)) {
3438                 if (__exist_node_summaries(sbi)) {
3439                         struct f2fs_summary *ns = &sum->entries[0];
3440                         int i;
3441                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3442                                 ns->version = 0;
3443                                 ns->ofs_in_node = 0;
3444                         }
3445                 } else {
3446                         err = f2fs_restore_node_summary(sbi, segno, sum);
3447                         if (err)
3448                                 goto out;
3449                 }
3450         }
3451
3452         /* set uncompleted segment to curseg */
3453         curseg = CURSEG_I(sbi, type);
3454         mutex_lock(&curseg->curseg_mutex);
3455
3456         /* update journal info */
3457         down_write(&curseg->journal_rwsem);
3458         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3459         up_write(&curseg->journal_rwsem);
3460
3461         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3462         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3463         curseg->next_segno = segno;
3464         reset_curseg(sbi, type, 0);
3465         curseg->alloc_type = ckpt->alloc_type[type];
3466         curseg->next_blkoff = blk_off;
3467         mutex_unlock(&curseg->curseg_mutex);
3468 out:
3469         f2fs_put_page(new, 1);
3470         return err;
3471 }
3472
3473 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3474 {
3475         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3476         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3477         int type = CURSEG_HOT_DATA;
3478         int err;
3479
3480         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3481                 int npages = f2fs_npages_for_summary_flush(sbi, true);
3482
3483                 if (npages >= 2)
3484                         f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3485                                                         META_CP, true);
3486
3487                 /* restore for compacted data summary */
3488                 err = read_compacted_summaries(sbi);
3489                 if (err)
3490                         return err;
3491                 type = CURSEG_HOT_NODE;
3492         }
3493
3494         if (__exist_node_summaries(sbi))
3495                 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3496                                         NR_CURSEG_TYPE - type, META_CP, true);
3497
3498         for (; type <= CURSEG_COLD_NODE; type++) {
3499                 err = read_normal_summaries(sbi, type);
3500                 if (err)
3501                         return err;
3502         }
3503
3504         /* sanity check for summary blocks */
3505         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3506                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3507                 return -EINVAL;
3508
3509         return 0;
3510 }
3511
3512 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3513 {
3514         struct page *page;
3515         unsigned char *kaddr;
3516         struct f2fs_summary *summary;
3517         struct curseg_info *seg_i;
3518         int written_size = 0;
3519         int i, j;
3520
3521         page = f2fs_grab_meta_page(sbi, blkaddr++);
3522         kaddr = (unsigned char *)page_address(page);
3523         memset(kaddr, 0, PAGE_SIZE);
3524
3525         /* Step 1: write nat cache */
3526         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3527         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3528         written_size += SUM_JOURNAL_SIZE;
3529
3530         /* Step 2: write sit cache */
3531         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3532         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3533         written_size += SUM_JOURNAL_SIZE;
3534
3535         /* Step 3: write summary entries */
3536         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3537                 unsigned short blkoff;
3538                 seg_i = CURSEG_I(sbi, i);
3539                 if (sbi->ckpt->alloc_type[i] == SSR)
3540                         blkoff = sbi->blocks_per_seg;
3541                 else
3542                         blkoff = curseg_blkoff(sbi, i);
3543
3544                 for (j = 0; j < blkoff; j++) {
3545                         if (!page) {
3546                                 page = f2fs_grab_meta_page(sbi, blkaddr++);
3547                                 kaddr = (unsigned char *)page_address(page);
3548                                 memset(kaddr, 0, PAGE_SIZE);
3549                                 written_size = 0;
3550                         }
3551                         summary = (struct f2fs_summary *)(kaddr + written_size);
3552                         *summary = seg_i->sum_blk->entries[j];
3553                         written_size += SUMMARY_SIZE;
3554
3555                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3556                                                         SUM_FOOTER_SIZE)
3557                                 continue;
3558
3559                         set_page_dirty(page);
3560                         f2fs_put_page(page, 1);
3561                         page = NULL;
3562                 }
3563         }
3564         if (page) {
3565                 set_page_dirty(page);
3566                 f2fs_put_page(page, 1);
3567         }
3568 }
3569
3570 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3571                                         block_t blkaddr, int type)
3572 {
3573         int i, end;
3574         if (IS_DATASEG(type))
3575                 end = type + NR_CURSEG_DATA_TYPE;
3576         else
3577                 end = type + NR_CURSEG_NODE_TYPE;
3578
3579         for (i = type; i < end; i++)
3580                 write_current_sum_page(sbi, i, blkaddr + (i - type));
3581 }
3582
3583 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3584 {
3585         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3586                 write_compacted_summaries(sbi, start_blk);
3587         else
3588                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3589 }
3590
3591 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3592 {
3593         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3594 }
3595
3596 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3597                                         unsigned int val, int alloc)
3598 {
3599         int i;
3600
3601         if (type == NAT_JOURNAL) {
3602                 for (i = 0; i < nats_in_cursum(journal); i++) {
3603                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3604                                 return i;
3605                 }
3606                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3607                         return update_nats_in_cursum(journal, 1);
3608         } else if (type == SIT_JOURNAL) {
3609                 for (i = 0; i < sits_in_cursum(journal); i++)
3610                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3611                                 return i;
3612                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3613                         return update_sits_in_cursum(journal, 1);
3614         }
3615         return -1;
3616 }
3617
3618 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3619                                         unsigned int segno)
3620 {
3621         return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
3622 }
3623
3624 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3625                                         unsigned int start)
3626 {
3627         struct sit_info *sit_i = SIT_I(sbi);
3628         struct page *page;
3629         pgoff_t src_off, dst_off;
3630
3631         src_off = current_sit_addr(sbi, start);
3632         dst_off = next_sit_addr(sbi, src_off);
3633
3634         page = f2fs_grab_meta_page(sbi, dst_off);
3635         seg_info_to_sit_page(sbi, page, start);
3636
3637         set_page_dirty(page);
3638         set_to_next_sit(sit_i, start);
3639
3640         return page;
3641 }
3642
3643 static struct sit_entry_set *grab_sit_entry_set(void)
3644 {
3645         struct sit_entry_set *ses =
3646                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3647
3648         ses->entry_cnt = 0;
3649         INIT_LIST_HEAD(&ses->set_list);
3650         return ses;
3651 }
3652
3653 static void release_sit_entry_set(struct sit_entry_set *ses)
3654 {
3655         list_del(&ses->set_list);
3656         kmem_cache_free(sit_entry_set_slab, ses);
3657 }
3658
3659 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3660                                                 struct list_head *head)
3661 {
3662         struct sit_entry_set *next = ses;
3663
3664         if (list_is_last(&ses->set_list, head))
3665                 return;
3666
3667         list_for_each_entry_continue(next, head, set_list)
3668                 if (ses->entry_cnt <= next->entry_cnt)
3669                         break;
3670
3671         list_move_tail(&ses->set_list, &next->set_list);
3672 }
3673
3674 static void add_sit_entry(unsigned int segno, struct list_head *head)
3675 {
3676         struct sit_entry_set *ses;
3677         unsigned int start_segno = START_SEGNO(segno);
3678
3679         list_for_each_entry(ses, head, set_list) {
3680                 if (ses->start_segno == start_segno) {
3681                         ses->entry_cnt++;
3682                         adjust_sit_entry_set(ses, head);
3683                         return;
3684                 }
3685         }
3686
3687         ses = grab_sit_entry_set();
3688
3689         ses->start_segno = start_segno;
3690         ses->entry_cnt++;
3691         list_add(&ses->set_list, head);
3692 }
3693
3694 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3695 {
3696         struct f2fs_sm_info *sm_info = SM_I(sbi);
3697         struct list_head *set_list = &sm_info->sit_entry_set;
3698         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3699         unsigned int segno;
3700
3701         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3702                 add_sit_entry(segno, set_list);
3703 }
3704
3705 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3706 {
3707         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3708         struct f2fs_journal *journal = curseg->journal;
3709         int i;
3710
3711         down_write(&curseg->journal_rwsem);
3712         for (i = 0; i < sits_in_cursum(journal); i++) {
3713                 unsigned int segno;
3714                 bool dirtied;
3715
3716                 segno = le32_to_cpu(segno_in_journal(journal, i));
3717                 dirtied = __mark_sit_entry_dirty(sbi, segno);
3718
3719                 if (!dirtied)
3720                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3721         }
3722         update_sits_in_cursum(journal, -i);
3723         up_write(&curseg->journal_rwsem);
3724 }
3725
3726 /*
3727  * CP calls this function, which flushes SIT entries including sit_journal,
3728  * and moves prefree segs to free segs.
3729  */
3730 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3731 {
3732         struct sit_info *sit_i = SIT_I(sbi);
3733         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3734         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3735         struct f2fs_journal *journal = curseg->journal;
3736         struct sit_entry_set *ses, *tmp;
3737         struct list_head *head = &SM_I(sbi)->sit_entry_set;
3738         bool to_journal = true;
3739         struct seg_entry *se;
3740
3741         down_write(&sit_i->sentry_lock);
3742
3743         if (!sit_i->dirty_sentries)
3744                 goto out;
3745
3746         /*
3747          * add and account sit entries of dirty bitmap in sit entry
3748          * set temporarily
3749          */
3750         add_sits_in_set(sbi);
3751
3752         /*
3753          * if there are no enough space in journal to store dirty sit
3754          * entries, remove all entries from journal and add and account
3755          * them in sit entry set.
3756          */
3757         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3758                 remove_sits_in_journal(sbi);
3759
3760         /*
3761          * there are two steps to flush sit entries:
3762          * #1, flush sit entries to journal in current cold data summary block.
3763          * #2, flush sit entries to sit page.
3764          */
3765         list_for_each_entry_safe(ses, tmp, head, set_list) {
3766                 struct page *page = NULL;
3767                 struct f2fs_sit_block *raw_sit = NULL;
3768                 unsigned int start_segno = ses->start_segno;
3769                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3770                                                 (unsigned long)MAIN_SEGS(sbi));
3771                 unsigned int segno = start_segno;
3772
3773                 if (to_journal &&
3774                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3775                         to_journal = false;
3776
3777                 if (to_journal) {
3778                         down_write(&curseg->journal_rwsem);
3779                 } else {
3780                         page = get_next_sit_page(sbi, start_segno);
3781                         raw_sit = page_address(page);
3782                 }
3783
3784                 /* flush dirty sit entries in region of current sit set */
3785                 for_each_set_bit_from(segno, bitmap, end) {
3786                         int offset, sit_offset;
3787
3788                         se = get_seg_entry(sbi, segno);
3789 #ifdef CONFIG_F2FS_CHECK_FS
3790                         if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3791                                                 SIT_VBLOCK_MAP_SIZE))
3792                                 f2fs_bug_on(sbi, 1);
3793 #endif
3794
3795                         /* add discard candidates */
3796                         if (!(cpc->reason & CP_DISCARD)) {
3797                                 cpc->trim_start = segno;
3798                                 add_discard_addrs(sbi, cpc, false);
3799                         }
3800
3801                         if (to_journal) {
3802                                 offset = f2fs_lookup_journal_in_cursum(journal,
3803                                                         SIT_JOURNAL, segno, 1);
3804                                 f2fs_bug_on(sbi, offset < 0);
3805                                 segno_in_journal(journal, offset) =
3806                                                         cpu_to_le32(segno);
3807                                 seg_info_to_raw_sit(se,
3808                                         &sit_in_journal(journal, offset));
3809                                 check_block_count(sbi, segno,
3810                                         &sit_in_journal(journal, offset));
3811                         } else {
3812                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3813                                 seg_info_to_raw_sit(se,
3814                                                 &raw_sit->entries[sit_offset]);
3815                                 check_block_count(sbi, segno,
3816                                                 &raw_sit->entries[sit_offset]);
3817                         }
3818
3819                         __clear_bit(segno, bitmap);
3820                         sit_i->dirty_sentries--;
3821                         ses->entry_cnt--;
3822                 }
3823
3824                 if (to_journal)
3825                         up_write(&curseg->journal_rwsem);
3826                 else
3827                         f2fs_put_page(page, 1);
3828
3829                 f2fs_bug_on(sbi, ses->entry_cnt);
3830                 release_sit_entry_set(ses);
3831         }
3832
3833         f2fs_bug_on(sbi, !list_empty(head));
3834         f2fs_bug_on(sbi, sit_i->dirty_sentries);
3835 out:
3836         if (cpc->reason & CP_DISCARD) {
3837                 __u64 trim_start = cpc->trim_start;
3838
3839                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3840                         add_discard_addrs(sbi, cpc, false);
3841
3842                 cpc->trim_start = trim_start;
3843         }
3844         up_write(&sit_i->sentry_lock);
3845
3846         set_prefree_as_free_segments(sbi);
3847 }
3848
3849 static int build_sit_info(struct f2fs_sb_info *sbi)
3850 {
3851         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3852         struct sit_info *sit_i;
3853         unsigned int sit_segs, start;
3854         char *src_bitmap;
3855         unsigned int bitmap_size;
3856
3857         /* allocate memory for SIT information */
3858         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3859         if (!sit_i)
3860                 return -ENOMEM;
3861
3862         SM_I(sbi)->sit_info = sit_i;
3863
3864         sit_i->sentries =
3865                 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3866                                               MAIN_SEGS(sbi)),
3867                               GFP_KERNEL);
3868         if (!sit_i->sentries)
3869                 return -ENOMEM;
3870
3871         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3872         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3873                                                                 GFP_KERNEL);
3874         if (!sit_i->dirty_sentries_bitmap)
3875                 return -ENOMEM;
3876
3877         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3878                 sit_i->sentries[start].cur_valid_map
3879                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3880                 sit_i->sentries[start].ckpt_valid_map
3881                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3882                 if (!sit_i->sentries[start].cur_valid_map ||
3883                                 !sit_i->sentries[start].ckpt_valid_map)
3884                         return -ENOMEM;
3885
3886 #ifdef CONFIG_F2FS_CHECK_FS
3887                 sit_i->sentries[start].cur_valid_map_mir
3888                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3889                 if (!sit_i->sentries[start].cur_valid_map_mir)
3890                         return -ENOMEM;
3891 #endif
3892
3893                 sit_i->sentries[start].discard_map
3894                         = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3895                                                         GFP_KERNEL);
3896                 if (!sit_i->sentries[start].discard_map)
3897                         return -ENOMEM;
3898         }
3899
3900         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3901         if (!sit_i->tmp_map)
3902                 return -ENOMEM;
3903
3904         if (__is_large_section(sbi)) {
3905                 sit_i->sec_entries =
3906                         f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
3907                                                       MAIN_SECS(sbi)),
3908                                       GFP_KERNEL);
3909                 if (!sit_i->sec_entries)
3910                         return -ENOMEM;
3911         }
3912
3913         /* get information related with SIT */
3914         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3915
3916         /* setup SIT bitmap from ckeckpoint pack */
3917         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3918         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3919
3920         sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3921         if (!sit_i->sit_bitmap)
3922                 return -ENOMEM;
3923
3924 #ifdef CONFIG_F2FS_CHECK_FS
3925         sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3926         if (!sit_i->sit_bitmap_mir)
3927                 return -ENOMEM;
3928 #endif
3929
3930         /* init SIT information */
3931         sit_i->s_ops = &default_salloc_ops;
3932
3933         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3934         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
3935         sit_i->written_valid_blocks = 0;
3936         sit_i->bitmap_size = bitmap_size;
3937         sit_i->dirty_sentries = 0;
3938         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3939         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3940         sit_i->mounted_time = ktime_get_real_seconds();
3941         init_rwsem(&sit_i->sentry_lock);
3942         return 0;
3943 }
3944
3945 static int build_free_segmap(struct f2fs_sb_info *sbi)
3946 {
3947         struct free_segmap_info *free_i;
3948         unsigned int bitmap_size, sec_bitmap_size;
3949
3950         /* allocate memory for free segmap information */
3951         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
3952         if (!free_i)
3953                 return -ENOMEM;
3954
3955         SM_I(sbi)->free_info = free_i;
3956
3957         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3958         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
3959         if (!free_i->free_segmap)
3960                 return -ENOMEM;
3961
3962         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3963         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
3964         if (!free_i->free_secmap)
3965                 return -ENOMEM;
3966
3967         /* set all segments as dirty temporarily */
3968         memset(free_i->free_segmap, 0xff, bitmap_size);
3969         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3970
3971         /* init free segmap information */
3972         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
3973         free_i->free_segments = 0;
3974         free_i->free_sections = 0;
3975         spin_lock_init(&free_i->segmap_lock);
3976         return 0;
3977 }
3978
3979 static int build_curseg(struct f2fs_sb_info *sbi)
3980 {
3981         struct curseg_info *array;
3982         int i;
3983
3984         array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
3985                              GFP_KERNEL);
3986         if (!array)
3987                 return -ENOMEM;
3988
3989         SM_I(sbi)->curseg_array = array;
3990
3991         for (i = 0; i < NR_CURSEG_TYPE; i++) {
3992                 mutex_init(&array[i].curseg_mutex);
3993                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
3994                 if (!array[i].sum_blk)
3995                         return -ENOMEM;
3996                 init_rwsem(&array[i].journal_rwsem);
3997                 array[i].journal = f2fs_kzalloc(sbi,
3998                                 sizeof(struct f2fs_journal), GFP_KERNEL);
3999                 if (!array[i].journal)
4000                         return -ENOMEM;
4001                 array[i].segno = NULL_SEGNO;
4002                 array[i].next_blkoff = 0;
4003         }
4004         return restore_curseg_summaries(sbi);
4005 }
4006
4007 static int build_sit_entries(struct f2fs_sb_info *sbi)
4008 {
4009         struct sit_info *sit_i = SIT_I(sbi);
4010         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4011         struct f2fs_journal *journal = curseg->journal;
4012         struct seg_entry *se;
4013         struct f2fs_sit_entry sit;
4014         int sit_blk_cnt = SIT_BLK_CNT(sbi);
4015         unsigned int i, start, end;
4016         unsigned int readed, start_blk = 0;
4017         int err = 0;
4018         block_t total_node_blocks = 0;
4019
4020         do {
4021                 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
4022                                                         META_SIT, true);
4023
4024                 start = start_blk * sit_i->sents_per_block;
4025                 end = (start_blk + readed) * sit_i->sents_per_block;
4026
4027                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4028                         struct f2fs_sit_block *sit_blk;
4029                         struct page *page;
4030
4031                         se = &sit_i->sentries[start];
4032                         page = get_current_sit_page(sbi, start);
4033                         if (IS_ERR(page))
4034                                 return PTR_ERR(page);
4035                         sit_blk = (struct f2fs_sit_block *)page_address(page);
4036                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4037                         f2fs_put_page(page, 1);
4038
4039                         err = check_block_count(sbi, start, &sit);
4040                         if (err)
4041                                 return err;
4042                         seg_info_from_raw_sit(se, &sit);
4043                         if (IS_NODESEG(se->type))
4044                                 total_node_blocks += se->valid_blocks;
4045
4046                         /* build discard map only one time */
4047                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4048                                 memset(se->discard_map, 0xff,
4049                                         SIT_VBLOCK_MAP_SIZE);
4050                         } else {
4051                                 memcpy(se->discard_map,
4052                                         se->cur_valid_map,
4053                                         SIT_VBLOCK_MAP_SIZE);
4054                                 sbi->discard_blks +=
4055                                         sbi->blocks_per_seg -
4056                                         se->valid_blocks;
4057                         }
4058
4059                         if (__is_large_section(sbi))
4060                                 get_sec_entry(sbi, start)->valid_blocks +=
4061                                                         se->valid_blocks;
4062                 }
4063                 start_blk += readed;
4064         } while (start_blk < sit_blk_cnt);
4065
4066         down_read(&curseg->journal_rwsem);
4067         for (i = 0; i < sits_in_cursum(journal); i++) {
4068                 unsigned int old_valid_blocks;
4069
4070                 start = le32_to_cpu(segno_in_journal(journal, i));
4071                 if (start >= MAIN_SEGS(sbi)) {
4072                         f2fs_msg(sbi->sb, KERN_ERR,
4073                                         "Wrong journal entry on segno %u",
4074                                         start);
4075                         set_sbi_flag(sbi, SBI_NEED_FSCK);
4076                         err = -EINVAL;
4077                         break;
4078                 }
4079
4080                 se = &sit_i->sentries[start];
4081                 sit = sit_in_journal(journal, i);
4082
4083                 old_valid_blocks = se->valid_blocks;
4084                 if (IS_NODESEG(se->type))
4085                         total_node_blocks -= old_valid_blocks;
4086
4087                 err = check_block_count(sbi, start, &sit);
4088                 if (err)
4089                         break;
4090                 seg_info_from_raw_sit(se, &sit);
4091                 if (IS_NODESEG(se->type))
4092                         total_node_blocks += se->valid_blocks;
4093
4094                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4095                         memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4096                 } else {
4097                         memcpy(se->discard_map, se->cur_valid_map,
4098                                                 SIT_VBLOCK_MAP_SIZE);
4099                         sbi->discard_blks += old_valid_blocks;
4100                         sbi->discard_blks -= se->valid_blocks;
4101                 }
4102
4103                 if (__is_large_section(sbi)) {
4104                         get_sec_entry(sbi, start)->valid_blocks +=
4105                                                         se->valid_blocks;
4106                         get_sec_entry(sbi, start)->valid_blocks -=
4107                                                         old_valid_blocks;
4108                 }
4109         }
4110         up_read(&curseg->journal_rwsem);
4111
4112         if (!err && total_node_blocks != valid_node_count(sbi)) {
4113                 f2fs_msg(sbi->sb, KERN_ERR,
4114                         "SIT is corrupted node# %u vs %u",
4115                         total_node_blocks, valid_node_count(sbi));
4116                 set_sbi_flag(sbi, SBI_NEED_FSCK);
4117                 err = -EINVAL;
4118         }
4119
4120         return err;
4121 }
4122
4123 static void init_free_segmap(struct f2fs_sb_info *sbi)
4124 {
4125         unsigned int start;
4126         int type;
4127
4128         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4129                 struct seg_entry *sentry = get_seg_entry(sbi, start);
4130                 if (!sentry->valid_blocks)
4131                         __set_free(sbi, start);
4132                 else
4133                         SIT_I(sbi)->written_valid_blocks +=
4134                                                 sentry->valid_blocks;
4135         }
4136
4137         /* set use the current segments */
4138         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4139                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4140                 __set_test_and_inuse(sbi, curseg_t->segno);
4141         }
4142 }
4143
4144 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4145 {
4146         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4147         struct free_segmap_info *free_i = FREE_I(sbi);
4148         unsigned int segno = 0, offset = 0;
4149         unsigned short valid_blocks;
4150
4151         while (1) {
4152                 /* find dirty segment based on free segmap */
4153                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4154                 if (segno >= MAIN_SEGS(sbi))
4155                         break;
4156                 offset = segno + 1;
4157                 valid_blocks = get_valid_blocks(sbi, segno, false);
4158                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
4159                         continue;
4160                 if (valid_blocks > sbi->blocks_per_seg) {
4161                         f2fs_bug_on(sbi, 1);
4162                         continue;
4163                 }
4164                 mutex_lock(&dirty_i->seglist_lock);
4165                 __locate_dirty_segment(sbi, segno, DIRTY);
4166                 mutex_unlock(&dirty_i->seglist_lock);
4167         }
4168 }
4169
4170 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4171 {
4172         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4173         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4174
4175         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4176         if (!dirty_i->victim_secmap)
4177                 return -ENOMEM;
4178         return 0;
4179 }
4180
4181 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4182 {
4183         struct dirty_seglist_info *dirty_i;
4184         unsigned int bitmap_size, i;
4185
4186         /* allocate memory for dirty segments list information */
4187         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4188                                                                 GFP_KERNEL);
4189         if (!dirty_i)
4190                 return -ENOMEM;
4191
4192         SM_I(sbi)->dirty_info = dirty_i;
4193         mutex_init(&dirty_i->seglist_lock);
4194
4195         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4196
4197         for (i = 0; i < NR_DIRTY_TYPE; i++) {
4198                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4199                                                                 GFP_KERNEL);
4200                 if (!dirty_i->dirty_segmap[i])
4201                         return -ENOMEM;
4202         }
4203
4204         init_dirty_segmap(sbi);
4205         return init_victim_secmap(sbi);
4206 }
4207
4208 /*
4209  * Update min, max modified time for cost-benefit GC algorithm
4210  */
4211 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4212 {
4213         struct sit_info *sit_i = SIT_I(sbi);
4214         unsigned int segno;
4215
4216         down_write(&sit_i->sentry_lock);
4217
4218         sit_i->min_mtime = ULLONG_MAX;
4219
4220         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4221                 unsigned int i;
4222                 unsigned long long mtime = 0;
4223
4224                 for (i = 0; i < sbi->segs_per_sec; i++)
4225                         mtime += get_seg_entry(sbi, segno + i)->mtime;
4226
4227                 mtime = div_u64(mtime, sbi->segs_per_sec);
4228
4229                 if (sit_i->min_mtime > mtime)
4230                         sit_i->min_mtime = mtime;
4231         }
4232         sit_i->max_mtime = get_mtime(sbi, false);
4233         up_write(&sit_i->sentry_lock);
4234 }
4235
4236 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
4237 {
4238         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4239         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4240         struct f2fs_sm_info *sm_info;
4241         int err;
4242
4243         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
4244         if (!sm_info)
4245                 return -ENOMEM;
4246
4247         /* init sm info */
4248         sbi->sm_info = sm_info;
4249         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4250         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4251         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4252         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4253         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4254         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4255         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
4256         sm_info->rec_prefree_segments = sm_info->main_segments *
4257                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
4258         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4259                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4260
4261         if (!test_opt(sbi, LFS))
4262                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
4263         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
4264         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
4265         sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
4266         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
4267         sm_info->min_ssr_sections = reserved_sections(sbi);
4268
4269         INIT_LIST_HEAD(&sm_info->sit_entry_set);
4270
4271         init_rwsem(&sm_info->curseg_lock);
4272
4273         if (!f2fs_readonly(sbi->sb)) {
4274                 err = f2fs_create_flush_cmd_control(sbi);
4275                 if (err)
4276                         return err;
4277         }
4278
4279         err = create_discard_cmd_control(sbi);
4280         if (err)
4281                 return err;
4282
4283         err = build_sit_info(sbi);
4284         if (err)
4285                 return err;
4286         err = build_free_segmap(sbi);
4287         if (err)
4288                 return err;
4289         err = build_curseg(sbi);
4290         if (err)
4291                 return err;
4292
4293         /* reinit free segmap based on SIT */
4294         err = build_sit_entries(sbi);
4295         if (err)
4296                 return err;
4297
4298         init_free_segmap(sbi);
4299         err = build_dirty_segmap(sbi);
4300         if (err)
4301                 return err;
4302
4303         init_min_max_mtime(sbi);
4304         return 0;
4305 }
4306
4307 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4308                 enum dirty_type dirty_type)
4309 {
4310         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4311
4312         mutex_lock(&dirty_i->seglist_lock);
4313         kvfree(dirty_i->dirty_segmap[dirty_type]);
4314         dirty_i->nr_dirty[dirty_type] = 0;
4315         mutex_unlock(&dirty_i->seglist_lock);
4316 }
4317
4318 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
4319 {
4320         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4321         kvfree(dirty_i->victim_secmap);
4322 }
4323
4324 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4325 {
4326         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4327         int i;
4328
4329         if (!dirty_i)
4330                 return;
4331
4332         /* discard pre-free/dirty segments list */
4333         for (i = 0; i < NR_DIRTY_TYPE; i++)
4334                 discard_dirty_segmap(sbi, i);
4335
4336         destroy_victim_secmap(sbi);
4337         SM_I(sbi)->dirty_info = NULL;
4338         kvfree(dirty_i);
4339 }
4340
4341 static void destroy_curseg(struct f2fs_sb_info *sbi)
4342 {
4343         struct curseg_info *array = SM_I(sbi)->curseg_array;
4344         int i;
4345
4346         if (!array)
4347                 return;
4348         SM_I(sbi)->curseg_array = NULL;
4349         for (i = 0; i < NR_CURSEG_TYPE; i++) {
4350                 kvfree(array[i].sum_blk);
4351                 kvfree(array[i].journal);
4352         }
4353         kvfree(array);
4354 }
4355
4356 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4357 {
4358         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4359         if (!free_i)
4360                 return;
4361         SM_I(sbi)->free_info = NULL;
4362         kvfree(free_i->free_segmap);
4363         kvfree(free_i->free_secmap);
4364         kvfree(free_i);
4365 }
4366
4367 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4368 {
4369         struct sit_info *sit_i = SIT_I(sbi);
4370         unsigned int start;
4371
4372         if (!sit_i)
4373                 return;
4374
4375         if (sit_i->sentries) {
4376                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4377                         kvfree(sit_i->sentries[start].cur_valid_map);
4378 #ifdef CONFIG_F2FS_CHECK_FS
4379                         kvfree(sit_i->sentries[start].cur_valid_map_mir);
4380 #endif
4381                         kvfree(sit_i->sentries[start].ckpt_valid_map);
4382                         kvfree(sit_i->sentries[start].discard_map);
4383                 }
4384         }
4385         kvfree(sit_i->tmp_map);
4386
4387         kvfree(sit_i->sentries);
4388         kvfree(sit_i->sec_entries);
4389         kvfree(sit_i->dirty_sentries_bitmap);
4390
4391         SM_I(sbi)->sit_info = NULL;
4392         kvfree(sit_i->sit_bitmap);
4393 #ifdef CONFIG_F2FS_CHECK_FS
4394         kvfree(sit_i->sit_bitmap_mir);
4395 #endif
4396         kvfree(sit_i);
4397 }
4398
4399 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
4400 {
4401         struct f2fs_sm_info *sm_info = SM_I(sbi);
4402
4403         if (!sm_info)
4404                 return;
4405         f2fs_destroy_flush_cmd_control(sbi, true);
4406         destroy_discard_cmd_control(sbi);
4407         destroy_dirty_segmap(sbi);
4408         destroy_curseg(sbi);
4409         destroy_free_segmap(sbi);
4410         destroy_sit_info(sbi);
4411         sbi->sm_info = NULL;
4412         kvfree(sm_info);
4413 }
4414
4415 int __init f2fs_create_segment_manager_caches(void)
4416 {
4417         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4418                         sizeof(struct discard_entry));
4419         if (!discard_entry_slab)
4420                 goto fail;
4421
4422         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4423                         sizeof(struct discard_cmd));
4424         if (!discard_cmd_slab)
4425                 goto destroy_discard_entry;
4426
4427         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4428                         sizeof(struct sit_entry_set));
4429         if (!sit_entry_set_slab)
4430                 goto destroy_discard_cmd;
4431
4432         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4433                         sizeof(struct inmem_pages));
4434         if (!inmem_entry_slab)
4435                 goto destroy_sit_entry_set;
4436         return 0;
4437
4438 destroy_sit_entry_set:
4439         kmem_cache_destroy(sit_entry_set_slab);
4440 destroy_discard_cmd:
4441         kmem_cache_destroy(discard_cmd_slab);
4442 destroy_discard_entry:
4443         kmem_cache_destroy(discard_entry_slab);
4444 fail:
4445         return -ENOMEM;
4446 }
4447
4448 void f2fs_destroy_segment_manager_caches(void)
4449 {
4450         kmem_cache_destroy(sit_entry_set_slab);
4451         kmem_cache_destroy(discard_cmd_slab);
4452         kmem_cache_destroy(discard_entry_slab);
4453         kmem_cache_destroy(inmem_entry_slab);
4454 }