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 (f2fs_lfs_mode(sbi))
 176                 return false;
 177         if (sbi->gc_mode == GC_URGENT_HIGH)
 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 inmem_pages *new;
 189
 190         f2fs_trace_pid(page);
 191
 192         f2fs_set_page_private(page, ATOMIC_WRITTEN_PAGE);
 193
 194         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 195
 196         /* add atomic page indices to the list */
 197         new->page = page;
 198         INIT_LIST_HEAD(&new->list);
 199
 200         /* increase reference count with clean state */
 201         get_page(page);
 202         mutex_lock(&F2FS_I(inode)->inmem_lock);
 203         list_add_tail(&new->list, &F2FS_I(inode)->inmem_pages);
 204         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 205         mutex_unlock(&F2FS_I(inode)->inmem_lock);
 206
 207         trace_f2fs_register_inmem_page(page, INMEM);
 208 }
 209
 210 static int __revoke_inmem_pages(struct inode *inode,
 211                                 struct list_head *head, bool drop, bool recover,
 212                                 bool trylock)
 213 {
 214         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 215         struct inmem_pages *cur, *tmp;
 216         int err = 0;
 217
 218         list_for_each_entry_safe(cur, tmp, head, list) {
 219                 struct page *page = cur->page;
 220
 221                 if (drop)
 222                         trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 223
 224                 if (trylock) {
 225                         /*
 226                          * to avoid deadlock in between page lock and
 227                          * inmem_lock.
 228                          */
 229                         if (!trylock_page(page))
 230                                 continue;
 231                 } else {
 232                         lock_page(page);
 233                 }
 234
 235                 f2fs_wait_on_page_writeback(page, DATA, true, true);
 236
 237                 if (recover) {
 238                         struct dnode_of_data dn;
 239                         struct node_info ni;
 240
 241                         trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 242 retry:
 243                         set_new_dnode(&dn, inode, NULL, NULL, 0);
 244                         err = f2fs_get_dnode_of_data(&dn, page->index,
 245                                                                 LOOKUP_NODE);
 246                         if (err) {
 247                                 if (err == -ENOMEM) {
 248                                         congestion_wait(BLK_RW_ASYNC,
 249                                                         DEFAULT_IO_TIMEOUT);
 250                                         cond_resched();
 251                                         goto retry;
 252                                 }
 253                                 err = -EAGAIN;
 254                                 goto next;
 255                         }
 256
 257                         err = f2fs_get_node_info(sbi, dn.nid, &ni);
 258                         if (err) {
 259                                 f2fs_put_dnode(&dn);
 260                                 return err;
 261                         }
 262
 263                         if (cur->old_addr == NEW_ADDR) {
 264                                 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
 265                                 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
 266                         } else
 267                                 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 268                                         cur->old_addr, ni.version, true, true);
 269                         f2fs_put_dnode(&dn);
 270                 }
 271 next:
 272                 /* we don't need to invalidate this in the sccessful status */
 273                 if (drop || recover) {
 274                         ClearPageUptodate(page);
 275                         clear_cold_data(page);
 276                 }
 277                 f2fs_clear_page_private(page);
 278                 f2fs_put_page(page, 1);
 279
 280                 list_del(&cur->list);
 281                 kmem_cache_free(inmem_entry_slab, cur);
 282                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 283         }
 284         return err;
 285 }
 286
 287 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
 288 {
 289         struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
 290         struct inode *inode;
 291         struct f2fs_inode_info *fi;
 292         unsigned int count = sbi->atomic_files;
 293         unsigned int looped = 0;
 294 next:
 295         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 296         if (list_empty(head)) {
 297                 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 298                 return;
 299         }
 300         fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
 301         inode = igrab(&fi->vfs_inode);
 302         if (inode)
 303                 list_move_tail(&fi->inmem_ilist, head);
 304         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 305
 306         if (inode) {
 307                 if (gc_failure) {
 308                         if (!fi->i_gc_failures[GC_FAILURE_ATOMIC])
 309                                 goto skip;
 310                 }
 311                 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
 312                 f2fs_drop_inmem_pages(inode);
 313 skip:
 314                 iput(inode);
 315         }
 316         congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
 317         cond_resched();
 318         if (gc_failure) {
 319                 if (++looped >= count)
 320                         return;
 321         }
 322         goto next;
 323 }
 324
 325 void f2fs_drop_inmem_pages(struct inode *inode)
 326 {
 327         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 328         struct f2fs_inode_info *fi = F2FS_I(inode);
 329
 330         while (!list_empty(&fi->inmem_pages)) {
 331                 mutex_lock(&fi->inmem_lock);
 332                 __revoke_inmem_pages(inode, &fi->inmem_pages,
 333                                                 true, false, true);
 334                 mutex_unlock(&fi->inmem_lock);
 335         }
 336
 337         fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
 338
 339         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 340         if (!list_empty(&fi->inmem_ilist))
 341                 list_del_init(&fi->inmem_ilist);
 342         if (f2fs_is_atomic_file(inode)) {
 343                 clear_inode_flag(inode, FI_ATOMIC_FILE);
 344                 sbi->atomic_files--;
 345         }
 346         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 347 }
 348
 349 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
 350 {
 351         struct f2fs_inode_info *fi = F2FS_I(inode);
 352         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 353         struct list_head *head = &fi->inmem_pages;
 354         struct inmem_pages *cur = NULL;
 355
 356         f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
 357
 358         mutex_lock(&fi->inmem_lock);
 359         list_for_each_entry(cur, head, list) {
 360                 if (cur->page == page)
 361                         break;
 362         }
 363
 364         f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
 365         list_del(&cur->list);
 366         mutex_unlock(&fi->inmem_lock);
 367
 368         dec_page_count(sbi, F2FS_INMEM_PAGES);
 369         kmem_cache_free(inmem_entry_slab, cur);
 370
 371         ClearPageUptodate(page);
 372         f2fs_clear_page_private(page);
 373         f2fs_put_page(page, 0);
 374
 375         trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 376 }
 377
 378 static int __f2fs_commit_inmem_pages(struct inode *inode)
 379 {
 380         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 381         struct f2fs_inode_info *fi = F2FS_I(inode);
 382         struct inmem_pages *cur, *tmp;
 383         struct f2fs_io_info fio = {
 384                 .sbi = sbi,
 385                 .ino = inode->i_ino,
 386                 .type = DATA,
 387                 .op = REQ_OP_WRITE,
 388                 .op_flags = REQ_SYNC | REQ_PRIO,
 389                 .io_type = FS_DATA_IO,
 390         };
 391         struct list_head revoke_list;
 392         bool submit_bio = false;
 393         int err = 0;
 394
 395         INIT_LIST_HEAD(&revoke_list);
 396
 397         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 398                 struct page *page = cur->page;
 399
 400                 lock_page(page);
 401                 if (page->mapping == inode->i_mapping) {
 402                         trace_f2fs_commit_inmem_page(page, INMEM);
 403
 404                         f2fs_wait_on_page_writeback(page, DATA, true, true);
 405
 406                         set_page_dirty(page);
 407                         if (clear_page_dirty_for_io(page)) {
 408                                 inode_dec_dirty_pages(inode);
 409                                 f2fs_remove_dirty_inode(inode);
 410                         }
 411 retry:
 412                         fio.page = page;
 413                         fio.old_blkaddr = NULL_ADDR;
 414                         fio.encrypted_page = NULL;
 415                         fio.need_lock = LOCK_DONE;
 416                         err = f2fs_do_write_data_page(&fio);
 417                         if (err) {
 418                                 if (err == -ENOMEM) {
 419                                         congestion_wait(BLK_RW_ASYNC,
 420                                                         DEFAULT_IO_TIMEOUT);
 421                                         cond_resched();
 422                                         goto retry;
 423                                 }
 424                                 unlock_page(page);
 425                                 break;
 426                         }
 427                         /* record old blkaddr for revoking */
 428                         cur->old_addr = fio.old_blkaddr;
 429                         submit_bio = true;
 430                 }
 431                 unlock_page(page);
 432                 list_move_tail(&cur->list, &revoke_list);
 433         }
 434
 435         if (submit_bio)
 436                 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
 437
 438         if (err) {
 439                 /*
 440                  * try to revoke all committed pages, but still we could fail
 441                  * due to no memory or other reason, if that happened, EAGAIN
 442                  * will be returned, which means in such case, transaction is
 443                  * already not integrity, caller should use journal to do the
 444                  * recovery or rewrite & commit last transaction. For other
 445                  * error number, revoking was done by filesystem itself.
 446                  */
 447                 err = __revoke_inmem_pages(inode, &revoke_list,
 448                                                 false, true, false);
 449
 450                 /* drop all uncommitted pages */
 451                 __revoke_inmem_pages(inode, &fi->inmem_pages,
 452                                                 true, false, false);
 453         } else {
 454                 __revoke_inmem_pages(inode, &revoke_list,
 455                                                 false, false, false);
 456         }
 457
 458         return err;
 459 }
 460
 461 int f2fs_commit_inmem_pages(struct inode *inode)
 462 {
 463         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 464         struct f2fs_inode_info *fi = F2FS_I(inode);
 465         int err;
 466
 467         f2fs_balance_fs(sbi, true);
 468
 469         down_write(&fi->i_gc_rwsem[WRITE]);
 470
 471         f2fs_lock_op(sbi);
 472         set_inode_flag(inode, FI_ATOMIC_COMMIT);
 473
 474         mutex_lock(&fi->inmem_lock);
 475         err = __f2fs_commit_inmem_pages(inode);
 476         mutex_unlock(&fi->inmem_lock);
 477
 478         clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 479
 480         f2fs_unlock_op(sbi);
 481         up_write(&fi->i_gc_rwsem[WRITE]);
 482
 483         return err;
 484 }
 485
 486 /*
 487  * This function balances dirty node and dentry pages.
 488  * In addition, it controls garbage collection.
 489  */
 490 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 491 {
 492         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 493                 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
 494                 f2fs_stop_checkpoint(sbi, false);
 495         }
 496
 497         /* balance_fs_bg is able to be pending */
 498         if (need && excess_cached_nats(sbi))
 499                 f2fs_balance_fs_bg(sbi, false);
 500
 501         if (!f2fs_is_checkpoint_ready(sbi))
 502                 return;
 503
 504         /*
 505          * We should do GC or end up with checkpoint, if there are so many dirty
 506          * dir/node pages without enough free segments.
 507          */
 508         if (has_not_enough_free_secs(sbi, 0, 0)) {
 509                 down_write(&sbi->gc_lock);
 510                 f2fs_gc(sbi, false, false, NULL_SEGNO);
 511         }
 512 }
 513
 514 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
 515 {
 516         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 517                 return;
 518
 519         /* try to shrink extent cache when there is no enough memory */
 520         if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
 521                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 522
 523         /* check the # of cached NAT entries */
 524         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
 525                 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 526
 527         if (!f2fs_available_free_memory(sbi, FREE_NIDS))
 528                 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
 529         else
 530                 f2fs_build_free_nids(sbi, false, false);
 531
 532         if (!is_idle(sbi, REQ_TIME) &&
 533                 (!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
 534                 return;
 535
 536         /* checkpoint is the only way to shrink partial cached entries */
 537         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
 538                         !f2fs_available_free_memory(sbi, INO_ENTRIES) ||
 539                         excess_prefree_segs(sbi) ||
 540                         excess_dirty_nats(sbi) ||
 541                         excess_dirty_nodes(sbi) ||
 542                         f2fs_time_over(sbi, CP_TIME)) {
 543                 if (test_opt(sbi, DATA_FLUSH) && from_bg) {
 544                         struct blk_plug plug;
 545
 546                         mutex_lock(&sbi->flush_lock);
 547
 548                         blk_start_plug(&plug);
 549                         f2fs_sync_dirty_inodes(sbi, FILE_INODE);
 550                         blk_finish_plug(&plug);
 551
 552                         mutex_unlock(&sbi->flush_lock);
 553                 }
 554                 f2fs_sync_fs(sbi->sb, true);
 555                 stat_inc_bg_cp_count(sbi->stat_info);
 556         }
 557 }
 558
 559 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 560                                 struct block_device *bdev)
 561 {
 562         struct bio *bio;
 563         int ret;
 564
 565         bio = f2fs_bio_alloc(sbi, 0, false);
 566         if (!bio)
 567                 return -ENOMEM;
 568
 569         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
 570         bio_set_dev(bio, bdev);
 571         ret = submit_bio_wait(bio);
 572         bio_put(bio);
 573
 574         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 575                                 test_opt(sbi, FLUSH_MERGE), ret);
 576         return ret;
 577 }
 578
 579 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 580 {
 581         int ret = 0;
 582         int i;
 583
 584         if (!f2fs_is_multi_device(sbi))
 585                 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 586
 587         for (i = 0; i < sbi->s_ndevs; i++) {
 588                 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
 589                         continue;
 590                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 591                 if (ret)
 592                         break;
 593         }
 594         return ret;
 595 }
 596
 597 static int issue_flush_thread(void *data)
 598 {
 599         struct f2fs_sb_info *sbi = data;
 600         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 601         wait_queue_head_t *q = &fcc->flush_wait_queue;
 602 repeat:
 603         if (kthread_should_stop())
 604                 return 0;
 605
 606         sb_start_intwrite(sbi->sb);
 607
 608         if (!llist_empty(&fcc->issue_list)) {
 609                 struct flush_cmd *cmd, *next;
 610                 int ret;
 611
 612                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 613                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 614
 615                 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 616
 617                 ret = submit_flush_wait(sbi, cmd->ino);
 618                 atomic_inc(&fcc->issued_flush);
 619
 620                 llist_for_each_entry_safe(cmd, next,
 621                                           fcc->dispatch_list, llnode) {
 622                         cmd->ret = ret;
 623                         complete(&cmd->wait);
 624                 }
 625                 fcc->dispatch_list = NULL;
 626         }
 627
 628         sb_end_intwrite(sbi->sb);
 629
 630         wait_event_interruptible(*q,
 631                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 632         goto repeat;
 633 }
 634
 635 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 636 {
 637         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 638         struct flush_cmd cmd;
 639         int ret;
 640
 641         if (test_opt(sbi, NOBARRIER))
 642                 return 0;
 643
 644         if (!test_opt(sbi, FLUSH_MERGE)) {
 645                 atomic_inc(&fcc->queued_flush);
 646                 ret = submit_flush_wait(sbi, ino);
 647                 atomic_dec(&fcc->queued_flush);
 648                 atomic_inc(&fcc->issued_flush);
 649                 return ret;
 650         }
 651
 652         if (atomic_inc_return(&fcc->queued_flush) == 1 ||
 653             f2fs_is_multi_device(sbi)) {
 654                 ret = submit_flush_wait(sbi, ino);
 655                 atomic_dec(&fcc->queued_flush);
 656
 657                 atomic_inc(&fcc->issued_flush);
 658                 return ret;
 659         }
 660
 661         cmd.ino = ino;
 662         init_completion(&cmd.wait);
 663
 664         llist_add(&cmd.llnode, &fcc->issue_list);
 665
 666         /* update issue_list before we wake up issue_flush thread */
 667         smp_mb();
 668
 669         if (waitqueue_active(&fcc->flush_wait_queue))
 670                 wake_up(&fcc->flush_wait_queue);
 671
 672         if (fcc->f2fs_issue_flush) {
 673                 wait_for_completion(&cmd.wait);
 674                 atomic_dec(&fcc->queued_flush);
 675         } else {
 676                 struct llist_node *list;
 677
 678                 list = llist_del_all(&fcc->issue_list);
 679                 if (!list) {
 680                         wait_for_completion(&cmd.wait);
 681                         atomic_dec(&fcc->queued_flush);
 682                 } else {
 683                         struct flush_cmd *tmp, *next;
 684
 685                         ret = submit_flush_wait(sbi, ino);
 686
 687                         llist_for_each_entry_safe(tmp, next, list, llnode) {
 688                                 if (tmp == &cmd) {
 689                                         cmd.ret = ret;
 690                                         atomic_dec(&fcc->queued_flush);
 691                                         continue;
 692                                 }
 693                                 tmp->ret = ret;
 694                                 complete(&tmp->wait);
 695                         }
 696                 }
 697         }
 698
 699         return cmd.ret;
 700 }
 701
 702 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
 703 {
 704         dev_t dev = sbi->sb->s_bdev->bd_dev;
 705         struct flush_cmd_control *fcc;
 706         int err = 0;
 707
 708         if (SM_I(sbi)->fcc_info) {
 709                 fcc = SM_I(sbi)->fcc_info;
 710                 if (fcc->f2fs_issue_flush)
 711                         return err;
 712                 goto init_thread;
 713         }
 714
 715         fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
 716         if (!fcc)
 717                 return -ENOMEM;
 718         atomic_set(&fcc->issued_flush, 0);
 719         atomic_set(&fcc->queued_flush, 0);
 720         init_waitqueue_head(&fcc->flush_wait_queue);
 721         init_llist_head(&fcc->issue_list);
 722         SM_I(sbi)->fcc_info = fcc;
 723         if (!test_opt(sbi, FLUSH_MERGE))
 724                 return err;
 725
 726 init_thread:
 727         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 728                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 729         if (IS_ERR(fcc->f2fs_issue_flush)) {
 730                 err = PTR_ERR(fcc->f2fs_issue_flush);
 731                 kvfree(fcc);
 732                 SM_I(sbi)->fcc_info = NULL;
 733                 return err;
 734         }
 735
 736         return err;
 737 }
 738
 739 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 740 {
 741         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 742
 743         if (fcc && fcc->f2fs_issue_flush) {
 744                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 745
 746                 fcc->f2fs_issue_flush = NULL;
 747                 kthread_stop(flush_thread);
 748         }
 749         if (free) {
 750                 kvfree(fcc);
 751                 SM_I(sbi)->fcc_info = NULL;
 752         }
 753 }
 754
 755 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
 756 {
 757         int ret = 0, i;
 758
 759         if (!f2fs_is_multi_device(sbi))
 760                 return 0;
 761
 762         for (i = 1; i < sbi->s_ndevs; i++) {
 763                 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
 764                         continue;
 765                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 766                 if (ret)
 767                         break;
 768
 769                 spin_lock(&sbi->dev_lock);
 770                 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
 771                 spin_unlock(&sbi->dev_lock);
 772         }
 773
 774         return ret;
 775 }
 776
 777 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 778                 enum dirty_type dirty_type)
 779 {
 780         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 781
 782         /* need not be added */
 783         if (IS_CURSEG(sbi, segno))
 784                 return;
 785
 786         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 787                 dirty_i->nr_dirty[dirty_type]++;
 788
 789         if (dirty_type == DIRTY) {
 790                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 791                 enum dirty_type t = sentry->type;
 792
 793                 if (unlikely(t >= DIRTY)) {
 794                         f2fs_bug_on(sbi, 1);
 795                         return;
 796                 }
 797                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 798                         dirty_i->nr_dirty[t]++;
 799
 800                 if (__is_large_section(sbi)) {
 801                         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 802                         block_t valid_blocks =
 803                                 get_valid_blocks(sbi, segno, true);
 804
 805                         f2fs_bug_on(sbi, unlikely(!valid_blocks ||
 806                                         valid_blocks == BLKS_PER_SEC(sbi)));
 807
 808                         if (!IS_CURSEC(sbi, secno))
 809                                 set_bit(secno, dirty_i->dirty_secmap);
 810                 }
 811         }
 812 }
 813
 814 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 815                 enum dirty_type dirty_type)
 816 {
 817         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 818         block_t valid_blocks;
 819
 820         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 821                 dirty_i->nr_dirty[dirty_type]--;
 822
 823         if (dirty_type == DIRTY) {
 824                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 825                 enum dirty_type t = sentry->type;
 826
 827                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 828                         dirty_i->nr_dirty[t]--;
 829
 830                 valid_blocks = get_valid_blocks(sbi, segno, true);
 831                 if (valid_blocks == 0) {
 832                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 833                                                 dirty_i->victim_secmap);
 834 #ifdef CONFIG_F2FS_CHECK_FS
 835                         clear_bit(segno, SIT_I(sbi)->invalid_segmap);
 836 #endif
 837                 }
 838                 if (__is_large_section(sbi)) {
 839                         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 840
 841                         if (!valid_blocks ||
 842                                         valid_blocks == BLKS_PER_SEC(sbi)) {
 843                                 clear_bit(secno, dirty_i->dirty_secmap);
 844                                 return;
 845                         }
 846
 847                         if (!IS_CURSEC(sbi, secno))
 848                                 set_bit(secno, dirty_i->dirty_secmap);
 849                 }
 850         }
 851 }
 852
 853 /*
 854  * Should not occur error such as -ENOMEM.
 855  * Adding dirty entry into seglist is not critical operation.
 856  * If a given segment is one of current working segments, it won't be added.
 857  */
 858 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 859 {
 860         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 861         unsigned short valid_blocks, ckpt_valid_blocks;
 862         unsigned int usable_blocks;
 863
 864         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 865                 return;
 866
 867         usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
 868         mutex_lock(&dirty_i->seglist_lock);
 869
 870         valid_blocks = get_valid_blocks(sbi, segno, false);
 871         ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno);
 872
 873         if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
 874                 ckpt_valid_blocks == usable_blocks)) {
 875                 __locate_dirty_segment(sbi, segno, PRE);
 876                 __remove_dirty_segment(sbi, segno, DIRTY);
 877         } else if (valid_blocks < usable_blocks) {
 878                 __locate_dirty_segment(sbi, segno, DIRTY);
 879         } else {
 880                 /* Recovery routine with SSR needs this */
 881                 __remove_dirty_segment(sbi, segno, DIRTY);
 882         }
 883
 884         mutex_unlock(&dirty_i->seglist_lock);
 885 }
 886
 887 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
 888 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
 889 {
 890         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 891         unsigned int segno;
 892
 893         mutex_lock(&dirty_i->seglist_lock);
 894         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 895                 if (get_valid_blocks(sbi, segno, false))
 896                         continue;
 897                 if (IS_CURSEG(sbi, segno))
 898                         continue;
 899                 __locate_dirty_segment(sbi, segno, PRE);
 900                 __remove_dirty_segment(sbi, segno, DIRTY);
 901         }
 902         mutex_unlock(&dirty_i->seglist_lock);
 903 }
 904
 905 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
 906 {
 907         int ovp_hole_segs =
 908                 (overprovision_segments(sbi) - reserved_segments(sbi));
 909         block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
 910         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 911         block_t holes[2] = {0, 0};      /* DATA and NODE */
 912         block_t unusable;
 913         struct seg_entry *se;
 914         unsigned int segno;
 915
 916         mutex_lock(&dirty_i->seglist_lock);
 917         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 918                 se = get_seg_entry(sbi, segno);
 919                 if (IS_NODESEG(se->type))
 920                         holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
 921                                                         se->valid_blocks;
 922                 else
 923                         holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
 924                                                         se->valid_blocks;
 925         }
 926         mutex_unlock(&dirty_i->seglist_lock);
 927
 928         unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
 929         if (unusable > ovp_holes)
 930                 return unusable - ovp_holes;
 931         return 0;
 932 }
 933
 934 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
 935 {
 936         int ovp_hole_segs =
 937                 (overprovision_segments(sbi) - reserved_segments(sbi));
 938         if (unusable > F2FS_OPTION(sbi).unusable_cap)
 939                 return -EAGAIN;
 940         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
 941                 dirty_segments(sbi) > ovp_hole_segs)
 942                 return -EAGAIN;
 943         return 0;
 944 }
 945
 946 /* This is only used by SBI_CP_DISABLED */
 947 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
 948 {
 949         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 950         unsigned int segno = 0;
 951
 952         mutex_lock(&dirty_i->seglist_lock);
 953         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 954                 if (get_valid_blocks(sbi, segno, false))
 955                         continue;
 956                 if (get_ckpt_valid_blocks(sbi, segno))
 957                         continue;
 958                 mutex_unlock(&dirty_i->seglist_lock);
 959                 return segno;
 960         }
 961         mutex_unlock(&dirty_i->seglist_lock);
 962         return NULL_SEGNO;
 963 }
 964
 965 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 966                 struct block_device *bdev, block_t lstart,
 967                 block_t start, block_t len)
 968 {
 969         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 970         struct list_head *pend_list;
 971         struct discard_cmd *dc;
 972
 973         f2fs_bug_on(sbi, !len);
 974
 975         pend_list = &dcc->pend_list[plist_idx(len)];
 976
 977         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 978         INIT_LIST_HEAD(&dc->list);
 979         dc->bdev = bdev;
 980         dc->lstart = lstart;
 981         dc->start = start;
 982         dc->len = len;
 983         dc->ref = 0;
 984         dc->state = D_PREP;
 985         dc->queued = 0;
 986         dc->error = 0;
 987         init_completion(&dc->wait);
 988         list_add_tail(&dc->list, pend_list);
 989         spin_lock_init(&dc->lock);
 990         dc->bio_ref = 0;
 991         atomic_inc(&dcc->discard_cmd_cnt);
 992         dcc->undiscard_blks += len;
 993
 994         return dc;
 995 }
 996
 997 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 998                                 struct block_device *bdev, block_t lstart,
 999                                 block_t start, block_t len,
1000                                 struct rb_node *parent, struct rb_node **p,
1001                                 bool leftmost)
1002 {
1003         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1004         struct discard_cmd *dc;
1005
1006         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1007
1008         rb_link_node(&dc->rb_node, parent, p);
1009         rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
1010
1011         return dc;
1012 }
1013
1014 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1015                                                         struct discard_cmd *dc)
1016 {
1017         if (dc->state == D_DONE)
1018                 atomic_sub(dc->queued, &dcc->queued_discard);
1019
1020         list_del(&dc->list);
1021         rb_erase_cached(&dc->rb_node, &dcc->root);
1022         dcc->undiscard_blks -= dc->len;
1023
1024         kmem_cache_free(discard_cmd_slab, dc);
1025
1026         atomic_dec(&dcc->discard_cmd_cnt);
1027 }
1028
1029 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1030                                                         struct discard_cmd *dc)
1031 {
1032         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1033         unsigned long flags;
1034
1035         trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
1036
1037         spin_lock_irqsave(&dc->lock, flags);
1038         if (dc->bio_ref) {
1039                 spin_unlock_irqrestore(&dc->lock, flags);
1040                 return;
1041         }
1042         spin_unlock_irqrestore(&dc->lock, flags);
1043
1044         f2fs_bug_on(sbi, dc->ref);
1045
1046         if (dc->error == -EOPNOTSUPP)
1047                 dc->error = 0;
1048
1049         if (dc->error)
1050                 printk_ratelimited(
1051                         "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
1052                         KERN_INFO, sbi->sb->s_id,
1053                         dc->lstart, dc->start, dc->len, dc->error);
1054         __detach_discard_cmd(dcc, dc);
1055 }
1056
1057 static void f2fs_submit_discard_endio(struct bio *bio)
1058 {
1059         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1060         unsigned long flags;
1061
1062         spin_lock_irqsave(&dc->lock, flags);
1063         if (!dc->error)
1064                 dc->error = blk_status_to_errno(bio->bi_status);
1065         dc->bio_ref--;
1066         if (!dc->bio_ref && dc->state == D_SUBMIT) {
1067                 dc->state = D_DONE;
1068                 complete_all(&dc->wait);
1069         }
1070         spin_unlock_irqrestore(&dc->lock, flags);
1071         bio_put(bio);
1072 }
1073
1074 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1075                                 block_t start, block_t end)
1076 {
1077 #ifdef CONFIG_F2FS_CHECK_FS
1078         struct seg_entry *sentry;
1079         unsigned int segno;
1080         block_t blk = start;
1081         unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1082         unsigned long *map;
1083
1084         while (blk < end) {
1085                 segno = GET_SEGNO(sbi, blk);
1086                 sentry = get_seg_entry(sbi, segno);
1087                 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1088
1089                 if (end < START_BLOCK(sbi, segno + 1))
1090                         size = GET_BLKOFF_FROM_SEG0(sbi, end);
1091                 else
1092                         size = max_blocks;
1093                 map = (unsigned long *)(sentry->cur_valid_map);
1094                 offset = __find_rev_next_bit(map, size, offset);
1095                 f2fs_bug_on(sbi, offset != size);
1096                 blk = START_BLOCK(sbi, segno + 1);
1097         }
1098 #endif
1099 }
1100
1101 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1102                                 struct discard_policy *dpolicy,
1103                                 int discard_type, unsigned int granularity)
1104 {
1105         /* common policy */
1106         dpolicy->type = discard_type;
1107         dpolicy->sync = true;
1108         dpolicy->ordered = false;
1109         dpolicy->granularity = granularity;
1110
1111         dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1112         dpolicy->io_aware_gran = MAX_PLIST_NUM;
1113         dpolicy->timeout = false;
1114
1115         if (discard_type == DPOLICY_BG) {
1116                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1117                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1118                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1119                 dpolicy->io_aware = true;
1120                 dpolicy->sync = false;
1121                 dpolicy->ordered = true;
1122                 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1123                         dpolicy->granularity = 1;
1124                         dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1125                 }
1126         } else if (discard_type == DPOLICY_FORCE) {
1127                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1128                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1129                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1130                 dpolicy->io_aware = false;
1131         } else if (discard_type == DPOLICY_FSTRIM) {
1132                 dpolicy->io_aware = false;
1133         } else if (discard_type == DPOLICY_UMOUNT) {
1134                 dpolicy->io_aware = false;
1135                 /* we need to issue all to keep CP_TRIMMED_FLAG */
1136                 dpolicy->granularity = 1;
1137                 dpolicy->timeout = true;
1138         }
1139 }
1140
1141 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1142                                 struct block_device *bdev, block_t lstart,
1143                                 block_t start, block_t len);
1144 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1145 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1146                                                 struct discard_policy *dpolicy,
1147                                                 struct discard_cmd *dc,
1148                                                 unsigned int *issued)
1149 {
1150         struct block_device *bdev = dc->bdev;
1151         struct request_queue *q = bdev_get_queue(bdev);
1152         unsigned int max_discard_blocks =
1153                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1154         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1155         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1156                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1157         int flag = dpolicy->sync ? REQ_SYNC : 0;
1158         block_t lstart, start, len, total_len;
1159         int err = 0;
1160
1161         if (dc->state != D_PREP)
1162                 return 0;
1163
1164         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1165                 return 0;
1166
1167         trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1168
1169         lstart = dc->lstart;
1170         start = dc->start;
1171         len = dc->len;
1172         total_len = len;
1173
1174         dc->len = 0;
1175
1176         while (total_len && *issued < dpolicy->max_requests && !err) {
1177                 struct bio *bio = NULL;
1178                 unsigned long flags;
1179                 bool last = true;
1180
1181                 if (len > max_discard_blocks) {
1182                         len = max_discard_blocks;
1183                         last = false;
1184                 }
1185
1186                 (*issued)++;
1187                 if (*issued == dpolicy->max_requests)
1188                         last = true;
1189
1190                 dc->len += len;
1191
1192                 if (time_to_inject(sbi, FAULT_DISCARD)) {
1193                         f2fs_show_injection_info(sbi, FAULT_DISCARD);
1194                         err = -EIO;
1195                         goto submit;
1196                 }
1197                 err = __blkdev_issue_discard(bdev,
1198                                         SECTOR_FROM_BLOCK(start),
1199                                         SECTOR_FROM_BLOCK(len),
1200                                         GFP_NOFS, 0, &bio);
1201 submit:
1202                 if (err) {
1203                         spin_lock_irqsave(&dc->lock, flags);
1204                         if (dc->state == D_PARTIAL)
1205                                 dc->state = D_SUBMIT;
1206                         spin_unlock_irqrestore(&dc->lock, flags);
1207
1208                         break;
1209                 }
1210
1211                 f2fs_bug_on(sbi, !bio);
1212
1213                 /*
1214                  * should keep before submission to avoid D_DONE
1215                  * right away
1216                  */
1217                 spin_lock_irqsave(&dc->lock, flags);
1218                 if (last)
1219                         dc->state = D_SUBMIT;
1220                 else
1221                         dc->state = D_PARTIAL;
1222                 dc->bio_ref++;
1223                 spin_unlock_irqrestore(&dc->lock, flags);
1224
1225                 atomic_inc(&dcc->queued_discard);
1226                 dc->queued++;
1227                 list_move_tail(&dc->list, wait_list);
1228
1229                 /* sanity check on discard range */
1230                 __check_sit_bitmap(sbi, lstart, lstart + len);
1231
1232                 bio->bi_private = dc;
1233                 bio->bi_end_io = f2fs_submit_discard_endio;
1234                 bio->bi_opf |= flag;
1235                 submit_bio(bio);
1236
1237                 atomic_inc(&dcc->issued_discard);
1238
1239                 f2fs_update_iostat(sbi, FS_DISCARD, 1);
1240
1241                 lstart += len;
1242                 start += len;
1243                 total_len -= len;
1244                 len = total_len;
1245         }
1246
1247         if (!err && len) {
1248                 dcc->undiscard_blks -= len;
1249                 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1250         }
1251         return err;
1252 }
1253
1254 static void __insert_discard_tree(struct f2fs_sb_info *sbi,
1255                                 struct block_device *bdev, block_t lstart,
1256                                 block_t start, block_t len,
1257                                 struct rb_node **insert_p,
1258                                 struct rb_node *insert_parent)
1259 {
1260         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1261         struct rb_node **p;
1262         struct rb_node *parent = NULL;
1263         bool leftmost = true;
1264
1265         if (insert_p && insert_parent) {
1266                 parent = insert_parent;
1267                 p = insert_p;
1268                 goto do_insert;
1269         }
1270
1271         p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1272                                                         lstart, &leftmost);
1273 do_insert:
1274         __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1275                                                                 p, leftmost);
1276 }
1277
1278 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1279                                                 struct discard_cmd *dc)
1280 {
1281         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1282 }
1283
1284 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1285                                 struct discard_cmd *dc, block_t blkaddr)
1286 {
1287         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1288         struct discard_info di = dc->di;
1289         bool modified = false;
1290
1291         if (dc->state == D_DONE || dc->len == 1) {
1292                 __remove_discard_cmd(sbi, dc);
1293                 return;
1294         }
1295
1296         dcc->undiscard_blks -= di.len;
1297
1298         if (blkaddr > di.lstart) {
1299                 dc->len = blkaddr - dc->lstart;
1300                 dcc->undiscard_blks += dc->len;
1301                 __relocate_discard_cmd(dcc, dc);
1302                 modified = true;
1303         }
1304
1305         if (blkaddr < di.lstart + di.len - 1) {
1306                 if (modified) {
1307                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1308                                         di.start + blkaddr + 1 - di.lstart,
1309                                         di.lstart + di.len - 1 - blkaddr,
1310                                         NULL, NULL);
1311                 } else {
1312                         dc->lstart++;
1313                         dc->len--;
1314                         dc->start++;
1315                         dcc->undiscard_blks += dc->len;
1316                         __relocate_discard_cmd(dcc, dc);
1317                 }
1318         }
1319 }
1320
1321 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1322                                 struct block_device *bdev, block_t lstart,
1323                                 block_t start, block_t len)
1324 {
1325         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1326         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1327         struct discard_cmd *dc;
1328         struct discard_info di = {0};
1329         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1330         struct request_queue *q = bdev_get_queue(bdev);
1331         unsigned int max_discard_blocks =
1332                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1333         block_t end = lstart + len;
1334
1335         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1336                                         NULL, lstart,
1337                                         (struct rb_entry **)&prev_dc,
1338                                         (struct rb_entry **)&next_dc,
1339                                         &insert_p, &insert_parent, true, NULL);
1340         if (dc)
1341                 prev_dc = dc;
1342
1343         if (!prev_dc) {
1344                 di.lstart = lstart;
1345                 di.len = next_dc ? next_dc->lstart - lstart : len;
1346                 di.len = min(di.len, len);
1347                 di.start = start;
1348         }
1349
1350         while (1) {
1351                 struct rb_node *node;
1352                 bool merged = false;
1353                 struct discard_cmd *tdc = NULL;
1354
1355                 if (prev_dc) {
1356                         di.lstart = prev_dc->lstart + prev_dc->len;
1357                         if (di.lstart < lstart)
1358                                 di.lstart = lstart;
1359                         if (di.lstart >= end)
1360                                 break;
1361
1362                         if (!next_dc || next_dc->lstart > end)
1363                                 di.len = end - di.lstart;
1364                         else
1365                                 di.len = next_dc->lstart - di.lstart;
1366                         di.start = start + di.lstart - lstart;
1367                 }
1368
1369                 if (!di.len)
1370                         goto next;
1371
1372                 if (prev_dc && prev_dc->state == D_PREP &&
1373                         prev_dc->bdev == bdev &&
1374                         __is_discard_back_mergeable(&di, &prev_dc->di,
1375                                                         max_discard_blocks)) {
1376                         prev_dc->di.len += di.len;
1377                         dcc->undiscard_blks += di.len;
1378                         __relocate_discard_cmd(dcc, prev_dc);
1379                         di = prev_dc->di;
1380                         tdc = prev_dc;
1381                         merged = true;
1382                 }
1383
1384                 if (next_dc && next_dc->state == D_PREP &&
1385                         next_dc->bdev == bdev &&
1386                         __is_discard_front_mergeable(&di, &next_dc->di,
1387                                                         max_discard_blocks)) {
1388                         next_dc->di.lstart = di.lstart;
1389                         next_dc->di.len += di.len;
1390                         next_dc->di.start = di.start;
1391                         dcc->undiscard_blks += di.len;
1392                         __relocate_discard_cmd(dcc, next_dc);
1393                         if (tdc)
1394                                 __remove_discard_cmd(sbi, tdc);
1395                         merged = true;
1396                 }
1397
1398                 if (!merged) {
1399                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1400                                                         di.len, NULL, NULL);
1401                 }
1402  next:
1403                 prev_dc = next_dc;
1404                 if (!prev_dc)
1405                         break;
1406
1407                 node = rb_next(&prev_dc->rb_node);
1408                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1409         }
1410 }
1411
1412 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1413                 struct block_device *bdev, block_t blkstart, block_t blklen)
1414 {
1415         block_t lblkstart = blkstart;
1416
1417         if (!f2fs_bdev_support_discard(bdev))
1418                 return 0;
1419
1420         trace_f2fs_queue_discard(bdev, blkstart, blklen);
1421
1422         if (f2fs_is_multi_device(sbi)) {
1423                 int devi = f2fs_target_device_index(sbi, blkstart);
1424
1425                 blkstart -= FDEV(devi).start_blk;
1426         }
1427         mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1428         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1429         mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1430         return 0;
1431 }
1432
1433 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1434                                         struct discard_policy *dpolicy)
1435 {
1436         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1437         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1438         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1439         struct discard_cmd *dc;
1440         struct blk_plug plug;
1441         unsigned int pos = dcc->next_pos;
1442         unsigned int issued = 0;
1443         bool io_interrupted = false;
1444
1445         mutex_lock(&dcc->cmd_lock);
1446         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1447                                         NULL, pos,
1448                                         (struct rb_entry **)&prev_dc,
1449                                         (struct rb_entry **)&next_dc,
1450                                         &insert_p, &insert_parent, true, NULL);
1451         if (!dc)
1452                 dc = next_dc;
1453
1454         blk_start_plug(&plug);
1455
1456         while (dc) {
1457                 struct rb_node *node;
1458                 int err = 0;
1459
1460                 if (dc->state != D_PREP)
1461                         goto next;
1462
1463                 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1464                         io_interrupted = true;
1465                         break;
1466                 }
1467
1468                 dcc->next_pos = dc->lstart + dc->len;
1469                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1470
1471                 if (issued >= dpolicy->max_requests)
1472                         break;
1473 next:
1474                 node = rb_next(&dc->rb_node);
1475                 if (err)
1476                         __remove_discard_cmd(sbi, dc);
1477                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1478         }
1479
1480         blk_finish_plug(&plug);
1481
1482         if (!dc)
1483                 dcc->next_pos = 0;
1484
1485         mutex_unlock(&dcc->cmd_lock);
1486
1487         if (!issued && io_interrupted)
1488                 issued = -1;
1489
1490         return issued;
1491 }
1492 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1493                                         struct discard_policy *dpolicy);
1494
1495 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1496                                         struct discard_policy *dpolicy)
1497 {
1498         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1499         struct list_head *pend_list;
1500         struct discard_cmd *dc, *tmp;
1501         struct blk_plug plug;
1502         int i, issued;
1503         bool io_interrupted = false;
1504
1505         if (dpolicy->timeout)
1506                 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1507
1508 retry:
1509         issued = 0;
1510         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1511                 if (dpolicy->timeout &&
1512                                 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1513                         break;
1514
1515                 if (i + 1 < dpolicy->granularity)
1516                         break;
1517
1518                 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1519                         return __issue_discard_cmd_orderly(sbi, dpolicy);
1520
1521                 pend_list = &dcc->pend_list[i];
1522
1523                 mutex_lock(&dcc->cmd_lock);
1524                 if (list_empty(pend_list))
1525                         goto next;
1526                 if (unlikely(dcc->rbtree_check))
1527                         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1528                                                         &dcc->root, false));
1529                 blk_start_plug(&plug);
1530                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1531                         f2fs_bug_on(sbi, dc->state != D_PREP);
1532
1533                         if (dpolicy->timeout &&
1534                                 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1535                                 break;
1536
1537                         if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1538                                                 !is_idle(sbi, DISCARD_TIME)) {
1539                                 io_interrupted = true;
1540                                 break;
1541                         }
1542
1543                         __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1544
1545                         if (issued >= dpolicy->max_requests)
1546                                 break;
1547                 }
1548                 blk_finish_plug(&plug);
1549 next:
1550                 mutex_unlock(&dcc->cmd_lock);
1551
1552                 if (issued >= dpolicy->max_requests || io_interrupted)
1553                         break;
1554         }
1555
1556         if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1557                 __wait_all_discard_cmd(sbi, dpolicy);
1558                 goto retry;
1559         }
1560
1561         if (!issued && io_interrupted)
1562                 issued = -1;
1563
1564         return issued;
1565 }
1566
1567 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1568 {
1569         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1570         struct list_head *pend_list;
1571         struct discard_cmd *dc, *tmp;
1572         int i;
1573         bool dropped = false;
1574
1575         mutex_lock(&dcc->cmd_lock);
1576         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1577                 pend_list = &dcc->pend_list[i];
1578                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1579                         f2fs_bug_on(sbi, dc->state != D_PREP);
1580                         __remove_discard_cmd(sbi, dc);
1581                         dropped = true;
1582                 }
1583         }
1584         mutex_unlock(&dcc->cmd_lock);
1585
1586         return dropped;
1587 }
1588
1589 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1590 {
1591         __drop_discard_cmd(sbi);
1592 }
1593
1594 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1595                                                         struct discard_cmd *dc)
1596 {
1597         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1598         unsigned int len = 0;
1599
1600         wait_for_completion_io(&dc->wait);
1601         mutex_lock(&dcc->cmd_lock);
1602         f2fs_bug_on(sbi, dc->state != D_DONE);
1603         dc->ref--;
1604         if (!dc->ref) {
1605                 if (!dc->error)
1606                         len = dc->len;
1607                 __remove_discard_cmd(sbi, dc);
1608         }
1609         mutex_unlock(&dcc->cmd_lock);
1610
1611         return len;
1612 }
1613
1614 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1615                                                 struct discard_policy *dpolicy,
1616                                                 block_t start, block_t end)
1617 {
1618         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1619         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1620                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1621         struct discard_cmd *dc, *tmp;
1622         bool need_wait;
1623         unsigned int trimmed = 0;
1624
1625 next:
1626         need_wait = false;
1627
1628         mutex_lock(&dcc->cmd_lock);
1629         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1630                 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1631                         continue;
1632                 if (dc->len < dpolicy->granularity)
1633                         continue;
1634                 if (dc->state == D_DONE && !dc->ref) {
1635                         wait_for_completion_io(&dc->wait);
1636                         if (!dc->error)
1637                                 trimmed += dc->len;
1638                         __remove_discard_cmd(sbi, dc);
1639                 } else {
1640                         dc->ref++;
1641                         need_wait = true;
1642                         break;
1643                 }
1644         }
1645         mutex_unlock(&dcc->cmd_lock);
1646
1647         if (need_wait) {
1648                 trimmed += __wait_one_discard_bio(sbi, dc);
1649                 goto next;
1650         }
1651
1652         return trimmed;
1653 }
1654
1655 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1656                                                 struct discard_policy *dpolicy)
1657 {
1658         struct discard_policy dp;
1659         unsigned int discard_blks;
1660
1661         if (dpolicy)
1662                 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1663
1664         /* wait all */
1665         __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1666         discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1667         __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1668         discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1669
1670         return discard_blks;
1671 }
1672
1673 /* This should be covered by global mutex, &sit_i->sentry_lock */
1674 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1675 {
1676         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1677         struct discard_cmd *dc;
1678         bool need_wait = false;
1679
1680         mutex_lock(&dcc->cmd_lock);
1681         dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1682                                                         NULL, blkaddr);
1683         if (dc) {
1684                 if (dc->state == D_PREP) {
1685                         __punch_discard_cmd(sbi, dc, blkaddr);
1686                 } else {
1687                         dc->ref++;
1688                         need_wait = true;
1689                 }
1690         }
1691         mutex_unlock(&dcc->cmd_lock);
1692
1693         if (need_wait)
1694                 __wait_one_discard_bio(sbi, dc);
1695 }
1696
1697 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1698 {
1699         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1700
1701         if (dcc && dcc->f2fs_issue_discard) {
1702                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1703
1704                 dcc->f2fs_issue_discard = NULL;
1705                 kthread_stop(discard_thread);
1706         }
1707 }
1708
1709 /* This comes from f2fs_put_super */
1710 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1711 {
1712         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1713         struct discard_policy dpolicy;
1714         bool dropped;
1715
1716         __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1717                                         dcc->discard_granularity);
1718         __issue_discard_cmd(sbi, &dpolicy);
1719         dropped = __drop_discard_cmd(sbi);
1720
1721         /* just to make sure there is no pending discard commands */
1722         __wait_all_discard_cmd(sbi, NULL);
1723
1724         f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1725         return dropped;
1726 }
1727
1728 static int issue_discard_thread(void *data)
1729 {
1730         struct f2fs_sb_info *sbi = data;
1731         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1732         wait_queue_head_t *q = &dcc->discard_wait_queue;
1733         struct discard_policy dpolicy;
1734         unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1735         int issued;
1736
1737         set_freezable();
1738
1739         do {
1740                 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1741                                         dcc->discard_granularity);
1742
1743                 wait_event_interruptible_timeout(*q,
1744                                 kthread_should_stop() || freezing(current) ||
1745                                 dcc->discard_wake,
1746                                 msecs_to_jiffies(wait_ms));
1747
1748                 if (dcc->discard_wake)
1749                         dcc->discard_wake = 0;
1750
1751                 /* clean up pending candidates before going to sleep */
1752                 if (atomic_read(&dcc->queued_discard))
1753                         __wait_all_discard_cmd(sbi, NULL);
1754
1755                 if (try_to_freeze())
1756                         continue;
1757                 if (f2fs_readonly(sbi->sb))
1758                         continue;
1759                 if (kthread_should_stop())
1760                         return 0;
1761                 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1762                         wait_ms = dpolicy.max_interval;
1763                         continue;
1764                 }
1765
1766                 if (sbi->gc_mode == GC_URGENT_HIGH)
1767                         __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1768
1769                 sb_start_intwrite(sbi->sb);
1770
1771                 issued = __issue_discard_cmd(sbi, &dpolicy);
1772                 if (issued > 0) {
1773                         __wait_all_discard_cmd(sbi, &dpolicy);
1774                         wait_ms = dpolicy.min_interval;
1775                 } else if (issued == -1){
1776                         wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1777                         if (!wait_ms)
1778                                 wait_ms = dpolicy.mid_interval;
1779                 } else {
1780                         wait_ms = dpolicy.max_interval;
1781                 }
1782
1783                 sb_end_intwrite(sbi->sb);
1784
1785         } while (!kthread_should_stop());
1786         return 0;
1787 }
1788
1789 #ifdef CONFIG_BLK_DEV_ZONED
1790 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1791                 struct block_device *bdev, block_t blkstart, block_t blklen)
1792 {
1793         sector_t sector, nr_sects;
1794         block_t lblkstart = blkstart;
1795         int devi = 0;
1796
1797         if (f2fs_is_multi_device(sbi)) {
1798                 devi = f2fs_target_device_index(sbi, blkstart);
1799                 if (blkstart < FDEV(devi).start_blk ||
1800                     blkstart > FDEV(devi).end_blk) {
1801                         f2fs_err(sbi, "Invalid block %x", blkstart);
1802                         return -EIO;
1803                 }
1804                 blkstart -= FDEV(devi).start_blk;
1805         }
1806
1807         /* For sequential zones, reset the zone write pointer */
1808         if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1809                 sector = SECTOR_FROM_BLOCK(blkstart);
1810                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1811
1812                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1813                                 nr_sects != bdev_zone_sectors(bdev)) {
1814                         f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1815                                  devi, sbi->s_ndevs ? FDEV(devi).path : "",
1816                                  blkstart, blklen);
1817                         return -EIO;
1818                 }
1819                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1820                 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1821                                         sector, nr_sects, GFP_NOFS);
1822         }
1823
1824         /* For conventional zones, use regular discard if supported */
1825         return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1826 }
1827 #endif
1828
1829 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1830                 struct block_device *bdev, block_t blkstart, block_t blklen)
1831 {
1832 #ifdef CONFIG_BLK_DEV_ZONED
1833         if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1834                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1835 #endif
1836         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1837 }
1838
1839 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1840                                 block_t blkstart, block_t blklen)
1841 {
1842         sector_t start = blkstart, len = 0;
1843         struct block_device *bdev;
1844         struct seg_entry *se;
1845         unsigned int offset;
1846         block_t i;
1847         int err = 0;
1848
1849         bdev = f2fs_target_device(sbi, blkstart, NULL);
1850
1851         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1852                 if (i != start) {
1853                         struct block_device *bdev2 =
1854                                 f2fs_target_device(sbi, i, NULL);
1855
1856                         if (bdev2 != bdev) {
1857                                 err = __issue_discard_async(sbi, bdev,
1858                                                 start, len);
1859                                 if (err)
1860                                         return err;
1861                                 bdev = bdev2;
1862                                 start = i;
1863                                 len = 0;
1864                         }
1865                 }
1866
1867                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1868                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1869
1870                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1871                         sbi->discard_blks--;
1872         }
1873
1874         if (len)
1875                 err = __issue_discard_async(sbi, bdev, start, len);
1876         return err;
1877 }
1878
1879 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1880                                                         bool check_only)
1881 {
1882         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1883         int max_blocks = sbi->blocks_per_seg;
1884         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1885         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1886         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1887         unsigned long *discard_map = (unsigned long *)se->discard_map;
1888         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1889         unsigned int start = 0, end = -1;
1890         bool force = (cpc->reason & CP_DISCARD);
1891         struct discard_entry *de = NULL;
1892         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1893         int i;
1894
1895         if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
1896                 return false;
1897
1898         if (!force) {
1899                 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1900                         SM_I(sbi)->dcc_info->nr_discards >=
1901                                 SM_I(sbi)->dcc_info->max_discards)
1902                         return false;
1903         }
1904
1905         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1906         for (i = 0; i < entries; i++)
1907                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1908                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1909
1910         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1911                                 SM_I(sbi)->dcc_info->max_discards) {
1912                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1913                 if (start >= max_blocks)
1914                         break;
1915
1916                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1917                 if (force && start && end != max_blocks
1918                                         && (end - start) < cpc->trim_minlen)
1919                         continue;
1920
1921                 if (check_only)
1922                         return true;
1923
1924                 if (!de) {
1925                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1926                                                                 GFP_F2FS_ZERO);
1927                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1928                         list_add_tail(&de->list, head);
1929                 }
1930
1931                 for (i = start; i < end; i++)
1932                         __set_bit_le(i, (void *)de->discard_map);
1933
1934                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1935         }
1936         return false;
1937 }
1938
1939 static void release_discard_addr(struct discard_entry *entry)
1940 {
1941         list_del(&entry->list);
1942         kmem_cache_free(discard_entry_slab, entry);
1943 }
1944
1945 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1946 {
1947         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1948         struct discard_entry *entry, *this;
1949
1950         /* drop caches */
1951         list_for_each_entry_safe(entry, this, head, list)
1952                 release_discard_addr(entry);
1953 }
1954
1955 /*
1956  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1957  */
1958 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1959 {
1960         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1961         unsigned int segno;
1962
1963         mutex_lock(&dirty_i->seglist_lock);
1964         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1965                 __set_test_and_free(sbi, segno, false);
1966         mutex_unlock(&dirty_i->seglist_lock);
1967 }
1968
1969 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1970                                                 struct cp_control *cpc)
1971 {
1972         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1973         struct list_head *head = &dcc->entry_list;
1974         struct discard_entry *entry, *this;
1975         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1976         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1977         unsigned int start = 0, end = -1;
1978         unsigned int secno, start_segno;
1979         bool force = (cpc->reason & CP_DISCARD);
1980         bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
1981
1982         mutex_lock(&dirty_i->seglist_lock);
1983
1984         while (1) {
1985                 int i;
1986
1987                 if (need_align && end != -1)
1988                         end--;
1989                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1990                 if (start >= MAIN_SEGS(sbi))
1991                         break;
1992                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1993                                                                 start + 1);
1994
1995                 if (need_align) {
1996                         start = rounddown(start, sbi->segs_per_sec);
1997                         end = roundup(end, sbi->segs_per_sec);
1998                 }
1999
2000                 for (i = start; i < end; i++) {
2001                         if (test_and_clear_bit(i, prefree_map))
2002                                 dirty_i->nr_dirty[PRE]--;
2003                 }
2004
2005                 if (!f2fs_realtime_discard_enable(sbi))
2006                         continue;
2007
2008                 if (force && start >= cpc->trim_start &&
2009                                         (end - 1) <= cpc->trim_end)
2010                                 continue;
2011
2012                 if (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi)) {
2013                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2014                                 (end - start) << sbi->log_blocks_per_seg);
2015                         continue;
2016                 }
2017 next:
2018                 secno = GET_SEC_FROM_SEG(sbi, start);
2019                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
2020                 if (!IS_CURSEC(sbi, secno) &&
2021                         !get_valid_blocks(sbi, start, true))
2022                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2023                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
2024
2025                 start = start_segno + sbi->segs_per_sec;
2026                 if (start < end)
2027                         goto next;
2028                 else
2029                         end = start - 1;
2030         }
2031         mutex_unlock(&dirty_i->seglist_lock);
2032
2033         /* send small discards */
2034         list_for_each_entry_safe(entry, this, head, list) {
2035                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2036                 bool is_valid = test_bit_le(0, entry->discard_map);
2037
2038 find_next:
2039                 if (is_valid) {
2040                         next_pos = find_next_zero_bit_le(entry->discard_map,
2041                                         sbi->blocks_per_seg, cur_pos);
2042                         len = next_pos - cur_pos;
2043
2044                         if (f2fs_sb_has_blkzoned(sbi) ||
2045                             (force && len < cpc->trim_minlen))
2046                                 goto skip;
2047
2048                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2049                                                                         len);
2050                         total_len += len;
2051                 } else {
2052                         next_pos = find_next_bit_le(entry->discard_map,
2053                                         sbi->blocks_per_seg, cur_pos);
2054                 }
2055 skip:
2056                 cur_pos = next_pos;
2057                 is_valid = !is_valid;
2058
2059                 if (cur_pos < sbi->blocks_per_seg)
2060                         goto find_next;
2061
2062                 release_discard_addr(entry);
2063                 dcc->nr_discards -= total_len;
2064         }
2065
2066         wake_up_discard_thread(sbi, false);
2067 }
2068
2069 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2070 {
2071         dev_t dev = sbi->sb->s_bdev->bd_dev;
2072         struct discard_cmd_control *dcc;
2073         int err = 0, i;
2074
2075         if (SM_I(sbi)->dcc_info) {
2076                 dcc = SM_I(sbi)->dcc_info;
2077                 goto init_thread;
2078         }
2079
2080         dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2081         if (!dcc)
2082                 return -ENOMEM;
2083
2084         dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2085         INIT_LIST_HEAD(&dcc->entry_list);
2086         for (i = 0; i < MAX_PLIST_NUM; i++)
2087                 INIT_LIST_HEAD(&dcc->pend_list[i]);
2088         INIT_LIST_HEAD(&dcc->wait_list);
2089         INIT_LIST_HEAD(&dcc->fstrim_list);
2090         mutex_init(&dcc->cmd_lock);
2091         atomic_set(&dcc->issued_discard, 0);
2092         atomic_set(&dcc->queued_discard, 0);
2093         atomic_set(&dcc->discard_cmd_cnt, 0);
2094         dcc->nr_discards = 0;
2095         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2096         dcc->undiscard_blks = 0;
2097         dcc->next_pos = 0;
2098         dcc->root = RB_ROOT_CACHED;
2099         dcc->rbtree_check = false;
2100
2101         init_waitqueue_head(&dcc->discard_wait_queue);
2102         SM_I(sbi)->dcc_info = dcc;
2103 init_thread:
2104         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2105                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2106         if (IS_ERR(dcc->f2fs_issue_discard)) {
2107                 err = PTR_ERR(dcc->f2fs_issue_discard);
2108                 kvfree(dcc);
2109                 SM_I(sbi)->dcc_info = NULL;
2110                 return err;
2111         }
2112
2113         return err;
2114 }
2115
2116 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2117 {
2118         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2119
2120         if (!dcc)
2121                 return;
2122
2123         f2fs_stop_discard_thread(sbi);
2124
2125         /*
2126          * Recovery can cache discard commands, so in error path of
2127          * fill_super(), it needs to give a chance to handle them.
2128          */
2129         if (unlikely(atomic_read(&dcc->discard_cmd_cnt)))
2130                 f2fs_issue_discard_timeout(sbi);
2131
2132         kvfree(dcc);
2133         SM_I(sbi)->dcc_info = NULL;
2134 }
2135
2136 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2137 {
2138         struct sit_info *sit_i = SIT_I(sbi);
2139
2140         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2141                 sit_i->dirty_sentries++;
2142                 return false;
2143         }
2144
2145         return true;
2146 }
2147
2148 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2149                                         unsigned int segno, int modified)
2150 {
2151         struct seg_entry *se = get_seg_entry(sbi, segno);
2152         se->type = type;
2153         if (modified)
2154                 __mark_sit_entry_dirty(sbi, segno);
2155 }
2156
2157 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2158                                                                 block_t blkaddr)
2159 {
2160         unsigned int segno = GET_SEGNO(sbi, blkaddr);
2161
2162         if (segno == NULL_SEGNO)
2163                 return 0;
2164         return get_seg_entry(sbi, segno)->mtime;
2165 }
2166
2167 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2168                                                 unsigned long long old_mtime)
2169 {
2170         struct seg_entry *se;
2171         unsigned int segno = GET_SEGNO(sbi, blkaddr);
2172         unsigned long long ctime = get_mtime(sbi, false);
2173         unsigned long long mtime = old_mtime ? old_mtime : ctime;
2174
2175         if (segno == NULL_SEGNO)
2176                 return;
2177
2178         se = get_seg_entry(sbi, segno);
2179
2180         if (!se->mtime)
2181                 se->mtime = mtime;
2182         else
2183                 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2184                                                 se->valid_blocks + 1);
2185
2186         if (ctime > SIT_I(sbi)->max_mtime)
2187                 SIT_I(sbi)->max_mtime = ctime;
2188 }
2189
2190 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2191 {
2192         struct seg_entry *se;
2193         unsigned int segno, offset;
2194         long int new_vblocks;
2195         bool exist;
2196 #ifdef CONFIG_F2FS_CHECK_FS
2197         bool mir_exist;
2198 #endif
2199
2200         segno = GET_SEGNO(sbi, blkaddr);
2201
2202         se = get_seg_entry(sbi, segno);
2203         new_vblocks = se->valid_blocks + del;
2204         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2205
2206         f2fs_bug_on(sbi, (new_vblocks < 0 ||
2207                         (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2208
2209         se->valid_blocks = new_vblocks;
2210
2211         /* Update valid block bitmap */
2212         if (del > 0) {
2213                 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2214 #ifdef CONFIG_F2FS_CHECK_FS
2215                 mir_exist = f2fs_test_and_set_bit(offset,
2216                                                 se->cur_valid_map_mir);
2217                 if (unlikely(exist != mir_exist)) {
2218                         f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2219                                  blkaddr, exist);
2220                         f2fs_bug_on(sbi, 1);
2221                 }
2222 #endif
2223                 if (unlikely(exist)) {
2224                         f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2225                                  blkaddr);
2226                         f2fs_bug_on(sbi, 1);
2227                         se->valid_blocks--;
2228                         del = 0;
2229                 }
2230
2231                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
2232                         sbi->discard_blks--;
2233
2234                 /*
2235                  * SSR should never reuse block which is checkpointed
2236                  * or newly invalidated.
2237                  */
2238                 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2239                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2240                                 se->ckpt_valid_blocks++;
2241                 }
2242         } else {
2243                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2244 #ifdef CONFIG_F2FS_CHECK_FS
2245                 mir_exist = f2fs_test_and_clear_bit(offset,
2246                                                 se->cur_valid_map_mir);
2247                 if (unlikely(exist != mir_exist)) {
2248                         f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2249                                  blkaddr, exist);
2250                         f2fs_bug_on(sbi, 1);
2251                 }
2252 #endif
2253                 if (unlikely(!exist)) {
2254                         f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2255                                  blkaddr);
2256                         f2fs_bug_on(sbi, 1);
2257                         se->valid_blocks++;
2258                         del = 0;
2259                 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2260                         /*
2261                          * If checkpoints are off, we must not reuse data that
2262                          * was used in the previous checkpoint. If it was used
2263                          * before, we must track that to know how much space we
2264                          * really have.
2265                          */
2266                         if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2267                                 spin_lock(&sbi->stat_lock);
2268                                 sbi->unusable_block_count++;
2269                                 spin_unlock(&sbi->stat_lock);
2270                         }
2271                 }
2272
2273                 if (f2fs_test_and_clear_bit(offset, se->discard_map))
2274                         sbi->discard_blks++;
2275         }
2276         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2277                 se->ckpt_valid_blocks += del;
2278
2279         __mark_sit_entry_dirty(sbi, segno);
2280
2281         /* update total number of valid blocks to be written in ckpt area */
2282         SIT_I(sbi)->written_valid_blocks += del;
2283
2284         if (__is_large_section(sbi))
2285                 get_sec_entry(sbi, segno)->valid_blocks += del;
2286 }
2287
2288 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2289 {
2290         unsigned int segno = GET_SEGNO(sbi, addr);
2291         struct sit_info *sit_i = SIT_I(sbi);
2292
2293         f2fs_bug_on(sbi, addr == NULL_ADDR);
2294         if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2295                 return;
2296
2297         invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2298
2299         /* add it into sit main buffer */
2300         down_write(&sit_i->sentry_lock);
2301
2302         update_segment_mtime(sbi, addr, 0);
2303         update_sit_entry(sbi, addr, -1);
2304
2305         /* add it into dirty seglist */
2306         locate_dirty_segment(sbi, segno);
2307
2308         up_write(&sit_i->sentry_lock);
2309 }
2310
2311 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2312 {
2313         struct sit_info *sit_i = SIT_I(sbi);
2314         unsigned int segno, offset;
2315         struct seg_entry *se;
2316         bool is_cp = false;
2317
2318         if (!__is_valid_data_blkaddr(blkaddr))
2319                 return true;
2320
2321         down_read(&sit_i->sentry_lock);
2322
2323         segno = GET_SEGNO(sbi, blkaddr);
2324         se = get_seg_entry(sbi, segno);
2325         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2326
2327         if (f2fs_test_bit(offset, se->ckpt_valid_map))
2328                 is_cp = true;
2329
2330         up_read(&sit_i->sentry_lock);
2331
2332         return is_cp;
2333 }
2334
2335 /*
2336  * This function should be resided under the curseg_mutex lock
2337  */
2338 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2339                                         struct f2fs_summary *sum)
2340 {
2341         struct curseg_info *curseg = CURSEG_I(sbi, type);
2342         void *addr = curseg->sum_blk;
2343         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2344         memcpy(addr, sum, sizeof(struct f2fs_summary));
2345 }
2346
2347 /*
2348  * Calculate the number of current summary pages for writing
2349  */
2350 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2351 {
2352         int valid_sum_count = 0;
2353         int i, sum_in_page;
2354
2355         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2356                 if (sbi->ckpt->alloc_type[i] == SSR)
2357                         valid_sum_count += sbi->blocks_per_seg;
2358                 else {
2359                         if (for_ra)
2360                                 valid_sum_count += le16_to_cpu(
2361                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2362                         else
2363                                 valid_sum_count += curseg_blkoff(sbi, i);
2364                 }
2365         }
2366
2367         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2368                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2369         if (valid_sum_count <= sum_in_page)
2370                 return 1;
2371         else if ((valid_sum_count - sum_in_page) <=
2372                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2373                 return 2;
2374         return 3;
2375 }
2376
2377 /*
2378  * Caller should put this summary page
2379  */
2380 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2381 {
2382         return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
2383 }
2384
2385 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2386                                         void *src, block_t blk_addr)
2387 {
2388         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2389
2390         memcpy(page_address(page), src, PAGE_SIZE);
2391         set_page_dirty(page);
2392         f2fs_put_page(page, 1);
2393 }
2394
2395 static void write_sum_page(struct f2fs_sb_info *sbi,
2396                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
2397 {
2398         f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2399 }
2400
2401 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2402                                                 int type, block_t blk_addr)
2403 {
2404         struct curseg_info *curseg = CURSEG_I(sbi, type);
2405         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2406         struct f2fs_summary_block *src = curseg->sum_blk;
2407         struct f2fs_summary_block *dst;
2408
2409         dst = (struct f2fs_summary_block *)page_address(page);
2410         memset(dst, 0, PAGE_SIZE);
2411
2412         mutex_lock(&curseg->curseg_mutex);
2413
2414         down_read(&curseg->journal_rwsem);
2415         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2416         up_read(&curseg->journal_rwsem);
2417
2418         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2419         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2420
2421         mutex_unlock(&curseg->curseg_mutex);
2422
2423         set_page_dirty(page);
2424         f2fs_put_page(page, 1);
2425 }
2426
2427 static int is_next_segment_free(struct f2fs_sb_info *sbi,
2428                                 struct curseg_info *curseg, int type)
2429 {
2430         unsigned int segno = curseg->segno + 1;
2431         struct free_segmap_info *free_i = FREE_I(sbi);
2432
2433         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2434                 return !test_bit(segno, free_i->free_segmap);
2435         return 0;
2436 }
2437
2438 /*
2439  * Find a new segment from the free segments bitmap to right order
2440  * This function should be returned with success, otherwise BUG
2441  */
2442 static void get_new_segment(struct f2fs_sb_info *sbi,
2443                         unsigned int *newseg, bool new_sec, int dir)
2444 {
2445         struct free_segmap_info *free_i = FREE_I(sbi);
2446         unsigned int segno, secno, zoneno;
2447         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2448         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2449         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2450         unsigned int left_start = hint;
2451         bool init = true;
2452         int go_left = 0;
2453         int i;
2454
2455         spin_lock(&free_i->segmap_lock);
2456
2457         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2458                 segno = find_next_zero_bit(free_i->free_segmap,
2459                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2460                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2461                         goto got_it;
2462         }
2463 find_other_zone:
2464         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2465         if (secno >= MAIN_SECS(sbi)) {
2466                 if (dir == ALLOC_RIGHT) {
2467                         secno = find_next_zero_bit(free_i->free_secmap,
2468                                                         MAIN_SECS(sbi), 0);
2469                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2470                 } else {
2471                         go_left = 1;
2472                         left_start = hint - 1;
2473                 }
2474         }
2475         if (go_left == 0)
2476                 goto skip_left;
2477
2478         while (test_bit(left_start, free_i->free_secmap)) {
2479                 if (left_start > 0) {
2480                         left_start--;
2481                         continue;
2482                 }
2483                 left_start = find_next_zero_bit(free_i->free_secmap,
2484                                                         MAIN_SECS(sbi), 0);
2485                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2486                 break;
2487         }
2488         secno = left_start;
2489 skip_left:
2490         segno = GET_SEG_FROM_SEC(sbi, secno);
2491         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2492
2493         /* give up on finding another zone */
2494         if (!init)
2495                 goto got_it;
2496         if (sbi->secs_per_zone == 1)
2497                 goto got_it;
2498         if (zoneno == old_zoneno)
2499                 goto got_it;
2500         if (dir == ALLOC_LEFT) {
2501                 if (!go_left && zoneno + 1 >= total_zones)
2502                         goto got_it;
2503                 if (go_left && zoneno == 0)
2504                         goto got_it;
2505         }
2506         for (i = 0; i < NR_CURSEG_TYPE; i++)
2507                 if (CURSEG_I(sbi, i)->zone == zoneno)
2508                         break;
2509
2510         if (i < NR_CURSEG_TYPE) {
2511                 /* zone is in user, try another */
2512                 if (go_left)
2513                         hint = zoneno * sbi->secs_per_zone - 1;
2514                 else if (zoneno + 1 >= total_zones)
2515                         hint = 0;
2516                 else
2517                         hint = (zoneno + 1) * sbi->secs_per_zone;
2518                 init = false;
2519                 goto find_other_zone;
2520         }
2521 got_it:
2522         /* set it as dirty segment in free segmap */
2523         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2524         __set_inuse(sbi, segno);
2525         *newseg = segno;
2526         spin_unlock(&free_i->segmap_lock);
2527 }
2528
2529 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2530 {
2531         struct curseg_info *curseg = CURSEG_I(sbi, type);
2532         struct summary_footer *sum_footer;
2533         unsigned short seg_type = curseg->seg_type;
2534
2535         curseg->inited = true;
2536         curseg->segno = curseg->next_segno;
2537         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2538         curseg->next_blkoff = 0;
2539         curseg->next_segno = NULL_SEGNO;
2540
2541         sum_footer = &(curseg->sum_blk->footer);
2542         memset(sum_footer, 0, sizeof(struct summary_footer));
2543
2544         sanity_check_seg_type(sbi, seg_type);
2545
2546         if (IS_DATASEG(seg_type))
2547                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2548         if (IS_NODESEG(seg_type))
2549                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2550         __set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
2551 }
2552
2553 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2554 {
2555         struct curseg_info *curseg = CURSEG_I(sbi, type);
2556         unsigned short seg_type = curseg->seg_type;
2557
2558         sanity_check_seg_type(sbi, seg_type);
2559
2560         /* if segs_per_sec is large than 1, we need to keep original policy. */
2561         if (__is_large_section(sbi))
2562                 return curseg->segno;
2563
2564         /* inmem log may not locate on any segment after mount */
2565         if (!curseg->inited)
2566                 return 0;
2567
2568         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2569                 return 0;
2570
2571         if (test_opt(sbi, NOHEAP) &&
2572                 (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type)))
2573                 return 0;
2574
2575         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2576                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2577
2578         /* find segments from 0 to reuse freed segments */
2579         if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2580                 return 0;
2581
2582         return curseg->segno;
2583 }
2584
2585 /*
2586  * Allocate a current working segment.
2587  * This function always allocates a free segment in LFS manner.
2588  */
2589 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2590 {
2591         struct curseg_info *curseg = CURSEG_I(sbi, type);
2592         unsigned short seg_type = curseg->seg_type;
2593         unsigned int segno = curseg->segno;
2594         int dir = ALLOC_LEFT;
2595
2596         if (curseg->inited)
2597                 write_sum_page(sbi, curseg->sum_blk,
2598                                 GET_SUM_BLOCK(sbi, segno));
2599         if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA)
2600                 dir = ALLOC_RIGHT;
2601
2602         if (test_opt(sbi, NOHEAP))
2603                 dir = ALLOC_RIGHT;
2604
2605         segno = __get_next_segno(sbi, type);
2606         get_new_segment(sbi, &segno, new_sec, dir);
2607         curseg->next_segno = segno;
2608         reset_curseg(sbi, type, 1);
2609         curseg->alloc_type = LFS;
2610 }
2611
2612 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2613                         struct curseg_info *seg, block_t start)
2614 {
2615         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2616         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2617         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2618         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2619         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2620         int i, pos;
2621
2622         for (i = 0; i < entries; i++)
2623                 target_map[i] = ckpt_map[i] | cur_map[i];
2624
2625         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2626
2627         seg->next_blkoff = pos;
2628 }
2629
2630 /*
2631  * If a segment is written by LFS manner, next block offset is just obtained
2632  * by increasing the current block offset. However, if a segment is written by
2633  * SSR manner, next block offset obtained by calling __next_free_blkoff
2634  */
2635 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2636                                 struct curseg_info *seg)
2637 {
2638         if (seg->alloc_type == SSR)
2639                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2640         else
2641                 seg->next_blkoff++;
2642 }
2643
2644 /*
2645  * This function always allocates a used segment(from dirty seglist) by SSR
2646  * manner, so it should recover the existing segment information of valid blocks
2647  */
2648 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool flush)
2649 {
2650         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2651         struct curseg_info *curseg = CURSEG_I(sbi, type);
2652         unsigned int new_segno = curseg->next_segno;
2653         struct f2fs_summary_block *sum_node;
2654         struct page *sum_page;
2655
2656         if (flush)
2657                 write_sum_page(sbi, curseg->sum_blk,
2658                                         GET_SUM_BLOCK(sbi, curseg->segno));
2659
2660         __set_test_and_inuse(sbi, new_segno);
2661
2662         mutex_lock(&dirty_i->seglist_lock);
2663         __remove_dirty_segment(sbi, new_segno, PRE);
2664         __remove_dirty_segment(sbi, new_segno, DIRTY);
2665         mutex_unlock(&dirty_i->seglist_lock);
2666
2667         reset_curseg(sbi, type, 1);
2668         curseg->alloc_type = SSR;
2669         __next_free_blkoff(sbi, curseg, 0);
2670
2671         sum_page = f2fs_get_sum_page(sbi, new_segno);
2672         f2fs_bug_on(sbi, IS_ERR(sum_page));
2673         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2674         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2675         f2fs_put_page(sum_page, 1);
2676 }
2677
2678 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2679                                 int alloc_mode, unsigned long long age);
2680
2681 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type,
2682                                         int target_type, int alloc_mode,
2683                                         unsigned long long age)
2684 {
2685         struct curseg_info *curseg = CURSEG_I(sbi, type);
2686
2687         curseg->seg_type = target_type;
2688
2689         if (get_ssr_segment(sbi, type, alloc_mode, age)) {
2690                 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
2691
2692                 curseg->seg_type = se->type;
2693                 change_curseg(sbi, type, true);
2694         } else {
2695                 /* allocate cold segment by default */
2696                 curseg->seg_type = CURSEG_COLD_DATA;
2697                 new_curseg(sbi, type, true);
2698         }
2699         stat_inc_seg_type(sbi, curseg);
2700 }
2701
2702 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
2703 {
2704         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
2705
2706         if (!sbi->am.atgc_enabled)
2707                 return;
2708
2709         down_read(&SM_I(sbi)->curseg_lock);
2710
2711         mutex_lock(&curseg->curseg_mutex);
2712         down_write(&SIT_I(sbi)->sentry_lock);
2713
2714         get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0);
2715
2716         up_write(&SIT_I(sbi)->sentry_lock);
2717         mutex_unlock(&curseg->curseg_mutex);
2718
2719         up_read(&SM_I(sbi)->curseg_lock);
2720
2721 }
2722 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
2723 {
2724         __f2fs_init_atgc_curseg(sbi);
2725 }
2726
2727 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2728 {
2729         struct curseg_info *curseg = CURSEG_I(sbi, type);
2730
2731         mutex_lock(&curseg->curseg_mutex);
2732         if (!curseg->inited)
2733                 goto out;
2734
2735         if (get_valid_blocks(sbi, curseg->segno, false)) {
2736                 write_sum_page(sbi, curseg->sum_blk,
2737                                 GET_SUM_BLOCK(sbi, curseg->segno));
2738         } else {
2739                 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2740                 __set_test_and_free(sbi, curseg->segno, true);
2741                 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2742         }
2743 out:
2744         mutex_unlock(&curseg->curseg_mutex);
2745 }
2746
2747 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
2748 {
2749         __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2750
2751         if (sbi->am.atgc_enabled)
2752                 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2753 }
2754
2755 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2756 {
2757         struct curseg_info *curseg = CURSEG_I(sbi, type);
2758
2759         mutex_lock(&curseg->curseg_mutex);
2760         if (!curseg->inited)
2761                 goto out;
2762         if (get_valid_blocks(sbi, curseg->segno, false))
2763                 goto out;
2764
2765         mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2766         __set_test_and_inuse(sbi, curseg->segno);
2767         mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2768 out:
2769         mutex_unlock(&curseg->curseg_mutex);
2770 }
2771
2772 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
2773 {
2774         __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2775
2776         if (sbi->am.atgc_enabled)
2777                 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2778 }
2779
2780 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2781                                 int alloc_mode, unsigned long long age)
2782 {
2783         struct curseg_info *curseg = CURSEG_I(sbi, type);
2784         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2785         unsigned segno = NULL_SEGNO;
2786         unsigned short seg_type = curseg->seg_type;
2787         int i, cnt;
2788         bool reversed = false;
2789
2790         sanity_check_seg_type(sbi, seg_type);
2791
2792         /* f2fs_need_SSR() already forces to do this */
2793         if (!v_ops->get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) {
2794                 curseg->next_segno = segno;
2795                 return 1;
2796         }
2797
2798         /* For node segments, let's do SSR more intensively */
2799         if (IS_NODESEG(seg_type)) {
2800                 if (seg_type >= CURSEG_WARM_NODE) {
2801                         reversed = true;
2802                         i = CURSEG_COLD_NODE;
2803                 } else {
2804                         i = CURSEG_HOT_NODE;
2805                 }
2806                 cnt = NR_CURSEG_NODE_TYPE;
2807         } else {
2808                 if (seg_type >= CURSEG_WARM_DATA) {
2809                         reversed = true;
2810                         i = CURSEG_COLD_DATA;
2811                 } else {
2812                         i = CURSEG_HOT_DATA;
2813                 }
2814                 cnt = NR_CURSEG_DATA_TYPE;
2815         }
2816
2817         for (; cnt-- > 0; reversed ? i-- : i++) {
2818                 if (i == seg_type)
2819                         continue;
2820                 if (!v_ops->get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) {
2821                         curseg->next_segno = segno;
2822                         return 1;
2823                 }
2824         }
2825
2826         /* find valid_blocks=0 in dirty list */
2827         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2828                 segno = get_free_segment(sbi);
2829                 if (segno != NULL_SEGNO) {
2830                         curseg->next_segno = segno;
2831                         return 1;
2832                 }
2833         }
2834         return 0;
2835 }
2836
2837 /*
2838  * flush out current segment and replace it with new segment
2839  * This function should be returned with success, otherwise BUG
2840  */
2841 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2842                                                 int type, bool force)
2843 {
2844         struct curseg_info *curseg = CURSEG_I(sbi, type);
2845
2846         if (force)
2847                 new_curseg(sbi, type, true);
2848         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2849                                         curseg->seg_type == CURSEG_WARM_NODE)
2850                 new_curseg(sbi, type, false);
2851         else if (curseg->alloc_type == LFS &&
2852                         is_next_segment_free(sbi, curseg, type) &&
2853                         likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2854                 new_curseg(sbi, type, false);
2855         else if (f2fs_need_SSR(sbi) &&
2856                         get_ssr_segment(sbi, type, SSR, 0))
2857                 change_curseg(sbi, type, true);
2858         else
2859                 new_curseg(sbi, type, false);
2860
2861         stat_inc_seg_type(sbi, curseg);
2862 }
2863
2864 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2865                                         unsigned int start, unsigned int end)
2866 {
2867         struct curseg_info *curseg = CURSEG_I(sbi, type);
2868         unsigned int segno;
2869
2870         down_read(&SM_I(sbi)->curseg_lock);
2871         mutex_lock(&curseg->curseg_mutex);
2872         down_write(&SIT_I(sbi)->sentry_lock);
2873
2874         segno = CURSEG_I(sbi, type)->segno;
2875         if (segno < start || segno > end)
2876                 goto unlock;
2877
2878         if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
2879                 change_curseg(sbi, type, true);
2880         else
2881                 new_curseg(sbi, type, true);
2882
2883         stat_inc_seg_type(sbi, curseg);
2884
2885         locate_dirty_segment(sbi, segno);
2886 unlock:
2887         up_write(&SIT_I(sbi)->sentry_lock);
2888
2889         if (segno != curseg->segno)
2890                 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2891                             type, segno, curseg->segno);
2892
2893         mutex_unlock(&curseg->curseg_mutex);
2894         up_read(&SM_I(sbi)->curseg_lock);
2895 }
2896
2897 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type)
2898 {
2899         struct curseg_info *curseg = CURSEG_I(sbi, type);
2900         unsigned int old_segno;
2901
2902         if (!curseg->inited)
2903                 goto alloc;
2904
2905         if (!curseg->next_blkoff &&
2906                 !get_valid_blocks(sbi, curseg->segno, false) &&
2907                 !get_ckpt_valid_blocks(sbi, curseg->segno))
2908                 return;
2909
2910 alloc:
2911         old_segno = curseg->segno;
2912         SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
2913         locate_dirty_segment(sbi, old_segno);
2914 }
2915
2916 void f2fs_allocate_new_segment(struct f2fs_sb_info *sbi, int type)
2917 {
2918         down_write(&SIT_I(sbi)->sentry_lock);
2919         __allocate_new_segment(sbi, type);
2920         up_write(&SIT_I(sbi)->sentry_lock);
2921 }
2922
2923 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2924 {
2925         int i;
2926
2927         down_write(&SIT_I(sbi)->sentry_lock);
2928         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
2929                 __allocate_new_segment(sbi, i);
2930         up_write(&SIT_I(sbi)->sentry_lock);
2931 }
2932
2933 static const struct segment_allocation default_salloc_ops = {
2934         .allocate_segment = allocate_segment_by_default,
2935 };
2936
2937 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2938                                                 struct cp_control *cpc)
2939 {
2940         __u64 trim_start = cpc->trim_start;
2941         bool has_candidate = false;
2942
2943         down_write(&SIT_I(sbi)->sentry_lock);
2944         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2945                 if (add_discard_addrs(sbi, cpc, true)) {
2946                         has_candidate = true;
2947                         break;
2948                 }
2949         }
2950         up_write(&SIT_I(sbi)->sentry_lock);
2951
2952         cpc->trim_start = trim_start;
2953         return has_candidate;
2954 }
2955
2956 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2957                                         struct discard_policy *dpolicy,
2958                                         unsigned int start, unsigned int end)
2959 {
2960         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2961         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2962         struct rb_node **insert_p = NULL, *insert_parent = NULL;
2963         struct discard_cmd *dc;
2964         struct blk_plug plug;
2965         int issued;
2966         unsigned int trimmed = 0;
2967
2968 next:
2969         issued = 0;
2970
2971         mutex_lock(&dcc->cmd_lock);
2972         if (unlikely(dcc->rbtree_check))
2973                 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2974                                                         &dcc->root, false));
2975
2976         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2977                                         NULL, start,
2978                                         (struct rb_entry **)&prev_dc,
2979                                         (struct rb_entry **)&next_dc,
2980                                         &insert_p, &insert_parent, true, NULL);
2981         if (!dc)
2982                 dc = next_dc;
2983
2984         blk_start_plug(&plug);
2985
2986         while (dc && dc->lstart <= end) {
2987                 struct rb_node *node;
2988                 int err = 0;
2989
2990                 if (dc->len < dpolicy->granularity)
2991                         goto skip;
2992
2993                 if (dc->state != D_PREP) {
2994                         list_move_tail(&dc->list, &dcc->fstrim_list);
2995                         goto skip;
2996                 }
2997
2998                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2999
3000                 if (issued >= dpolicy->max_requests) {
3001                         start = dc->lstart + dc->len;
3002
3003                         if (err)
3004                                 __remove_discard_cmd(sbi, dc);
3005
3006                         blk_finish_plug(&plug);
3007                         mutex_unlock(&dcc->cmd_lock);
3008                         trimmed += __wait_all_discard_cmd(sbi, NULL);
3009                         congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
3010                         goto next;
3011                 }
3012 skip:
3013                 node = rb_next(&dc->rb_node);
3014                 if (err)
3015                         __remove_discard_cmd(sbi, dc);
3016                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3017
3018                 if (fatal_signal_pending(current))
3019                         break;
3020         }
3021
3022         blk_finish_plug(&plug);
3023         mutex_unlock(&dcc->cmd_lock);
3024
3025         return trimmed;
3026 }
3027
3028 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3029 {
3030         __u64 start = F2FS_BYTES_TO_BLK(range->start);
3031         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
3032         unsigned int start_segno, end_segno;
3033         block_t start_block, end_block;
3034         struct cp_control cpc;
3035         struct discard_policy dpolicy;
3036         unsigned long long trimmed = 0;
3037         int err = 0;
3038         bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
3039
3040         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
3041                 return -EINVAL;
3042
3043         if (end < MAIN_BLKADDR(sbi))
3044                 goto out;
3045
3046         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
3047                 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
3048                 return -EFSCORRUPTED;
3049         }
3050
3051         /* start/end segment number in main_area */
3052         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3053         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3054                                                 GET_SEGNO(sbi, end);
3055         if (need_align) {
3056                 start_segno = rounddown(start_segno, sbi->segs_per_sec);
3057                 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
3058         }
3059
3060         cpc.reason = CP_DISCARD;
3061         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
3062         cpc.trim_start = start_segno;
3063         cpc.trim_end = end_segno;
3064
3065         if (sbi->discard_blks == 0)
3066                 goto out;
3067
3068         down_write(&sbi->gc_lock);
3069         err = f2fs_write_checkpoint(sbi, &cpc);
3070         up_write(&sbi->gc_lock);
3071         if (err)
3072                 goto out;
3073
3074         /*
3075          * We filed discard candidates, but actually we don't need to wait for
3076          * all of them, since they'll be issued in idle time along with runtime
3077          * discard option. User configuration looks like using runtime discard
3078          * or periodic fstrim instead of it.
3079          */
3080         if (f2fs_realtime_discard_enable(sbi))
3081                 goto out;
3082
3083         start_block = START_BLOCK(sbi, start_segno);
3084         end_block = START_BLOCK(sbi, end_segno + 1);
3085
3086         __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3087         trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3088                                         start_block, end_block);
3089
3090         trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3091                                         start_block, end_block);
3092 out:
3093         if (!err)
3094                 range->len = F2FS_BLK_TO_BYTES(trimmed);
3095         return err;
3096 }
3097
3098 static bool __has_curseg_space(struct f2fs_sb_info *sbi,
3099                                         struct curseg_info *curseg)
3100 {
3101         return curseg->next_blkoff < f2fs_usable_blks_in_seg(sbi,
3102                                                         curseg->segno);
3103 }
3104
3105 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
3106 {
3107         switch (hint) {
3108         case WRITE_LIFE_SHORT:
3109                 return CURSEG_HOT_DATA;
3110         case WRITE_LIFE_EXTREME:
3111                 return CURSEG_COLD_DATA;
3112         default:
3113                 return CURSEG_WARM_DATA;
3114         }
3115 }
3116
3117 /* This returns write hints for each segment type. This hints will be
3118  * passed down to block layer. There are mapping tables which depend on
3119  * the mount option 'whint_mode'.
3120  *
3121  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
3122  *
3123  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
3124  *
3125  * User                  F2FS                     Block
3126  * ----                  ----                     -----
3127  *                       META                     WRITE_LIFE_NOT_SET
3128  *                       HOT_NODE                 "
3129  *                       WARM_NODE                "
3130  *                       COLD_NODE                "
3131  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
3132  * extension list        "                        "
3133  *
3134  * -- buffered io
3135  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3136  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3137  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
3138  * WRITE_LIFE_NONE       "                        "
3139  * WRITE_LIFE_MEDIUM     "                        "
3140  * WRITE_LIFE_LONG       "                        "
3141  *
3142  * -- direct io
3143  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3144  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3145  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
3146  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
3147  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
3148  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
3149  *
3150  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
3151  *
3152  * User                  F2FS                     Block
3153  * ----                  ----                     -----
3154  *                       META                     WRITE_LIFE_MEDIUM;
3155  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
3156  *                       WARM_NODE                "
3157  *                       COLD_NODE                WRITE_LIFE_NONE
3158  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
3159  * extension list        "                        "
3160  *
3161  * -- buffered io
3162  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3163  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3164  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
3165  * WRITE_LIFE_NONE       "                        "
3166  * WRITE_LIFE_MEDIUM     "                        "
3167  * WRITE_LIFE_LONG       "                        "
3168  *
3169  * -- direct io
3170  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3171  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3172  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
3173  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
3174  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
3175  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
3176  */
3177
3178 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
3179                                 enum page_type type, enum temp_type temp)
3180 {
3181         if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
3182                 if (type == DATA) {
3183                         if (temp == WARM)
3184                                 return WRITE_LIFE_NOT_SET;
3185                         else if (temp == HOT)
3186                                 return WRITE_LIFE_SHORT;
3187                         else if (temp == COLD)
3188                                 return WRITE_LIFE_EXTREME;
3189                 } else {
3190                         return WRITE_LIFE_NOT_SET;
3191                 }
3192         } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
3193                 if (type == DATA) {
3194                         if (temp == WARM)
3195                                 return WRITE_LIFE_LONG;
3196                         else if (temp == HOT)
3197                                 return WRITE_LIFE_SHORT;
3198                         else if (temp == COLD)
3199                                 return WRITE_LIFE_EXTREME;
3200                 } else if (type == NODE) {
3201                         if (temp == WARM || temp == HOT)
3202                                 return WRITE_LIFE_NOT_SET;
3203                         else if (temp == COLD)
3204                                 return WRITE_LIFE_NONE;
3205                 } else if (type == META) {
3206                         return WRITE_LIFE_MEDIUM;
3207                 }
3208         }
3209         return WRITE_LIFE_NOT_SET;
3210 }
3211
3212 static int __get_segment_type_2(struct f2fs_io_info *fio)
3213 {
3214         if (fio->type == DATA)
3215                 return CURSEG_HOT_DATA;
3216         else
3217                 return CURSEG_HOT_NODE;
3218 }
3219
3220 static int __get_segment_type_4(struct f2fs_io_info *fio)
3221 {
3222         if (fio->type == DATA) {
3223                 struct inode *inode = fio->page->mapping->host;
3224
3225                 if (S_ISDIR(inode->i_mode))
3226                         return CURSEG_HOT_DATA;
3227                 else
3228                         return CURSEG_COLD_DATA;
3229         } else {
3230                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3231                         return CURSEG_WARM_NODE;
3232                 else
3233                         return CURSEG_COLD_NODE;
3234         }
3235 }
3236
3237 static int __get_segment_type_6(struct f2fs_io_info *fio)
3238 {
3239         if (fio->type == DATA) {
3240                 struct inode *inode = fio->page->mapping->host;
3241
3242                 if (is_cold_data(fio->page)) {
3243                         if (fio->sbi->am.atgc_enabled)
3244                                 return CURSEG_ALL_DATA_ATGC;
3245                         else
3246                                 return CURSEG_COLD_DATA;
3247                 }
3248                 if (file_is_cold(inode) || f2fs_compressed_file(inode))
3249                         return CURSEG_COLD_DATA;
3250                 if (file_is_hot(inode) ||
3251                                 is_inode_flag_set(inode, FI_HOT_DATA) ||
3252                                 f2fs_is_atomic_file(inode) ||
3253                                 f2fs_is_volatile_file(inode))
3254                         return CURSEG_HOT_DATA;
3255                 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3256         } else {
3257                 if (IS_DNODE(fio->page))
3258                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3259                                                 CURSEG_HOT_NODE;
3260                 return CURSEG_COLD_NODE;
3261         }
3262 }
3263
3264 static int __get_segment_type(struct f2fs_io_info *fio)
3265 {
3266         int type = 0;
3267
3268         switch (F2FS_OPTION(fio->sbi).active_logs) {
3269         case 2:
3270                 type = __get_segment_type_2(fio);
3271                 break;
3272         case 4:
3273                 type = __get_segment_type_4(fio);
3274                 break;
3275         case 6:
3276                 type = __get_segment_type_6(fio);
3277                 break;
3278         default:
3279                 f2fs_bug_on(fio->sbi, true);
3280         }
3281
3282         if (IS_HOT(type))
3283                 fio->temp = HOT;
3284         else if (IS_WARM(type))
3285                 fio->temp = WARM;
3286         else
3287                 fio->temp = COLD;
3288         return type;
3289 }
3290
3291 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3292                 block_t old_blkaddr, block_t *new_blkaddr,
3293                 struct f2fs_summary *sum, int type,
3294                 struct f2fs_io_info *fio)
3295 {
3296         struct sit_info *sit_i = SIT_I(sbi);
3297         struct curseg_info *curseg = CURSEG_I(sbi, type);
3298         unsigned long long old_mtime;
3299         bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3300         struct seg_entry *se = NULL;
3301
3302         down_read(&SM_I(sbi)->curseg_lock);
3303
3304         mutex_lock(&curseg->curseg_mutex);
3305         down_write(&sit_i->sentry_lock);
3306
3307         if (from_gc) {
3308                 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3309                 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3310                 sanity_check_seg_type(sbi, se->type);
3311                 f2fs_bug_on(sbi, IS_NODESEG(se->type));
3312         }
3313         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3314
3315         f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg);
3316
3317         f2fs_wait_discard_bio(sbi, *new_blkaddr);
3318
3319         /*
3320          * __add_sum_entry should be resided under the curseg_mutex
3321          * because, this function updates a summary entry in the
3322          * current summary block.
3323          */
3324         __add_sum_entry(sbi, type, sum);
3325
3326         __refresh_next_blkoff(sbi, curseg);
3327
3328         stat_inc_block_count(sbi, curseg);
3329
3330         if (from_gc) {
3331                 old_mtime = get_segment_mtime(sbi, old_blkaddr);
3332         } else {
3333                 update_segment_mtime(sbi, old_blkaddr, 0);
3334                 old_mtime = 0;
3335         }
3336         update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3337
3338         /*
3339          * SIT information should be updated before segment allocation,
3340          * since SSR needs latest valid block information.
3341          */
3342         update_sit_entry(sbi, *new_blkaddr, 1);
3343         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3344                 update_sit_entry(sbi, old_blkaddr, -1);
3345
3346         if (!__has_curseg_space(sbi, curseg)) {
3347                 if (from_gc)
3348                         get_atssr_segment(sbi, type, se->type,
3349                                                 AT_SSR, se->mtime);
3350                 else
3351                         sit_i->s_ops->allocate_segment(sbi, type, false);
3352         }
3353         /*
3354          * segment dirty status should be updated after segment allocation,
3355          * so we just need to update status only one time after previous
3356          * segment being closed.
3357          */
3358         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3359         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3360
3361         up_write(&sit_i->sentry_lock);
3362
3363         if (page && IS_NODESEG(type)) {
3364                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3365
3366                 f2fs_inode_chksum_set(sbi, page);
3367         }
3368
3369         if (F2FS_IO_ALIGNED(sbi))
3370                 fio->retry = false;
3371
3372         if (fio) {
3373                 struct f2fs_bio_info *io;
3374
3375                 INIT_LIST_HEAD(&fio->list);
3376                 fio->in_list = true;
3377                 io = sbi->write_io[fio->type] + fio->temp;
3378                 spin_lock(&io->io_lock);
3379                 list_add_tail(&fio->list, &io->io_list);
3380                 spin_unlock(&io->io_lock);
3381         }
3382
3383         mutex_unlock(&curseg->curseg_mutex);
3384
3385         up_read(&SM_I(sbi)->curseg_lock);
3386 }
3387
3388 static void update_device_state(struct f2fs_io_info *fio)
3389 {
3390         struct f2fs_sb_info *sbi = fio->sbi;
3391         unsigned int devidx;
3392
3393         if (!f2fs_is_multi_device(sbi))
3394                 return;
3395
3396         devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3397
3398         /* update device state for fsync */
3399         f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
3400
3401         /* update device state for checkpoint */
3402         if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3403                 spin_lock(&sbi->dev_lock);
3404                 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3405                 spin_unlock(&sbi->dev_lock);
3406         }
3407 }
3408
3409 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3410 {
3411         int type = __get_segment_type(fio);
3412         bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3413
3414         if (keep_order)
3415                 down_read(&fio->sbi->io_order_lock);
3416 reallocate:
3417         f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3418                         &fio->new_blkaddr, sum, type, fio);
3419         if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3420                 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3421                                         fio->old_blkaddr, fio->old_blkaddr);
3422
3423         /* writeout dirty page into bdev */
3424         f2fs_submit_page_write(fio);
3425         if (fio->retry) {
3426                 fio->old_blkaddr = fio->new_blkaddr;
3427                 goto reallocate;
3428         }
3429
3430         update_device_state(fio);
3431
3432         if (keep_order)
3433                 up_read(&fio->sbi->io_order_lock);
3434 }
3435
3436 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3437                                         enum iostat_type io_type)
3438 {
3439         struct f2fs_io_info fio = {
3440                 .sbi = sbi,
3441                 .type = META,
3442                 .temp = HOT,
3443                 .op = REQ_OP_WRITE,
3444                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3445                 .old_blkaddr = page->index,
3446                 .new_blkaddr = page->index,
3447                 .page = page,
3448                 .encrypted_page = NULL,
3449                 .in_list = false,
3450         };
3451
3452         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3453                 fio.op_flags &= ~REQ_META;
3454
3455         set_page_writeback(page);
3456         ClearPageError(page);
3457         f2fs_submit_page_write(&fio);
3458
3459         stat_inc_meta_count(sbi, page->index);
3460         f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3461 }
3462
3463 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3464 {
3465         struct f2fs_summary sum;
3466
3467         set_summary(&sum, nid, 0, 0);
3468         do_write_page(&sum, fio);
3469
3470         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3471 }
3472
3473 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3474                                         struct f2fs_io_info *fio)
3475 {
3476         struct f2fs_sb_info *sbi = fio->sbi;
3477         struct f2fs_summary sum;
3478
3479         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3480         set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3481         do_write_page(&sum, fio);
3482         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3483
3484         f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3485 }
3486
3487 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3488 {
3489         int err;
3490         struct f2fs_sb_info *sbi = fio->sbi;
3491         unsigned int segno;
3492
3493         fio->new_blkaddr = fio->old_blkaddr;
3494         /* i/o temperature is needed for passing down write hints */
3495         __get_segment_type(fio);
3496
3497         segno = GET_SEGNO(sbi, fio->new_blkaddr);
3498
3499         if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3500                 set_sbi_flag(sbi, SBI_NEED_FSCK);
3501                 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3502                           __func__, segno);
3503                 return -EFSCORRUPTED;
3504         }
3505
3506         stat_inc_inplace_blocks(fio->sbi);
3507
3508         if (fio->bio && !(SM_I(sbi)->ipu_policy & (1 << F2FS_IPU_NOCACHE)))
3509                 err = f2fs_merge_page_bio(fio);
3510         else
3511                 err = f2fs_submit_page_bio(fio);
3512         if (!err) {
3513                 update_device_state(fio);
3514                 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3515         }
3516
3517         return err;
3518 }
3519
3520 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3521                                                 unsigned int segno)
3522 {
3523         int i;
3524
3525         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3526                 if (CURSEG_I(sbi, i)->segno == segno)
3527                         break;
3528         }
3529         return i;
3530 }
3531
3532 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3533                                 block_t old_blkaddr, block_t new_blkaddr,
3534                                 bool recover_curseg, bool recover_newaddr,
3535                                 bool from_gc)
3536 {
3537         struct sit_info *sit_i = SIT_I(sbi);
3538         struct curseg_info *curseg;
3539         unsigned int segno, old_cursegno;
3540         struct seg_entry *se;
3541         int type;
3542         unsigned short old_blkoff;
3543
3544         segno = GET_SEGNO(sbi, new_blkaddr);
3545         se = get_seg_entry(sbi, segno);
3546         type = se->type;
3547
3548         down_write(&SM_I(sbi)->curseg_lock);
3549
3550         if (!recover_curseg) {
3551                 /* for recovery flow */
3552                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3553                         if (old_blkaddr == NULL_ADDR)
3554                                 type = CURSEG_COLD_DATA;
3555                         else
3556                                 type = CURSEG_WARM_DATA;
3557                 }
3558         } else {
3559                 if (IS_CURSEG(sbi, segno)) {
3560                         /* se->type is volatile as SSR allocation */
3561                         type = __f2fs_get_curseg(sbi, segno);
3562                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3563                 } else {
3564                         type = CURSEG_WARM_DATA;
3565                 }
3566         }
3567
3568         f2fs_bug_on(sbi, !IS_DATASEG(type));
3569         curseg = CURSEG_I(sbi, type);
3570
3571         mutex_lock(&curseg->curseg_mutex);
3572         down_write(&sit_i->sentry_lock);
3573
3574         old_cursegno = curseg->segno;
3575         old_blkoff = curseg->next_blkoff;
3576
3577         /* change the current segment */
3578         if (segno != curseg->segno) {
3579                 curseg->next_segno = segno;
3580                 change_curseg(sbi, type, true);
3581         }
3582
3583         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3584         __add_sum_entry(sbi, type, sum);
3585
3586         if (!recover_curseg || recover_newaddr) {
3587                 if (!from_gc)
3588                         update_segment_mtime(sbi, new_blkaddr, 0);
3589                 update_sit_entry(sbi, new_blkaddr, 1);
3590         }
3591         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3592                 invalidate_mapping_pages(META_MAPPING(sbi),
3593                                         old_blkaddr, old_blkaddr);
3594                 if (!from_gc)
3595                         update_segment_mtime(sbi, old_blkaddr, 0);
3596                 update_sit_entry(sbi, old_blkaddr, -1);
3597         }
3598
3599         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3600         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3601
3602         locate_dirty_segment(sbi, old_cursegno);
3603
3604         if (recover_curseg) {
3605                 if (old_cursegno != curseg->segno) {
3606                         curseg->next_segno = old_cursegno;
3607                         change_curseg(sbi, type, true);
3608                 }
3609                 curseg->next_blkoff = old_blkoff;
3610         }
3611
3612         up_write(&sit_i->sentry_lock);
3613         mutex_unlock(&curseg->curseg_mutex);
3614         up_write(&SM_I(sbi)->curseg_lock);
3615 }
3616
3617 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3618                                 block_t old_addr, block_t new_addr,
3619                                 unsigned char version, bool recover_curseg,
3620                                 bool recover_newaddr)
3621 {
3622         struct f2fs_summary sum;
3623
3624         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3625
3626         f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3627                                         recover_curseg, recover_newaddr, false);
3628
3629         f2fs_update_data_blkaddr(dn, new_addr);
3630 }
3631
3632 void f2fs_wait_on_page_writeback(struct page *page,
3633                                 enum page_type type, bool ordered, bool locked)
3634 {
3635         if (PageWriteback(page)) {
3636                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3637
3638                 /* submit cached LFS IO */
3639                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3640                 /* sbumit cached IPU IO */
3641                 f2fs_submit_merged_ipu_write(sbi, NULL, page);
3642                 if (ordered) {
3643                         wait_on_page_writeback(page);
3644                         f2fs_bug_on(sbi, locked && PageWriteback(page));
3645                 } else {
3646                         wait_for_stable_page(page);
3647                 }
3648         }
3649 }
3650
3651 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3652 {
3653         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3654         struct page *cpage;
3655
3656         if (!f2fs_post_read_required(inode))
3657                 return;
3658
3659         if (!__is_valid_data_blkaddr(blkaddr))
3660                 return;
3661
3662         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3663         if (cpage) {
3664                 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3665                 f2fs_put_page(cpage, 1);
3666         }
3667 }
3668
3669 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3670                                                                 block_t len)
3671 {
3672         block_t i;
3673
3674         for (i = 0; i < len; i++)
3675                 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3676 }
3677
3678 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3679 {
3680         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3681         struct curseg_info *seg_i;
3682         unsigned char *kaddr;
3683         struct page *page;
3684         block_t start;
3685         int i, j, offset;
3686
3687         start = start_sum_block(sbi);
3688
3689         page = f2fs_get_meta_page(sbi, start++);
3690         if (IS_ERR(page))
3691                 return PTR_ERR(page);
3692         kaddr = (unsigned char *)page_address(page);
3693
3694         /* Step 1: restore nat cache */
3695         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3696         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3697
3698         /* Step 2: restore sit cache */
3699         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3700         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3701         offset = 2 * SUM_JOURNAL_SIZE;
3702
3703         /* Step 3: restore summary entries */
3704         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3705                 unsigned short blk_off;
3706                 unsigned int segno;
3707
3708                 seg_i = CURSEG_I(sbi, i);
3709                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3710                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3711                 seg_i->next_segno = segno;
3712                 reset_curseg(sbi, i, 0);
3713                 seg_i->alloc_type = ckpt->alloc_type[i];
3714                 seg_i->next_blkoff = blk_off;
3715
3716                 if (seg_i->alloc_type == SSR)
3717                         blk_off = sbi->blocks_per_seg;
3718
3719                 for (j = 0; j < blk_off; j++) {
3720                         struct f2fs_summary *s;
3721                         s = (struct f2fs_summary *)(kaddr + offset);
3722                         seg_i->sum_blk->entries[j] = *s;
3723                         offset += SUMMARY_SIZE;
3724                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3725                                                 SUM_FOOTER_SIZE)
3726                                 continue;
3727
3728                         f2fs_put_page(page, 1);
3729                         page = NULL;
3730
3731                         page = f2fs_get_meta_page(sbi, start++);
3732                         if (IS_ERR(page))
3733                                 return PTR_ERR(page);
3734                         kaddr = (unsigned char *)page_address(page);
3735                         offset = 0;
3736                 }
3737         }
3738         f2fs_put_page(page, 1);
3739         return 0;
3740 }
3741
3742 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3743 {
3744         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3745         struct f2fs_summary_block *sum;
3746         struct curseg_info *curseg;
3747         struct page *new;
3748         unsigned short blk_off;
3749         unsigned int segno = 0;
3750         block_t blk_addr = 0;
3751         int err = 0;
3752
3753         /* get segment number and block addr */
3754         if (IS_DATASEG(type)) {
3755                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3756                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3757                                                         CURSEG_HOT_DATA]);
3758                 if (__exist_node_summaries(sbi))
3759                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
3760                 else
3761                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3762         } else {
3763                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3764                                                         CURSEG_HOT_NODE]);
3765                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3766                                                         CURSEG_HOT_NODE]);
3767                 if (__exist_node_summaries(sbi))
3768                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3769                                                         type - CURSEG_HOT_NODE);
3770                 else
3771                         blk_addr = GET_SUM_BLOCK(sbi, segno);
3772         }
3773
3774         new = f2fs_get_meta_page(sbi, blk_addr);
3775         if (IS_ERR(new))
3776                 return PTR_ERR(new);
3777         sum = (struct f2fs_summary_block *)page_address(new);
3778
3779         if (IS_NODESEG(type)) {
3780                 if (__exist_node_summaries(sbi)) {
3781                         struct f2fs_summary *ns = &sum->entries[0];
3782                         int i;
3783                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3784                                 ns->version = 0;
3785                                 ns->ofs_in_node = 0;
3786                         }
3787                 } else {
3788                         err = f2fs_restore_node_summary(sbi, segno, sum);
3789                         if (err)
3790                                 goto out;
3791                 }
3792         }
3793
3794         /* set uncompleted segment to curseg */
3795         curseg = CURSEG_I(sbi, type);
3796         mutex_lock(&curseg->curseg_mutex);
3797
3798         /* update journal info */
3799         down_write(&curseg->journal_rwsem);
3800         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3801         up_write(&curseg->journal_rwsem);
3802
3803         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3804         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3805         curseg->next_segno = segno;
3806         reset_curseg(sbi, type, 0);
3807         curseg->alloc_type = ckpt->alloc_type[type];
3808         curseg->next_blkoff = blk_off;
3809         mutex_unlock(&curseg->curseg_mutex);
3810 out:
3811         f2fs_put_page(new, 1);
3812         return err;
3813 }
3814
3815 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3816 {
3817         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3818         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3819         int type = CURSEG_HOT_DATA;
3820         int err;
3821
3822         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3823                 int npages = f2fs_npages_for_summary_flush(sbi, true);
3824
3825                 if (npages >= 2)
3826                         f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3827                                                         META_CP, true);
3828
3829                 /* restore for compacted data summary */
3830                 err = read_compacted_summaries(sbi);
3831                 if (err)
3832                         return err;
3833                 type = CURSEG_HOT_NODE;
3834         }
3835
3836         if (__exist_node_summaries(sbi))
3837                 f2fs_ra_meta_pages(sbi,
3838                                 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
3839                                 NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
3840
3841         for (; type <= CURSEG_COLD_NODE; type++) {
3842                 err = read_normal_summaries(sbi, type);
3843                 if (err)
3844                         return err;
3845         }
3846
3847         /* sanity check for summary blocks */
3848         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3849                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3850                 f2fs_err(sbi, "invalid journal entries nats %u sits %u\n",
3851                          nats_in_cursum(nat_j), sits_in_cursum(sit_j));
3852                 return -EINVAL;
3853         }
3854
3855         return 0;
3856 }
3857
3858 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3859 {
3860         struct page *page;
3861         unsigned char *kaddr;
3862         struct f2fs_summary *summary;
3863         struct curseg_info *seg_i;
3864         int written_size = 0;
3865         int i, j;
3866
3867         page = f2fs_grab_meta_page(sbi, blkaddr++);
3868         kaddr = (unsigned char *)page_address(page);
3869         memset(kaddr, 0, PAGE_SIZE);
3870
3871         /* Step 1: write nat cache */
3872         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3873         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3874         written_size += SUM_JOURNAL_SIZE;
3875
3876         /* Step 2: write sit cache */
3877         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3878         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3879         written_size += SUM_JOURNAL_SIZE;
3880
3881         /* Step 3: write summary entries */
3882         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3883                 unsigned short blkoff;
3884                 seg_i = CURSEG_I(sbi, i);
3885                 if (sbi->ckpt->alloc_type[i] == SSR)
3886                         blkoff = sbi->blocks_per_seg;
3887                 else
3888                         blkoff = curseg_blkoff(sbi, i);
3889
3890                 for (j = 0; j < blkoff; j++) {
3891                         if (!page) {
3892                                 page = f2fs_grab_meta_page(sbi, blkaddr++);
3893                                 kaddr = (unsigned char *)page_address(page);
3894                                 memset(kaddr, 0, PAGE_SIZE);
3895                                 written_size = 0;
3896                         }
3897                         summary = (struct f2fs_summary *)(kaddr + written_size);
3898                         *summary = seg_i->sum_blk->entries[j];
3899                         written_size += SUMMARY_SIZE;
3900
3901                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3902                                                         SUM_FOOTER_SIZE)
3903                                 continue;
3904
3905                         set_page_dirty(page);
3906                         f2fs_put_page(page, 1);
3907                         page = NULL;
3908                 }
3909         }
3910         if (page) {
3911                 set_page_dirty(page);
3912                 f2fs_put_page(page, 1);
3913         }
3914 }
3915
3916 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3917                                         block_t blkaddr, int type)
3918 {
3919         int i, end;
3920         if (IS_DATASEG(type))
3921                 end = type + NR_CURSEG_DATA_TYPE;
3922         else
3923                 end = type + NR_CURSEG_NODE_TYPE;
3924
3925         for (i = type; i < end; i++)
3926                 write_current_sum_page(sbi, i, blkaddr + (i - type));
3927 }
3928
3929 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3930 {
3931         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3932                 write_compacted_summaries(sbi, start_blk);
3933         else
3934                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3935 }
3936
3937 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3938 {
3939         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3940 }
3941
3942 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3943                                         unsigned int val, int alloc)
3944 {
3945         int i;
3946
3947         if (type == NAT_JOURNAL) {
3948                 for (i = 0; i < nats_in_cursum(journal); i++) {
3949                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3950                                 return i;
3951                 }
3952                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3953                         return update_nats_in_cursum(journal, 1);
3954         } else if (type == SIT_JOURNAL) {
3955                 for (i = 0; i < sits_in_cursum(journal); i++)
3956                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3957                                 return i;
3958                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3959                         return update_sits_in_cursum(journal, 1);
3960         }
3961         return -1;
3962 }
3963
3964 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3965                                         unsigned int segno)
3966 {
3967         return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
3968 }
3969
3970 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3971                                         unsigned int start)
3972 {
3973         struct sit_info *sit_i = SIT_I(sbi);
3974         struct page *page;
3975         pgoff_t src_off, dst_off;
3976
3977         src_off = current_sit_addr(sbi, start);
3978         dst_off = next_sit_addr(sbi, src_off);
3979
3980         page = f2fs_grab_meta_page(sbi, dst_off);
3981         seg_info_to_sit_page(sbi, page, start);
3982
3983         set_page_dirty(page);
3984         set_to_next_sit(sit_i, start);
3985
3986         return page;
3987 }
3988
3989 static struct sit_entry_set *grab_sit_entry_set(void)
3990 {
3991         struct sit_entry_set *ses =
3992                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3993
3994         ses->entry_cnt = 0;
3995         INIT_LIST_HEAD(&ses->set_list);
3996         return ses;
3997 }
3998
3999 static void release_sit_entry_set(struct sit_entry_set *ses)
4000 {
4001         list_del(&ses->set_list);
4002         kmem_cache_free(sit_entry_set_slab, ses);
4003 }
4004
4005 static void adjust_sit_entry_set(struct sit_entry_set *ses,
4006                                                 struct list_head *head)
4007 {
4008         struct sit_entry_set *next = ses;
4009
4010         if (list_is_last(&ses->set_list, head))
4011                 return;
4012
4013         list_for_each_entry_continue(next, head, set_list)
4014                 if (ses->entry_cnt <= next->entry_cnt)
4015                         break;
4016
4017         list_move_tail(&ses->set_list, &next->set_list);
4018 }
4019
4020 static void add_sit_entry(unsigned int segno, struct list_head *head)
4021 {
4022         struct sit_entry_set *ses;
4023         unsigned int start_segno = START_SEGNO(segno);
4024
4025         list_for_each_entry(ses, head, set_list) {
4026                 if (ses->start_segno == start_segno) {
4027                         ses->entry_cnt++;
4028                         adjust_sit_entry_set(ses, head);
4029                         return;
4030                 }
4031         }
4032
4033         ses = grab_sit_entry_set();
4034
4035         ses->start_segno = start_segno;
4036         ses->entry_cnt++;
4037         list_add(&ses->set_list, head);
4038 }
4039
4040 static void add_sits_in_set(struct f2fs_sb_info *sbi)
4041 {
4042         struct f2fs_sm_info *sm_info = SM_I(sbi);
4043         struct list_head *set_list = &sm_info->sit_entry_set;
4044         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
4045         unsigned int segno;
4046
4047         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
4048                 add_sit_entry(segno, set_list);
4049 }
4050
4051 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
4052 {
4053         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4054         struct f2fs_journal *journal = curseg->journal;
4055         int i;
4056
4057         down_write(&curseg->journal_rwsem);
4058         for (i = 0; i < sits_in_cursum(journal); i++) {
4059                 unsigned int segno;
4060                 bool dirtied;
4061
4062                 segno = le32_to_cpu(segno_in_journal(journal, i));
4063                 dirtied = __mark_sit_entry_dirty(sbi, segno);
4064
4065                 if (!dirtied)
4066                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
4067         }
4068         update_sits_in_cursum(journal, -i);
4069         up_write(&curseg->journal_rwsem);
4070 }
4071
4072 /*
4073  * CP calls this function, which flushes SIT entries including sit_journal,
4074  * and moves prefree segs to free segs.
4075  */
4076 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
4077 {
4078         struct sit_info *sit_i = SIT_I(sbi);
4079         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4080         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4081         struct f2fs_journal *journal = curseg->journal;
4082         struct sit_entry_set *ses, *tmp;
4083         struct list_head *head = &SM_I(sbi)->sit_entry_set;
4084         bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4085         struct seg_entry *se;
4086
4087         down_write(&sit_i->sentry_lock);
4088
4089         if (!sit_i->dirty_sentries)
4090                 goto out;
4091
4092         /*
4093          * add and account sit entries of dirty bitmap in sit entry
4094          * set temporarily
4095          */
4096         add_sits_in_set(sbi);
4097
4098         /*
4099          * if there are no enough space in journal to store dirty sit
4100          * entries, remove all entries from journal and add and account
4101          * them in sit entry set.
4102          */
4103         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4104                                                                 !to_journal)
4105                 remove_sits_in_journal(sbi);
4106
4107         /*
4108          * there are two steps to flush sit entries:
4109          * #1, flush sit entries to journal in current cold data summary block.
4110          * #2, flush sit entries to sit page.
4111          */
4112         list_for_each_entry_safe(ses, tmp, head, set_list) {
4113                 struct page *page = NULL;
4114                 struct f2fs_sit_block *raw_sit = NULL;
4115                 unsigned int start_segno = ses->start_segno;
4116                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4117                                                 (unsigned long)MAIN_SEGS(sbi));
4118                 unsigned int segno = start_segno;
4119
4120                 if (to_journal &&
4121                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4122                         to_journal = false;
4123
4124                 if (to_journal) {
4125                         down_write(&curseg->journal_rwsem);
4126                 } else {
4127                         page = get_next_sit_page(sbi, start_segno);
4128                         raw_sit = page_address(page);
4129                 }
4130
4131                 /* flush dirty sit entries in region of current sit set */
4132                 for_each_set_bit_from(segno, bitmap, end) {
4133                         int offset, sit_offset;
4134
4135                         se = get_seg_entry(sbi, segno);
4136 #ifdef CONFIG_F2FS_CHECK_FS
4137                         if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4138                                                 SIT_VBLOCK_MAP_SIZE))
4139                                 f2fs_bug_on(sbi, 1);
4140 #endif
4141
4142                         /* add discard candidates */
4143                         if (!(cpc->reason & CP_DISCARD)) {
4144                                 cpc->trim_start = segno;
4145                                 add_discard_addrs(sbi, cpc, false);
4146                         }
4147
4148                         if (to_journal) {
4149                                 offset = f2fs_lookup_journal_in_cursum(journal,
4150                                                         SIT_JOURNAL, segno, 1);
4151                                 f2fs_bug_on(sbi, offset < 0);
4152                                 segno_in_journal(journal, offset) =
4153                                                         cpu_to_le32(segno);
4154                                 seg_info_to_raw_sit(se,
4155                                         &sit_in_journal(journal, offset));
4156                                 check_block_count(sbi, segno,
4157                                         &sit_in_journal(journal, offset));
4158                         } else {
4159                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4160                                 seg_info_to_raw_sit(se,
4161                                                 &raw_sit->entries[sit_offset]);
4162                                 check_block_count(sbi, segno,
4163                                                 &raw_sit->entries[sit_offset]);
4164                         }
4165
4166                         __clear_bit(segno, bitmap);
4167                         sit_i->dirty_sentries--;
4168                         ses->entry_cnt--;
4169                 }
4170
4171                 if (to_journal)
4172                         up_write(&curseg->journal_rwsem);
4173                 else
4174                         f2fs_put_page(page, 1);
4175
4176                 f2fs_bug_on(sbi, ses->entry_cnt);
4177                 release_sit_entry_set(ses);
4178         }
4179
4180         f2fs_bug_on(sbi, !list_empty(head));
4181         f2fs_bug_on(sbi, sit_i->dirty_sentries);
4182 out:
4183         if (cpc->reason & CP_DISCARD) {
4184                 __u64 trim_start = cpc->trim_start;
4185
4186                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4187                         add_discard_addrs(sbi, cpc, false);
4188
4189                 cpc->trim_start = trim_start;
4190         }
4191         up_write(&sit_i->sentry_lock);
4192
4193         set_prefree_as_free_segments(sbi);
4194 }
4195
4196 static int build_sit_info(struct f2fs_sb_info *sbi)
4197 {
4198         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4199         struct sit_info *sit_i;
4200         unsigned int sit_segs, start;
4201         char *src_bitmap, *bitmap;
4202         unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4203
4204         /* allocate memory for SIT information */
4205         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4206         if (!sit_i)
4207                 return -ENOMEM;
4208
4209         SM_I(sbi)->sit_info = sit_i;
4210
4211         sit_i->sentries =
4212                 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4213                                               MAIN_SEGS(sbi)),
4214                               GFP_KERNEL);
4215         if (!sit_i->sentries)
4216                 return -ENOMEM;
4217
4218         main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4219         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4220                                                                 GFP_KERNEL);
4221         if (!sit_i->dirty_sentries_bitmap)
4222                 return -ENOMEM;
4223
4224 #ifdef CONFIG_F2FS_CHECK_FS
4225         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 4;
4226 #else
4227         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 3;
4228 #endif
4229         sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4230         if (!sit_i->bitmap)
4231                 return -ENOMEM;
4232
4233         bitmap = sit_i->bitmap;
4234
4235         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4236                 sit_i->sentries[start].cur_valid_map = bitmap;
4237                 bitmap += SIT_VBLOCK_MAP_SIZE;
4238
4239                 sit_i->sentries[start].ckpt_valid_map = bitmap;
4240                 bitmap += SIT_VBLOCK_MAP_SIZE;
4241
4242 #ifdef CONFIG_F2FS_CHECK_FS
4243                 sit_i->sentries[start].cur_valid_map_mir = bitmap;
4244                 bitmap += SIT_VBLOCK_MAP_SIZE;
4245 #endif
4246
4247                 sit_i->sentries[start].discard_map = bitmap;
4248                 bitmap += SIT_VBLOCK_MAP_SIZE;
4249         }
4250
4251         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4252         if (!sit_i->tmp_map)
4253                 return -ENOMEM;
4254
4255         if (__is_large_section(sbi)) {
4256                 sit_i->sec_entries =
4257                         f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4258                                                       MAIN_SECS(sbi)),
4259                                       GFP_KERNEL);
4260                 if (!sit_i->sec_entries)
4261                         return -ENOMEM;
4262         }
4263
4264         /* get information related with SIT */
4265         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4266
4267         /* setup SIT bitmap from ckeckpoint pack */
4268         sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4269         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4270
4271         sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4272         if (!sit_i->sit_bitmap)
4273                 return -ENOMEM;
4274
4275 #ifdef CONFIG_F2FS_CHECK_FS
4276         sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4277                                         sit_bitmap_size, GFP_KERNEL);
4278         if (!sit_i->sit_bitmap_mir)
4279                 return -ENOMEM;
4280
4281         sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4282                                         main_bitmap_size, GFP_KERNEL);
4283         if (!sit_i->invalid_segmap)
4284                 return -ENOMEM;
4285 #endif
4286
4287         /* init SIT information */
4288         sit_i->s_ops = &default_salloc_ops;
4289
4290         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4291         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4292         sit_i->written_valid_blocks = 0;
4293         sit_i->bitmap_size = sit_bitmap_size;
4294         sit_i->dirty_sentries = 0;
4295         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4296         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4297         sit_i->mounted_time = ktime_get_boottime_seconds();
4298         init_rwsem(&sit_i->sentry_lock);
4299         return 0;
4300 }
4301
4302 static int build_free_segmap(struct f2fs_sb_info *sbi)
4303 {
4304         struct free_segmap_info *free_i;
4305         unsigned int bitmap_size, sec_bitmap_size;
4306
4307         /* allocate memory for free segmap information */
4308         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4309         if (!free_i)
4310                 return -ENOMEM;
4311
4312         SM_I(sbi)->free_info = free_i;
4313
4314         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4315         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4316         if (!free_i->free_segmap)
4317                 return -ENOMEM;
4318
4319         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4320         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4321         if (!free_i->free_secmap)
4322                 return -ENOMEM;
4323
4324         /* set all segments as dirty temporarily */
4325         memset(free_i->free_segmap, 0xff, bitmap_size);
4326         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4327
4328         /* init free segmap information */
4329         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4330         free_i->free_segments = 0;
4331         free_i->free_sections = 0;
4332         spin_lock_init(&free_i->segmap_lock);
4333         return 0;
4334 }
4335
4336 static int build_curseg(struct f2fs_sb_info *sbi)
4337 {
4338         struct curseg_info *array;
4339         int i;
4340
4341         array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4342                                         sizeof(*array)), GFP_KERNEL);
4343         if (!array)
4344                 return -ENOMEM;
4345
4346         SM_I(sbi)->curseg_array = array;
4347
4348         for (i = 0; i < NO_CHECK_TYPE; i++) {
4349                 mutex_init(&array[i].curseg_mutex);
4350                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4351                 if (!array[i].sum_blk)
4352                         return -ENOMEM;
4353                 init_rwsem(&array[i].journal_rwsem);
4354                 array[i].journal = f2fs_kzalloc(sbi,
4355                                 sizeof(struct f2fs_journal), GFP_KERNEL);
4356                 if (!array[i].journal)
4357                         return -ENOMEM;
4358                 if (i < NR_PERSISTENT_LOG)
4359                         array[i].seg_type = CURSEG_HOT_DATA + i;
4360                 else if (i == CURSEG_COLD_DATA_PINNED)
4361                         array[i].seg_type = CURSEG_COLD_DATA;
4362                 else if (i == CURSEG_ALL_DATA_ATGC)
4363                         array[i].seg_type = CURSEG_COLD_DATA;
4364                 array[i].segno = NULL_SEGNO;
4365                 array[i].next_blkoff = 0;
4366                 array[i].inited = false;
4367         }
4368         return restore_curseg_summaries(sbi);
4369 }
4370
4371 static int build_sit_entries(struct f2fs_sb_info *sbi)
4372 {
4373         struct sit_info *sit_i = SIT_I(sbi);
4374         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4375         struct f2fs_journal *journal = curseg->journal;
4376         struct seg_entry *se;
4377         struct f2fs_sit_entry sit;
4378         int sit_blk_cnt = SIT_BLK_CNT(sbi);
4379         unsigned int i, start, end;
4380         unsigned int readed, start_blk = 0;
4381         int err = 0;
4382         block_t total_node_blocks = 0;
4383
4384         do {
4385                 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
4386                                                         META_SIT, true);
4387
4388                 start = start_blk * sit_i->sents_per_block;
4389                 end = (start_blk + readed) * sit_i->sents_per_block;
4390
4391                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4392                         struct f2fs_sit_block *sit_blk;
4393                         struct page *page;
4394
4395                         se = &sit_i->sentries[start];
4396                         page = get_current_sit_page(sbi, start);
4397                         if (IS_ERR(page))
4398                                 return PTR_ERR(page);
4399                         sit_blk = (struct f2fs_sit_block *)page_address(page);
4400                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4401                         f2fs_put_page(page, 1);
4402
4403                         err = check_block_count(sbi, start, &sit);
4404                         if (err)
4405                                 return err;
4406                         seg_info_from_raw_sit(se, &sit);
4407                         if (IS_NODESEG(se->type))
4408                                 total_node_blocks += se->valid_blocks;
4409
4410                         /* build discard map only one time */
4411                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4412                                 memset(se->discard_map, 0xff,
4413                                         SIT_VBLOCK_MAP_SIZE);
4414                         } else {
4415                                 memcpy(se->discard_map,
4416                                         se->cur_valid_map,
4417                                         SIT_VBLOCK_MAP_SIZE);
4418                                 sbi->discard_blks +=
4419                                         sbi->blocks_per_seg -
4420                                         se->valid_blocks;
4421                         }
4422
4423                         if (__is_large_section(sbi))
4424                                 get_sec_entry(sbi, start)->valid_blocks +=
4425                                                         se->valid_blocks;
4426                 }
4427                 start_blk += readed;
4428         } while (start_blk < sit_blk_cnt);
4429
4430         down_read(&curseg->journal_rwsem);
4431         for (i = 0; i < sits_in_cursum(journal); i++) {
4432                 unsigned int old_valid_blocks;
4433
4434                 start = le32_to_cpu(segno_in_journal(journal, i));
4435                 if (start >= MAIN_SEGS(sbi)) {
4436                         f2fs_err(sbi, "Wrong journal entry on segno %u",
4437                                  start);
4438                         err = -EFSCORRUPTED;
4439                         break;
4440                 }
4441
4442                 se = &sit_i->sentries[start];
4443                 sit = sit_in_journal(journal, i);
4444
4445                 old_valid_blocks = se->valid_blocks;
4446                 if (IS_NODESEG(se->type))
4447                         total_node_blocks -= old_valid_blocks;
4448
4449                 err = check_block_count(sbi, start, &sit);
4450                 if (err)
4451                         break;
4452                 seg_info_from_raw_sit(se, &sit);
4453                 if (IS_NODESEG(se->type))
4454                         total_node_blocks += se->valid_blocks;
4455
4456                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4457                         memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4458                 } else {
4459                         memcpy(se->discard_map, se->cur_valid_map,
4460                                                 SIT_VBLOCK_MAP_SIZE);
4461                         sbi->discard_blks += old_valid_blocks;
4462                         sbi->discard_blks -= se->valid_blocks;
4463                 }
4464
4465                 if (__is_large_section(sbi)) {
4466                         get_sec_entry(sbi, start)->valid_blocks +=
4467                                                         se->valid_blocks;
4468                         get_sec_entry(sbi, start)->valid_blocks -=
4469                                                         old_valid_blocks;
4470                 }
4471         }
4472         up_read(&curseg->journal_rwsem);
4473
4474         if (!err && total_node_blocks != valid_node_count(sbi)) {
4475                 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4476                          total_node_blocks, valid_node_count(sbi));
4477                 err = -EFSCORRUPTED;
4478         }
4479
4480         return err;
4481 }
4482
4483 static void init_free_segmap(struct f2fs_sb_info *sbi)
4484 {
4485         unsigned int start;
4486         int type;
4487         struct seg_entry *sentry;
4488
4489         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4490                 if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4491                         continue;
4492                 sentry = get_seg_entry(sbi, start);
4493                 if (!sentry->valid_blocks)
4494                         __set_free(sbi, start);
4495                 else
4496                         SIT_I(sbi)->written_valid_blocks +=
4497                                                 sentry->valid_blocks;
4498         }
4499
4500         /* set use the current segments */
4501         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4502                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4503                 __set_test_and_inuse(sbi, curseg_t->segno);
4504         }
4505 }
4506
4507 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4508 {
4509         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4510         struct free_segmap_info *free_i = FREE_I(sbi);
4511         unsigned int segno = 0, offset = 0, secno;
4512         block_t valid_blocks, usable_blks_in_seg;
4513         block_t blks_per_sec = BLKS_PER_SEC(sbi);
4514
4515         while (1) {
4516                 /* find dirty segment based on free segmap */
4517                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4518                 if (segno >= MAIN_SEGS(sbi))
4519                         break;
4520                 offset = segno + 1;
4521                 valid_blocks = get_valid_blocks(sbi, segno, false);
4522                 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4523                 if (valid_blocks == usable_blks_in_seg || !valid_blocks)
4524                         continue;
4525                 if (valid_blocks > usable_blks_in_seg) {
4526                         f2fs_bug_on(sbi, 1);
4527                         continue;
4528                 }
4529                 mutex_lock(&dirty_i->seglist_lock);
4530                 __locate_dirty_segment(sbi, segno, DIRTY);
4531                 mutex_unlock(&dirty_i->seglist_lock);
4532         }
4533
4534         if (!__is_large_section(sbi))
4535                 return;
4536
4537         mutex_lock(&dirty_i->seglist_lock);
4538         for (segno = 0; segno < MAIN_SECS(sbi); segno += blks_per_sec) {
4539                 valid_blocks = get_valid_blocks(sbi, segno, true);
4540                 secno = GET_SEC_FROM_SEG(sbi, segno);
4541
4542                 if (!valid_blocks || valid_blocks == blks_per_sec)
4543                         continue;
4544                 if (IS_CURSEC(sbi, secno))
4545                         continue;
4546                 set_bit(secno, dirty_i->dirty_secmap);
4547         }
4548         mutex_unlock(&dirty_i->seglist_lock);
4549 }
4550
4551 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4552 {
4553         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4554         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4555
4556         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4557         if (!dirty_i->victim_secmap)
4558                 return -ENOMEM;
4559         return 0;
4560 }
4561
4562 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4563 {
4564         struct dirty_seglist_info *dirty_i;
4565         unsigned int bitmap_size, i;
4566
4567         /* allocate memory for dirty segments list information */
4568         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4569                                                                 GFP_KERNEL);
4570         if (!dirty_i)
4571                 return -ENOMEM;
4572
4573         SM_I(sbi)->dirty_info = dirty_i;
4574         mutex_init(&dirty_i->seglist_lock);
4575
4576         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4577
4578         for (i = 0; i < NR_DIRTY_TYPE; i++) {
4579                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4580                                                                 GFP_KERNEL);
4581                 if (!dirty_i->dirty_segmap[i])
4582                         return -ENOMEM;
4583         }
4584
4585         if (__is_large_section(sbi)) {
4586                 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4587                 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
4588                                                 bitmap_size, GFP_KERNEL);
4589                 if (!dirty_i->dirty_secmap)
4590                         return -ENOMEM;
4591         }
4592
4593         init_dirty_segmap(sbi);
4594         return init_victim_secmap(sbi);
4595 }
4596
4597 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4598 {
4599         int i;
4600
4601         /*
4602          * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4603          * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4604          */
4605         for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4606                 struct curseg_info *curseg = CURSEG_I(sbi, i);
4607                 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4608                 unsigned int blkofs = curseg->next_blkoff;
4609
4610                 sanity_check_seg_type(sbi, curseg->seg_type);
4611
4612                 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4613                         goto out;
4614
4615                 if (curseg->alloc_type == SSR)
4616                         continue;
4617
4618                 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4619                         if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4620                                 continue;
4621 out:
4622                         f2fs_err(sbi,
4623                                  "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4624                                  i, curseg->segno, curseg->alloc_type,
4625                                  curseg->next_blkoff, blkofs);
4626                         return -EFSCORRUPTED;
4627                 }
4628         }
4629         return 0;
4630 }
4631
4632 #ifdef CONFIG_BLK_DEV_ZONED
4633
4634 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4635                                     struct f2fs_dev_info *fdev,
4636                                     struct blk_zone *zone)
4637 {
4638         unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno;
4639         block_t zone_block, wp_block, last_valid_block;
4640         unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4641         int i, s, b, ret;
4642         struct seg_entry *se;
4643
4644         if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4645                 return 0;
4646
4647         wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block);
4648         wp_segno = GET_SEGNO(sbi, wp_block);
4649         wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4650         zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4651         zone_segno = GET_SEGNO(sbi, zone_block);
4652         zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno);
4653
4654         if (zone_segno >= MAIN_SEGS(sbi))
4655                 return 0;
4656
4657         /*
4658          * Skip check of zones cursegs point to, since
4659          * fix_curseg_write_pointer() checks them.
4660          */
4661         for (i = 0; i < NO_CHECK_TYPE; i++)
4662                 if (zone_secno == GET_SEC_FROM_SEG(sbi,
4663                                                    CURSEG_I(sbi, i)->segno))
4664                         return 0;
4665
4666         /*
4667          * Get last valid block of the zone.
4668          */
4669         last_valid_block = zone_block - 1;
4670         for (s = sbi->segs_per_sec - 1; s >= 0; s--) {
4671                 segno = zone_segno + s;
4672                 se = get_seg_entry(sbi, segno);
4673                 for (b = sbi->blocks_per_seg - 1; b >= 0; b--)
4674                         if (f2fs_test_bit(b, se->cur_valid_map)) {
4675                                 last_valid_block = START_BLOCK(sbi, segno) + b;
4676                                 break;
4677                         }
4678                 if (last_valid_block >= zone_block)
4679                         break;
4680         }
4681
4682         /*
4683          * If last valid block is beyond the write pointer, report the
4684          * inconsistency. This inconsistency does not cause write error
4685          * because the zone will not be selected for write operation until
4686          * it get discarded. Just report it.
4687          */
4688         if (last_valid_block >= wp_block) {
4689                 f2fs_notice(sbi, "Valid block beyond write pointer: "
4690                             "valid block[0x%x,0x%x] wp[0x%x,0x%x]",
4691                             GET_SEGNO(sbi, last_valid_block),
4692                             GET_BLKOFF_FROM_SEG0(sbi, last_valid_block),
4693                             wp_segno, wp_blkoff);
4694                 return 0;
4695         }
4696
4697         /*
4698          * If there is no valid block in the zone and if write pointer is
4699          * not at zone start, reset the write pointer.
4700          */
4701         if (last_valid_block + 1 == zone_block && zone->wp != zone->start) {
4702                 f2fs_notice(sbi,
4703                             "Zone without valid block has non-zero write "
4704                             "pointer. Reset the write pointer: wp[0x%x,0x%x]",
4705                             wp_segno, wp_blkoff);
4706                 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
4707                                         zone->len >> log_sectors_per_block);
4708                 if (ret) {
4709                         f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4710                                  fdev->path, ret);
4711                         return ret;
4712                 }
4713         }
4714
4715         return 0;
4716 }
4717
4718 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
4719                                                   block_t zone_blkaddr)
4720 {
4721         int i;
4722
4723         for (i = 0; i < sbi->s_ndevs; i++) {
4724                 if (!bdev_is_zoned(FDEV(i).bdev))
4725                         continue;
4726                 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
4727                                 zone_blkaddr <= FDEV(i).end_blk))
4728                         return &FDEV(i);
4729         }
4730
4731         return NULL;
4732 }
4733
4734 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
4735                               void *data) {
4736         memcpy(data, zone, sizeof(struct blk_zone));
4737         return 0;
4738 }
4739
4740 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
4741 {
4742         struct curseg_info *cs = CURSEG_I(sbi, type);
4743         struct f2fs_dev_info *zbd;
4744         struct blk_zone zone;
4745         unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
4746         block_t cs_zone_block, wp_block;
4747         unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4748         sector_t zone_sector;
4749         int err;
4750
4751         cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4752         cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4753
4754         zbd = get_target_zoned_dev(sbi, cs_zone_block);
4755         if (!zbd)
4756                 return 0;
4757
4758         /* report zone for the sector the curseg points to */
4759         zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4760                 << log_sectors_per_block;
4761         err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4762                                   report_one_zone_cb, &zone);
4763         if (err != 1) {
4764                 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4765                          zbd->path, err);
4766                 return err;
4767         }
4768
4769         if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4770                 return 0;
4771
4772         wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
4773         wp_segno = GET_SEGNO(sbi, wp_block);
4774         wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4775         wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
4776
4777         if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
4778                 wp_sector_off == 0)
4779                 return 0;
4780
4781         f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
4782                     "curseg[0x%x,0x%x] wp[0x%x,0x%x]",
4783                     type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff);
4784
4785         f2fs_notice(sbi, "Assign new section to curseg[%d]: "
4786                     "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff);
4787         allocate_segment_by_default(sbi, type, true);
4788
4789         /* check consistency of the zone curseg pointed to */
4790         if (check_zone_write_pointer(sbi, zbd, &zone))
4791                 return -EIO;
4792
4793         /* check newly assigned zone */
4794         cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4795         cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4796
4797         zbd = get_target_zoned_dev(sbi, cs_zone_block);
4798         if (!zbd)
4799                 return 0;
4800
4801         zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4802                 << log_sectors_per_block;
4803         err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4804                                   report_one_zone_cb, &zone);
4805         if (err != 1) {
4806                 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4807                          zbd->path, err);
4808                 return err;
4809         }
4810
4811         if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4812                 return 0;
4813
4814         if (zone.wp != zone.start) {
4815                 f2fs_notice(sbi,
4816                             "New zone for curseg[%d] is not yet discarded. "
4817                             "Reset the zone: curseg[0x%x,0x%x]",
4818                             type, cs->segno, cs->next_blkoff);
4819                 err = __f2fs_issue_discard_zone(sbi, zbd->bdev,
4820                                 zone_sector >> log_sectors_per_block,
4821                                 zone.len >> log_sectors_per_block);
4822                 if (err) {
4823                         f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4824                                  zbd->path, err);
4825                         return err;
4826                 }
4827         }
4828
4829         return 0;
4830 }
4831
4832 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4833 {
4834         int i, ret;
4835
4836         for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4837                 ret = fix_curseg_write_pointer(sbi, i);
4838                 if (ret)
4839                         return ret;
4840         }
4841
4842         return 0;
4843 }
4844
4845 struct check_zone_write_pointer_args {
4846         struct f2fs_sb_info *sbi;
4847         struct f2fs_dev_info *fdev;
4848 };
4849
4850 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
4851                                       void *data) {
4852         struct check_zone_write_pointer_args *args;
4853         args = (struct check_zone_write_pointer_args *)data;
4854
4855         return check_zone_write_pointer(args->sbi, args->fdev, zone);
4856 }
4857
4858 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4859 {
4860         int i, ret;
4861         struct check_zone_write_pointer_args args;
4862
4863         for (i = 0; i < sbi->s_ndevs; i++) {
4864                 if (!bdev_is_zoned(FDEV(i).bdev))
4865                         continue;
4866
4867                 args.sbi = sbi;
4868                 args.fdev = &FDEV(i);
4869                 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
4870                                           check_zone_write_pointer_cb, &args);
4871                 if (ret < 0)
4872                         return ret;
4873         }
4874
4875         return 0;
4876 }
4877
4878 static bool is_conv_zone(struct f2fs_sb_info *sbi, unsigned int zone_idx,
4879                                                 unsigned int dev_idx)
4880 {
4881         if (!bdev_is_zoned(FDEV(dev_idx).bdev))
4882                 return true;
4883         return !test_bit(zone_idx, FDEV(dev_idx).blkz_seq);
4884 }
4885
4886 /* Return the zone index in the given device */
4887 static unsigned int get_zone_idx(struct f2fs_sb_info *sbi, unsigned int secno,
4888                                         int dev_idx)
4889 {
4890         block_t sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
4891
4892         return (sec_start_blkaddr - FDEV(dev_idx).start_blk) >>
4893                                                 sbi->log_blocks_per_blkz;
4894 }
4895
4896 /*
4897  * Return the usable segments in a section based on the zone's
4898  * corresponding zone capacity. Zone is equal to a section.
4899  */
4900 static inline unsigned int f2fs_usable_zone_segs_in_sec(
4901                 struct f2fs_sb_info *sbi, unsigned int segno)
4902 {
4903         unsigned int dev_idx, zone_idx, unusable_segs_in_sec;
4904
4905         dev_idx = f2fs_target_device_index(sbi, START_BLOCK(sbi, segno));
4906         zone_idx = get_zone_idx(sbi, GET_SEC_FROM_SEG(sbi, segno), dev_idx);
4907
4908         /* Conventional zone's capacity is always equal to zone size */
4909         if (is_conv_zone(sbi, zone_idx, dev_idx))
4910                 return sbi->segs_per_sec;
4911
4912         /*
4913          * If the zone_capacity_blocks array is NULL, then zone capacity
4914          * is equal to the zone size for all zones
4915          */
4916         if (!FDEV(dev_idx).zone_capacity_blocks)
4917                 return sbi->segs_per_sec;
4918
4919         /* Get the segment count beyond zone capacity block */
4920         unusable_segs_in_sec = (sbi->blocks_per_blkz -
4921                                 FDEV(dev_idx).zone_capacity_blocks[zone_idx]) >>
4922                                 sbi->log_blocks_per_seg;
4923         return sbi->segs_per_sec - unusable_segs_in_sec;
4924 }
4925
4926 /*
4927  * Return the number of usable blocks in a segment. The number of blocks
4928  * returned is always equal to the number of blocks in a segment for
4929  * segments fully contained within a sequential zone capacity or a
4930  * conventional zone. For segments partially contained in a sequential
4931  * zone capacity, the number of usable blocks up to the zone capacity
4932  * is returned. 0 is returned in all other cases.
4933  */
4934 static inline unsigned int f2fs_usable_zone_blks_in_seg(
4935                         struct f2fs_sb_info *sbi, unsigned int segno)
4936 {
4937         block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
4938         unsigned int zone_idx, dev_idx, secno;
4939
4940         secno = GET_SEC_FROM_SEG(sbi, segno);
4941         seg_start = START_BLOCK(sbi, segno);
4942         dev_idx = f2fs_target_device_index(sbi, seg_start);
4943         zone_idx = get_zone_idx(sbi, secno, dev_idx);
4944
4945         /*
4946          * Conventional zone's capacity is always equal to zone size,
4947          * so, blocks per segment is unchanged.
4948          */
4949         if (is_conv_zone(sbi, zone_idx, dev_idx))
4950                 return sbi->blocks_per_seg;
4951
4952         if (!FDEV(dev_idx).zone_capacity_blocks)
4953                 return sbi->blocks_per_seg;
4954
4955         sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
4956         sec_cap_blkaddr = sec_start_blkaddr +
4957                                 FDEV(dev_idx).zone_capacity_blocks[zone_idx];
4958
4959         /*
4960          * If segment starts before zone capacity and spans beyond
4961          * zone capacity, then usable blocks are from seg start to
4962          * zone capacity. If the segment starts after the zone capacity,
4963          * then there are no usable blocks.
4964          */
4965         if (seg_start >= sec_cap_blkaddr)
4966                 return 0;
4967         if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr)
4968                 return sec_cap_blkaddr - seg_start;
4969
4970         return sbi->blocks_per_seg;
4971 }
4972 #else
4973 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4974 {
4975         return 0;
4976 }
4977
4978 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4979 {
4980         return 0;
4981 }
4982
4983 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
4984                                                         unsigned int segno)
4985 {
4986         return 0;
4987 }
4988
4989 static inline unsigned int f2fs_usable_zone_segs_in_sec(struct f2fs_sb_info *sbi,
4990                                                         unsigned int segno)
4991 {
4992         return 0;
4993 }
4994 #endif
4995 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
4996                                         unsigned int segno)
4997 {
4998         if (f2fs_sb_has_blkzoned(sbi))
4999                 return f2fs_usable_zone_blks_in_seg(sbi, segno);
5000
5001         return sbi->blocks_per_seg;
5002 }
5003
5004 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi,
5005                                         unsigned int segno)
5006 {
5007         if (f2fs_sb_has_blkzoned(sbi))
5008                 return f2fs_usable_zone_segs_in_sec(sbi, segno);
5009
5010         return sbi->segs_per_sec;
5011 }
5012
5013 /*
5014  * Update min, max modified time for cost-benefit GC algorithm
5015  */
5016 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5017 {
5018         struct sit_info *sit_i = SIT_I(sbi);
5019         unsigned int segno;
5020
5021         down_write(&sit_i->sentry_lock);
5022
5023         sit_i->min_mtime = ULLONG_MAX;
5024
5025         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
5026                 unsigned int i;
5027                 unsigned long long mtime = 0;
5028
5029                 for (i = 0; i < sbi->segs_per_sec; i++)
5030                         mtime += get_seg_entry(sbi, segno + i)->mtime;
5031
5032                 mtime = div_u64(mtime, sbi->segs_per_sec);
5033
5034                 if (sit_i->min_mtime > mtime)
5035                         sit_i->min_mtime = mtime;
5036         }
5037         sit_i->max_mtime = get_mtime(sbi, false);
5038         sit_i->dirty_max_mtime = 0;
5039         up_write(&sit_i->sentry_lock);
5040 }
5041
5042 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
5043 {
5044         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5045         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
5046         struct f2fs_sm_info *sm_info;
5047         int err;
5048
5049         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
5050         if (!sm_info)
5051                 return -ENOMEM;
5052
5053         /* init sm info */
5054         sbi->sm_info = sm_info;
5055         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5056         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5057         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5058         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5059         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5060         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5061         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
5062         sm_info->rec_prefree_segments = sm_info->main_segments *
5063                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
5064         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5065                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5066
5067         if (!f2fs_lfs_mode(sbi))
5068                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
5069         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
5070         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
5071         sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
5072         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
5073         sm_info->min_ssr_sections = reserved_sections(sbi);
5074
5075         INIT_LIST_HEAD(&sm_info->sit_entry_set);
5076
5077         init_rwsem(&sm_info->curseg_lock);
5078
5079         if (!f2fs_readonly(sbi->sb)) {
5080                 err = f2fs_create_flush_cmd_control(sbi);
5081                 if (err)
5082                         return err;
5083         }
5084
5085         err = create_discard_cmd_control(sbi);
5086         if (err)
5087                 return err;
5088
5089         err = build_sit_info(sbi);
5090         if (err)
5091                 return err;
5092         err = build_free_segmap(sbi);
5093         if (err)
5094                 return err;
5095         err = build_curseg(sbi);
5096         if (err)
5097                 return err;
5098
5099         /* reinit free segmap based on SIT */
5100         err = build_sit_entries(sbi);
5101         if (err)
5102                 return err;
5103
5104         init_free_segmap(sbi);
5105         err = build_dirty_segmap(sbi);
5106         if (err)
5107                 return err;
5108
5109         err = sanity_check_curseg(sbi);
5110         if (err)
5111                 return err;
5112
5113         init_min_max_mtime(sbi);
5114         return 0;
5115 }
5116
5117 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5118                 enum dirty_type dirty_type)
5119 {
5120         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5121
5122         mutex_lock(&dirty_i->seglist_lock);
5123         kvfree(dirty_i->dirty_segmap[dirty_type]);
5124         dirty_i->nr_dirty[dirty_type] = 0;
5125         mutex_unlock(&dirty_i->seglist_lock);
5126 }
5127
5128 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5129 {
5130         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5131         kvfree(dirty_i->victim_secmap);
5132 }
5133
5134 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5135 {
5136         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5137         int i;
5138
5139         if (!dirty_i)
5140                 return;
5141
5142         /* discard pre-free/dirty segments list */
5143         for (i = 0; i < NR_DIRTY_TYPE; i++)
5144                 discard_dirty_segmap(sbi, i);
5145
5146         if (__is_large_section(sbi)) {
5147                 mutex_lock(&dirty_i->seglist_lock);
5148                 kvfree(dirty_i->dirty_secmap);
5149                 mutex_unlock(&dirty_i->seglist_lock);
5150         }
5151
5152         destroy_victim_secmap(sbi);
5153         SM_I(sbi)->dirty_info = NULL;
5154         kvfree(dirty_i);
5155 }
5156
5157 static void destroy_curseg(struct f2fs_sb_info *sbi)
5158 {
5159         struct curseg_info *array = SM_I(sbi)->curseg_array;
5160         int i;
5161
5162         if (!array)
5163                 return;
5164         SM_I(sbi)->curseg_array = NULL;
5165         for (i = 0; i < NR_CURSEG_TYPE; i++) {
5166                 kvfree(array[i].sum_blk);
5167                 kvfree(array[i].journal);
5168         }
5169         kvfree(array);
5170 }
5171
5172 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5173 {
5174         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5175         if (!free_i)
5176                 return;
5177         SM_I(sbi)->free_info = NULL;
5178         kvfree(free_i->free_segmap);
5179         kvfree(free_i->free_secmap);
5180         kvfree(free_i);
5181 }
5182
5183 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5184 {
5185         struct sit_info *sit_i = SIT_I(sbi);
5186
5187         if (!sit_i)
5188                 return;
5189
5190         if (sit_i->sentries)
5191                 kvfree(sit_i->bitmap);
5192         kvfree(sit_i->tmp_map);
5193
5194         kvfree(sit_i->sentries);
5195         kvfree(sit_i->sec_entries);
5196         kvfree(sit_i->dirty_sentries_bitmap);
5197
5198         SM_I(sbi)->sit_info = NULL;
5199         kvfree(sit_i->sit_bitmap);
5200 #ifdef CONFIG_F2FS_CHECK_FS
5201         kvfree(sit_i->sit_bitmap_mir);
5202         kvfree(sit_i->invalid_segmap);
5203 #endif
5204         kvfree(sit_i);
5205 }
5206
5207 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5208 {
5209         struct f2fs_sm_info *sm_info = SM_I(sbi);
5210
5211         if (!sm_info)
5212                 return;
5213         f2fs_destroy_flush_cmd_control(sbi, true);
5214         destroy_discard_cmd_control(sbi);
5215         destroy_dirty_segmap(sbi);
5216         destroy_curseg(sbi);
5217         destroy_free_segmap(sbi);
5218         destroy_sit_info(sbi);
5219         sbi->sm_info = NULL;
5220         kvfree(sm_info);
5221 }
5222
5223 int __init f2fs_create_segment_manager_caches(void)
5224 {
5225         discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5226                         sizeof(struct discard_entry));
5227         if (!discard_entry_slab)
5228                 goto fail;
5229
5230         discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5231                         sizeof(struct discard_cmd));
5232         if (!discard_cmd_slab)
5233                 goto destroy_discard_entry;
5234
5235         sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5236                         sizeof(struct sit_entry_set));
5237         if (!sit_entry_set_slab)
5238                 goto destroy_discard_cmd;
5239
5240         inmem_entry_slab = f2fs_kmem_cache_create("f2fs_inmem_page_entry",
5241                         sizeof(struct inmem_pages));
5242         if (!inmem_entry_slab)
5243                 goto destroy_sit_entry_set;
5244         return 0;
5245
5246 destroy_sit_entry_set:
5247         kmem_cache_destroy(sit_entry_set_slab);
5248 destroy_discard_cmd:
5249         kmem_cache_destroy(discard_cmd_slab);
5250 destroy_discard_entry:
5251         kmem_cache_destroy(discard_entry_slab);
5252 fail:
5253         return -ENOMEM;
5254 }
5255
5256 void f2fs_destroy_segment_manager_caches(void)
5257 {
5258         kmem_cache_destroy(sit_entry_set_slab);
5259         kmem_cache_destroy(discard_cmd_slab);
5260         kmem_cache_destroy(discard_entry_slab);
5261         kmem_cache_destroy(inmem_entry_slab);
5262 }