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/events/f2fs.h>
  24
  25 #define __reverse_ffz(x) __reverse_ffs(~(x))
  26
  27 static struct kmem_cache *discard_entry_slab;
  28 static struct kmem_cache *discard_cmd_slab;
  29 static struct kmem_cache *sit_entry_set_slab;
  30 static struct kmem_cache *inmem_entry_slab;
  31
  32 static unsigned long __reverse_ulong(unsigned char *str)
  33 {
  34         unsigned long tmp = 0;
  35         int shift = 24, idx = 0;
  36
  37 #if BITS_PER_LONG == 64
  38         shift = 56;
  39 #endif
  40         while (shift >= 0) {
  41                 tmp |= (unsigned long)str[idx++] << shift;
  42                 shift -= BITS_PER_BYTE;
  43         }
  44         return tmp;
  45 }
  46
  47 /*
  48  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  49  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  50  */
  51 static inline unsigned long __reverse_ffs(unsigned long word)
  52 {
  53         int num = 0;
  54
  55 #if BITS_PER_LONG == 64
  56         if ((word & 0xffffffff00000000UL) == 0)
  57                 num += 32;
  58         else
  59                 word >>= 32;
  60 #endif
  61         if ((word & 0xffff0000) == 0)
  62                 num += 16;
  63         else
  64                 word >>= 16;
  65
  66         if ((word & 0xff00) == 0)
  67                 num += 8;
  68         else
  69                 word >>= 8;
  70
  71         if ((word & 0xf0) == 0)
  72                 num += 4;
  73         else
  74                 word >>= 4;
  75
  76         if ((word & 0xc) == 0)
  77                 num += 2;
  78         else
  79                 word >>= 2;
  80
  81         if ((word & 0x2) == 0)
  82                 num += 1;
  83         return num;
  84 }
  85
  86 /*
  87  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  88  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  89  * @size must be integral times of unsigned long.
  90  * Example:
  91  *                             MSB <--> LSB
  92  *   f2fs_set_bit(0, bitmap) => 1000 0000
  93  *   f2fs_set_bit(7, bitmap) => 0000 0001
  94  */
  95 static unsigned long __find_rev_next_bit(const unsigned long *addr,
  96                         unsigned long size, unsigned long offset)
  97 {
  98         const unsigned long *p = addr + BIT_WORD(offset);
  99         unsigned long result = size;
 100         unsigned long tmp;
 101
 102         if (offset >= size)
 103                 return size;
 104
 105         size -= (offset & ~(BITS_PER_LONG - 1));
 106         offset %= BITS_PER_LONG;
 107
 108         while (1) {
 109                 if (*p == 0)
 110                         goto pass;
 111
 112                 tmp = __reverse_ulong((unsigned char *)p);
 113
 114                 tmp &= ~0UL >> offset;
 115                 if (size < BITS_PER_LONG)
 116                         tmp &= (~0UL << (BITS_PER_LONG - size));
 117                 if (tmp)
 118                         goto found;
 119 pass:
 120                 if (size <= BITS_PER_LONG)
 121                         break;
 122                 size -= BITS_PER_LONG;
 123                 offset = 0;
 124                 p++;
 125         }
 126         return result;
 127 found:
 128         return result - size + __reverse_ffs(tmp);
 129 }
 130
 131 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 132                         unsigned long size, unsigned long offset)
 133 {
 134         const unsigned long *p = addr + BIT_WORD(offset);
 135         unsigned long result = size;
 136         unsigned long tmp;
 137
 138         if (offset >= size)
 139                 return size;
 140
 141         size -= (offset & ~(BITS_PER_LONG - 1));
 142         offset %= BITS_PER_LONG;
 143
 144         while (1) {
 145                 if (*p == ~0UL)
 146                         goto pass;
 147
 148                 tmp = __reverse_ulong((unsigned char *)p);
 149
 150                 if (offset)
 151                         tmp |= ~0UL << (BITS_PER_LONG - offset);
 152                 if (size < BITS_PER_LONG)
 153                         tmp |= ~0UL >> size;
 154                 if (tmp != ~0UL)
 155                         goto found;
 156 pass:
 157                 if (size <= BITS_PER_LONG)
 158                         break;
 159                 size -= BITS_PER_LONG;
 160                 offset = 0;
 161                 p++;
 162         }
 163         return result;
 164 found:
 165         return result - size + __reverse_ffz(tmp);
 166 }
 167
 168 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
 169 {
 170         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 171         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 172         int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
 173
 174         if (f2fs_lfs_mode(sbi))
 175                 return false;
 176         if (sbi->gc_mode == GC_URGENT_HIGH)
 177                 return true;
 178         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
 179                 return true;
 180
 181         return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
 182                         SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
 183 }
 184
 185 void f2fs_register_inmem_page(struct inode *inode, struct page *page)
 186 {
 187         struct inmem_pages *new;
 188
 189         if (PagePrivate(page))
 190                 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
 191         else
 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         do {
 331                 mutex_lock(&fi->inmem_lock);
 332                 if (list_empty(&fi->inmem_pages)) {
 333                         fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
 334
 335                         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 336                         if (!list_empty(&fi->inmem_ilist))
 337                                 list_del_init(&fi->inmem_ilist);
 338                         if (f2fs_is_atomic_file(inode)) {
 339                                 clear_inode_flag(inode, FI_ATOMIC_FILE);
 340                                 sbi->atomic_files--;
 341                         }
 342                         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 343
 344                         mutex_unlock(&fi->inmem_lock);
 345                         break;
 346                 }
 347                 __revoke_inmem_pages(inode, &fi->inmem_pages,
 348                                                 true, false, true);
 349                 mutex_unlock(&fi->inmem_lock);
 350         } while (1);
 351 }
 352
 353 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
 354 {
 355         struct f2fs_inode_info *fi = F2FS_I(inode);
 356         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 357         struct list_head *head = &fi->inmem_pages;
 358         struct inmem_pages *cur = NULL;
 359
 360         f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
 361
 362         mutex_lock(&fi->inmem_lock);
 363         list_for_each_entry(cur, head, list) {
 364                 if (cur->page == page)
 365                         break;
 366         }
 367
 368         f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
 369         list_del(&cur->list);
 370         mutex_unlock(&fi->inmem_lock);
 371
 372         dec_page_count(sbi, F2FS_INMEM_PAGES);
 373         kmem_cache_free(inmem_entry_slab, cur);
 374
 375         ClearPageUptodate(page);
 376         f2fs_clear_page_private(page);
 377         f2fs_put_page(page, 0);
 378
 379         trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 380 }
 381
 382 static int __f2fs_commit_inmem_pages(struct inode *inode)
 383 {
 384         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 385         struct f2fs_inode_info *fi = F2FS_I(inode);
 386         struct inmem_pages *cur, *tmp;
 387         struct f2fs_io_info fio = {
 388                 .sbi = sbi,
 389                 .ino = inode->i_ino,
 390                 .type = DATA,
 391                 .op = REQ_OP_WRITE,
 392                 .op_flags = REQ_SYNC | REQ_PRIO,
 393                 .io_type = FS_DATA_IO,
 394         };
 395         struct list_head revoke_list;
 396         bool submit_bio = false;
 397         int err = 0;
 398
 399         INIT_LIST_HEAD(&revoke_list);
 400
 401         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 402                 struct page *page = cur->page;
 403
 404                 lock_page(page);
 405                 if (page->mapping == inode->i_mapping) {
 406                         trace_f2fs_commit_inmem_page(page, INMEM);
 407
 408                         f2fs_wait_on_page_writeback(page, DATA, true, true);
 409
 410                         set_page_dirty(page);
 411                         if (clear_page_dirty_for_io(page)) {
 412                                 inode_dec_dirty_pages(inode);
 413                                 f2fs_remove_dirty_inode(inode);
 414                         }
 415 retry:
 416                         fio.page = page;
 417                         fio.old_blkaddr = NULL_ADDR;
 418                         fio.encrypted_page = NULL;
 419                         fio.need_lock = LOCK_DONE;
 420                         err = f2fs_do_write_data_page(&fio);
 421                         if (err) {
 422                                 if (err == -ENOMEM) {
 423                                         congestion_wait(BLK_RW_ASYNC,
 424                                                         DEFAULT_IO_TIMEOUT);
 425                                         cond_resched();
 426                                         goto retry;
 427                                 }
 428                                 unlock_page(page);
 429                                 break;
 430                         }
 431                         /* record old blkaddr for revoking */
 432                         cur->old_addr = fio.old_blkaddr;
 433                         submit_bio = true;
 434                 }
 435                 unlock_page(page);
 436                 list_move_tail(&cur->list, &revoke_list);
 437         }
 438
 439         if (submit_bio)
 440                 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
 441
 442         if (err) {
 443                 /*
 444                  * try to revoke all committed pages, but still we could fail
 445                  * due to no memory or other reason, if that happened, EAGAIN
 446                  * will be returned, which means in such case, transaction is
 447                  * already not integrity, caller should use journal to do the
 448                  * recovery or rewrite & commit last transaction. For other
 449                  * error number, revoking was done by filesystem itself.
 450                  */
 451                 err = __revoke_inmem_pages(inode, &revoke_list,
 452                                                 false, true, false);
 453
 454                 /* drop all uncommitted pages */
 455                 __revoke_inmem_pages(inode, &fi->inmem_pages,
 456                                                 true, false, false);
 457         } else {
 458                 __revoke_inmem_pages(inode, &revoke_list,
 459                                                 false, false, false);
 460         }
 461
 462         return err;
 463 }
 464
 465 int f2fs_commit_inmem_pages(struct inode *inode)
 466 {
 467         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 468         struct f2fs_inode_info *fi = F2FS_I(inode);
 469         int err;
 470
 471         f2fs_balance_fs(sbi, true);
 472
 473         down_write(&fi->i_gc_rwsem[WRITE]);
 474
 475         f2fs_lock_op(sbi);
 476         set_inode_flag(inode, FI_ATOMIC_COMMIT);
 477
 478         mutex_lock(&fi->inmem_lock);
 479         err = __f2fs_commit_inmem_pages(inode);
 480         mutex_unlock(&fi->inmem_lock);
 481
 482         clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 483
 484         f2fs_unlock_op(sbi);
 485         up_write(&fi->i_gc_rwsem[WRITE]);
 486
 487         return err;
 488 }
 489
 490 /*
 491  * This function balances dirty node and dentry pages.
 492  * In addition, it controls garbage collection.
 493  */
 494 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 495 {
 496         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 497                 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
 498                 f2fs_stop_checkpoint(sbi, false);
 499         }
 500
 501         /* balance_fs_bg is able to be pending */
 502         if (need && excess_cached_nats(sbi))
 503                 f2fs_balance_fs_bg(sbi, false);
 504
 505         if (!f2fs_is_checkpoint_ready(sbi))
 506                 return;
 507
 508         /*
 509          * We should do GC or end up with checkpoint, if there are so many dirty
 510          * dir/node pages without enough free segments.
 511          */
 512         if (has_not_enough_free_secs(sbi, 0, 0)) {
 513                 if (test_opt(sbi, GC_MERGE) && sbi->gc_thread &&
 514                                         sbi->gc_thread->f2fs_gc_task) {
 515                         DEFINE_WAIT(wait);
 516
 517                         prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait,
 518                                                 TASK_UNINTERRUPTIBLE);
 519                         wake_up(&sbi->gc_thread->gc_wait_queue_head);
 520                         io_schedule();
 521                         finish_wait(&sbi->gc_thread->fggc_wq, &wait);
 522                 } else {
 523                         down_write(&sbi->gc_lock);
 524                         f2fs_gc(sbi, false, false, false, NULL_SEGNO);
 525                 }
 526         }
 527 }
 528
 529 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
 530 {
 531         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 532                 return;
 533
 534         /* try to shrink extent cache when there is no enough memory */
 535         if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
 536                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 537
 538         /* check the # of cached NAT entries */
 539         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
 540                 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 541
 542         if (!f2fs_available_free_memory(sbi, FREE_NIDS))
 543                 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
 544         else
 545                 f2fs_build_free_nids(sbi, false, false);
 546
 547         if (excess_dirty_nats(sbi) || excess_dirty_nodes(sbi) ||
 548                 excess_prefree_segs(sbi))
 549                 goto do_sync;
 550
 551         /* there is background inflight IO or foreground operation recently */
 552         if (is_inflight_io(sbi, REQ_TIME) ||
 553                 (!f2fs_time_over(sbi, REQ_TIME) && rwsem_is_locked(&sbi->cp_rwsem)))
 554                 return;
 555
 556         /* exceed periodical checkpoint timeout threshold */
 557         if (f2fs_time_over(sbi, CP_TIME))
 558                 goto do_sync;
 559
 560         /* checkpoint is the only way to shrink partial cached entries */
 561         if (f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
 562                 f2fs_available_free_memory(sbi, INO_ENTRIES))
 563                 return;
 564
 565 do_sync:
 566         if (test_opt(sbi, DATA_FLUSH) && from_bg) {
 567                 struct blk_plug plug;
 568
 569                 mutex_lock(&sbi->flush_lock);
 570
 571                 blk_start_plug(&plug);
 572                 f2fs_sync_dirty_inodes(sbi, FILE_INODE);
 573                 blk_finish_plug(&plug);
 574
 575                 mutex_unlock(&sbi->flush_lock);
 576         }
 577         f2fs_sync_fs(sbi->sb, true);
 578         stat_inc_bg_cp_count(sbi->stat_info);
 579 }
 580
 581 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 582                                 struct block_device *bdev)
 583 {
 584         int ret = blkdev_issue_flush(bdev);
 585
 586         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 587                                 test_opt(sbi, FLUSH_MERGE), ret);
 588         return ret;
 589 }
 590
 591 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 592 {
 593         int ret = 0;
 594         int i;
 595
 596         if (!f2fs_is_multi_device(sbi))
 597                 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 598
 599         for (i = 0; i < sbi->s_ndevs; i++) {
 600                 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
 601                         continue;
 602                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 603                 if (ret)
 604                         break;
 605         }
 606         return ret;
 607 }
 608
 609 static int issue_flush_thread(void *data)
 610 {
 611         struct f2fs_sb_info *sbi = data;
 612         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 613         wait_queue_head_t *q = &fcc->flush_wait_queue;
 614 repeat:
 615         if (kthread_should_stop())
 616                 return 0;
 617
 618         if (!llist_empty(&fcc->issue_list)) {
 619                 struct flush_cmd *cmd, *next;
 620                 int ret;
 621
 622                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 623                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 624
 625                 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 626
 627                 ret = submit_flush_wait(sbi, cmd->ino);
 628                 atomic_inc(&fcc->issued_flush);
 629
 630                 llist_for_each_entry_safe(cmd, next,
 631                                           fcc->dispatch_list, llnode) {
 632                         cmd->ret = ret;
 633                         complete(&cmd->wait);
 634                 }
 635                 fcc->dispatch_list = NULL;
 636         }
 637
 638         wait_event_interruptible(*q,
 639                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 640         goto repeat;
 641 }
 642
 643 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 644 {
 645         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 646         struct flush_cmd cmd;
 647         int ret;
 648
 649         if (test_opt(sbi, NOBARRIER))
 650                 return 0;
 651
 652         if (!test_opt(sbi, FLUSH_MERGE)) {
 653                 atomic_inc(&fcc->queued_flush);
 654                 ret = submit_flush_wait(sbi, ino);
 655                 atomic_dec(&fcc->queued_flush);
 656                 atomic_inc(&fcc->issued_flush);
 657                 return ret;
 658         }
 659
 660         if (atomic_inc_return(&fcc->queued_flush) == 1 ||
 661             f2fs_is_multi_device(sbi)) {
 662                 ret = submit_flush_wait(sbi, ino);
 663                 atomic_dec(&fcc->queued_flush);
 664
 665                 atomic_inc(&fcc->issued_flush);
 666                 return ret;
 667         }
 668
 669         cmd.ino = ino;
 670         init_completion(&cmd.wait);
 671
 672         llist_add(&cmd.llnode, &fcc->issue_list);
 673
 674         /*
 675          * update issue_list before we wake up issue_flush thread, this
 676          * smp_mb() pairs with another barrier in ___wait_event(), see
 677          * more details in comments of waitqueue_active().
 678          */
 679         smp_mb();
 680
 681         if (waitqueue_active(&fcc->flush_wait_queue))
 682                 wake_up(&fcc->flush_wait_queue);
 683
 684         if (fcc->f2fs_issue_flush) {
 685                 wait_for_completion(&cmd.wait);
 686                 atomic_dec(&fcc->queued_flush);
 687         } else {
 688                 struct llist_node *list;
 689
 690                 list = llist_del_all(&fcc->issue_list);
 691                 if (!list) {
 692                         wait_for_completion(&cmd.wait);
 693                         atomic_dec(&fcc->queued_flush);
 694                 } else {
 695                         struct flush_cmd *tmp, *next;
 696
 697                         ret = submit_flush_wait(sbi, ino);
 698
 699                         llist_for_each_entry_safe(tmp, next, list, llnode) {
 700                                 if (tmp == &cmd) {
 701                                         cmd.ret = ret;
 702                                         atomic_dec(&fcc->queued_flush);
 703                                         continue;
 704                                 }
 705                                 tmp->ret = ret;
 706                                 complete(&tmp->wait);
 707                         }
 708                 }
 709         }
 710
 711         return cmd.ret;
 712 }
 713
 714 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
 715 {
 716         dev_t dev = sbi->sb->s_bdev->bd_dev;
 717         struct flush_cmd_control *fcc;
 718         int err = 0;
 719
 720         if (SM_I(sbi)->fcc_info) {
 721                 fcc = SM_I(sbi)->fcc_info;
 722                 if (fcc->f2fs_issue_flush)
 723                         return err;
 724                 goto init_thread;
 725         }
 726
 727         fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
 728         if (!fcc)
 729                 return -ENOMEM;
 730         atomic_set(&fcc->issued_flush, 0);
 731         atomic_set(&fcc->queued_flush, 0);
 732         init_waitqueue_head(&fcc->flush_wait_queue);
 733         init_llist_head(&fcc->issue_list);
 734         SM_I(sbi)->fcc_info = fcc;
 735         if (!test_opt(sbi, FLUSH_MERGE))
 736                 return err;
 737
 738 init_thread:
 739         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 740                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 741         if (IS_ERR(fcc->f2fs_issue_flush)) {
 742                 err = PTR_ERR(fcc->f2fs_issue_flush);
 743                 kfree(fcc);
 744                 SM_I(sbi)->fcc_info = NULL;
 745                 return err;
 746         }
 747
 748         return err;
 749 }
 750
 751 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 752 {
 753         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 754
 755         if (fcc && fcc->f2fs_issue_flush) {
 756                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 757
 758                 fcc->f2fs_issue_flush = NULL;
 759                 kthread_stop(flush_thread);
 760         }
 761         if (free) {
 762                 kfree(fcc);
 763                 SM_I(sbi)->fcc_info = NULL;
 764         }
 765 }
 766
 767 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
 768 {
 769         int ret = 0, i;
 770
 771         if (!f2fs_is_multi_device(sbi))
 772                 return 0;
 773
 774         if (test_opt(sbi, NOBARRIER))
 775                 return 0;
 776
 777         for (i = 1; i < sbi->s_ndevs; i++) {
 778                 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
 779                         continue;
 780                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 781                 if (ret)
 782                         break;
 783
 784                 spin_lock(&sbi->dev_lock);
 785                 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
 786                 spin_unlock(&sbi->dev_lock);
 787         }
 788
 789         return ret;
 790 }
 791
 792 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 793                 enum dirty_type dirty_type)
 794 {
 795         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 796
 797         /* need not be added */
 798         if (IS_CURSEG(sbi, segno))
 799                 return;
 800
 801         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 802                 dirty_i->nr_dirty[dirty_type]++;
 803
 804         if (dirty_type == DIRTY) {
 805                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 806                 enum dirty_type t = sentry->type;
 807
 808                 if (unlikely(t >= DIRTY)) {
 809                         f2fs_bug_on(sbi, 1);
 810                         return;
 811                 }
 812                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 813                         dirty_i->nr_dirty[t]++;
 814
 815                 if (__is_large_section(sbi)) {
 816                         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 817                         block_t valid_blocks =
 818                                 get_valid_blocks(sbi, segno, true);
 819
 820                         f2fs_bug_on(sbi, unlikely(!valid_blocks ||
 821                                         valid_blocks == BLKS_PER_SEC(sbi)));
 822
 823                         if (!IS_CURSEC(sbi, secno))
 824                                 set_bit(secno, dirty_i->dirty_secmap);
 825                 }
 826         }
 827 }
 828
 829 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 830                 enum dirty_type dirty_type)
 831 {
 832         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 833         block_t valid_blocks;
 834
 835         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 836                 dirty_i->nr_dirty[dirty_type]--;
 837
 838         if (dirty_type == DIRTY) {
 839                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 840                 enum dirty_type t = sentry->type;
 841
 842                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 843                         dirty_i->nr_dirty[t]--;
 844
 845                 valid_blocks = get_valid_blocks(sbi, segno, true);
 846                 if (valid_blocks == 0) {
 847                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 848                                                 dirty_i->victim_secmap);
 849 #ifdef CONFIG_F2FS_CHECK_FS
 850                         clear_bit(segno, SIT_I(sbi)->invalid_segmap);
 851 #endif
 852                 }
 853                 if (__is_large_section(sbi)) {
 854                         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 855
 856                         if (!valid_blocks ||
 857                                         valid_blocks == BLKS_PER_SEC(sbi)) {
 858                                 clear_bit(secno, dirty_i->dirty_secmap);
 859                                 return;
 860                         }
 861
 862                         if (!IS_CURSEC(sbi, secno))
 863                                 set_bit(secno, dirty_i->dirty_secmap);
 864                 }
 865         }
 866 }
 867
 868 /*
 869  * Should not occur error such as -ENOMEM.
 870  * Adding dirty entry into seglist is not critical operation.
 871  * If a given segment is one of current working segments, it won't be added.
 872  */
 873 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 874 {
 875         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 876         unsigned short valid_blocks, ckpt_valid_blocks;
 877         unsigned int usable_blocks;
 878
 879         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 880                 return;
 881
 882         usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
 883         mutex_lock(&dirty_i->seglist_lock);
 884
 885         valid_blocks = get_valid_blocks(sbi, segno, false);
 886         ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false);
 887
 888         if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
 889                 ckpt_valid_blocks == usable_blocks)) {
 890                 __locate_dirty_segment(sbi, segno, PRE);
 891                 __remove_dirty_segment(sbi, segno, DIRTY);
 892         } else if (valid_blocks < usable_blocks) {
 893                 __locate_dirty_segment(sbi, segno, DIRTY);
 894         } else {
 895                 /* Recovery routine with SSR needs this */
 896                 __remove_dirty_segment(sbi, segno, DIRTY);
 897         }
 898
 899         mutex_unlock(&dirty_i->seglist_lock);
 900 }
 901
 902 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
 903 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
 904 {
 905         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 906         unsigned int segno;
 907
 908         mutex_lock(&dirty_i->seglist_lock);
 909         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 910                 if (get_valid_blocks(sbi, segno, false))
 911                         continue;
 912                 if (IS_CURSEG(sbi, segno))
 913                         continue;
 914                 __locate_dirty_segment(sbi, segno, PRE);
 915                 __remove_dirty_segment(sbi, segno, DIRTY);
 916         }
 917         mutex_unlock(&dirty_i->seglist_lock);
 918 }
 919
 920 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
 921 {
 922         int ovp_hole_segs =
 923                 (overprovision_segments(sbi) - reserved_segments(sbi));
 924         block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
 925         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 926         block_t holes[2] = {0, 0};      /* DATA and NODE */
 927         block_t unusable;
 928         struct seg_entry *se;
 929         unsigned int segno;
 930
 931         mutex_lock(&dirty_i->seglist_lock);
 932         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 933                 se = get_seg_entry(sbi, segno);
 934                 if (IS_NODESEG(se->type))
 935                         holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
 936                                                         se->valid_blocks;
 937                 else
 938                         holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
 939                                                         se->valid_blocks;
 940         }
 941         mutex_unlock(&dirty_i->seglist_lock);
 942
 943         unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
 944         if (unusable > ovp_holes)
 945                 return unusable - ovp_holes;
 946         return 0;
 947 }
 948
 949 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
 950 {
 951         int ovp_hole_segs =
 952                 (overprovision_segments(sbi) - reserved_segments(sbi));
 953         if (unusable > F2FS_OPTION(sbi).unusable_cap)
 954                 return -EAGAIN;
 955         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
 956                 dirty_segments(sbi) > ovp_hole_segs)
 957                 return -EAGAIN;
 958         return 0;
 959 }
 960
 961 /* This is only used by SBI_CP_DISABLED */
 962 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
 963 {
 964         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 965         unsigned int segno = 0;
 966
 967         mutex_lock(&dirty_i->seglist_lock);
 968         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 969                 if (get_valid_blocks(sbi, segno, false))
 970                         continue;
 971                 if (get_ckpt_valid_blocks(sbi, segno, false))
 972                         continue;
 973                 mutex_unlock(&dirty_i->seglist_lock);
 974                 return segno;
 975         }
 976         mutex_unlock(&dirty_i->seglist_lock);
 977         return NULL_SEGNO;
 978 }
 979
 980 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 981                 struct block_device *bdev, block_t lstart,
 982                 block_t start, block_t len)
 983 {
 984         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 985         struct list_head *pend_list;
 986         struct discard_cmd *dc;
 987
 988         f2fs_bug_on(sbi, !len);
 989
 990         pend_list = &dcc->pend_list[plist_idx(len)];
 991
 992         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 993         INIT_LIST_HEAD(&dc->list);
 994         dc->bdev = bdev;
 995         dc->lstart = lstart;
 996         dc->start = start;
 997         dc->len = len;
 998         dc->ref = 0;
 999         dc->state = D_PREP;
1000         dc->queued = 0;
1001         dc->error = 0;
1002         init_completion(&dc->wait);
1003         list_add_tail(&dc->list, pend_list);
1004         spin_lock_init(&dc->lock);
1005         dc->bio_ref = 0;
1006         atomic_inc(&dcc->discard_cmd_cnt);
1007         dcc->undiscard_blks += len;
1008
1009         return dc;
1010 }
1011
1012 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
1013                                 struct block_device *bdev, block_t lstart,
1014                                 block_t start, block_t len,
1015                                 struct rb_node *parent, struct rb_node **p,
1016                                 bool leftmost)
1017 {
1018         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1019         struct discard_cmd *dc;
1020
1021         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1022
1023         rb_link_node(&dc->rb_node, parent, p);
1024         rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
1025
1026         return dc;
1027 }
1028
1029 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1030                                                         struct discard_cmd *dc)
1031 {
1032         if (dc->state == D_DONE)
1033                 atomic_sub(dc->queued, &dcc->queued_discard);
1034
1035         list_del(&dc->list);
1036         rb_erase_cached(&dc->rb_node, &dcc->root);
1037         dcc->undiscard_blks -= dc->len;
1038
1039         kmem_cache_free(discard_cmd_slab, dc);
1040
1041         atomic_dec(&dcc->discard_cmd_cnt);
1042 }
1043
1044 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1045                                                         struct discard_cmd *dc)
1046 {
1047         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1048         unsigned long flags;
1049
1050         trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
1051
1052         spin_lock_irqsave(&dc->lock, flags);
1053         if (dc->bio_ref) {
1054                 spin_unlock_irqrestore(&dc->lock, flags);
1055                 return;
1056         }
1057         spin_unlock_irqrestore(&dc->lock, flags);
1058
1059         f2fs_bug_on(sbi, dc->ref);
1060
1061         if (dc->error == -EOPNOTSUPP)
1062                 dc->error = 0;
1063
1064         if (dc->error)
1065                 printk_ratelimited(
1066                         "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
1067                         KERN_INFO, sbi->sb->s_id,
1068                         dc->lstart, dc->start, dc->len, dc->error);
1069         __detach_discard_cmd(dcc, dc);
1070 }
1071
1072 static void f2fs_submit_discard_endio(struct bio *bio)
1073 {
1074         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1075         unsigned long flags;
1076
1077         spin_lock_irqsave(&dc->lock, flags);
1078         if (!dc->error)
1079                 dc->error = blk_status_to_errno(bio->bi_status);
1080         dc->bio_ref--;
1081         if (!dc->bio_ref && dc->state == D_SUBMIT) {
1082                 dc->state = D_DONE;
1083                 complete_all(&dc->wait);
1084         }
1085         spin_unlock_irqrestore(&dc->lock, flags);
1086         bio_put(bio);
1087 }
1088
1089 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1090                                 block_t start, block_t end)
1091 {
1092 #ifdef CONFIG_F2FS_CHECK_FS
1093         struct seg_entry *sentry;
1094         unsigned int segno;
1095         block_t blk = start;
1096         unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1097         unsigned long *map;
1098
1099         while (blk < end) {
1100                 segno = GET_SEGNO(sbi, blk);
1101                 sentry = get_seg_entry(sbi, segno);
1102                 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1103
1104                 if (end < START_BLOCK(sbi, segno + 1))
1105                         size = GET_BLKOFF_FROM_SEG0(sbi, end);
1106                 else
1107                         size = max_blocks;
1108                 map = (unsigned long *)(sentry->cur_valid_map);
1109                 offset = __find_rev_next_bit(map, size, offset);
1110                 f2fs_bug_on(sbi, offset != size);
1111                 blk = START_BLOCK(sbi, segno + 1);
1112         }
1113 #endif
1114 }
1115
1116 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1117                                 struct discard_policy *dpolicy,
1118                                 int discard_type, unsigned int granularity)
1119 {
1120         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1121
1122         /* common policy */
1123         dpolicy->type = discard_type;
1124         dpolicy->sync = true;
1125         dpolicy->ordered = false;
1126         dpolicy->granularity = granularity;
1127
1128         dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1129         dpolicy->io_aware_gran = MAX_PLIST_NUM;
1130         dpolicy->timeout = false;
1131
1132         if (discard_type == DPOLICY_BG) {
1133                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1134                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1135                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1136                 dpolicy->io_aware = true;
1137                 dpolicy->sync = false;
1138                 dpolicy->ordered = true;
1139                 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1140                         dpolicy->granularity = 1;
1141                         if (atomic_read(&dcc->discard_cmd_cnt))
1142                                 dpolicy->max_interval =
1143                                         DEF_MIN_DISCARD_ISSUE_TIME;
1144                 }
1145         } else if (discard_type == DPOLICY_FORCE) {
1146                 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1147                 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1148                 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1149                 dpolicy->io_aware = false;
1150         } else if (discard_type == DPOLICY_FSTRIM) {
1151                 dpolicy->io_aware = false;
1152         } else if (discard_type == DPOLICY_UMOUNT) {
1153                 dpolicy->io_aware = false;
1154                 /* we need to issue all to keep CP_TRIMMED_FLAG */
1155                 dpolicy->granularity = 1;
1156                 dpolicy->timeout = true;
1157         }
1158 }
1159
1160 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1161                                 struct block_device *bdev, block_t lstart,
1162                                 block_t start, block_t len);
1163 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1164 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1165                                                 struct discard_policy *dpolicy,
1166                                                 struct discard_cmd *dc,
1167                                                 unsigned int *issued)
1168 {
1169         struct block_device *bdev = dc->bdev;
1170         struct request_queue *q = bdev_get_queue(bdev);
1171         unsigned int max_discard_blocks =
1172                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1173         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1174         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1175                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1176         int flag = dpolicy->sync ? REQ_SYNC : 0;
1177         block_t lstart, start, len, total_len;
1178         int err = 0;
1179
1180         if (dc->state != D_PREP)
1181                 return 0;
1182
1183         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1184                 return 0;
1185
1186         trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1187
1188         lstart = dc->lstart;
1189         start = dc->start;
1190         len = dc->len;
1191         total_len = len;
1192
1193         dc->len = 0;
1194
1195         while (total_len && *issued < dpolicy->max_requests && !err) {
1196                 struct bio *bio = NULL;
1197                 unsigned long flags;
1198                 bool last = true;
1199
1200                 if (len > max_discard_blocks) {
1201                         len = max_discard_blocks;
1202                         last = false;
1203                 }
1204
1205                 (*issued)++;
1206                 if (*issued == dpolicy->max_requests)
1207                         last = true;
1208
1209                 dc->len += len;
1210
1211                 if (time_to_inject(sbi, FAULT_DISCARD)) {
1212                         f2fs_show_injection_info(sbi, FAULT_DISCARD);
1213                         err = -EIO;
1214                         goto submit;
1215                 }
1216                 err = __blkdev_issue_discard(bdev,
1217                                         SECTOR_FROM_BLOCK(start),
1218                                         SECTOR_FROM_BLOCK(len),
1219                                         GFP_NOFS, 0, &bio);
1220 submit:
1221                 if (err) {
1222                         spin_lock_irqsave(&dc->lock, flags);
1223                         if (dc->state == D_PARTIAL)
1224                                 dc->state = D_SUBMIT;
1225                         spin_unlock_irqrestore(&dc->lock, flags);
1226
1227                         break;
1228                 }
1229
1230                 f2fs_bug_on(sbi, !bio);
1231
1232                 /*
1233                  * should keep before submission to avoid D_DONE
1234                  * right away
1235                  */
1236                 spin_lock_irqsave(&dc->lock, flags);
1237                 if (last)
1238                         dc->state = D_SUBMIT;
1239                 else
1240                         dc->state = D_PARTIAL;
1241                 dc->bio_ref++;
1242                 spin_unlock_irqrestore(&dc->lock, flags);
1243
1244                 atomic_inc(&dcc->queued_discard);
1245                 dc->queued++;
1246                 list_move_tail(&dc->list, wait_list);
1247
1248                 /* sanity check on discard range */
1249                 __check_sit_bitmap(sbi, lstart, lstart + len);
1250
1251                 bio->bi_private = dc;
1252                 bio->bi_end_io = f2fs_submit_discard_endio;
1253                 bio->bi_opf |= flag;
1254                 submit_bio(bio);
1255
1256                 atomic_inc(&dcc->issued_discard);
1257
1258                 f2fs_update_iostat(sbi, FS_DISCARD, 1);
1259
1260                 lstart += len;
1261                 start += len;
1262                 total_len -= len;
1263                 len = total_len;
1264         }
1265
1266         if (!err && len) {
1267                 dcc->undiscard_blks -= len;
1268                 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1269         }
1270         return err;
1271 }
1272
1273 static void __insert_discard_tree(struct f2fs_sb_info *sbi,
1274                                 struct block_device *bdev, block_t lstart,
1275                                 block_t start, block_t len,
1276                                 struct rb_node **insert_p,
1277                                 struct rb_node *insert_parent)
1278 {
1279         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1280         struct rb_node **p;
1281         struct rb_node *parent = NULL;
1282         bool leftmost = true;
1283
1284         if (insert_p && insert_parent) {
1285                 parent = insert_parent;
1286                 p = insert_p;
1287                 goto do_insert;
1288         }
1289
1290         p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1291                                                         lstart, &leftmost);
1292 do_insert:
1293         __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1294                                                                 p, leftmost);
1295 }
1296
1297 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1298                                                 struct discard_cmd *dc)
1299 {
1300         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1301 }
1302
1303 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1304                                 struct discard_cmd *dc, block_t blkaddr)
1305 {
1306         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1307         struct discard_info di = dc->di;
1308         bool modified = false;
1309
1310         if (dc->state == D_DONE || dc->len == 1) {
1311                 __remove_discard_cmd(sbi, dc);
1312                 return;
1313         }
1314
1315         dcc->undiscard_blks -= di.len;
1316
1317         if (blkaddr > di.lstart) {
1318                 dc->len = blkaddr - dc->lstart;
1319                 dcc->undiscard_blks += dc->len;
1320                 __relocate_discard_cmd(dcc, dc);
1321                 modified = true;
1322         }
1323
1324         if (blkaddr < di.lstart + di.len - 1) {
1325                 if (modified) {
1326                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1327                                         di.start + blkaddr + 1 - di.lstart,
1328                                         di.lstart + di.len - 1 - blkaddr,
1329                                         NULL, NULL);
1330                 } else {
1331                         dc->lstart++;
1332                         dc->len--;
1333                         dc->start++;
1334                         dcc->undiscard_blks += dc->len;
1335                         __relocate_discard_cmd(dcc, dc);
1336                 }
1337         }
1338 }
1339
1340 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1341                                 struct block_device *bdev, block_t lstart,
1342                                 block_t start, block_t len)
1343 {
1344         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1345         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1346         struct discard_cmd *dc;
1347         struct discard_info di = {0};
1348         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1349         struct request_queue *q = bdev_get_queue(bdev);
1350         unsigned int max_discard_blocks =
1351                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1352         block_t end = lstart + len;
1353
1354         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1355                                         NULL, lstart,
1356                                         (struct rb_entry **)&prev_dc,
1357                                         (struct rb_entry **)&next_dc,
1358                                         &insert_p, &insert_parent, true, NULL);
1359         if (dc)
1360                 prev_dc = dc;
1361
1362         if (!prev_dc) {
1363                 di.lstart = lstart;
1364                 di.len = next_dc ? next_dc->lstart - lstart : len;
1365                 di.len = min(di.len, len);
1366                 di.start = start;
1367         }
1368
1369         while (1) {
1370                 struct rb_node *node;
1371                 bool merged = false;
1372                 struct discard_cmd *tdc = NULL;
1373
1374                 if (prev_dc) {
1375                         di.lstart = prev_dc->lstart + prev_dc->len;
1376                         if (di.lstart < lstart)
1377                                 di.lstart = lstart;
1378                         if (di.lstart >= end)
1379                                 break;
1380
1381                         if (!next_dc || next_dc->lstart > end)
1382                                 di.len = end - di.lstart;
1383                         else
1384                                 di.len = next_dc->lstart - di.lstart;
1385                         di.start = start + di.lstart - lstart;
1386                 }
1387
1388                 if (!di.len)
1389                         goto next;
1390
1391                 if (prev_dc && prev_dc->state == D_PREP &&
1392                         prev_dc->bdev == bdev &&
1393                         __is_discard_back_mergeable(&di, &prev_dc->di,
1394                                                         max_discard_blocks)) {
1395                         prev_dc->di.len += di.len;
1396                         dcc->undiscard_blks += di.len;
1397                         __relocate_discard_cmd(dcc, prev_dc);
1398                         di = prev_dc->di;
1399                         tdc = prev_dc;
1400                         merged = true;
1401                 }
1402
1403                 if (next_dc && next_dc->state == D_PREP &&
1404                         next_dc->bdev == bdev &&
1405                         __is_discard_front_mergeable(&di, &next_dc->di,
1406                                                         max_discard_blocks)) {
1407                         next_dc->di.lstart = di.lstart;
1408                         next_dc->di.len += di.len;
1409                         next_dc->di.start = di.start;
1410                         dcc->undiscard_blks += di.len;
1411                         __relocate_discard_cmd(dcc, next_dc);
1412                         if (tdc)
1413                                 __remove_discard_cmd(sbi, tdc);
1414                         merged = true;
1415                 }
1416
1417                 if (!merged) {
1418                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1419                                                         di.len, NULL, NULL);
1420                 }
1421  next:
1422                 prev_dc = next_dc;
1423                 if (!prev_dc)
1424                         break;
1425
1426                 node = rb_next(&prev_dc->rb_node);
1427                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1428         }
1429 }
1430
1431 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1432                 struct block_device *bdev, block_t blkstart, block_t blklen)
1433 {
1434         block_t lblkstart = blkstart;
1435
1436         if (!f2fs_bdev_support_discard(bdev))
1437                 return 0;
1438
1439         trace_f2fs_queue_discard(bdev, blkstart, blklen);
1440
1441         if (f2fs_is_multi_device(sbi)) {
1442                 int devi = f2fs_target_device_index(sbi, blkstart);
1443
1444                 blkstart -= FDEV(devi).start_blk;
1445         }
1446         mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1447         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1448         mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1449         return 0;
1450 }
1451
1452 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1453                                         struct discard_policy *dpolicy)
1454 {
1455         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1456         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1457         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1458         struct discard_cmd *dc;
1459         struct blk_plug plug;
1460         unsigned int pos = dcc->next_pos;
1461         unsigned int issued = 0;
1462         bool io_interrupted = false;
1463
1464         mutex_lock(&dcc->cmd_lock);
1465         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1466                                         NULL, pos,
1467                                         (struct rb_entry **)&prev_dc,
1468                                         (struct rb_entry **)&next_dc,
1469                                         &insert_p, &insert_parent, true, NULL);
1470         if (!dc)
1471                 dc = next_dc;
1472
1473         blk_start_plug(&plug);
1474
1475         while (dc) {
1476                 struct rb_node *node;
1477                 int err = 0;
1478
1479                 if (dc->state != D_PREP)
1480                         goto next;
1481
1482                 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1483                         io_interrupted = true;
1484                         break;
1485                 }
1486
1487                 dcc->next_pos = dc->lstart + dc->len;
1488                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1489
1490                 if (issued >= dpolicy->max_requests)
1491                         break;
1492 next:
1493                 node = rb_next(&dc->rb_node);
1494                 if (err)
1495                         __remove_discard_cmd(sbi, dc);
1496                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1497         }
1498
1499         blk_finish_plug(&plug);
1500
1501         if (!dc)
1502                 dcc->next_pos = 0;
1503
1504         mutex_unlock(&dcc->cmd_lock);
1505
1506         if (!issued && io_interrupted)
1507                 issued = -1;
1508
1509         return issued;
1510 }
1511 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1512                                         struct discard_policy *dpolicy);
1513
1514 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1515                                         struct discard_policy *dpolicy)
1516 {
1517         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1518         struct list_head *pend_list;
1519         struct discard_cmd *dc, *tmp;
1520         struct blk_plug plug;
1521         int i, issued;
1522         bool io_interrupted = false;
1523
1524         if (dpolicy->timeout)
1525                 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1526
1527 retry:
1528         issued = 0;
1529         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1530                 if (dpolicy->timeout &&
1531                                 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1532                         break;
1533
1534                 if (i + 1 < dpolicy->granularity)
1535                         break;
1536
1537                 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1538                         return __issue_discard_cmd_orderly(sbi, dpolicy);
1539
1540                 pend_list = &dcc->pend_list[i];
1541
1542                 mutex_lock(&dcc->cmd_lock);
1543                 if (list_empty(pend_list))
1544                         goto next;
1545                 if (unlikely(dcc->rbtree_check))
1546                         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1547                                                         &dcc->root, false));
1548                 blk_start_plug(&plug);
1549                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1550                         f2fs_bug_on(sbi, dc->state != D_PREP);
1551
1552                         if (dpolicy->timeout &&
1553                                 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1554                                 break;
1555
1556                         if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1557                                                 !is_idle(sbi, DISCARD_TIME)) {
1558                                 io_interrupted = true;
1559                                 break;
1560                         }
1561
1562                         __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1563
1564                         if (issued >= dpolicy->max_requests)
1565                                 break;
1566                 }
1567                 blk_finish_plug(&plug);
1568 next:
1569                 mutex_unlock(&dcc->cmd_lock);
1570
1571                 if (issued >= dpolicy->max_requests || io_interrupted)
1572                         break;
1573         }
1574
1575         if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1576                 __wait_all_discard_cmd(sbi, dpolicy);
1577                 goto retry;
1578         }
1579
1580         if (!issued && io_interrupted)
1581                 issued = -1;
1582
1583         return issued;
1584 }
1585
1586 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1587 {
1588         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1589         struct list_head *pend_list;
1590         struct discard_cmd *dc, *tmp;
1591         int i;
1592         bool dropped = false;
1593
1594         mutex_lock(&dcc->cmd_lock);
1595         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1596                 pend_list = &dcc->pend_list[i];
1597                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1598                         f2fs_bug_on(sbi, dc->state != D_PREP);
1599                         __remove_discard_cmd(sbi, dc);
1600                         dropped = true;
1601                 }
1602         }
1603         mutex_unlock(&dcc->cmd_lock);
1604
1605         return dropped;
1606 }
1607
1608 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1609 {
1610         __drop_discard_cmd(sbi);
1611 }
1612
1613 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1614                                                         struct discard_cmd *dc)
1615 {
1616         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1617         unsigned int len = 0;
1618
1619         wait_for_completion_io(&dc->wait);
1620         mutex_lock(&dcc->cmd_lock);
1621         f2fs_bug_on(sbi, dc->state != D_DONE);
1622         dc->ref--;
1623         if (!dc->ref) {
1624                 if (!dc->error)
1625                         len = dc->len;
1626                 __remove_discard_cmd(sbi, dc);
1627         }
1628         mutex_unlock(&dcc->cmd_lock);
1629
1630         return len;
1631 }
1632
1633 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1634                                                 struct discard_policy *dpolicy,
1635                                                 block_t start, block_t end)
1636 {
1637         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1638         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1639                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1640         struct discard_cmd *dc, *tmp;
1641         bool need_wait;
1642         unsigned int trimmed = 0;
1643
1644 next:
1645         need_wait = false;
1646
1647         mutex_lock(&dcc->cmd_lock);
1648         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1649                 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1650                         continue;
1651                 if (dc->len < dpolicy->granularity)
1652                         continue;
1653                 if (dc->state == D_DONE && !dc->ref) {
1654                         wait_for_completion_io(&dc->wait);
1655                         if (!dc->error)
1656                                 trimmed += dc->len;
1657                         __remove_discard_cmd(sbi, dc);
1658                 } else {
1659                         dc->ref++;
1660                         need_wait = true;
1661                         break;
1662                 }
1663         }
1664         mutex_unlock(&dcc->cmd_lock);
1665
1666         if (need_wait) {
1667                 trimmed += __wait_one_discard_bio(sbi, dc);
1668                 goto next;
1669         }
1670
1671         return trimmed;
1672 }
1673
1674 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1675                                                 struct discard_policy *dpolicy)
1676 {
1677         struct discard_policy dp;
1678         unsigned int discard_blks;
1679
1680         if (dpolicy)
1681                 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1682
1683         /* wait all */
1684         __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1685         discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1686         __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1687         discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1688
1689         return discard_blks;
1690 }
1691
1692 /* This should be covered by global mutex, &sit_i->sentry_lock */
1693 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1694 {
1695         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1696         struct discard_cmd *dc;
1697         bool need_wait = false;
1698
1699         mutex_lock(&dcc->cmd_lock);
1700         dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1701                                                         NULL, blkaddr);
1702         if (dc) {
1703                 if (dc->state == D_PREP) {
1704                         __punch_discard_cmd(sbi, dc, blkaddr);
1705                 } else {
1706                         dc->ref++;
1707                         need_wait = true;
1708                 }
1709         }
1710         mutex_unlock(&dcc->cmd_lock);
1711
1712         if (need_wait)
1713                 __wait_one_discard_bio(sbi, dc);
1714 }
1715
1716 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1717 {
1718         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1719
1720         if (dcc && dcc->f2fs_issue_discard) {
1721                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1722
1723                 dcc->f2fs_issue_discard = NULL;
1724                 kthread_stop(discard_thread);
1725         }
1726 }
1727
1728 /* This comes from f2fs_put_super */
1729 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1730 {
1731         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1732         struct discard_policy dpolicy;
1733         bool dropped;
1734
1735         __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1736                                         dcc->discard_granularity);
1737         __issue_discard_cmd(sbi, &dpolicy);
1738         dropped = __drop_discard_cmd(sbi);
1739
1740         /* just to make sure there is no pending discard commands */
1741         __wait_all_discard_cmd(sbi, NULL);
1742
1743         f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1744         return dropped;
1745 }
1746
1747 static int issue_discard_thread(void *data)
1748 {
1749         struct f2fs_sb_info *sbi = data;
1750         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1751         wait_queue_head_t *q = &dcc->discard_wait_queue;
1752         struct discard_policy dpolicy;
1753         unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1754         int issued;
1755
1756         set_freezable();
1757
1758         do {
1759                 if (sbi->gc_mode == GC_URGENT_HIGH ||
1760                         !f2fs_available_free_memory(sbi, DISCARD_CACHE))
1761                         __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1762                 else
1763                         __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1764                                                 dcc->discard_granularity);
1765
1766                 if (!atomic_read(&dcc->discard_cmd_cnt))
1767                        wait_ms = dpolicy.max_interval;
1768
1769                 wait_event_interruptible_timeout(*q,
1770                                 kthread_should_stop() || freezing(current) ||
1771                                 dcc->discard_wake,
1772                                 msecs_to_jiffies(wait_ms));
1773
1774                 if (dcc->discard_wake)
1775                         dcc->discard_wake = 0;
1776
1777                 /* clean up pending candidates before going to sleep */
1778                 if (atomic_read(&dcc->queued_discard))
1779                         __wait_all_discard_cmd(sbi, NULL);
1780
1781                 if (try_to_freeze())
1782                         continue;
1783                 if (f2fs_readonly(sbi->sb))
1784                         continue;
1785                 if (kthread_should_stop())
1786                         return 0;
1787                 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1788                         wait_ms = dpolicy.max_interval;
1789                         continue;
1790                 }
1791                 if (!atomic_read(&dcc->discard_cmd_cnt))
1792                         continue;
1793
1794                 sb_start_intwrite(sbi->sb);
1795
1796                 issued = __issue_discard_cmd(sbi, &dpolicy);
1797                 if (issued > 0) {
1798                         __wait_all_discard_cmd(sbi, &dpolicy);
1799                         wait_ms = dpolicy.min_interval;
1800                 } else if (issued == -1) {
1801                         wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1802                         if (!wait_ms)
1803                                 wait_ms = dpolicy.mid_interval;
1804                 } else {
1805                         wait_ms = dpolicy.max_interval;
1806                 }
1807
1808                 sb_end_intwrite(sbi->sb);
1809
1810         } while (!kthread_should_stop());
1811         return 0;
1812 }
1813
1814 #ifdef CONFIG_BLK_DEV_ZONED
1815 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1816                 struct block_device *bdev, block_t blkstart, block_t blklen)
1817 {
1818         sector_t sector, nr_sects;
1819         block_t lblkstart = blkstart;
1820         int devi = 0;
1821
1822         if (f2fs_is_multi_device(sbi)) {
1823                 devi = f2fs_target_device_index(sbi, blkstart);
1824                 if (blkstart < FDEV(devi).start_blk ||
1825                     blkstart > FDEV(devi).end_blk) {
1826                         f2fs_err(sbi, "Invalid block %x", blkstart);
1827                         return -EIO;
1828                 }
1829                 blkstart -= FDEV(devi).start_blk;
1830         }
1831
1832         /* For sequential zones, reset the zone write pointer */
1833         if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1834                 sector = SECTOR_FROM_BLOCK(blkstart);
1835                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1836
1837                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1838                                 nr_sects != bdev_zone_sectors(bdev)) {
1839                         f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1840                                  devi, sbi->s_ndevs ? FDEV(devi).path : "",
1841                                  blkstart, blklen);
1842                         return -EIO;
1843                 }
1844                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1845                 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1846                                         sector, nr_sects, GFP_NOFS);
1847         }
1848
1849         /* For conventional zones, use regular discard if supported */
1850         return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1851 }
1852 #endif
1853
1854 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1855                 struct block_device *bdev, block_t blkstart, block_t blklen)
1856 {
1857 #ifdef CONFIG_BLK_DEV_ZONED
1858         if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1859                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1860 #endif
1861         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1862 }
1863
1864 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1865                                 block_t blkstart, block_t blklen)
1866 {
1867         sector_t start = blkstart, len = 0;
1868         struct block_device *bdev;
1869         struct seg_entry *se;
1870         unsigned int offset;
1871         block_t i;
1872         int err = 0;
1873
1874         bdev = f2fs_target_device(sbi, blkstart, NULL);
1875
1876         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1877                 if (i != start) {
1878                         struct block_device *bdev2 =
1879                                 f2fs_target_device(sbi, i, NULL);
1880
1881                         if (bdev2 != bdev) {
1882                                 err = __issue_discard_async(sbi, bdev,
1883                                                 start, len);
1884                                 if (err)
1885                                         return err;
1886                                 bdev = bdev2;
1887                                 start = i;
1888                                 len = 0;
1889                         }
1890                 }
1891
1892                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1893                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1894
1895                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1896                         sbi->discard_blks--;
1897         }
1898
1899         if (len)
1900                 err = __issue_discard_async(sbi, bdev, start, len);
1901         return err;
1902 }
1903
1904 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1905                                                         bool check_only)
1906 {
1907         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1908         int max_blocks = sbi->blocks_per_seg;
1909         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1910         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1911         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1912         unsigned long *discard_map = (unsigned long *)se->discard_map;
1913         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1914         unsigned int start = 0, end = -1;
1915         bool force = (cpc->reason & CP_DISCARD);
1916         struct discard_entry *de = NULL;
1917         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1918         int i;
1919
1920         if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
1921                 return false;
1922
1923         if (!force) {
1924                 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1925                         SM_I(sbi)->dcc_info->nr_discards >=
1926                                 SM_I(sbi)->dcc_info->max_discards)
1927                         return false;
1928         }
1929
1930         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1931         for (i = 0; i < entries; i++)
1932                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1933                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1934
1935         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1936                                 SM_I(sbi)->dcc_info->max_discards) {
1937                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1938                 if (start >= max_blocks)
1939                         break;
1940
1941                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1942                 if (force && start && end != max_blocks
1943                                         && (end - start) < cpc->trim_minlen)
1944                         continue;
1945
1946                 if (check_only)
1947                         return true;
1948
1949                 if (!de) {
1950                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1951                                                                 GFP_F2FS_ZERO);
1952                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1953                         list_add_tail(&de->list, head);
1954                 }
1955
1956                 for (i = start; i < end; i++)
1957                         __set_bit_le(i, (void *)de->discard_map);
1958
1959                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1960         }
1961         return false;
1962 }
1963
1964 static void release_discard_addr(struct discard_entry *entry)
1965 {
1966         list_del(&entry->list);
1967         kmem_cache_free(discard_entry_slab, entry);
1968 }
1969
1970 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1971 {
1972         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1973         struct discard_entry *entry, *this;
1974
1975         /* drop caches */
1976         list_for_each_entry_safe(entry, this, head, list)
1977                 release_discard_addr(entry);
1978 }
1979
1980 /*
1981  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1982  */
1983 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1984 {
1985         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1986         unsigned int segno;
1987
1988         mutex_lock(&dirty_i->seglist_lock);
1989         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1990                 __set_test_and_free(sbi, segno, false);
1991         mutex_unlock(&dirty_i->seglist_lock);
1992 }
1993
1994 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1995                                                 struct cp_control *cpc)
1996 {
1997         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1998         struct list_head *head = &dcc->entry_list;
1999         struct discard_entry *entry, *this;
2000         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2001         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
2002         unsigned int start = 0, end = -1;
2003         unsigned int secno, start_segno;
2004         bool force = (cpc->reason & CP_DISCARD);
2005         bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
2006
2007         mutex_lock(&dirty_i->seglist_lock);
2008
2009         while (1) {
2010                 int i;
2011
2012                 if (need_align && end != -1)
2013                         end--;
2014                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
2015                 if (start >= MAIN_SEGS(sbi))
2016                         break;
2017                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
2018                                                                 start + 1);
2019
2020                 if (need_align) {
2021                         start = rounddown(start, sbi->segs_per_sec);
2022                         end = roundup(end, sbi->segs_per_sec);
2023                 }
2024
2025                 for (i = start; i < end; i++) {
2026                         if (test_and_clear_bit(i, prefree_map))
2027                                 dirty_i->nr_dirty[PRE]--;
2028                 }
2029
2030                 if (!f2fs_realtime_discard_enable(sbi))
2031                         continue;
2032
2033                 if (force && start >= cpc->trim_start &&
2034                                         (end - 1) <= cpc->trim_end)
2035                                 continue;
2036
2037                 if (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi)) {
2038                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2039                                 (end - start) << sbi->log_blocks_per_seg);
2040                         continue;
2041                 }
2042 next:
2043                 secno = GET_SEC_FROM_SEG(sbi, start);
2044                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
2045                 if (!IS_CURSEC(sbi, secno) &&
2046                         !get_valid_blocks(sbi, start, true))
2047                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2048                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
2049
2050                 start = start_segno + sbi->segs_per_sec;
2051                 if (start < end)
2052                         goto next;
2053                 else
2054                         end = start - 1;
2055         }
2056         mutex_unlock(&dirty_i->seglist_lock);
2057
2058         /* send small discards */
2059         list_for_each_entry_safe(entry, this, head, list) {
2060                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2061                 bool is_valid = test_bit_le(0, entry->discard_map);
2062
2063 find_next:
2064                 if (is_valid) {
2065                         next_pos = find_next_zero_bit_le(entry->discard_map,
2066                                         sbi->blocks_per_seg, cur_pos);
2067                         len = next_pos - cur_pos;
2068
2069                         if (f2fs_sb_has_blkzoned(sbi) ||
2070                             (force && len < cpc->trim_minlen))
2071                                 goto skip;
2072
2073                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2074                                                                         len);
2075                         total_len += len;
2076                 } else {
2077                         next_pos = find_next_bit_le(entry->discard_map,
2078                                         sbi->blocks_per_seg, cur_pos);
2079                 }
2080 skip:
2081                 cur_pos = next_pos;
2082                 is_valid = !is_valid;
2083
2084                 if (cur_pos < sbi->blocks_per_seg)
2085                         goto find_next;
2086
2087                 release_discard_addr(entry);
2088                 dcc->nr_discards -= total_len;
2089         }
2090
2091         wake_up_discard_thread(sbi, false);
2092 }
2093
2094 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2095 {
2096         dev_t dev = sbi->sb->s_bdev->bd_dev;
2097         struct discard_cmd_control *dcc;
2098         int err = 0, i;
2099
2100         if (SM_I(sbi)->dcc_info) {
2101                 dcc = SM_I(sbi)->dcc_info;
2102                 goto init_thread;
2103         }
2104
2105         dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2106         if (!dcc)
2107                 return -ENOMEM;
2108
2109         dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2110         INIT_LIST_HEAD(&dcc->entry_list);
2111         for (i = 0; i < MAX_PLIST_NUM; i++)
2112                 INIT_LIST_HEAD(&dcc->pend_list[i]);
2113         INIT_LIST_HEAD(&dcc->wait_list);
2114         INIT_LIST_HEAD(&dcc->fstrim_list);
2115         mutex_init(&dcc->cmd_lock);
2116         atomic_set(&dcc->issued_discard, 0);
2117         atomic_set(&dcc->queued_discard, 0);
2118         atomic_set(&dcc->discard_cmd_cnt, 0);
2119         dcc->nr_discards = 0;
2120         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2121         dcc->undiscard_blks = 0;
2122         dcc->next_pos = 0;
2123         dcc->root = RB_ROOT_CACHED;
2124         dcc->rbtree_check = false;
2125
2126         init_waitqueue_head(&dcc->discard_wait_queue);
2127         SM_I(sbi)->dcc_info = dcc;
2128 init_thread:
2129         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2130                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2131         if (IS_ERR(dcc->f2fs_issue_discard)) {
2132                 err = PTR_ERR(dcc->f2fs_issue_discard);
2133                 kfree(dcc);
2134                 SM_I(sbi)->dcc_info = NULL;
2135                 return err;
2136         }
2137
2138         return err;
2139 }
2140
2141 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2142 {
2143         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2144
2145         if (!dcc)
2146                 return;
2147
2148         f2fs_stop_discard_thread(sbi);
2149
2150         /*
2151          * Recovery can cache discard commands, so in error path of
2152          * fill_super(), it needs to give a chance to handle them.
2153          */
2154         if (unlikely(atomic_read(&dcc->discard_cmd_cnt)))
2155                 f2fs_issue_discard_timeout(sbi);
2156
2157         kfree(dcc);
2158         SM_I(sbi)->dcc_info = NULL;
2159 }
2160
2161 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2162 {
2163         struct sit_info *sit_i = SIT_I(sbi);
2164
2165         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2166                 sit_i->dirty_sentries++;
2167                 return false;
2168         }
2169
2170         return true;
2171 }
2172
2173 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2174                                         unsigned int segno, int modified)
2175 {
2176         struct seg_entry *se = get_seg_entry(sbi, segno);
2177
2178         se->type = type;
2179         if (modified)
2180                 __mark_sit_entry_dirty(sbi, segno);
2181 }
2182
2183 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2184                                                                 block_t blkaddr)
2185 {
2186         unsigned int segno = GET_SEGNO(sbi, blkaddr);
2187
2188         if (segno == NULL_SEGNO)
2189                 return 0;
2190         return get_seg_entry(sbi, segno)->mtime;
2191 }
2192
2193 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2194                                                 unsigned long long old_mtime)
2195 {
2196         struct seg_entry *se;
2197         unsigned int segno = GET_SEGNO(sbi, blkaddr);
2198         unsigned long long ctime = get_mtime(sbi, false);
2199         unsigned long long mtime = old_mtime ? old_mtime : ctime;
2200
2201         if (segno == NULL_SEGNO)
2202                 return;
2203
2204         se = get_seg_entry(sbi, segno);
2205
2206         if (!se->mtime)
2207                 se->mtime = mtime;
2208         else
2209                 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2210                                                 se->valid_blocks + 1);
2211
2212         if (ctime > SIT_I(sbi)->max_mtime)
2213                 SIT_I(sbi)->max_mtime = ctime;
2214 }
2215
2216 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2217 {
2218         struct seg_entry *se;
2219         unsigned int segno, offset;
2220         long int new_vblocks;
2221         bool exist;
2222 #ifdef CONFIG_F2FS_CHECK_FS
2223         bool mir_exist;
2224 #endif
2225
2226         segno = GET_SEGNO(sbi, blkaddr);
2227
2228         se = get_seg_entry(sbi, segno);
2229         new_vblocks = se->valid_blocks + del;
2230         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2231
2232         f2fs_bug_on(sbi, (new_vblocks < 0 ||
2233                         (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2234
2235         se->valid_blocks = new_vblocks;
2236
2237         /* Update valid block bitmap */
2238         if (del > 0) {
2239                 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2240 #ifdef CONFIG_F2FS_CHECK_FS
2241                 mir_exist = f2fs_test_and_set_bit(offset,
2242                                                 se->cur_valid_map_mir);
2243                 if (unlikely(exist != mir_exist)) {
2244                         f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2245                                  blkaddr, exist);
2246                         f2fs_bug_on(sbi, 1);
2247                 }
2248 #endif
2249                 if (unlikely(exist)) {
2250                         f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2251                                  blkaddr);
2252                         f2fs_bug_on(sbi, 1);
2253                         se->valid_blocks--;
2254                         del = 0;
2255                 }
2256
2257                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
2258                         sbi->discard_blks--;
2259
2260                 /*
2261                  * SSR should never reuse block which is checkpointed
2262                  * or newly invalidated.
2263                  */
2264                 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2265                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2266                                 se->ckpt_valid_blocks++;
2267                 }
2268         } else {
2269                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2270 #ifdef CONFIG_F2FS_CHECK_FS
2271                 mir_exist = f2fs_test_and_clear_bit(offset,
2272                                                 se->cur_valid_map_mir);
2273                 if (unlikely(exist != mir_exist)) {
2274                         f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2275                                  blkaddr, exist);
2276                         f2fs_bug_on(sbi, 1);
2277                 }
2278 #endif
2279                 if (unlikely(!exist)) {
2280                         f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2281                                  blkaddr);
2282                         f2fs_bug_on(sbi, 1);
2283                         se->valid_blocks++;
2284                         del = 0;
2285                 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2286                         /*
2287                          * If checkpoints are off, we must not reuse data that
2288                          * was used in the previous checkpoint. If it was used
2289                          * before, we must track that to know how much space we
2290                          * really have.
2291                          */
2292                         if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2293                                 spin_lock(&sbi->stat_lock);
2294                                 sbi->unusable_block_count++;
2295                                 spin_unlock(&sbi->stat_lock);
2296                         }
2297                 }
2298
2299                 if (f2fs_test_and_clear_bit(offset, se->discard_map))
2300                         sbi->discard_blks++;
2301         }
2302         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2303                 se->ckpt_valid_blocks += del;
2304
2305         __mark_sit_entry_dirty(sbi, segno);
2306
2307         /* update total number of valid blocks to be written in ckpt area */
2308         SIT_I(sbi)->written_valid_blocks += del;
2309
2310         if (__is_large_section(sbi))
2311                 get_sec_entry(sbi, segno)->valid_blocks += del;
2312 }
2313
2314 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2315 {
2316         unsigned int segno = GET_SEGNO(sbi, addr);
2317         struct sit_info *sit_i = SIT_I(sbi);
2318
2319         f2fs_bug_on(sbi, addr == NULL_ADDR);
2320         if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2321                 return;
2322
2323         invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2324
2325         /* add it into sit main buffer */
2326         down_write(&sit_i->sentry_lock);
2327
2328         update_segment_mtime(sbi, addr, 0);
2329         update_sit_entry(sbi, addr, -1);
2330
2331         /* add it into dirty seglist */
2332         locate_dirty_segment(sbi, segno);
2333
2334         up_write(&sit_i->sentry_lock);
2335 }
2336
2337 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2338 {
2339         struct sit_info *sit_i = SIT_I(sbi);
2340         unsigned int segno, offset;
2341         struct seg_entry *se;
2342         bool is_cp = false;
2343
2344         if (!__is_valid_data_blkaddr(blkaddr))
2345                 return true;
2346
2347         down_read(&sit_i->sentry_lock);
2348
2349         segno = GET_SEGNO(sbi, blkaddr);
2350         se = get_seg_entry(sbi, segno);
2351         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2352
2353         if (f2fs_test_bit(offset, se->ckpt_valid_map))
2354                 is_cp = true;
2355
2356         up_read(&sit_i->sentry_lock);
2357
2358         return is_cp;
2359 }
2360
2361 /*
2362  * This function should be resided under the curseg_mutex lock
2363  */
2364 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2365                                         struct f2fs_summary *sum)
2366 {
2367         struct curseg_info *curseg = CURSEG_I(sbi, type);
2368         void *addr = curseg->sum_blk;
2369
2370         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2371         memcpy(addr, sum, sizeof(struct f2fs_summary));
2372 }
2373
2374 /*
2375  * Calculate the number of current summary pages for writing
2376  */
2377 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2378 {
2379         int valid_sum_count = 0;
2380         int i, sum_in_page;
2381
2382         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2383                 if (sbi->ckpt->alloc_type[i] == SSR)
2384                         valid_sum_count += sbi->blocks_per_seg;
2385                 else {
2386                         if (for_ra)
2387                                 valid_sum_count += le16_to_cpu(
2388                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2389                         else
2390                                 valid_sum_count += curseg_blkoff(sbi, i);
2391                 }
2392         }
2393
2394         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2395                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2396         if (valid_sum_count <= sum_in_page)
2397                 return 1;
2398         else if ((valid_sum_count - sum_in_page) <=
2399                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2400                 return 2;
2401         return 3;
2402 }
2403
2404 /*
2405  * Caller should put this summary page
2406  */
2407 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2408 {
2409         if (unlikely(f2fs_cp_error(sbi)))
2410                 return ERR_PTR(-EIO);
2411         return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno));
2412 }
2413
2414 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2415                                         void *src, block_t blk_addr)
2416 {
2417         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2418
2419         memcpy(page_address(page), src, PAGE_SIZE);
2420         set_page_dirty(page);
2421         f2fs_put_page(page, 1);
2422 }
2423
2424 static void write_sum_page(struct f2fs_sb_info *sbi,
2425                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
2426 {
2427         f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2428 }
2429
2430 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2431                                                 int type, block_t blk_addr)
2432 {
2433         struct curseg_info *curseg = CURSEG_I(sbi, type);
2434         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2435         struct f2fs_summary_block *src = curseg->sum_blk;
2436         struct f2fs_summary_block *dst;
2437
2438         dst = (struct f2fs_summary_block *)page_address(page);
2439         memset(dst, 0, PAGE_SIZE);
2440
2441         mutex_lock(&curseg->curseg_mutex);
2442
2443         down_read(&curseg->journal_rwsem);
2444         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2445         up_read(&curseg->journal_rwsem);
2446
2447         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2448         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2449
2450         mutex_unlock(&curseg->curseg_mutex);
2451
2452         set_page_dirty(page);
2453         f2fs_put_page(page, 1);
2454 }
2455
2456 static int is_next_segment_free(struct f2fs_sb_info *sbi,
2457                                 struct curseg_info *curseg, int type)
2458 {
2459         unsigned int segno = curseg->segno + 1;
2460         struct free_segmap_info *free_i = FREE_I(sbi);
2461
2462         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2463                 return !test_bit(segno, free_i->free_segmap);
2464         return 0;
2465 }
2466
2467 /*
2468  * Find a new segment from the free segments bitmap to right order
2469  * This function should be returned with success, otherwise BUG
2470  */
2471 static void get_new_segment(struct f2fs_sb_info *sbi,
2472                         unsigned int *newseg, bool new_sec, int dir)
2473 {
2474         struct free_segmap_info *free_i = FREE_I(sbi);
2475         unsigned int segno, secno, zoneno;
2476         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2477         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2478         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2479         unsigned int left_start = hint;
2480         bool init = true;
2481         int go_left = 0;
2482         int i;
2483
2484         spin_lock(&free_i->segmap_lock);
2485
2486         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2487                 segno = find_next_zero_bit(free_i->free_segmap,
2488                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2489                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2490                         goto got_it;
2491         }
2492 find_other_zone:
2493         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2494         if (secno >= MAIN_SECS(sbi)) {
2495                 if (dir == ALLOC_RIGHT) {
2496                         secno = find_next_zero_bit(free_i->free_secmap,
2497                                                         MAIN_SECS(sbi), 0);
2498                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2499                 } else {
2500                         go_left = 1;
2501                         left_start = hint - 1;
2502                 }
2503         }
2504         if (go_left == 0)
2505                 goto skip_left;
2506
2507         while (test_bit(left_start, free_i->free_secmap)) {
2508                 if (left_start > 0) {
2509                         left_start--;
2510                         continue;
2511                 }
2512                 left_start = find_next_zero_bit(free_i->free_secmap,
2513                                                         MAIN_SECS(sbi), 0);
2514                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2515                 break;
2516         }
2517         secno = left_start;
2518 skip_left:
2519         segno = GET_SEG_FROM_SEC(sbi, secno);
2520         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2521
2522         /* give up on finding another zone */
2523         if (!init)
2524                 goto got_it;
2525         if (sbi->secs_per_zone == 1)
2526                 goto got_it;
2527         if (zoneno == old_zoneno)
2528                 goto got_it;
2529         if (dir == ALLOC_LEFT) {
2530                 if (!go_left && zoneno + 1 >= total_zones)
2531                         goto got_it;
2532                 if (go_left && zoneno == 0)
2533                         goto got_it;
2534         }
2535         for (i = 0; i < NR_CURSEG_TYPE; i++)
2536                 if (CURSEG_I(sbi, i)->zone == zoneno)
2537                         break;
2538
2539         if (i < NR_CURSEG_TYPE) {
2540                 /* zone is in user, try another */
2541                 if (go_left)
2542                         hint = zoneno * sbi->secs_per_zone - 1;
2543                 else if (zoneno + 1 >= total_zones)
2544                         hint = 0;
2545                 else
2546                         hint = (zoneno + 1) * sbi->secs_per_zone;
2547                 init = false;
2548                 goto find_other_zone;
2549         }
2550 got_it:
2551         /* set it as dirty segment in free segmap */
2552         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2553         __set_inuse(sbi, segno);
2554         *newseg = segno;
2555         spin_unlock(&free_i->segmap_lock);
2556 }
2557
2558 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2559 {
2560         struct curseg_info *curseg = CURSEG_I(sbi, type);
2561         struct summary_footer *sum_footer;
2562         unsigned short seg_type = curseg->seg_type;
2563
2564         curseg->inited = true;
2565         curseg->segno = curseg->next_segno;
2566         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2567         curseg->next_blkoff = 0;
2568         curseg->next_segno = NULL_SEGNO;
2569
2570         sum_footer = &(curseg->sum_blk->footer);
2571         memset(sum_footer, 0, sizeof(struct summary_footer));
2572
2573         sanity_check_seg_type(sbi, seg_type);
2574
2575         if (IS_DATASEG(seg_type))
2576                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2577         if (IS_NODESEG(seg_type))
2578                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2579         __set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
2580 }
2581
2582 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2583 {
2584         struct curseg_info *curseg = CURSEG_I(sbi, type);
2585         unsigned short seg_type = curseg->seg_type;
2586
2587         sanity_check_seg_type(sbi, seg_type);
2588
2589         /* if segs_per_sec is large than 1, we need to keep original policy. */
2590         if (__is_large_section(sbi))
2591                 return curseg->segno;
2592
2593         /* inmem log may not locate on any segment after mount */
2594         if (!curseg->inited)
2595                 return 0;
2596
2597         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2598                 return 0;
2599
2600         if (test_opt(sbi, NOHEAP) &&
2601                 (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type)))
2602                 return 0;
2603
2604         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2605                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2606
2607         /* find segments from 0 to reuse freed segments */
2608         if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2609                 return 0;
2610
2611         return curseg->segno;
2612 }
2613
2614 /*
2615  * Allocate a current working segment.
2616  * This function always allocates a free segment in LFS manner.
2617  */
2618 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2619 {
2620         struct curseg_info *curseg = CURSEG_I(sbi, type);
2621         unsigned short seg_type = curseg->seg_type;
2622         unsigned int segno = curseg->segno;
2623         int dir = ALLOC_LEFT;
2624
2625         if (curseg->inited)
2626                 write_sum_page(sbi, curseg->sum_blk,
2627                                 GET_SUM_BLOCK(sbi, segno));
2628         if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA)
2629                 dir = ALLOC_RIGHT;
2630
2631         if (test_opt(sbi, NOHEAP))
2632                 dir = ALLOC_RIGHT;
2633
2634         segno = __get_next_segno(sbi, type);
2635         get_new_segment(sbi, &segno, new_sec, dir);
2636         curseg->next_segno = segno;
2637         reset_curseg(sbi, type, 1);
2638         curseg->alloc_type = LFS;
2639 }
2640
2641 static int __next_free_blkoff(struct f2fs_sb_info *sbi,
2642                                         int segno, block_t start)
2643 {
2644         struct seg_entry *se = get_seg_entry(sbi, segno);
2645         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2646         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2647         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2648         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2649         int i;
2650
2651         for (i = 0; i < entries; i++)
2652                 target_map[i] = ckpt_map[i] | cur_map[i];
2653
2654         return __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2655 }
2656
2657 /*
2658  * If a segment is written by LFS manner, next block offset is just obtained
2659  * by increasing the current block offset. However, if a segment is written by
2660  * SSR manner, next block offset obtained by calling __next_free_blkoff
2661  */
2662 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2663                                 struct curseg_info *seg)
2664 {
2665         if (seg->alloc_type == SSR)
2666                 seg->next_blkoff =
2667                         __next_free_blkoff(sbi, seg->segno,
2668                                                 seg->next_blkoff + 1);
2669         else
2670                 seg->next_blkoff++;
2671 }
2672
2673 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno)
2674 {
2675         return __next_free_blkoff(sbi, segno, 0) < sbi->blocks_per_seg;
2676 }
2677
2678 /*
2679  * This function always allocates a used segment(from dirty seglist) by SSR
2680  * manner, so it should recover the existing segment information of valid blocks
2681  */
2682 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool flush)
2683 {
2684         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2685         struct curseg_info *curseg = CURSEG_I(sbi, type);
2686         unsigned int new_segno = curseg->next_segno;
2687         struct f2fs_summary_block *sum_node;
2688         struct page *sum_page;
2689
2690         if (flush)
2691                 write_sum_page(sbi, curseg->sum_blk,
2692                                         GET_SUM_BLOCK(sbi, curseg->segno));
2693
2694         __set_test_and_inuse(sbi, new_segno);
2695
2696         mutex_lock(&dirty_i->seglist_lock);
2697         __remove_dirty_segment(sbi, new_segno, PRE);
2698         __remove_dirty_segment(sbi, new_segno, DIRTY);
2699         mutex_unlock(&dirty_i->seglist_lock);
2700
2701         reset_curseg(sbi, type, 1);
2702         curseg->alloc_type = SSR;
2703         curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0);
2704
2705         sum_page = f2fs_get_sum_page(sbi, new_segno);
2706         if (IS_ERR(sum_page)) {
2707                 /* GC won't be able to use stale summary pages by cp_error */
2708                 memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
2709                 return;
2710         }
2711         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2712         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2713         f2fs_put_page(sum_page, 1);
2714 }
2715
2716 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2717                                 int alloc_mode, unsigned long long age);
2718
2719 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type,
2720                                         int target_type, int alloc_mode,
2721                                         unsigned long long age)
2722 {
2723         struct curseg_info *curseg = CURSEG_I(sbi, type);
2724
2725         curseg->seg_type = target_type;
2726
2727         if (get_ssr_segment(sbi, type, alloc_mode, age)) {
2728                 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
2729
2730                 curseg->seg_type = se->type;
2731                 change_curseg(sbi, type, true);
2732         } else {
2733                 /* allocate cold segment by default */
2734                 curseg->seg_type = CURSEG_COLD_DATA;
2735                 new_curseg(sbi, type, true);
2736         }
2737         stat_inc_seg_type(sbi, curseg);
2738 }
2739
2740 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
2741 {
2742         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
2743
2744         if (!sbi->am.atgc_enabled)
2745                 return;
2746
2747         down_read(&SM_I(sbi)->curseg_lock);
2748
2749         mutex_lock(&curseg->curseg_mutex);
2750         down_write(&SIT_I(sbi)->sentry_lock);
2751
2752         get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0);
2753
2754         up_write(&SIT_I(sbi)->sentry_lock);
2755         mutex_unlock(&curseg->curseg_mutex);
2756
2757         up_read(&SM_I(sbi)->curseg_lock);
2758
2759 }
2760 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
2761 {
2762         __f2fs_init_atgc_curseg(sbi);
2763 }
2764
2765 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2766 {
2767         struct curseg_info *curseg = CURSEG_I(sbi, type);
2768
2769         mutex_lock(&curseg->curseg_mutex);
2770         if (!curseg->inited)
2771                 goto out;
2772
2773         if (get_valid_blocks(sbi, curseg->segno, false)) {
2774                 write_sum_page(sbi, curseg->sum_blk,
2775                                 GET_SUM_BLOCK(sbi, curseg->segno));
2776         } else {
2777                 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2778                 __set_test_and_free(sbi, curseg->segno, true);
2779                 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2780         }
2781 out:
2782         mutex_unlock(&curseg->curseg_mutex);
2783 }
2784
2785 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
2786 {
2787         __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2788
2789         if (sbi->am.atgc_enabled)
2790                 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2791 }
2792
2793 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2794 {
2795         struct curseg_info *curseg = CURSEG_I(sbi, type);
2796
2797         mutex_lock(&curseg->curseg_mutex);
2798         if (!curseg->inited)
2799                 goto out;
2800         if (get_valid_blocks(sbi, curseg->segno, false))
2801                 goto out;
2802
2803         mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2804         __set_test_and_inuse(sbi, curseg->segno);
2805         mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2806 out:
2807         mutex_unlock(&curseg->curseg_mutex);
2808 }
2809
2810 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
2811 {
2812         __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2813
2814         if (sbi->am.atgc_enabled)
2815                 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2816 }
2817
2818 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2819                                 int alloc_mode, unsigned long long age)
2820 {
2821         struct curseg_info *curseg = CURSEG_I(sbi, type);
2822         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2823         unsigned segno = NULL_SEGNO;
2824         unsigned short seg_type = curseg->seg_type;
2825         int i, cnt;
2826         bool reversed = false;
2827
2828         sanity_check_seg_type(sbi, seg_type);
2829
2830         /* f2fs_need_SSR() already forces to do this */
2831         if (!v_ops->get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) {
2832                 curseg->next_segno = segno;
2833                 return 1;
2834         }
2835
2836         /* For node segments, let's do SSR more intensively */
2837         if (IS_NODESEG(seg_type)) {
2838                 if (seg_type >= CURSEG_WARM_NODE) {
2839                         reversed = true;
2840                         i = CURSEG_COLD_NODE;
2841                 } else {
2842                         i = CURSEG_HOT_NODE;
2843                 }
2844                 cnt = NR_CURSEG_NODE_TYPE;
2845         } else {
2846                 if (seg_type >= CURSEG_WARM_DATA) {
2847                         reversed = true;
2848                         i = CURSEG_COLD_DATA;
2849                 } else {
2850                         i = CURSEG_HOT_DATA;
2851                 }
2852                 cnt = NR_CURSEG_DATA_TYPE;
2853         }
2854
2855         for (; cnt-- > 0; reversed ? i-- : i++) {
2856                 if (i == seg_type)
2857                         continue;
2858                 if (!v_ops->get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) {
2859                         curseg->next_segno = segno;
2860                         return 1;
2861                 }
2862         }
2863
2864         /* find valid_blocks=0 in dirty list */
2865         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2866                 segno = get_free_segment(sbi);
2867                 if (segno != NULL_SEGNO) {
2868                         curseg->next_segno = segno;
2869                         return 1;
2870                 }
2871         }
2872         return 0;
2873 }
2874
2875 /*
2876  * flush out current segment and replace it with new segment
2877  * This function should be returned with success, otherwise BUG
2878  */
2879 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2880                                                 int type, bool force)
2881 {
2882         struct curseg_info *curseg = CURSEG_I(sbi, type);
2883
2884         if (force)
2885                 new_curseg(sbi, type, true);
2886         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2887                                         curseg->seg_type == CURSEG_WARM_NODE)
2888                 new_curseg(sbi, type, false);
2889         else if (curseg->alloc_type == LFS &&
2890                         is_next_segment_free(sbi, curseg, type) &&
2891                         likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2892                 new_curseg(sbi, type, false);
2893         else if (f2fs_need_SSR(sbi) &&
2894                         get_ssr_segment(sbi, type, SSR, 0))
2895                 change_curseg(sbi, type, true);
2896         else
2897                 new_curseg(sbi, type, false);
2898
2899         stat_inc_seg_type(sbi, curseg);
2900 }
2901
2902 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2903                                         unsigned int start, unsigned int end)
2904 {
2905         struct curseg_info *curseg = CURSEG_I(sbi, type);
2906         unsigned int segno;
2907
2908         down_read(&SM_I(sbi)->curseg_lock);
2909         mutex_lock(&curseg->curseg_mutex);
2910         down_write(&SIT_I(sbi)->sentry_lock);
2911
2912         segno = CURSEG_I(sbi, type)->segno;
2913         if (segno < start || segno > end)
2914                 goto unlock;
2915
2916         if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
2917                 change_curseg(sbi, type, true);
2918         else
2919                 new_curseg(sbi, type, true);
2920
2921         stat_inc_seg_type(sbi, curseg);
2922
2923         locate_dirty_segment(sbi, segno);
2924 unlock:
2925         up_write(&SIT_I(sbi)->sentry_lock);
2926
2927         if (segno != curseg->segno)
2928                 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2929                             type, segno, curseg->segno);
2930
2931         mutex_unlock(&curseg->curseg_mutex);
2932         up_read(&SM_I(sbi)->curseg_lock);
2933 }
2934
2935 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type,
2936                                                 bool new_sec, bool force)
2937 {
2938         struct curseg_info *curseg = CURSEG_I(sbi, type);
2939         unsigned int old_segno;
2940
2941         if (!curseg->inited)
2942                 goto alloc;
2943
2944         if (force || curseg->next_blkoff ||
2945                 get_valid_blocks(sbi, curseg->segno, new_sec))
2946                 goto alloc;
2947
2948         if (!get_ckpt_valid_blocks(sbi, curseg->segno, new_sec))
2949                 return;
2950 alloc:
2951         old_segno = curseg->segno;
2952         SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
2953         locate_dirty_segment(sbi, old_segno);
2954 }
2955
2956 static void __allocate_new_section(struct f2fs_sb_info *sbi,
2957                                                 int type, bool force)
2958 {
2959         __allocate_new_segment(sbi, type, true, force);
2960 }
2961
2962 void f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force)
2963 {
2964         down_read(&SM_I(sbi)->curseg_lock);
2965         down_write(&SIT_I(sbi)->sentry_lock);
2966         __allocate_new_section(sbi, type, force);
2967         up_write(&SIT_I(sbi)->sentry_lock);
2968         up_read(&SM_I(sbi)->curseg_lock);
2969 }
2970
2971 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2972 {
2973         int i;
2974
2975         down_read(&SM_I(sbi)->curseg_lock);
2976         down_write(&SIT_I(sbi)->sentry_lock);
2977         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
2978                 __allocate_new_segment(sbi, i, false, false);
2979         up_write(&SIT_I(sbi)->sentry_lock);
2980         up_read(&SM_I(sbi)->curseg_lock);
2981 }
2982
2983 static const struct segment_allocation default_salloc_ops = {
2984         .allocate_segment = allocate_segment_by_default,
2985 };
2986
2987 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2988                                                 struct cp_control *cpc)
2989 {
2990         __u64 trim_start = cpc->trim_start;
2991         bool has_candidate = false;
2992
2993         down_write(&SIT_I(sbi)->sentry_lock);
2994         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2995                 if (add_discard_addrs(sbi, cpc, true)) {
2996                         has_candidate = true;
2997                         break;
2998                 }
2999         }
3000         up_write(&SIT_I(sbi)->sentry_lock);
3001
3002         cpc->trim_start = trim_start;
3003         return has_candidate;
3004 }
3005
3006 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
3007                                         struct discard_policy *dpolicy,
3008                                         unsigned int start, unsigned int end)
3009 {
3010         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
3011         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
3012         struct rb_node **insert_p = NULL, *insert_parent = NULL;
3013         struct discard_cmd *dc;
3014         struct blk_plug plug;
3015         int issued;
3016         unsigned int trimmed = 0;
3017
3018 next:
3019         issued = 0;
3020
3021         mutex_lock(&dcc->cmd_lock);
3022         if (unlikely(dcc->rbtree_check))
3023                 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
3024                                                         &dcc->root, false));
3025
3026         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
3027                                         NULL, start,
3028                                         (struct rb_entry **)&prev_dc,
3029                                         (struct rb_entry **)&next_dc,
3030                                         &insert_p, &insert_parent, true, NULL);
3031         if (!dc)
3032                 dc = next_dc;
3033
3034         blk_start_plug(&plug);
3035
3036         while (dc && dc->lstart <= end) {
3037                 struct rb_node *node;
3038                 int err = 0;
3039
3040                 if (dc->len < dpolicy->granularity)
3041                         goto skip;
3042
3043                 if (dc->state != D_PREP) {
3044                         list_move_tail(&dc->list, &dcc->fstrim_list);
3045                         goto skip;
3046                 }
3047
3048                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
3049
3050                 if (issued >= dpolicy->max_requests) {
3051                         start = dc->lstart + dc->len;
3052
3053                         if (err)
3054                                 __remove_discard_cmd(sbi, dc);
3055
3056                         blk_finish_plug(&plug);
3057                         mutex_unlock(&dcc->cmd_lock);
3058                         trimmed += __wait_all_discard_cmd(sbi, NULL);
3059                         congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
3060                         goto next;
3061                 }
3062 skip:
3063                 node = rb_next(&dc->rb_node);
3064                 if (err)
3065                         __remove_discard_cmd(sbi, dc);
3066                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3067
3068                 if (fatal_signal_pending(current))
3069                         break;
3070         }
3071
3072         blk_finish_plug(&plug);
3073         mutex_unlock(&dcc->cmd_lock);
3074
3075         return trimmed;
3076 }
3077
3078 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3079 {
3080         __u64 start = F2FS_BYTES_TO_BLK(range->start);
3081         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
3082         unsigned int start_segno, end_segno;
3083         block_t start_block, end_block;
3084         struct cp_control cpc;
3085         struct discard_policy dpolicy;
3086         unsigned long long trimmed = 0;
3087         int err = 0;
3088         bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
3089
3090         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
3091                 return -EINVAL;
3092
3093         if (end < MAIN_BLKADDR(sbi))
3094                 goto out;
3095
3096         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
3097                 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
3098                 return -EFSCORRUPTED;
3099         }
3100
3101         /* start/end segment number in main_area */
3102         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3103         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3104                                                 GET_SEGNO(sbi, end);
3105         if (need_align) {
3106                 start_segno = rounddown(start_segno, sbi->segs_per_sec);
3107                 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
3108         }
3109
3110         cpc.reason = CP_DISCARD;
3111         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
3112         cpc.trim_start = start_segno;
3113         cpc.trim_end = end_segno;
3114
3115         if (sbi->discard_blks == 0)
3116                 goto out;
3117
3118         down_write(&sbi->gc_lock);
3119         err = f2fs_write_checkpoint(sbi, &cpc);
3120         up_write(&sbi->gc_lock);
3121         if (err)
3122                 goto out;
3123
3124         /*
3125          * We filed discard candidates, but actually we don't need to wait for
3126          * all of them, since they'll be issued in idle time along with runtime
3127          * discard option. User configuration looks like using runtime discard
3128          * or periodic fstrim instead of it.
3129          */
3130         if (f2fs_realtime_discard_enable(sbi))
3131                 goto out;
3132
3133         start_block = START_BLOCK(sbi, start_segno);
3134         end_block = START_BLOCK(sbi, end_segno + 1);
3135
3136         __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3137         trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3138                                         start_block, end_block);
3139
3140         trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3141                                         start_block, end_block);
3142 out:
3143         if (!err)
3144                 range->len = F2FS_BLK_TO_BYTES(trimmed);
3145         return err;
3146 }
3147
3148 static bool __has_curseg_space(struct f2fs_sb_info *sbi,
3149                                         struct curseg_info *curseg)
3150 {
3151         return curseg->next_blkoff < f2fs_usable_blks_in_seg(sbi,
3152                                                         curseg->segno);
3153 }
3154
3155 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
3156 {
3157         switch (hint) {
3158         case WRITE_LIFE_SHORT:
3159                 return CURSEG_HOT_DATA;
3160         case WRITE_LIFE_EXTREME:
3161                 return CURSEG_COLD_DATA;
3162         default:
3163                 return CURSEG_WARM_DATA;
3164         }
3165 }
3166
3167 /* This returns write hints for each segment type. This hints will be
3168  * passed down to block layer. There are mapping tables which depend on
3169  * the mount option 'whint_mode'.
3170  *
3171  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
3172  *
3173  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
3174  *
3175  * User                  F2FS                     Block
3176  * ----                  ----                     -----
3177  *                       META                     WRITE_LIFE_NOT_SET
3178  *                       HOT_NODE                 "
3179  *                       WARM_NODE                "
3180  *                       COLD_NODE                "
3181  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
3182  * extension list        "                        "
3183  *
3184  * -- buffered io
3185  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3186  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3187  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
3188  * WRITE_LIFE_NONE       "                        "
3189  * WRITE_LIFE_MEDIUM     "                        "
3190  * WRITE_LIFE_LONG       "                        "
3191  *
3192  * -- direct io
3193  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3194  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3195  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
3196  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
3197  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
3198  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
3199  *
3200  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
3201  *
3202  * User                  F2FS                     Block
3203  * ----                  ----                     -----
3204  *                       META                     WRITE_LIFE_MEDIUM;
3205  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
3206  *                       WARM_NODE                "
3207  *                       COLD_NODE                WRITE_LIFE_NONE
3208  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
3209  * extension list        "                        "
3210  *
3211  * -- buffered io
3212  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3213  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3214  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
3215  * WRITE_LIFE_NONE       "                        "
3216  * WRITE_LIFE_MEDIUM     "                        "
3217  * WRITE_LIFE_LONG       "                        "
3218  *
3219  * -- direct io
3220  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3221  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3222  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
3223  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
3224  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
3225  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
3226  */
3227
3228 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
3229                                 enum page_type type, enum temp_type temp)
3230 {
3231         if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
3232                 if (type == DATA) {
3233                         if (temp == WARM)
3234                                 return WRITE_LIFE_NOT_SET;
3235                         else if (temp == HOT)
3236                                 return WRITE_LIFE_SHORT;
3237                         else if (temp == COLD)
3238                                 return WRITE_LIFE_EXTREME;
3239                 } else {
3240                         return WRITE_LIFE_NOT_SET;
3241                 }
3242         } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
3243                 if (type == DATA) {
3244                         if (temp == WARM)
3245                                 return WRITE_LIFE_LONG;
3246                         else if (temp == HOT)
3247                                 return WRITE_LIFE_SHORT;
3248                         else if (temp == COLD)
3249                                 return WRITE_LIFE_EXTREME;
3250                 } else if (type == NODE) {
3251                         if (temp == WARM || temp == HOT)
3252                                 return WRITE_LIFE_NOT_SET;
3253                         else if (temp == COLD)
3254                                 return WRITE_LIFE_NONE;
3255                 } else if (type == META) {
3256                         return WRITE_LIFE_MEDIUM;
3257                 }
3258         }
3259         return WRITE_LIFE_NOT_SET;
3260 }
3261
3262 static int __get_segment_type_2(struct f2fs_io_info *fio)
3263 {
3264         if (fio->type == DATA)
3265                 return CURSEG_HOT_DATA;
3266         else
3267                 return CURSEG_HOT_NODE;
3268 }
3269
3270 static int __get_segment_type_4(struct f2fs_io_info *fio)
3271 {
3272         if (fio->type == DATA) {
3273                 struct inode *inode = fio->page->mapping->host;
3274
3275                 if (S_ISDIR(inode->i_mode))
3276                         return CURSEG_HOT_DATA;
3277                 else
3278                         return CURSEG_COLD_DATA;
3279         } else {
3280                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3281                         return CURSEG_WARM_NODE;
3282                 else
3283                         return CURSEG_COLD_NODE;
3284         }
3285 }
3286
3287 static int __get_segment_type_6(struct f2fs_io_info *fio)
3288 {
3289         if (fio->type == DATA) {
3290                 struct inode *inode = fio->page->mapping->host;
3291
3292                 if (is_cold_data(fio->page)) {
3293                         if (fio->sbi->am.atgc_enabled &&
3294                                 (fio->io_type == FS_DATA_IO) &&
3295                                 (fio->sbi->gc_mode != GC_URGENT_HIGH))
3296                                 return CURSEG_ALL_DATA_ATGC;
3297                         else
3298                                 return CURSEG_COLD_DATA;
3299                 }
3300                 if (file_is_cold(inode) || f2fs_need_compress_data(inode))
3301                         return CURSEG_COLD_DATA;
3302                 if (file_is_hot(inode) ||
3303                                 is_inode_flag_set(inode, FI_HOT_DATA) ||
3304                                 f2fs_is_atomic_file(inode) ||
3305                                 f2fs_is_volatile_file(inode))
3306                         return CURSEG_HOT_DATA;
3307                 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3308         } else {
3309                 if (IS_DNODE(fio->page))
3310                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3311                                                 CURSEG_HOT_NODE;
3312                 return CURSEG_COLD_NODE;
3313         }
3314 }
3315
3316 static int __get_segment_type(struct f2fs_io_info *fio)
3317 {
3318         int type = 0;
3319
3320         switch (F2FS_OPTION(fio->sbi).active_logs) {
3321         case 2:
3322                 type = __get_segment_type_2(fio);
3323                 break;
3324         case 4:
3325                 type = __get_segment_type_4(fio);
3326                 break;
3327         case 6:
3328                 type = __get_segment_type_6(fio);
3329                 break;
3330         default:
3331                 f2fs_bug_on(fio->sbi, true);
3332         }
3333
3334         if (IS_HOT(type))
3335                 fio->temp = HOT;
3336         else if (IS_WARM(type))
3337                 fio->temp = WARM;
3338         else
3339                 fio->temp = COLD;
3340         return type;
3341 }
3342
3343 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3344                 block_t old_blkaddr, block_t *new_blkaddr,
3345                 struct f2fs_summary *sum, int type,
3346                 struct f2fs_io_info *fio)
3347 {
3348         struct sit_info *sit_i = SIT_I(sbi);
3349         struct curseg_info *curseg = CURSEG_I(sbi, type);
3350         unsigned long long old_mtime;
3351         bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3352         struct seg_entry *se = NULL;
3353
3354         down_read(&SM_I(sbi)->curseg_lock);
3355
3356         mutex_lock(&curseg->curseg_mutex);
3357         down_write(&sit_i->sentry_lock);
3358
3359         if (from_gc) {
3360                 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3361                 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3362                 sanity_check_seg_type(sbi, se->type);
3363                 f2fs_bug_on(sbi, IS_NODESEG(se->type));
3364         }
3365         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3366
3367         f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg);
3368
3369         f2fs_wait_discard_bio(sbi, *new_blkaddr);
3370
3371         /*
3372          * __add_sum_entry should be resided under the curseg_mutex
3373          * because, this function updates a summary entry in the
3374          * current summary block.
3375          */
3376         __add_sum_entry(sbi, type, sum);
3377
3378         __refresh_next_blkoff(sbi, curseg);
3379
3380         stat_inc_block_count(sbi, curseg);
3381
3382         if (from_gc) {
3383                 old_mtime = get_segment_mtime(sbi, old_blkaddr);
3384         } else {
3385                 update_segment_mtime(sbi, old_blkaddr, 0);
3386                 old_mtime = 0;
3387         }
3388         update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3389
3390         /*
3391          * SIT information should be updated before segment allocation,
3392          * since SSR needs latest valid block information.
3393          */
3394         update_sit_entry(sbi, *new_blkaddr, 1);
3395         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3396                 update_sit_entry(sbi, old_blkaddr, -1);
3397
3398         if (!__has_curseg_space(sbi, curseg)) {
3399                 if (from_gc)
3400                         get_atssr_segment(sbi, type, se->type,
3401                                                 AT_SSR, se->mtime);
3402                 else
3403                         sit_i->s_ops->allocate_segment(sbi, type, false);
3404         }
3405         /*
3406          * segment dirty status should be updated after segment allocation,
3407          * so we just need to update status only one time after previous
3408          * segment being closed.
3409          */
3410         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3411         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3412
3413         up_write(&sit_i->sentry_lock);
3414
3415         if (page && IS_NODESEG(type)) {
3416                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3417
3418                 f2fs_inode_chksum_set(sbi, page);
3419         }
3420
3421         if (fio) {
3422                 struct f2fs_bio_info *io;
3423
3424                 if (F2FS_IO_ALIGNED(sbi))
3425                         fio->retry = false;
3426
3427                 INIT_LIST_HEAD(&fio->list);
3428                 fio->in_list = true;
3429                 io = sbi->write_io[fio->type] + fio->temp;
3430                 spin_lock(&io->io_lock);
3431                 list_add_tail(&fio->list, &io->io_list);
3432                 spin_unlock(&io->io_lock);
3433         }
3434
3435         mutex_unlock(&curseg->curseg_mutex);
3436
3437         up_read(&SM_I(sbi)->curseg_lock);
3438 }
3439
3440 static void update_device_state(struct f2fs_io_info *fio)
3441 {
3442         struct f2fs_sb_info *sbi = fio->sbi;
3443         unsigned int devidx;
3444
3445         if (!f2fs_is_multi_device(sbi))
3446                 return;
3447
3448         devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3449
3450         /* update device state for fsync */
3451         f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
3452
3453         /* update device state for checkpoint */
3454         if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3455                 spin_lock(&sbi->dev_lock);
3456                 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3457                 spin_unlock(&sbi->dev_lock);
3458         }
3459 }
3460
3461 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3462 {
3463         int type = __get_segment_type(fio);
3464         bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3465
3466         if (keep_order)
3467                 down_read(&fio->sbi->io_order_lock);
3468 reallocate:
3469         f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3470                         &fio->new_blkaddr, sum, type, fio);
3471         if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3472                 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3473                                         fio->old_blkaddr, fio->old_blkaddr);
3474
3475         /* writeout dirty page into bdev */
3476         f2fs_submit_page_write(fio);
3477         if (fio->retry) {
3478                 fio->old_blkaddr = fio->new_blkaddr;
3479                 goto reallocate;
3480         }
3481
3482         update_device_state(fio);
3483
3484         if (keep_order)
3485                 up_read(&fio->sbi->io_order_lock);
3486 }
3487
3488 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3489                                         enum iostat_type io_type)
3490 {
3491         struct f2fs_io_info fio = {
3492                 .sbi = sbi,
3493                 .type = META,
3494                 .temp = HOT,
3495                 .op = REQ_OP_WRITE,
3496                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3497                 .old_blkaddr = page->index,
3498                 .new_blkaddr = page->index,
3499                 .page = page,
3500                 .encrypted_page = NULL,
3501                 .in_list = false,
3502         };
3503
3504         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3505                 fio.op_flags &= ~REQ_META;
3506
3507         set_page_writeback(page);
3508         ClearPageError(page);
3509         f2fs_submit_page_write(&fio);
3510
3511         stat_inc_meta_count(sbi, page->index);
3512         f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3513 }
3514
3515 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3516 {
3517         struct f2fs_summary sum;
3518
3519         set_summary(&sum, nid, 0, 0);
3520         do_write_page(&sum, fio);
3521
3522         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3523 }
3524
3525 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3526                                         struct f2fs_io_info *fio)
3527 {
3528         struct f2fs_sb_info *sbi = fio->sbi;
3529         struct f2fs_summary sum;
3530
3531         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3532         set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3533         do_write_page(&sum, fio);
3534         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3535
3536         f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3537 }
3538
3539 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3540 {
3541         int err;
3542         struct f2fs_sb_info *sbi = fio->sbi;
3543         unsigned int segno;
3544
3545         fio->new_blkaddr = fio->old_blkaddr;
3546         /* i/o temperature is needed for passing down write hints */
3547         __get_segment_type(fio);
3548
3549         segno = GET_SEGNO(sbi, fio->new_blkaddr);
3550
3551         if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3552                 set_sbi_flag(sbi, SBI_NEED_FSCK);
3553                 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3554                           __func__, segno);
3555                 err = -EFSCORRUPTED;
3556                 goto drop_bio;
3557         }
3558
3559         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) || f2fs_cp_error(sbi)) {
3560                 err = -EIO;
3561                 goto drop_bio;
3562         }
3563
3564         stat_inc_inplace_blocks(fio->sbi);
3565
3566         if (fio->bio && !(SM_I(sbi)->ipu_policy & (1 << F2FS_IPU_NOCACHE)))
3567                 err = f2fs_merge_page_bio(fio);
3568         else
3569                 err = f2fs_submit_page_bio(fio);
3570         if (!err) {
3571                 update_device_state(fio);
3572                 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3573         }
3574
3575         return err;
3576 drop_bio:
3577         if (fio->bio && *(fio->bio)) {
3578                 struct bio *bio = *(fio->bio);
3579
3580                 bio->bi_status = BLK_STS_IOERR;
3581                 bio_endio(bio);
3582                 *(fio->bio) = NULL;
3583         }
3584         return err;
3585 }
3586
3587 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3588                                                 unsigned int segno)
3589 {
3590         int i;
3591
3592         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3593                 if (CURSEG_I(sbi, i)->segno == segno)
3594                         break;
3595         }
3596         return i;
3597 }
3598
3599 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3600                                 block_t old_blkaddr, block_t new_blkaddr,
3601                                 bool recover_curseg, bool recover_newaddr,
3602                                 bool from_gc)
3603 {
3604         struct sit_info *sit_i = SIT_I(sbi);
3605         struct curseg_info *curseg;
3606         unsigned int segno, old_cursegno;
3607         struct seg_entry *se;
3608         int type;
3609         unsigned short old_blkoff;
3610         unsigned char old_alloc_type;
3611
3612         segno = GET_SEGNO(sbi, new_blkaddr);
3613         se = get_seg_entry(sbi, segno);
3614         type = se->type;
3615
3616         down_write(&SM_I(sbi)->curseg_lock);
3617
3618         if (!recover_curseg) {
3619                 /* for recovery flow */
3620                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3621                         if (old_blkaddr == NULL_ADDR)
3622                                 type = CURSEG_COLD_DATA;
3623                         else
3624                                 type = CURSEG_WARM_DATA;
3625                 }
3626         } else {
3627                 if (IS_CURSEG(sbi, segno)) {
3628                         /* se->type is volatile as SSR allocation */
3629                         type = __f2fs_get_curseg(sbi, segno);
3630                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3631                 } else {
3632                         type = CURSEG_WARM_DATA;
3633                 }
3634         }
3635
3636         f2fs_bug_on(sbi, !IS_DATASEG(type));
3637         curseg = CURSEG_I(sbi, type);
3638
3639         mutex_lock(&curseg->curseg_mutex);
3640         down_write(&sit_i->sentry_lock);
3641
3642         old_cursegno = curseg->segno;
3643         old_blkoff = curseg->next_blkoff;
3644         old_alloc_type = curseg->alloc_type;
3645
3646         /* change the current segment */
3647         if (segno != curseg->segno) {
3648                 curseg->next_segno = segno;
3649                 change_curseg(sbi, type, true);
3650         }
3651
3652         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3653         __add_sum_entry(sbi, type, sum);
3654
3655         if (!recover_curseg || recover_newaddr) {
3656                 if (!from_gc)
3657                         update_segment_mtime(sbi, new_blkaddr, 0);
3658                 update_sit_entry(sbi, new_blkaddr, 1);
3659         }
3660         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3661                 invalidate_mapping_pages(META_MAPPING(sbi),
3662                                         old_blkaddr, old_blkaddr);
3663                 if (!from_gc)
3664                         update_segment_mtime(sbi, old_blkaddr, 0);
3665                 update_sit_entry(sbi, old_blkaddr, -1);
3666         }
3667
3668         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3669         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3670
3671         locate_dirty_segment(sbi, old_cursegno);
3672
3673         if (recover_curseg) {
3674                 if (old_cursegno != curseg->segno) {
3675                         curseg->next_segno = old_cursegno;
3676                         change_curseg(sbi, type, true);
3677                 }
3678                 curseg->next_blkoff = old_blkoff;
3679                 curseg->alloc_type = old_alloc_type;
3680         }
3681
3682         up_write(&sit_i->sentry_lock);
3683         mutex_unlock(&curseg->curseg_mutex);
3684         up_write(&SM_I(sbi)->curseg_lock);
3685 }
3686
3687 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3688                                 block_t old_addr, block_t new_addr,
3689                                 unsigned char version, bool recover_curseg,
3690                                 bool recover_newaddr)
3691 {
3692         struct f2fs_summary sum;
3693
3694         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3695
3696         f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3697                                         recover_curseg, recover_newaddr, false);
3698
3699         f2fs_update_data_blkaddr(dn, new_addr);
3700 }
3701
3702 void f2fs_wait_on_page_writeback(struct page *page,
3703                                 enum page_type type, bool ordered, bool locked)
3704 {
3705         if (PageWriteback(page)) {
3706                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3707
3708                 /* submit cached LFS IO */
3709                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3710                 /* sbumit cached IPU IO */
3711                 f2fs_submit_merged_ipu_write(sbi, NULL, page);
3712                 if (ordered) {
3713                         wait_on_page_writeback(page);
3714                         f2fs_bug_on(sbi, locked && PageWriteback(page));
3715                 } else {
3716                         wait_for_stable_page(page);
3717                 }
3718         }
3719 }
3720
3721 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3722 {
3723         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3724         struct page *cpage;
3725
3726         if (!f2fs_post_read_required(inode))
3727                 return;
3728
3729         if (!__is_valid_data_blkaddr(blkaddr))
3730                 return;
3731
3732         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3733         if (cpage) {
3734                 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3735                 f2fs_put_page(cpage, 1);
3736         }
3737 }
3738
3739 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3740                                                                 block_t len)
3741 {
3742         block_t i;
3743
3744         for (i = 0; i < len; i++)
3745                 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3746 }
3747
3748 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3749 {
3750         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3751         struct curseg_info *seg_i;
3752         unsigned char *kaddr;
3753         struct page *page;
3754         block_t start;
3755         int i, j, offset;
3756
3757         start = start_sum_block(sbi);
3758
3759         page = f2fs_get_meta_page(sbi, start++);
3760         if (IS_ERR(page))
3761                 return PTR_ERR(page);
3762         kaddr = (unsigned char *)page_address(page);
3763
3764         /* Step 1: restore nat cache */
3765         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3766         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3767
3768         /* Step 2: restore sit cache */
3769         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3770         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3771         offset = 2 * SUM_JOURNAL_SIZE;
3772
3773         /* Step 3: restore summary entries */
3774         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3775                 unsigned short blk_off;
3776                 unsigned int segno;
3777
3778                 seg_i = CURSEG_I(sbi, i);
3779                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3780                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3781                 seg_i->next_segno = segno;
3782                 reset_curseg(sbi, i, 0);
3783                 seg_i->alloc_type = ckpt->alloc_type[i];
3784                 seg_i->next_blkoff = blk_off;
3785
3786                 if (seg_i->alloc_type == SSR)
3787                         blk_off = sbi->blocks_per_seg;
3788
3789                 for (j = 0; j < blk_off; j++) {
3790                         struct f2fs_summary *s;
3791
3792                         s = (struct f2fs_summary *)(kaddr + offset);
3793                         seg_i->sum_blk->entries[j] = *s;
3794                         offset += SUMMARY_SIZE;
3795                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3796                                                 SUM_FOOTER_SIZE)
3797                                 continue;
3798
3799                         f2fs_put_page(page, 1);
3800                         page = NULL;
3801
3802                         page = f2fs_get_meta_page(sbi, start++);
3803                         if (IS_ERR(page))
3804                                 return PTR_ERR(page);
3805                         kaddr = (unsigned char *)page_address(page);
3806                         offset = 0;
3807                 }
3808         }
3809         f2fs_put_page(page, 1);
3810         return 0;
3811 }
3812
3813 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3814 {
3815         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3816         struct f2fs_summary_block *sum;
3817         struct curseg_info *curseg;
3818         struct page *new;
3819         unsigned short blk_off;
3820         unsigned int segno = 0;
3821         block_t blk_addr = 0;
3822         int err = 0;
3823
3824         /* get segment number and block addr */
3825         if (IS_DATASEG(type)) {
3826                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3827                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3828                                                         CURSEG_HOT_DATA]);
3829                 if (__exist_node_summaries(sbi))
3830                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
3831                 else
3832                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3833         } else {
3834                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3835                                                         CURSEG_HOT_NODE]);
3836                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3837                                                         CURSEG_HOT_NODE]);
3838                 if (__exist_node_summaries(sbi))
3839                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3840                                                         type - CURSEG_HOT_NODE);
3841                 else
3842                         blk_addr = GET_SUM_BLOCK(sbi, segno);
3843         }
3844
3845         new = f2fs_get_meta_page(sbi, blk_addr);
3846         if (IS_ERR(new))
3847                 return PTR_ERR(new);
3848         sum = (struct f2fs_summary_block *)page_address(new);
3849
3850         if (IS_NODESEG(type)) {
3851                 if (__exist_node_summaries(sbi)) {
3852                         struct f2fs_summary *ns = &sum->entries[0];
3853                         int i;
3854
3855                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3856                                 ns->version = 0;
3857                                 ns->ofs_in_node = 0;
3858                         }
3859                 } else {
3860                         err = f2fs_restore_node_summary(sbi, segno, sum);
3861                         if (err)
3862                                 goto out;
3863                 }
3864         }
3865
3866         /* set uncompleted segment to curseg */
3867         curseg = CURSEG_I(sbi, type);
3868         mutex_lock(&curseg->curseg_mutex);
3869
3870         /* update journal info */
3871         down_write(&curseg->journal_rwsem);
3872         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3873         up_write(&curseg->journal_rwsem);
3874
3875         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3876         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3877         curseg->next_segno = segno;
3878         reset_curseg(sbi, type, 0);
3879         curseg->alloc_type = ckpt->alloc_type[type];
3880         curseg->next_blkoff = blk_off;
3881         mutex_unlock(&curseg->curseg_mutex);
3882 out:
3883         f2fs_put_page(new, 1);
3884         return err;
3885 }
3886
3887 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3888 {
3889         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3890         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3891         int type = CURSEG_HOT_DATA;
3892         int err;
3893
3894         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3895                 int npages = f2fs_npages_for_summary_flush(sbi, true);
3896
3897                 if (npages >= 2)
3898                         f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3899                                                         META_CP, true);
3900
3901                 /* restore for compacted data summary */
3902                 err = read_compacted_summaries(sbi);
3903                 if (err)
3904                         return err;
3905                 type = CURSEG_HOT_NODE;
3906         }
3907
3908         if (__exist_node_summaries(sbi))
3909                 f2fs_ra_meta_pages(sbi,
3910                                 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
3911                                 NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
3912
3913         for (; type <= CURSEG_COLD_NODE; type++) {
3914                 err = read_normal_summaries(sbi, type);
3915                 if (err)
3916                         return err;
3917         }
3918
3919         /* sanity check for summary blocks */
3920         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3921                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3922                 f2fs_err(sbi, "invalid journal entries nats %u sits %u\n",
3923                          nats_in_cursum(nat_j), sits_in_cursum(sit_j));
3924                 return -EINVAL;
3925         }
3926
3927         return 0;
3928 }
3929
3930 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3931 {
3932         struct page *page;
3933         unsigned char *kaddr;
3934         struct f2fs_summary *summary;
3935         struct curseg_info *seg_i;
3936         int written_size = 0;
3937         int i, j;
3938
3939         page = f2fs_grab_meta_page(sbi, blkaddr++);
3940         kaddr = (unsigned char *)page_address(page);
3941         memset(kaddr, 0, PAGE_SIZE);
3942
3943         /* Step 1: write nat cache */
3944         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3945         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3946         written_size += SUM_JOURNAL_SIZE;
3947
3948         /* Step 2: write sit cache */
3949         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3950         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3951         written_size += SUM_JOURNAL_SIZE;
3952
3953         /* Step 3: write summary entries */
3954         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3955                 unsigned short blkoff;
3956
3957                 seg_i = CURSEG_I(sbi, i);
3958                 if (sbi->ckpt->alloc_type[i] == SSR)
3959                         blkoff = sbi->blocks_per_seg;
3960                 else
3961                         blkoff = curseg_blkoff(sbi, i);
3962
3963                 for (j = 0; j < blkoff; j++) {
3964                         if (!page) {
3965                                 page = f2fs_grab_meta_page(sbi, blkaddr++);
3966                                 kaddr = (unsigned char *)page_address(page);
3967                                 memset(kaddr, 0, PAGE_SIZE);
3968                                 written_size = 0;
3969                         }
3970                         summary = (struct f2fs_summary *)(kaddr + written_size);
3971                         *summary = seg_i->sum_blk->entries[j];
3972                         written_size += SUMMARY_SIZE;
3973
3974                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3975                                                         SUM_FOOTER_SIZE)
3976                                 continue;
3977
3978                         set_page_dirty(page);
3979                         f2fs_put_page(page, 1);
3980                         page = NULL;
3981                 }
3982         }
3983         if (page) {
3984                 set_page_dirty(page);
3985                 f2fs_put_page(page, 1);
3986         }
3987 }
3988
3989 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3990                                         block_t blkaddr, int type)
3991 {
3992         int i, end;
3993
3994         if (IS_DATASEG(type))
3995                 end = type + NR_CURSEG_DATA_TYPE;
3996         else
3997                 end = type + NR_CURSEG_NODE_TYPE;
3998
3999         for (i = type; i < end; i++)
4000                 write_current_sum_page(sbi, i, blkaddr + (i - type));
4001 }
4002
4003 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4004 {
4005         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
4006                 write_compacted_summaries(sbi, start_blk);
4007         else
4008                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
4009 }
4010
4011 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4012 {
4013         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
4014 }
4015
4016 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
4017                                         unsigned int val, int alloc)
4018 {
4019         int i;
4020
4021         if (type == NAT_JOURNAL) {
4022                 for (i = 0; i < nats_in_cursum(journal); i++) {
4023                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
4024                                 return i;
4025                 }
4026                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
4027                         return update_nats_in_cursum(journal, 1);
4028         } else if (type == SIT_JOURNAL) {
4029                 for (i = 0; i < sits_in_cursum(journal); i++)
4030                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
4031                                 return i;
4032                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
4033                         return update_sits_in_cursum(journal, 1);
4034         }
4035         return -1;
4036 }
4037
4038 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
4039                                         unsigned int segno)
4040 {
4041         return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
4042 }
4043
4044 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
4045                                         unsigned int start)
4046 {
4047         struct sit_info *sit_i = SIT_I(sbi);
4048         struct page *page;
4049         pgoff_t src_off, dst_off;
4050
4051         src_off = current_sit_addr(sbi, start);
4052         dst_off = next_sit_addr(sbi, src_off);
4053
4054         page = f2fs_grab_meta_page(sbi, dst_off);
4055         seg_info_to_sit_page(sbi, page, start);
4056
4057         set_page_dirty(page);
4058         set_to_next_sit(sit_i, start);
4059
4060         return page;
4061 }
4062
4063 static struct sit_entry_set *grab_sit_entry_set(void)
4064 {
4065         struct sit_entry_set *ses =
4066                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
4067
4068         ses->entry_cnt = 0;
4069         INIT_LIST_HEAD(&ses->set_list);
4070         return ses;
4071 }
4072
4073 static void release_sit_entry_set(struct sit_entry_set *ses)
4074 {
4075         list_del(&ses->set_list);
4076         kmem_cache_free(sit_entry_set_slab, ses);
4077 }
4078
4079 static void adjust_sit_entry_set(struct sit_entry_set *ses,
4080                                                 struct list_head *head)
4081 {
4082         struct sit_entry_set *next = ses;
4083
4084         if (list_is_last(&ses->set_list, head))
4085                 return;
4086
4087         list_for_each_entry_continue(next, head, set_list)
4088                 if (ses->entry_cnt <= next->entry_cnt)
4089                         break;
4090
4091         list_move_tail(&ses->set_list, &next->set_list);
4092 }
4093
4094 static void add_sit_entry(unsigned int segno, struct list_head *head)
4095 {
4096         struct sit_entry_set *ses;
4097         unsigned int start_segno = START_SEGNO(segno);
4098
4099         list_for_each_entry(ses, head, set_list) {
4100                 if (ses->start_segno == start_segno) {
4101                         ses->entry_cnt++;
4102                         adjust_sit_entry_set(ses, head);
4103                         return;
4104                 }
4105         }
4106
4107         ses = grab_sit_entry_set();
4108
4109         ses->start_segno = start_segno;
4110         ses->entry_cnt++;
4111         list_add(&ses->set_list, head);
4112 }
4113
4114 static void add_sits_in_set(struct f2fs_sb_info *sbi)
4115 {
4116         struct f2fs_sm_info *sm_info = SM_I(sbi);
4117         struct list_head *set_list = &sm_info->sit_entry_set;
4118         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
4119         unsigned int segno;
4120
4121         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
4122                 add_sit_entry(segno, set_list);
4123 }
4124
4125 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
4126 {
4127         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4128         struct f2fs_journal *journal = curseg->journal;
4129         int i;
4130
4131         down_write(&curseg->journal_rwsem);
4132         for (i = 0; i < sits_in_cursum(journal); i++) {
4133                 unsigned int segno;
4134                 bool dirtied;
4135
4136                 segno = le32_to_cpu(segno_in_journal(journal, i));
4137                 dirtied = __mark_sit_entry_dirty(sbi, segno);
4138
4139                 if (!dirtied)
4140                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
4141         }
4142         update_sits_in_cursum(journal, -i);
4143         up_write(&curseg->journal_rwsem);
4144 }
4145
4146 /*
4147  * CP calls this function, which flushes SIT entries including sit_journal,
4148  * and moves prefree segs to free segs.
4149  */
4150 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
4151 {
4152         struct sit_info *sit_i = SIT_I(sbi);
4153         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4154         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4155         struct f2fs_journal *journal = curseg->journal;
4156         struct sit_entry_set *ses, *tmp;
4157         struct list_head *head = &SM_I(sbi)->sit_entry_set;
4158         bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4159         struct seg_entry *se;
4160
4161         down_write(&sit_i->sentry_lock);
4162
4163         if (!sit_i->dirty_sentries)
4164                 goto out;
4165
4166         /*
4167          * add and account sit entries of dirty bitmap in sit entry
4168          * set temporarily
4169          */
4170         add_sits_in_set(sbi);
4171
4172         /*
4173          * if there are no enough space in journal to store dirty sit
4174          * entries, remove all entries from journal and add and account
4175          * them in sit entry set.
4176          */
4177         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4178                                                                 !to_journal)
4179                 remove_sits_in_journal(sbi);
4180
4181         /*
4182          * there are two steps to flush sit entries:
4183          * #1, flush sit entries to journal in current cold data summary block.
4184          * #2, flush sit entries to sit page.
4185          */
4186         list_for_each_entry_safe(ses, tmp, head, set_list) {
4187                 struct page *page = NULL;
4188                 struct f2fs_sit_block *raw_sit = NULL;
4189                 unsigned int start_segno = ses->start_segno;
4190                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4191                                                 (unsigned long)MAIN_SEGS(sbi));
4192                 unsigned int segno = start_segno;
4193
4194                 if (to_journal &&
4195                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4196                         to_journal = false;
4197
4198                 if (to_journal) {
4199                         down_write(&curseg->journal_rwsem);
4200                 } else {
4201                         page = get_next_sit_page(sbi, start_segno);
4202                         raw_sit = page_address(page);
4203                 }
4204
4205                 /* flush dirty sit entries in region of current sit set */
4206                 for_each_set_bit_from(segno, bitmap, end) {
4207                         int offset, sit_offset;
4208
4209                         se = get_seg_entry(sbi, segno);
4210 #ifdef CONFIG_F2FS_CHECK_FS
4211                         if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4212                                                 SIT_VBLOCK_MAP_SIZE))
4213                                 f2fs_bug_on(sbi, 1);
4214 #endif
4215
4216                         /* add discard candidates */
4217                         if (!(cpc->reason & CP_DISCARD)) {
4218                                 cpc->trim_start = segno;
4219                                 add_discard_addrs(sbi, cpc, false);
4220                         }
4221
4222                         if (to_journal) {
4223                                 offset = f2fs_lookup_journal_in_cursum(journal,
4224                                                         SIT_JOURNAL, segno, 1);
4225                                 f2fs_bug_on(sbi, offset < 0);
4226                                 segno_in_journal(journal, offset) =
4227                                                         cpu_to_le32(segno);
4228                                 seg_info_to_raw_sit(se,
4229                                         &sit_in_journal(journal, offset));
4230                                 check_block_count(sbi, segno,
4231                                         &sit_in_journal(journal, offset));
4232                         } else {
4233                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4234                                 seg_info_to_raw_sit(se,
4235                                                 &raw_sit->entries[sit_offset]);
4236                                 check_block_count(sbi, segno,
4237                                                 &raw_sit->entries[sit_offset]);
4238                         }
4239
4240                         __clear_bit(segno, bitmap);
4241                         sit_i->dirty_sentries--;
4242                         ses->entry_cnt--;
4243                 }
4244
4245                 if (to_journal)
4246                         up_write(&curseg->journal_rwsem);
4247                 else
4248                         f2fs_put_page(page, 1);
4249
4250                 f2fs_bug_on(sbi, ses->entry_cnt);
4251                 release_sit_entry_set(ses);
4252         }
4253
4254         f2fs_bug_on(sbi, !list_empty(head));
4255         f2fs_bug_on(sbi, sit_i->dirty_sentries);
4256 out:
4257         if (cpc->reason & CP_DISCARD) {
4258                 __u64 trim_start = cpc->trim_start;
4259
4260                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4261                         add_discard_addrs(sbi, cpc, false);
4262
4263                 cpc->trim_start = trim_start;
4264         }
4265         up_write(&sit_i->sentry_lock);
4266
4267         set_prefree_as_free_segments(sbi);
4268 }
4269
4270 static int build_sit_info(struct f2fs_sb_info *sbi)
4271 {
4272         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4273         struct sit_info *sit_i;
4274         unsigned int sit_segs, start;
4275         char *src_bitmap, *bitmap;
4276         unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4277
4278         /* allocate memory for SIT information */
4279         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4280         if (!sit_i)
4281                 return -ENOMEM;
4282
4283         SM_I(sbi)->sit_info = sit_i;
4284
4285         sit_i->sentries =
4286                 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4287                                               MAIN_SEGS(sbi)),
4288                               GFP_KERNEL);
4289         if (!sit_i->sentries)
4290                 return -ENOMEM;
4291
4292         main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4293         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4294                                                                 GFP_KERNEL);
4295         if (!sit_i->dirty_sentries_bitmap)
4296                 return -ENOMEM;
4297
4298 #ifdef CONFIG_F2FS_CHECK_FS
4299         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 4;
4300 #else
4301         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 3;
4302 #endif
4303         sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4304         if (!sit_i->bitmap)
4305                 return -ENOMEM;
4306
4307         bitmap = sit_i->bitmap;
4308
4309         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4310                 sit_i->sentries[start].cur_valid_map = bitmap;
4311                 bitmap += SIT_VBLOCK_MAP_SIZE;
4312
4313                 sit_i->sentries[start].ckpt_valid_map = bitmap;
4314                 bitmap += SIT_VBLOCK_MAP_SIZE;
4315
4316 #ifdef CONFIG_F2FS_CHECK_FS
4317                 sit_i->sentries[start].cur_valid_map_mir = bitmap;
4318                 bitmap += SIT_VBLOCK_MAP_SIZE;
4319 #endif
4320
4321                 sit_i->sentries[start].discard_map = bitmap;
4322                 bitmap += SIT_VBLOCK_MAP_SIZE;
4323         }
4324
4325         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4326         if (!sit_i->tmp_map)
4327                 return -ENOMEM;
4328
4329         if (__is_large_section(sbi)) {
4330                 sit_i->sec_entries =
4331                         f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4332                                                       MAIN_SECS(sbi)),
4333                                       GFP_KERNEL);
4334                 if (!sit_i->sec_entries)
4335                         return -ENOMEM;
4336         }
4337
4338         /* get information related with SIT */
4339         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4340
4341         /* setup SIT bitmap from ckeckpoint pack */
4342         sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4343         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4344
4345         sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4346         if (!sit_i->sit_bitmap)
4347                 return -ENOMEM;
4348
4349 #ifdef CONFIG_F2FS_CHECK_FS
4350         sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4351                                         sit_bitmap_size, GFP_KERNEL);
4352         if (!sit_i->sit_bitmap_mir)
4353                 return -ENOMEM;
4354
4355         sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4356                                         main_bitmap_size, GFP_KERNEL);
4357         if (!sit_i->invalid_segmap)
4358                 return -ENOMEM;
4359 #endif
4360
4361         /* init SIT information */
4362         sit_i->s_ops = &default_salloc_ops;
4363
4364         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4365         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4366         sit_i->written_valid_blocks = 0;
4367         sit_i->bitmap_size = sit_bitmap_size;
4368         sit_i->dirty_sentries = 0;
4369         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4370         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4371         sit_i->mounted_time = ktime_get_boottime_seconds();
4372         init_rwsem(&sit_i->sentry_lock);
4373         return 0;
4374 }
4375
4376 static int build_free_segmap(struct f2fs_sb_info *sbi)
4377 {
4378         struct free_segmap_info *free_i;
4379         unsigned int bitmap_size, sec_bitmap_size;
4380
4381         /* allocate memory for free segmap information */
4382         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4383         if (!free_i)
4384                 return -ENOMEM;
4385
4386         SM_I(sbi)->free_info = free_i;
4387
4388         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4389         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4390         if (!free_i->free_segmap)
4391                 return -ENOMEM;
4392
4393         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4394         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4395         if (!free_i->free_secmap)
4396                 return -ENOMEM;
4397
4398         /* set all segments as dirty temporarily */
4399         memset(free_i->free_segmap, 0xff, bitmap_size);
4400         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4401
4402         /* init free segmap information */
4403         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4404         free_i->free_segments = 0;
4405         free_i->free_sections = 0;
4406         spin_lock_init(&free_i->segmap_lock);
4407         return 0;
4408 }
4409
4410 static int build_curseg(struct f2fs_sb_info *sbi)
4411 {
4412         struct curseg_info *array;
4413         int i;
4414
4415         array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4416                                         sizeof(*array)), GFP_KERNEL);
4417         if (!array)
4418                 return -ENOMEM;
4419
4420         SM_I(sbi)->curseg_array = array;
4421
4422         for (i = 0; i < NO_CHECK_TYPE; i++) {
4423                 mutex_init(&array[i].curseg_mutex);
4424                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4425                 if (!array[i].sum_blk)
4426                         return -ENOMEM;
4427                 init_rwsem(&array[i].journal_rwsem);
4428                 array[i].journal = f2fs_kzalloc(sbi,
4429                                 sizeof(struct f2fs_journal), GFP_KERNEL);
4430                 if (!array[i].journal)
4431                         return -ENOMEM;
4432                 if (i < NR_PERSISTENT_LOG)
4433                         array[i].seg_type = CURSEG_HOT_DATA + i;
4434                 else if (i == CURSEG_COLD_DATA_PINNED)
4435                         array[i].seg_type = CURSEG_COLD_DATA;
4436                 else if (i == CURSEG_ALL_DATA_ATGC)
4437                         array[i].seg_type = CURSEG_COLD_DATA;
4438                 array[i].segno = NULL_SEGNO;
4439                 array[i].next_blkoff = 0;
4440                 array[i].inited = false;
4441         }
4442         return restore_curseg_summaries(sbi);
4443 }
4444
4445 static int build_sit_entries(struct f2fs_sb_info *sbi)
4446 {
4447         struct sit_info *sit_i = SIT_I(sbi);
4448         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4449         struct f2fs_journal *journal = curseg->journal;
4450         struct seg_entry *se;
4451         struct f2fs_sit_entry sit;
4452         int sit_blk_cnt = SIT_BLK_CNT(sbi);
4453         unsigned int i, start, end;
4454         unsigned int readed, start_blk = 0;
4455         int err = 0;
4456         block_t total_node_blocks = 0;
4457
4458         do {
4459                 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS,
4460                                                         META_SIT, true);
4461
4462                 start = start_blk * sit_i->sents_per_block;
4463                 end = (start_blk + readed) * sit_i->sents_per_block;
4464
4465                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4466                         struct f2fs_sit_block *sit_blk;
4467                         struct page *page;
4468
4469                         se = &sit_i->sentries[start];
4470                         page = get_current_sit_page(sbi, start);
4471                         if (IS_ERR(page))
4472                                 return PTR_ERR(page);
4473                         sit_blk = (struct f2fs_sit_block *)page_address(page);
4474                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4475                         f2fs_put_page(page, 1);
4476
4477                         err = check_block_count(sbi, start, &sit);
4478                         if (err)
4479                                 return err;
4480                         seg_info_from_raw_sit(se, &sit);
4481                         if (IS_NODESEG(se->type))
4482                                 total_node_blocks += se->valid_blocks;
4483
4484                         /* build discard map only one time */
4485                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4486                                 memset(se->discard_map, 0xff,
4487                                         SIT_VBLOCK_MAP_SIZE);
4488                         } else {
4489                                 memcpy(se->discard_map,
4490                                         se->cur_valid_map,
4491                                         SIT_VBLOCK_MAP_SIZE);
4492                                 sbi->discard_blks +=
4493                                         sbi->blocks_per_seg -
4494                                         se->valid_blocks;
4495                         }
4496
4497                         if (__is_large_section(sbi))
4498                                 get_sec_entry(sbi, start)->valid_blocks +=
4499                                                         se->valid_blocks;
4500                 }
4501                 start_blk += readed;
4502         } while (start_blk < sit_blk_cnt);
4503
4504         down_read(&curseg->journal_rwsem);
4505         for (i = 0; i < sits_in_cursum(journal); i++) {
4506                 unsigned int old_valid_blocks;
4507
4508                 start = le32_to_cpu(segno_in_journal(journal, i));
4509                 if (start >= MAIN_SEGS(sbi)) {
4510                         f2fs_err(sbi, "Wrong journal entry on segno %u",
4511                                  start);
4512                         err = -EFSCORRUPTED;
4513                         break;
4514                 }
4515
4516                 se = &sit_i->sentries[start];
4517                 sit = sit_in_journal(journal, i);
4518
4519                 old_valid_blocks = se->valid_blocks;
4520                 if (IS_NODESEG(se->type))
4521                         total_node_blocks -= old_valid_blocks;
4522
4523                 err = check_block_count(sbi, start, &sit);
4524                 if (err)
4525                         break;
4526                 seg_info_from_raw_sit(se, &sit);
4527                 if (IS_NODESEG(se->type))
4528                         total_node_blocks += se->valid_blocks;
4529
4530                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4531                         memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4532                 } else {
4533                         memcpy(se->discard_map, se->cur_valid_map,
4534                                                 SIT_VBLOCK_MAP_SIZE);
4535                         sbi->discard_blks += old_valid_blocks;
4536                         sbi->discard_blks -= se->valid_blocks;
4537                 }
4538
4539                 if (__is_large_section(sbi)) {
4540                         get_sec_entry(sbi, start)->valid_blocks +=
4541                                                         se->valid_blocks;
4542                         get_sec_entry(sbi, start)->valid_blocks -=
4543                                                         old_valid_blocks;
4544                 }
4545         }
4546         up_read(&curseg->journal_rwsem);
4547
4548         if (!err && total_node_blocks != valid_node_count(sbi)) {
4549                 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4550                          total_node_blocks, valid_node_count(sbi));
4551                 err = -EFSCORRUPTED;
4552         }
4553
4554         return err;
4555 }
4556
4557 static void init_free_segmap(struct f2fs_sb_info *sbi)
4558 {
4559         unsigned int start;
4560         int type;
4561         struct seg_entry *sentry;
4562
4563         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4564                 if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4565                         continue;
4566                 sentry = get_seg_entry(sbi, start);
4567                 if (!sentry->valid_blocks)
4568                         __set_free(sbi, start);
4569                 else
4570                         SIT_I(sbi)->written_valid_blocks +=
4571                                                 sentry->valid_blocks;
4572         }
4573
4574         /* set use the current segments */
4575         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4576                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4577
4578                 __set_test_and_inuse(sbi, curseg_t->segno);
4579         }
4580 }
4581
4582 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4583 {
4584         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4585         struct free_segmap_info *free_i = FREE_I(sbi);
4586         unsigned int segno = 0, offset = 0, secno;
4587         block_t valid_blocks, usable_blks_in_seg;
4588         block_t blks_per_sec = BLKS_PER_SEC(sbi);
4589
4590         while (1) {
4591                 /* find dirty segment based on free segmap */
4592                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4593                 if (segno >= MAIN_SEGS(sbi))
4594                         break;
4595                 offset = segno + 1;
4596                 valid_blocks = get_valid_blocks(sbi, segno, false);
4597                 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4598                 if (valid_blocks == usable_blks_in_seg || !valid_blocks)
4599                         continue;
4600                 if (valid_blocks > usable_blks_in_seg) {
4601                         f2fs_bug_on(sbi, 1);
4602                         continue;
4603                 }
4604                 mutex_lock(&dirty_i->seglist_lock);
4605                 __locate_dirty_segment(sbi, segno, DIRTY);
4606                 mutex_unlock(&dirty_i->seglist_lock);
4607         }
4608
4609         if (!__is_large_section(sbi))
4610                 return;
4611
4612         mutex_lock(&dirty_i->seglist_lock);
4613         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4614                 valid_blocks = get_valid_blocks(sbi, segno, true);
4615                 secno = GET_SEC_FROM_SEG(sbi, segno);
4616
4617                 if (!valid_blocks || valid_blocks == blks_per_sec)
4618                         continue;
4619                 if (IS_CURSEC(sbi, secno))
4620                         continue;
4621                 set_bit(secno, dirty_i->dirty_secmap);
4622         }
4623         mutex_unlock(&dirty_i->seglist_lock);
4624 }
4625
4626 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4627 {
4628         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4629         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4630
4631         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4632         if (!dirty_i->victim_secmap)
4633                 return -ENOMEM;
4634         return 0;
4635 }
4636
4637 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4638 {
4639         struct dirty_seglist_info *dirty_i;
4640         unsigned int bitmap_size, i;
4641
4642         /* allocate memory for dirty segments list information */
4643         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4644                                                                 GFP_KERNEL);
4645         if (!dirty_i)
4646                 return -ENOMEM;
4647
4648         SM_I(sbi)->dirty_info = dirty_i;
4649         mutex_init(&dirty_i->seglist_lock);
4650
4651         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4652
4653         for (i = 0; i < NR_DIRTY_TYPE; i++) {
4654                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4655                                                                 GFP_KERNEL);
4656                 if (!dirty_i->dirty_segmap[i])
4657                         return -ENOMEM;
4658         }
4659
4660         if (__is_large_section(sbi)) {
4661                 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4662                 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
4663                                                 bitmap_size, GFP_KERNEL);
4664                 if (!dirty_i->dirty_secmap)
4665                         return -ENOMEM;
4666         }
4667
4668         init_dirty_segmap(sbi);
4669         return init_victim_secmap(sbi);
4670 }
4671
4672 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4673 {
4674         int i;
4675
4676         /*
4677          * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4678          * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4679          */
4680         for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4681                 struct curseg_info *curseg = CURSEG_I(sbi, i);
4682                 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4683                 unsigned int blkofs = curseg->next_blkoff;
4684
4685                 sanity_check_seg_type(sbi, curseg->seg_type);
4686
4687                 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4688                         goto out;
4689
4690                 if (curseg->alloc_type == SSR)
4691                         continue;
4692
4693                 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4694                         if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4695                                 continue;
4696 out:
4697                         f2fs_err(sbi,
4698                                  "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4699                                  i, curseg->segno, curseg->alloc_type,
4700                                  curseg->next_blkoff, blkofs);
4701                         return -EFSCORRUPTED;
4702                 }
4703         }
4704         return 0;
4705 }
4706
4707 #ifdef CONFIG_BLK_DEV_ZONED
4708
4709 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4710                                     struct f2fs_dev_info *fdev,
4711                                     struct blk_zone *zone)
4712 {
4713         unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno;
4714         block_t zone_block, wp_block, last_valid_block;
4715         unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4716         int i, s, b, ret;
4717         struct seg_entry *se;
4718
4719         if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4720                 return 0;
4721
4722         wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block);
4723         wp_segno = GET_SEGNO(sbi, wp_block);
4724         wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4725         zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4726         zone_segno = GET_SEGNO(sbi, zone_block);
4727         zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno);
4728
4729         if (zone_segno >= MAIN_SEGS(sbi))
4730                 return 0;
4731
4732         /*
4733          * Skip check of zones cursegs point to, since
4734          * fix_curseg_write_pointer() checks them.
4735          */
4736         for (i = 0; i < NO_CHECK_TYPE; i++)
4737                 if (zone_secno == GET_SEC_FROM_SEG(sbi,
4738                                                    CURSEG_I(sbi, i)->segno))
4739                         return 0;
4740
4741         /*
4742          * Get last valid block of the zone.
4743          */
4744         last_valid_block = zone_block - 1;
4745         for (s = sbi->segs_per_sec - 1; s >= 0; s--) {
4746                 segno = zone_segno + s;
4747                 se = get_seg_entry(sbi, segno);
4748                 for (b = sbi->blocks_per_seg - 1; b >= 0; b--)
4749                         if (f2fs_test_bit(b, se->cur_valid_map)) {
4750                                 last_valid_block = START_BLOCK(sbi, segno) + b;
4751                                 break;
4752                         }
4753                 if (last_valid_block >= zone_block)
4754                         break;
4755         }
4756
4757         /*
4758          * If last valid block is beyond the write pointer, report the
4759          * inconsistency. This inconsistency does not cause write error
4760          * because the zone will not be selected for write operation until
4761          * it get discarded. Just report it.
4762          */
4763         if (last_valid_block >= wp_block) {
4764                 f2fs_notice(sbi, "Valid block beyond write pointer: "
4765                             "valid block[0x%x,0x%x] wp[0x%x,0x%x]",
4766                             GET_SEGNO(sbi, last_valid_block),
4767                             GET_BLKOFF_FROM_SEG0(sbi, last_valid_block),
4768                             wp_segno, wp_blkoff);
4769                 return 0;
4770         }
4771
4772         /*
4773          * If there is no valid block in the zone and if write pointer is
4774          * not at zone start, reset the write pointer.
4775          */
4776         if (last_valid_block + 1 == zone_block && zone->wp != zone->start) {
4777                 f2fs_notice(sbi,
4778                             "Zone without valid block has non-zero write "
4779                             "pointer. Reset the write pointer: wp[0x%x,0x%x]",
4780                             wp_segno, wp_blkoff);
4781                 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
4782                                         zone->len >> log_sectors_per_block);
4783                 if (ret) {
4784                         f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4785                                  fdev->path, ret);
4786                         return ret;
4787                 }
4788         }
4789
4790         return 0;
4791 }
4792
4793 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
4794                                                   block_t zone_blkaddr)
4795 {
4796         int i;
4797
4798         for (i = 0; i < sbi->s_ndevs; i++) {
4799                 if (!bdev_is_zoned(FDEV(i).bdev))
4800                         continue;
4801                 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
4802                                 zone_blkaddr <= FDEV(i).end_blk))
4803                         return &FDEV(i);
4804         }
4805
4806         return NULL;
4807 }
4808
4809 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
4810                               void *data)
4811 {
4812         memcpy(data, zone, sizeof(struct blk_zone));
4813         return 0;
4814 }
4815
4816 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
4817 {
4818         struct curseg_info *cs = CURSEG_I(sbi, type);
4819         struct f2fs_dev_info *zbd;
4820         struct blk_zone zone;
4821         unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
4822         block_t cs_zone_block, wp_block;
4823         unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4824         sector_t zone_sector;
4825         int err;
4826
4827         cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4828         cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4829
4830         zbd = get_target_zoned_dev(sbi, cs_zone_block);
4831         if (!zbd)
4832                 return 0;
4833
4834         /* report zone for the sector the curseg points to */
4835         zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4836                 << log_sectors_per_block;
4837         err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4838                                   report_one_zone_cb, &zone);
4839         if (err != 1) {
4840                 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4841                          zbd->path, err);
4842                 return err;
4843         }
4844
4845         if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4846                 return 0;
4847
4848         wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
4849         wp_segno = GET_SEGNO(sbi, wp_block);
4850         wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4851         wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
4852
4853         if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
4854                 wp_sector_off == 0)
4855                 return 0;
4856
4857         f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
4858                     "curseg[0x%x,0x%x] wp[0x%x,0x%x]",
4859                     type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff);
4860
4861         f2fs_notice(sbi, "Assign new section to curseg[%d]: "
4862                     "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff);
4863
4864         f2fs_allocate_new_section(sbi, type, true);
4865
4866         /* check consistency of the zone curseg pointed to */
4867         if (check_zone_write_pointer(sbi, zbd, &zone))
4868                 return -EIO;
4869
4870         /* check newly assigned zone */
4871         cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4872         cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4873
4874         zbd = get_target_zoned_dev(sbi, cs_zone_block);
4875         if (!zbd)
4876                 return 0;
4877
4878         zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4879                 << log_sectors_per_block;
4880         err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4881                                   report_one_zone_cb, &zone);
4882         if (err != 1) {
4883                 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4884                          zbd->path, err);
4885                 return err;
4886         }
4887
4888         if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4889                 return 0;
4890
4891         if (zone.wp != zone.start) {
4892                 f2fs_notice(sbi,
4893                             "New zone for curseg[%d] is not yet discarded. "
4894                             "Reset the zone: curseg[0x%x,0x%x]",
4895                             type, cs->segno, cs->next_blkoff);
4896                 err = __f2fs_issue_discard_zone(sbi, zbd->bdev,
4897                                 zone_sector >> log_sectors_per_block,
4898                                 zone.len >> log_sectors_per_block);
4899                 if (err) {
4900                         f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4901                                  zbd->path, err);
4902                         return err;
4903                 }
4904         }
4905
4906         return 0;
4907 }
4908
4909 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4910 {
4911         int i, ret;
4912
4913         for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4914                 ret = fix_curseg_write_pointer(sbi, i);
4915                 if (ret)
4916                         return ret;
4917         }
4918
4919         return 0;
4920 }
4921
4922 struct check_zone_write_pointer_args {
4923         struct f2fs_sb_info *sbi;
4924         struct f2fs_dev_info *fdev;
4925 };
4926
4927 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
4928                                       void *data)
4929 {
4930         struct check_zone_write_pointer_args *args;
4931
4932         args = (struct check_zone_write_pointer_args *)data;
4933
4934         return check_zone_write_pointer(args->sbi, args->fdev, zone);
4935 }
4936
4937 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4938 {
4939         int i, ret;
4940         struct check_zone_write_pointer_args args;
4941
4942         for (i = 0; i < sbi->s_ndevs; i++) {
4943                 if (!bdev_is_zoned(FDEV(i).bdev))
4944                         continue;
4945
4946                 args.sbi = sbi;
4947                 args.fdev = &FDEV(i);
4948                 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
4949                                           check_zone_write_pointer_cb, &args);
4950                 if (ret < 0)
4951                         return ret;
4952         }
4953
4954         return 0;
4955 }
4956
4957 static bool is_conv_zone(struct f2fs_sb_info *sbi, unsigned int zone_idx,
4958                                                 unsigned int dev_idx)
4959 {
4960         if (!bdev_is_zoned(FDEV(dev_idx).bdev))
4961                 return true;
4962         return !test_bit(zone_idx, FDEV(dev_idx).blkz_seq);
4963 }
4964
4965 /* Return the zone index in the given device */
4966 static unsigned int get_zone_idx(struct f2fs_sb_info *sbi, unsigned int secno,
4967                                         int dev_idx)
4968 {
4969         block_t sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
4970
4971         return (sec_start_blkaddr - FDEV(dev_idx).start_blk) >>
4972                                                 sbi->log_blocks_per_blkz;
4973 }
4974
4975 /*
4976  * Return the usable segments in a section based on the zone's
4977  * corresponding zone capacity. Zone is equal to a section.
4978  */
4979 static inline unsigned int f2fs_usable_zone_segs_in_sec(
4980                 struct f2fs_sb_info *sbi, unsigned int segno)
4981 {
4982         unsigned int dev_idx, zone_idx, unusable_segs_in_sec;
4983
4984         dev_idx = f2fs_target_device_index(sbi, START_BLOCK(sbi, segno));
4985         zone_idx = get_zone_idx(sbi, GET_SEC_FROM_SEG(sbi, segno), dev_idx);
4986
4987         /* Conventional zone's capacity is always equal to zone size */
4988         if (is_conv_zone(sbi, zone_idx, dev_idx))
4989                 return sbi->segs_per_sec;
4990
4991         /*
4992          * If the zone_capacity_blocks array is NULL, then zone capacity
4993          * is equal to the zone size for all zones
4994          */
4995         if (!FDEV(dev_idx).zone_capacity_blocks)
4996                 return sbi->segs_per_sec;
4997
4998         /* Get the segment count beyond zone capacity block */
4999         unusable_segs_in_sec = (sbi->blocks_per_blkz -
5000                                 FDEV(dev_idx).zone_capacity_blocks[zone_idx]) >>
5001                                 sbi->log_blocks_per_seg;
5002         return sbi->segs_per_sec - unusable_segs_in_sec;
5003 }
5004
5005 /*
5006  * Return the number of usable blocks in a segment. The number of blocks
5007  * returned is always equal to the number of blocks in a segment for
5008  * segments fully contained within a sequential zone capacity or a
5009  * conventional zone. For segments partially contained in a sequential
5010  * zone capacity, the number of usable blocks up to the zone capacity
5011  * is returned. 0 is returned in all other cases.
5012  */
5013 static inline unsigned int f2fs_usable_zone_blks_in_seg(
5014                         struct f2fs_sb_info *sbi, unsigned int segno)
5015 {
5016         block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
5017         unsigned int zone_idx, dev_idx, secno;
5018
5019         secno = GET_SEC_FROM_SEG(sbi, segno);
5020         seg_start = START_BLOCK(sbi, segno);
5021         dev_idx = f2fs_target_device_index(sbi, seg_start);
5022         zone_idx = get_zone_idx(sbi, secno, dev_idx);
5023
5024         /*
5025          * Conventional zone's capacity is always equal to zone size,
5026          * so, blocks per segment is unchanged.
5027          */
5028         if (is_conv_zone(sbi, zone_idx, dev_idx))
5029                 return sbi->blocks_per_seg;
5030
5031         if (!FDEV(dev_idx).zone_capacity_blocks)
5032                 return sbi->blocks_per_seg;
5033
5034         sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
5035         sec_cap_blkaddr = sec_start_blkaddr +
5036                                 FDEV(dev_idx).zone_capacity_blocks[zone_idx];
5037
5038         /*
5039          * If segment starts before zone capacity and spans beyond
5040          * zone capacity, then usable blocks are from seg start to
5041          * zone capacity. If the segment starts after the zone capacity,
5042          * then there are no usable blocks.
5043          */
5044         if (seg_start >= sec_cap_blkaddr)
5045                 return 0;
5046         if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr)
5047                 return sec_cap_blkaddr - seg_start;
5048
5049         return sbi->blocks_per_seg;
5050 }
5051 #else
5052 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5053 {
5054         return 0;
5055 }
5056
5057 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5058 {
5059         return 0;
5060 }
5061
5062 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
5063                                                         unsigned int segno)
5064 {
5065         return 0;
5066 }
5067
5068 static inline unsigned int f2fs_usable_zone_segs_in_sec(struct f2fs_sb_info *sbi,
5069                                                         unsigned int segno)
5070 {
5071         return 0;
5072 }
5073 #endif
5074 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
5075                                         unsigned int segno)
5076 {
5077         if (f2fs_sb_has_blkzoned(sbi))
5078                 return f2fs_usable_zone_blks_in_seg(sbi, segno);
5079
5080         return sbi->blocks_per_seg;
5081 }
5082
5083 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi,
5084                                         unsigned int segno)
5085 {
5086         if (f2fs_sb_has_blkzoned(sbi))
5087                 return f2fs_usable_zone_segs_in_sec(sbi, segno);
5088
5089         return sbi->segs_per_sec;
5090 }
5091
5092 /*
5093  * Update min, max modified time for cost-benefit GC algorithm
5094  */
5095 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5096 {
5097         struct sit_info *sit_i = SIT_I(sbi);
5098         unsigned int segno;
5099
5100         down_write(&sit_i->sentry_lock);
5101
5102         sit_i->min_mtime = ULLONG_MAX;
5103
5104         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
5105                 unsigned int i;
5106                 unsigned long long mtime = 0;
5107
5108                 for (i = 0; i < sbi->segs_per_sec; i++)
5109                         mtime += get_seg_entry(sbi, segno + i)->mtime;
5110
5111                 mtime = div_u64(mtime, sbi->segs_per_sec);
5112
5113                 if (sit_i->min_mtime > mtime)
5114                         sit_i->min_mtime = mtime;
5115         }
5116         sit_i->max_mtime = get_mtime(sbi, false);
5117         sit_i->dirty_max_mtime = 0;
5118         up_write(&sit_i->sentry_lock);
5119 }
5120
5121 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
5122 {
5123         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5124         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
5125         struct f2fs_sm_info *sm_info;
5126         int err;
5127
5128         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
5129         if (!sm_info)
5130                 return -ENOMEM;
5131
5132         /* init sm info */
5133         sbi->sm_info = sm_info;
5134         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5135         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5136         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5137         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5138         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5139         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5140         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
5141         sm_info->rec_prefree_segments = sm_info->main_segments *
5142                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
5143         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5144                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5145
5146         if (!f2fs_lfs_mode(sbi))
5147                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
5148         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
5149         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
5150         sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
5151         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
5152         sm_info->min_ssr_sections = reserved_sections(sbi);
5153
5154         INIT_LIST_HEAD(&sm_info->sit_entry_set);
5155
5156         init_rwsem(&sm_info->curseg_lock);
5157
5158         if (!f2fs_readonly(sbi->sb)) {
5159                 err = f2fs_create_flush_cmd_control(sbi);
5160                 if (err)
5161                         return err;
5162         }
5163
5164         err = create_discard_cmd_control(sbi);
5165         if (err)
5166                 return err;
5167
5168         err = build_sit_info(sbi);
5169         if (err)
5170                 return err;
5171         err = build_free_segmap(sbi);
5172         if (err)
5173                 return err;
5174         err = build_curseg(sbi);
5175         if (err)
5176                 return err;
5177
5178         /* reinit free segmap based on SIT */
5179         err = build_sit_entries(sbi);
5180         if (err)
5181                 return err;
5182
5183         init_free_segmap(sbi);
5184         err = build_dirty_segmap(sbi);
5185         if (err)
5186                 return err;
5187
5188         err = sanity_check_curseg(sbi);
5189         if (err)
5190                 return err;
5191
5192         init_min_max_mtime(sbi);
5193         return 0;
5194 }
5195
5196 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5197                 enum dirty_type dirty_type)
5198 {
5199         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5200
5201         mutex_lock(&dirty_i->seglist_lock);
5202         kvfree(dirty_i->dirty_segmap[dirty_type]);
5203         dirty_i->nr_dirty[dirty_type] = 0;
5204         mutex_unlock(&dirty_i->seglist_lock);
5205 }
5206
5207 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5208 {
5209         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5210
5211         kvfree(dirty_i->victim_secmap);
5212 }
5213
5214 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5215 {
5216         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5217         int i;
5218
5219         if (!dirty_i)
5220                 return;
5221
5222         /* discard pre-free/dirty segments list */
5223         for (i = 0; i < NR_DIRTY_TYPE; i++)
5224                 discard_dirty_segmap(sbi, i);
5225
5226         if (__is_large_section(sbi)) {
5227                 mutex_lock(&dirty_i->seglist_lock);
5228                 kvfree(dirty_i->dirty_secmap);
5229                 mutex_unlock(&dirty_i->seglist_lock);
5230         }
5231
5232         destroy_victim_secmap(sbi);
5233         SM_I(sbi)->dirty_info = NULL;
5234         kfree(dirty_i);
5235 }
5236
5237 static void destroy_curseg(struct f2fs_sb_info *sbi)
5238 {
5239         struct curseg_info *array = SM_I(sbi)->curseg_array;
5240         int i;
5241
5242         if (!array)
5243                 return;
5244         SM_I(sbi)->curseg_array = NULL;
5245         for (i = 0; i < NR_CURSEG_TYPE; i++) {
5246                 kfree(array[i].sum_blk);
5247                 kfree(array[i].journal);
5248         }
5249         kfree(array);
5250 }
5251
5252 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5253 {
5254         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5255
5256         if (!free_i)
5257                 return;
5258         SM_I(sbi)->free_info = NULL;
5259         kvfree(free_i->free_segmap);
5260         kvfree(free_i->free_secmap);
5261         kfree(free_i);
5262 }
5263
5264 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5265 {
5266         struct sit_info *sit_i = SIT_I(sbi);
5267
5268         if (!sit_i)
5269                 return;
5270
5271         if (sit_i->sentries)
5272                 kvfree(sit_i->bitmap);
5273         kfree(sit_i->tmp_map);
5274
5275         kvfree(sit_i->sentries);
5276         kvfree(sit_i->sec_entries);
5277         kvfree(sit_i->dirty_sentries_bitmap);
5278
5279         SM_I(sbi)->sit_info = NULL;
5280         kvfree(sit_i->sit_bitmap);
5281 #ifdef CONFIG_F2FS_CHECK_FS
5282         kvfree(sit_i->sit_bitmap_mir);
5283         kvfree(sit_i->invalid_segmap);
5284 #endif
5285         kfree(sit_i);
5286 }
5287
5288 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5289 {
5290         struct f2fs_sm_info *sm_info = SM_I(sbi);
5291
5292         if (!sm_info)
5293                 return;
5294         f2fs_destroy_flush_cmd_control(sbi, true);
5295         destroy_discard_cmd_control(sbi);
5296         destroy_dirty_segmap(sbi);
5297         destroy_curseg(sbi);
5298         destroy_free_segmap(sbi);
5299         destroy_sit_info(sbi);
5300         sbi->sm_info = NULL;
5301         kfree(sm_info);
5302 }
5303
5304 int __init f2fs_create_segment_manager_caches(void)
5305 {
5306         discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5307                         sizeof(struct discard_entry));
5308         if (!discard_entry_slab)
5309                 goto fail;
5310
5311         discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5312                         sizeof(struct discard_cmd));
5313         if (!discard_cmd_slab)
5314                 goto destroy_discard_entry;
5315
5316         sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5317                         sizeof(struct sit_entry_set));
5318         if (!sit_entry_set_slab)
5319                 goto destroy_discard_cmd;
5320
5321         inmem_entry_slab = f2fs_kmem_cache_create("f2fs_inmem_page_entry",
5322                         sizeof(struct inmem_pages));
5323         if (!inmem_entry_slab)
5324                 goto destroy_sit_entry_set;
5325         return 0;
5326
5327 destroy_sit_entry_set:
5328         kmem_cache_destroy(sit_entry_set_slab);
5329 destroy_discard_cmd:
5330         kmem_cache_destroy(discard_cmd_slab);
5331 destroy_discard_entry:
5332         kmem_cache_destroy(discard_entry_slab);
5333 fail:
5334         return -ENOMEM;
5335 }
5336
5337 void f2fs_destroy_segment_manager_caches(void)
5338 {
5339         kmem_cache_destroy(sit_entry_set_slab);
5340         kmem_cache_destroy(discard_cmd_slab);
5341         kmem_cache_destroy(discard_entry_slab);
5342         kmem_cache_destroy(inmem_entry_slab);
5343 }