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/sched/mm.h>
  13 #include <linux/prefetch.h>
  14 #include <linux/kthread.h>
  15 #include <linux/swap.h>
  16 #include <linux/timer.h>
  17 #include <linux/freezer.h>
  18 #include <linux/sched/signal.h>
  19 #include <linux/random.h>
  20
  21 #include "f2fs.h"
  22 #include "segment.h"
  23 #include "node.h"
  24 #include "gc.h"
  25 #include "iostat.h"
  26 #include <trace/events/f2fs.h>
  27
  28 #define __reverse_ffz(x) __reverse_ffs(~(x))
  29
  30 static struct kmem_cache *discard_entry_slab;
  31 static struct kmem_cache *discard_cmd_slab;
  32 static struct kmem_cache *sit_entry_set_slab;
  33 static struct kmem_cache *inmem_entry_slab;
  34
  35 static unsigned long __reverse_ulong(unsigned char *str)
  36 {
  37         unsigned long tmp = 0;
  38         int shift = 24, idx = 0;
  39
  40 #if BITS_PER_LONG == 64
  41         shift = 56;
  42 #endif
  43         while (shift >= 0) {
  44                 tmp |= (unsigned long)str[idx++] << shift;
  45                 shift -= BITS_PER_BYTE;
  46         }
  47         return tmp;
  48 }
  49
  50 /*
  51  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  52  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  53  */
  54 static inline unsigned long __reverse_ffs(unsigned long word)
  55 {
  56         int num = 0;
  57
  58 #if BITS_PER_LONG == 64
  59         if ((word & 0xffffffff00000000UL) == 0)
  60                 num += 32;
  61         else
  62                 word >>= 32;
  63 #endif
  64         if ((word & 0xffff0000) == 0)
  65                 num += 16;
  66         else
  67                 word >>= 16;
  68
  69         if ((word & 0xff00) == 0)
  70                 num += 8;
  71         else
  72                 word >>= 8;
  73
  74         if ((word & 0xf0) == 0)
  75                 num += 4;
  76         else
  77                 word >>= 4;
  78
  79         if ((word & 0xc) == 0)
  80                 num += 2;
  81         else
  82                 word >>= 2;
  83
  84         if ((word & 0x2) == 0)
  85                 num += 1;
  86         return num;
  87 }
  88
  89 /*
  90  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  91  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  92  * @size must be integral times of unsigned long.
  93  * Example:
  94  *                             MSB <--> LSB
  95  *   f2fs_set_bit(0, bitmap) => 1000 0000
  96  *   f2fs_set_bit(7, bitmap) => 0000 0001
  97  */
  98 static unsigned long __find_rev_next_bit(const unsigned long *addr,
  99                         unsigned long size, unsigned long offset)
 100 {
 101         const unsigned long *p = addr + BIT_WORD(offset);
 102         unsigned long result = size;
 103         unsigned long tmp;
 104
 105         if (offset >= size)
 106                 return size;
 107
 108         size -= (offset & ~(BITS_PER_LONG - 1));
 109         offset %= BITS_PER_LONG;
 110
 111         while (1) {
 112                 if (*p == 0)
 113                         goto pass;
 114
 115                 tmp = __reverse_ulong((unsigned char *)p);
 116
 117                 tmp &= ~0UL >> offset;
 118                 if (size < BITS_PER_LONG)
 119                         tmp &= (~0UL << (BITS_PER_LONG - size));
 120                 if (tmp)
 121                         goto found;
 122 pass:
 123                 if (size <= BITS_PER_LONG)
 124                         break;
 125                 size -= BITS_PER_LONG;
 126                 offset = 0;
 127                 p++;
 128         }
 129         return result;
 130 found:
 131         return result - size + __reverse_ffs(tmp);
 132 }
 133
 134 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 135                         unsigned long size, unsigned long offset)
 136 {
 137         const unsigned long *p = addr + BIT_WORD(offset);
 138         unsigned long result = size;
 139         unsigned long tmp;
 140
 141         if (offset >= size)
 142                 return size;
 143
 144         size -= (offset & ~(BITS_PER_LONG - 1));
 145         offset %= BITS_PER_LONG;
 146
 147         while (1) {
 148                 if (*p == ~0UL)
 149                         goto pass;
 150
 151                 tmp = __reverse_ulong((unsigned char *)p);
 152
 153                 if (offset)
 154                         tmp |= ~0UL << (BITS_PER_LONG - offset);
 155                 if (size < BITS_PER_LONG)
 156                         tmp |= ~0UL >> size;
 157                 if (tmp != ~0UL)
 158                         goto found;
 159 pass:
 160                 if (size <= BITS_PER_LONG)
 161                         break;
 162                 size -= BITS_PER_LONG;
 163                 offset = 0;
 164                 p++;
 165         }
 166         return result;
 167 found:
 168         return result - size + __reverse_ffz(tmp);
 169 }
 170
 171 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
 172 {
 173         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 174         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 175         int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
 176
 177         if (f2fs_lfs_mode(sbi))
 178                 return false;
 179         if (sbi->gc_mode == GC_URGENT_HIGH)
 180                 return true;
 181         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
 182                 return true;
 183
 184         return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
 185                         SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
 186 }
 187
 188 void f2fs_register_inmem_page(struct inode *inode, struct page *page)
 189 {
 190         struct inmem_pages *new;
 191
 192         set_page_private_atomic(page);
 193
 194         new = f2fs_kmem_cache_alloc(inmem_entry_slab,
 195                                         GFP_NOFS, true, NULL);
 196
 197         /* add atomic page indices to the list */
 198         new->page = page;
 199         INIT_LIST_HEAD(&new->list);
 200
 201         /* increase reference count with clean state */
 202         get_page(page);
 203         mutex_lock(&F2FS_I(inode)->inmem_lock);
 204         list_add_tail(&new->list, &F2FS_I(inode)->inmem_pages);
 205         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 206         mutex_unlock(&F2FS_I(inode)->inmem_lock);
 207
 208         trace_f2fs_register_inmem_page(page, INMEM);
 209 }
 210
 211 static int __revoke_inmem_pages(struct inode *inode,
 212                                 struct list_head *head, bool drop, bool recover,
 213                                 bool trylock)
 214 {
 215         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 216         struct inmem_pages *cur, *tmp;
 217         int err = 0;
 218
 219         list_for_each_entry_safe(cur, tmp, head, list) {
 220                 struct page *page = cur->page;
 221
 222                 if (drop)
 223                         trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 224
 225                 if (trylock) {
 226                         /*
 227                          * to avoid deadlock in between page lock and
 228                          * inmem_lock.
 229                          */
 230                         if (!trylock_page(page))
 231                                 continue;
 232                 } else {
 233                         lock_page(page);
 234                 }
 235
 236                 f2fs_wait_on_page_writeback(page, DATA, true, true);
 237
 238                 if (recover) {
 239                         struct dnode_of_data dn;
 240                         struct node_info ni;
 241
 242                         trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 243 retry:
 244                         set_new_dnode(&dn, inode, NULL, NULL, 0);
 245                         err = f2fs_get_dnode_of_data(&dn, page->index,
 246                                                                 LOOKUP_NODE);
 247                         if (err) {
 248                                 if (err == -ENOMEM) {
 249                                         memalloc_retry_wait(GFP_NOFS);
 250                                         goto retry;
 251                                 }
 252                                 err = -EAGAIN;
 253                                 goto next;
 254                         }
 255
 256                         err = f2fs_get_node_info(sbi, dn.nid, &ni, false);
 257                         if (err) {
 258                                 f2fs_put_dnode(&dn);
 259                                 return err;
 260                         }
 261
 262                         if (cur->old_addr == NEW_ADDR) {
 263                                 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
 264                                 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
 265                         } else
 266                                 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 267                                         cur->old_addr, ni.version, true, true);
 268                         f2fs_put_dnode(&dn);
 269                 }
 270 next:
 271                 /* we don't need to invalidate this in the sccessful status */
 272                 if (drop || recover) {
 273                         ClearPageUptodate(page);
 274                         clear_page_private_gcing(page);
 275                 }
 276                 detach_page_private(page);
 277                 set_page_private(page, 0);
 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         f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
 317         if (gc_failure) {
 318                 if (++looped >= count)
 319                         return;
 320         }
 321         goto next;
 322 }
 323
 324 void f2fs_drop_inmem_pages(struct inode *inode)
 325 {
 326         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 327         struct f2fs_inode_info *fi = F2FS_I(inode);
 328
 329         do {
 330                 mutex_lock(&fi->inmem_lock);
 331                 if (list_empty(&fi->inmem_pages)) {
 332                         fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
 333
 334                         spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 335                         if (!list_empty(&fi->inmem_ilist))
 336                                 list_del_init(&fi->inmem_ilist);
 337                         if (f2fs_is_atomic_file(inode)) {
 338                                 clear_inode_flag(inode, FI_ATOMIC_FILE);
 339                                 sbi->atomic_files--;
 340                         }
 341                         spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 342
 343                         mutex_unlock(&fi->inmem_lock);
 344                         break;
 345                 }
 346                 __revoke_inmem_pages(inode, &fi->inmem_pages,
 347                                                 true, false, true);
 348                 mutex_unlock(&fi->inmem_lock);
 349         } while (1);
 350 }
 351
 352 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
 353 {
 354         struct f2fs_inode_info *fi = F2FS_I(inode);
 355         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 356         struct list_head *head = &fi->inmem_pages;
 357         struct inmem_pages *cur = NULL;
 358
 359         f2fs_bug_on(sbi, !page_private_atomic(page));
 360
 361         mutex_lock(&fi->inmem_lock);
 362         list_for_each_entry(cur, head, list) {
 363                 if (cur->page == page)
 364                         break;
 365         }
 366
 367         f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
 368         list_del(&cur->list);
 369         mutex_unlock(&fi->inmem_lock);
 370
 371         dec_page_count(sbi, F2FS_INMEM_PAGES);
 372         kmem_cache_free(inmem_entry_slab, cur);
 373
 374         ClearPageUptodate(page);
 375         clear_page_private_atomic(page);
 376         f2fs_put_page(page, 0);
 377
 378         detach_page_private(page);
 379         set_page_private(page, 0);
 380
 381         trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 382 }
 383
 384 static int __f2fs_commit_inmem_pages(struct inode *inode)
 385 {
 386         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 387         struct f2fs_inode_info *fi = F2FS_I(inode);
 388         struct inmem_pages *cur, *tmp;
 389         struct f2fs_io_info fio = {
 390                 .sbi = sbi,
 391                 .ino = inode->i_ino,
 392                 .type = DATA,
 393                 .op = REQ_OP_WRITE,
 394                 .op_flags = REQ_SYNC | REQ_PRIO,
 395                 .io_type = FS_DATA_IO,
 396         };
 397         struct list_head revoke_list;
 398         bool submit_bio = false;
 399         int err = 0;
 400
 401         INIT_LIST_HEAD(&revoke_list);
 402
 403         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 404                 struct page *page = cur->page;
 405
 406                 lock_page(page);
 407                 if (page->mapping == inode->i_mapping) {
 408                         trace_f2fs_commit_inmem_page(page, INMEM);
 409
 410                         f2fs_wait_on_page_writeback(page, DATA, true, true);
 411
 412                         set_page_dirty(page);
 413                         if (clear_page_dirty_for_io(page)) {
 414                                 inode_dec_dirty_pages(inode);
 415                                 f2fs_remove_dirty_inode(inode);
 416                         }
 417 retry:
 418                         fio.page = page;
 419                         fio.old_blkaddr = NULL_ADDR;
 420                         fio.encrypted_page = NULL;
 421                         fio.need_lock = LOCK_DONE;
 422                         err = f2fs_do_write_data_page(&fio);
 423                         if (err) {
 424                                 if (err == -ENOMEM) {
 425                                         memalloc_retry_wait(GFP_NOFS);
 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         f2fs_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         f2fs_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                         f2fs_down_write(&sbi->gc_lock);
 524                         f2fs_gc(sbi, false, false, false, NULL_SEGNO);
 525                 }
 526         }
 527 }
 528
 529 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi)
 530 {
 531         int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2;
 532         unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS);
 533         unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA);
 534         unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES);
 535         unsigned int meta = get_pages(sbi, F2FS_DIRTY_META);
 536         unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA);
 537         unsigned int threshold = sbi->blocks_per_seg * factor *
 538                                         DEFAULT_DIRTY_THRESHOLD;
 539         unsigned int global_threshold = threshold * 3 / 2;
 540
 541         if (dents >= threshold || qdata >= threshold ||
 542                 nodes >= threshold || meta >= threshold ||
 543                 imeta >= threshold)
 544                 return true;
 545         return dents + qdata + nodes + meta + imeta >  global_threshold;
 546 }
 547
 548 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
 549 {
 550         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 551                 return;
 552
 553         /* try to shrink extent cache when there is no enough memory */
 554         if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
 555                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 556
 557         /* check the # of cached NAT entries */
 558         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
 559                 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 560
 561         if (!f2fs_available_free_memory(sbi, FREE_NIDS))
 562                 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
 563         else
 564                 f2fs_build_free_nids(sbi, false, false);
 565
 566         if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) ||
 567                 excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi))
 568                 goto do_sync;
 569
 570         /* there is background inflight IO or foreground operation recently */
 571         if (is_inflight_io(sbi, REQ_TIME) ||
 572                 (!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem)))
 573                 return;
 574
 575         /* exceed periodical checkpoint timeout threshold */
 576         if (f2fs_time_over(sbi, CP_TIME))
 577                 goto do_sync;
 578
 579         /* checkpoint is the only way to shrink partial cached entries */
 580         if (f2fs_available_free_memory(sbi, NAT_ENTRIES) &&
 581                 f2fs_available_free_memory(sbi, INO_ENTRIES))
 582                 return;
 583
 584 do_sync:
 585         if (test_opt(sbi, DATA_FLUSH) && from_bg) {
 586                 struct blk_plug plug;
 587
 588                 mutex_lock(&sbi->flush_lock);
 589
 590                 blk_start_plug(&plug);
 591                 f2fs_sync_dirty_inodes(sbi, FILE_INODE);
 592                 blk_finish_plug(&plug);
 593
 594                 mutex_unlock(&sbi->flush_lock);
 595         }
 596         f2fs_sync_fs(sbi->sb, true);
 597         stat_inc_bg_cp_count(sbi->stat_info);
 598 }
 599
 600 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 601                                 struct block_device *bdev)
 602 {
 603         int ret = blkdev_issue_flush(bdev);
 604
 605         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 606                                 test_opt(sbi, FLUSH_MERGE), ret);
 607         return ret;
 608 }
 609
 610 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 611 {
 612         int ret = 0;
 613         int i;
 614
 615         if (!f2fs_is_multi_device(sbi))
 616                 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 617
 618         for (i = 0; i < sbi->s_ndevs; i++) {
 619                 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
 620                         continue;
 621                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 622                 if (ret)
 623                         break;
 624         }
 625         return ret;
 626 }
 627
 628 static int issue_flush_thread(void *data)
 629 {
 630         struct f2fs_sb_info *sbi = data;
 631         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 632         wait_queue_head_t *q = &fcc->flush_wait_queue;
 633 repeat:
 634         if (kthread_should_stop())
 635                 return 0;
 636
 637         if (!llist_empty(&fcc->issue_list)) {
 638                 struct flush_cmd *cmd, *next;
 639                 int ret;
 640
 641                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 642                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 643
 644                 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 645
 646                 ret = submit_flush_wait(sbi, cmd->ino);
 647                 atomic_inc(&fcc->issued_flush);
 648
 649                 llist_for_each_entry_safe(cmd, next,
 650                                           fcc->dispatch_list, llnode) {
 651                         cmd->ret = ret;
 652                         complete(&cmd->wait);
 653                 }
 654                 fcc->dispatch_list = NULL;
 655         }
 656
 657         wait_event_interruptible(*q,
 658                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 659         goto repeat;
 660 }
 661
 662 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 663 {
 664         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 665         struct flush_cmd cmd;
 666         int ret;
 667
 668         if (test_opt(sbi, NOBARRIER))
 669                 return 0;
 670
 671         if (!test_opt(sbi, FLUSH_MERGE)) {
 672                 atomic_inc(&fcc->queued_flush);
 673                 ret = submit_flush_wait(sbi, ino);
 674                 atomic_dec(&fcc->queued_flush);
 675                 atomic_inc(&fcc->issued_flush);
 676                 return ret;
 677         }
 678
 679         if (atomic_inc_return(&fcc->queued_flush) == 1 ||
 680             f2fs_is_multi_device(sbi)) {
 681                 ret = submit_flush_wait(sbi, ino);
 682                 atomic_dec(&fcc->queued_flush);
 683
 684                 atomic_inc(&fcc->issued_flush);
 685                 return ret;
 686         }
 687
 688         cmd.ino = ino;
 689         init_completion(&cmd.wait);
 690
 691         llist_add(&cmd.llnode, &fcc->issue_list);
 692
 693         /*
 694          * update issue_list before we wake up issue_flush thread, this
 695          * smp_mb() pairs with another barrier in ___wait_event(), see
 696          * more details in comments of waitqueue_active().
 697          */
 698         smp_mb();
 699
 700         if (waitqueue_active(&fcc->flush_wait_queue))
 701                 wake_up(&fcc->flush_wait_queue);
 702
 703         if (fcc->f2fs_issue_flush) {
 704                 wait_for_completion(&cmd.wait);
 705                 atomic_dec(&fcc->queued_flush);
 706         } else {
 707                 struct llist_node *list;
 708
 709                 list = llist_del_all(&fcc->issue_list);
 710                 if (!list) {
 711                         wait_for_completion(&cmd.wait);
 712                         atomic_dec(&fcc->queued_flush);
 713                 } else {
 714                         struct flush_cmd *tmp, *next;
 715
 716                         ret = submit_flush_wait(sbi, ino);
 717
 718                         llist_for_each_entry_safe(tmp, next, list, llnode) {
 719                                 if (tmp == &cmd) {
 720                                         cmd.ret = ret;
 721                                         atomic_dec(&fcc->queued_flush);
 722                                         continue;
 723                                 }
 724                                 tmp->ret = ret;
 725                                 complete(&tmp->wait);
 726                         }
 727                 }
 728         }
 729
 730         return cmd.ret;
 731 }
 732
 733 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
 734 {
 735         dev_t dev = sbi->sb->s_bdev->bd_dev;
 736         struct flush_cmd_control *fcc;
 737         int err = 0;
 738
 739         if (SM_I(sbi)->fcc_info) {
 740                 fcc = SM_I(sbi)->fcc_info;
 741                 if (fcc->f2fs_issue_flush)
 742                         return err;
 743                 goto init_thread;
 744         }
 745
 746         fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
 747         if (!fcc)
 748                 return -ENOMEM;
 749         atomic_set(&fcc->issued_flush, 0);
 750         atomic_set(&fcc->queued_flush, 0);
 751         init_waitqueue_head(&fcc->flush_wait_queue);
 752         init_llist_head(&fcc->issue_list);
 753         SM_I(sbi)->fcc_info = fcc;
 754         if (!test_opt(sbi, FLUSH_MERGE))
 755                 return err;
 756
 757 init_thread:
 758         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 759                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 760         if (IS_ERR(fcc->f2fs_issue_flush)) {
 761                 err = PTR_ERR(fcc->f2fs_issue_flush);
 762                 kfree(fcc);
 763                 SM_I(sbi)->fcc_info = NULL;
 764                 return err;
 765         }
 766
 767         return err;
 768 }
 769
 770 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 771 {
 772         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 773
 774         if (fcc && fcc->f2fs_issue_flush) {
 775                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 776
 777                 fcc->f2fs_issue_flush = NULL;
 778                 kthread_stop(flush_thread);
 779         }
 780         if (free) {
 781                 kfree(fcc);
 782                 SM_I(sbi)->fcc_info = NULL;
 783         }
 784 }
 785
 786 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
 787 {
 788         int ret = 0, i;
 789
 790         if (!f2fs_is_multi_device(sbi))
 791                 return 0;
 792
 793         if (test_opt(sbi, NOBARRIER))
 794                 return 0;
 795
 796         for (i = 1; i < sbi->s_ndevs; i++) {
 797                 int count = DEFAULT_RETRY_IO_COUNT;
 798
 799                 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
 800                         continue;
 801
 802                 do {
 803                         ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 804                         if (ret)
 805                                 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
 806                 } while (ret && --count);
 807
 808                 if (ret) {
 809                         f2fs_stop_checkpoint(sbi, false);
 810                         break;
 811                 }
 812
 813                 spin_lock(&sbi->dev_lock);
 814                 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
 815                 spin_unlock(&sbi->dev_lock);
 816         }
 817
 818         return ret;
 819 }
 820
 821 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 822                 enum dirty_type dirty_type)
 823 {
 824         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 825
 826         /* need not be added */
 827         if (IS_CURSEG(sbi, segno))
 828                 return;
 829
 830         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 831                 dirty_i->nr_dirty[dirty_type]++;
 832
 833         if (dirty_type == DIRTY) {
 834                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 835                 enum dirty_type t = sentry->type;
 836
 837                 if (unlikely(t >= DIRTY)) {
 838                         f2fs_bug_on(sbi, 1);
 839                         return;
 840                 }
 841                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 842                         dirty_i->nr_dirty[t]++;
 843
 844                 if (__is_large_section(sbi)) {
 845                         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 846                         block_t valid_blocks =
 847                                 get_valid_blocks(sbi, segno, true);
 848
 849                         f2fs_bug_on(sbi, unlikely(!valid_blocks ||
 850                                         valid_blocks == BLKS_PER_SEC(sbi)));
 851
 852                         if (!IS_CURSEC(sbi, secno))
 853                                 set_bit(secno, dirty_i->dirty_secmap);
 854                 }
 855         }
 856 }
 857
 858 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 859                 enum dirty_type dirty_type)
 860 {
 861         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 862         block_t valid_blocks;
 863
 864         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 865                 dirty_i->nr_dirty[dirty_type]--;
 866
 867         if (dirty_type == DIRTY) {
 868                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 869                 enum dirty_type t = sentry->type;
 870
 871                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 872                         dirty_i->nr_dirty[t]--;
 873
 874                 valid_blocks = get_valid_blocks(sbi, segno, true);
 875                 if (valid_blocks == 0) {
 876                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 877                                                 dirty_i->victim_secmap);
 878 #ifdef CONFIG_F2FS_CHECK_FS
 879                         clear_bit(segno, SIT_I(sbi)->invalid_segmap);
 880 #endif
 881                 }
 882                 if (__is_large_section(sbi)) {
 883                         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 884
 885                         if (!valid_blocks ||
 886                                         valid_blocks == BLKS_PER_SEC(sbi)) {
 887                                 clear_bit(secno, dirty_i->dirty_secmap);
 888                                 return;
 889                         }
 890
 891                         if (!IS_CURSEC(sbi, secno))
 892                                 set_bit(secno, dirty_i->dirty_secmap);
 893                 }
 894         }
 895 }
 896
 897 /*
 898  * Should not occur error such as -ENOMEM.
 899  * Adding dirty entry into seglist is not critical operation.
 900  * If a given segment is one of current working segments, it won't be added.
 901  */
 902 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 903 {
 904         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 905         unsigned short valid_blocks, ckpt_valid_blocks;
 906         unsigned int usable_blocks;
 907
 908         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 909                 return;
 910
 911         usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
 912         mutex_lock(&dirty_i->seglist_lock);
 913
 914         valid_blocks = get_valid_blocks(sbi, segno, false);
 915         ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false);
 916
 917         if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
 918                 ckpt_valid_blocks == usable_blocks)) {
 919                 __locate_dirty_segment(sbi, segno, PRE);
 920                 __remove_dirty_segment(sbi, segno, DIRTY);
 921         } else if (valid_blocks < usable_blocks) {
 922                 __locate_dirty_segment(sbi, segno, DIRTY);
 923         } else {
 924                 /* Recovery routine with SSR needs this */
 925                 __remove_dirty_segment(sbi, segno, DIRTY);
 926         }
 927
 928         mutex_unlock(&dirty_i->seglist_lock);
 929 }
 930
 931 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
 932 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
 933 {
 934         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 935         unsigned int segno;
 936
 937         mutex_lock(&dirty_i->seglist_lock);
 938         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 939                 if (get_valid_blocks(sbi, segno, false))
 940                         continue;
 941                 if (IS_CURSEG(sbi, segno))
 942                         continue;
 943                 __locate_dirty_segment(sbi, segno, PRE);
 944                 __remove_dirty_segment(sbi, segno, DIRTY);
 945         }
 946         mutex_unlock(&dirty_i->seglist_lock);
 947 }
 948
 949 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
 950 {
 951         int ovp_hole_segs =
 952                 (overprovision_segments(sbi) - reserved_segments(sbi));
 953         block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
 954         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 955         block_t holes[2] = {0, 0};      /* DATA and NODE */
 956         block_t unusable;
 957         struct seg_entry *se;
 958         unsigned int segno;
 959
 960         mutex_lock(&dirty_i->seglist_lock);
 961         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 962                 se = get_seg_entry(sbi, segno);
 963                 if (IS_NODESEG(se->type))
 964                         holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
 965                                                         se->valid_blocks;
 966                 else
 967                         holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
 968                                                         se->valid_blocks;
 969         }
 970         mutex_unlock(&dirty_i->seglist_lock);
 971
 972         unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
 973         if (unusable > ovp_holes)
 974                 return unusable - ovp_holes;
 975         return 0;
 976 }
 977
 978 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
 979 {
 980         int ovp_hole_segs =
 981                 (overprovision_segments(sbi) - reserved_segments(sbi));
 982         if (unusable > F2FS_OPTION(sbi).unusable_cap)
 983                 return -EAGAIN;
 984         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
 985                 dirty_segments(sbi) > ovp_hole_segs)
 986                 return -EAGAIN;
 987         return 0;
 988 }
 989
 990 /* This is only used by SBI_CP_DISABLED */
 991 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
 992 {
 993         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 994         unsigned int segno = 0;
 995
 996         mutex_lock(&dirty_i->seglist_lock);
 997         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 998                 if (get_valid_blocks(sbi, segno, false))
 999                         continue;
1000                 if (get_ckpt_valid_blocks(sbi, segno, false))
1001                         continue;
1002                 mutex_unlock(&dirty_i->seglist_lock);
1003                 return segno;
1004         }
1005         mutex_unlock(&dirty_i->seglist_lock);
1006         return NULL_SEGNO;
1007 }
1008
1009 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
1010                 struct block_device *bdev, block_t lstart,
1011                 block_t start, block_t len)
1012 {
1013         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1014         struct list_head *pend_list;
1015         struct discard_cmd *dc;
1016
1017         f2fs_bug_on(sbi, !len);
1018
1019         pend_list = &dcc->pend_list[plist_idx(len)];
1020
1021         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL);
1022         INIT_LIST_HEAD(&dc->list);
1023         dc->bdev = bdev;
1024         dc->lstart = lstart;
1025         dc->start = start;
1026         dc->len = len;
1027         dc->ref = 0;
1028         dc->state = D_PREP;
1029         dc->queued = 0;
1030         dc->error = 0;
1031         init_completion(&dc->wait);
1032         list_add_tail(&dc->list, pend_list);
1033         spin_lock_init(&dc->lock);
1034         dc->bio_ref = 0;
1035         atomic_inc(&dcc->discard_cmd_cnt);
1036         dcc->undiscard_blks += len;
1037
1038         return dc;
1039 }
1040
1041 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
1042                                 struct block_device *bdev, block_t lstart,
1043                                 block_t start, block_t len,
1044                                 struct rb_node *parent, struct rb_node **p,
1045                                 bool leftmost)
1046 {
1047         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1048         struct discard_cmd *dc;
1049
1050         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1051
1052         rb_link_node(&dc->rb_node, parent, p);
1053         rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
1054
1055         return dc;
1056 }
1057
1058 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1059                                                         struct discard_cmd *dc)
1060 {
1061         if (dc->state == D_DONE)
1062                 atomic_sub(dc->queued, &dcc->queued_discard);
1063
1064         list_del(&dc->list);
1065         rb_erase_cached(&dc->rb_node, &dcc->root);
1066         dcc->undiscard_blks -= dc->len;
1067
1068         kmem_cache_free(discard_cmd_slab, dc);
1069
1070         atomic_dec(&dcc->discard_cmd_cnt);
1071 }
1072
1073 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1074                                                         struct discard_cmd *dc)
1075 {
1076         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1077         unsigned long flags;
1078
1079         trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
1080
1081         spin_lock_irqsave(&dc->lock, flags);
1082         if (dc->bio_ref) {
1083                 spin_unlock_irqrestore(&dc->lock, flags);
1084                 return;
1085         }
1086         spin_unlock_irqrestore(&dc->lock, flags);
1087
1088         f2fs_bug_on(sbi, dc->ref);
1089
1090         if (dc->error == -EOPNOTSUPP)
1091                 dc->error = 0;
1092
1093         if (dc->error)
1094                 printk_ratelimited(
1095                         "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
1096                         KERN_INFO, sbi->sb->s_id,
1097                         dc->lstart, dc->start, dc->len, dc->error);
1098         __detach_discard_cmd(dcc, dc);
1099 }
1100
1101 static void f2fs_submit_discard_endio(struct bio *bio)
1102 {
1103         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1104         unsigned long flags;
1105
1106         spin_lock_irqsave(&dc->lock, flags);
1107         if (!dc->error)
1108                 dc->error = blk_status_to_errno(bio->bi_status);
1109         dc->bio_ref--;
1110         if (!dc->bio_ref && dc->state == D_SUBMIT) {
1111                 dc->state = D_DONE;
1112                 complete_all(&dc->wait);
1113         }
1114         spin_unlock_irqrestore(&dc->lock, flags);
1115         bio_put(bio);
1116 }
1117
1118 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1119                                 block_t start, block_t end)
1120 {
1121 #ifdef CONFIG_F2FS_CHECK_FS
1122         struct seg_entry *sentry;
1123         unsigned int segno;
1124         block_t blk = start;
1125         unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1126         unsigned long *map;
1127
1128         while (blk < end) {
1129                 segno = GET_SEGNO(sbi, blk);
1130                 sentry = get_seg_entry(sbi, segno);
1131                 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1132
1133                 if (end < START_BLOCK(sbi, segno + 1))
1134                         size = GET_BLKOFF_FROM_SEG0(sbi, end);
1135                 else
1136                         size = max_blocks;
1137                 map = (unsigned long *)(sentry->cur_valid_map);
1138                 offset = __find_rev_next_bit(map, size, offset);
1139                 f2fs_bug_on(sbi, offset != size);
1140                 blk = START_BLOCK(sbi, segno + 1);
1141         }
1142 #endif
1143 }
1144
1145 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1146                                 struct discard_policy *dpolicy,
1147                                 int discard_type, unsigned int granularity)
1148 {
1149         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1150
1151         /* common policy */
1152         dpolicy->type = discard_type;
1153         dpolicy->sync = true;
1154         dpolicy->ordered = false;
1155         dpolicy->granularity = granularity;
1156
1157         dpolicy->max_requests = dcc->max_discard_request;
1158         dpolicy->io_aware_gran = MAX_PLIST_NUM;
1159         dpolicy->timeout = false;
1160
1161         if (discard_type == DPOLICY_BG) {
1162                 dpolicy->min_interval = dcc->min_discard_issue_time;
1163                 dpolicy->mid_interval = dcc->mid_discard_issue_time;
1164                 dpolicy->max_interval = dcc->max_discard_issue_time;
1165                 dpolicy->io_aware = true;
1166                 dpolicy->sync = false;
1167                 dpolicy->ordered = true;
1168                 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1169                         dpolicy->granularity = 1;
1170                         if (atomic_read(&dcc->discard_cmd_cnt))
1171                                 dpolicy->max_interval =
1172                                         dcc->min_discard_issue_time;
1173                 }
1174         } else if (discard_type == DPOLICY_FORCE) {
1175                 dpolicy->min_interval = dcc->min_discard_issue_time;
1176                 dpolicy->mid_interval = dcc->mid_discard_issue_time;
1177                 dpolicy->max_interval = dcc->max_discard_issue_time;
1178                 dpolicy->io_aware = false;
1179         } else if (discard_type == DPOLICY_FSTRIM) {
1180                 dpolicy->io_aware = false;
1181         } else if (discard_type == DPOLICY_UMOUNT) {
1182                 dpolicy->io_aware = false;
1183                 /* we need to issue all to keep CP_TRIMMED_FLAG */
1184                 dpolicy->granularity = 1;
1185                 dpolicy->timeout = true;
1186         }
1187 }
1188
1189 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1190                                 struct block_device *bdev, block_t lstart,
1191                                 block_t start, block_t len);
1192 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1193 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1194                                                 struct discard_policy *dpolicy,
1195                                                 struct discard_cmd *dc,
1196                                                 unsigned int *issued)
1197 {
1198         struct block_device *bdev = dc->bdev;
1199         struct request_queue *q = bdev_get_queue(bdev);
1200         unsigned int max_discard_blocks =
1201                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1202         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1203         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1204                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1205         int flag = dpolicy->sync ? REQ_SYNC : 0;
1206         block_t lstart, start, len, total_len;
1207         int err = 0;
1208
1209         if (dc->state != D_PREP)
1210                 return 0;
1211
1212         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1213                 return 0;
1214
1215         trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1216
1217         lstart = dc->lstart;
1218         start = dc->start;
1219         len = dc->len;
1220         total_len = len;
1221
1222         dc->len = 0;
1223
1224         while (total_len && *issued < dpolicy->max_requests && !err) {
1225                 struct bio *bio = NULL;
1226                 unsigned long flags;
1227                 bool last = true;
1228
1229                 if (len > max_discard_blocks) {
1230                         len = max_discard_blocks;
1231                         last = false;
1232                 }
1233
1234                 (*issued)++;
1235                 if (*issued == dpolicy->max_requests)
1236                         last = true;
1237
1238                 dc->len += len;
1239
1240                 if (time_to_inject(sbi, FAULT_DISCARD)) {
1241                         f2fs_show_injection_info(sbi, FAULT_DISCARD);
1242                         err = -EIO;
1243                         goto submit;
1244                 }
1245                 err = __blkdev_issue_discard(bdev,
1246                                         SECTOR_FROM_BLOCK(start),
1247                                         SECTOR_FROM_BLOCK(len),
1248                                         GFP_NOFS, 0, &bio);
1249 submit:
1250                 if (err) {
1251                         spin_lock_irqsave(&dc->lock, flags);
1252                         if (dc->state == D_PARTIAL)
1253                                 dc->state = D_SUBMIT;
1254                         spin_unlock_irqrestore(&dc->lock, flags);
1255
1256                         break;
1257                 }
1258
1259                 f2fs_bug_on(sbi, !bio);
1260
1261                 /*
1262                  * should keep before submission to avoid D_DONE
1263                  * right away
1264                  */
1265                 spin_lock_irqsave(&dc->lock, flags);
1266                 if (last)
1267                         dc->state = D_SUBMIT;
1268                 else
1269                         dc->state = D_PARTIAL;
1270                 dc->bio_ref++;
1271                 spin_unlock_irqrestore(&dc->lock, flags);
1272
1273                 atomic_inc(&dcc->queued_discard);
1274                 dc->queued++;
1275                 list_move_tail(&dc->list, wait_list);
1276
1277                 /* sanity check on discard range */
1278                 __check_sit_bitmap(sbi, lstart, lstart + len);
1279
1280                 bio->bi_private = dc;
1281                 bio->bi_end_io = f2fs_submit_discard_endio;
1282                 bio->bi_opf |= flag;
1283                 submit_bio(bio);
1284
1285                 atomic_inc(&dcc->issued_discard);
1286
1287                 f2fs_update_iostat(sbi, FS_DISCARD, 1);
1288
1289                 lstart += len;
1290                 start += len;
1291                 total_len -= len;
1292                 len = total_len;
1293         }
1294
1295         if (!err && len) {
1296                 dcc->undiscard_blks -= len;
1297                 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1298         }
1299         return err;
1300 }
1301
1302 static void __insert_discard_tree(struct f2fs_sb_info *sbi,
1303                                 struct block_device *bdev, block_t lstart,
1304                                 block_t start, block_t len,
1305                                 struct rb_node **insert_p,
1306                                 struct rb_node *insert_parent)
1307 {
1308         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1309         struct rb_node **p;
1310         struct rb_node *parent = NULL;
1311         bool leftmost = true;
1312
1313         if (insert_p && insert_parent) {
1314                 parent = insert_parent;
1315                 p = insert_p;
1316                 goto do_insert;
1317         }
1318
1319         p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1320                                                         lstart, &leftmost);
1321 do_insert:
1322         __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1323                                                                 p, leftmost);
1324 }
1325
1326 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1327                                                 struct discard_cmd *dc)
1328 {
1329         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1330 }
1331
1332 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1333                                 struct discard_cmd *dc, block_t blkaddr)
1334 {
1335         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1336         struct discard_info di = dc->di;
1337         bool modified = false;
1338
1339         if (dc->state == D_DONE || dc->len == 1) {
1340                 __remove_discard_cmd(sbi, dc);
1341                 return;
1342         }
1343
1344         dcc->undiscard_blks -= di.len;
1345
1346         if (blkaddr > di.lstart) {
1347                 dc->len = blkaddr - dc->lstart;
1348                 dcc->undiscard_blks += dc->len;
1349                 __relocate_discard_cmd(dcc, dc);
1350                 modified = true;
1351         }
1352
1353         if (blkaddr < di.lstart + di.len - 1) {
1354                 if (modified) {
1355                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1356                                         di.start + blkaddr + 1 - di.lstart,
1357                                         di.lstart + di.len - 1 - blkaddr,
1358                                         NULL, NULL);
1359                 } else {
1360                         dc->lstart++;
1361                         dc->len--;
1362                         dc->start++;
1363                         dcc->undiscard_blks += dc->len;
1364                         __relocate_discard_cmd(dcc, dc);
1365                 }
1366         }
1367 }
1368
1369 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1370                                 struct block_device *bdev, block_t lstart,
1371                                 block_t start, block_t len)
1372 {
1373         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1374         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1375         struct discard_cmd *dc;
1376         struct discard_info di = {0};
1377         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1378         struct request_queue *q = bdev_get_queue(bdev);
1379         unsigned int max_discard_blocks =
1380                         SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1381         block_t end = lstart + len;
1382
1383         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1384                                         NULL, lstart,
1385                                         (struct rb_entry **)&prev_dc,
1386                                         (struct rb_entry **)&next_dc,
1387                                         &insert_p, &insert_parent, true, NULL);
1388         if (dc)
1389                 prev_dc = dc;
1390
1391         if (!prev_dc) {
1392                 di.lstart = lstart;
1393                 di.len = next_dc ? next_dc->lstart - lstart : len;
1394                 di.len = min(di.len, len);
1395                 di.start = start;
1396         }
1397
1398         while (1) {
1399                 struct rb_node *node;
1400                 bool merged = false;
1401                 struct discard_cmd *tdc = NULL;
1402
1403                 if (prev_dc) {
1404                         di.lstart = prev_dc->lstart + prev_dc->len;
1405                         if (di.lstart < lstart)
1406                                 di.lstart = lstart;
1407                         if (di.lstart >= end)
1408                                 break;
1409
1410                         if (!next_dc || next_dc->lstart > end)
1411                                 di.len = end - di.lstart;
1412                         else
1413                                 di.len = next_dc->lstart - di.lstart;
1414                         di.start = start + di.lstart - lstart;
1415                 }
1416
1417                 if (!di.len)
1418                         goto next;
1419
1420                 if (prev_dc && prev_dc->state == D_PREP &&
1421                         prev_dc->bdev == bdev &&
1422                         __is_discard_back_mergeable(&di, &prev_dc->di,
1423                                                         max_discard_blocks)) {
1424                         prev_dc->di.len += di.len;
1425                         dcc->undiscard_blks += di.len;
1426                         __relocate_discard_cmd(dcc, prev_dc);
1427                         di = prev_dc->di;
1428                         tdc = prev_dc;
1429                         merged = true;
1430                 }
1431
1432                 if (next_dc && next_dc->state == D_PREP &&
1433                         next_dc->bdev == bdev &&
1434                         __is_discard_front_mergeable(&di, &next_dc->di,
1435                                                         max_discard_blocks)) {
1436                         next_dc->di.lstart = di.lstart;
1437                         next_dc->di.len += di.len;
1438                         next_dc->di.start = di.start;
1439                         dcc->undiscard_blks += di.len;
1440                         __relocate_discard_cmd(dcc, next_dc);
1441                         if (tdc)
1442                                 __remove_discard_cmd(sbi, tdc);
1443                         merged = true;
1444                 }
1445
1446                 if (!merged) {
1447                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1448                                                         di.len, NULL, NULL);
1449                 }
1450  next:
1451                 prev_dc = next_dc;
1452                 if (!prev_dc)
1453                         break;
1454
1455                 node = rb_next(&prev_dc->rb_node);
1456                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1457         }
1458 }
1459
1460 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1461                 struct block_device *bdev, block_t blkstart, block_t blklen)
1462 {
1463         block_t lblkstart = blkstart;
1464
1465         if (!f2fs_bdev_support_discard(bdev))
1466                 return 0;
1467
1468         trace_f2fs_queue_discard(bdev, blkstart, blklen);
1469
1470         if (f2fs_is_multi_device(sbi)) {
1471                 int devi = f2fs_target_device_index(sbi, blkstart);
1472
1473                 blkstart -= FDEV(devi).start_blk;
1474         }
1475         mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1476         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1477         mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1478         return 0;
1479 }
1480
1481 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1482                                         struct discard_policy *dpolicy)
1483 {
1484         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1485         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1486         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1487         struct discard_cmd *dc;
1488         struct blk_plug plug;
1489         unsigned int pos = dcc->next_pos;
1490         unsigned int issued = 0;
1491         bool io_interrupted = false;
1492
1493         mutex_lock(&dcc->cmd_lock);
1494         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1495                                         NULL, pos,
1496                                         (struct rb_entry **)&prev_dc,
1497                                         (struct rb_entry **)&next_dc,
1498                                         &insert_p, &insert_parent, true, NULL);
1499         if (!dc)
1500                 dc = next_dc;
1501
1502         blk_start_plug(&plug);
1503
1504         while (dc) {
1505                 struct rb_node *node;
1506                 int err = 0;
1507
1508                 if (dc->state != D_PREP)
1509                         goto next;
1510
1511                 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1512                         io_interrupted = true;
1513                         break;
1514                 }
1515
1516                 dcc->next_pos = dc->lstart + dc->len;
1517                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1518
1519                 if (issued >= dpolicy->max_requests)
1520                         break;
1521 next:
1522                 node = rb_next(&dc->rb_node);
1523                 if (err)
1524                         __remove_discard_cmd(sbi, dc);
1525                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1526         }
1527
1528         blk_finish_plug(&plug);
1529
1530         if (!dc)
1531                 dcc->next_pos = 0;
1532
1533         mutex_unlock(&dcc->cmd_lock);
1534
1535         if (!issued && io_interrupted)
1536                 issued = -1;
1537
1538         return issued;
1539 }
1540 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1541                                         struct discard_policy *dpolicy);
1542
1543 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1544                                         struct discard_policy *dpolicy)
1545 {
1546         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1547         struct list_head *pend_list;
1548         struct discard_cmd *dc, *tmp;
1549         struct blk_plug plug;
1550         int i, issued;
1551         bool io_interrupted = false;
1552
1553         if (dpolicy->timeout)
1554                 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1555
1556 retry:
1557         issued = 0;
1558         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1559                 if (dpolicy->timeout &&
1560                                 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1561                         break;
1562
1563                 if (i + 1 < dpolicy->granularity)
1564                         break;
1565
1566                 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1567                         return __issue_discard_cmd_orderly(sbi, dpolicy);
1568
1569                 pend_list = &dcc->pend_list[i];
1570
1571                 mutex_lock(&dcc->cmd_lock);
1572                 if (list_empty(pend_list))
1573                         goto next;
1574                 if (unlikely(dcc->rbtree_check))
1575                         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1576                                                         &dcc->root, false));
1577                 blk_start_plug(&plug);
1578                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1579                         f2fs_bug_on(sbi, dc->state != D_PREP);
1580
1581                         if (dpolicy->timeout &&
1582                                 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1583                                 break;
1584
1585                         if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1586                                                 !is_idle(sbi, DISCARD_TIME)) {
1587                                 io_interrupted = true;
1588                                 break;
1589                         }
1590
1591                         __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1592
1593                         if (issued >= dpolicy->max_requests)
1594                                 break;
1595                 }
1596                 blk_finish_plug(&plug);
1597 next:
1598                 mutex_unlock(&dcc->cmd_lock);
1599
1600                 if (issued >= dpolicy->max_requests || io_interrupted)
1601                         break;
1602         }
1603
1604         if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1605                 __wait_all_discard_cmd(sbi, dpolicy);
1606                 goto retry;
1607         }
1608
1609         if (!issued && io_interrupted)
1610                 issued = -1;
1611
1612         return issued;
1613 }
1614
1615 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1616 {
1617         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1618         struct list_head *pend_list;
1619         struct discard_cmd *dc, *tmp;
1620         int i;
1621         bool dropped = false;
1622
1623         mutex_lock(&dcc->cmd_lock);
1624         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1625                 pend_list = &dcc->pend_list[i];
1626                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1627                         f2fs_bug_on(sbi, dc->state != D_PREP);
1628                         __remove_discard_cmd(sbi, dc);
1629                         dropped = true;
1630                 }
1631         }
1632         mutex_unlock(&dcc->cmd_lock);
1633
1634         return dropped;
1635 }
1636
1637 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1638 {
1639         __drop_discard_cmd(sbi);
1640 }
1641
1642 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1643                                                         struct discard_cmd *dc)
1644 {
1645         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1646         unsigned int len = 0;
1647
1648         wait_for_completion_io(&dc->wait);
1649         mutex_lock(&dcc->cmd_lock);
1650         f2fs_bug_on(sbi, dc->state != D_DONE);
1651         dc->ref--;
1652         if (!dc->ref) {
1653                 if (!dc->error)
1654                         len = dc->len;
1655                 __remove_discard_cmd(sbi, dc);
1656         }
1657         mutex_unlock(&dcc->cmd_lock);
1658
1659         return len;
1660 }
1661
1662 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1663                                                 struct discard_policy *dpolicy,
1664                                                 block_t start, block_t end)
1665 {
1666         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1667         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1668                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1669         struct discard_cmd *dc, *tmp;
1670         bool need_wait;
1671         unsigned int trimmed = 0;
1672
1673 next:
1674         need_wait = false;
1675
1676         mutex_lock(&dcc->cmd_lock);
1677         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1678                 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1679                         continue;
1680                 if (dc->len < dpolicy->granularity)
1681                         continue;
1682                 if (dc->state == D_DONE && !dc->ref) {
1683                         wait_for_completion_io(&dc->wait);
1684                         if (!dc->error)
1685                                 trimmed += dc->len;
1686                         __remove_discard_cmd(sbi, dc);
1687                 } else {
1688                         dc->ref++;
1689                         need_wait = true;
1690                         break;
1691                 }
1692         }
1693         mutex_unlock(&dcc->cmd_lock);
1694
1695         if (need_wait) {
1696                 trimmed += __wait_one_discard_bio(sbi, dc);
1697                 goto next;
1698         }
1699
1700         return trimmed;
1701 }
1702
1703 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1704                                                 struct discard_policy *dpolicy)
1705 {
1706         struct discard_policy dp;
1707         unsigned int discard_blks;
1708
1709         if (dpolicy)
1710                 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1711
1712         /* wait all */
1713         __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1714         discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1715         __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1716         discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1717
1718         return discard_blks;
1719 }
1720
1721 /* This should be covered by global mutex, &sit_i->sentry_lock */
1722 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1723 {
1724         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1725         struct discard_cmd *dc;
1726         bool need_wait = false;
1727
1728         mutex_lock(&dcc->cmd_lock);
1729         dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1730                                                         NULL, blkaddr);
1731         if (dc) {
1732                 if (dc->state == D_PREP) {
1733                         __punch_discard_cmd(sbi, dc, blkaddr);
1734                 } else {
1735                         dc->ref++;
1736                         need_wait = true;
1737                 }
1738         }
1739         mutex_unlock(&dcc->cmd_lock);
1740
1741         if (need_wait)
1742                 __wait_one_discard_bio(sbi, dc);
1743 }
1744
1745 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1746 {
1747         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1748
1749         if (dcc && dcc->f2fs_issue_discard) {
1750                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1751
1752                 dcc->f2fs_issue_discard = NULL;
1753                 kthread_stop(discard_thread);
1754         }
1755 }
1756
1757 /* This comes from f2fs_put_super */
1758 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1759 {
1760         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1761         struct discard_policy dpolicy;
1762         bool dropped;
1763
1764         __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1765                                         dcc->discard_granularity);
1766         __issue_discard_cmd(sbi, &dpolicy);
1767         dropped = __drop_discard_cmd(sbi);
1768
1769         /* just to make sure there is no pending discard commands */
1770         __wait_all_discard_cmd(sbi, NULL);
1771
1772         f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1773         return dropped;
1774 }
1775
1776 static int issue_discard_thread(void *data)
1777 {
1778         struct f2fs_sb_info *sbi = data;
1779         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1780         wait_queue_head_t *q = &dcc->discard_wait_queue;
1781         struct discard_policy dpolicy;
1782         unsigned int wait_ms = dcc->min_discard_issue_time;
1783         int issued;
1784
1785         set_freezable();
1786
1787         do {
1788                 if (sbi->gc_mode == GC_URGENT_HIGH ||
1789                         !f2fs_available_free_memory(sbi, DISCARD_CACHE))
1790                         __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1791                 else
1792                         __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1793                                                 dcc->discard_granularity);
1794
1795                 if (!atomic_read(&dcc->discard_cmd_cnt))
1796                        wait_ms = dpolicy.max_interval;
1797
1798                 wait_event_interruptible_timeout(*q,
1799                                 kthread_should_stop() || freezing(current) ||
1800                                 dcc->discard_wake,
1801                                 msecs_to_jiffies(wait_ms));
1802
1803                 if (dcc->discard_wake)
1804                         dcc->discard_wake = 0;
1805
1806                 /* clean up pending candidates before going to sleep */
1807                 if (atomic_read(&dcc->queued_discard))
1808                         __wait_all_discard_cmd(sbi, NULL);
1809
1810                 if (try_to_freeze())
1811                         continue;
1812                 if (f2fs_readonly(sbi->sb))
1813                         continue;
1814                 if (kthread_should_stop())
1815                         return 0;
1816                 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1817                         wait_ms = dpolicy.max_interval;
1818                         continue;
1819                 }
1820                 if (!atomic_read(&dcc->discard_cmd_cnt))
1821                         continue;
1822
1823                 sb_start_intwrite(sbi->sb);
1824
1825                 issued = __issue_discard_cmd(sbi, &dpolicy);
1826                 if (issued > 0) {
1827                         __wait_all_discard_cmd(sbi, &dpolicy);
1828                         wait_ms = dpolicy.min_interval;
1829                 } else if (issued == -1) {
1830                         wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1831                         if (!wait_ms)
1832                                 wait_ms = dpolicy.mid_interval;
1833                 } else {
1834                         wait_ms = dpolicy.max_interval;
1835                 }
1836
1837                 sb_end_intwrite(sbi->sb);
1838
1839         } while (!kthread_should_stop());
1840         return 0;
1841 }
1842
1843 #ifdef CONFIG_BLK_DEV_ZONED
1844 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1845                 struct block_device *bdev, block_t blkstart, block_t blklen)
1846 {
1847         sector_t sector, nr_sects;
1848         block_t lblkstart = blkstart;
1849         int devi = 0;
1850
1851         if (f2fs_is_multi_device(sbi)) {
1852                 devi = f2fs_target_device_index(sbi, blkstart);
1853                 if (blkstart < FDEV(devi).start_blk ||
1854                     blkstart > FDEV(devi).end_blk) {
1855                         f2fs_err(sbi, "Invalid block %x", blkstart);
1856                         return -EIO;
1857                 }
1858                 blkstart -= FDEV(devi).start_blk;
1859         }
1860
1861         /* For sequential zones, reset the zone write pointer */
1862         if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1863                 sector = SECTOR_FROM_BLOCK(blkstart);
1864                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1865
1866                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1867                                 nr_sects != bdev_zone_sectors(bdev)) {
1868                         f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1869                                  devi, sbi->s_ndevs ? FDEV(devi).path : "",
1870                                  blkstart, blklen);
1871                         return -EIO;
1872                 }
1873                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1874                 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1875                                         sector, nr_sects, GFP_NOFS);
1876         }
1877
1878         /* For conventional zones, use regular discard if supported */
1879         return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1880 }
1881 #endif
1882
1883 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1884                 struct block_device *bdev, block_t blkstart, block_t blklen)
1885 {
1886 #ifdef CONFIG_BLK_DEV_ZONED
1887         if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1888                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1889 #endif
1890         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1891 }
1892
1893 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1894                                 block_t blkstart, block_t blklen)
1895 {
1896         sector_t start = blkstart, len = 0;
1897         struct block_device *bdev;
1898         struct seg_entry *se;
1899         unsigned int offset;
1900         block_t i;
1901         int err = 0;
1902
1903         bdev = f2fs_target_device(sbi, blkstart, NULL);
1904
1905         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1906                 if (i != start) {
1907                         struct block_device *bdev2 =
1908                                 f2fs_target_device(sbi, i, NULL);
1909
1910                         if (bdev2 != bdev) {
1911                                 err = __issue_discard_async(sbi, bdev,
1912                                                 start, len);
1913                                 if (err)
1914                                         return err;
1915                                 bdev = bdev2;
1916                                 start = i;
1917                                 len = 0;
1918                         }
1919                 }
1920
1921                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1922                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1923
1924                 if (f2fs_block_unit_discard(sbi) &&
1925                                 !f2fs_test_and_set_bit(offset, se->discard_map))
1926                         sbi->discard_blks--;
1927         }
1928
1929         if (len)
1930                 err = __issue_discard_async(sbi, bdev, start, len);
1931         return err;
1932 }
1933
1934 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1935                                                         bool check_only)
1936 {
1937         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1938         int max_blocks = sbi->blocks_per_seg;
1939         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1940         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1941         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1942         unsigned long *discard_map = (unsigned long *)se->discard_map;
1943         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1944         unsigned int start = 0, end = -1;
1945         bool force = (cpc->reason & CP_DISCARD);
1946         struct discard_entry *de = NULL;
1947         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1948         int i;
1949
1950         if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi) ||
1951                         !f2fs_block_unit_discard(sbi))
1952                 return false;
1953
1954         if (!force) {
1955                 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1956                         SM_I(sbi)->dcc_info->nr_discards >=
1957                                 SM_I(sbi)->dcc_info->max_discards)
1958                         return false;
1959         }
1960
1961         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1962         for (i = 0; i < entries; i++)
1963                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1964                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1965
1966         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1967                                 SM_I(sbi)->dcc_info->max_discards) {
1968                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1969                 if (start >= max_blocks)
1970                         break;
1971
1972                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1973                 if (force && start && end != max_blocks
1974                                         && (end - start) < cpc->trim_minlen)
1975                         continue;
1976
1977                 if (check_only)
1978                         return true;
1979
1980                 if (!de) {
1981                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1982                                                 GFP_F2FS_ZERO, true, NULL);
1983                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1984                         list_add_tail(&de->list, head);
1985                 }
1986
1987                 for (i = start; i < end; i++)
1988                         __set_bit_le(i, (void *)de->discard_map);
1989
1990                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1991         }
1992         return false;
1993 }
1994
1995 static void release_discard_addr(struct discard_entry *entry)
1996 {
1997         list_del(&entry->list);
1998         kmem_cache_free(discard_entry_slab, entry);
1999 }
2000
2001 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
2002 {
2003         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
2004         struct discard_entry *entry, *this;
2005
2006         /* drop caches */
2007         list_for_each_entry_safe(entry, this, head, list)
2008                 release_discard_addr(entry);
2009 }
2010
2011 /*
2012  * Should call f2fs_clear_prefree_segments after checkpoint is done.
2013  */
2014 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
2015 {
2016         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2017         unsigned int segno;
2018
2019         mutex_lock(&dirty_i->seglist_lock);
2020         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
2021                 __set_test_and_free(sbi, segno, false);
2022         mutex_unlock(&dirty_i->seglist_lock);
2023 }
2024
2025 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
2026                                                 struct cp_control *cpc)
2027 {
2028         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2029         struct list_head *head = &dcc->entry_list;
2030         struct discard_entry *entry, *this;
2031         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2032         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
2033         unsigned int start = 0, end = -1;
2034         unsigned int secno, start_segno;
2035         bool force = (cpc->reason & CP_DISCARD);
2036         bool section_alignment = F2FS_OPTION(sbi).discard_unit ==
2037                                                 DISCARD_UNIT_SECTION;
2038
2039         if (f2fs_lfs_mode(sbi) && __is_large_section(sbi))
2040                 section_alignment = true;
2041
2042         mutex_lock(&dirty_i->seglist_lock);
2043
2044         while (1) {
2045                 int i;
2046
2047                 if (section_alignment && end != -1)
2048                         end--;
2049                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
2050                 if (start >= MAIN_SEGS(sbi))
2051                         break;
2052                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
2053                                                                 start + 1);
2054
2055                 if (section_alignment) {
2056                         start = rounddown(start, sbi->segs_per_sec);
2057                         end = roundup(end, sbi->segs_per_sec);
2058                 }
2059
2060                 for (i = start; i < end; i++) {
2061                         if (test_and_clear_bit(i, prefree_map))
2062                                 dirty_i->nr_dirty[PRE]--;
2063                 }
2064
2065                 if (!f2fs_realtime_discard_enable(sbi))
2066                         continue;
2067
2068                 if (force && start >= cpc->trim_start &&
2069                                         (end - 1) <= cpc->trim_end)
2070                                 continue;
2071
2072                 if (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi)) {
2073                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2074                                 (end - start) << sbi->log_blocks_per_seg);
2075                         continue;
2076                 }
2077 next:
2078                 secno = GET_SEC_FROM_SEG(sbi, start);
2079                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
2080                 if (!IS_CURSEC(sbi, secno) &&
2081                         !get_valid_blocks(sbi, start, true))
2082                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2083                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
2084
2085                 start = start_segno + sbi->segs_per_sec;
2086                 if (start < end)
2087                         goto next;
2088                 else
2089                         end = start - 1;
2090         }
2091         mutex_unlock(&dirty_i->seglist_lock);
2092
2093         if (!f2fs_block_unit_discard(sbi))
2094                 goto wakeup;
2095
2096         /* send small discards */
2097         list_for_each_entry_safe(entry, this, head, list) {
2098                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2099                 bool is_valid = test_bit_le(0, entry->discard_map);
2100
2101 find_next:
2102                 if (is_valid) {
2103                         next_pos = find_next_zero_bit_le(entry->discard_map,
2104                                         sbi->blocks_per_seg, cur_pos);
2105                         len = next_pos - cur_pos;
2106
2107                         if (f2fs_sb_has_blkzoned(sbi) ||
2108                             (force && len < cpc->trim_minlen))
2109                                 goto skip;
2110
2111                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2112                                                                         len);
2113                         total_len += len;
2114                 } else {
2115                         next_pos = find_next_bit_le(entry->discard_map,
2116                                         sbi->blocks_per_seg, cur_pos);
2117                 }
2118 skip:
2119                 cur_pos = next_pos;
2120                 is_valid = !is_valid;
2121
2122                 if (cur_pos < sbi->blocks_per_seg)
2123                         goto find_next;
2124
2125                 release_discard_addr(entry);
2126                 dcc->nr_discards -= total_len;
2127         }
2128
2129 wakeup:
2130         wake_up_discard_thread(sbi, false);
2131 }
2132
2133 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi)
2134 {
2135         dev_t dev = sbi->sb->s_bdev->bd_dev;
2136         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2137         int err = 0;
2138
2139         if (!f2fs_realtime_discard_enable(sbi))
2140                 return 0;
2141
2142         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2143                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2144         if (IS_ERR(dcc->f2fs_issue_discard))
2145                 err = PTR_ERR(dcc->f2fs_issue_discard);
2146
2147         return err;
2148 }
2149
2150 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2151 {
2152         struct discard_cmd_control *dcc;
2153         int err = 0, i;
2154
2155         if (SM_I(sbi)->dcc_info) {
2156                 dcc = SM_I(sbi)->dcc_info;
2157                 goto init_thread;
2158         }
2159
2160         dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2161         if (!dcc)
2162                 return -ENOMEM;
2163
2164         dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2165         if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT)
2166                 dcc->discard_granularity = sbi->blocks_per_seg;
2167         else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION)
2168                 dcc->discard_granularity = BLKS_PER_SEC(sbi);
2169
2170         INIT_LIST_HEAD(&dcc->entry_list);
2171         for (i = 0; i < MAX_PLIST_NUM; i++)
2172                 INIT_LIST_HEAD(&dcc->pend_list[i]);
2173         INIT_LIST_HEAD(&dcc->wait_list);
2174         INIT_LIST_HEAD(&dcc->fstrim_list);
2175         mutex_init(&dcc->cmd_lock);
2176         atomic_set(&dcc->issued_discard, 0);
2177         atomic_set(&dcc->queued_discard, 0);
2178         atomic_set(&dcc->discard_cmd_cnt, 0);
2179         dcc->nr_discards = 0;
2180         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2181         dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST;
2182         dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME;
2183         dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME;
2184         dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME;
2185         dcc->undiscard_blks = 0;
2186         dcc->next_pos = 0;
2187         dcc->root = RB_ROOT_CACHED;
2188         dcc->rbtree_check = false;
2189
2190         init_waitqueue_head(&dcc->discard_wait_queue);
2191         SM_I(sbi)->dcc_info = dcc;
2192 init_thread:
2193         err = f2fs_start_discard_thread(sbi);
2194         if (err) {
2195                 kfree(dcc);
2196                 SM_I(sbi)->dcc_info = NULL;
2197         }
2198
2199         return err;
2200 }
2201
2202 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2203 {
2204         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2205
2206         if (!dcc)
2207                 return;
2208
2209         f2fs_stop_discard_thread(sbi);
2210
2211         /*
2212          * Recovery can cache discard commands, so in error path of
2213          * fill_super(), it needs to give a chance to handle them.
2214          */
2215         if (unlikely(atomic_read(&dcc->discard_cmd_cnt)))
2216                 f2fs_issue_discard_timeout(sbi);
2217
2218         kfree(dcc);
2219         SM_I(sbi)->dcc_info = NULL;
2220 }
2221
2222 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2223 {
2224         struct sit_info *sit_i = SIT_I(sbi);
2225
2226         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2227                 sit_i->dirty_sentries++;
2228                 return false;
2229         }
2230
2231         return true;
2232 }
2233
2234 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2235                                         unsigned int segno, int modified)
2236 {
2237         struct seg_entry *se = get_seg_entry(sbi, segno);
2238
2239         se->type = type;
2240         if (modified)
2241                 __mark_sit_entry_dirty(sbi, segno);
2242 }
2243
2244 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2245                                                                 block_t blkaddr)
2246 {
2247         unsigned int segno = GET_SEGNO(sbi, blkaddr);
2248
2249         if (segno == NULL_SEGNO)
2250                 return 0;
2251         return get_seg_entry(sbi, segno)->mtime;
2252 }
2253
2254 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2255                                                 unsigned long long old_mtime)
2256 {
2257         struct seg_entry *se;
2258         unsigned int segno = GET_SEGNO(sbi, blkaddr);
2259         unsigned long long ctime = get_mtime(sbi, false);
2260         unsigned long long mtime = old_mtime ? old_mtime : ctime;
2261
2262         if (segno == NULL_SEGNO)
2263                 return;
2264
2265         se = get_seg_entry(sbi, segno);
2266
2267         if (!se->mtime)
2268                 se->mtime = mtime;
2269         else
2270                 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2271                                                 se->valid_blocks + 1);
2272
2273         if (ctime > SIT_I(sbi)->max_mtime)
2274                 SIT_I(sbi)->max_mtime = ctime;
2275 }
2276
2277 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2278 {
2279         struct seg_entry *se;
2280         unsigned int segno, offset;
2281         long int new_vblocks;
2282         bool exist;
2283 #ifdef CONFIG_F2FS_CHECK_FS
2284         bool mir_exist;
2285 #endif
2286
2287         segno = GET_SEGNO(sbi, blkaddr);
2288
2289         se = get_seg_entry(sbi, segno);
2290         new_vblocks = se->valid_blocks + del;
2291         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2292
2293         f2fs_bug_on(sbi, (new_vblocks < 0 ||
2294                         (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2295
2296         se->valid_blocks = new_vblocks;
2297
2298         /* Update valid block bitmap */
2299         if (del > 0) {
2300                 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2301 #ifdef CONFIG_F2FS_CHECK_FS
2302                 mir_exist = f2fs_test_and_set_bit(offset,
2303                                                 se->cur_valid_map_mir);
2304                 if (unlikely(exist != mir_exist)) {
2305                         f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2306                                  blkaddr, exist);
2307                         f2fs_bug_on(sbi, 1);
2308                 }
2309 #endif
2310                 if (unlikely(exist)) {
2311                         f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2312                                  blkaddr);
2313                         f2fs_bug_on(sbi, 1);
2314                         se->valid_blocks--;
2315                         del = 0;
2316                 }
2317
2318                 if (f2fs_block_unit_discard(sbi) &&
2319                                 !f2fs_test_and_set_bit(offset, se->discard_map))
2320                         sbi->discard_blks--;
2321
2322                 /*
2323                  * SSR should never reuse block which is checkpointed
2324                  * or newly invalidated.
2325                  */
2326                 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2327                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2328                                 se->ckpt_valid_blocks++;
2329                 }
2330         } else {
2331                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2332 #ifdef CONFIG_F2FS_CHECK_FS
2333                 mir_exist = f2fs_test_and_clear_bit(offset,
2334                                                 se->cur_valid_map_mir);
2335                 if (unlikely(exist != mir_exist)) {
2336                         f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2337                                  blkaddr, exist);
2338                         f2fs_bug_on(sbi, 1);
2339                 }
2340 #endif
2341                 if (unlikely(!exist)) {
2342                         f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2343                                  blkaddr);
2344                         f2fs_bug_on(sbi, 1);
2345                         se->valid_blocks++;
2346                         del = 0;
2347                 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2348                         /*
2349                          * If checkpoints are off, we must not reuse data that
2350                          * was used in the previous checkpoint. If it was used
2351                          * before, we must track that to know how much space we
2352                          * really have.
2353                          */
2354                         if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2355                                 spin_lock(&sbi->stat_lock);
2356                                 sbi->unusable_block_count++;
2357                                 spin_unlock(&sbi->stat_lock);
2358                         }
2359                 }
2360
2361                 if (f2fs_block_unit_discard(sbi) &&
2362                         f2fs_test_and_clear_bit(offset, se->discard_map))
2363                         sbi->discard_blks++;
2364         }
2365         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2366                 se->ckpt_valid_blocks += del;
2367
2368         __mark_sit_entry_dirty(sbi, segno);
2369
2370         /* update total number of valid blocks to be written in ckpt area */
2371         SIT_I(sbi)->written_valid_blocks += del;
2372
2373         if (__is_large_section(sbi))
2374                 get_sec_entry(sbi, segno)->valid_blocks += del;
2375 }
2376
2377 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2378 {
2379         unsigned int segno = GET_SEGNO(sbi, addr);
2380         struct sit_info *sit_i = SIT_I(sbi);
2381
2382         f2fs_bug_on(sbi, addr == NULL_ADDR);
2383         if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2384                 return;
2385
2386         invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2387         f2fs_invalidate_compress_page(sbi, addr);
2388
2389         /* add it into sit main buffer */
2390         down_write(&sit_i->sentry_lock);
2391
2392         update_segment_mtime(sbi, addr, 0);
2393         update_sit_entry(sbi, addr, -1);
2394
2395         /* add it into dirty seglist */
2396         locate_dirty_segment(sbi, segno);
2397
2398         up_write(&sit_i->sentry_lock);
2399 }
2400
2401 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2402 {
2403         struct sit_info *sit_i = SIT_I(sbi);
2404         unsigned int segno, offset;
2405         struct seg_entry *se;
2406         bool is_cp = false;
2407
2408         if (!__is_valid_data_blkaddr(blkaddr))
2409                 return true;
2410
2411         down_read(&sit_i->sentry_lock);
2412
2413         segno = GET_SEGNO(sbi, blkaddr);
2414         se = get_seg_entry(sbi, segno);
2415         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2416
2417         if (f2fs_test_bit(offset, se->ckpt_valid_map))
2418                 is_cp = true;
2419
2420         up_read(&sit_i->sentry_lock);
2421
2422         return is_cp;
2423 }
2424
2425 /*
2426  * This function should be resided under the curseg_mutex lock
2427  */
2428 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2429                                         struct f2fs_summary *sum)
2430 {
2431         struct curseg_info *curseg = CURSEG_I(sbi, type);
2432         void *addr = curseg->sum_blk;
2433
2434         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2435         memcpy(addr, sum, sizeof(struct f2fs_summary));
2436 }
2437
2438 /*
2439  * Calculate the number of current summary pages for writing
2440  */
2441 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2442 {
2443         int valid_sum_count = 0;
2444         int i, sum_in_page;
2445
2446         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2447                 if (sbi->ckpt->alloc_type[i] == SSR)
2448                         valid_sum_count += sbi->blocks_per_seg;
2449                 else {
2450                         if (for_ra)
2451                                 valid_sum_count += le16_to_cpu(
2452                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2453                         else
2454                                 valid_sum_count += curseg_blkoff(sbi, i);
2455                 }
2456         }
2457
2458         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2459                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2460         if (valid_sum_count <= sum_in_page)
2461                 return 1;
2462         else if ((valid_sum_count - sum_in_page) <=
2463                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2464                 return 2;
2465         return 3;
2466 }
2467
2468 /*
2469  * Caller should put this summary page
2470  */
2471 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2472 {
2473         if (unlikely(f2fs_cp_error(sbi)))
2474                 return ERR_PTR(-EIO);
2475         return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno));
2476 }
2477
2478 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2479                                         void *src, block_t blk_addr)
2480 {
2481         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2482
2483         memcpy(page_address(page), src, PAGE_SIZE);
2484         set_page_dirty(page);
2485         f2fs_put_page(page, 1);
2486 }
2487
2488 static void write_sum_page(struct f2fs_sb_info *sbi,
2489                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
2490 {
2491         f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2492 }
2493
2494 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2495                                                 int type, block_t blk_addr)
2496 {
2497         struct curseg_info *curseg = CURSEG_I(sbi, type);
2498         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2499         struct f2fs_summary_block *src = curseg->sum_blk;
2500         struct f2fs_summary_block *dst;
2501
2502         dst = (struct f2fs_summary_block *)page_address(page);
2503         memset(dst, 0, PAGE_SIZE);
2504
2505         mutex_lock(&curseg->curseg_mutex);
2506
2507         down_read(&curseg->journal_rwsem);
2508         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2509         up_read(&curseg->journal_rwsem);
2510
2511         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2512         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2513
2514         mutex_unlock(&curseg->curseg_mutex);
2515
2516         set_page_dirty(page);
2517         f2fs_put_page(page, 1);
2518 }
2519
2520 static int is_next_segment_free(struct f2fs_sb_info *sbi,
2521                                 struct curseg_info *curseg, int type)
2522 {
2523         unsigned int segno = curseg->segno + 1;
2524         struct free_segmap_info *free_i = FREE_I(sbi);
2525
2526         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2527                 return !test_bit(segno, free_i->free_segmap);
2528         return 0;
2529 }
2530
2531 /*
2532  * Find a new segment from the free segments bitmap to right order
2533  * This function should be returned with success, otherwise BUG
2534  */
2535 static void get_new_segment(struct f2fs_sb_info *sbi,
2536                         unsigned int *newseg, bool new_sec, int dir)
2537 {
2538         struct free_segmap_info *free_i = FREE_I(sbi);
2539         unsigned int segno, secno, zoneno;
2540         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2541         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2542         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2543         unsigned int left_start = hint;
2544         bool init = true;
2545         int go_left = 0;
2546         int i;
2547
2548         spin_lock(&free_i->segmap_lock);
2549
2550         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2551                 segno = find_next_zero_bit(free_i->free_segmap,
2552                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2553                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2554                         goto got_it;
2555         }
2556 find_other_zone:
2557         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2558         if (secno >= MAIN_SECS(sbi)) {
2559                 if (dir == ALLOC_RIGHT) {
2560                         secno = find_first_zero_bit(free_i->free_secmap,
2561                                                         MAIN_SECS(sbi));
2562                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2563                 } else {
2564                         go_left = 1;
2565                         left_start = hint - 1;
2566                 }
2567         }
2568         if (go_left == 0)
2569                 goto skip_left;
2570
2571         while (test_bit(left_start, free_i->free_secmap)) {
2572                 if (left_start > 0) {
2573                         left_start--;
2574                         continue;
2575                 }
2576                 left_start = find_first_zero_bit(free_i->free_secmap,
2577                                                         MAIN_SECS(sbi));
2578                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2579                 break;
2580         }
2581         secno = left_start;
2582 skip_left:
2583         segno = GET_SEG_FROM_SEC(sbi, secno);
2584         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2585
2586         /* give up on finding another zone */
2587         if (!init)
2588                 goto got_it;
2589         if (sbi->secs_per_zone == 1)
2590                 goto got_it;
2591         if (zoneno == old_zoneno)
2592                 goto got_it;
2593         if (dir == ALLOC_LEFT) {
2594                 if (!go_left && zoneno + 1 >= total_zones)
2595                         goto got_it;
2596                 if (go_left && zoneno == 0)
2597                         goto got_it;
2598         }
2599         for (i = 0; i < NR_CURSEG_TYPE; i++)
2600                 if (CURSEG_I(sbi, i)->zone == zoneno)
2601                         break;
2602
2603         if (i < NR_CURSEG_TYPE) {
2604                 /* zone is in user, try another */
2605                 if (go_left)
2606                         hint = zoneno * sbi->secs_per_zone - 1;
2607                 else if (zoneno + 1 >= total_zones)
2608                         hint = 0;
2609                 else
2610                         hint = (zoneno + 1) * sbi->secs_per_zone;
2611                 init = false;
2612                 goto find_other_zone;
2613         }
2614 got_it:
2615         /* set it as dirty segment in free segmap */
2616         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2617         __set_inuse(sbi, segno);
2618         *newseg = segno;
2619         spin_unlock(&free_i->segmap_lock);
2620 }
2621
2622 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2623 {
2624         struct curseg_info *curseg = CURSEG_I(sbi, type);
2625         struct summary_footer *sum_footer;
2626         unsigned short seg_type = curseg->seg_type;
2627
2628         curseg->inited = true;
2629         curseg->segno = curseg->next_segno;
2630         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2631         curseg->next_blkoff = 0;
2632         curseg->next_segno = NULL_SEGNO;
2633
2634         sum_footer = &(curseg->sum_blk->footer);
2635         memset(sum_footer, 0, sizeof(struct summary_footer));
2636
2637         sanity_check_seg_type(sbi, seg_type);
2638
2639         if (IS_DATASEG(seg_type))
2640                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2641         if (IS_NODESEG(seg_type))
2642                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2643         __set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
2644 }
2645
2646 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2647 {
2648         struct curseg_info *curseg = CURSEG_I(sbi, type);
2649         unsigned short seg_type = curseg->seg_type;
2650
2651         sanity_check_seg_type(sbi, seg_type);
2652         if (f2fs_need_rand_seg(sbi))
2653                 return prandom_u32() % (MAIN_SECS(sbi) * sbi->segs_per_sec);
2654
2655         /* if segs_per_sec is large than 1, we need to keep original policy. */
2656         if (__is_large_section(sbi))
2657                 return curseg->segno;
2658
2659         /* inmem log may not locate on any segment after mount */
2660         if (!curseg->inited)
2661                 return 0;
2662
2663         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2664                 return 0;
2665
2666         if (test_opt(sbi, NOHEAP) &&
2667                 (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type)))
2668                 return 0;
2669
2670         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2671                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2672
2673         /* find segments from 0 to reuse freed segments */
2674         if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2675                 return 0;
2676
2677         return curseg->segno;
2678 }
2679
2680 /*
2681  * Allocate a current working segment.
2682  * This function always allocates a free segment in LFS manner.
2683  */
2684 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2685 {
2686         struct curseg_info *curseg = CURSEG_I(sbi, type);
2687         unsigned short seg_type = curseg->seg_type;
2688         unsigned int segno = curseg->segno;
2689         int dir = ALLOC_LEFT;
2690
2691         if (curseg->inited)
2692                 write_sum_page(sbi, curseg->sum_blk,
2693                                 GET_SUM_BLOCK(sbi, segno));
2694         if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA)
2695                 dir = ALLOC_RIGHT;
2696
2697         if (test_opt(sbi, NOHEAP))
2698                 dir = ALLOC_RIGHT;
2699
2700         segno = __get_next_segno(sbi, type);
2701         get_new_segment(sbi, &segno, new_sec, dir);
2702         curseg->next_segno = segno;
2703         reset_curseg(sbi, type, 1);
2704         curseg->alloc_type = LFS;
2705         if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
2706                 curseg->fragment_remained_chunk =
2707                                 prandom_u32() % sbi->max_fragment_chunk + 1;
2708 }
2709
2710 static int __next_free_blkoff(struct f2fs_sb_info *sbi,
2711                                         int segno, block_t start)
2712 {
2713         struct seg_entry *se = get_seg_entry(sbi, segno);
2714         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2715         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2716         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2717         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2718         int i;
2719
2720         for (i = 0; i < entries; i++)
2721                 target_map[i] = ckpt_map[i] | cur_map[i];
2722
2723         return __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2724 }
2725
2726 /*
2727  * If a segment is written by LFS manner, next block offset is just obtained
2728  * by increasing the current block offset. However, if a segment is written by
2729  * SSR manner, next block offset obtained by calling __next_free_blkoff
2730  */
2731 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2732                                 struct curseg_info *seg)
2733 {
2734         if (seg->alloc_type == SSR) {
2735                 seg->next_blkoff =
2736                         __next_free_blkoff(sbi, seg->segno,
2737                                                 seg->next_blkoff + 1);
2738         } else {
2739                 seg->next_blkoff++;
2740                 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) {
2741                         /* To allocate block chunks in different sizes, use random number */
2742                         if (--seg->fragment_remained_chunk <= 0) {
2743                                 seg->fragment_remained_chunk =
2744                                    prandom_u32() % sbi->max_fragment_chunk + 1;
2745                                 seg->next_blkoff +=
2746                                    prandom_u32() % sbi->max_fragment_hole + 1;
2747                         }
2748                 }
2749         }
2750 }
2751
2752 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno)
2753 {
2754         return __next_free_blkoff(sbi, segno, 0) < sbi->blocks_per_seg;
2755 }
2756
2757 /*
2758  * This function always allocates a used segment(from dirty seglist) by SSR
2759  * manner, so it should recover the existing segment information of valid blocks
2760  */
2761 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool flush)
2762 {
2763         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2764         struct curseg_info *curseg = CURSEG_I(sbi, type);
2765         unsigned int new_segno = curseg->next_segno;
2766         struct f2fs_summary_block *sum_node;
2767         struct page *sum_page;
2768
2769         if (flush)
2770                 write_sum_page(sbi, curseg->sum_blk,
2771                                         GET_SUM_BLOCK(sbi, curseg->segno));
2772
2773         __set_test_and_inuse(sbi, new_segno);
2774
2775         mutex_lock(&dirty_i->seglist_lock);
2776         __remove_dirty_segment(sbi, new_segno, PRE);
2777         __remove_dirty_segment(sbi, new_segno, DIRTY);
2778         mutex_unlock(&dirty_i->seglist_lock);
2779
2780         reset_curseg(sbi, type, 1);
2781         curseg->alloc_type = SSR;
2782         curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0);
2783
2784         sum_page = f2fs_get_sum_page(sbi, new_segno);
2785         if (IS_ERR(sum_page)) {
2786                 /* GC won't be able to use stale summary pages by cp_error */
2787                 memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
2788                 return;
2789         }
2790         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2791         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2792         f2fs_put_page(sum_page, 1);
2793 }
2794
2795 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2796                                 int alloc_mode, unsigned long long age);
2797
2798 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type,
2799                                         int target_type, int alloc_mode,
2800                                         unsigned long long age)
2801 {
2802         struct curseg_info *curseg = CURSEG_I(sbi, type);
2803
2804         curseg->seg_type = target_type;
2805
2806         if (get_ssr_segment(sbi, type, alloc_mode, age)) {
2807                 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
2808
2809                 curseg->seg_type = se->type;
2810                 change_curseg(sbi, type, true);
2811         } else {
2812                 /* allocate cold segment by default */
2813                 curseg->seg_type = CURSEG_COLD_DATA;
2814                 new_curseg(sbi, type, true);
2815         }
2816         stat_inc_seg_type(sbi, curseg);
2817 }
2818
2819 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
2820 {
2821         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
2822
2823         if (!sbi->am.atgc_enabled)
2824                 return;
2825
2826         f2fs_down_read(&SM_I(sbi)->curseg_lock);
2827
2828         mutex_lock(&curseg->curseg_mutex);
2829         down_write(&SIT_I(sbi)->sentry_lock);
2830
2831         get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0);
2832
2833         up_write(&SIT_I(sbi)->sentry_lock);
2834         mutex_unlock(&curseg->curseg_mutex);
2835
2836         f2fs_up_read(&SM_I(sbi)->curseg_lock);
2837
2838 }
2839 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
2840 {
2841         __f2fs_init_atgc_curseg(sbi);
2842 }
2843
2844 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2845 {
2846         struct curseg_info *curseg = CURSEG_I(sbi, type);
2847
2848         mutex_lock(&curseg->curseg_mutex);
2849         if (!curseg->inited)
2850                 goto out;
2851
2852         if (get_valid_blocks(sbi, curseg->segno, false)) {
2853                 write_sum_page(sbi, curseg->sum_blk,
2854                                 GET_SUM_BLOCK(sbi, curseg->segno));
2855         } else {
2856                 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2857                 __set_test_and_free(sbi, curseg->segno, true);
2858                 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2859         }
2860 out:
2861         mutex_unlock(&curseg->curseg_mutex);
2862 }
2863
2864 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
2865 {
2866         __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2867
2868         if (sbi->am.atgc_enabled)
2869                 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2870 }
2871
2872 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2873 {
2874         struct curseg_info *curseg = CURSEG_I(sbi, type);
2875
2876         mutex_lock(&curseg->curseg_mutex);
2877         if (!curseg->inited)
2878                 goto out;
2879         if (get_valid_blocks(sbi, curseg->segno, false))
2880                 goto out;
2881
2882         mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2883         __set_test_and_inuse(sbi, curseg->segno);
2884         mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2885 out:
2886         mutex_unlock(&curseg->curseg_mutex);
2887 }
2888
2889 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
2890 {
2891         __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2892
2893         if (sbi->am.atgc_enabled)
2894                 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2895 }
2896
2897 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2898                                 int alloc_mode, unsigned long long age)
2899 {
2900         struct curseg_info *curseg = CURSEG_I(sbi, type);
2901         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2902         unsigned segno = NULL_SEGNO;
2903         unsigned short seg_type = curseg->seg_type;
2904         int i, cnt;
2905         bool reversed = false;
2906
2907         sanity_check_seg_type(sbi, seg_type);
2908
2909         /* f2fs_need_SSR() already forces to do this */
2910         if (!v_ops->get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) {
2911                 curseg->next_segno = segno;
2912                 return 1;
2913         }
2914
2915         /* For node segments, let's do SSR more intensively */
2916         if (IS_NODESEG(seg_type)) {
2917                 if (seg_type >= CURSEG_WARM_NODE) {
2918                         reversed = true;
2919                         i = CURSEG_COLD_NODE;
2920                 } else {
2921                         i = CURSEG_HOT_NODE;
2922                 }
2923                 cnt = NR_CURSEG_NODE_TYPE;
2924         } else {
2925                 if (seg_type >= CURSEG_WARM_DATA) {
2926                         reversed = true;
2927                         i = CURSEG_COLD_DATA;
2928                 } else {
2929                         i = CURSEG_HOT_DATA;
2930                 }
2931                 cnt = NR_CURSEG_DATA_TYPE;
2932         }
2933
2934         for (; cnt-- > 0; reversed ? i-- : i++) {
2935                 if (i == seg_type)
2936                         continue;
2937                 if (!v_ops->get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) {
2938                         curseg->next_segno = segno;
2939                         return 1;
2940                 }
2941         }
2942
2943         /* find valid_blocks=0 in dirty list */
2944         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2945                 segno = get_free_segment(sbi);
2946                 if (segno != NULL_SEGNO) {
2947                         curseg->next_segno = segno;
2948                         return 1;
2949                 }
2950         }
2951         return 0;
2952 }
2953
2954 /*
2955  * flush out current segment and replace it with new segment
2956  * This function should be returned with success, otherwise BUG
2957  */
2958 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2959                                                 int type, bool force)
2960 {
2961         struct curseg_info *curseg = CURSEG_I(sbi, type);
2962
2963         if (force)
2964                 new_curseg(sbi, type, true);
2965         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2966                                         curseg->seg_type == CURSEG_WARM_NODE)
2967                 new_curseg(sbi, type, false);
2968         else if (curseg->alloc_type == LFS &&
2969                         is_next_segment_free(sbi, curseg, type) &&
2970                         likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2971                 new_curseg(sbi, type, false);
2972         else if (f2fs_need_SSR(sbi) &&
2973                         get_ssr_segment(sbi, type, SSR, 0))
2974                 change_curseg(sbi, type, true);
2975         else
2976                 new_curseg(sbi, type, false);
2977
2978         stat_inc_seg_type(sbi, curseg);
2979 }
2980
2981 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2982                                         unsigned int start, unsigned int end)
2983 {
2984         struct curseg_info *curseg = CURSEG_I(sbi, type);
2985         unsigned int segno;
2986
2987         f2fs_down_read(&SM_I(sbi)->curseg_lock);
2988         mutex_lock(&curseg->curseg_mutex);
2989         down_write(&SIT_I(sbi)->sentry_lock);
2990
2991         segno = CURSEG_I(sbi, type)->segno;
2992         if (segno < start || segno > end)
2993                 goto unlock;
2994
2995         if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
2996                 change_curseg(sbi, type, true);
2997         else
2998                 new_curseg(sbi, type, true);
2999
3000         stat_inc_seg_type(sbi, curseg);
3001
3002         locate_dirty_segment(sbi, segno);
3003 unlock:
3004         up_write(&SIT_I(sbi)->sentry_lock);
3005
3006         if (segno != curseg->segno)
3007                 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
3008                             type, segno, curseg->segno);
3009
3010         mutex_unlock(&curseg->curseg_mutex);
3011         f2fs_up_read(&SM_I(sbi)->curseg_lock);
3012 }
3013
3014 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type,
3015                                                 bool new_sec, bool force)
3016 {
3017         struct curseg_info *curseg = CURSEG_I(sbi, type);
3018         unsigned int old_segno;
3019
3020         if (!curseg->inited)
3021                 goto alloc;
3022
3023         if (force || curseg->next_blkoff ||
3024                 get_valid_blocks(sbi, curseg->segno, new_sec))
3025                 goto alloc;
3026
3027         if (!get_ckpt_valid_blocks(sbi, curseg->segno, new_sec))
3028                 return;
3029 alloc:
3030         old_segno = curseg->segno;
3031         SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
3032         locate_dirty_segment(sbi, old_segno);
3033 }
3034
3035 static void __allocate_new_section(struct f2fs_sb_info *sbi,
3036                                                 int type, bool force)
3037 {
3038         __allocate_new_segment(sbi, type, true, force);
3039 }
3040
3041 void f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force)
3042 {
3043         f2fs_down_read(&SM_I(sbi)->curseg_lock);
3044         down_write(&SIT_I(sbi)->sentry_lock);
3045         __allocate_new_section(sbi, type, force);
3046         up_write(&SIT_I(sbi)->sentry_lock);
3047         f2fs_up_read(&SM_I(sbi)->curseg_lock);
3048 }
3049
3050 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
3051 {
3052         int i;
3053
3054         f2fs_down_read(&SM_I(sbi)->curseg_lock);
3055         down_write(&SIT_I(sbi)->sentry_lock);
3056         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
3057                 __allocate_new_segment(sbi, i, false, false);
3058         up_write(&SIT_I(sbi)->sentry_lock);
3059         f2fs_up_read(&SM_I(sbi)->curseg_lock);
3060 }
3061
3062 static const struct segment_allocation default_salloc_ops = {
3063         .allocate_segment = allocate_segment_by_default,
3064 };
3065
3066 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
3067                                                 struct cp_control *cpc)
3068 {
3069         __u64 trim_start = cpc->trim_start;
3070         bool has_candidate = false;
3071
3072         down_write(&SIT_I(sbi)->sentry_lock);
3073         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
3074                 if (add_discard_addrs(sbi, cpc, true)) {
3075                         has_candidate = true;
3076                         break;
3077                 }
3078         }
3079         up_write(&SIT_I(sbi)->sentry_lock);
3080
3081         cpc->trim_start = trim_start;
3082         return has_candidate;
3083 }
3084
3085 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
3086                                         struct discard_policy *dpolicy,
3087                                         unsigned int start, unsigned int end)
3088 {
3089         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
3090         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
3091         struct rb_node **insert_p = NULL, *insert_parent = NULL;
3092         struct discard_cmd *dc;
3093         struct blk_plug plug;
3094         int issued;
3095         unsigned int trimmed = 0;
3096
3097 next:
3098         issued = 0;
3099
3100         mutex_lock(&dcc->cmd_lock);
3101         if (unlikely(dcc->rbtree_check))
3102                 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
3103                                                         &dcc->root, false));
3104
3105         dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
3106                                         NULL, start,
3107                                         (struct rb_entry **)&prev_dc,
3108                                         (struct rb_entry **)&next_dc,
3109                                         &insert_p, &insert_parent, true, NULL);
3110         if (!dc)
3111                 dc = next_dc;
3112
3113         blk_start_plug(&plug);
3114
3115         while (dc && dc->lstart <= end) {
3116                 struct rb_node *node;
3117                 int err = 0;
3118
3119                 if (dc->len < dpolicy->granularity)
3120                         goto skip;
3121
3122                 if (dc->state != D_PREP) {
3123                         list_move_tail(&dc->list, &dcc->fstrim_list);
3124                         goto skip;
3125                 }
3126
3127                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
3128
3129                 if (issued >= dpolicy->max_requests) {
3130                         start = dc->lstart + dc->len;
3131
3132                         if (err)
3133                                 __remove_discard_cmd(sbi, dc);
3134
3135                         blk_finish_plug(&plug);
3136                         mutex_unlock(&dcc->cmd_lock);
3137                         trimmed += __wait_all_discard_cmd(sbi, NULL);
3138                         f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
3139                         goto next;
3140                 }
3141 skip:
3142                 node = rb_next(&dc->rb_node);
3143                 if (err)
3144                         __remove_discard_cmd(sbi, dc);
3145                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3146
3147                 if (fatal_signal_pending(current))
3148                         break;
3149         }
3150
3151         blk_finish_plug(&plug);
3152         mutex_unlock(&dcc->cmd_lock);
3153
3154         return trimmed;
3155 }
3156
3157 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3158 {
3159         __u64 start = F2FS_BYTES_TO_BLK(range->start);
3160         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
3161         unsigned int start_segno, end_segno;
3162         block_t start_block, end_block;
3163         struct cp_control cpc;
3164         struct discard_policy dpolicy;
3165         unsigned long long trimmed = 0;
3166         int err = 0;
3167         bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
3168
3169         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
3170                 return -EINVAL;
3171
3172         if (end < MAIN_BLKADDR(sbi))
3173                 goto out;
3174
3175         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
3176                 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
3177                 return -EFSCORRUPTED;
3178         }
3179
3180         /* start/end segment number in main_area */
3181         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3182         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3183                                                 GET_SEGNO(sbi, end);
3184         if (need_align) {
3185                 start_segno = rounddown(start_segno, sbi->segs_per_sec);
3186                 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
3187         }
3188
3189         cpc.reason = CP_DISCARD;
3190         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
3191         cpc.trim_start = start_segno;
3192         cpc.trim_end = end_segno;
3193
3194         if (sbi->discard_blks == 0)
3195                 goto out;
3196
3197         f2fs_down_write(&sbi->gc_lock);
3198         err = f2fs_write_checkpoint(sbi, &cpc);
3199         f2fs_up_write(&sbi->gc_lock);
3200         if (err)
3201                 goto out;
3202
3203         /*
3204          * We filed discard candidates, but actually we don't need to wait for
3205          * all of them, since they'll be issued in idle time along with runtime
3206          * discard option. User configuration looks like using runtime discard
3207          * or periodic fstrim instead of it.
3208          */
3209         if (f2fs_realtime_discard_enable(sbi))
3210                 goto out;
3211
3212         start_block = START_BLOCK(sbi, start_segno);
3213         end_block = START_BLOCK(sbi, end_segno + 1);
3214
3215         __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3216         trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3217                                         start_block, end_block);
3218
3219         trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3220                                         start_block, end_block);
3221 out:
3222         if (!err)
3223                 range->len = F2FS_BLK_TO_BYTES(trimmed);
3224         return err;
3225 }
3226
3227 static bool __has_curseg_space(struct f2fs_sb_info *sbi,
3228                                         struct curseg_info *curseg)
3229 {
3230         return curseg->next_blkoff < f2fs_usable_blks_in_seg(sbi,
3231                                                         curseg->segno);
3232 }
3233
3234 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
3235 {
3236         switch (hint) {
3237         case WRITE_LIFE_SHORT:
3238                 return CURSEG_HOT_DATA;
3239         case WRITE_LIFE_EXTREME:
3240                 return CURSEG_COLD_DATA;
3241         default:
3242                 return CURSEG_WARM_DATA;
3243         }
3244 }
3245
3246 static int __get_segment_type_2(struct f2fs_io_info *fio)
3247 {
3248         if (fio->type == DATA)
3249                 return CURSEG_HOT_DATA;
3250         else
3251                 return CURSEG_HOT_NODE;
3252 }
3253
3254 static int __get_segment_type_4(struct f2fs_io_info *fio)
3255 {
3256         if (fio->type == DATA) {
3257                 struct inode *inode = fio->page->mapping->host;
3258
3259                 if (S_ISDIR(inode->i_mode))
3260                         return CURSEG_HOT_DATA;
3261                 else
3262                         return CURSEG_COLD_DATA;
3263         } else {
3264                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3265                         return CURSEG_WARM_NODE;
3266                 else
3267                         return CURSEG_COLD_NODE;
3268         }
3269 }
3270
3271 static int __get_segment_type_6(struct f2fs_io_info *fio)
3272 {
3273         if (fio->type == DATA) {
3274                 struct inode *inode = fio->page->mapping->host;
3275
3276                 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
3277                         return CURSEG_COLD_DATA_PINNED;
3278
3279                 if (page_private_gcing(fio->page)) {
3280                         if (fio->sbi->am.atgc_enabled &&
3281                                 (fio->io_type == FS_DATA_IO) &&
3282                                 (fio->sbi->gc_mode != GC_URGENT_HIGH))
3283                                 return CURSEG_ALL_DATA_ATGC;
3284                         else
3285                                 return CURSEG_COLD_DATA;
3286                 }
3287                 if (file_is_cold(inode) || f2fs_need_compress_data(inode))
3288                         return CURSEG_COLD_DATA;
3289                 if (file_is_hot(inode) ||
3290                                 is_inode_flag_set(inode, FI_HOT_DATA) ||
3291                                 f2fs_is_atomic_file(inode) ||
3292                                 f2fs_is_volatile_file(inode))
3293                         return CURSEG_HOT_DATA;
3294                 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3295         } else {
3296                 if (IS_DNODE(fio->page))
3297                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3298                                                 CURSEG_HOT_NODE;
3299                 return CURSEG_COLD_NODE;
3300         }
3301 }
3302
3303 static int __get_segment_type(struct f2fs_io_info *fio)
3304 {
3305         int type = 0;
3306
3307         switch (F2FS_OPTION(fio->sbi).active_logs) {
3308         case 2:
3309                 type = __get_segment_type_2(fio);
3310                 break;
3311         case 4:
3312                 type = __get_segment_type_4(fio);
3313                 break;
3314         case 6:
3315                 type = __get_segment_type_6(fio);
3316                 break;
3317         default:
3318                 f2fs_bug_on(fio->sbi, true);
3319         }
3320
3321         if (IS_HOT(type))
3322                 fio->temp = HOT;
3323         else if (IS_WARM(type))
3324                 fio->temp = WARM;
3325         else
3326                 fio->temp = COLD;
3327         return type;
3328 }
3329
3330 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3331                 block_t old_blkaddr, block_t *new_blkaddr,
3332                 struct f2fs_summary *sum, int type,
3333                 struct f2fs_io_info *fio)
3334 {
3335         struct sit_info *sit_i = SIT_I(sbi);
3336         struct curseg_info *curseg = CURSEG_I(sbi, type);
3337         unsigned long long old_mtime;
3338         bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3339         struct seg_entry *se = NULL;
3340
3341         f2fs_down_read(&SM_I(sbi)->curseg_lock);
3342
3343         mutex_lock(&curseg->curseg_mutex);
3344         down_write(&sit_i->sentry_lock);
3345
3346         if (from_gc) {
3347                 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3348                 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3349                 sanity_check_seg_type(sbi, se->type);
3350                 f2fs_bug_on(sbi, IS_NODESEG(se->type));
3351         }
3352         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3353
3354         f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg);
3355
3356         f2fs_wait_discard_bio(sbi, *new_blkaddr);
3357
3358         /*
3359          * __add_sum_entry should be resided under the curseg_mutex
3360          * because, this function updates a summary entry in the
3361          * current summary block.
3362          */
3363         __add_sum_entry(sbi, type, sum);
3364
3365         __refresh_next_blkoff(sbi, curseg);
3366
3367         stat_inc_block_count(sbi, curseg);
3368
3369         if (from_gc) {
3370                 old_mtime = get_segment_mtime(sbi, old_blkaddr);
3371         } else {
3372                 update_segment_mtime(sbi, old_blkaddr, 0);
3373                 old_mtime = 0;
3374         }
3375         update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3376
3377         /*
3378          * SIT information should be updated before segment allocation,
3379          * since SSR needs latest valid block information.
3380          */
3381         update_sit_entry(sbi, *new_blkaddr, 1);
3382         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3383                 update_sit_entry(sbi, old_blkaddr, -1);
3384
3385         if (!__has_curseg_space(sbi, curseg)) {
3386                 if (from_gc)
3387                         get_atssr_segment(sbi, type, se->type,
3388                                                 AT_SSR, se->mtime);
3389                 else
3390                         sit_i->s_ops->allocate_segment(sbi, type, false);
3391         }
3392         /*
3393          * segment dirty status should be updated after segment allocation,
3394          * so we just need to update status only one time after previous
3395          * segment being closed.
3396          */
3397         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3398         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3399
3400         up_write(&sit_i->sentry_lock);
3401
3402         if (page && IS_NODESEG(type)) {
3403                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3404
3405                 f2fs_inode_chksum_set(sbi, page);
3406         }
3407
3408         if (fio) {
3409                 struct f2fs_bio_info *io;
3410
3411                 if (F2FS_IO_ALIGNED(sbi))
3412                         fio->retry = false;
3413
3414                 INIT_LIST_HEAD(&fio->list);
3415                 fio->in_list = true;
3416                 io = sbi->write_io[fio->type] + fio->temp;
3417                 spin_lock(&io->io_lock);
3418                 list_add_tail(&fio->list, &io->io_list);
3419                 spin_unlock(&io->io_lock);
3420         }
3421
3422         mutex_unlock(&curseg->curseg_mutex);
3423
3424         f2fs_up_read(&SM_I(sbi)->curseg_lock);
3425 }
3426
3427 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino,
3428                                         block_t blkaddr, unsigned int blkcnt)
3429 {
3430         if (!f2fs_is_multi_device(sbi))
3431                 return;
3432
3433         while (1) {
3434                 unsigned int devidx = f2fs_target_device_index(sbi, blkaddr);
3435                 unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1;
3436
3437                 /* update device state for fsync */
3438                 f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO);
3439
3440                 /* update device state for checkpoint */
3441                 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3442                         spin_lock(&sbi->dev_lock);
3443                         f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3444                         spin_unlock(&sbi->dev_lock);
3445                 }
3446
3447                 if (blkcnt <= blks)
3448                         break;
3449                 blkcnt -= blks;
3450                 blkaddr += blks;
3451         }
3452 }
3453
3454 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3455 {
3456         int type = __get_segment_type(fio);
3457         bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3458
3459         if (keep_order)
3460                 f2fs_down_read(&fio->sbi->io_order_lock);
3461 reallocate:
3462         f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3463                         &fio->new_blkaddr, sum, type, fio);
3464         if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO) {
3465                 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3466                                         fio->old_blkaddr, fio->old_blkaddr);
3467                 f2fs_invalidate_compress_page(fio->sbi, fio->old_blkaddr);
3468         }
3469
3470         /* writeout dirty page into bdev */
3471         f2fs_submit_page_write(fio);
3472         if (fio->retry) {
3473                 fio->old_blkaddr = fio->new_blkaddr;
3474                 goto reallocate;
3475         }
3476
3477         f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1);
3478
3479         if (keep_order)
3480                 f2fs_up_read(&fio->sbi->io_order_lock);
3481 }
3482
3483 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3484                                         enum iostat_type io_type)
3485 {
3486         struct f2fs_io_info fio = {
3487                 .sbi = sbi,
3488                 .type = META,
3489                 .temp = HOT,
3490                 .op = REQ_OP_WRITE,
3491                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3492                 .old_blkaddr = page->index,
3493                 .new_blkaddr = page->index,
3494                 .page = page,
3495                 .encrypted_page = NULL,
3496                 .in_list = false,
3497         };
3498
3499         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3500                 fio.op_flags &= ~REQ_META;
3501
3502         set_page_writeback(page);
3503         ClearPageError(page);
3504         f2fs_submit_page_write(&fio);
3505
3506         stat_inc_meta_count(sbi, page->index);
3507         f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3508 }
3509
3510 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3511 {
3512         struct f2fs_summary sum;
3513
3514         set_summary(&sum, nid, 0, 0);
3515         do_write_page(&sum, fio);
3516
3517         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3518 }
3519
3520 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3521                                         struct f2fs_io_info *fio)
3522 {
3523         struct f2fs_sb_info *sbi = fio->sbi;
3524         struct f2fs_summary sum;
3525
3526         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3527         set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3528         do_write_page(&sum, fio);
3529         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3530
3531         f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3532 }
3533
3534 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3535 {
3536         int err;
3537         struct f2fs_sb_info *sbi = fio->sbi;
3538         unsigned int segno;
3539
3540         fio->new_blkaddr = fio->old_blkaddr;
3541         /* i/o temperature is needed for passing down write hints */
3542         __get_segment_type(fio);
3543
3544         segno = GET_SEGNO(sbi, fio->new_blkaddr);
3545
3546         if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3547                 set_sbi_flag(sbi, SBI_NEED_FSCK);
3548                 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3549                           __func__, segno);
3550                 err = -EFSCORRUPTED;
3551                 goto drop_bio;
3552         }
3553
3554         if (f2fs_cp_error(sbi)) {
3555                 err = -EIO;
3556                 goto drop_bio;
3557         }
3558
3559         invalidate_mapping_pages(META_MAPPING(sbi),
3560                                 fio->new_blkaddr, fio->new_blkaddr);
3561
3562         stat_inc_inplace_blocks(fio->sbi);
3563
3564         if (fio->bio && !(SM_I(sbi)->ipu_policy & (1 << F2FS_IPU_NOCACHE)))
3565                 err = f2fs_merge_page_bio(fio);
3566         else
3567                 err = f2fs_submit_page_bio(fio);
3568         if (!err) {
3569                 f2fs_update_device_state(fio->sbi, fio->ino,
3570                                                 fio->new_blkaddr, 1);
3571                 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3572         }
3573
3574         return err;
3575 drop_bio:
3576         if (fio->bio && *(fio->bio)) {
3577                 struct bio *bio = *(fio->bio);
3578
3579                 bio->bi_status = BLK_STS_IOERR;
3580                 bio_endio(bio);
3581                 *(fio->bio) = NULL;
3582         }
3583         return err;
3584 }
3585
3586 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3587                                                 unsigned int segno)
3588 {
3589         int i;
3590
3591         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3592                 if (CURSEG_I(sbi, i)->segno == segno)
3593                         break;
3594         }
3595         return i;
3596 }
3597
3598 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3599                                 block_t old_blkaddr, block_t new_blkaddr,
3600                                 bool recover_curseg, bool recover_newaddr,
3601                                 bool from_gc)
3602 {
3603         struct sit_info *sit_i = SIT_I(sbi);
3604         struct curseg_info *curseg;
3605         unsigned int segno, old_cursegno;
3606         struct seg_entry *se;
3607         int type;
3608         unsigned short old_blkoff;
3609         unsigned char old_alloc_type;
3610
3611         segno = GET_SEGNO(sbi, new_blkaddr);
3612         se = get_seg_entry(sbi, segno);
3613         type = se->type;
3614
3615         f2fs_down_write(&SM_I(sbi)->curseg_lock);
3616
3617         if (!recover_curseg) {
3618                 /* for recovery flow */
3619                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3620                         if (old_blkaddr == NULL_ADDR)
3621                                 type = CURSEG_COLD_DATA;
3622                         else
3623                                 type = CURSEG_WARM_DATA;
3624                 }
3625         } else {
3626                 if (IS_CURSEG(sbi, segno)) {
3627                         /* se->type is volatile as SSR allocation */
3628                         type = __f2fs_get_curseg(sbi, segno);
3629                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3630                 } else {
3631                         type = CURSEG_WARM_DATA;
3632                 }
3633         }
3634
3635         f2fs_bug_on(sbi, !IS_DATASEG(type));
3636         curseg = CURSEG_I(sbi, type);
3637
3638         mutex_lock(&curseg->curseg_mutex);
3639         down_write(&sit_i->sentry_lock);
3640
3641         old_cursegno = curseg->segno;
3642         old_blkoff = curseg->next_blkoff;
3643         old_alloc_type = curseg->alloc_type;
3644
3645         /* change the current segment */
3646         if (segno != curseg->segno) {
3647                 curseg->next_segno = segno;
3648                 change_curseg(sbi, type, true);
3649         }
3650
3651         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3652         __add_sum_entry(sbi, type, sum);
3653
3654         if (!recover_curseg || recover_newaddr) {
3655                 if (!from_gc)
3656                         update_segment_mtime(sbi, new_blkaddr, 0);
3657                 update_sit_entry(sbi, new_blkaddr, 1);
3658         }
3659         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3660                 invalidate_mapping_pages(META_MAPPING(sbi),
3661                                         old_blkaddr, old_blkaddr);
3662                 f2fs_invalidate_compress_page(sbi, 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         f2fs_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",
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,
4067                                                 GFP_NOFS, true, NULL);
4068
4069         ses->entry_cnt = 0;
4070         INIT_LIST_HEAD(&ses->set_list);
4071         return ses;
4072 }
4073
4074 static void release_sit_entry_set(struct sit_entry_set *ses)
4075 {
4076         list_del(&ses->set_list);
4077         kmem_cache_free(sit_entry_set_slab, ses);
4078 }
4079
4080 static void adjust_sit_entry_set(struct sit_entry_set *ses,
4081                                                 struct list_head *head)
4082 {
4083         struct sit_entry_set *next = ses;
4084
4085         if (list_is_last(&ses->set_list, head))
4086                 return;
4087
4088         list_for_each_entry_continue(next, head, set_list)
4089                 if (ses->entry_cnt <= next->entry_cnt)
4090                         break;
4091
4092         list_move_tail(&ses->set_list, &next->set_list);
4093 }
4094
4095 static void add_sit_entry(unsigned int segno, struct list_head *head)
4096 {
4097         struct sit_entry_set *ses;
4098         unsigned int start_segno = START_SEGNO(segno);
4099
4100         list_for_each_entry(ses, head, set_list) {
4101                 if (ses->start_segno == start_segno) {
4102                         ses->entry_cnt++;
4103                         adjust_sit_entry_set(ses, head);
4104                         return;
4105                 }
4106         }
4107
4108         ses = grab_sit_entry_set();
4109
4110         ses->start_segno = start_segno;
4111         ses->entry_cnt++;
4112         list_add(&ses->set_list, head);
4113 }
4114
4115 static void add_sits_in_set(struct f2fs_sb_info *sbi)
4116 {
4117         struct f2fs_sm_info *sm_info = SM_I(sbi);
4118         struct list_head *set_list = &sm_info->sit_entry_set;
4119         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
4120         unsigned int segno;
4121
4122         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
4123                 add_sit_entry(segno, set_list);
4124 }
4125
4126 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
4127 {
4128         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4129         struct f2fs_journal *journal = curseg->journal;
4130         int i;
4131
4132         down_write(&curseg->journal_rwsem);
4133         for (i = 0; i < sits_in_cursum(journal); i++) {
4134                 unsigned int segno;
4135                 bool dirtied;
4136
4137                 segno = le32_to_cpu(segno_in_journal(journal, i));
4138                 dirtied = __mark_sit_entry_dirty(sbi, segno);
4139
4140                 if (!dirtied)
4141                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
4142         }
4143         update_sits_in_cursum(journal, -i);
4144         up_write(&curseg->journal_rwsem);
4145 }
4146
4147 /*
4148  * CP calls this function, which flushes SIT entries including sit_journal,
4149  * and moves prefree segs to free segs.
4150  */
4151 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
4152 {
4153         struct sit_info *sit_i = SIT_I(sbi);
4154         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4155         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4156         struct f2fs_journal *journal = curseg->journal;
4157         struct sit_entry_set *ses, *tmp;
4158         struct list_head *head = &SM_I(sbi)->sit_entry_set;
4159         bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4160         struct seg_entry *se;
4161
4162         down_write(&sit_i->sentry_lock);
4163
4164         if (!sit_i->dirty_sentries)
4165                 goto out;
4166
4167         /*
4168          * add and account sit entries of dirty bitmap in sit entry
4169          * set temporarily
4170          */
4171         add_sits_in_set(sbi);
4172
4173         /*
4174          * if there are no enough space in journal to store dirty sit
4175          * entries, remove all entries from journal and add and account
4176          * them in sit entry set.
4177          */
4178         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4179                                                                 !to_journal)
4180                 remove_sits_in_journal(sbi);
4181
4182         /*
4183          * there are two steps to flush sit entries:
4184          * #1, flush sit entries to journal in current cold data summary block.
4185          * #2, flush sit entries to sit page.
4186          */
4187         list_for_each_entry_safe(ses, tmp, head, set_list) {
4188                 struct page *page = NULL;
4189                 struct f2fs_sit_block *raw_sit = NULL;
4190                 unsigned int start_segno = ses->start_segno;
4191                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4192                                                 (unsigned long)MAIN_SEGS(sbi));
4193                 unsigned int segno = start_segno;
4194
4195                 if (to_journal &&
4196                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4197                         to_journal = false;
4198
4199                 if (to_journal) {
4200                         down_write(&curseg->journal_rwsem);
4201                 } else {
4202                         page = get_next_sit_page(sbi, start_segno);
4203                         raw_sit = page_address(page);
4204                 }
4205
4206                 /* flush dirty sit entries in region of current sit set */
4207                 for_each_set_bit_from(segno, bitmap, end) {
4208                         int offset, sit_offset;
4209
4210                         se = get_seg_entry(sbi, segno);
4211 #ifdef CONFIG_F2FS_CHECK_FS
4212                         if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4213                                                 SIT_VBLOCK_MAP_SIZE))
4214                                 f2fs_bug_on(sbi, 1);
4215 #endif
4216
4217                         /* add discard candidates */
4218                         if (!(cpc->reason & CP_DISCARD)) {
4219                                 cpc->trim_start = segno;
4220                                 add_discard_addrs(sbi, cpc, false);
4221                         }
4222
4223                         if (to_journal) {
4224                                 offset = f2fs_lookup_journal_in_cursum(journal,
4225                                                         SIT_JOURNAL, segno, 1);
4226                                 f2fs_bug_on(sbi, offset < 0);
4227                                 segno_in_journal(journal, offset) =
4228                                                         cpu_to_le32(segno);
4229                                 seg_info_to_raw_sit(se,
4230                                         &sit_in_journal(journal, offset));
4231                                 check_block_count(sbi, segno,
4232                                         &sit_in_journal(journal, offset));
4233                         } else {
4234                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4235                                 seg_info_to_raw_sit(se,
4236                                                 &raw_sit->entries[sit_offset]);
4237                                 check_block_count(sbi, segno,
4238                                                 &raw_sit->entries[sit_offset]);
4239                         }
4240
4241                         __clear_bit(segno, bitmap);
4242                         sit_i->dirty_sentries--;
4243                         ses->entry_cnt--;
4244                 }
4245
4246                 if (to_journal)
4247                         up_write(&curseg->journal_rwsem);
4248                 else
4249                         f2fs_put_page(page, 1);
4250
4251                 f2fs_bug_on(sbi, ses->entry_cnt);
4252                 release_sit_entry_set(ses);
4253         }
4254
4255         f2fs_bug_on(sbi, !list_empty(head));
4256         f2fs_bug_on(sbi, sit_i->dirty_sentries);
4257 out:
4258         if (cpc->reason & CP_DISCARD) {
4259                 __u64 trim_start = cpc->trim_start;
4260
4261                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4262                         add_discard_addrs(sbi, cpc, false);
4263
4264                 cpc->trim_start = trim_start;
4265         }
4266         up_write(&sit_i->sentry_lock);
4267
4268         set_prefree_as_free_segments(sbi);
4269 }
4270
4271 static int build_sit_info(struct f2fs_sb_info *sbi)
4272 {
4273         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4274         struct sit_info *sit_i;
4275         unsigned int sit_segs, start;
4276         char *src_bitmap, *bitmap;
4277         unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4278         unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0;
4279
4280         /* allocate memory for SIT information */
4281         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4282         if (!sit_i)
4283                 return -ENOMEM;
4284
4285         SM_I(sbi)->sit_info = sit_i;
4286
4287         sit_i->sentries =
4288                 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4289                                               MAIN_SEGS(sbi)),
4290                               GFP_KERNEL);
4291         if (!sit_i->sentries)
4292                 return -ENOMEM;
4293
4294         main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4295         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4296                                                                 GFP_KERNEL);
4297         if (!sit_i->dirty_sentries_bitmap)
4298                 return -ENOMEM;
4299
4300 #ifdef CONFIG_F2FS_CHECK_FS
4301         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map);
4302 #else
4303         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map);
4304 #endif
4305         sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4306         if (!sit_i->bitmap)
4307                 return -ENOMEM;
4308
4309         bitmap = sit_i->bitmap;
4310
4311         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4312                 sit_i->sentries[start].cur_valid_map = bitmap;
4313                 bitmap += SIT_VBLOCK_MAP_SIZE;
4314
4315                 sit_i->sentries[start].ckpt_valid_map = bitmap;
4316                 bitmap += SIT_VBLOCK_MAP_SIZE;
4317
4318 #ifdef CONFIG_F2FS_CHECK_FS
4319                 sit_i->sentries[start].cur_valid_map_mir = bitmap;
4320                 bitmap += SIT_VBLOCK_MAP_SIZE;
4321 #endif
4322
4323                 if (discard_map) {
4324                         sit_i->sentries[start].discard_map = bitmap;
4325                         bitmap += SIT_VBLOCK_MAP_SIZE;
4326                 }
4327         }
4328
4329         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4330         if (!sit_i->tmp_map)
4331                 return -ENOMEM;
4332
4333         if (__is_large_section(sbi)) {
4334                 sit_i->sec_entries =
4335                         f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4336                                                       MAIN_SECS(sbi)),
4337                                       GFP_KERNEL);
4338                 if (!sit_i->sec_entries)
4339                         return -ENOMEM;
4340         }
4341
4342         /* get information related with SIT */
4343         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4344
4345         /* setup SIT bitmap from ckeckpoint pack */
4346         sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4347         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4348
4349         sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4350         if (!sit_i->sit_bitmap)
4351                 return -ENOMEM;
4352
4353 #ifdef CONFIG_F2FS_CHECK_FS
4354         sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4355                                         sit_bitmap_size, GFP_KERNEL);
4356         if (!sit_i->sit_bitmap_mir)
4357                 return -ENOMEM;
4358
4359         sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4360                                         main_bitmap_size, GFP_KERNEL);
4361         if (!sit_i->invalid_segmap)
4362                 return -ENOMEM;
4363 #endif
4364
4365         /* init SIT information */
4366         sit_i->s_ops = &default_salloc_ops;
4367
4368         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4369         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4370         sit_i->written_valid_blocks = 0;
4371         sit_i->bitmap_size = sit_bitmap_size;
4372         sit_i->dirty_sentries = 0;
4373         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4374         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4375         sit_i->mounted_time = ktime_get_boottime_seconds();
4376         init_rwsem(&sit_i->sentry_lock);
4377         return 0;
4378 }
4379
4380 static int build_free_segmap(struct f2fs_sb_info *sbi)
4381 {
4382         struct free_segmap_info *free_i;
4383         unsigned int bitmap_size, sec_bitmap_size;
4384
4385         /* allocate memory for free segmap information */
4386         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4387         if (!free_i)
4388                 return -ENOMEM;
4389
4390         SM_I(sbi)->free_info = free_i;
4391
4392         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4393         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4394         if (!free_i->free_segmap)
4395                 return -ENOMEM;
4396
4397         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4398         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4399         if (!free_i->free_secmap)
4400                 return -ENOMEM;
4401
4402         /* set all segments as dirty temporarily */
4403         memset(free_i->free_segmap, 0xff, bitmap_size);
4404         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4405
4406         /* init free segmap information */
4407         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4408         free_i->free_segments = 0;
4409         free_i->free_sections = 0;
4410         spin_lock_init(&free_i->segmap_lock);
4411         return 0;
4412 }
4413
4414 static int build_curseg(struct f2fs_sb_info *sbi)
4415 {
4416         struct curseg_info *array;
4417         int i;
4418
4419         array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4420                                         sizeof(*array)), GFP_KERNEL);
4421         if (!array)
4422                 return -ENOMEM;
4423
4424         SM_I(sbi)->curseg_array = array;
4425
4426         for (i = 0; i < NO_CHECK_TYPE; i++) {
4427                 mutex_init(&array[i].curseg_mutex);
4428                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4429                 if (!array[i].sum_blk)
4430                         return -ENOMEM;
4431                 init_rwsem(&array[i].journal_rwsem);
4432                 array[i].journal = f2fs_kzalloc(sbi,
4433                                 sizeof(struct f2fs_journal), GFP_KERNEL);
4434                 if (!array[i].journal)
4435                         return -ENOMEM;
4436                 if (i < NR_PERSISTENT_LOG)
4437                         array[i].seg_type = CURSEG_HOT_DATA + i;
4438                 else if (i == CURSEG_COLD_DATA_PINNED)
4439                         array[i].seg_type = CURSEG_COLD_DATA;
4440                 else if (i == CURSEG_ALL_DATA_ATGC)
4441                         array[i].seg_type = CURSEG_COLD_DATA;
4442                 array[i].segno = NULL_SEGNO;
4443                 array[i].next_blkoff = 0;
4444                 array[i].inited = false;
4445         }
4446         return restore_curseg_summaries(sbi);
4447 }
4448
4449 static int build_sit_entries(struct f2fs_sb_info *sbi)
4450 {
4451         struct sit_info *sit_i = SIT_I(sbi);
4452         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4453         struct f2fs_journal *journal = curseg->journal;
4454         struct seg_entry *se;
4455         struct f2fs_sit_entry sit;
4456         int sit_blk_cnt = SIT_BLK_CNT(sbi);
4457         unsigned int i, start, end;
4458         unsigned int readed, start_blk = 0;
4459         int err = 0;
4460         block_t total_node_blocks = 0;
4461
4462         do {
4463                 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS,
4464                                                         META_SIT, true);
4465
4466                 start = start_blk * sit_i->sents_per_block;
4467                 end = (start_blk + readed) * sit_i->sents_per_block;
4468
4469                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4470                         struct f2fs_sit_block *sit_blk;
4471                         struct page *page;
4472
4473                         se = &sit_i->sentries[start];
4474                         page = get_current_sit_page(sbi, start);
4475                         if (IS_ERR(page))
4476                                 return PTR_ERR(page);
4477                         sit_blk = (struct f2fs_sit_block *)page_address(page);
4478                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4479                         f2fs_put_page(page, 1);
4480
4481                         err = check_block_count(sbi, start, &sit);
4482                         if (err)
4483                                 return err;
4484                         seg_info_from_raw_sit(se, &sit);
4485                         if (IS_NODESEG(se->type))
4486                                 total_node_blocks += se->valid_blocks;
4487
4488                         if (f2fs_block_unit_discard(sbi)) {
4489                                 /* build discard map only one time */
4490                                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4491                                         memset(se->discard_map, 0xff,
4492                                                 SIT_VBLOCK_MAP_SIZE);
4493                                 } else {
4494                                         memcpy(se->discard_map,
4495                                                 se->cur_valid_map,
4496                                                 SIT_VBLOCK_MAP_SIZE);
4497                                         sbi->discard_blks +=
4498                                                 sbi->blocks_per_seg -
4499                                                 se->valid_blocks;
4500                                 }
4501                         }
4502
4503                         if (__is_large_section(sbi))
4504                                 get_sec_entry(sbi, start)->valid_blocks +=
4505                                                         se->valid_blocks;
4506                 }
4507                 start_blk += readed;
4508         } while (start_blk < sit_blk_cnt);
4509
4510         down_read(&curseg->journal_rwsem);
4511         for (i = 0; i < sits_in_cursum(journal); i++) {
4512                 unsigned int old_valid_blocks;
4513
4514                 start = le32_to_cpu(segno_in_journal(journal, i));
4515                 if (start >= MAIN_SEGS(sbi)) {
4516                         f2fs_err(sbi, "Wrong journal entry on segno %u",
4517                                  start);
4518                         err = -EFSCORRUPTED;
4519                         break;
4520                 }
4521
4522                 se = &sit_i->sentries[start];
4523                 sit = sit_in_journal(journal, i);
4524
4525                 old_valid_blocks = se->valid_blocks;
4526                 if (IS_NODESEG(se->type))
4527                         total_node_blocks -= old_valid_blocks;
4528
4529                 err = check_block_count(sbi, start, &sit);
4530                 if (err)
4531                         break;
4532                 seg_info_from_raw_sit(se, &sit);
4533                 if (IS_NODESEG(se->type))
4534                         total_node_blocks += se->valid_blocks;
4535
4536                 if (f2fs_block_unit_discard(sbi)) {
4537                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4538                                 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4539                         } else {
4540                                 memcpy(se->discard_map, se->cur_valid_map,
4541                                                         SIT_VBLOCK_MAP_SIZE);
4542                                 sbi->discard_blks += old_valid_blocks;
4543                                 sbi->discard_blks -= se->valid_blocks;
4544                         }
4545                 }
4546
4547                 if (__is_large_section(sbi)) {
4548                         get_sec_entry(sbi, start)->valid_blocks +=
4549                                                         se->valid_blocks;
4550                         get_sec_entry(sbi, start)->valid_blocks -=
4551                                                         old_valid_blocks;
4552                 }
4553         }
4554         up_read(&curseg->journal_rwsem);
4555
4556         if (!err && total_node_blocks != valid_node_count(sbi)) {
4557                 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4558                          total_node_blocks, valid_node_count(sbi));
4559                 err = -EFSCORRUPTED;
4560         }
4561
4562         return err;
4563 }
4564
4565 static void init_free_segmap(struct f2fs_sb_info *sbi)
4566 {
4567         unsigned int start;
4568         int type;
4569         struct seg_entry *sentry;
4570
4571         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4572                 if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4573                         continue;
4574                 sentry = get_seg_entry(sbi, start);
4575                 if (!sentry->valid_blocks)
4576                         __set_free(sbi, start);
4577                 else
4578                         SIT_I(sbi)->written_valid_blocks +=
4579                                                 sentry->valid_blocks;
4580         }
4581
4582         /* set use the current segments */
4583         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4584                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4585
4586                 __set_test_and_inuse(sbi, curseg_t->segno);
4587         }
4588 }
4589
4590 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4591 {
4592         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4593         struct free_segmap_info *free_i = FREE_I(sbi);
4594         unsigned int segno = 0, offset = 0, secno;
4595         block_t valid_blocks, usable_blks_in_seg;
4596         block_t blks_per_sec = BLKS_PER_SEC(sbi);
4597
4598         while (1) {
4599                 /* find dirty segment based on free segmap */
4600                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4601                 if (segno >= MAIN_SEGS(sbi))
4602                         break;
4603                 offset = segno + 1;
4604                 valid_blocks = get_valid_blocks(sbi, segno, false);
4605                 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4606                 if (valid_blocks == usable_blks_in_seg || !valid_blocks)
4607                         continue;
4608                 if (valid_blocks > usable_blks_in_seg) {
4609                         f2fs_bug_on(sbi, 1);
4610                         continue;
4611                 }
4612                 mutex_lock(&dirty_i->seglist_lock);
4613                 __locate_dirty_segment(sbi, segno, DIRTY);
4614                 mutex_unlock(&dirty_i->seglist_lock);
4615         }
4616
4617         if (!__is_large_section(sbi))
4618                 return;
4619
4620         mutex_lock(&dirty_i->seglist_lock);
4621         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4622                 valid_blocks = get_valid_blocks(sbi, segno, true);
4623                 secno = GET_SEC_FROM_SEG(sbi, segno);
4624
4625                 if (!valid_blocks || valid_blocks == blks_per_sec)
4626                         continue;
4627                 if (IS_CURSEC(sbi, secno))
4628                         continue;
4629                 set_bit(secno, dirty_i->dirty_secmap);
4630         }
4631         mutex_unlock(&dirty_i->seglist_lock);
4632 }
4633
4634 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4635 {
4636         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4637         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4638
4639         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4640         if (!dirty_i->victim_secmap)
4641                 return -ENOMEM;
4642         return 0;
4643 }
4644
4645 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4646 {
4647         struct dirty_seglist_info *dirty_i;
4648         unsigned int bitmap_size, i;
4649
4650         /* allocate memory for dirty segments list information */
4651         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4652                                                                 GFP_KERNEL);
4653         if (!dirty_i)
4654                 return -ENOMEM;
4655
4656         SM_I(sbi)->dirty_info = dirty_i;
4657         mutex_init(&dirty_i->seglist_lock);
4658
4659         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4660
4661         for (i = 0; i < NR_DIRTY_TYPE; i++) {
4662                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4663                                                                 GFP_KERNEL);
4664                 if (!dirty_i->dirty_segmap[i])
4665                         return -ENOMEM;
4666         }
4667
4668         if (__is_large_section(sbi)) {
4669                 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4670                 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
4671                                                 bitmap_size, GFP_KERNEL);
4672                 if (!dirty_i->dirty_secmap)
4673                         return -ENOMEM;
4674         }
4675
4676         init_dirty_segmap(sbi);
4677         return init_victim_secmap(sbi);
4678 }
4679
4680 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4681 {
4682         int i;
4683
4684         /*
4685          * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4686          * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4687          */
4688         for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4689                 struct curseg_info *curseg = CURSEG_I(sbi, i);
4690                 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4691                 unsigned int blkofs = curseg->next_blkoff;
4692
4693                 if (f2fs_sb_has_readonly(sbi) &&
4694                         i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE)
4695                         continue;
4696
4697                 sanity_check_seg_type(sbi, curseg->seg_type);
4698
4699                 if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) {
4700                         f2fs_err(sbi,
4701                                  "Current segment has invalid alloc_type:%d",
4702                                  curseg->alloc_type);
4703                         return -EFSCORRUPTED;
4704                 }
4705
4706                 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4707                         goto out;
4708
4709                 if (curseg->alloc_type == SSR)
4710                         continue;
4711
4712                 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4713                         if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4714                                 continue;
4715 out:
4716                         f2fs_err(sbi,
4717                                  "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4718                                  i, curseg->segno, curseg->alloc_type,
4719                                  curseg->next_blkoff, blkofs);
4720                         return -EFSCORRUPTED;
4721                 }
4722         }
4723         return 0;
4724 }
4725
4726 #ifdef CONFIG_BLK_DEV_ZONED
4727
4728 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4729                                     struct f2fs_dev_info *fdev,
4730                                     struct blk_zone *zone)
4731 {
4732         unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno;
4733         block_t zone_block, wp_block, last_valid_block;
4734         unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4735         int i, s, b, ret;
4736         struct seg_entry *se;
4737
4738         if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4739                 return 0;
4740
4741         wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block);
4742         wp_segno = GET_SEGNO(sbi, wp_block);
4743         wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4744         zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4745         zone_segno = GET_SEGNO(sbi, zone_block);
4746         zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno);
4747
4748         if (zone_segno >= MAIN_SEGS(sbi))
4749                 return 0;
4750
4751         /*
4752          * Skip check of zones cursegs point to, since
4753          * fix_curseg_write_pointer() checks them.
4754          */
4755         for (i = 0; i < NO_CHECK_TYPE; i++)
4756                 if (zone_secno == GET_SEC_FROM_SEG(sbi,
4757                                                    CURSEG_I(sbi, i)->segno))
4758                         return 0;
4759
4760         /*
4761          * Get last valid block of the zone.
4762          */
4763         last_valid_block = zone_block - 1;
4764         for (s = sbi->segs_per_sec - 1; s >= 0; s--) {
4765                 segno = zone_segno + s;
4766                 se = get_seg_entry(sbi, segno);
4767                 for (b = sbi->blocks_per_seg - 1; b >= 0; b--)
4768                         if (f2fs_test_bit(b, se->cur_valid_map)) {
4769                                 last_valid_block = START_BLOCK(sbi, segno) + b;
4770                                 break;
4771                         }
4772                 if (last_valid_block >= zone_block)
4773                         break;
4774         }
4775
4776         /*
4777          * If last valid block is beyond the write pointer, report the
4778          * inconsistency. This inconsistency does not cause write error
4779          * because the zone will not be selected for write operation until
4780          * it get discarded. Just report it.
4781          */
4782         if (last_valid_block >= wp_block) {
4783                 f2fs_notice(sbi, "Valid block beyond write pointer: "
4784                             "valid block[0x%x,0x%x] wp[0x%x,0x%x]",
4785                             GET_SEGNO(sbi, last_valid_block),
4786                             GET_BLKOFF_FROM_SEG0(sbi, last_valid_block),
4787                             wp_segno, wp_blkoff);
4788                 return 0;
4789         }
4790
4791         /*
4792          * If there is no valid block in the zone and if write pointer is
4793          * not at zone start, reset the write pointer.
4794          */
4795         if (last_valid_block + 1 == zone_block && zone->wp != zone->start) {
4796                 f2fs_notice(sbi,
4797                             "Zone without valid block has non-zero write "
4798                             "pointer. Reset the write pointer: wp[0x%x,0x%x]",
4799                             wp_segno, wp_blkoff);
4800                 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
4801                                         zone->len >> log_sectors_per_block);
4802                 if (ret) {
4803                         f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4804                                  fdev->path, ret);
4805                         return ret;
4806                 }
4807         }
4808
4809         return 0;
4810 }
4811
4812 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
4813                                                   block_t zone_blkaddr)
4814 {
4815         int i;
4816
4817         for (i = 0; i < sbi->s_ndevs; i++) {
4818                 if (!bdev_is_zoned(FDEV(i).bdev))
4819                         continue;
4820                 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
4821                                 zone_blkaddr <= FDEV(i).end_blk))
4822                         return &FDEV(i);
4823         }
4824
4825         return NULL;
4826 }
4827
4828 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
4829                               void *data)
4830 {
4831         memcpy(data, zone, sizeof(struct blk_zone));
4832         return 0;
4833 }
4834
4835 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
4836 {
4837         struct curseg_info *cs = CURSEG_I(sbi, type);
4838         struct f2fs_dev_info *zbd;
4839         struct blk_zone zone;
4840         unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
4841         block_t cs_zone_block, wp_block;
4842         unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4843         sector_t zone_sector;
4844         int err;
4845
4846         cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4847         cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4848
4849         zbd = get_target_zoned_dev(sbi, cs_zone_block);
4850         if (!zbd)
4851                 return 0;
4852
4853         /* report zone for the sector the curseg points to */
4854         zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4855                 << log_sectors_per_block;
4856         err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4857                                   report_one_zone_cb, &zone);
4858         if (err != 1) {
4859                 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4860                          zbd->path, err);
4861                 return err;
4862         }
4863
4864         if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4865                 return 0;
4866
4867         wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
4868         wp_segno = GET_SEGNO(sbi, wp_block);
4869         wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4870         wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
4871
4872         if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
4873                 wp_sector_off == 0)
4874                 return 0;
4875
4876         f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
4877                     "curseg[0x%x,0x%x] wp[0x%x,0x%x]",
4878                     type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff);
4879
4880         f2fs_notice(sbi, "Assign new section to curseg[%d]: "
4881                     "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff);
4882
4883         f2fs_allocate_new_section(sbi, type, true);
4884
4885         /* check consistency of the zone curseg pointed to */
4886         if (check_zone_write_pointer(sbi, zbd, &zone))
4887                 return -EIO;
4888
4889         /* check newly assigned zone */
4890         cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4891         cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4892
4893         zbd = get_target_zoned_dev(sbi, cs_zone_block);
4894         if (!zbd)
4895                 return 0;
4896
4897         zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4898                 << log_sectors_per_block;
4899         err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4900                                   report_one_zone_cb, &zone);
4901         if (err != 1) {
4902                 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4903                          zbd->path, err);
4904                 return err;
4905         }
4906
4907         if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4908                 return 0;
4909
4910         if (zone.wp != zone.start) {
4911                 f2fs_notice(sbi,
4912                             "New zone for curseg[%d] is not yet discarded. "
4913                             "Reset the zone: curseg[0x%x,0x%x]",
4914                             type, cs->segno, cs->next_blkoff);
4915                 err = __f2fs_issue_discard_zone(sbi, zbd->bdev,
4916                                 zone_sector >> log_sectors_per_block,
4917                                 zone.len >> log_sectors_per_block);
4918                 if (err) {
4919                         f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4920                                  zbd->path, err);
4921                         return err;
4922                 }
4923         }
4924
4925         return 0;
4926 }
4927
4928 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4929 {
4930         int i, ret;
4931
4932         for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4933                 ret = fix_curseg_write_pointer(sbi, i);
4934                 if (ret)
4935                         return ret;
4936         }
4937
4938         return 0;
4939 }
4940
4941 struct check_zone_write_pointer_args {
4942         struct f2fs_sb_info *sbi;
4943         struct f2fs_dev_info *fdev;
4944 };
4945
4946 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
4947                                       void *data)
4948 {
4949         struct check_zone_write_pointer_args *args;
4950
4951         args = (struct check_zone_write_pointer_args *)data;
4952
4953         return check_zone_write_pointer(args->sbi, args->fdev, zone);
4954 }
4955
4956 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4957 {
4958         int i, ret;
4959         struct check_zone_write_pointer_args args;
4960
4961         for (i = 0; i < sbi->s_ndevs; i++) {
4962                 if (!bdev_is_zoned(FDEV(i).bdev))
4963                         continue;
4964
4965                 args.sbi = sbi;
4966                 args.fdev = &FDEV(i);
4967                 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
4968                                           check_zone_write_pointer_cb, &args);
4969                 if (ret < 0)
4970                         return ret;
4971         }
4972
4973         return 0;
4974 }
4975
4976 static bool is_conv_zone(struct f2fs_sb_info *sbi, unsigned int zone_idx,
4977                                                 unsigned int dev_idx)
4978 {
4979         if (!bdev_is_zoned(FDEV(dev_idx).bdev))
4980                 return true;
4981         return !test_bit(zone_idx, FDEV(dev_idx).blkz_seq);
4982 }
4983
4984 /* Return the zone index in the given device */
4985 static unsigned int get_zone_idx(struct f2fs_sb_info *sbi, unsigned int secno,
4986                                         int dev_idx)
4987 {
4988         block_t sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
4989
4990         return (sec_start_blkaddr - FDEV(dev_idx).start_blk) >>
4991                                                 sbi->log_blocks_per_blkz;
4992 }
4993
4994 /*
4995  * Return the usable segments in a section based on the zone's
4996  * corresponding zone capacity. Zone is equal to a section.
4997  */
4998 static inline unsigned int f2fs_usable_zone_segs_in_sec(
4999                 struct f2fs_sb_info *sbi, unsigned int segno)
5000 {
5001         unsigned int dev_idx, zone_idx, unusable_segs_in_sec;
5002
5003         dev_idx = f2fs_target_device_index(sbi, START_BLOCK(sbi, segno));
5004         zone_idx = get_zone_idx(sbi, GET_SEC_FROM_SEG(sbi, segno), dev_idx);
5005
5006         /* Conventional zone's capacity is always equal to zone size */
5007         if (is_conv_zone(sbi, zone_idx, dev_idx))
5008                 return sbi->segs_per_sec;
5009
5010         /*
5011          * If the zone_capacity_blocks array is NULL, then zone capacity
5012          * is equal to the zone size for all zones
5013          */
5014         if (!FDEV(dev_idx).zone_capacity_blocks)
5015                 return sbi->segs_per_sec;
5016
5017         /* Get the segment count beyond zone capacity block */
5018         unusable_segs_in_sec = (sbi->blocks_per_blkz -
5019                                 FDEV(dev_idx).zone_capacity_blocks[zone_idx]) >>
5020                                 sbi->log_blocks_per_seg;
5021         return sbi->segs_per_sec - unusable_segs_in_sec;
5022 }
5023
5024 /*
5025  * Return the number of usable blocks in a segment. The number of blocks
5026  * returned is always equal to the number of blocks in a segment for
5027  * segments fully contained within a sequential zone capacity or a
5028  * conventional zone. For segments partially contained in a sequential
5029  * zone capacity, the number of usable blocks up to the zone capacity
5030  * is returned. 0 is returned in all other cases.
5031  */
5032 static inline unsigned int f2fs_usable_zone_blks_in_seg(
5033                         struct f2fs_sb_info *sbi, unsigned int segno)
5034 {
5035         block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
5036         unsigned int zone_idx, dev_idx, secno;
5037
5038         secno = GET_SEC_FROM_SEG(sbi, segno);
5039         seg_start = START_BLOCK(sbi, segno);
5040         dev_idx = f2fs_target_device_index(sbi, seg_start);
5041         zone_idx = get_zone_idx(sbi, secno, dev_idx);
5042
5043         /*
5044          * Conventional zone's capacity is always equal to zone size,
5045          * so, blocks per segment is unchanged.
5046          */
5047         if (is_conv_zone(sbi, zone_idx, dev_idx))
5048                 return sbi->blocks_per_seg;
5049
5050         if (!FDEV(dev_idx).zone_capacity_blocks)
5051                 return sbi->blocks_per_seg;
5052
5053         sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
5054         sec_cap_blkaddr = sec_start_blkaddr +
5055                                 FDEV(dev_idx).zone_capacity_blocks[zone_idx];
5056
5057         /*
5058          * If segment starts before zone capacity and spans beyond
5059          * zone capacity, then usable blocks are from seg start to
5060          * zone capacity. If the segment starts after the zone capacity,
5061          * then there are no usable blocks.
5062          */
5063         if (seg_start >= sec_cap_blkaddr)
5064                 return 0;
5065         if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr)
5066                 return sec_cap_blkaddr - seg_start;
5067
5068         return sbi->blocks_per_seg;
5069 }
5070 #else
5071 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5072 {
5073         return 0;
5074 }
5075
5076 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5077 {
5078         return 0;
5079 }
5080
5081 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
5082                                                         unsigned int segno)
5083 {
5084         return 0;
5085 }
5086
5087 static inline unsigned int f2fs_usable_zone_segs_in_sec(struct f2fs_sb_info *sbi,
5088                                                         unsigned int segno)
5089 {
5090         return 0;
5091 }
5092 #endif
5093 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
5094                                         unsigned int segno)
5095 {
5096         if (f2fs_sb_has_blkzoned(sbi))
5097                 return f2fs_usable_zone_blks_in_seg(sbi, segno);
5098
5099         return sbi->blocks_per_seg;
5100 }
5101
5102 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi,
5103                                         unsigned int segno)
5104 {
5105         if (f2fs_sb_has_blkzoned(sbi))
5106                 return f2fs_usable_zone_segs_in_sec(sbi, segno);
5107
5108         return sbi->segs_per_sec;
5109 }
5110
5111 /*
5112  * Update min, max modified time for cost-benefit GC algorithm
5113  */
5114 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5115 {
5116         struct sit_info *sit_i = SIT_I(sbi);
5117         unsigned int segno;
5118
5119         down_write(&sit_i->sentry_lock);
5120
5121         sit_i->min_mtime = ULLONG_MAX;
5122
5123         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
5124                 unsigned int i;
5125                 unsigned long long mtime = 0;
5126
5127                 for (i = 0; i < sbi->segs_per_sec; i++)
5128                         mtime += get_seg_entry(sbi, segno + i)->mtime;
5129
5130                 mtime = div_u64(mtime, sbi->segs_per_sec);
5131
5132                 if (sit_i->min_mtime > mtime)
5133                         sit_i->min_mtime = mtime;
5134         }
5135         sit_i->max_mtime = get_mtime(sbi, false);
5136         sit_i->dirty_max_mtime = 0;
5137         up_write(&sit_i->sentry_lock);
5138 }
5139
5140 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
5141 {
5142         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5143         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
5144         struct f2fs_sm_info *sm_info;
5145         int err;
5146
5147         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
5148         if (!sm_info)
5149                 return -ENOMEM;
5150
5151         /* init sm info */
5152         sbi->sm_info = sm_info;
5153         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5154         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5155         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5156         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5157         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5158         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5159         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
5160         sm_info->rec_prefree_segments = sm_info->main_segments *
5161                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
5162         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5163                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5164
5165         if (!f2fs_lfs_mode(sbi))
5166                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
5167         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
5168         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
5169         sm_info->min_seq_blocks = sbi->blocks_per_seg;
5170         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
5171         sm_info->min_ssr_sections = reserved_sections(sbi);
5172
5173         INIT_LIST_HEAD(&sm_info->sit_entry_set);
5174
5175         init_f2fs_rwsem(&sm_info->curseg_lock);
5176
5177         if (!f2fs_readonly(sbi->sb)) {
5178                 err = f2fs_create_flush_cmd_control(sbi);
5179                 if (err)
5180                         return err;
5181         }
5182
5183         err = create_discard_cmd_control(sbi);
5184         if (err)
5185                 return err;
5186
5187         err = build_sit_info(sbi);
5188         if (err)
5189                 return err;
5190         err = build_free_segmap(sbi);
5191         if (err)
5192                 return err;
5193         err = build_curseg(sbi);
5194         if (err)
5195                 return err;
5196
5197         /* reinit free segmap based on SIT */
5198         err = build_sit_entries(sbi);
5199         if (err)
5200                 return err;
5201
5202         init_free_segmap(sbi);
5203         err = build_dirty_segmap(sbi);
5204         if (err)
5205                 return err;
5206
5207         err = sanity_check_curseg(sbi);
5208         if (err)
5209                 return err;
5210
5211         init_min_max_mtime(sbi);
5212         return 0;
5213 }
5214
5215 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5216                 enum dirty_type dirty_type)
5217 {
5218         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5219
5220         mutex_lock(&dirty_i->seglist_lock);
5221         kvfree(dirty_i->dirty_segmap[dirty_type]);
5222         dirty_i->nr_dirty[dirty_type] = 0;
5223         mutex_unlock(&dirty_i->seglist_lock);
5224 }
5225
5226 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5227 {
5228         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5229
5230         kvfree(dirty_i->victim_secmap);
5231 }
5232
5233 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5234 {
5235         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5236         int i;
5237
5238         if (!dirty_i)
5239                 return;
5240
5241         /* discard pre-free/dirty segments list */
5242         for (i = 0; i < NR_DIRTY_TYPE; i++)
5243                 discard_dirty_segmap(sbi, i);
5244
5245         if (__is_large_section(sbi)) {
5246                 mutex_lock(&dirty_i->seglist_lock);
5247                 kvfree(dirty_i->dirty_secmap);
5248                 mutex_unlock(&dirty_i->seglist_lock);
5249         }
5250
5251         destroy_victim_secmap(sbi);
5252         SM_I(sbi)->dirty_info = NULL;
5253         kfree(dirty_i);
5254 }
5255
5256 static void destroy_curseg(struct f2fs_sb_info *sbi)
5257 {
5258         struct curseg_info *array = SM_I(sbi)->curseg_array;
5259         int i;
5260
5261         if (!array)
5262                 return;
5263         SM_I(sbi)->curseg_array = NULL;
5264         for (i = 0; i < NR_CURSEG_TYPE; i++) {
5265                 kfree(array[i].sum_blk);
5266                 kfree(array[i].journal);
5267         }
5268         kfree(array);
5269 }
5270
5271 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5272 {
5273         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5274
5275         if (!free_i)
5276                 return;
5277         SM_I(sbi)->free_info = NULL;
5278         kvfree(free_i->free_segmap);
5279         kvfree(free_i->free_secmap);
5280         kfree(free_i);
5281 }
5282
5283 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5284 {
5285         struct sit_info *sit_i = SIT_I(sbi);
5286
5287         if (!sit_i)
5288                 return;
5289
5290         if (sit_i->sentries)
5291                 kvfree(sit_i->bitmap);
5292         kfree(sit_i->tmp_map);
5293
5294         kvfree(sit_i->sentries);
5295         kvfree(sit_i->sec_entries);
5296         kvfree(sit_i->dirty_sentries_bitmap);
5297
5298         SM_I(sbi)->sit_info = NULL;
5299         kvfree(sit_i->sit_bitmap);
5300 #ifdef CONFIG_F2FS_CHECK_FS
5301         kvfree(sit_i->sit_bitmap_mir);
5302         kvfree(sit_i->invalid_segmap);
5303 #endif
5304         kfree(sit_i);
5305 }
5306
5307 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5308 {
5309         struct f2fs_sm_info *sm_info = SM_I(sbi);
5310
5311         if (!sm_info)
5312                 return;
5313         f2fs_destroy_flush_cmd_control(sbi, true);
5314         destroy_discard_cmd_control(sbi);
5315         destroy_dirty_segmap(sbi);
5316         destroy_curseg(sbi);
5317         destroy_free_segmap(sbi);
5318         destroy_sit_info(sbi);
5319         sbi->sm_info = NULL;
5320         kfree(sm_info);
5321 }
5322
5323 int __init f2fs_create_segment_manager_caches(void)
5324 {
5325         discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5326                         sizeof(struct discard_entry));
5327         if (!discard_entry_slab)
5328                 goto fail;
5329
5330         discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5331                         sizeof(struct discard_cmd));
5332         if (!discard_cmd_slab)
5333                 goto destroy_discard_entry;
5334
5335         sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5336                         sizeof(struct sit_entry_set));
5337         if (!sit_entry_set_slab)
5338                 goto destroy_discard_cmd;
5339
5340         inmem_entry_slab = f2fs_kmem_cache_create("f2fs_inmem_page_entry",
5341                         sizeof(struct inmem_pages));
5342         if (!inmem_entry_slab)
5343                 goto destroy_sit_entry_set;
5344         return 0;
5345
5346 destroy_sit_entry_set:
5347         kmem_cache_destroy(sit_entry_set_slab);
5348 destroy_discard_cmd:
5349         kmem_cache_destroy(discard_cmd_slab);
5350 destroy_discard_entry:
5351         kmem_cache_destroy(discard_entry_slab);
5352 fail:
5353         return -ENOMEM;
5354 }
5355
5356 void f2fs_destroy_segment_manager_caches(void)
5357 {
5358         kmem_cache_destroy(sit_entry_set_slab);
5359         kmem_cache_destroy(discard_cmd_slab);
5360         kmem_cache_destroy(discard_entry_slab);
5361         kmem_cache_destroy(inmem_entry_slab);
5362 }