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 *revoke_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_abort_atomic_write(struct inode *inode, bool clean)
 189 {
 190         struct f2fs_inode_info *fi = F2FS_I(inode);
 191
 192         if (!f2fs_is_atomic_file(inode))
 193                 return;
 194
 195         release_atomic_write_cnt(inode);
 196         clear_inode_flag(inode, FI_ATOMIC_COMMITTED);
 197         clear_inode_flag(inode, FI_ATOMIC_REPLACE);
 198         clear_inode_flag(inode, FI_ATOMIC_FILE);
 199         stat_dec_atomic_inode(inode);
 200
 201         F2FS_I(inode)->atomic_write_task = NULL;
 202
 203         if (clean) {
 204                 truncate_inode_pages_final(inode->i_mapping);
 205                 f2fs_i_size_write(inode, fi->original_i_size);
 206                 fi->original_i_size = 0;
 207         }
 208 }
 209
 210 static int __replace_atomic_write_block(struct inode *inode, pgoff_t index,
 211                         block_t new_addr, block_t *old_addr, bool recover)
 212 {
 213         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 214         struct dnode_of_data dn;
 215         struct node_info ni;
 216         int err;
 217
 218 retry:
 219         set_new_dnode(&dn, inode, NULL, NULL, 0);
 220         err = f2fs_get_dnode_of_data(&dn, index, ALLOC_NODE);
 221         if (err) {
 222                 if (err == -ENOMEM) {
 223                         f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
 224                         goto retry;
 225                 }
 226                 return err;
 227         }
 228
 229         err = f2fs_get_node_info(sbi, dn.nid, &ni, false);
 230         if (err) {
 231                 f2fs_put_dnode(&dn);
 232                 return err;
 233         }
 234
 235         if (recover) {
 236                 /* dn.data_blkaddr is always valid */
 237                 if (!__is_valid_data_blkaddr(new_addr)) {
 238                         if (new_addr == NULL_ADDR)
 239                                 dec_valid_block_count(sbi, inode, 1);
 240                         f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
 241                         f2fs_update_data_blkaddr(&dn, new_addr);
 242                 } else {
 243                         f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 244                                 new_addr, ni.version, true, true);
 245                 }
 246         } else {
 247                 blkcnt_t count = 1;
 248
 249                 err = inc_valid_block_count(sbi, inode, &count);
 250                 if (err) {
 251                         f2fs_put_dnode(&dn);
 252                         return err;
 253                 }
 254
 255                 *old_addr = dn.data_blkaddr;
 256                 f2fs_truncate_data_blocks_range(&dn, 1);
 257                 dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count);
 258
 259                 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr,
 260                                         ni.version, true, false);
 261         }
 262
 263         f2fs_put_dnode(&dn);
 264
 265         trace_f2fs_replace_atomic_write_block(inode, F2FS_I(inode)->cow_inode,
 266                         index, old_addr ? *old_addr : 0, new_addr, recover);
 267         return 0;
 268 }
 269
 270 static void __complete_revoke_list(struct inode *inode, struct list_head *head,
 271                                         bool revoke)
 272 {
 273         struct revoke_entry *cur, *tmp;
 274         pgoff_t start_index = 0;
 275         bool truncate = is_inode_flag_set(inode, FI_ATOMIC_REPLACE);
 276
 277         list_for_each_entry_safe(cur, tmp, head, list) {
 278                 if (revoke) {
 279                         __replace_atomic_write_block(inode, cur->index,
 280                                                 cur->old_addr, NULL, true);
 281                 } else if (truncate) {
 282                         f2fs_truncate_hole(inode, start_index, cur->index);
 283                         start_index = cur->index + 1;
 284                 }
 285
 286                 list_del(&cur->list);
 287                 kmem_cache_free(revoke_entry_slab, cur);
 288         }
 289
 290         if (!revoke && truncate)
 291                 f2fs_do_truncate_blocks(inode, start_index * PAGE_SIZE, false);
 292 }
 293
 294 static int __f2fs_commit_atomic_write(struct inode *inode)
 295 {
 296         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 297         struct f2fs_inode_info *fi = F2FS_I(inode);
 298         struct inode *cow_inode = fi->cow_inode;
 299         struct revoke_entry *new;
 300         struct list_head revoke_list;
 301         block_t blkaddr;
 302         struct dnode_of_data dn;
 303         pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
 304         pgoff_t off = 0, blen, index;
 305         int ret = 0, i;
 306
 307         INIT_LIST_HEAD(&revoke_list);
 308
 309         while (len) {
 310                 blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len);
 311
 312                 set_new_dnode(&dn, cow_inode, NULL, NULL, 0);
 313                 ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA);
 314                 if (ret && ret != -ENOENT) {
 315                         goto out;
 316                 } else if (ret == -ENOENT) {
 317                         ret = 0;
 318                         if (dn.max_level == 0)
 319                                 goto out;
 320                         goto next;
 321                 }
 322
 323                 blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_page, cow_inode),
 324                                 len);
 325                 index = off;
 326                 for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) {
 327                         blkaddr = f2fs_data_blkaddr(&dn);
 328
 329                         if (!__is_valid_data_blkaddr(blkaddr)) {
 330                                 continue;
 331                         } else if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
 332                                         DATA_GENERIC_ENHANCE)) {
 333                                 f2fs_put_dnode(&dn);
 334                                 ret = -EFSCORRUPTED;
 335                                 f2fs_handle_error(sbi,
 336                                                 ERROR_INVALID_BLKADDR);
 337                                 goto out;
 338                         }
 339
 340                         new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS,
 341                                                         true, NULL);
 342
 343                         ret = __replace_atomic_write_block(inode, index, blkaddr,
 344                                                         &new->old_addr, false);
 345                         if (ret) {
 346                                 f2fs_put_dnode(&dn);
 347                                 kmem_cache_free(revoke_entry_slab, new);
 348                                 goto out;
 349                         }
 350
 351                         f2fs_update_data_blkaddr(&dn, NULL_ADDR);
 352                         new->index = index;
 353                         list_add_tail(&new->list, &revoke_list);
 354                 }
 355                 f2fs_put_dnode(&dn);
 356 next:
 357                 off += blen;
 358                 len -= blen;
 359         }
 360
 361 out:
 362         if (ret) {
 363                 sbi->revoked_atomic_block += fi->atomic_write_cnt;
 364         } else {
 365                 sbi->committed_atomic_block += fi->atomic_write_cnt;
 366                 set_inode_flag(inode, FI_ATOMIC_COMMITTED);
 367         }
 368
 369         __complete_revoke_list(inode, &revoke_list, ret ? true : false);
 370
 371         return ret;
 372 }
 373
 374 int f2fs_commit_atomic_write(struct inode *inode)
 375 {
 376         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 377         struct f2fs_inode_info *fi = F2FS_I(inode);
 378         int err;
 379
 380         err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX);
 381         if (err)
 382                 return err;
 383
 384         f2fs_down_write(&fi->i_gc_rwsem[WRITE]);
 385         f2fs_lock_op(sbi);
 386
 387         err = __f2fs_commit_atomic_write(inode);
 388
 389         f2fs_unlock_op(sbi);
 390         f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
 391
 392         return err;
 393 }
 394
 395 /*
 396  * This function balances dirty node and dentry pages.
 397  * In addition, it controls garbage collection.
 398  */
 399 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 400 {
 401         if (time_to_inject(sbi, FAULT_CHECKPOINT))
 402                 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT);
 403
 404         /* balance_fs_bg is able to be pending */
 405         if (need && excess_cached_nats(sbi))
 406                 f2fs_balance_fs_bg(sbi, false);
 407
 408         if (!f2fs_is_checkpoint_ready(sbi))
 409                 return;
 410
 411         /*
 412          * We should do GC or end up with checkpoint, if there are so many dirty
 413          * dir/node pages without enough free segments.
 414          */
 415         if (has_enough_free_secs(sbi, 0, 0))
 416                 return;
 417
 418         if (test_opt(sbi, GC_MERGE) && sbi->gc_thread &&
 419                                 sbi->gc_thread->f2fs_gc_task) {
 420                 DEFINE_WAIT(wait);
 421
 422                 prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait,
 423                                         TASK_UNINTERRUPTIBLE);
 424                 wake_up(&sbi->gc_thread->gc_wait_queue_head);
 425                 io_schedule();
 426                 finish_wait(&sbi->gc_thread->fggc_wq, &wait);
 427         } else {
 428                 struct f2fs_gc_control gc_control = {
 429                         .victim_segno = NULL_SEGNO,
 430                         .init_gc_type = BG_GC,
 431                         .no_bg_gc = true,
 432                         .should_migrate_blocks = false,
 433                         .err_gc_skipped = false,
 434                         .nr_free_secs = 1 };
 435                 f2fs_down_write(&sbi->gc_lock);
 436                 f2fs_gc(sbi, &gc_control);
 437         }
 438 }
 439
 440 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi)
 441 {
 442         int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2;
 443         unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS);
 444         unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA);
 445         unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES);
 446         unsigned int meta = get_pages(sbi, F2FS_DIRTY_META);
 447         unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA);
 448         unsigned int threshold = sbi->blocks_per_seg * factor *
 449                                         DEFAULT_DIRTY_THRESHOLD;
 450         unsigned int global_threshold = threshold * 3 / 2;
 451
 452         if (dents >= threshold || qdata >= threshold ||
 453                 nodes >= threshold || meta >= threshold ||
 454                 imeta >= threshold)
 455                 return true;
 456         return dents + qdata + nodes + meta + imeta >  global_threshold;
 457 }
 458
 459 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
 460 {
 461         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 462                 return;
 463
 464         /* try to shrink extent cache when there is no enough memory */
 465         if (!f2fs_available_free_memory(sbi, READ_EXTENT_CACHE))
 466                 f2fs_shrink_read_extent_tree(sbi,
 467                                 READ_EXTENT_CACHE_SHRINK_NUMBER);
 468
 469         /* try to shrink age extent cache when there is no enough memory */
 470         if (!f2fs_available_free_memory(sbi, AGE_EXTENT_CACHE))
 471                 f2fs_shrink_age_extent_tree(sbi,
 472                                 AGE_EXTENT_CACHE_SHRINK_NUMBER);
 473
 474         /* check the # of cached NAT entries */
 475         if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
 476                 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 477
 478         if (!f2fs_available_free_memory(sbi, FREE_NIDS))
 479                 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
 480         else
 481                 f2fs_build_free_nids(sbi, false, false);
 482
 483         if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) ||
 484                 excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi))
 485                 goto do_sync;
 486
 487         /* there is background inflight IO or foreground operation recently */
 488         if (is_inflight_io(sbi, REQ_TIME) ||
 489                 (!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem)))
 490                 return;
 491
 492         /* exceed periodical checkpoint timeout threshold */
 493         if (f2fs_time_over(sbi, CP_TIME))
 494                 goto do_sync;
 495
 496         /* checkpoint is the only way to shrink partial cached entries */
 497         if (f2fs_available_free_memory(sbi, NAT_ENTRIES) &&
 498                 f2fs_available_free_memory(sbi, INO_ENTRIES))
 499                 return;
 500
 501 do_sync:
 502         if (test_opt(sbi, DATA_FLUSH) && from_bg) {
 503                 struct blk_plug plug;
 504
 505                 mutex_lock(&sbi->flush_lock);
 506
 507                 blk_start_plug(&plug);
 508                 f2fs_sync_dirty_inodes(sbi, FILE_INODE, false);
 509                 blk_finish_plug(&plug);
 510
 511                 mutex_unlock(&sbi->flush_lock);
 512         }
 513         f2fs_sync_fs(sbi->sb, 1);
 514         stat_inc_bg_cp_count(sbi->stat_info);
 515 }
 516
 517 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 518                                 struct block_device *bdev)
 519 {
 520         int ret = blkdev_issue_flush(bdev);
 521
 522         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 523                                 test_opt(sbi, FLUSH_MERGE), ret);
 524         if (!ret)
 525                 f2fs_update_iostat(sbi, NULL, FS_FLUSH_IO, 0);
 526         return ret;
 527 }
 528
 529 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 530 {
 531         int ret = 0;
 532         int i;
 533
 534         if (!f2fs_is_multi_device(sbi))
 535                 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 536
 537         for (i = 0; i < sbi->s_ndevs; i++) {
 538                 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
 539                         continue;
 540                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 541                 if (ret)
 542                         break;
 543         }
 544         return ret;
 545 }
 546
 547 static int issue_flush_thread(void *data)
 548 {
 549         struct f2fs_sb_info *sbi = data;
 550         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 551         wait_queue_head_t *q = &fcc->flush_wait_queue;
 552 repeat:
 553         if (kthread_should_stop())
 554                 return 0;
 555
 556         if (!llist_empty(&fcc->issue_list)) {
 557                 struct flush_cmd *cmd, *next;
 558                 int ret;
 559
 560                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 561                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 562
 563                 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 564
 565                 ret = submit_flush_wait(sbi, cmd->ino);
 566                 atomic_inc(&fcc->issued_flush);
 567
 568                 llist_for_each_entry_safe(cmd, next,
 569                                           fcc->dispatch_list, llnode) {
 570                         cmd->ret = ret;
 571                         complete(&cmd->wait);
 572                 }
 573                 fcc->dispatch_list = NULL;
 574         }
 575
 576         wait_event_interruptible(*q,
 577                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 578         goto repeat;
 579 }
 580
 581 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 582 {
 583         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 584         struct flush_cmd cmd;
 585         int ret;
 586
 587         if (test_opt(sbi, NOBARRIER))
 588                 return 0;
 589
 590         if (!test_opt(sbi, FLUSH_MERGE)) {
 591                 atomic_inc(&fcc->queued_flush);
 592                 ret = submit_flush_wait(sbi, ino);
 593                 atomic_dec(&fcc->queued_flush);
 594                 atomic_inc(&fcc->issued_flush);
 595                 return ret;
 596         }
 597
 598         if (atomic_inc_return(&fcc->queued_flush) == 1 ||
 599             f2fs_is_multi_device(sbi)) {
 600                 ret = submit_flush_wait(sbi, ino);
 601                 atomic_dec(&fcc->queued_flush);
 602
 603                 atomic_inc(&fcc->issued_flush);
 604                 return ret;
 605         }
 606
 607         cmd.ino = ino;
 608         init_completion(&cmd.wait);
 609
 610         llist_add(&cmd.llnode, &fcc->issue_list);
 611
 612         /*
 613          * update issue_list before we wake up issue_flush thread, this
 614          * smp_mb() pairs with another barrier in ___wait_event(), see
 615          * more details in comments of waitqueue_active().
 616          */
 617         smp_mb();
 618
 619         if (waitqueue_active(&fcc->flush_wait_queue))
 620                 wake_up(&fcc->flush_wait_queue);
 621
 622         if (fcc->f2fs_issue_flush) {
 623                 wait_for_completion(&cmd.wait);
 624                 atomic_dec(&fcc->queued_flush);
 625         } else {
 626                 struct llist_node *list;
 627
 628                 list = llist_del_all(&fcc->issue_list);
 629                 if (!list) {
 630                         wait_for_completion(&cmd.wait);
 631                         atomic_dec(&fcc->queued_flush);
 632                 } else {
 633                         struct flush_cmd *tmp, *next;
 634
 635                         ret = submit_flush_wait(sbi, ino);
 636
 637                         llist_for_each_entry_safe(tmp, next, list, llnode) {
 638                                 if (tmp == &cmd) {
 639                                         cmd.ret = ret;
 640                                         atomic_dec(&fcc->queued_flush);
 641                                         continue;
 642                                 }
 643                                 tmp->ret = ret;
 644                                 complete(&tmp->wait);
 645                         }
 646                 }
 647         }
 648
 649         return cmd.ret;
 650 }
 651
 652 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
 653 {
 654         dev_t dev = sbi->sb->s_bdev->bd_dev;
 655         struct flush_cmd_control *fcc;
 656
 657         if (SM_I(sbi)->fcc_info) {
 658                 fcc = SM_I(sbi)->fcc_info;
 659                 if (fcc->f2fs_issue_flush)
 660                         return 0;
 661                 goto init_thread;
 662         }
 663
 664         fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
 665         if (!fcc)
 666                 return -ENOMEM;
 667         atomic_set(&fcc->issued_flush, 0);
 668         atomic_set(&fcc->queued_flush, 0);
 669         init_waitqueue_head(&fcc->flush_wait_queue);
 670         init_llist_head(&fcc->issue_list);
 671         SM_I(sbi)->fcc_info = fcc;
 672         if (!test_opt(sbi, FLUSH_MERGE))
 673                 return 0;
 674
 675 init_thread:
 676         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 677                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 678         if (IS_ERR(fcc->f2fs_issue_flush)) {
 679                 int err = PTR_ERR(fcc->f2fs_issue_flush);
 680
 681                 fcc->f2fs_issue_flush = NULL;
 682                 return err;
 683         }
 684
 685         return 0;
 686 }
 687
 688 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 689 {
 690         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 691
 692         if (fcc && fcc->f2fs_issue_flush) {
 693                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 694
 695                 fcc->f2fs_issue_flush = NULL;
 696                 kthread_stop(flush_thread);
 697         }
 698         if (free) {
 699                 kfree(fcc);
 700                 SM_I(sbi)->fcc_info = NULL;
 701         }
 702 }
 703
 704 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
 705 {
 706         int ret = 0, i;
 707
 708         if (!f2fs_is_multi_device(sbi))
 709                 return 0;
 710
 711         if (test_opt(sbi, NOBARRIER))
 712                 return 0;
 713
 714         for (i = 1; i < sbi->s_ndevs; i++) {
 715                 int count = DEFAULT_RETRY_IO_COUNT;
 716
 717                 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
 718                         continue;
 719
 720                 do {
 721                         ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 722                         if (ret)
 723                                 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
 724                 } while (ret && --count);
 725
 726                 if (ret) {
 727                         f2fs_stop_checkpoint(sbi, false,
 728                                         STOP_CP_REASON_FLUSH_FAIL);
 729                         break;
 730                 }
 731
 732                 spin_lock(&sbi->dev_lock);
 733                 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
 734                 spin_unlock(&sbi->dev_lock);
 735         }
 736
 737         return ret;
 738 }
 739
 740 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 741                 enum dirty_type dirty_type)
 742 {
 743         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 744
 745         /* need not be added */
 746         if (IS_CURSEG(sbi, segno))
 747                 return;
 748
 749         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 750                 dirty_i->nr_dirty[dirty_type]++;
 751
 752         if (dirty_type == DIRTY) {
 753                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 754                 enum dirty_type t = sentry->type;
 755
 756                 if (unlikely(t >= DIRTY)) {
 757                         f2fs_bug_on(sbi, 1);
 758                         return;
 759                 }
 760                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 761                         dirty_i->nr_dirty[t]++;
 762
 763                 if (__is_large_section(sbi)) {
 764                         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 765                         block_t valid_blocks =
 766                                 get_valid_blocks(sbi, segno, true);
 767
 768                         f2fs_bug_on(sbi, unlikely(!valid_blocks ||
 769                                         valid_blocks == CAP_BLKS_PER_SEC(sbi)));
 770
 771                         if (!IS_CURSEC(sbi, secno))
 772                                 set_bit(secno, dirty_i->dirty_secmap);
 773                 }
 774         }
 775 }
 776
 777 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 778                 enum dirty_type dirty_type)
 779 {
 780         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 781         block_t valid_blocks;
 782
 783         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 784                 dirty_i->nr_dirty[dirty_type]--;
 785
 786         if (dirty_type == DIRTY) {
 787                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 788                 enum dirty_type t = sentry->type;
 789
 790                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 791                         dirty_i->nr_dirty[t]--;
 792
 793                 valid_blocks = get_valid_blocks(sbi, segno, true);
 794                 if (valid_blocks == 0) {
 795                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 796                                                 dirty_i->victim_secmap);
 797 #ifdef CONFIG_F2FS_CHECK_FS
 798                         clear_bit(segno, SIT_I(sbi)->invalid_segmap);
 799 #endif
 800                 }
 801                 if (__is_large_section(sbi)) {
 802                         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 803
 804                         if (!valid_blocks ||
 805                                         valid_blocks == CAP_BLKS_PER_SEC(sbi)) {
 806                                 clear_bit(secno, dirty_i->dirty_secmap);
 807                                 return;
 808                         }
 809
 810                         if (!IS_CURSEC(sbi, secno))
 811                                 set_bit(secno, dirty_i->dirty_secmap);
 812                 }
 813         }
 814 }
 815
 816 /*
 817  * Should not occur error such as -ENOMEM.
 818  * Adding dirty entry into seglist is not critical operation.
 819  * If a given segment is one of current working segments, it won't be added.
 820  */
 821 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 822 {
 823         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 824         unsigned short valid_blocks, ckpt_valid_blocks;
 825         unsigned int usable_blocks;
 826
 827         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 828                 return;
 829
 830         usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
 831         mutex_lock(&dirty_i->seglist_lock);
 832
 833         valid_blocks = get_valid_blocks(sbi, segno, false);
 834         ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false);
 835
 836         if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
 837                 ckpt_valid_blocks == usable_blocks)) {
 838                 __locate_dirty_segment(sbi, segno, PRE);
 839                 __remove_dirty_segment(sbi, segno, DIRTY);
 840         } else if (valid_blocks < usable_blocks) {
 841                 __locate_dirty_segment(sbi, segno, DIRTY);
 842         } else {
 843                 /* Recovery routine with SSR needs this */
 844                 __remove_dirty_segment(sbi, segno, DIRTY);
 845         }
 846
 847         mutex_unlock(&dirty_i->seglist_lock);
 848 }
 849
 850 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
 851 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
 852 {
 853         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 854         unsigned int segno;
 855
 856         mutex_lock(&dirty_i->seglist_lock);
 857         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 858                 if (get_valid_blocks(sbi, segno, false))
 859                         continue;
 860                 if (IS_CURSEG(sbi, segno))
 861                         continue;
 862                 __locate_dirty_segment(sbi, segno, PRE);
 863                 __remove_dirty_segment(sbi, segno, DIRTY);
 864         }
 865         mutex_unlock(&dirty_i->seglist_lock);
 866 }
 867
 868 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
 869 {
 870         int ovp_hole_segs =
 871                 (overprovision_segments(sbi) - reserved_segments(sbi));
 872         block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
 873         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 874         block_t holes[2] = {0, 0};      /* DATA and NODE */
 875         block_t unusable;
 876         struct seg_entry *se;
 877         unsigned int segno;
 878
 879         mutex_lock(&dirty_i->seglist_lock);
 880         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 881                 se = get_seg_entry(sbi, segno);
 882                 if (IS_NODESEG(se->type))
 883                         holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
 884                                                         se->valid_blocks;
 885                 else
 886                         holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
 887                                                         se->valid_blocks;
 888         }
 889         mutex_unlock(&dirty_i->seglist_lock);
 890
 891         unusable = max(holes[DATA], holes[NODE]);
 892         if (unusable > ovp_holes)
 893                 return unusable - ovp_holes;
 894         return 0;
 895 }
 896
 897 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
 898 {
 899         int ovp_hole_segs =
 900                 (overprovision_segments(sbi) - reserved_segments(sbi));
 901         if (unusable > F2FS_OPTION(sbi).unusable_cap)
 902                 return -EAGAIN;
 903         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
 904                 dirty_segments(sbi) > ovp_hole_segs)
 905                 return -EAGAIN;
 906         return 0;
 907 }
 908
 909 /* This is only used by SBI_CP_DISABLED */
 910 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
 911 {
 912         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 913         unsigned int segno = 0;
 914
 915         mutex_lock(&dirty_i->seglist_lock);
 916         for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
 917                 if (get_valid_blocks(sbi, segno, false))
 918                         continue;
 919                 if (get_ckpt_valid_blocks(sbi, segno, false))
 920                         continue;
 921                 mutex_unlock(&dirty_i->seglist_lock);
 922                 return segno;
 923         }
 924         mutex_unlock(&dirty_i->seglist_lock);
 925         return NULL_SEGNO;
 926 }
 927
 928 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 929                 struct block_device *bdev, block_t lstart,
 930                 block_t start, block_t len)
 931 {
 932         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 933         struct list_head *pend_list;
 934         struct discard_cmd *dc;
 935
 936         f2fs_bug_on(sbi, !len);
 937
 938         pend_list = &dcc->pend_list[plist_idx(len)];
 939
 940         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL);
 941         INIT_LIST_HEAD(&dc->list);
 942         dc->bdev = bdev;
 943         dc->di.lstart = lstart;
 944         dc->di.start = start;
 945         dc->di.len = len;
 946         dc->ref = 0;
 947         dc->state = D_PREP;
 948         dc->queued = 0;
 949         dc->error = 0;
 950         init_completion(&dc->wait);
 951         list_add_tail(&dc->list, pend_list);
 952         spin_lock_init(&dc->lock);
 953         dc->bio_ref = 0;
 954         atomic_inc(&dcc->discard_cmd_cnt);
 955         dcc->undiscard_blks += len;
 956
 957         return dc;
 958 }
 959
 960 static bool f2fs_check_discard_tree(struct f2fs_sb_info *sbi)
 961 {
 962 #ifdef CONFIG_F2FS_CHECK_FS
 963         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 964         struct rb_node *cur = rb_first_cached(&dcc->root), *next;
 965         struct discard_cmd *cur_dc, *next_dc;
 966
 967         while (cur) {
 968                 next = rb_next(cur);
 969                 if (!next)
 970                         return true;
 971
 972                 cur_dc = rb_entry(cur, struct discard_cmd, rb_node);
 973                 next_dc = rb_entry(next, struct discard_cmd, rb_node);
 974
 975                 if (cur_dc->di.lstart + cur_dc->di.len > next_dc->di.lstart) {
 976                         f2fs_info(sbi, "broken discard_rbtree, "
 977                                 "cur(%u, %u) next(%u, %u)",
 978                                 cur_dc->di.lstart, cur_dc->di.len,
 979                                 next_dc->di.lstart, next_dc->di.len);
 980                         return false;
 981                 }
 982                 cur = next;
 983         }
 984 #endif
 985         return true;
 986 }
 987
 988 static struct discard_cmd *__lookup_discard_cmd(struct f2fs_sb_info *sbi,
 989                                                 block_t blkaddr)
 990 {
 991         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 992         struct rb_node *node = dcc->root.rb_root.rb_node;
 993         struct discard_cmd *dc;
 994
 995         while (node) {
 996                 dc = rb_entry(node, struct discard_cmd, rb_node);
 997
 998                 if (blkaddr < dc->di.lstart)
 999                         node = node->rb_left;
1000                 else if (blkaddr >= dc->di.lstart + dc->di.len)
1001                         node = node->rb_right;
1002                 else
1003                         return dc;
1004         }
1005         return NULL;
1006 }
1007
1008 static struct discard_cmd *__lookup_discard_cmd_ret(struct rb_root_cached *root,
1009                                 block_t blkaddr,
1010                                 struct discard_cmd **prev_entry,
1011                                 struct discard_cmd **next_entry,
1012                                 struct rb_node ***insert_p,
1013                                 struct rb_node **insert_parent)
1014 {
1015         struct rb_node **pnode = &root->rb_root.rb_node;
1016         struct rb_node *parent = NULL, *tmp_node;
1017         struct discard_cmd *dc;
1018
1019         *insert_p = NULL;
1020         *insert_parent = NULL;
1021         *prev_entry = NULL;
1022         *next_entry = NULL;
1023
1024         if (RB_EMPTY_ROOT(&root->rb_root))
1025                 return NULL;
1026
1027         while (*pnode) {
1028                 parent = *pnode;
1029                 dc = rb_entry(*pnode, struct discard_cmd, rb_node);
1030
1031                 if (blkaddr < dc->di.lstart)
1032                         pnode = &(*pnode)->rb_left;
1033                 else if (blkaddr >= dc->di.lstart + dc->di.len)
1034                         pnode = &(*pnode)->rb_right;
1035                 else
1036                         goto lookup_neighbors;
1037         }
1038
1039         *insert_p = pnode;
1040         *insert_parent = parent;
1041
1042         dc = rb_entry(parent, struct discard_cmd, rb_node);
1043         tmp_node = parent;
1044         if (parent && blkaddr > dc->di.lstart)
1045                 tmp_node = rb_next(parent);
1046         *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1047
1048         tmp_node = parent;
1049         if (parent && blkaddr < dc->di.lstart)
1050                 tmp_node = rb_prev(parent);
1051         *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1052         return NULL;
1053
1054 lookup_neighbors:
1055         /* lookup prev node for merging backward later */
1056         tmp_node = rb_prev(&dc->rb_node);
1057         *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1058
1059         /* lookup next node for merging frontward later */
1060         tmp_node = rb_next(&dc->rb_node);
1061         *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1062         return dc;
1063 }
1064
1065 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1066                                                         struct discard_cmd *dc)
1067 {
1068         if (dc->state == D_DONE)
1069                 atomic_sub(dc->queued, &dcc->queued_discard);
1070
1071         list_del(&dc->list);
1072         rb_erase_cached(&dc->rb_node, &dcc->root);
1073         dcc->undiscard_blks -= dc->di.len;
1074
1075         kmem_cache_free(discard_cmd_slab, dc);
1076
1077         atomic_dec(&dcc->discard_cmd_cnt);
1078 }
1079
1080 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1081                                                         struct discard_cmd *dc)
1082 {
1083         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1084         unsigned long flags;
1085
1086         trace_f2fs_remove_discard(dc->bdev, dc->di.start, dc->di.len);
1087
1088         spin_lock_irqsave(&dc->lock, flags);
1089         if (dc->bio_ref) {
1090                 spin_unlock_irqrestore(&dc->lock, flags);
1091                 return;
1092         }
1093         spin_unlock_irqrestore(&dc->lock, flags);
1094
1095         f2fs_bug_on(sbi, dc->ref);
1096
1097         if (dc->error == -EOPNOTSUPP)
1098                 dc->error = 0;
1099
1100         if (dc->error)
1101                 printk_ratelimited(
1102                         "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
1103                         KERN_INFO, sbi->sb->s_id,
1104                         dc->di.lstart, dc->di.start, dc->di.len, dc->error);
1105         __detach_discard_cmd(dcc, dc);
1106 }
1107
1108 static void f2fs_submit_discard_endio(struct bio *bio)
1109 {
1110         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1111         unsigned long flags;
1112
1113         spin_lock_irqsave(&dc->lock, flags);
1114         if (!dc->error)
1115                 dc->error = blk_status_to_errno(bio->bi_status);
1116         dc->bio_ref--;
1117         if (!dc->bio_ref && dc->state == D_SUBMIT) {
1118                 dc->state = D_DONE;
1119                 complete_all(&dc->wait);
1120         }
1121         spin_unlock_irqrestore(&dc->lock, flags);
1122         bio_put(bio);
1123 }
1124
1125 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1126                                 block_t start, block_t end)
1127 {
1128 #ifdef CONFIG_F2FS_CHECK_FS
1129         struct seg_entry *sentry;
1130         unsigned int segno;
1131         block_t blk = start;
1132         unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1133         unsigned long *map;
1134
1135         while (blk < end) {
1136                 segno = GET_SEGNO(sbi, blk);
1137                 sentry = get_seg_entry(sbi, segno);
1138                 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1139
1140                 if (end < START_BLOCK(sbi, segno + 1))
1141                         size = GET_BLKOFF_FROM_SEG0(sbi, end);
1142                 else
1143                         size = max_blocks;
1144                 map = (unsigned long *)(sentry->cur_valid_map);
1145                 offset = __find_rev_next_bit(map, size, offset);
1146                 f2fs_bug_on(sbi, offset != size);
1147                 blk = START_BLOCK(sbi, segno + 1);
1148         }
1149 #endif
1150 }
1151
1152 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1153                                 struct discard_policy *dpolicy,
1154                                 int discard_type, unsigned int granularity)
1155 {
1156         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1157
1158         /* common policy */
1159         dpolicy->type = discard_type;
1160         dpolicy->sync = true;
1161         dpolicy->ordered = false;
1162         dpolicy->granularity = granularity;
1163
1164         dpolicy->max_requests = dcc->max_discard_request;
1165         dpolicy->io_aware_gran = dcc->discard_io_aware_gran;
1166         dpolicy->timeout = false;
1167
1168         if (discard_type == DPOLICY_BG) {
1169                 dpolicy->min_interval = dcc->min_discard_issue_time;
1170                 dpolicy->mid_interval = dcc->mid_discard_issue_time;
1171                 dpolicy->max_interval = dcc->max_discard_issue_time;
1172                 dpolicy->io_aware = true;
1173                 dpolicy->sync = false;
1174                 dpolicy->ordered = true;
1175                 if (utilization(sbi) > dcc->discard_urgent_util) {
1176                         dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1177                         if (atomic_read(&dcc->discard_cmd_cnt))
1178                                 dpolicy->max_interval =
1179                                         dcc->min_discard_issue_time;
1180                 }
1181         } else if (discard_type == DPOLICY_FORCE) {
1182                 dpolicy->min_interval = dcc->min_discard_issue_time;
1183                 dpolicy->mid_interval = dcc->mid_discard_issue_time;
1184                 dpolicy->max_interval = dcc->max_discard_issue_time;
1185                 dpolicy->io_aware = false;
1186         } else if (discard_type == DPOLICY_FSTRIM) {
1187                 dpolicy->io_aware = false;
1188         } else if (discard_type == DPOLICY_UMOUNT) {
1189                 dpolicy->io_aware = false;
1190                 /* we need to issue all to keep CP_TRIMMED_FLAG */
1191                 dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1192                 dpolicy->timeout = true;
1193         }
1194 }
1195
1196 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1197                                 struct block_device *bdev, block_t lstart,
1198                                 block_t start, block_t len);
1199 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1200 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1201                                 struct discard_policy *dpolicy,
1202                                 struct discard_cmd *dc, int *issued)
1203 {
1204         struct block_device *bdev = dc->bdev;
1205         unsigned int max_discard_blocks =
1206                         SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1207         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1208         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1209                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1210         blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0;
1211         block_t lstart, start, len, total_len;
1212         int err = 0;
1213
1214         if (dc->state != D_PREP)
1215                 return 0;
1216
1217         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1218                 return 0;
1219
1220         trace_f2fs_issue_discard(bdev, dc->di.start, dc->di.len);
1221
1222         lstart = dc->di.lstart;
1223         start = dc->di.start;
1224         len = dc->di.len;
1225         total_len = len;
1226
1227         dc->di.len = 0;
1228
1229         while (total_len && *issued < dpolicy->max_requests && !err) {
1230                 struct bio *bio = NULL;
1231                 unsigned long flags;
1232                 bool last = true;
1233
1234                 if (len > max_discard_blocks) {
1235                         len = max_discard_blocks;
1236                         last = false;
1237                 }
1238
1239                 (*issued)++;
1240                 if (*issued == dpolicy->max_requests)
1241                         last = true;
1242
1243                 dc->di.len += len;
1244
1245                 if (time_to_inject(sbi, FAULT_DISCARD)) {
1246                         err = -EIO;
1247                 } else {
1248                         err = __blkdev_issue_discard(bdev,
1249                                         SECTOR_FROM_BLOCK(start),
1250                                         SECTOR_FROM_BLOCK(len),
1251                                         GFP_NOFS, &bio);
1252                 }
1253                 if (err) {
1254                         spin_lock_irqsave(&dc->lock, flags);
1255                         if (dc->state == D_PARTIAL)
1256                                 dc->state = D_SUBMIT;
1257                         spin_unlock_irqrestore(&dc->lock, flags);
1258
1259                         break;
1260                 }
1261
1262                 f2fs_bug_on(sbi, !bio);
1263
1264                 /*
1265                  * should keep before submission to avoid D_DONE
1266                  * right away
1267                  */
1268                 spin_lock_irqsave(&dc->lock, flags);
1269                 if (last)
1270                         dc->state = D_SUBMIT;
1271                 else
1272                         dc->state = D_PARTIAL;
1273                 dc->bio_ref++;
1274                 spin_unlock_irqrestore(&dc->lock, flags);
1275
1276                 atomic_inc(&dcc->queued_discard);
1277                 dc->queued++;
1278                 list_move_tail(&dc->list, wait_list);
1279
1280                 /* sanity check on discard range */
1281                 __check_sit_bitmap(sbi, lstart, lstart + len);
1282
1283                 bio->bi_private = dc;
1284                 bio->bi_end_io = f2fs_submit_discard_endio;
1285                 bio->bi_opf |= flag;
1286                 submit_bio(bio);
1287
1288                 atomic_inc(&dcc->issued_discard);
1289
1290                 f2fs_update_iostat(sbi, NULL, FS_DISCARD_IO, len * F2FS_BLKSIZE);
1291
1292                 lstart += len;
1293                 start += len;
1294                 total_len -= len;
1295                 len = total_len;
1296         }
1297
1298         if (!err && len) {
1299                 dcc->undiscard_blks -= len;
1300                 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1301         }
1302         return err;
1303 }
1304
1305 static void __insert_discard_cmd(struct f2fs_sb_info *sbi,
1306                                 struct block_device *bdev, block_t lstart,
1307                                 block_t start, block_t len)
1308 {
1309         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1310         struct rb_node **p = &dcc->root.rb_root.rb_node;
1311         struct rb_node *parent = NULL;
1312         struct discard_cmd *dc;
1313         bool leftmost = true;
1314
1315         /* look up rb tree to find parent node */
1316         while (*p) {
1317                 parent = *p;
1318                 dc = rb_entry(parent, struct discard_cmd, rb_node);
1319
1320                 if (lstart < dc->di.lstart) {
1321                         p = &(*p)->rb_left;
1322                 } else if (lstart >= dc->di.lstart + dc->di.len) {
1323                         p = &(*p)->rb_right;
1324                         leftmost = false;
1325                 } else {
1326                         f2fs_bug_on(sbi, 1);
1327                 }
1328         }
1329
1330         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1331
1332         rb_link_node(&dc->rb_node, parent, p);
1333         rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
1334 }
1335
1336 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1337                                                 struct discard_cmd *dc)
1338 {
1339         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->di.len)]);
1340 }
1341
1342 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1343                                 struct discard_cmd *dc, block_t blkaddr)
1344 {
1345         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1346         struct discard_info di = dc->di;
1347         bool modified = false;
1348
1349         if (dc->state == D_DONE || dc->di.len == 1) {
1350                 __remove_discard_cmd(sbi, dc);
1351                 return;
1352         }
1353
1354         dcc->undiscard_blks -= di.len;
1355
1356         if (blkaddr > di.lstart) {
1357                 dc->di.len = blkaddr - dc->di.lstart;
1358                 dcc->undiscard_blks += dc->di.len;
1359                 __relocate_discard_cmd(dcc, dc);
1360                 modified = true;
1361         }
1362
1363         if (blkaddr < di.lstart + di.len - 1) {
1364                 if (modified) {
1365                         __insert_discard_cmd(sbi, dc->bdev, blkaddr + 1,
1366                                         di.start + blkaddr + 1 - di.lstart,
1367                                         di.lstart + di.len - 1 - blkaddr);
1368                 } else {
1369                         dc->di.lstart++;
1370                         dc->di.len--;
1371                         dc->di.start++;
1372                         dcc->undiscard_blks += dc->di.len;
1373                         __relocate_discard_cmd(dcc, dc);
1374                 }
1375         }
1376 }
1377
1378 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1379                                 struct block_device *bdev, block_t lstart,
1380                                 block_t start, block_t len)
1381 {
1382         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1383         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1384         struct discard_cmd *dc;
1385         struct discard_info di = {0};
1386         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1387         unsigned int max_discard_blocks =
1388                         SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1389         block_t end = lstart + len;
1390
1391         dc = __lookup_discard_cmd_ret(&dcc->root, lstart,
1392                                 &prev_dc, &next_dc, &insert_p, &insert_parent);
1393         if (dc)
1394                 prev_dc = dc;
1395
1396         if (!prev_dc) {
1397                 di.lstart = lstart;
1398                 di.len = next_dc ? next_dc->di.lstart - lstart : len;
1399                 di.len = min(di.len, len);
1400                 di.start = start;
1401         }
1402
1403         while (1) {
1404                 struct rb_node *node;
1405                 bool merged = false;
1406                 struct discard_cmd *tdc = NULL;
1407
1408                 if (prev_dc) {
1409                         di.lstart = prev_dc->di.lstart + prev_dc->di.len;
1410                         if (di.lstart < lstart)
1411                                 di.lstart = lstart;
1412                         if (di.lstart >= end)
1413                                 break;
1414
1415                         if (!next_dc || next_dc->di.lstart > end)
1416                                 di.len = end - di.lstart;
1417                         else
1418                                 di.len = next_dc->di.lstart - di.lstart;
1419                         di.start = start + di.lstart - lstart;
1420                 }
1421
1422                 if (!di.len)
1423                         goto next;
1424
1425                 if (prev_dc && prev_dc->state == D_PREP &&
1426                         prev_dc->bdev == bdev &&
1427                         __is_discard_back_mergeable(&di, &prev_dc->di,
1428                                                         max_discard_blocks)) {
1429                         prev_dc->di.len += di.len;
1430                         dcc->undiscard_blks += di.len;
1431                         __relocate_discard_cmd(dcc, prev_dc);
1432                         di = prev_dc->di;
1433                         tdc = prev_dc;
1434                         merged = true;
1435                 }
1436
1437                 if (next_dc && next_dc->state == D_PREP &&
1438                         next_dc->bdev == bdev &&
1439                         __is_discard_front_mergeable(&di, &next_dc->di,
1440                                                         max_discard_blocks)) {
1441                         next_dc->di.lstart = di.lstart;
1442                         next_dc->di.len += di.len;
1443                         next_dc->di.start = di.start;
1444                         dcc->undiscard_blks += di.len;
1445                         __relocate_discard_cmd(dcc, next_dc);
1446                         if (tdc)
1447                                 __remove_discard_cmd(sbi, tdc);
1448                         merged = true;
1449                 }
1450
1451                 if (!merged)
1452                         __insert_discard_cmd(sbi, bdev,
1453                                                 di.lstart, di.start, di.len);
1454  next:
1455                 prev_dc = next_dc;
1456                 if (!prev_dc)
1457                         break;
1458
1459                 node = rb_next(&prev_dc->rb_node);
1460                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1461         }
1462 }
1463
1464 static void __queue_discard_cmd(struct f2fs_sb_info *sbi,
1465                 struct block_device *bdev, block_t blkstart, block_t blklen)
1466 {
1467         block_t lblkstart = blkstart;
1468
1469         if (!f2fs_bdev_support_discard(bdev))
1470                 return;
1471
1472         trace_f2fs_queue_discard(bdev, blkstart, blklen);
1473
1474         if (f2fs_is_multi_device(sbi)) {
1475                 int devi = f2fs_target_device_index(sbi, blkstart);
1476
1477                 blkstart -= FDEV(devi).start_blk;
1478         }
1479         mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1480         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1481         mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1482 }
1483
1484 static void __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1485                 struct discard_policy *dpolicy, int *issued)
1486 {
1487         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1488         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1489         struct rb_node **insert_p = NULL, *insert_parent = NULL;
1490         struct discard_cmd *dc;
1491         struct blk_plug plug;
1492         bool io_interrupted = false;
1493
1494         mutex_lock(&dcc->cmd_lock);
1495         dc = __lookup_discard_cmd_ret(&dcc->root, dcc->next_pos,
1496                                 &prev_dc, &next_dc, &insert_p, &insert_parent);
1497         if (!dc)
1498                 dc = next_dc;
1499
1500         blk_start_plug(&plug);
1501
1502         while (dc) {
1503                 struct rb_node *node;
1504                 int err = 0;
1505
1506                 if (dc->state != D_PREP)
1507                         goto next;
1508
1509                 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1510                         io_interrupted = true;
1511                         break;
1512                 }
1513
1514                 dcc->next_pos = dc->di.lstart + dc->di.len;
1515                 err = __submit_discard_cmd(sbi, dpolicy, dc, issued);
1516
1517                 if (*issued >= dpolicy->max_requests)
1518                         break;
1519 next:
1520                 node = rb_next(&dc->rb_node);
1521                 if (err)
1522                         __remove_discard_cmd(sbi, dc);
1523                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1524         }
1525
1526         blk_finish_plug(&plug);
1527
1528         if (!dc)
1529                 dcc->next_pos = 0;
1530
1531         mutex_unlock(&dcc->cmd_lock);
1532
1533         if (!(*issued) && io_interrupted)
1534                 *issued = -1;
1535 }
1536 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1537                                         struct discard_policy *dpolicy);
1538
1539 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1540                                         struct discard_policy *dpolicy)
1541 {
1542         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1543         struct list_head *pend_list;
1544         struct discard_cmd *dc, *tmp;
1545         struct blk_plug plug;
1546         int i, issued;
1547         bool io_interrupted = false;
1548
1549         if (dpolicy->timeout)
1550                 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1551
1552 retry:
1553         issued = 0;
1554         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1555                 if (dpolicy->timeout &&
1556                                 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1557                         break;
1558
1559                 if (i + 1 < dpolicy->granularity)
1560                         break;
1561
1562                 if (i + 1 < dcc->max_ordered_discard && dpolicy->ordered) {
1563                         __issue_discard_cmd_orderly(sbi, dpolicy, &issued);
1564                         return issued;
1565                 }
1566
1567                 pend_list = &dcc->pend_list[i];
1568
1569                 mutex_lock(&dcc->cmd_lock);
1570                 if (list_empty(pend_list))
1571                         goto next;
1572                 if (unlikely(dcc->rbtree_check))
1573                         f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
1574                 blk_start_plug(&plug);
1575                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1576                         f2fs_bug_on(sbi, dc->state != D_PREP);
1577
1578                         if (dpolicy->timeout &&
1579                                 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1580                                 break;
1581
1582                         if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1583                                                 !is_idle(sbi, DISCARD_TIME)) {
1584                                 io_interrupted = true;
1585                                 break;
1586                         }
1587
1588                         __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1589
1590                         if (issued >= dpolicy->max_requests)
1591                                 break;
1592                 }
1593                 blk_finish_plug(&plug);
1594 next:
1595                 mutex_unlock(&dcc->cmd_lock);
1596
1597                 if (issued >= dpolicy->max_requests || io_interrupted)
1598                         break;
1599         }
1600
1601         if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1602                 __wait_all_discard_cmd(sbi, dpolicy);
1603                 goto retry;
1604         }
1605
1606         if (!issued && io_interrupted)
1607                 issued = -1;
1608
1609         return issued;
1610 }
1611
1612 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1613 {
1614         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1615         struct list_head *pend_list;
1616         struct discard_cmd *dc, *tmp;
1617         int i;
1618         bool dropped = false;
1619
1620         mutex_lock(&dcc->cmd_lock);
1621         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1622                 pend_list = &dcc->pend_list[i];
1623                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1624                         f2fs_bug_on(sbi, dc->state != D_PREP);
1625                         __remove_discard_cmd(sbi, dc);
1626                         dropped = true;
1627                 }
1628         }
1629         mutex_unlock(&dcc->cmd_lock);
1630
1631         return dropped;
1632 }
1633
1634 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1635 {
1636         __drop_discard_cmd(sbi);
1637 }
1638
1639 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1640                                                         struct discard_cmd *dc)
1641 {
1642         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1643         unsigned int len = 0;
1644
1645         wait_for_completion_io(&dc->wait);
1646         mutex_lock(&dcc->cmd_lock);
1647         f2fs_bug_on(sbi, dc->state != D_DONE);
1648         dc->ref--;
1649         if (!dc->ref) {
1650                 if (!dc->error)
1651                         len = dc->di.len;
1652                 __remove_discard_cmd(sbi, dc);
1653         }
1654         mutex_unlock(&dcc->cmd_lock);
1655
1656         return len;
1657 }
1658
1659 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1660                                                 struct discard_policy *dpolicy,
1661                                                 block_t start, block_t end)
1662 {
1663         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1664         struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1665                                         &(dcc->fstrim_list) : &(dcc->wait_list);
1666         struct discard_cmd *dc = NULL, *iter, *tmp;
1667         unsigned int trimmed = 0;
1668
1669 next:
1670         dc = NULL;
1671
1672         mutex_lock(&dcc->cmd_lock);
1673         list_for_each_entry_safe(iter, tmp, wait_list, list) {
1674                 if (iter->di.lstart + iter->di.len <= start ||
1675                                         end <= iter->di.lstart)
1676                         continue;
1677                 if (iter->di.len < dpolicy->granularity)
1678                         continue;
1679                 if (iter->state == D_DONE && !iter->ref) {
1680                         wait_for_completion_io(&iter->wait);
1681                         if (!iter->error)
1682                                 trimmed += iter->di.len;
1683                         __remove_discard_cmd(sbi, iter);
1684                 } else {
1685                         iter->ref++;
1686                         dc = iter;
1687                         break;
1688                 }
1689         }
1690         mutex_unlock(&dcc->cmd_lock);
1691
1692         if (dc) {
1693                 trimmed += __wait_one_discard_bio(sbi, dc);
1694                 goto next;
1695         }
1696
1697         return trimmed;
1698 }
1699
1700 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1701                                                 struct discard_policy *dpolicy)
1702 {
1703         struct discard_policy dp;
1704         unsigned int discard_blks;
1705
1706         if (dpolicy)
1707                 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1708
1709         /* wait all */
1710         __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, MIN_DISCARD_GRANULARITY);
1711         discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1712         __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, MIN_DISCARD_GRANULARITY);
1713         discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1714
1715         return discard_blks;
1716 }
1717
1718 /* This should be covered by global mutex, &sit_i->sentry_lock */
1719 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1720 {
1721         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1722         struct discard_cmd *dc;
1723         bool need_wait = false;
1724
1725         mutex_lock(&dcc->cmd_lock);
1726         dc = __lookup_discard_cmd(sbi, blkaddr);
1727         if (dc) {
1728                 if (dc->state == D_PREP) {
1729                         __punch_discard_cmd(sbi, dc, blkaddr);
1730                 } else {
1731                         dc->ref++;
1732                         need_wait = true;
1733                 }
1734         }
1735         mutex_unlock(&dcc->cmd_lock);
1736
1737         if (need_wait)
1738                 __wait_one_discard_bio(sbi, dc);
1739 }
1740
1741 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1742 {
1743         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1744
1745         if (dcc && dcc->f2fs_issue_discard) {
1746                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1747
1748                 dcc->f2fs_issue_discard = NULL;
1749                 kthread_stop(discard_thread);
1750         }
1751 }
1752
1753 /**
1754  * f2fs_issue_discard_timeout() - Issue all discard cmd within UMOUNT_DISCARD_TIMEOUT
1755  * @sbi: the f2fs_sb_info data for discard cmd to issue
1756  *
1757  * When UMOUNT_DISCARD_TIMEOUT is exceeded, all remaining discard commands will be dropped
1758  *
1759  * Return true if issued all discard cmd or no discard cmd need issue, otherwise return false.
1760  */
1761 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1762 {
1763         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1764         struct discard_policy dpolicy;
1765         bool dropped;
1766
1767         if (!atomic_read(&dcc->discard_cmd_cnt))
1768                 return true;
1769
1770         __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1771                                         dcc->discard_granularity);
1772         __issue_discard_cmd(sbi, &dpolicy);
1773         dropped = __drop_discard_cmd(sbi);
1774
1775         /* just to make sure there is no pending discard commands */
1776         __wait_all_discard_cmd(sbi, NULL);
1777
1778         f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1779         return !dropped;
1780 }
1781
1782 static int issue_discard_thread(void *data)
1783 {
1784         struct f2fs_sb_info *sbi = data;
1785         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1786         wait_queue_head_t *q = &dcc->discard_wait_queue;
1787         struct discard_policy dpolicy;
1788         unsigned int wait_ms = dcc->min_discard_issue_time;
1789         int issued;
1790
1791         set_freezable();
1792
1793         do {
1794                 wait_event_interruptible_timeout(*q,
1795                                 kthread_should_stop() || freezing(current) ||
1796                                 dcc->discard_wake,
1797                                 msecs_to_jiffies(wait_ms));
1798
1799                 if (sbi->gc_mode == GC_URGENT_HIGH ||
1800                         !f2fs_available_free_memory(sbi, DISCARD_CACHE))
1801                         __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE,
1802                                                 MIN_DISCARD_GRANULARITY);
1803                 else
1804                         __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1805                                                 dcc->discard_granularity);
1806
1807                 if (dcc->discard_wake)
1808                         dcc->discard_wake = false;
1809
1810                 /* clean up pending candidates before going to sleep */
1811                 if (atomic_read(&dcc->queued_discard))
1812                         __wait_all_discard_cmd(sbi, NULL);
1813
1814                 if (try_to_freeze())
1815                         continue;
1816                 if (f2fs_readonly(sbi->sb))
1817                         continue;
1818                 if (kthread_should_stop())
1819                         return 0;
1820                 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) ||
1821                         !atomic_read(&dcc->discard_cmd_cnt)) {
1822                         wait_ms = dpolicy.max_interval;
1823                         continue;
1824                 }
1825
1826                 sb_start_intwrite(sbi->sb);
1827
1828                 issued = __issue_discard_cmd(sbi, &dpolicy);
1829                 if (issued > 0) {
1830                         __wait_all_discard_cmd(sbi, &dpolicy);
1831                         wait_ms = dpolicy.min_interval;
1832                 } else if (issued == -1) {
1833                         wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1834                         if (!wait_ms)
1835                                 wait_ms = dpolicy.mid_interval;
1836                 } else {
1837                         wait_ms = dpolicy.max_interval;
1838                 }
1839                 if (!atomic_read(&dcc->discard_cmd_cnt))
1840                         wait_ms = dpolicy.max_interval;
1841
1842                 sb_end_intwrite(sbi->sb);
1843
1844         } while (!kthread_should_stop());
1845         return 0;
1846 }
1847
1848 #ifdef CONFIG_BLK_DEV_ZONED
1849 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1850                 struct block_device *bdev, block_t blkstart, block_t blklen)
1851 {
1852         sector_t sector, nr_sects;
1853         block_t lblkstart = blkstart;
1854         int devi = 0;
1855         u64 remainder = 0;
1856
1857         if (f2fs_is_multi_device(sbi)) {
1858                 devi = f2fs_target_device_index(sbi, blkstart);
1859                 if (blkstart < FDEV(devi).start_blk ||
1860                     blkstart > FDEV(devi).end_blk) {
1861                         f2fs_err(sbi, "Invalid block %x", blkstart);
1862                         return -EIO;
1863                 }
1864                 blkstart -= FDEV(devi).start_blk;
1865         }
1866
1867         /* For sequential zones, reset the zone write pointer */
1868         if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1869                 sector = SECTOR_FROM_BLOCK(blkstart);
1870                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1871                 div64_u64_rem(sector, bdev_zone_sectors(bdev), &remainder);
1872
1873                 if (remainder || nr_sects != bdev_zone_sectors(bdev)) {
1874                         f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1875                                  devi, sbi->s_ndevs ? FDEV(devi).path : "",
1876                                  blkstart, blklen);
1877                         return -EIO;
1878                 }
1879                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1880                 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1881                                         sector, nr_sects, GFP_NOFS);
1882         }
1883
1884         /* For conventional zones, use regular discard if supported */
1885         __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1886         return 0;
1887 }
1888 #endif
1889
1890 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1891                 struct block_device *bdev, block_t blkstart, block_t blklen)
1892 {
1893 #ifdef CONFIG_BLK_DEV_ZONED
1894         if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1895                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1896 #endif
1897         __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1898         return 0;
1899 }
1900
1901 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1902                                 block_t blkstart, block_t blklen)
1903 {
1904         sector_t start = blkstart, len = 0;
1905         struct block_device *bdev;
1906         struct seg_entry *se;
1907         unsigned int offset;
1908         block_t i;
1909         int err = 0;
1910
1911         bdev = f2fs_target_device(sbi, blkstart, NULL);
1912
1913         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1914                 if (i != start) {
1915                         struct block_device *bdev2 =
1916                                 f2fs_target_device(sbi, i, NULL);
1917
1918                         if (bdev2 != bdev) {
1919                                 err = __issue_discard_async(sbi, bdev,
1920                                                 start, len);
1921                                 if (err)
1922                                         return err;
1923                                 bdev = bdev2;
1924                                 start = i;
1925                                 len = 0;
1926                         }
1927                 }
1928
1929                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1930                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1931
1932                 if (f2fs_block_unit_discard(sbi) &&
1933                                 !f2fs_test_and_set_bit(offset, se->discard_map))
1934                         sbi->discard_blks--;
1935         }
1936
1937         if (len)
1938                 err = __issue_discard_async(sbi, bdev, start, len);
1939         return err;
1940 }
1941
1942 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1943                                                         bool check_only)
1944 {
1945         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1946         int max_blocks = sbi->blocks_per_seg;
1947         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1948         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1949         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1950         unsigned long *discard_map = (unsigned long *)se->discard_map;
1951         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1952         unsigned int start = 0, end = -1;
1953         bool force = (cpc->reason & CP_DISCARD);
1954         struct discard_entry *de = NULL;
1955         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1956         int i;
1957
1958         if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi) ||
1959                         !f2fs_block_unit_discard(sbi))
1960                 return false;
1961
1962         if (!force) {
1963                 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1964                         SM_I(sbi)->dcc_info->nr_discards >=
1965                                 SM_I(sbi)->dcc_info->max_discards)
1966                         return false;
1967         }
1968
1969         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1970         for (i = 0; i < entries; i++)
1971                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1972                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1973
1974         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1975                                 SM_I(sbi)->dcc_info->max_discards) {
1976                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1977                 if (start >= max_blocks)
1978                         break;
1979
1980                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1981                 if (force && start && end != max_blocks
1982                                         && (end - start) < cpc->trim_minlen)
1983                         continue;
1984
1985                 if (check_only)
1986                         return true;
1987
1988                 if (!de) {
1989                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1990                                                 GFP_F2FS_ZERO, true, NULL);
1991                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1992                         list_add_tail(&de->list, head);
1993                 }
1994
1995                 for (i = start; i < end; i++)
1996                         __set_bit_le(i, (void *)de->discard_map);
1997
1998                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1999         }
2000         return false;
2001 }
2002
2003 static void release_discard_addr(struct discard_entry *entry)
2004 {
2005         list_del(&entry->list);
2006         kmem_cache_free(discard_entry_slab, entry);
2007 }
2008
2009 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
2010 {
2011         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
2012         struct discard_entry *entry, *this;
2013
2014         /* drop caches */
2015         list_for_each_entry_safe(entry, this, head, list)
2016                 release_discard_addr(entry);
2017 }
2018
2019 /*
2020  * Should call f2fs_clear_prefree_segments after checkpoint is done.
2021  */
2022 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
2023 {
2024         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2025         unsigned int segno;
2026
2027         mutex_lock(&dirty_i->seglist_lock);
2028         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
2029                 __set_test_and_free(sbi, segno, false);
2030         mutex_unlock(&dirty_i->seglist_lock);
2031 }
2032
2033 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
2034                                                 struct cp_control *cpc)
2035 {
2036         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2037         struct list_head *head = &dcc->entry_list;
2038         struct discard_entry *entry, *this;
2039         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2040         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
2041         unsigned int start = 0, end = -1;
2042         unsigned int secno, start_segno;
2043         bool force = (cpc->reason & CP_DISCARD);
2044         bool section_alignment = F2FS_OPTION(sbi).discard_unit ==
2045                                                 DISCARD_UNIT_SECTION;
2046
2047         if (f2fs_lfs_mode(sbi) && __is_large_section(sbi))
2048                 section_alignment = true;
2049
2050         mutex_lock(&dirty_i->seglist_lock);
2051
2052         while (1) {
2053                 int i;
2054
2055                 if (section_alignment && end != -1)
2056                         end--;
2057                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
2058                 if (start >= MAIN_SEGS(sbi))
2059                         break;
2060                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
2061                                                                 start + 1);
2062
2063                 if (section_alignment) {
2064                         start = rounddown(start, sbi->segs_per_sec);
2065                         end = roundup(end, sbi->segs_per_sec);
2066                 }
2067
2068                 for (i = start; i < end; i++) {
2069                         if (test_and_clear_bit(i, prefree_map))
2070                                 dirty_i->nr_dirty[PRE]--;
2071                 }
2072
2073                 if (!f2fs_realtime_discard_enable(sbi))
2074                         continue;
2075
2076                 if (force && start >= cpc->trim_start &&
2077                                         (end - 1) <= cpc->trim_end)
2078                         continue;
2079
2080                 /* Should cover 2MB zoned device for zone-based reset */
2081                 if (!f2fs_sb_has_blkzoned(sbi) &&
2082                     (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi))) {
2083                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2084                                 (end - start) << sbi->log_blocks_per_seg);
2085                         continue;
2086                 }
2087 next:
2088                 secno = GET_SEC_FROM_SEG(sbi, start);
2089                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
2090                 if (!IS_CURSEC(sbi, secno) &&
2091                         !get_valid_blocks(sbi, start, true))
2092                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2093                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
2094
2095                 start = start_segno + sbi->segs_per_sec;
2096                 if (start < end)
2097                         goto next;
2098                 else
2099                         end = start - 1;
2100         }
2101         mutex_unlock(&dirty_i->seglist_lock);
2102
2103         if (!f2fs_block_unit_discard(sbi))
2104                 goto wakeup;
2105
2106         /* send small discards */
2107         list_for_each_entry_safe(entry, this, head, list) {
2108                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2109                 bool is_valid = test_bit_le(0, entry->discard_map);
2110
2111 find_next:
2112                 if (is_valid) {
2113                         next_pos = find_next_zero_bit_le(entry->discard_map,
2114                                         sbi->blocks_per_seg, cur_pos);
2115                         len = next_pos - cur_pos;
2116
2117                         if (f2fs_sb_has_blkzoned(sbi) ||
2118                             (force && len < cpc->trim_minlen))
2119                                 goto skip;
2120
2121                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2122                                                                         len);
2123                         total_len += len;
2124                 } else {
2125                         next_pos = find_next_bit_le(entry->discard_map,
2126                                         sbi->blocks_per_seg, cur_pos);
2127                 }
2128 skip:
2129                 cur_pos = next_pos;
2130                 is_valid = !is_valid;
2131
2132                 if (cur_pos < sbi->blocks_per_seg)
2133                         goto find_next;
2134
2135                 release_discard_addr(entry);
2136                 dcc->nr_discards -= total_len;
2137         }
2138
2139 wakeup:
2140         wake_up_discard_thread(sbi, false);
2141 }
2142
2143 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi)
2144 {
2145         dev_t dev = sbi->sb->s_bdev->bd_dev;
2146         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2147         int err = 0;
2148
2149         if (!f2fs_realtime_discard_enable(sbi))
2150                 return 0;
2151
2152         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2153                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2154         if (IS_ERR(dcc->f2fs_issue_discard)) {
2155                 err = PTR_ERR(dcc->f2fs_issue_discard);
2156                 dcc->f2fs_issue_discard = NULL;
2157         }
2158
2159         return err;
2160 }
2161
2162 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2163 {
2164         struct discard_cmd_control *dcc;
2165         int err = 0, i;
2166
2167         if (SM_I(sbi)->dcc_info) {
2168                 dcc = SM_I(sbi)->dcc_info;
2169                 goto init_thread;
2170         }
2171
2172         dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2173         if (!dcc)
2174                 return -ENOMEM;
2175
2176         dcc->discard_io_aware_gran = MAX_PLIST_NUM;
2177         dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2178         dcc->max_ordered_discard = DEFAULT_MAX_ORDERED_DISCARD_GRANULARITY;
2179         if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT)
2180                 dcc->discard_granularity = sbi->blocks_per_seg;
2181         else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION)
2182                 dcc->discard_granularity = BLKS_PER_SEC(sbi);
2183
2184         INIT_LIST_HEAD(&dcc->entry_list);
2185         for (i = 0; i < MAX_PLIST_NUM; i++)
2186                 INIT_LIST_HEAD(&dcc->pend_list[i]);
2187         INIT_LIST_HEAD(&dcc->wait_list);
2188         INIT_LIST_HEAD(&dcc->fstrim_list);
2189         mutex_init(&dcc->cmd_lock);
2190         atomic_set(&dcc->issued_discard, 0);
2191         atomic_set(&dcc->queued_discard, 0);
2192         atomic_set(&dcc->discard_cmd_cnt, 0);
2193         dcc->nr_discards = 0;
2194         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2195         dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST;
2196         dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME;
2197         dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME;
2198         dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME;
2199         dcc->discard_urgent_util = DEF_DISCARD_URGENT_UTIL;
2200         dcc->undiscard_blks = 0;
2201         dcc->next_pos = 0;
2202         dcc->root = RB_ROOT_CACHED;
2203         dcc->rbtree_check = false;
2204
2205         init_waitqueue_head(&dcc->discard_wait_queue);
2206         SM_I(sbi)->dcc_info = dcc;
2207 init_thread:
2208         err = f2fs_start_discard_thread(sbi);
2209         if (err) {
2210                 kfree(dcc);
2211                 SM_I(sbi)->dcc_info = NULL;
2212         }
2213
2214         return err;
2215 }
2216
2217 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2218 {
2219         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2220
2221         if (!dcc)
2222                 return;
2223
2224         f2fs_stop_discard_thread(sbi);
2225
2226         /*
2227          * Recovery can cache discard commands, so in error path of
2228          * fill_super(), it needs to give a chance to handle them.
2229          */
2230         f2fs_issue_discard_timeout(sbi);
2231
2232         kfree(dcc);
2233         SM_I(sbi)->dcc_info = NULL;
2234 }
2235
2236 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2237 {
2238         struct sit_info *sit_i = SIT_I(sbi);
2239
2240         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2241                 sit_i->dirty_sentries++;
2242                 return false;
2243         }
2244
2245         return true;
2246 }
2247
2248 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2249                                         unsigned int segno, int modified)
2250 {
2251         struct seg_entry *se = get_seg_entry(sbi, segno);
2252
2253         se->type = type;
2254         if (modified)
2255                 __mark_sit_entry_dirty(sbi, segno);
2256 }
2257
2258 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2259                                                                 block_t blkaddr)
2260 {
2261         unsigned int segno = GET_SEGNO(sbi, blkaddr);
2262
2263         if (segno == NULL_SEGNO)
2264                 return 0;
2265         return get_seg_entry(sbi, segno)->mtime;
2266 }
2267
2268 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2269                                                 unsigned long long old_mtime)
2270 {
2271         struct seg_entry *se;
2272         unsigned int segno = GET_SEGNO(sbi, blkaddr);
2273         unsigned long long ctime = get_mtime(sbi, false);
2274         unsigned long long mtime = old_mtime ? old_mtime : ctime;
2275
2276         if (segno == NULL_SEGNO)
2277                 return;
2278
2279         se = get_seg_entry(sbi, segno);
2280
2281         if (!se->mtime)
2282                 se->mtime = mtime;
2283         else
2284                 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2285                                                 se->valid_blocks + 1);
2286
2287         if (ctime > SIT_I(sbi)->max_mtime)
2288                 SIT_I(sbi)->max_mtime = ctime;
2289 }
2290
2291 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2292 {
2293         struct seg_entry *se;
2294         unsigned int segno, offset;
2295         long int new_vblocks;
2296         bool exist;
2297 #ifdef CONFIG_F2FS_CHECK_FS
2298         bool mir_exist;
2299 #endif
2300
2301         segno = GET_SEGNO(sbi, blkaddr);
2302
2303         se = get_seg_entry(sbi, segno);
2304         new_vblocks = se->valid_blocks + del;
2305         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2306
2307         f2fs_bug_on(sbi, (new_vblocks < 0 ||
2308                         (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2309
2310         se->valid_blocks = new_vblocks;
2311
2312         /* Update valid block bitmap */
2313         if (del > 0) {
2314                 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2315 #ifdef CONFIG_F2FS_CHECK_FS
2316                 mir_exist = f2fs_test_and_set_bit(offset,
2317                                                 se->cur_valid_map_mir);
2318                 if (unlikely(exist != mir_exist)) {
2319                         f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2320                                  blkaddr, exist);
2321                         f2fs_bug_on(sbi, 1);
2322                 }
2323 #endif
2324                 if (unlikely(exist)) {
2325                         f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2326                                  blkaddr);
2327                         f2fs_bug_on(sbi, 1);
2328                         se->valid_blocks--;
2329                         del = 0;
2330                 }
2331
2332                 if (f2fs_block_unit_discard(sbi) &&
2333                                 !f2fs_test_and_set_bit(offset, se->discard_map))
2334                         sbi->discard_blks--;
2335
2336                 /*
2337                  * SSR should never reuse block which is checkpointed
2338                  * or newly invalidated.
2339                  */
2340                 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2341                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2342                                 se->ckpt_valid_blocks++;
2343                 }
2344         } else {
2345                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2346 #ifdef CONFIG_F2FS_CHECK_FS
2347                 mir_exist = f2fs_test_and_clear_bit(offset,
2348                                                 se->cur_valid_map_mir);
2349                 if (unlikely(exist != mir_exist)) {
2350                         f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2351                                  blkaddr, exist);
2352                         f2fs_bug_on(sbi, 1);
2353                 }
2354 #endif
2355                 if (unlikely(!exist)) {
2356                         f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2357                                  blkaddr);
2358                         f2fs_bug_on(sbi, 1);
2359                         se->valid_blocks++;
2360                         del = 0;
2361                 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2362                         /*
2363                          * If checkpoints are off, we must not reuse data that
2364                          * was used in the previous checkpoint. If it was used
2365                          * before, we must track that to know how much space we
2366                          * really have.
2367                          */
2368                         if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2369                                 spin_lock(&sbi->stat_lock);
2370                                 sbi->unusable_block_count++;
2371                                 spin_unlock(&sbi->stat_lock);
2372                         }
2373                 }
2374
2375                 if (f2fs_block_unit_discard(sbi) &&
2376                         f2fs_test_and_clear_bit(offset, se->discard_map))
2377                         sbi->discard_blks++;
2378         }
2379         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2380                 se->ckpt_valid_blocks += del;
2381
2382         __mark_sit_entry_dirty(sbi, segno);
2383
2384         /* update total number of valid blocks to be written in ckpt area */
2385         SIT_I(sbi)->written_valid_blocks += del;
2386
2387         if (__is_large_section(sbi))
2388                 get_sec_entry(sbi, segno)->valid_blocks += del;
2389 }
2390
2391 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2392 {
2393         unsigned int segno = GET_SEGNO(sbi, addr);
2394         struct sit_info *sit_i = SIT_I(sbi);
2395
2396         f2fs_bug_on(sbi, addr == NULL_ADDR);
2397         if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2398                 return;
2399
2400         invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2401         f2fs_invalidate_compress_page(sbi, addr);
2402
2403         /* add it into sit main buffer */
2404         down_write(&sit_i->sentry_lock);
2405
2406         update_segment_mtime(sbi, addr, 0);
2407         update_sit_entry(sbi, addr, -1);
2408
2409         /* add it into dirty seglist */
2410         locate_dirty_segment(sbi, segno);
2411
2412         up_write(&sit_i->sentry_lock);
2413 }
2414
2415 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2416 {
2417         struct sit_info *sit_i = SIT_I(sbi);
2418         unsigned int segno, offset;
2419         struct seg_entry *se;
2420         bool is_cp = false;
2421
2422         if (!__is_valid_data_blkaddr(blkaddr))
2423                 return true;
2424
2425         down_read(&sit_i->sentry_lock);
2426
2427         segno = GET_SEGNO(sbi, blkaddr);
2428         se = get_seg_entry(sbi, segno);
2429         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2430
2431         if (f2fs_test_bit(offset, se->ckpt_valid_map))
2432                 is_cp = true;
2433
2434         up_read(&sit_i->sentry_lock);
2435
2436         return is_cp;
2437 }
2438
2439 static unsigned short f2fs_curseg_valid_blocks(struct f2fs_sb_info *sbi, int type)
2440 {
2441         struct curseg_info *curseg = CURSEG_I(sbi, type);
2442
2443         if (sbi->ckpt->alloc_type[type] == SSR)
2444                 return sbi->blocks_per_seg;
2445         return curseg->next_blkoff;
2446 }
2447
2448 /*
2449  * Calculate the number of current summary pages for writing
2450  */
2451 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2452 {
2453         int valid_sum_count = 0;
2454         int i, sum_in_page;
2455
2456         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2457                 if (sbi->ckpt->alloc_type[i] != SSR && for_ra)
2458                         valid_sum_count +=
2459                                 le16_to_cpu(F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2460                 else
2461                         valid_sum_count += f2fs_curseg_valid_blocks(sbi, i);
2462         }
2463
2464         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2465                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2466         if (valid_sum_count <= sum_in_page)
2467                 return 1;
2468         else if ((valid_sum_count - sum_in_page) <=
2469                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2470                 return 2;
2471         return 3;
2472 }
2473
2474 /*
2475  * Caller should put this summary page
2476  */
2477 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2478 {
2479         if (unlikely(f2fs_cp_error(sbi)))
2480                 return ERR_PTR(-EIO);
2481         return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno));
2482 }
2483
2484 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2485                                         void *src, block_t blk_addr)
2486 {
2487         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2488
2489         memcpy(page_address(page), src, PAGE_SIZE);
2490         set_page_dirty(page);
2491         f2fs_put_page(page, 1);
2492 }
2493
2494 static void write_sum_page(struct f2fs_sb_info *sbi,
2495                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
2496 {
2497         f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2498 }
2499
2500 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2501                                                 int type, block_t blk_addr)
2502 {
2503         struct curseg_info *curseg = CURSEG_I(sbi, type);
2504         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2505         struct f2fs_summary_block *src = curseg->sum_blk;
2506         struct f2fs_summary_block *dst;
2507
2508         dst = (struct f2fs_summary_block *)page_address(page);
2509         memset(dst, 0, PAGE_SIZE);
2510
2511         mutex_lock(&curseg->curseg_mutex);
2512
2513         down_read(&curseg->journal_rwsem);
2514         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2515         up_read(&curseg->journal_rwsem);
2516
2517         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2518         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2519
2520         mutex_unlock(&curseg->curseg_mutex);
2521
2522         set_page_dirty(page);
2523         f2fs_put_page(page, 1);
2524 }
2525
2526 static int is_next_segment_free(struct f2fs_sb_info *sbi,
2527                                 struct curseg_info *curseg, int type)
2528 {
2529         unsigned int segno = curseg->segno + 1;
2530         struct free_segmap_info *free_i = FREE_I(sbi);
2531
2532         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2533                 return !test_bit(segno, free_i->free_segmap);
2534         return 0;
2535 }
2536
2537 /*
2538  * Find a new segment from the free segments bitmap to right order
2539  * This function should be returned with success, otherwise BUG
2540  */
2541 static void get_new_segment(struct f2fs_sb_info *sbi,
2542                         unsigned int *newseg, bool new_sec, int dir)
2543 {
2544         struct free_segmap_info *free_i = FREE_I(sbi);
2545         unsigned int segno, secno, zoneno;
2546         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2547         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2548         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2549         unsigned int left_start = hint;
2550         bool init = true;
2551         int go_left = 0;
2552         int i;
2553
2554         spin_lock(&free_i->segmap_lock);
2555
2556         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2557                 segno = find_next_zero_bit(free_i->free_segmap,
2558                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2559                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2560                         goto got_it;
2561         }
2562 find_other_zone:
2563         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2564         if (secno >= MAIN_SECS(sbi)) {
2565                 if (dir == ALLOC_RIGHT) {
2566                         secno = find_first_zero_bit(free_i->free_secmap,
2567                                                         MAIN_SECS(sbi));
2568                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2569                 } else {
2570                         go_left = 1;
2571                         left_start = hint - 1;
2572                 }
2573         }
2574         if (go_left == 0)
2575                 goto skip_left;
2576
2577         while (test_bit(left_start, free_i->free_secmap)) {
2578                 if (left_start > 0) {
2579                         left_start--;
2580                         continue;
2581                 }
2582                 left_start = find_first_zero_bit(free_i->free_secmap,
2583                                                         MAIN_SECS(sbi));
2584                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2585                 break;
2586         }
2587         secno = left_start;
2588 skip_left:
2589         segno = GET_SEG_FROM_SEC(sbi, secno);
2590         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2591
2592         /* give up on finding another zone */
2593         if (!init)
2594                 goto got_it;
2595         if (sbi->secs_per_zone == 1)
2596                 goto got_it;
2597         if (zoneno == old_zoneno)
2598                 goto got_it;
2599         if (dir == ALLOC_LEFT) {
2600                 if (!go_left && zoneno + 1 >= total_zones)
2601                         goto got_it;
2602                 if (go_left && zoneno == 0)
2603                         goto got_it;
2604         }
2605         for (i = 0; i < NR_CURSEG_TYPE; i++)
2606                 if (CURSEG_I(sbi, i)->zone == zoneno)
2607                         break;
2608
2609         if (i < NR_CURSEG_TYPE) {
2610                 /* zone is in user, try another */
2611                 if (go_left)
2612                         hint = zoneno * sbi->secs_per_zone - 1;
2613                 else if (zoneno + 1 >= total_zones)
2614                         hint = 0;
2615                 else
2616                         hint = (zoneno + 1) * sbi->secs_per_zone;
2617                 init = false;
2618                 goto find_other_zone;
2619         }
2620 got_it:
2621         /* set it as dirty segment in free segmap */
2622         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2623         __set_inuse(sbi, segno);
2624         *newseg = segno;
2625         spin_unlock(&free_i->segmap_lock);
2626 }
2627
2628 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2629 {
2630         struct curseg_info *curseg = CURSEG_I(sbi, type);
2631         struct summary_footer *sum_footer;
2632         unsigned short seg_type = curseg->seg_type;
2633
2634         curseg->inited = true;
2635         curseg->segno = curseg->next_segno;
2636         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2637         curseg->next_blkoff = 0;
2638         curseg->next_segno = NULL_SEGNO;
2639
2640         sum_footer = &(curseg->sum_blk->footer);
2641         memset(sum_footer, 0, sizeof(struct summary_footer));
2642
2643         sanity_check_seg_type(sbi, seg_type);
2644
2645         if (IS_DATASEG(seg_type))
2646                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2647         if (IS_NODESEG(seg_type))
2648                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2649         __set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
2650 }
2651
2652 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2653 {
2654         struct curseg_info *curseg = CURSEG_I(sbi, type);
2655         unsigned short seg_type = curseg->seg_type;
2656
2657         sanity_check_seg_type(sbi, seg_type);
2658         if (f2fs_need_rand_seg(sbi))
2659                 return get_random_u32_below(MAIN_SECS(sbi) * sbi->segs_per_sec);
2660
2661         /* if segs_per_sec is large than 1, we need to keep original policy. */
2662         if (__is_large_section(sbi))
2663                 return curseg->segno;
2664
2665         /* inmem log may not locate on any segment after mount */
2666         if (!curseg->inited)
2667                 return 0;
2668
2669         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2670                 return 0;
2671
2672         if (test_opt(sbi, NOHEAP) &&
2673                 (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type)))
2674                 return 0;
2675
2676         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2677                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2678
2679         /* find segments from 0 to reuse freed segments */
2680         if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2681                 return 0;
2682
2683         return curseg->segno;
2684 }
2685
2686 /*
2687  * Allocate a current working segment.
2688  * This function always allocates a free segment in LFS manner.
2689  */
2690 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2691 {
2692         struct curseg_info *curseg = CURSEG_I(sbi, type);
2693         unsigned short seg_type = curseg->seg_type;
2694         unsigned int segno = curseg->segno;
2695         int dir = ALLOC_LEFT;
2696
2697         if (curseg->inited)
2698                 write_sum_page(sbi, curseg->sum_blk,
2699                                 GET_SUM_BLOCK(sbi, segno));
2700         if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA)
2701                 dir = ALLOC_RIGHT;
2702
2703         if (test_opt(sbi, NOHEAP))
2704                 dir = ALLOC_RIGHT;
2705
2706         segno = __get_next_segno(sbi, type);
2707         get_new_segment(sbi, &segno, new_sec, dir);
2708         curseg->next_segno = segno;
2709         reset_curseg(sbi, type, 1);
2710         curseg->alloc_type = LFS;
2711         if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
2712                 curseg->fragment_remained_chunk =
2713                                 get_random_u32_inclusive(1, sbi->max_fragment_chunk);
2714 }
2715
2716 static int __next_free_blkoff(struct f2fs_sb_info *sbi,
2717                                         int segno, block_t start)
2718 {
2719         struct seg_entry *se = get_seg_entry(sbi, segno);
2720         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2721         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2722         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2723         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2724         int i;
2725
2726         for (i = 0; i < entries; i++)
2727                 target_map[i] = ckpt_map[i] | cur_map[i];
2728
2729         return __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2730 }
2731
2732 static int f2fs_find_next_ssr_block(struct f2fs_sb_info *sbi,
2733                 struct curseg_info *seg)
2734 {
2735         return __next_free_blkoff(sbi, seg->segno, seg->next_blkoff + 1);
2736 }
2737
2738 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno)
2739 {
2740         return __next_free_blkoff(sbi, segno, 0) < sbi->blocks_per_seg;
2741 }
2742
2743 /*
2744  * This function always allocates a used segment(from dirty seglist) by SSR
2745  * manner, so it should recover the existing segment information of valid blocks
2746  */
2747 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2748 {
2749         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2750         struct curseg_info *curseg = CURSEG_I(sbi, type);
2751         unsigned int new_segno = curseg->next_segno;
2752         struct f2fs_summary_block *sum_node;
2753         struct page *sum_page;
2754
2755         write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, curseg->segno));
2756
2757         __set_test_and_inuse(sbi, new_segno);
2758
2759         mutex_lock(&dirty_i->seglist_lock);
2760         __remove_dirty_segment(sbi, new_segno, PRE);
2761         __remove_dirty_segment(sbi, new_segno, DIRTY);
2762         mutex_unlock(&dirty_i->seglist_lock);
2763
2764         reset_curseg(sbi, type, 1);
2765         curseg->alloc_type = SSR;
2766         curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0);
2767
2768         sum_page = f2fs_get_sum_page(sbi, new_segno);
2769         if (IS_ERR(sum_page)) {
2770                 /* GC won't be able to use stale summary pages by cp_error */
2771                 memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
2772                 return;
2773         }
2774         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2775         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2776         f2fs_put_page(sum_page, 1);
2777 }
2778
2779 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2780                                 int alloc_mode, unsigned long long age);
2781
2782 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type,
2783                                         int target_type, int alloc_mode,
2784                                         unsigned long long age)
2785 {
2786         struct curseg_info *curseg = CURSEG_I(sbi, type);
2787
2788         curseg->seg_type = target_type;
2789
2790         if (get_ssr_segment(sbi, type, alloc_mode, age)) {
2791                 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
2792
2793                 curseg->seg_type = se->type;
2794                 change_curseg(sbi, type);
2795         } else {
2796                 /* allocate cold segment by default */
2797                 curseg->seg_type = CURSEG_COLD_DATA;
2798                 new_curseg(sbi, type, true);
2799         }
2800         stat_inc_seg_type(sbi, curseg);
2801 }
2802
2803 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
2804 {
2805         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
2806
2807         if (!sbi->am.atgc_enabled)
2808                 return;
2809
2810         f2fs_down_read(&SM_I(sbi)->curseg_lock);
2811
2812         mutex_lock(&curseg->curseg_mutex);
2813         down_write(&SIT_I(sbi)->sentry_lock);
2814
2815         get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0);
2816
2817         up_write(&SIT_I(sbi)->sentry_lock);
2818         mutex_unlock(&curseg->curseg_mutex);
2819
2820         f2fs_up_read(&SM_I(sbi)->curseg_lock);
2821
2822 }
2823 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
2824 {
2825         __f2fs_init_atgc_curseg(sbi);
2826 }
2827
2828 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2829 {
2830         struct curseg_info *curseg = CURSEG_I(sbi, type);
2831
2832         mutex_lock(&curseg->curseg_mutex);
2833         if (!curseg->inited)
2834                 goto out;
2835
2836         if (get_valid_blocks(sbi, curseg->segno, false)) {
2837                 write_sum_page(sbi, curseg->sum_blk,
2838                                 GET_SUM_BLOCK(sbi, curseg->segno));
2839         } else {
2840                 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2841                 __set_test_and_free(sbi, curseg->segno, true);
2842                 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2843         }
2844 out:
2845         mutex_unlock(&curseg->curseg_mutex);
2846 }
2847
2848 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
2849 {
2850         __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2851
2852         if (sbi->am.atgc_enabled)
2853                 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2854 }
2855
2856 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2857 {
2858         struct curseg_info *curseg = CURSEG_I(sbi, type);
2859
2860         mutex_lock(&curseg->curseg_mutex);
2861         if (!curseg->inited)
2862                 goto out;
2863         if (get_valid_blocks(sbi, curseg->segno, false))
2864                 goto out;
2865
2866         mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2867         __set_test_and_inuse(sbi, curseg->segno);
2868         mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2869 out:
2870         mutex_unlock(&curseg->curseg_mutex);
2871 }
2872
2873 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
2874 {
2875         __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2876
2877         if (sbi->am.atgc_enabled)
2878                 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2879 }
2880
2881 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2882                                 int alloc_mode, unsigned long long age)
2883 {
2884         struct curseg_info *curseg = CURSEG_I(sbi, type);
2885         unsigned segno = NULL_SEGNO;
2886         unsigned short seg_type = curseg->seg_type;
2887         int i, cnt;
2888         bool reversed = false;
2889
2890         sanity_check_seg_type(sbi, seg_type);
2891
2892         /* f2fs_need_SSR() already forces to do this */
2893         if (!f2fs_get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) {
2894                 curseg->next_segno = segno;
2895                 return 1;
2896         }
2897
2898         /* For node segments, let's do SSR more intensively */
2899         if (IS_NODESEG(seg_type)) {
2900                 if (seg_type >= CURSEG_WARM_NODE) {
2901                         reversed = true;
2902                         i = CURSEG_COLD_NODE;
2903                 } else {
2904                         i = CURSEG_HOT_NODE;
2905                 }
2906                 cnt = NR_CURSEG_NODE_TYPE;
2907         } else {
2908                 if (seg_type >= CURSEG_WARM_DATA) {
2909                         reversed = true;
2910                         i = CURSEG_COLD_DATA;
2911                 } else {
2912                         i = CURSEG_HOT_DATA;
2913                 }
2914                 cnt = NR_CURSEG_DATA_TYPE;
2915         }
2916
2917         for (; cnt-- > 0; reversed ? i-- : i++) {
2918                 if (i == seg_type)
2919                         continue;
2920                 if (!f2fs_get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) {
2921                         curseg->next_segno = segno;
2922                         return 1;
2923                 }
2924         }
2925
2926         /* find valid_blocks=0 in dirty list */
2927         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2928                 segno = get_free_segment(sbi);
2929                 if (segno != NULL_SEGNO) {
2930                         curseg->next_segno = segno;
2931                         return 1;
2932                 }
2933         }
2934         return 0;
2935 }
2936
2937 static bool need_new_seg(struct f2fs_sb_info *sbi, int type)
2938 {
2939         struct curseg_info *curseg = CURSEG_I(sbi, type);
2940
2941         if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2942             curseg->seg_type == CURSEG_WARM_NODE)
2943                 return true;
2944         if (curseg->alloc_type == LFS &&
2945             is_next_segment_free(sbi, curseg, type) &&
2946             likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2947                 return true;
2948         if (!f2fs_need_SSR(sbi) || !get_ssr_segment(sbi, type, SSR, 0))
2949                 return true;
2950         return false;
2951 }
2952
2953 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2954                                         unsigned int start, unsigned int end)
2955 {
2956         struct curseg_info *curseg = CURSEG_I(sbi, type);
2957         unsigned int segno;
2958
2959         f2fs_down_read(&SM_I(sbi)->curseg_lock);
2960         mutex_lock(&curseg->curseg_mutex);
2961         down_write(&SIT_I(sbi)->sentry_lock);
2962
2963         segno = CURSEG_I(sbi, type)->segno;
2964         if (segno < start || segno > end)
2965                 goto unlock;
2966
2967         if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
2968                 change_curseg(sbi, type);
2969         else
2970                 new_curseg(sbi, type, true);
2971
2972         stat_inc_seg_type(sbi, curseg);
2973
2974         locate_dirty_segment(sbi, segno);
2975 unlock:
2976         up_write(&SIT_I(sbi)->sentry_lock);
2977
2978         if (segno != curseg->segno)
2979                 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2980                             type, segno, curseg->segno);
2981
2982         mutex_unlock(&curseg->curseg_mutex);
2983         f2fs_up_read(&SM_I(sbi)->curseg_lock);
2984 }
2985
2986 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type,
2987                                                 bool new_sec, bool force)
2988 {
2989         struct curseg_info *curseg = CURSEG_I(sbi, type);
2990         unsigned int old_segno;
2991
2992         if (!force && curseg->inited &&
2993             !curseg->next_blkoff &&
2994             !get_valid_blocks(sbi, curseg->segno, new_sec) &&
2995             !get_ckpt_valid_blocks(sbi, curseg->segno, new_sec))
2996                 return;
2997
2998         old_segno = curseg->segno;
2999         new_curseg(sbi, type, true);
3000         stat_inc_seg_type(sbi, curseg);
3001         locate_dirty_segment(sbi, old_segno);
3002 }
3003
3004 void f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force)
3005 {
3006         f2fs_down_read(&SM_I(sbi)->curseg_lock);
3007         down_write(&SIT_I(sbi)->sentry_lock);
3008         __allocate_new_segment(sbi, type, true, force);
3009         up_write(&SIT_I(sbi)->sentry_lock);
3010         f2fs_up_read(&SM_I(sbi)->curseg_lock);
3011 }
3012
3013 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
3014 {
3015         int i;
3016
3017         f2fs_down_read(&SM_I(sbi)->curseg_lock);
3018         down_write(&SIT_I(sbi)->sentry_lock);
3019         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
3020                 __allocate_new_segment(sbi, i, false, false);
3021         up_write(&SIT_I(sbi)->sentry_lock);
3022         f2fs_up_read(&SM_I(sbi)->curseg_lock);
3023 }
3024
3025 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
3026                                                 struct cp_control *cpc)
3027 {
3028         __u64 trim_start = cpc->trim_start;
3029         bool has_candidate = false;
3030
3031         down_write(&SIT_I(sbi)->sentry_lock);
3032         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
3033                 if (add_discard_addrs(sbi, cpc, true)) {
3034                         has_candidate = true;
3035                         break;
3036                 }
3037         }
3038         up_write(&SIT_I(sbi)->sentry_lock);
3039
3040         cpc->trim_start = trim_start;
3041         return has_candidate;
3042 }
3043
3044 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
3045                                         struct discard_policy *dpolicy,
3046                                         unsigned int start, unsigned int end)
3047 {
3048         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
3049         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
3050         struct rb_node **insert_p = NULL, *insert_parent = NULL;
3051         struct discard_cmd *dc;
3052         struct blk_plug plug;
3053         int issued;
3054         unsigned int trimmed = 0;
3055
3056 next:
3057         issued = 0;
3058
3059         mutex_lock(&dcc->cmd_lock);
3060         if (unlikely(dcc->rbtree_check))
3061                 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
3062
3063         dc = __lookup_discard_cmd_ret(&dcc->root, start,
3064                                 &prev_dc, &next_dc, &insert_p, &insert_parent);
3065         if (!dc)
3066                 dc = next_dc;
3067
3068         blk_start_plug(&plug);
3069
3070         while (dc && dc->di.lstart <= end) {
3071                 struct rb_node *node;
3072                 int err = 0;
3073
3074                 if (dc->di.len < dpolicy->granularity)
3075                         goto skip;
3076
3077                 if (dc->state != D_PREP) {
3078                         list_move_tail(&dc->list, &dcc->fstrim_list);
3079                         goto skip;
3080                 }
3081
3082                 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
3083
3084                 if (issued >= dpolicy->max_requests) {
3085                         start = dc->di.lstart + dc->di.len;
3086
3087                         if (err)
3088                                 __remove_discard_cmd(sbi, dc);
3089
3090                         blk_finish_plug(&plug);
3091                         mutex_unlock(&dcc->cmd_lock);
3092                         trimmed += __wait_all_discard_cmd(sbi, NULL);
3093                         f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
3094                         goto next;
3095                 }
3096 skip:
3097                 node = rb_next(&dc->rb_node);
3098                 if (err)
3099                         __remove_discard_cmd(sbi, dc);
3100                 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3101
3102                 if (fatal_signal_pending(current))
3103                         break;
3104         }
3105
3106         blk_finish_plug(&plug);
3107         mutex_unlock(&dcc->cmd_lock);
3108
3109         return trimmed;
3110 }
3111
3112 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3113 {
3114         __u64 start = F2FS_BYTES_TO_BLK(range->start);
3115         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
3116         unsigned int start_segno, end_segno;
3117         block_t start_block, end_block;
3118         struct cp_control cpc;
3119         struct discard_policy dpolicy;
3120         unsigned long long trimmed = 0;
3121         int err = 0;
3122         bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
3123
3124         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
3125                 return -EINVAL;
3126
3127         if (end < MAIN_BLKADDR(sbi))
3128                 goto out;
3129
3130         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
3131                 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
3132                 return -EFSCORRUPTED;
3133         }
3134
3135         /* start/end segment number in main_area */
3136         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3137         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3138                                                 GET_SEGNO(sbi, end);
3139         if (need_align) {
3140                 start_segno = rounddown(start_segno, sbi->segs_per_sec);
3141                 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
3142         }
3143
3144         cpc.reason = CP_DISCARD;
3145         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
3146         cpc.trim_start = start_segno;
3147         cpc.trim_end = end_segno;
3148
3149         if (sbi->discard_blks == 0)
3150                 goto out;
3151
3152         f2fs_down_write(&sbi->gc_lock);
3153         err = f2fs_write_checkpoint(sbi, &cpc);
3154         f2fs_up_write(&sbi->gc_lock);
3155         if (err)
3156                 goto out;
3157
3158         /*
3159          * We filed discard candidates, but actually we don't need to wait for
3160          * all of them, since they'll be issued in idle time along with runtime
3161          * discard option. User configuration looks like using runtime discard
3162          * or periodic fstrim instead of it.
3163          */
3164         if (f2fs_realtime_discard_enable(sbi))
3165                 goto out;
3166
3167         start_block = START_BLOCK(sbi, start_segno);
3168         end_block = START_BLOCK(sbi, end_segno + 1);
3169
3170         __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3171         trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3172                                         start_block, end_block);
3173
3174         trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3175                                         start_block, end_block);
3176 out:
3177         if (!err)
3178                 range->len = F2FS_BLK_TO_BYTES(trimmed);
3179         return err;
3180 }
3181
3182 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
3183 {
3184         switch (hint) {
3185         case WRITE_LIFE_SHORT:
3186                 return CURSEG_HOT_DATA;
3187         case WRITE_LIFE_EXTREME:
3188                 return CURSEG_COLD_DATA;
3189         default:
3190                 return CURSEG_WARM_DATA;
3191         }
3192 }
3193
3194 static int __get_segment_type_2(struct f2fs_io_info *fio)
3195 {
3196         if (fio->type == DATA)
3197                 return CURSEG_HOT_DATA;
3198         else
3199                 return CURSEG_HOT_NODE;
3200 }
3201
3202 static int __get_segment_type_4(struct f2fs_io_info *fio)
3203 {
3204         if (fio->type == DATA) {
3205                 struct inode *inode = fio->page->mapping->host;
3206
3207                 if (S_ISDIR(inode->i_mode))
3208                         return CURSEG_HOT_DATA;
3209                 else
3210                         return CURSEG_COLD_DATA;
3211         } else {
3212                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3213                         return CURSEG_WARM_NODE;
3214                 else
3215                         return CURSEG_COLD_NODE;
3216         }
3217 }
3218
3219 static int __get_age_segment_type(struct inode *inode, pgoff_t pgofs)
3220 {
3221         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3222         struct extent_info ei = {};
3223
3224         if (f2fs_lookup_age_extent_cache(inode, pgofs, &ei)) {
3225                 if (!ei.age)
3226                         return NO_CHECK_TYPE;
3227                 if (ei.age <= sbi->hot_data_age_threshold)
3228                         return CURSEG_HOT_DATA;
3229                 if (ei.age <= sbi->warm_data_age_threshold)
3230                         return CURSEG_WARM_DATA;
3231                 return CURSEG_COLD_DATA;
3232         }
3233         return NO_CHECK_TYPE;
3234 }
3235
3236 static int __get_segment_type_6(struct f2fs_io_info *fio)
3237 {
3238         if (fio->type == DATA) {
3239                 struct inode *inode = fio->page->mapping->host;
3240                 int type;
3241
3242                 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
3243                         return CURSEG_COLD_DATA_PINNED;
3244
3245                 if (page_private_gcing(fio->page)) {
3246                         if (fio->sbi->am.atgc_enabled &&
3247                                 (fio->io_type == FS_DATA_IO) &&
3248                                 (fio->sbi->gc_mode != GC_URGENT_HIGH))
3249                                 return CURSEG_ALL_DATA_ATGC;
3250                         else
3251                                 return CURSEG_COLD_DATA;
3252                 }
3253                 if (file_is_cold(inode) || f2fs_need_compress_data(inode))
3254                         return CURSEG_COLD_DATA;
3255
3256                 type = __get_age_segment_type(inode, fio->page->index);
3257                 if (type != NO_CHECK_TYPE)
3258                         return type;
3259
3260                 if (file_is_hot(inode) ||
3261                                 is_inode_flag_set(inode, FI_HOT_DATA) ||
3262                                 f2fs_is_cow_file(inode))
3263                         return CURSEG_HOT_DATA;
3264                 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3265         } else {
3266                 if (IS_DNODE(fio->page))
3267                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3268                                                 CURSEG_HOT_NODE;
3269                 return CURSEG_COLD_NODE;
3270         }
3271 }
3272
3273 static int __get_segment_type(struct f2fs_io_info *fio)
3274 {
3275         int type = 0;
3276
3277         switch (F2FS_OPTION(fio->sbi).active_logs) {
3278         case 2:
3279                 type = __get_segment_type_2(fio);
3280                 break;
3281         case 4:
3282                 type = __get_segment_type_4(fio);
3283                 break;
3284         case 6:
3285                 type = __get_segment_type_6(fio);
3286                 break;
3287         default:
3288                 f2fs_bug_on(fio->sbi, true);
3289         }
3290
3291         if (IS_HOT(type))
3292                 fio->temp = HOT;
3293         else if (IS_WARM(type))
3294                 fio->temp = WARM;
3295         else
3296                 fio->temp = COLD;
3297         return type;
3298 }
3299
3300 static void f2fs_randomize_chunk(struct f2fs_sb_info *sbi,
3301                 struct curseg_info *seg)
3302 {
3303         /* To allocate block chunks in different sizes, use random number */
3304         if (--seg->fragment_remained_chunk > 0)
3305                 return;
3306
3307         seg->fragment_remained_chunk =
3308                 get_random_u32_inclusive(1, sbi->max_fragment_chunk);
3309         seg->next_blkoff +=
3310                 get_random_u32_inclusive(1, sbi->max_fragment_hole);
3311 }
3312
3313 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3314                 block_t old_blkaddr, block_t *new_blkaddr,
3315                 struct f2fs_summary *sum, int type,
3316                 struct f2fs_io_info *fio)
3317 {
3318         struct sit_info *sit_i = SIT_I(sbi);
3319         struct curseg_info *curseg = CURSEG_I(sbi, type);
3320         unsigned long long old_mtime;
3321         bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3322         struct seg_entry *se = NULL;
3323         bool segment_full = false;
3324
3325         f2fs_down_read(&SM_I(sbi)->curseg_lock);
3326
3327         mutex_lock(&curseg->curseg_mutex);
3328         down_write(&sit_i->sentry_lock);
3329
3330         if (from_gc) {
3331                 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3332                 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3333                 sanity_check_seg_type(sbi, se->type);
3334                 f2fs_bug_on(sbi, IS_NODESEG(se->type));
3335         }
3336         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3337
3338         f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg);
3339
3340         f2fs_wait_discard_bio(sbi, *new_blkaddr);
3341
3342         curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
3343         if (curseg->alloc_type == SSR) {
3344                 curseg->next_blkoff = f2fs_find_next_ssr_block(sbi, curseg);
3345         } else {
3346                 curseg->next_blkoff++;
3347                 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
3348                         f2fs_randomize_chunk(sbi, curseg);
3349         }
3350         if (curseg->next_blkoff >= f2fs_usable_blks_in_seg(sbi, curseg->segno))
3351                 segment_full = true;
3352         stat_inc_block_count(sbi, curseg);
3353
3354         if (from_gc) {
3355                 old_mtime = get_segment_mtime(sbi, old_blkaddr);
3356         } else {
3357                 update_segment_mtime(sbi, old_blkaddr, 0);
3358                 old_mtime = 0;
3359         }
3360         update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3361
3362         /*
3363          * SIT information should be updated before segment allocation,
3364          * since SSR needs latest valid block information.
3365          */
3366         update_sit_entry(sbi, *new_blkaddr, 1);
3367         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3368                 update_sit_entry(sbi, old_blkaddr, -1);
3369
3370         /*
3371          * If the current segment is full, flush it out and replace it with a
3372          * new segment.
3373          */
3374         if (segment_full) {
3375                 if (from_gc) {
3376                         get_atssr_segment(sbi, type, se->type,
3377                                                 AT_SSR, se->mtime);
3378                 } else {
3379                         if (need_new_seg(sbi, type))
3380                                 new_curseg(sbi, type, false);
3381                         else
3382                                 change_curseg(sbi, type);
3383                         stat_inc_seg_type(sbi, curseg);
3384                 }
3385         }
3386         /*
3387          * segment dirty status should be updated after segment allocation,
3388          * so we just need to update status only one time after previous
3389          * segment being closed.
3390          */
3391         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3392         locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3393
3394         if (IS_DATASEG(type))
3395                 atomic64_inc(&sbi->allocated_data_blocks);
3396
3397         up_write(&sit_i->sentry_lock);
3398
3399         if (page && IS_NODESEG(type)) {
3400                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3401
3402                 f2fs_inode_chksum_set(sbi, page);
3403         }
3404
3405         if (fio) {
3406                 struct f2fs_bio_info *io;
3407
3408                 if (F2FS_IO_ALIGNED(sbi))
3409                         fio->retry = 0;
3410
3411                 INIT_LIST_HEAD(&fio->list);
3412                 fio->in_list = 1;
3413                 io = sbi->write_io[fio->type] + fio->temp;
3414                 spin_lock(&io->io_lock);
3415                 list_add_tail(&fio->list, &io->io_list);
3416                 spin_unlock(&io->io_lock);
3417         }
3418
3419         mutex_unlock(&curseg->curseg_mutex);
3420
3421         f2fs_up_read(&SM_I(sbi)->curseg_lock);
3422 }
3423
3424 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino,
3425                                         block_t blkaddr, unsigned int blkcnt)
3426 {
3427         if (!f2fs_is_multi_device(sbi))
3428                 return;
3429
3430         while (1) {
3431                 unsigned int devidx = f2fs_target_device_index(sbi, blkaddr);
3432                 unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1;
3433
3434                 /* update device state for fsync */
3435                 f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO);
3436
3437                 /* update device state for checkpoint */
3438                 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3439                         spin_lock(&sbi->dev_lock);
3440                         f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3441                         spin_unlock(&sbi->dev_lock);
3442                 }
3443
3444                 if (blkcnt <= blks)
3445                         break;
3446                 blkcnt -= blks;
3447                 blkaddr += blks;
3448         }
3449 }
3450
3451 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3452 {
3453         int type = __get_segment_type(fio);
3454         bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3455
3456         if (keep_order)
3457                 f2fs_down_read(&fio->sbi->io_order_lock);
3458 reallocate:
3459         f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3460                         &fio->new_blkaddr, sum, type, fio);
3461         if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO) {
3462                 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3463                                         fio->old_blkaddr, fio->old_blkaddr);
3464                 f2fs_invalidate_compress_page(fio->sbi, fio->old_blkaddr);
3465         }
3466
3467         /* writeout dirty page into bdev */
3468         f2fs_submit_page_write(fio);
3469         if (fio->retry) {
3470                 fio->old_blkaddr = fio->new_blkaddr;
3471                 goto reallocate;
3472         }
3473
3474         f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1);
3475
3476         if (keep_order)
3477                 f2fs_up_read(&fio->sbi->io_order_lock);
3478 }
3479
3480 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3481                                         enum iostat_type io_type)
3482 {
3483         struct f2fs_io_info fio = {
3484                 .sbi = sbi,
3485                 .type = META,
3486                 .temp = HOT,
3487                 .op = REQ_OP_WRITE,
3488                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3489                 .old_blkaddr = page->index,
3490                 .new_blkaddr = page->index,
3491                 .page = page,
3492                 .encrypted_page = NULL,
3493                 .in_list = 0,
3494         };
3495
3496         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3497                 fio.op_flags &= ~REQ_META;
3498
3499         set_page_writeback(page);
3500         f2fs_submit_page_write(&fio);
3501
3502         stat_inc_meta_count(sbi, page->index);
3503         f2fs_update_iostat(sbi, NULL, io_type, F2FS_BLKSIZE);
3504 }
3505
3506 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3507 {
3508         struct f2fs_summary sum;
3509
3510         set_summary(&sum, nid, 0, 0);
3511         do_write_page(&sum, fio);
3512
3513         f2fs_update_iostat(fio->sbi, NULL, fio->io_type, F2FS_BLKSIZE);
3514 }
3515
3516 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3517                                         struct f2fs_io_info *fio)
3518 {
3519         struct f2fs_sb_info *sbi = fio->sbi;
3520         struct f2fs_summary sum;
3521
3522         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3523         if (fio->io_type == FS_DATA_IO || fio->io_type == FS_CP_DATA_IO)
3524                 f2fs_update_age_extent_cache(dn);
3525         set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3526         do_write_page(&sum, fio);
3527         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3528
3529         f2fs_update_iostat(sbi, dn->inode, fio->io_type, F2FS_BLKSIZE);
3530 }
3531
3532 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3533 {
3534         int err;
3535         struct f2fs_sb_info *sbi = fio->sbi;
3536         unsigned int segno;
3537
3538         fio->new_blkaddr = fio->old_blkaddr;
3539         /* i/o temperature is needed for passing down write hints */
3540         __get_segment_type(fio);
3541
3542         segno = GET_SEGNO(sbi, fio->new_blkaddr);
3543
3544         if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3545                 set_sbi_flag(sbi, SBI_NEED_FSCK);
3546                 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3547                           __func__, segno);
3548                 err = -EFSCORRUPTED;
3549                 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
3550                 goto drop_bio;
3551         }
3552
3553         if (f2fs_cp_error(sbi)) {
3554                 err = -EIO;
3555                 goto drop_bio;
3556         }
3557
3558         if (fio->post_read)
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 && !IS_F2FS_IPU_NOCACHE(sbi))
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->page->mapping->host,
3572                                                 fio->io_type, F2FS_BLKSIZE);
3573         }
3574
3575         return err;
3576 drop_bio:
3577         if (fio->bio && *(fio->bio)) {
3578                 struct bio *bio = *(fio->bio);
3579
3580                 bio->bi_status = BLK_STS_IOERR;
3581                 bio_endio(bio);
3582                 *(fio->bio) = NULL;
3583         }
3584         return err;
3585 }
3586
3587 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3588                                                 unsigned int segno)
3589 {
3590         int i;
3591
3592         for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3593                 if (CURSEG_I(sbi, i)->segno == segno)
3594                         break;
3595         }
3596         return i;
3597 }
3598
3599 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3600                                 block_t old_blkaddr, block_t new_blkaddr,
3601                                 bool recover_curseg, bool recover_newaddr,
3602                                 bool from_gc)
3603 {
3604         struct sit_info *sit_i = SIT_I(sbi);
3605         struct curseg_info *curseg;
3606         unsigned int segno, old_cursegno;
3607         struct seg_entry *se;
3608         int type;
3609         unsigned short old_blkoff;
3610         unsigned char old_alloc_type;
3611
3612         segno = GET_SEGNO(sbi, new_blkaddr);
3613         se = get_seg_entry(sbi, segno);
3614         type = se->type;
3615
3616         f2fs_down_write(&SM_I(sbi)->curseg_lock);
3617
3618         if (!recover_curseg) {
3619                 /* for recovery flow */
3620                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3621                         if (old_blkaddr == NULL_ADDR)
3622                                 type = CURSEG_COLD_DATA;
3623                         else
3624                                 type = CURSEG_WARM_DATA;
3625                 }
3626         } else {
3627                 if (IS_CURSEG(sbi, segno)) {
3628                         /* se->type is volatile as SSR allocation */
3629                         type = __f2fs_get_curseg(sbi, segno);
3630                         f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3631                 } else {
3632                         type = CURSEG_WARM_DATA;
3633                 }
3634         }
3635
3636         f2fs_bug_on(sbi, !IS_DATASEG(type));
3637         curseg = CURSEG_I(sbi, type);
3638
3639         mutex_lock(&curseg->curseg_mutex);
3640         down_write(&sit_i->sentry_lock);
3641
3642         old_cursegno = curseg->segno;
3643         old_blkoff = curseg->next_blkoff;
3644         old_alloc_type = curseg->alloc_type;
3645
3646         /* change the current segment */
3647         if (segno != curseg->segno) {
3648                 curseg->next_segno = segno;
3649                 change_curseg(sbi, type);
3650         }
3651
3652         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3653         curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
3654
3655         if (!recover_curseg || recover_newaddr) {
3656                 if (!from_gc)
3657                         update_segment_mtime(sbi, new_blkaddr, 0);
3658                 update_sit_entry(sbi, new_blkaddr, 1);
3659         }
3660         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3661                 invalidate_mapping_pages(META_MAPPING(sbi),
3662                                         old_blkaddr, old_blkaddr);
3663                 f2fs_invalidate_compress_page(sbi, old_blkaddr);
3664                 if (!from_gc)
3665                         update_segment_mtime(sbi, old_blkaddr, 0);
3666                 update_sit_entry(sbi, old_blkaddr, -1);
3667         }
3668
3669         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3670         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3671
3672         locate_dirty_segment(sbi, old_cursegno);
3673
3674         if (recover_curseg) {
3675                 if (old_cursegno != curseg->segno) {
3676                         curseg->next_segno = old_cursegno;
3677                         change_curseg(sbi, type);
3678                 }
3679                 curseg->next_blkoff = old_blkoff;
3680                 curseg->alloc_type = old_alloc_type;
3681         }
3682
3683         up_write(&sit_i->sentry_lock);
3684         mutex_unlock(&curseg->curseg_mutex);
3685         f2fs_up_write(&SM_I(sbi)->curseg_lock);
3686 }
3687
3688 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3689                                 block_t old_addr, block_t new_addr,
3690                                 unsigned char version, bool recover_curseg,
3691                                 bool recover_newaddr)
3692 {
3693         struct f2fs_summary sum;
3694
3695         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3696
3697         f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3698                                         recover_curseg, recover_newaddr, false);
3699
3700         f2fs_update_data_blkaddr(dn, new_addr);
3701 }
3702
3703 void f2fs_wait_on_page_writeback(struct page *page,
3704                                 enum page_type type, bool ordered, bool locked)
3705 {
3706         if (PageWriteback(page)) {
3707                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3708
3709                 /* submit cached LFS IO */
3710                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3711                 /* submit cached IPU IO */
3712                 f2fs_submit_merged_ipu_write(sbi, NULL, page);
3713                 if (ordered) {
3714                         wait_on_page_writeback(page);
3715                         f2fs_bug_on(sbi, locked && PageWriteback(page));
3716                 } else {
3717                         wait_for_stable_page(page);
3718                 }
3719         }
3720 }
3721
3722 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3723 {
3724         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3725         struct page *cpage;
3726
3727         if (!f2fs_post_read_required(inode))
3728                 return;
3729
3730         if (!__is_valid_data_blkaddr(blkaddr))
3731                 return;
3732
3733         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3734         if (cpage) {
3735                 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3736                 f2fs_put_page(cpage, 1);
3737         }
3738 }
3739
3740 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3741                                                                 block_t len)
3742 {
3743         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3744         block_t i;
3745
3746         if (!f2fs_post_read_required(inode))
3747                 return;
3748
3749         for (i = 0; i < len; i++)
3750                 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3751
3752         invalidate_mapping_pages(META_MAPPING(sbi), blkaddr, blkaddr + len - 1);
3753 }
3754
3755 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3756 {
3757         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3758         struct curseg_info *seg_i;
3759         unsigned char *kaddr;
3760         struct page *page;
3761         block_t start;
3762         int i, j, offset;
3763
3764         start = start_sum_block(sbi);
3765
3766         page = f2fs_get_meta_page(sbi, start++);
3767         if (IS_ERR(page))
3768                 return PTR_ERR(page);
3769         kaddr = (unsigned char *)page_address(page);
3770
3771         /* Step 1: restore nat cache */
3772         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3773         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3774
3775         /* Step 2: restore sit cache */
3776         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3777         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3778         offset = 2 * SUM_JOURNAL_SIZE;
3779
3780         /* Step 3: restore summary entries */
3781         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3782                 unsigned short blk_off;
3783                 unsigned int segno;
3784
3785                 seg_i = CURSEG_I(sbi, i);
3786                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3787                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3788                 seg_i->next_segno = segno;
3789                 reset_curseg(sbi, i, 0);
3790                 seg_i->alloc_type = ckpt->alloc_type[i];
3791                 seg_i->next_blkoff = blk_off;
3792
3793                 if (seg_i->alloc_type == SSR)
3794                         blk_off = sbi->blocks_per_seg;
3795
3796                 for (j = 0; j < blk_off; j++) {
3797                         struct f2fs_summary *s;
3798
3799                         s = (struct f2fs_summary *)(kaddr + offset);
3800                         seg_i->sum_blk->entries[j] = *s;
3801                         offset += SUMMARY_SIZE;
3802                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3803                                                 SUM_FOOTER_SIZE)
3804                                 continue;
3805
3806                         f2fs_put_page(page, 1);
3807                         page = NULL;
3808
3809                         page = f2fs_get_meta_page(sbi, start++);
3810                         if (IS_ERR(page))
3811                                 return PTR_ERR(page);
3812                         kaddr = (unsigned char *)page_address(page);
3813                         offset = 0;
3814                 }
3815         }
3816         f2fs_put_page(page, 1);
3817         return 0;
3818 }
3819
3820 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3821 {
3822         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3823         struct f2fs_summary_block *sum;
3824         struct curseg_info *curseg;
3825         struct page *new;
3826         unsigned short blk_off;
3827         unsigned int segno = 0;
3828         block_t blk_addr = 0;
3829         int err = 0;
3830
3831         /* get segment number and block addr */
3832         if (IS_DATASEG(type)) {
3833                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3834                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3835                                                         CURSEG_HOT_DATA]);
3836                 if (__exist_node_summaries(sbi))
3837                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
3838                 else
3839                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3840         } else {
3841                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3842                                                         CURSEG_HOT_NODE]);
3843                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3844                                                         CURSEG_HOT_NODE]);
3845                 if (__exist_node_summaries(sbi))
3846                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3847                                                         type - CURSEG_HOT_NODE);
3848                 else
3849                         blk_addr = GET_SUM_BLOCK(sbi, segno);
3850         }
3851
3852         new = f2fs_get_meta_page(sbi, blk_addr);
3853         if (IS_ERR(new))
3854                 return PTR_ERR(new);
3855         sum = (struct f2fs_summary_block *)page_address(new);
3856
3857         if (IS_NODESEG(type)) {
3858                 if (__exist_node_summaries(sbi)) {
3859                         struct f2fs_summary *ns = &sum->entries[0];
3860                         int i;
3861
3862                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3863                                 ns->version = 0;
3864                                 ns->ofs_in_node = 0;
3865                         }
3866                 } else {
3867                         err = f2fs_restore_node_summary(sbi, segno, sum);
3868                         if (err)
3869                                 goto out;
3870                 }
3871         }
3872
3873         /* set uncompleted segment to curseg */
3874         curseg = CURSEG_I(sbi, type);
3875         mutex_lock(&curseg->curseg_mutex);
3876
3877         /* update journal info */
3878         down_write(&curseg->journal_rwsem);
3879         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3880         up_write(&curseg->journal_rwsem);
3881
3882         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3883         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3884         curseg->next_segno = segno;
3885         reset_curseg(sbi, type, 0);
3886         curseg->alloc_type = ckpt->alloc_type[type];
3887         curseg->next_blkoff = blk_off;
3888         mutex_unlock(&curseg->curseg_mutex);
3889 out:
3890         f2fs_put_page(new, 1);
3891         return err;
3892 }
3893
3894 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3895 {
3896         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3897         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3898         int type = CURSEG_HOT_DATA;
3899         int err;
3900
3901         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3902                 int npages = f2fs_npages_for_summary_flush(sbi, true);
3903
3904                 if (npages >= 2)
3905                         f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3906                                                         META_CP, true);
3907
3908                 /* restore for compacted data summary */
3909                 err = read_compacted_summaries(sbi);
3910                 if (err)
3911                         return err;
3912                 type = CURSEG_HOT_NODE;
3913         }
3914
3915         if (__exist_node_summaries(sbi))
3916                 f2fs_ra_meta_pages(sbi,
3917                                 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
3918                                 NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
3919
3920         for (; type <= CURSEG_COLD_NODE; type++) {
3921                 err = read_normal_summaries(sbi, type);
3922                 if (err)
3923                         return err;
3924         }
3925
3926         /* sanity check for summary blocks */
3927         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3928                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3929                 f2fs_err(sbi, "invalid journal entries nats %u sits %u",
3930                          nats_in_cursum(nat_j), sits_in_cursum(sit_j));
3931                 return -EINVAL;
3932         }
3933
3934         return 0;
3935 }
3936
3937 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3938 {
3939         struct page *page;
3940         unsigned char *kaddr;
3941         struct f2fs_summary *summary;
3942         struct curseg_info *seg_i;
3943         int written_size = 0;
3944         int i, j;
3945
3946         page = f2fs_grab_meta_page(sbi, blkaddr++);
3947         kaddr = (unsigned char *)page_address(page);
3948         memset(kaddr, 0, PAGE_SIZE);
3949
3950         /* Step 1: write nat cache */
3951         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3952         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3953         written_size += SUM_JOURNAL_SIZE;
3954
3955         /* Step 2: write sit cache */
3956         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3957         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3958         written_size += SUM_JOURNAL_SIZE;
3959
3960         /* Step 3: write summary entries */
3961         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3962                 seg_i = CURSEG_I(sbi, i);
3963                 for (j = 0; j < f2fs_curseg_valid_blocks(sbi, i); 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                         list_move_tail(&ses->set_list, &next->set_list);
4091                         return;
4092                 }
4093
4094         list_move_tail(&ses->set_list, head);
4095 }
4096
4097 static void add_sit_entry(unsigned int segno, struct list_head *head)
4098 {
4099         struct sit_entry_set *ses;
4100         unsigned int start_segno = START_SEGNO(segno);
4101
4102         list_for_each_entry(ses, head, set_list) {
4103                 if (ses->start_segno == start_segno) {
4104                         ses->entry_cnt++;
4105                         adjust_sit_entry_set(ses, head);
4106                         return;
4107                 }
4108         }
4109
4110         ses = grab_sit_entry_set();
4111
4112         ses->start_segno = start_segno;
4113         ses->entry_cnt++;
4114         list_add(&ses->set_list, head);
4115 }
4116
4117 static void add_sits_in_set(struct f2fs_sb_info *sbi)
4118 {
4119         struct f2fs_sm_info *sm_info = SM_I(sbi);
4120         struct list_head *set_list = &sm_info->sit_entry_set;
4121         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
4122         unsigned int segno;
4123
4124         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
4125                 add_sit_entry(segno, set_list);
4126 }
4127
4128 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
4129 {
4130         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4131         struct f2fs_journal *journal = curseg->journal;
4132         int i;
4133
4134         down_write(&curseg->journal_rwsem);
4135         for (i = 0; i < sits_in_cursum(journal); i++) {
4136                 unsigned int segno;
4137                 bool dirtied;
4138
4139                 segno = le32_to_cpu(segno_in_journal(journal, i));
4140                 dirtied = __mark_sit_entry_dirty(sbi, segno);
4141
4142                 if (!dirtied)
4143                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
4144         }
4145         update_sits_in_cursum(journal, -i);
4146         up_write(&curseg->journal_rwsem);
4147 }
4148
4149 /*
4150  * CP calls this function, which flushes SIT entries including sit_journal,
4151  * and moves prefree segs to free segs.
4152  */
4153 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
4154 {
4155         struct sit_info *sit_i = SIT_I(sbi);
4156         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4157         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4158         struct f2fs_journal *journal = curseg->journal;
4159         struct sit_entry_set *ses, *tmp;
4160         struct list_head *head = &SM_I(sbi)->sit_entry_set;
4161         bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4162         struct seg_entry *se;
4163
4164         down_write(&sit_i->sentry_lock);
4165
4166         if (!sit_i->dirty_sentries)
4167                 goto out;
4168
4169         /*
4170          * add and account sit entries of dirty bitmap in sit entry
4171          * set temporarily
4172          */
4173         add_sits_in_set(sbi);
4174
4175         /*
4176          * if there are no enough space in journal to store dirty sit
4177          * entries, remove all entries from journal and add and account
4178          * them in sit entry set.
4179          */
4180         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4181                                                                 !to_journal)
4182                 remove_sits_in_journal(sbi);
4183
4184         /*
4185          * there are two steps to flush sit entries:
4186          * #1, flush sit entries to journal in current cold data summary block.
4187          * #2, flush sit entries to sit page.
4188          */
4189         list_for_each_entry_safe(ses, tmp, head, set_list) {
4190                 struct page *page = NULL;
4191                 struct f2fs_sit_block *raw_sit = NULL;
4192                 unsigned int start_segno = ses->start_segno;
4193                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4194                                                 (unsigned long)MAIN_SEGS(sbi));
4195                 unsigned int segno = start_segno;
4196
4197                 if (to_journal &&
4198                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4199                         to_journal = false;
4200
4201                 if (to_journal) {
4202                         down_write(&curseg->journal_rwsem);
4203                 } else {
4204                         page = get_next_sit_page(sbi, start_segno);
4205                         raw_sit = page_address(page);
4206                 }
4207
4208                 /* flush dirty sit entries in region of current sit set */
4209                 for_each_set_bit_from(segno, bitmap, end) {
4210                         int offset, sit_offset;
4211
4212                         se = get_seg_entry(sbi, segno);
4213 #ifdef CONFIG_F2FS_CHECK_FS
4214                         if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4215                                                 SIT_VBLOCK_MAP_SIZE))
4216                                 f2fs_bug_on(sbi, 1);
4217 #endif
4218
4219                         /* add discard candidates */
4220                         if (!(cpc->reason & CP_DISCARD)) {
4221                                 cpc->trim_start = segno;
4222                                 add_discard_addrs(sbi, cpc, false);
4223                         }
4224
4225                         if (to_journal) {
4226                                 offset = f2fs_lookup_journal_in_cursum(journal,
4227                                                         SIT_JOURNAL, segno, 1);
4228                                 f2fs_bug_on(sbi, offset < 0);
4229                                 segno_in_journal(journal, offset) =
4230                                                         cpu_to_le32(segno);
4231                                 seg_info_to_raw_sit(se,
4232                                         &sit_in_journal(journal, offset));
4233                                 check_block_count(sbi, segno,
4234                                         &sit_in_journal(journal, offset));
4235                         } else {
4236                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4237                                 seg_info_to_raw_sit(se,
4238                                                 &raw_sit->entries[sit_offset]);
4239                                 check_block_count(sbi, segno,
4240                                                 &raw_sit->entries[sit_offset]);
4241                         }
4242
4243                         __clear_bit(segno, bitmap);
4244                         sit_i->dirty_sentries--;
4245                         ses->entry_cnt--;
4246                 }
4247
4248                 if (to_journal)
4249                         up_write(&curseg->journal_rwsem);
4250                 else
4251                         f2fs_put_page(page, 1);
4252
4253                 f2fs_bug_on(sbi, ses->entry_cnt);
4254                 release_sit_entry_set(ses);
4255         }
4256
4257         f2fs_bug_on(sbi, !list_empty(head));
4258         f2fs_bug_on(sbi, sit_i->dirty_sentries);
4259 out:
4260         if (cpc->reason & CP_DISCARD) {
4261                 __u64 trim_start = cpc->trim_start;
4262
4263                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4264                         add_discard_addrs(sbi, cpc, false);
4265
4266                 cpc->trim_start = trim_start;
4267         }
4268         up_write(&sit_i->sentry_lock);
4269
4270         set_prefree_as_free_segments(sbi);
4271 }
4272
4273 static int build_sit_info(struct f2fs_sb_info *sbi)
4274 {
4275         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4276         struct sit_info *sit_i;
4277         unsigned int sit_segs, start;
4278         char *src_bitmap, *bitmap;
4279         unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4280         unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0;
4281
4282         /* allocate memory for SIT information */
4283         sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4284         if (!sit_i)
4285                 return -ENOMEM;
4286
4287         SM_I(sbi)->sit_info = sit_i;
4288
4289         sit_i->sentries =
4290                 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4291                                               MAIN_SEGS(sbi)),
4292                               GFP_KERNEL);
4293         if (!sit_i->sentries)
4294                 return -ENOMEM;
4295
4296         main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4297         sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4298                                                                 GFP_KERNEL);
4299         if (!sit_i->dirty_sentries_bitmap)
4300                 return -ENOMEM;
4301
4302 #ifdef CONFIG_F2FS_CHECK_FS
4303         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map);
4304 #else
4305         bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map);
4306 #endif
4307         sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4308         if (!sit_i->bitmap)
4309                 return -ENOMEM;
4310
4311         bitmap = sit_i->bitmap;
4312
4313         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4314                 sit_i->sentries[start].cur_valid_map = bitmap;
4315                 bitmap += SIT_VBLOCK_MAP_SIZE;
4316
4317                 sit_i->sentries[start].ckpt_valid_map = bitmap;
4318                 bitmap += SIT_VBLOCK_MAP_SIZE;
4319
4320 #ifdef CONFIG_F2FS_CHECK_FS
4321                 sit_i->sentries[start].cur_valid_map_mir = bitmap;
4322                 bitmap += SIT_VBLOCK_MAP_SIZE;
4323 #endif
4324
4325                 if (discard_map) {
4326                         sit_i->sentries[start].discard_map = bitmap;
4327                         bitmap += SIT_VBLOCK_MAP_SIZE;
4328                 }
4329         }
4330
4331         sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4332         if (!sit_i->tmp_map)
4333                 return -ENOMEM;
4334
4335         if (__is_large_section(sbi)) {
4336                 sit_i->sec_entries =
4337                         f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4338                                                       MAIN_SECS(sbi)),
4339                                       GFP_KERNEL);
4340                 if (!sit_i->sec_entries)
4341                         return -ENOMEM;
4342         }
4343
4344         /* get information related with SIT */
4345         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4346
4347         /* setup SIT bitmap from ckeckpoint pack */
4348         sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4349         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4350
4351         sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4352         if (!sit_i->sit_bitmap)
4353                 return -ENOMEM;
4354
4355 #ifdef CONFIG_F2FS_CHECK_FS
4356         sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4357                                         sit_bitmap_size, GFP_KERNEL);
4358         if (!sit_i->sit_bitmap_mir)
4359                 return -ENOMEM;
4360
4361         sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4362                                         main_bitmap_size, GFP_KERNEL);
4363         if (!sit_i->invalid_segmap)
4364                 return -ENOMEM;
4365 #endif
4366
4367         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4368         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4369         sit_i->written_valid_blocks = 0;
4370         sit_i->bitmap_size = sit_bitmap_size;
4371         sit_i->dirty_sentries = 0;
4372         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4373         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4374         sit_i->mounted_time = ktime_get_boottime_seconds();
4375         init_rwsem(&sit_i->sentry_lock);
4376         return 0;
4377 }
4378
4379 static int build_free_segmap(struct f2fs_sb_info *sbi)
4380 {
4381         struct free_segmap_info *free_i;
4382         unsigned int bitmap_size, sec_bitmap_size;
4383
4384         /* allocate memory for free segmap information */
4385         free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4386         if (!free_i)
4387                 return -ENOMEM;
4388
4389         SM_I(sbi)->free_info = free_i;
4390
4391         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4392         free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4393         if (!free_i->free_segmap)
4394                 return -ENOMEM;
4395
4396         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4397         free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4398         if (!free_i->free_secmap)
4399                 return -ENOMEM;
4400
4401         /* set all segments as dirty temporarily */
4402         memset(free_i->free_segmap, 0xff, bitmap_size);
4403         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4404
4405         /* init free segmap information */
4406         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4407         free_i->free_segments = 0;
4408         free_i->free_sections = 0;
4409         spin_lock_init(&free_i->segmap_lock);
4410         return 0;
4411 }
4412
4413 static int build_curseg(struct f2fs_sb_info *sbi)
4414 {
4415         struct curseg_info *array;
4416         int i;
4417
4418         array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4419                                         sizeof(*array)), GFP_KERNEL);
4420         if (!array)
4421                 return -ENOMEM;
4422
4423         SM_I(sbi)->curseg_array = array;
4424
4425         for (i = 0; i < NO_CHECK_TYPE; i++) {
4426                 mutex_init(&array[i].curseg_mutex);
4427                 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4428                 if (!array[i].sum_blk)
4429                         return -ENOMEM;
4430                 init_rwsem(&array[i].journal_rwsem);
4431                 array[i].journal = f2fs_kzalloc(sbi,
4432                                 sizeof(struct f2fs_journal), GFP_KERNEL);
4433                 if (!array[i].journal)
4434                         return -ENOMEM;
4435                 if (i < NR_PERSISTENT_LOG)
4436                         array[i].seg_type = CURSEG_HOT_DATA + i;
4437                 else if (i == CURSEG_COLD_DATA_PINNED)
4438                         array[i].seg_type = CURSEG_COLD_DATA;
4439                 else if (i == CURSEG_ALL_DATA_ATGC)
4440                         array[i].seg_type = CURSEG_COLD_DATA;
4441                 array[i].segno = NULL_SEGNO;
4442                 array[i].next_blkoff = 0;
4443                 array[i].inited = false;
4444         }
4445         return restore_curseg_summaries(sbi);
4446 }
4447
4448 static int build_sit_entries(struct f2fs_sb_info *sbi)
4449 {
4450         struct sit_info *sit_i = SIT_I(sbi);
4451         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4452         struct f2fs_journal *journal = curseg->journal;
4453         struct seg_entry *se;
4454         struct f2fs_sit_entry sit;
4455         int sit_blk_cnt = SIT_BLK_CNT(sbi);
4456         unsigned int i, start, end;
4457         unsigned int readed, start_blk = 0;
4458         int err = 0;
4459         block_t sit_valid_blocks[2] = {0, 0};
4460
4461         do {
4462                 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS,
4463                                                         META_SIT, true);
4464
4465                 start = start_blk * sit_i->sents_per_block;
4466                 end = (start_blk + readed) * sit_i->sents_per_block;
4467
4468                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4469                         struct f2fs_sit_block *sit_blk;
4470                         struct page *page;
4471
4472                         se = &sit_i->sentries[start];
4473                         page = get_current_sit_page(sbi, start);
4474                         if (IS_ERR(page))
4475                                 return PTR_ERR(page);
4476                         sit_blk = (struct f2fs_sit_block *)page_address(page);
4477                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4478                         f2fs_put_page(page, 1);
4479
4480                         err = check_block_count(sbi, start, &sit);
4481                         if (err)
4482                                 return err;
4483                         seg_info_from_raw_sit(se, &sit);
4484
4485                         if (se->type >= NR_PERSISTENT_LOG) {
4486                                 f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4487                                                         se->type, start);
4488                                 f2fs_handle_error(sbi,
4489                                                 ERROR_INCONSISTENT_SUM_TYPE);
4490                                 return -EFSCORRUPTED;
4491                         }
4492
4493                         sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4494
4495                         if (f2fs_block_unit_discard(sbi)) {
4496                                 /* build discard map only one time */
4497                                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4498                                         memset(se->discard_map, 0xff,
4499                                                 SIT_VBLOCK_MAP_SIZE);
4500                                 } else {
4501                                         memcpy(se->discard_map,
4502                                                 se->cur_valid_map,
4503                                                 SIT_VBLOCK_MAP_SIZE);
4504                                         sbi->discard_blks +=
4505                                                 sbi->blocks_per_seg -
4506                                                 se->valid_blocks;
4507                                 }
4508                         }
4509
4510                         if (__is_large_section(sbi))
4511                                 get_sec_entry(sbi, start)->valid_blocks +=
4512                                                         se->valid_blocks;
4513                 }
4514                 start_blk += readed;
4515         } while (start_blk < sit_blk_cnt);
4516
4517         down_read(&curseg->journal_rwsem);
4518         for (i = 0; i < sits_in_cursum(journal); i++) {
4519                 unsigned int old_valid_blocks;
4520
4521                 start = le32_to_cpu(segno_in_journal(journal, i));
4522                 if (start >= MAIN_SEGS(sbi)) {
4523                         f2fs_err(sbi, "Wrong journal entry on segno %u",
4524                                  start);
4525                         err = -EFSCORRUPTED;
4526                         f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL);
4527                         break;
4528                 }
4529
4530                 se = &sit_i->sentries[start];
4531                 sit = sit_in_journal(journal, i);
4532
4533                 old_valid_blocks = se->valid_blocks;
4534
4535                 sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks;
4536
4537                 err = check_block_count(sbi, start, &sit);
4538                 if (err)
4539                         break;
4540                 seg_info_from_raw_sit(se, &sit);
4541
4542                 if (se->type >= NR_PERSISTENT_LOG) {
4543                         f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4544                                                         se->type, start);
4545                         err = -EFSCORRUPTED;
4546                         f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
4547                         break;
4548                 }
4549
4550                 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4551
4552                 if (f2fs_block_unit_discard(sbi)) {
4553                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4554                                 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4555                         } else {
4556                                 memcpy(se->discard_map, se->cur_valid_map,
4557                                                         SIT_VBLOCK_MAP_SIZE);
4558                                 sbi->discard_blks += old_valid_blocks;
4559                                 sbi->discard_blks -= se->valid_blocks;
4560                         }
4561                 }
4562
4563                 if (__is_large_section(sbi)) {
4564                         get_sec_entry(sbi, start)->valid_blocks +=
4565                                                         se->valid_blocks;
4566                         get_sec_entry(sbi, start)->valid_blocks -=
4567                                                         old_valid_blocks;
4568                 }
4569         }
4570         up_read(&curseg->journal_rwsem);
4571
4572         if (err)
4573                 return err;
4574
4575         if (sit_valid_blocks[NODE] != valid_node_count(sbi)) {
4576                 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4577                          sit_valid_blocks[NODE], valid_node_count(sbi));
4578                 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT);
4579                 return -EFSCORRUPTED;
4580         }
4581
4582         if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] >
4583                                 valid_user_blocks(sbi)) {
4584                 f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u",
4585                          sit_valid_blocks[DATA], sit_valid_blocks[NODE],
4586                          valid_user_blocks(sbi));
4587                 f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT);
4588                 return -EFSCORRUPTED;
4589         }
4590
4591         return 0;
4592 }
4593
4594 static void init_free_segmap(struct f2fs_sb_info *sbi)
4595 {
4596         unsigned int start;
4597         int type;
4598         struct seg_entry *sentry;
4599
4600         for (start = 0; start < MAIN_SEGS(sbi); start++) {
4601                 if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4602                         continue;
4603                 sentry = get_seg_entry(sbi, start);
4604                 if (!sentry->valid_blocks)
4605                         __set_free(sbi, start);
4606                 else
4607                         SIT_I(sbi)->written_valid_blocks +=
4608                                                 sentry->valid_blocks;
4609         }
4610
4611         /* set use the current segments */
4612         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4613                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4614
4615                 __set_test_and_inuse(sbi, curseg_t->segno);
4616         }
4617 }
4618
4619 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4620 {
4621         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4622         struct free_segmap_info *free_i = FREE_I(sbi);
4623         unsigned int segno = 0, offset = 0, secno;
4624         block_t valid_blocks, usable_blks_in_seg;
4625
4626         while (1) {
4627                 /* find dirty segment based on free segmap */
4628                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4629                 if (segno >= MAIN_SEGS(sbi))
4630                         break;
4631                 offset = segno + 1;
4632                 valid_blocks = get_valid_blocks(sbi, segno, false);
4633                 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4634                 if (valid_blocks == usable_blks_in_seg || !valid_blocks)
4635                         continue;
4636                 if (valid_blocks > usable_blks_in_seg) {
4637                         f2fs_bug_on(sbi, 1);
4638                         continue;
4639                 }
4640                 mutex_lock(&dirty_i->seglist_lock);
4641                 __locate_dirty_segment(sbi, segno, DIRTY);
4642                 mutex_unlock(&dirty_i->seglist_lock);
4643         }
4644
4645         if (!__is_large_section(sbi))
4646                 return;
4647
4648         mutex_lock(&dirty_i->seglist_lock);
4649         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4650                 valid_blocks = get_valid_blocks(sbi, segno, true);
4651                 secno = GET_SEC_FROM_SEG(sbi, segno);
4652
4653                 if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi))
4654                         continue;
4655                 if (IS_CURSEC(sbi, secno))
4656                         continue;
4657                 set_bit(secno, dirty_i->dirty_secmap);
4658         }
4659         mutex_unlock(&dirty_i->seglist_lock);
4660 }
4661
4662 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4663 {
4664         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4665         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4666
4667         dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4668         if (!dirty_i->victim_secmap)
4669                 return -ENOMEM;
4670
4671         dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4672         if (!dirty_i->pinned_secmap)
4673                 return -ENOMEM;
4674
4675         dirty_i->pinned_secmap_cnt = 0;
4676         dirty_i->enable_pin_section = true;
4677         return 0;
4678 }
4679
4680 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4681 {
4682         struct dirty_seglist_info *dirty_i;
4683         unsigned int bitmap_size, i;
4684
4685         /* allocate memory for dirty segments list information */
4686         dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4687                                                                 GFP_KERNEL);
4688         if (!dirty_i)
4689                 return -ENOMEM;
4690
4691         SM_I(sbi)->dirty_info = dirty_i;
4692         mutex_init(&dirty_i->seglist_lock);
4693
4694         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4695
4696         for (i = 0; i < NR_DIRTY_TYPE; i++) {
4697                 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4698                                                                 GFP_KERNEL);
4699                 if (!dirty_i->dirty_segmap[i])
4700                         return -ENOMEM;
4701         }
4702
4703         if (__is_large_section(sbi)) {
4704                 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4705                 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
4706                                                 bitmap_size, GFP_KERNEL);
4707                 if (!dirty_i->dirty_secmap)
4708                         return -ENOMEM;
4709         }
4710
4711         init_dirty_segmap(sbi);
4712         return init_victim_secmap(sbi);
4713 }
4714
4715 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4716 {
4717         int i;
4718
4719         /*
4720          * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4721          * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4722          */
4723         for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4724                 struct curseg_info *curseg = CURSEG_I(sbi, i);
4725                 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4726                 unsigned int blkofs = curseg->next_blkoff;
4727
4728                 if (f2fs_sb_has_readonly(sbi) &&
4729                         i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE)
4730                         continue;
4731
4732                 sanity_check_seg_type(sbi, curseg->seg_type);
4733
4734                 if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) {
4735                         f2fs_err(sbi,
4736                                  "Current segment has invalid alloc_type:%d",
4737                                  curseg->alloc_type);
4738                         f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
4739                         return -EFSCORRUPTED;
4740                 }
4741
4742                 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4743                         goto out;
4744
4745                 if (curseg->alloc_type == SSR)
4746                         continue;
4747
4748                 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4749                         if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4750                                 continue;
4751 out:
4752                         f2fs_err(sbi,
4753                                  "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4754                                  i, curseg->segno, curseg->alloc_type,
4755                                  curseg->next_blkoff, blkofs);
4756                         f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
4757                         return -EFSCORRUPTED;
4758                 }
4759         }
4760         return 0;
4761 }
4762
4763 #ifdef CONFIG_BLK_DEV_ZONED
4764
4765 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4766                                     struct f2fs_dev_info *fdev,
4767                                     struct blk_zone *zone)
4768 {
4769         unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno;
4770         block_t zone_block, wp_block, last_valid_block;
4771         unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4772         int i, s, b, ret;
4773         struct seg_entry *se;
4774
4775         if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4776                 return 0;
4777
4778         wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block);
4779         wp_segno = GET_SEGNO(sbi, wp_block);
4780         wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4781         zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4782         zone_segno = GET_SEGNO(sbi, zone_block);
4783         zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno);
4784
4785         if (zone_segno >= MAIN_SEGS(sbi))
4786                 return 0;
4787
4788         /*
4789          * Skip check of zones cursegs point to, since
4790          * fix_curseg_write_pointer() checks them.
4791          */
4792         for (i = 0; i < NO_CHECK_TYPE; i++)
4793                 if (zone_secno == GET_SEC_FROM_SEG(sbi,
4794                                                    CURSEG_I(sbi, i)->segno))
4795                         return 0;
4796
4797         /*
4798          * Get last valid block of the zone.
4799          */
4800         last_valid_block = zone_block - 1;
4801         for (s = sbi->segs_per_sec - 1; s >= 0; s--) {
4802                 segno = zone_segno + s;
4803                 se = get_seg_entry(sbi, segno);
4804                 for (b = sbi->blocks_per_seg - 1; b >= 0; b--)
4805                         if (f2fs_test_bit(b, se->cur_valid_map)) {
4806                                 last_valid_block = START_BLOCK(sbi, segno) + b;
4807                                 break;
4808                         }
4809                 if (last_valid_block >= zone_block)
4810                         break;
4811         }
4812
4813         /*
4814          * If last valid block is beyond the write pointer, report the
4815          * inconsistency. This inconsistency does not cause write error
4816          * because the zone will not be selected for write operation until
4817          * it get discarded. Just report it.
4818          */
4819         if (last_valid_block >= wp_block) {
4820                 f2fs_notice(sbi, "Valid block beyond write pointer: "
4821                             "valid block[0x%x,0x%x] wp[0x%x,0x%x]",
4822                             GET_SEGNO(sbi, last_valid_block),
4823                             GET_BLKOFF_FROM_SEG0(sbi, last_valid_block),
4824                             wp_segno, wp_blkoff);
4825                 return 0;
4826         }
4827
4828         /*
4829          * If there is no valid block in the zone and if write pointer is
4830          * not at zone start, reset the write pointer.
4831          */
4832         if (last_valid_block + 1 == zone_block && zone->wp != zone->start) {
4833                 f2fs_notice(sbi,
4834                             "Zone without valid block has non-zero write "
4835                             "pointer. Reset the write pointer: wp[0x%x,0x%x]",
4836                             wp_segno, wp_blkoff);
4837                 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
4838                                         zone->len >> log_sectors_per_block);
4839                 if (ret) {
4840                         f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4841                                  fdev->path, ret);
4842                         return ret;
4843                 }
4844         }
4845
4846         return 0;
4847 }
4848
4849 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
4850                                                   block_t zone_blkaddr)
4851 {
4852         int i;
4853
4854         for (i = 0; i < sbi->s_ndevs; i++) {
4855                 if (!bdev_is_zoned(FDEV(i).bdev))
4856                         continue;
4857                 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
4858                                 zone_blkaddr <= FDEV(i).end_blk))
4859                         return &FDEV(i);
4860         }
4861
4862         return NULL;
4863 }
4864
4865 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
4866                               void *data)
4867 {
4868         memcpy(data, zone, sizeof(struct blk_zone));
4869         return 0;
4870 }
4871
4872 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
4873 {
4874         struct curseg_info *cs = CURSEG_I(sbi, type);
4875         struct f2fs_dev_info *zbd;
4876         struct blk_zone zone;
4877         unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
4878         block_t cs_zone_block, wp_block;
4879         unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4880         sector_t zone_sector;
4881         int err;
4882
4883         cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4884         cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4885
4886         zbd = get_target_zoned_dev(sbi, cs_zone_block);
4887         if (!zbd)
4888                 return 0;
4889
4890         /* report zone for the sector the curseg points to */
4891         zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4892                 << log_sectors_per_block;
4893         err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4894                                   report_one_zone_cb, &zone);
4895         if (err != 1) {
4896                 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4897                          zbd->path, err);
4898                 return err;
4899         }
4900
4901         if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4902                 return 0;
4903
4904         wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
4905         wp_segno = GET_SEGNO(sbi, wp_block);
4906         wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4907         wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
4908
4909         if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
4910                 wp_sector_off == 0)
4911                 return 0;
4912
4913         f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
4914                     "curseg[0x%x,0x%x] wp[0x%x,0x%x]",
4915                     type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff);
4916
4917         f2fs_notice(sbi, "Assign new section to curseg[%d]: "
4918                     "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff);
4919
4920         f2fs_allocate_new_section(sbi, type, true);
4921
4922         /* check consistency of the zone curseg pointed to */
4923         if (check_zone_write_pointer(sbi, zbd, &zone))
4924                 return -EIO;
4925
4926         /* check newly assigned zone */
4927         cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4928         cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4929
4930         zbd = get_target_zoned_dev(sbi, cs_zone_block);
4931         if (!zbd)
4932                 return 0;
4933
4934         zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4935                 << log_sectors_per_block;
4936         err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4937                                   report_one_zone_cb, &zone);
4938         if (err != 1) {
4939                 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4940                          zbd->path, err);
4941                 return err;
4942         }
4943
4944         if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4945                 return 0;
4946
4947         if (zone.wp != zone.start) {
4948                 f2fs_notice(sbi,
4949                             "New zone for curseg[%d] is not yet discarded. "
4950                             "Reset the zone: curseg[0x%x,0x%x]",
4951                             type, cs->segno, cs->next_blkoff);
4952                 err = __f2fs_issue_discard_zone(sbi, zbd->bdev, cs_zone_block,
4953                                         zone.len >> log_sectors_per_block);
4954                 if (err) {
4955                         f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4956                                  zbd->path, err);
4957                         return err;
4958                 }
4959         }
4960
4961         return 0;
4962 }
4963
4964 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4965 {
4966         int i, ret;
4967
4968         for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4969                 ret = fix_curseg_write_pointer(sbi, i);
4970                 if (ret)
4971                         return ret;
4972         }
4973
4974         return 0;
4975 }
4976
4977 struct check_zone_write_pointer_args {
4978         struct f2fs_sb_info *sbi;
4979         struct f2fs_dev_info *fdev;
4980 };
4981
4982 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
4983                                       void *data)
4984 {
4985         struct check_zone_write_pointer_args *args;
4986
4987         args = (struct check_zone_write_pointer_args *)data;
4988
4989         return check_zone_write_pointer(args->sbi, args->fdev, zone);
4990 }
4991
4992 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4993 {
4994         int i, ret;
4995         struct check_zone_write_pointer_args args;
4996
4997         for (i = 0; i < sbi->s_ndevs; i++) {
4998                 if (!bdev_is_zoned(FDEV(i).bdev))
4999                         continue;
5000
5001                 args.sbi = sbi;
5002                 args.fdev = &FDEV(i);
5003                 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
5004                                           check_zone_write_pointer_cb, &args);
5005                 if (ret < 0)
5006                         return ret;
5007         }
5008
5009         return 0;
5010 }
5011
5012 /*
5013  * Return the number of usable blocks in a segment. The number of blocks
5014  * returned is always equal to the number of blocks in a segment for
5015  * segments fully contained within a sequential zone capacity or a
5016  * conventional zone. For segments partially contained in a sequential
5017  * zone capacity, the number of usable blocks up to the zone capacity
5018  * is returned. 0 is returned in all other cases.
5019  */
5020 static inline unsigned int f2fs_usable_zone_blks_in_seg(
5021                         struct f2fs_sb_info *sbi, unsigned int segno)
5022 {
5023         block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
5024         unsigned int secno;
5025
5026         if (!sbi->unusable_blocks_per_sec)
5027                 return sbi->blocks_per_seg;
5028
5029         secno = GET_SEC_FROM_SEG(sbi, segno);
5030         seg_start = START_BLOCK(sbi, segno);
5031         sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
5032         sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi);
5033
5034         /*
5035          * If segment starts before zone capacity and spans beyond
5036          * zone capacity, then usable blocks are from seg start to
5037          * zone capacity. If the segment starts after the zone capacity,
5038          * then there are no usable blocks.
5039          */
5040         if (seg_start >= sec_cap_blkaddr)
5041                 return 0;
5042         if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr)
5043                 return sec_cap_blkaddr - seg_start;
5044
5045         return sbi->blocks_per_seg;
5046 }
5047 #else
5048 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5049 {
5050         return 0;
5051 }
5052
5053 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5054 {
5055         return 0;
5056 }
5057
5058 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
5059                                                         unsigned int segno)
5060 {
5061         return 0;
5062 }
5063
5064 #endif
5065 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
5066                                         unsigned int segno)
5067 {
5068         if (f2fs_sb_has_blkzoned(sbi))
5069                 return f2fs_usable_zone_blks_in_seg(sbi, segno);
5070
5071         return sbi->blocks_per_seg;
5072 }
5073
5074 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi,
5075                                         unsigned int segno)
5076 {
5077         if (f2fs_sb_has_blkzoned(sbi))
5078                 return CAP_SEGS_PER_SEC(sbi);
5079
5080         return sbi->segs_per_sec;
5081 }
5082
5083 /*
5084  * Update min, max modified time for cost-benefit GC algorithm
5085  */
5086 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5087 {
5088         struct sit_info *sit_i = SIT_I(sbi);
5089         unsigned int segno;
5090
5091         down_write(&sit_i->sentry_lock);
5092
5093         sit_i->min_mtime = ULLONG_MAX;
5094
5095         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
5096                 unsigned int i;
5097                 unsigned long long mtime = 0;
5098
5099                 for (i = 0; i < sbi->segs_per_sec; i++)
5100                         mtime += get_seg_entry(sbi, segno + i)->mtime;
5101
5102                 mtime = div_u64(mtime, sbi->segs_per_sec);
5103
5104                 if (sit_i->min_mtime > mtime)
5105                         sit_i->min_mtime = mtime;
5106         }
5107         sit_i->max_mtime = get_mtime(sbi, false);
5108         sit_i->dirty_max_mtime = 0;
5109         up_write(&sit_i->sentry_lock);
5110 }
5111
5112 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
5113 {
5114         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5115         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
5116         struct f2fs_sm_info *sm_info;
5117         int err;
5118
5119         sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
5120         if (!sm_info)
5121                 return -ENOMEM;
5122
5123         /* init sm info */
5124         sbi->sm_info = sm_info;
5125         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5126         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5127         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5128         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5129         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5130         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5131         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
5132         sm_info->rec_prefree_segments = sm_info->main_segments *
5133                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
5134         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5135                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5136
5137         if (!f2fs_lfs_mode(sbi))
5138                 sm_info->ipu_policy = BIT(F2FS_IPU_FSYNC);
5139         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
5140         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
5141         sm_info->min_seq_blocks = sbi->blocks_per_seg;
5142         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
5143         sm_info->min_ssr_sections = reserved_sections(sbi);
5144
5145         INIT_LIST_HEAD(&sm_info->sit_entry_set);
5146
5147         init_f2fs_rwsem(&sm_info->curseg_lock);
5148
5149         err = f2fs_create_flush_cmd_control(sbi);
5150         if (err)
5151                 return err;
5152
5153         err = create_discard_cmd_control(sbi);
5154         if (err)
5155                 return err;
5156
5157         err = build_sit_info(sbi);
5158         if (err)
5159                 return err;
5160         err = build_free_segmap(sbi);
5161         if (err)
5162                 return err;
5163         err = build_curseg(sbi);
5164         if (err)
5165                 return err;
5166
5167         /* reinit free segmap based on SIT */
5168         err = build_sit_entries(sbi);
5169         if (err)
5170                 return err;
5171
5172         init_free_segmap(sbi);
5173         err = build_dirty_segmap(sbi);
5174         if (err)
5175                 return err;
5176
5177         err = sanity_check_curseg(sbi);
5178         if (err)
5179                 return err;
5180
5181         init_min_max_mtime(sbi);
5182         return 0;
5183 }
5184
5185 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5186                 enum dirty_type dirty_type)
5187 {
5188         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5189
5190         mutex_lock(&dirty_i->seglist_lock);
5191         kvfree(dirty_i->dirty_segmap[dirty_type]);
5192         dirty_i->nr_dirty[dirty_type] = 0;
5193         mutex_unlock(&dirty_i->seglist_lock);
5194 }
5195
5196 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5197 {
5198         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5199
5200         kvfree(dirty_i->pinned_secmap);
5201         kvfree(dirty_i->victim_secmap);
5202 }
5203
5204 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5205 {
5206         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5207         int i;
5208
5209         if (!dirty_i)
5210                 return;
5211
5212         /* discard pre-free/dirty segments list */
5213         for (i = 0; i < NR_DIRTY_TYPE; i++)
5214                 discard_dirty_segmap(sbi, i);
5215
5216         if (__is_large_section(sbi)) {
5217                 mutex_lock(&dirty_i->seglist_lock);
5218                 kvfree(dirty_i->dirty_secmap);
5219                 mutex_unlock(&dirty_i->seglist_lock);
5220         }
5221
5222         destroy_victim_secmap(sbi);
5223         SM_I(sbi)->dirty_info = NULL;
5224         kfree(dirty_i);
5225 }
5226
5227 static void destroy_curseg(struct f2fs_sb_info *sbi)
5228 {
5229         struct curseg_info *array = SM_I(sbi)->curseg_array;
5230         int i;
5231
5232         if (!array)
5233                 return;
5234         SM_I(sbi)->curseg_array = NULL;
5235         for (i = 0; i < NR_CURSEG_TYPE; i++) {
5236                 kfree(array[i].sum_blk);
5237                 kfree(array[i].journal);
5238         }
5239         kfree(array);
5240 }
5241
5242 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5243 {
5244         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5245
5246         if (!free_i)
5247                 return;
5248         SM_I(sbi)->free_info = NULL;
5249         kvfree(free_i->free_segmap);
5250         kvfree(free_i->free_secmap);
5251         kfree(free_i);
5252 }
5253
5254 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5255 {
5256         struct sit_info *sit_i = SIT_I(sbi);
5257
5258         if (!sit_i)
5259                 return;
5260
5261         if (sit_i->sentries)
5262                 kvfree(sit_i->bitmap);
5263         kfree(sit_i->tmp_map);
5264
5265         kvfree(sit_i->sentries);
5266         kvfree(sit_i->sec_entries);
5267         kvfree(sit_i->dirty_sentries_bitmap);
5268
5269         SM_I(sbi)->sit_info = NULL;
5270         kvfree(sit_i->sit_bitmap);
5271 #ifdef CONFIG_F2FS_CHECK_FS
5272         kvfree(sit_i->sit_bitmap_mir);
5273         kvfree(sit_i->invalid_segmap);
5274 #endif
5275         kfree(sit_i);
5276 }
5277
5278 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5279 {
5280         struct f2fs_sm_info *sm_info = SM_I(sbi);
5281
5282         if (!sm_info)
5283                 return;
5284         f2fs_destroy_flush_cmd_control(sbi, true);
5285         destroy_discard_cmd_control(sbi);
5286         destroy_dirty_segmap(sbi);
5287         destroy_curseg(sbi);
5288         destroy_free_segmap(sbi);
5289         destroy_sit_info(sbi);
5290         sbi->sm_info = NULL;
5291         kfree(sm_info);
5292 }
5293
5294 int __init f2fs_create_segment_manager_caches(void)
5295 {
5296         discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5297                         sizeof(struct discard_entry));
5298         if (!discard_entry_slab)
5299                 goto fail;
5300
5301         discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5302                         sizeof(struct discard_cmd));
5303         if (!discard_cmd_slab)
5304                 goto destroy_discard_entry;
5305
5306         sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5307                         sizeof(struct sit_entry_set));
5308         if (!sit_entry_set_slab)
5309                 goto destroy_discard_cmd;
5310
5311         revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry",
5312                         sizeof(struct revoke_entry));
5313         if (!revoke_entry_slab)
5314                 goto destroy_sit_entry_set;
5315         return 0;
5316
5317 destroy_sit_entry_set:
5318         kmem_cache_destroy(sit_entry_set_slab);
5319 destroy_discard_cmd:
5320         kmem_cache_destroy(discard_cmd_slab);
5321 destroy_discard_entry:
5322         kmem_cache_destroy(discard_entry_slab);
5323 fail:
5324         return -ENOMEM;
5325 }
5326
5327 void f2fs_destroy_segment_manager_caches(void)
5328 {
5329         kmem_cache_destroy(sit_entry_set_slab);
5330         kmem_cache_destroy(discard_cmd_slab);
5331         kmem_cache_destroy(discard_entry_slab);
5332         kmem_cache_destroy(revoke_entry_slab);
5333 }