fs/f2fs/segment.c

   1 /*
   2  * fs/f2fs/segment.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/bio.h>
  14 #include <linux/blkdev.h>
  15 #include <linux/prefetch.h>
  16 #include <linux/kthread.h>
  17 #include <linux/swap.h>
  18 #include <linux/timer.h>
  19 #include <linux/freezer.h>
  20 #include <linux/sched/signal.h>
  21
  22 #include "f2fs.h"
  23 #include "segment.h"
  24 #include "node.h"
  25 #include "gc.h"
  26 #include "trace.h"
  27 #include <trace/events/f2fs.h>
  28
  29 #define __reverse_ffz(x) __reverse_ffs(~(x))
  30
  31 static struct kmem_cache *discard_entry_slab;
  32 static struct kmem_cache *discard_cmd_slab;
  33 static struct kmem_cache *sit_entry_set_slab;
  34 static struct kmem_cache *inmem_entry_slab;
  35
  36 static unsigned long __reverse_ulong(unsigned char *str)
  37 {
  38         unsigned long tmp = 0;
  39         int shift = 24, idx = 0;
  40
  41 #if BITS_PER_LONG == 64
  42         shift = 56;
  43 #endif
  44         while (shift >= 0) {
  45                 tmp |= (unsigned long)str[idx++] << shift;
  46                 shift -= BITS_PER_BYTE;
  47         }
  48         return tmp;
  49 }
  50
  51 /*
  52  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  53  * MSB and LSB are reversed in a byte by f2fs_set_bit.
  54  */
  55 static inline unsigned long __reverse_ffs(unsigned long word)
  56 {
  57         int num = 0;
  58
  59 #if BITS_PER_LONG == 64
  60         if ((word & 0xffffffff00000000UL) == 0)
  61                 num += 32;
  62         else
  63                 word >>= 32;
  64 #endif
  65         if ((word & 0xffff0000) == 0)
  66                 num += 16;
  67         else
  68                 word >>= 16;
  69
  70         if ((word & 0xff00) == 0)
  71                 num += 8;
  72         else
  73                 word >>= 8;
  74
  75         if ((word & 0xf0) == 0)
  76                 num += 4;
  77         else
  78                 word >>= 4;
  79
  80         if ((word & 0xc) == 0)
  81                 num += 2;
  82         else
  83                 word >>= 2;
  84
  85         if ((word & 0x2) == 0)
  86                 num += 1;
  87         return num;
  88 }
  89
  90 /*
  91  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  92  * f2fs_set_bit makes MSB and LSB reversed in a byte.
  93  * @size must be integral times of unsigned long.
  94  * Example:
  95  *                             MSB <--> LSB
  96  *   f2fs_set_bit(0, bitmap) => 1000 0000
  97  *   f2fs_set_bit(7, bitmap) => 0000 0001
  98  */
  99 static unsigned long __find_rev_next_bit(const unsigned long *addr,
 100                         unsigned long size, unsigned long offset)
 101 {
 102         const unsigned long *p = addr + BIT_WORD(offset);
 103         unsigned long result = size;
 104         unsigned long tmp;
 105
 106         if (offset >= size)
 107                 return size;
 108
 109         size -= (offset & ~(BITS_PER_LONG - 1));
 110         offset %= BITS_PER_LONG;
 111
 112         while (1) {
 113                 if (*p == 0)
 114                         goto pass;
 115
 116                 tmp = __reverse_ulong((unsigned char *)p);
 117
 118                 tmp &= ~0UL >> offset;
 119                 if (size < BITS_PER_LONG)
 120                         tmp &= (~0UL << (BITS_PER_LONG - size));
 121                 if (tmp)
 122                         goto found;
 123 pass:
 124                 if (size <= BITS_PER_LONG)
 125                         break;
 126                 size -= BITS_PER_LONG;
 127                 offset = 0;
 128                 p++;
 129         }
 130         return result;
 131 found:
 132         return result - size + __reverse_ffs(tmp);
 133 }
 134
 135 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 136                         unsigned long size, unsigned long offset)
 137 {
 138         const unsigned long *p = addr + BIT_WORD(offset);
 139         unsigned long result = size;
 140         unsigned long tmp;
 141
 142         if (offset >= size)
 143                 return size;
 144
 145         size -= (offset & ~(BITS_PER_LONG - 1));
 146         offset %= BITS_PER_LONG;
 147
 148         while (1) {
 149                 if (*p == ~0UL)
 150                         goto pass;
 151
 152                 tmp = __reverse_ulong((unsigned char *)p);
 153
 154                 if (offset)
 155                         tmp |= ~0UL << (BITS_PER_LONG - offset);
 156                 if (size < BITS_PER_LONG)
 157                         tmp |= ~0UL >> size;
 158                 if (tmp != ~0UL)
 159                         goto found;
 160 pass:
 161                 if (size <= BITS_PER_LONG)
 162                         break;
 163                 size -= BITS_PER_LONG;
 164                 offset = 0;
 165                 p++;
 166         }
 167         return result;
 168 found:
 169         return result - size + __reverse_ffz(tmp);
 170 }
 171
 172 bool need_SSR(struct f2fs_sb_info *sbi)
 173 {
 174         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 175         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 176         int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
 177
 178         if (test_opt(sbi, LFS))
 179                 return false;
 180         if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
 181                 return true;
 182
 183         return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
 184                                                 2 * reserved_sections(sbi));
 185 }
 186
 187 void register_inmem_page(struct inode *inode, struct page *page)
 188 {
 189         struct f2fs_inode_info *fi = F2FS_I(inode);
 190         struct inmem_pages *new;
 191
 192         f2fs_trace_pid(page);
 193
 194         set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
 195         SetPagePrivate(page);
 196
 197         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 198
 199         /* add atomic page indices to the list */
 200         new->page = page;
 201         INIT_LIST_HEAD(&new->list);
 202
 203         /* increase reference count with clean state */
 204         mutex_lock(&fi->inmem_lock);
 205         get_page(page);
 206         list_add_tail(&new->list, &fi->inmem_pages);
 207         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 208         mutex_unlock(&fi->inmem_lock);
 209
 210         trace_f2fs_register_inmem_page(page, INMEM);
 211 }
 212
 213 static int __revoke_inmem_pages(struct inode *inode,
 214                                 struct list_head *head, bool drop, bool recover)
 215 {
 216         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 217         struct inmem_pages *cur, *tmp;
 218         int err = 0;
 219
 220         list_for_each_entry_safe(cur, tmp, head, list) {
 221                 struct page *page = cur->page;
 222
 223                 if (drop)
 224                         trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 225
 226                 lock_page(page);
 227
 228                 if (recover) {
 229                         struct dnode_of_data dn;
 230                         struct node_info ni;
 231
 232                         trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 233 retry:
 234                         set_new_dnode(&dn, inode, NULL, NULL, 0);
 235                         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
 236                         if (err) {
 237                                 if (err == -ENOMEM) {
 238                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 239                                         cond_resched();
 240                                         goto retry;
 241                                 }
 242                                 err = -EAGAIN;
 243                                 goto next;
 244                         }
 245                         get_node_info(sbi, dn.nid, &ni);
 246                         f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 247                                         cur->old_addr, ni.version, true, true);
 248                         f2fs_put_dnode(&dn);
 249                 }
 250 next:
 251                 /* we don't need to invalidate this in the sccessful status */
 252                 if (drop || recover)
 253                         ClearPageUptodate(page);
 254                 set_page_private(page, 0);
 255                 ClearPagePrivate(page);
 256                 f2fs_put_page(page, 1);
 257
 258                 list_del(&cur->list);
 259                 kmem_cache_free(inmem_entry_slab, cur);
 260                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 261         }
 262         return err;
 263 }
 264
 265 void drop_inmem_pages(struct inode *inode)
 266 {
 267         struct f2fs_inode_info *fi = F2FS_I(inode);
 268
 269         mutex_lock(&fi->inmem_lock);
 270         __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 271         mutex_unlock(&fi->inmem_lock);
 272
 273         clear_inode_flag(inode, FI_ATOMIC_FILE);
 274         clear_inode_flag(inode, FI_HOT_DATA);
 275         stat_dec_atomic_write(inode);
 276 }
 277
 278 void drop_inmem_page(struct inode *inode, struct page *page)
 279 {
 280         struct f2fs_inode_info *fi = F2FS_I(inode);
 281         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 282         struct list_head *head = &fi->inmem_pages;
 283         struct inmem_pages *cur = NULL;
 284
 285         f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
 286
 287         mutex_lock(&fi->inmem_lock);
 288         list_for_each_entry(cur, head, list) {
 289                 if (cur->page == page)
 290                         break;
 291         }
 292
 293         f2fs_bug_on(sbi, !cur || cur->page != page);
 294         list_del(&cur->list);
 295         mutex_unlock(&fi->inmem_lock);
 296
 297         dec_page_count(sbi, F2FS_INMEM_PAGES);
 298         kmem_cache_free(inmem_entry_slab, cur);
 299
 300         ClearPageUptodate(page);
 301         set_page_private(page, 0);
 302         ClearPagePrivate(page);
 303         f2fs_put_page(page, 0);
 304
 305         trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 306 }
 307
 308 static int __commit_inmem_pages(struct inode *inode,
 309                                         struct list_head *revoke_list)
 310 {
 311         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 312         struct f2fs_inode_info *fi = F2FS_I(inode);
 313         struct inmem_pages *cur, *tmp;
 314         struct f2fs_io_info fio = {
 315                 .sbi = sbi,
 316                 .type = DATA,
 317                 .op = REQ_OP_WRITE,
 318                 .op_flags = REQ_SYNC | REQ_PRIO,
 319                 .io_type = FS_DATA_IO,
 320         };
 321         pgoff_t last_idx = ULONG_MAX;
 322         int err = 0;
 323
 324         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 325                 struct page *page = cur->page;
 326
 327                 lock_page(page);
 328                 if (page->mapping == inode->i_mapping) {
 329                         trace_f2fs_commit_inmem_page(page, INMEM);
 330
 331                         set_page_dirty(page);
 332                         f2fs_wait_on_page_writeback(page, DATA, true);
 333                         if (clear_page_dirty_for_io(page)) {
 334                                 inode_dec_dirty_pages(inode);
 335                                 remove_dirty_inode(inode);
 336                         }
 337 retry:
 338                         fio.page = page;
 339                         fio.old_blkaddr = NULL_ADDR;
 340                         fio.encrypted_page = NULL;
 341                         fio.need_lock = LOCK_DONE;
 342                         err = do_write_data_page(&fio);
 343                         if (err) {
 344                                 if (err == -ENOMEM) {
 345                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 346                                         cond_resched();
 347                                         goto retry;
 348                                 }
 349                                 unlock_page(page);
 350                                 break;
 351                         }
 352                         /* record old blkaddr for revoking */
 353                         cur->old_addr = fio.old_blkaddr;
 354                         last_idx = page->index;
 355                 }
 356                 unlock_page(page);
 357                 list_move_tail(&cur->list, revoke_list);
 358         }
 359
 360         if (last_idx != ULONG_MAX)
 361                 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
 362
 363         if (!err)
 364                 __revoke_inmem_pages(inode, revoke_list, false, false);
 365
 366         return err;
 367 }
 368
 369 int commit_inmem_pages(struct inode *inode)
 370 {
 371         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 372         struct f2fs_inode_info *fi = F2FS_I(inode);
 373         struct list_head revoke_list;
 374         int err;
 375
 376         INIT_LIST_HEAD(&revoke_list);
 377         f2fs_balance_fs(sbi, true);
 378         f2fs_lock_op(sbi);
 379
 380         set_inode_flag(inode, FI_ATOMIC_COMMIT);
 381
 382         mutex_lock(&fi->inmem_lock);
 383         err = __commit_inmem_pages(inode, &revoke_list);
 384         if (err) {
 385                 int ret;
 386                 /*
 387                  * try to revoke all committed pages, but still we could fail
 388                  * due to no memory or other reason, if that happened, EAGAIN
 389                  * will be returned, which means in such case, transaction is
 390                  * already not integrity, caller should use journal to do the
 391                  * recovery or rewrite & commit last transaction. For other
 392                  * error number, revoking was done by filesystem itself.
 393                  */
 394                 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
 395                 if (ret)
 396                         err = ret;
 397
 398                 /* drop all uncommitted pages */
 399                 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 400         }
 401         mutex_unlock(&fi->inmem_lock);
 402
 403         clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 404
 405         f2fs_unlock_op(sbi);
 406         return err;
 407 }
 408
 409 /*
 410  * This function balances dirty node and dentry pages.
 411  * In addition, it controls garbage collection.
 412  */
 413 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 414 {
 415 #ifdef CONFIG_F2FS_FAULT_INJECTION
 416         if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 417                 f2fs_show_injection_info(FAULT_CHECKPOINT);
 418                 f2fs_stop_checkpoint(sbi, false);
 419         }
 420 #endif
 421
 422         /* balance_fs_bg is able to be pending */
 423         if (need && excess_cached_nats(sbi))
 424                 f2fs_balance_fs_bg(sbi);
 425
 426         /*
 427          * We should do GC or end up with checkpoint, if there are so many dirty
 428          * dir/node pages without enough free segments.
 429          */
 430         if (has_not_enough_free_secs(sbi, 0, 0)) {
 431                 mutex_lock(&sbi->gc_mutex);
 432                 f2fs_gc(sbi, false, false, NULL_SEGNO);
 433         }
 434 }
 435
 436 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 437 {
 438         /* try to shrink extent cache when there is no enough memory */
 439         if (!available_free_memory(sbi, EXTENT_CACHE))
 440                 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 441
 442         /* check the # of cached NAT entries */
 443         if (!available_free_memory(sbi, NAT_ENTRIES))
 444                 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 445
 446         if (!available_free_memory(sbi, FREE_NIDS))
 447                 try_to_free_nids(sbi, MAX_FREE_NIDS);
 448         else
 449                 build_free_nids(sbi, false, false);
 450
 451         if (!is_idle(sbi) && !excess_dirty_nats(sbi))
 452                 return;
 453
 454         /* checkpoint is the only way to shrink partial cached entries */
 455         if (!available_free_memory(sbi, NAT_ENTRIES) ||
 456                         !available_free_memory(sbi, INO_ENTRIES) ||
 457                         excess_prefree_segs(sbi) ||
 458                         excess_dirty_nats(sbi) ||
 459                         f2fs_time_over(sbi, CP_TIME)) {
 460                 if (test_opt(sbi, DATA_FLUSH)) {
 461                         struct blk_plug plug;
 462
 463                         blk_start_plug(&plug);
 464                         sync_dirty_inodes(sbi, FILE_INODE);
 465                         blk_finish_plug(&plug);
 466                 }
 467                 f2fs_sync_fs(sbi->sb, true);
 468                 stat_inc_bg_cp_count(sbi->stat_info);
 469         }
 470 }
 471
 472 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 473                                 struct block_device *bdev)
 474 {
 475         struct bio *bio = f2fs_bio_alloc(0);
 476         int ret;
 477
 478         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
 479         bio_set_dev(bio, bdev);
 480         ret = submit_bio_wait(bio);
 481         bio_put(bio);
 482
 483         trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 484                                 test_opt(sbi, FLUSH_MERGE), ret);
 485         return ret;
 486 }
 487
 488 static int submit_flush_wait(struct f2fs_sb_info *sbi)
 489 {
 490         int ret = __submit_flush_wait(sbi, sbi->sb->s_bdev);
 491         int i;
 492
 493         if (!sbi->s_ndevs || ret)
 494                 return ret;
 495
 496         for (i = 1; i < sbi->s_ndevs; i++) {
 497                 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 498                 if (ret)
 499                         break;
 500         }
 501         return ret;
 502 }
 503
 504 static int issue_flush_thread(void *data)
 505 {
 506         struct f2fs_sb_info *sbi = data;
 507         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 508         wait_queue_head_t *q = &fcc->flush_wait_queue;
 509 repeat:
 510         if (kthread_should_stop())
 511                 return 0;
 512
 513         sb_start_intwrite(sbi->sb);
 514
 515         if (!llist_empty(&fcc->issue_list)) {
 516                 struct flush_cmd *cmd, *next;
 517                 int ret;
 518
 519                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 520                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 521
 522                 ret = submit_flush_wait(sbi);
 523                 atomic_inc(&fcc->issued_flush);
 524
 525                 llist_for_each_entry_safe(cmd, next,
 526                                           fcc->dispatch_list, llnode) {
 527                         cmd->ret = ret;
 528                         complete(&cmd->wait);
 529                 }
 530                 fcc->dispatch_list = NULL;
 531         }
 532
 533         sb_end_intwrite(sbi->sb);
 534
 535         wait_event_interruptible(*q,
 536                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
 537         goto repeat;
 538 }
 539
 540 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
 541 {
 542         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 543         struct flush_cmd cmd;
 544         int ret;
 545
 546         if (test_opt(sbi, NOBARRIER))
 547                 return 0;
 548
 549         if (!test_opt(sbi, FLUSH_MERGE)) {
 550                 ret = submit_flush_wait(sbi);
 551                 atomic_inc(&fcc->issued_flush);
 552                 return ret;
 553         }
 554
 555         if (atomic_inc_return(&fcc->issing_flush) == 1) {
 556                 ret = submit_flush_wait(sbi);
 557                 atomic_dec(&fcc->issing_flush);
 558
 559                 atomic_inc(&fcc->issued_flush);
 560                 return ret;
 561         }
 562
 563         init_completion(&cmd.wait);
 564
 565         llist_add(&cmd.llnode, &fcc->issue_list);
 566
 567         /* update issue_list before we wake up issue_flush thread */
 568         smp_mb();
 569
 570         if (waitqueue_active(&fcc->flush_wait_queue))
 571                 wake_up(&fcc->flush_wait_queue);
 572
 573         if (fcc->f2fs_issue_flush) {
 574                 wait_for_completion(&cmd.wait);
 575                 atomic_dec(&fcc->issing_flush);
 576         } else {
 577                 struct llist_node *list;
 578
 579                 list = llist_del_all(&fcc->issue_list);
 580                 if (!list) {
 581                         wait_for_completion(&cmd.wait);
 582                         atomic_dec(&fcc->issing_flush);
 583                 } else {
 584                         struct flush_cmd *tmp, *next;
 585
 586                         ret = submit_flush_wait(sbi);
 587
 588                         llist_for_each_entry_safe(tmp, next, list, llnode) {
 589                                 if (tmp == &cmd) {
 590                                         cmd.ret = ret;
 591                                         atomic_dec(&fcc->issing_flush);
 592                                         continue;
 593                                 }
 594                                 tmp->ret = ret;
 595                                 complete(&tmp->wait);
 596                         }
 597                 }
 598         }
 599
 600         return cmd.ret;
 601 }
 602
 603 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
 604 {
 605         dev_t dev = sbi->sb->s_bdev->bd_dev;
 606         struct flush_cmd_control *fcc;
 607         int err = 0;
 608
 609         if (SM_I(sbi)->fcc_info) {
 610                 fcc = SM_I(sbi)->fcc_info;
 611                 if (fcc->f2fs_issue_flush)
 612                         return err;
 613                 goto init_thread;
 614         }
 615
 616         fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
 617         if (!fcc)
 618                 return -ENOMEM;
 619         atomic_set(&fcc->issued_flush, 0);
 620         atomic_set(&fcc->issing_flush, 0);
 621         init_waitqueue_head(&fcc->flush_wait_queue);
 622         init_llist_head(&fcc->issue_list);
 623         SM_I(sbi)->fcc_info = fcc;
 624         if (!test_opt(sbi, FLUSH_MERGE))
 625                 return err;
 626
 627 init_thread:
 628         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 629                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 630         if (IS_ERR(fcc->f2fs_issue_flush)) {
 631                 err = PTR_ERR(fcc->f2fs_issue_flush);
 632                 kfree(fcc);
 633                 SM_I(sbi)->fcc_info = NULL;
 634                 return err;
 635         }
 636
 637         return err;
 638 }
 639
 640 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 641 {
 642         struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 643
 644         if (fcc && fcc->f2fs_issue_flush) {
 645                 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 646
 647                 fcc->f2fs_issue_flush = NULL;
 648                 kthread_stop(flush_thread);
 649         }
 650         if (free) {
 651                 kfree(fcc);
 652                 SM_I(sbi)->fcc_info = NULL;
 653         }
 654 }
 655
 656 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 657                 enum dirty_type dirty_type)
 658 {
 659         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 660
 661         /* need not be added */
 662         if (IS_CURSEG(sbi, segno))
 663                 return;
 664
 665         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 666                 dirty_i->nr_dirty[dirty_type]++;
 667
 668         if (dirty_type == DIRTY) {
 669                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 670                 enum dirty_type t = sentry->type;
 671
 672                 if (unlikely(t >= DIRTY)) {
 673                         f2fs_bug_on(sbi, 1);
 674                         return;
 675                 }
 676                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 677                         dirty_i->nr_dirty[t]++;
 678         }
 679 }
 680
 681 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 682                 enum dirty_type dirty_type)
 683 {
 684         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 685
 686         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 687                 dirty_i->nr_dirty[dirty_type]--;
 688
 689         if (dirty_type == DIRTY) {
 690                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
 691                 enum dirty_type t = sentry->type;
 692
 693                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 694                         dirty_i->nr_dirty[t]--;
 695
 696                 if (get_valid_blocks(sbi, segno, true) == 0)
 697                         clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 698                                                 dirty_i->victim_secmap);
 699         }
 700 }
 701
 702 /*
 703  * Should not occur error such as -ENOMEM.
 704  * Adding dirty entry into seglist is not critical operation.
 705  * If a given segment is one of current working segments, it won't be added.
 706  */
 707 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 708 {
 709         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 710         unsigned short valid_blocks;
 711
 712         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 713                 return;
 714
 715         mutex_lock(&dirty_i->seglist_lock);
 716
 717         valid_blocks = get_valid_blocks(sbi, segno, false);
 718
 719         if (valid_blocks == 0) {
 720                 __locate_dirty_segment(sbi, segno, PRE);
 721                 __remove_dirty_segment(sbi, segno, DIRTY);
 722         } else if (valid_blocks < sbi->blocks_per_seg) {
 723                 __locate_dirty_segment(sbi, segno, DIRTY);
 724         } else {
 725                 /* Recovery routine with SSR needs this */
 726                 __remove_dirty_segment(sbi, segno, DIRTY);
 727         }
 728
 729         mutex_unlock(&dirty_i->seglist_lock);
 730 }
 731
 732 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 733                 struct block_device *bdev, block_t lstart,
 734                 block_t start, block_t len)
 735 {
 736         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 737         struct list_head *pend_list;
 738         struct discard_cmd *dc;
 739
 740         f2fs_bug_on(sbi, !len);
 741
 742         pend_list = &dcc->pend_list[plist_idx(len)];
 743
 744         dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 745         INIT_LIST_HEAD(&dc->list);
 746         dc->bdev = bdev;
 747         dc->lstart = lstart;
 748         dc->start = start;
 749         dc->len = len;
 750         dc->ref = 0;
 751         dc->state = D_PREP;
 752         dc->error = 0;
 753         init_completion(&dc->wait);
 754         list_add_tail(&dc->list, pend_list);
 755         atomic_inc(&dcc->discard_cmd_cnt);
 756         dcc->undiscard_blks += len;
 757
 758         return dc;
 759 }
 760
 761 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 762                                 struct block_device *bdev, block_t lstart,
 763                                 block_t start, block_t len,
 764                                 struct rb_node *parent, struct rb_node **p)
 765 {
 766         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 767         struct discard_cmd *dc;
 768
 769         dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
 770
 771         rb_link_node(&dc->rb_node, parent, p);
 772         rb_insert_color(&dc->rb_node, &dcc->root);
 773
 774         return dc;
 775 }
 776
 777 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
 778                                                         struct discard_cmd *dc)
 779 {
 780         if (dc->state == D_DONE)
 781                 atomic_dec(&dcc->issing_discard);
 782
 783         list_del(&dc->list);
 784         rb_erase(&dc->rb_node, &dcc->root);
 785         dcc->undiscard_blks -= dc->len;
 786
 787         kmem_cache_free(discard_cmd_slab, dc);
 788
 789         atomic_dec(&dcc->discard_cmd_cnt);
 790 }
 791
 792 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
 793                                                         struct discard_cmd *dc)
 794 {
 795         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 796
 797         f2fs_bug_on(sbi, dc->ref);
 798
 799         if (dc->error == -EOPNOTSUPP)
 800                 dc->error = 0;
 801
 802         if (dc->error)
 803                 f2fs_msg(sbi->sb, KERN_INFO,
 804                         "Issue discard(%u, %u, %u) failed, ret: %d",
 805                         dc->lstart, dc->start, dc->len, dc->error);
 806         __detach_discard_cmd(dcc, dc);
 807 }
 808
 809 static void f2fs_submit_discard_endio(struct bio *bio)
 810 {
 811         struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
 812
 813         dc->error = blk_status_to_errno(bio->bi_status);
 814         dc->state = D_DONE;
 815         complete_all(&dc->wait);
 816         bio_put(bio);
 817 }
 818
 819 void __check_sit_bitmap(struct f2fs_sb_info *sbi,
 820                                 block_t start, block_t end)
 821 {
 822 #ifdef CONFIG_F2FS_CHECK_FS
 823         struct seg_entry *sentry;
 824         unsigned int segno;
 825         block_t blk = start;
 826         unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
 827         unsigned long *map;
 828
 829         while (blk < end) {
 830                 segno = GET_SEGNO(sbi, blk);
 831                 sentry = get_seg_entry(sbi, segno);
 832                 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
 833
 834                 if (end < START_BLOCK(sbi, segno + 1))
 835                         size = GET_BLKOFF_FROM_SEG0(sbi, end);
 836                 else
 837                         size = max_blocks;
 838                 map = (unsigned long *)(sentry->cur_valid_map);
 839                 offset = __find_rev_next_bit(map, size, offset);
 840                 f2fs_bug_on(sbi, offset != size);
 841                 blk = START_BLOCK(sbi, segno + 1);
 842         }
 843 #endif
 844 }
 845
 846 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
 847 static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
 848                                 struct discard_cmd *dc)
 849 {
 850         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 851         struct bio *bio = NULL;
 852
 853         if (dc->state != D_PREP)
 854                 return;
 855
 856         trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
 857
 858         dc->error = __blkdev_issue_discard(dc->bdev,
 859                                 SECTOR_FROM_BLOCK(dc->start),
 860                                 SECTOR_FROM_BLOCK(dc->len),
 861                                 GFP_NOFS, 0, &bio);
 862         if (!dc->error) {
 863                 /* should keep before submission to avoid D_DONE right away */
 864                 dc->state = D_SUBMIT;
 865                 atomic_inc(&dcc->issued_discard);
 866                 atomic_inc(&dcc->issing_discard);
 867                 if (bio) {
 868                         bio->bi_private = dc;
 869                         bio->bi_end_io = f2fs_submit_discard_endio;
 870                         bio->bi_opf |= REQ_SYNC;
 871                         submit_bio(bio);
 872                         list_move_tail(&dc->list, &dcc->wait_list);
 873                         __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
 874
 875                         f2fs_update_iostat(sbi, FS_DISCARD, 1);
 876                 }
 877         } else {
 878                 __remove_discard_cmd(sbi, dc);
 879         }
 880 }
 881
 882 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
 883                                 struct block_device *bdev, block_t lstart,
 884                                 block_t start, block_t len,
 885                                 struct rb_node **insert_p,
 886                                 struct rb_node *insert_parent)
 887 {
 888         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 889         struct rb_node **p = &dcc->root.rb_node;
 890         struct rb_node *parent = NULL;
 891         struct discard_cmd *dc = NULL;
 892
 893         if (insert_p && insert_parent) {
 894                 parent = insert_parent;
 895                 p = insert_p;
 896                 goto do_insert;
 897         }
 898
 899         p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
 900 do_insert:
 901         dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
 902         if (!dc)
 903                 return NULL;
 904
 905         return dc;
 906 }
 907
 908 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
 909                                                 struct discard_cmd *dc)
 910 {
 911         list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
 912 }
 913
 914 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
 915                                 struct discard_cmd *dc, block_t blkaddr)
 916 {
 917         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 918         struct discard_info di = dc->di;
 919         bool modified = false;
 920
 921         if (dc->state == D_DONE || dc->len == 1) {
 922                 __remove_discard_cmd(sbi, dc);
 923                 return;
 924         }
 925
 926         dcc->undiscard_blks -= di.len;
 927
 928         if (blkaddr > di.lstart) {
 929                 dc->len = blkaddr - dc->lstart;
 930                 dcc->undiscard_blks += dc->len;
 931                 __relocate_discard_cmd(dcc, dc);
 932                 modified = true;
 933         }
 934
 935         if (blkaddr < di.lstart + di.len - 1) {
 936                 if (modified) {
 937                         __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
 938                                         di.start + blkaddr + 1 - di.lstart,
 939                                         di.lstart + di.len - 1 - blkaddr,
 940                                         NULL, NULL);
 941                 } else {
 942                         dc->lstart++;
 943                         dc->len--;
 944                         dc->start++;
 945                         dcc->undiscard_blks += dc->len;
 946                         __relocate_discard_cmd(dcc, dc);
 947                 }
 948         }
 949 }
 950
 951 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
 952                                 struct block_device *bdev, block_t lstart,
 953                                 block_t start, block_t len)
 954 {
 955         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 956         struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
 957         struct discard_cmd *dc;
 958         struct discard_info di = {0};
 959         struct rb_node **insert_p = NULL, *insert_parent = NULL;
 960         block_t end = lstart + len;
 961
 962         mutex_lock(&dcc->cmd_lock);
 963
 964         dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
 965                                         NULL, lstart,
 966                                         (struct rb_entry **)&prev_dc,
 967                                         (struct rb_entry **)&next_dc,
 968                                         &insert_p, &insert_parent, true);
 969         if (dc)
 970                 prev_dc = dc;
 971
 972         if (!prev_dc) {
 973                 di.lstart = lstart;
 974                 di.len = next_dc ? next_dc->lstart - lstart : len;
 975                 di.len = min(di.len, len);
 976                 di.start = start;
 977         }
 978
 979         while (1) {
 980                 struct rb_node *node;
 981                 bool merged = false;
 982                 struct discard_cmd *tdc = NULL;
 983
 984                 if (prev_dc) {
 985                         di.lstart = prev_dc->lstart + prev_dc->len;
 986                         if (di.lstart < lstart)
 987                                 di.lstart = lstart;
 988                         if (di.lstart >= end)
 989                                 break;
 990
 991                         if (!next_dc || next_dc->lstart > end)
 992                                 di.len = end - di.lstart;
 993                         else
 994                                 di.len = next_dc->lstart - di.lstart;
 995                         di.start = start + di.lstart - lstart;
 996                 }
 997
 998                 if (!di.len)
 999                         goto next;
1000
1001                 if (prev_dc && prev_dc->state == D_PREP &&
1002                         prev_dc->bdev == bdev &&
1003                         __is_discard_back_mergeable(&di, &prev_dc->di)) {
1004                         prev_dc->di.len += di.len;
1005                         dcc->undiscard_blks += di.len;
1006                         __relocate_discard_cmd(dcc, prev_dc);
1007                         di = prev_dc->di;
1008                         tdc = prev_dc;
1009                         merged = true;
1010                 }
1011
1012                 if (next_dc && next_dc->state == D_PREP &&
1013                         next_dc->bdev == bdev &&
1014                         __is_discard_front_mergeable(&di, &next_dc->di)) {
1015                         next_dc->di.lstart = di.lstart;
1016                         next_dc->di.len += di.len;
1017                         next_dc->di.start = di.start;
1018                         dcc->undiscard_blks += di.len;
1019                         __relocate_discard_cmd(dcc, next_dc);
1020                         if (tdc)
1021                                 __remove_discard_cmd(sbi, tdc);
1022                         merged = true;
1023                 }
1024
1025                 if (!merged) {
1026                         __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1027                                                         di.len, NULL, NULL);
1028                 }
1029  next:
1030                 prev_dc = next_dc;
1031                 if (!prev_dc)
1032                         break;
1033
1034                 node = rb_next(&prev_dc->rb_node);
1035                 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1036         }
1037
1038         mutex_unlock(&dcc->cmd_lock);
1039 }
1040
1041 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1042                 struct block_device *bdev, block_t blkstart, block_t blklen)
1043 {
1044         block_t lblkstart = blkstart;
1045
1046         trace_f2fs_queue_discard(bdev, blkstart, blklen);
1047
1048         if (sbi->s_ndevs) {
1049                 int devi = f2fs_target_device_index(sbi, blkstart);
1050
1051                 blkstart -= FDEV(devi).start_blk;
1052         }
1053         __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1054         return 0;
1055 }
1056
1057 static int __issue_discard_cmd(struct f2fs_sb_info *sbi, bool issue_cond)
1058 {
1059         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1060         struct list_head *pend_list;
1061         struct discard_cmd *dc, *tmp;
1062         struct blk_plug plug;
1063         int iter = 0, issued = 0;
1064         int i;
1065         bool io_interrupted = false;
1066
1067         mutex_lock(&dcc->cmd_lock);
1068         f2fs_bug_on(sbi,
1069                 !__check_rb_tree_consistence(sbi, &dcc->root));
1070         blk_start_plug(&plug);
1071         for (i = MAX_PLIST_NUM - 1;
1072                         i >= 0 && plist_issue(dcc->pend_list_tag[i]); i--) {
1073                 pend_list = &dcc->pend_list[i];
1074                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1075                         f2fs_bug_on(sbi, dc->state != D_PREP);
1076
1077                         /* Hurry up to finish fstrim */
1078                         if (dcc->pend_list_tag[i] & P_TRIM) {
1079                                 __submit_discard_cmd(sbi, dc);
1080                                 issued++;
1081
1082                                 if (fatal_signal_pending(current))
1083                                         break;
1084                                 continue;
1085                         }
1086
1087                         if (!issue_cond) {
1088                                 __submit_discard_cmd(sbi, dc);
1089                                 issued++;
1090                                 continue;
1091                         }
1092
1093                         if (is_idle(sbi)) {
1094                                 __submit_discard_cmd(sbi, dc);
1095                                 issued++;
1096                         } else {
1097                                 io_interrupted = true;
1098                         }
1099
1100                         if (++iter >= DISCARD_ISSUE_RATE)
1101                                 goto out;
1102                 }
1103                 if (list_empty(pend_list) && dcc->pend_list_tag[i] & P_TRIM)
1104                         dcc->pend_list_tag[i] &= (~P_TRIM);
1105         }
1106 out:
1107         blk_finish_plug(&plug);
1108         mutex_unlock(&dcc->cmd_lock);
1109
1110         if (!issued && io_interrupted)
1111                 issued = -1;
1112
1113         return issued;
1114 }
1115
1116 static void __drop_discard_cmd(struct f2fs_sb_info *sbi)
1117 {
1118         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1119         struct list_head *pend_list;
1120         struct discard_cmd *dc, *tmp;
1121         int i;
1122
1123         mutex_lock(&dcc->cmd_lock);
1124         for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1125                 pend_list = &dcc->pend_list[i];
1126                 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1127                         f2fs_bug_on(sbi, dc->state != D_PREP);
1128                         __remove_discard_cmd(sbi, dc);
1129                 }
1130         }
1131         mutex_unlock(&dcc->cmd_lock);
1132 }
1133
1134 static void __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1135                                                         struct discard_cmd *dc)
1136 {
1137         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1138
1139         wait_for_completion_io(&dc->wait);
1140         mutex_lock(&dcc->cmd_lock);
1141         f2fs_bug_on(sbi, dc->state != D_DONE);
1142         dc->ref--;
1143         if (!dc->ref)
1144                 __remove_discard_cmd(sbi, dc);
1145         mutex_unlock(&dcc->cmd_lock);
1146 }
1147
1148 static void __wait_discard_cmd(struct f2fs_sb_info *sbi, bool wait_cond)
1149 {
1150         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1151         struct list_head *wait_list = &(dcc->wait_list);
1152         struct discard_cmd *dc, *tmp;
1153         bool need_wait;
1154
1155 next:
1156         need_wait = false;
1157
1158         mutex_lock(&dcc->cmd_lock);
1159         list_for_each_entry_safe(dc, tmp, wait_list, list) {
1160                 if (!wait_cond || (dc->state == D_DONE && !dc->ref)) {
1161                         wait_for_completion_io(&dc->wait);
1162                         __remove_discard_cmd(sbi, dc);
1163                 } else {
1164                         dc->ref++;
1165                         need_wait = true;
1166                         break;
1167                 }
1168         }
1169         mutex_unlock(&dcc->cmd_lock);
1170
1171         if (need_wait) {
1172                 __wait_one_discard_bio(sbi, dc);
1173                 goto next;
1174         }
1175 }
1176
1177 /* This should be covered by global mutex, &sit_i->sentry_lock */
1178 void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1179 {
1180         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1181         struct discard_cmd *dc;
1182         bool need_wait = false;
1183
1184         mutex_lock(&dcc->cmd_lock);
1185         dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1186         if (dc) {
1187                 if (dc->state == D_PREP) {
1188                         __punch_discard_cmd(sbi, dc, blkaddr);
1189                 } else {
1190                         dc->ref++;
1191                         need_wait = true;
1192                 }
1193         }
1194         mutex_unlock(&dcc->cmd_lock);
1195
1196         if (need_wait)
1197                 __wait_one_discard_bio(sbi, dc);
1198 }
1199
1200 void stop_discard_thread(struct f2fs_sb_info *sbi)
1201 {
1202         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1203
1204         if (dcc && dcc->f2fs_issue_discard) {
1205                 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1206
1207                 dcc->f2fs_issue_discard = NULL;
1208                 kthread_stop(discard_thread);
1209         }
1210 }
1211
1212 /* This comes from f2fs_put_super and f2fs_trim_fs */
1213 void f2fs_wait_discard_bios(struct f2fs_sb_info *sbi, bool umount)
1214 {
1215         __issue_discard_cmd(sbi, false);
1216         __drop_discard_cmd(sbi);
1217         __wait_discard_cmd(sbi, !umount);
1218 }
1219
1220 static void mark_discard_range_all(struct f2fs_sb_info *sbi)
1221 {
1222         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1223         int i;
1224
1225         mutex_lock(&dcc->cmd_lock);
1226         for (i = 0; i < MAX_PLIST_NUM; i++)
1227                 dcc->pend_list_tag[i] |= P_TRIM;
1228         mutex_unlock(&dcc->cmd_lock);
1229 }
1230
1231 static int issue_discard_thread(void *data)
1232 {
1233         struct f2fs_sb_info *sbi = data;
1234         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1235         wait_queue_head_t *q = &dcc->discard_wait_queue;
1236         unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1237         int issued;
1238
1239         set_freezable();
1240
1241         do {
1242                 wait_event_interruptible_timeout(*q,
1243                                 kthread_should_stop() || freezing(current) ||
1244                                 dcc->discard_wake,
1245                                 msecs_to_jiffies(wait_ms));
1246                 if (try_to_freeze())
1247                         continue;
1248                 if (kthread_should_stop())
1249                         return 0;
1250
1251                 if (dcc->discard_wake) {
1252                         dcc->discard_wake = 0;
1253                         if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
1254                                 mark_discard_range_all(sbi);
1255                 }
1256
1257                 sb_start_intwrite(sbi->sb);
1258
1259                 issued = __issue_discard_cmd(sbi, true);
1260                 if (issued) {
1261                         __wait_discard_cmd(sbi, true);
1262                         wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1263                 } else {
1264                         wait_ms = DEF_MAX_DISCARD_ISSUE_TIME;
1265                 }
1266
1267                 sb_end_intwrite(sbi->sb);
1268
1269         } while (!kthread_should_stop());
1270         return 0;
1271 }
1272
1273 #ifdef CONFIG_BLK_DEV_ZONED
1274 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1275                 struct block_device *bdev, block_t blkstart, block_t blklen)
1276 {
1277         sector_t sector, nr_sects;
1278         block_t lblkstart = blkstart;
1279         int devi = 0;
1280
1281         if (sbi->s_ndevs) {
1282                 devi = f2fs_target_device_index(sbi, blkstart);
1283                 blkstart -= FDEV(devi).start_blk;
1284         }
1285
1286         /*
1287          * We need to know the type of the zone: for conventional zones,
1288          * use regular discard if the drive supports it. For sequential
1289          * zones, reset the zone write pointer.
1290          */
1291         switch (get_blkz_type(sbi, bdev, blkstart)) {
1292
1293         case BLK_ZONE_TYPE_CONVENTIONAL:
1294                 if (!blk_queue_discard(bdev_get_queue(bdev)))
1295                         return 0;
1296                 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1297         case BLK_ZONE_TYPE_SEQWRITE_REQ:
1298         case BLK_ZONE_TYPE_SEQWRITE_PREF:
1299                 sector = SECTOR_FROM_BLOCK(blkstart);
1300                 nr_sects = SECTOR_FROM_BLOCK(blklen);
1301
1302                 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1303                                 nr_sects != bdev_zone_sectors(bdev)) {
1304                         f2fs_msg(sbi->sb, KERN_INFO,
1305                                 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1306                                 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1307                                 blkstart, blklen);
1308                         return -EIO;
1309                 }
1310                 trace_f2fs_issue_reset_zone(bdev, blkstart);
1311                 return blkdev_reset_zones(bdev, sector,
1312                                           nr_sects, GFP_NOFS);
1313         default:
1314                 /* Unknown zone type: broken device ? */
1315                 return -EIO;
1316         }
1317 }
1318 #endif
1319
1320 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1321                 struct block_device *bdev, block_t blkstart, block_t blklen)
1322 {
1323 #ifdef CONFIG_BLK_DEV_ZONED
1324         if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
1325                                 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1326                 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1327 #endif
1328         return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1329 }
1330
1331 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1332                                 block_t blkstart, block_t blklen)
1333 {
1334         sector_t start = blkstart, len = 0;
1335         struct block_device *bdev;
1336         struct seg_entry *se;
1337         unsigned int offset;
1338         block_t i;
1339         int err = 0;
1340
1341         bdev = f2fs_target_device(sbi, blkstart, NULL);
1342
1343         for (i = blkstart; i < blkstart + blklen; i++, len++) {
1344                 if (i != start) {
1345                         struct block_device *bdev2 =
1346                                 f2fs_target_device(sbi, i, NULL);
1347
1348                         if (bdev2 != bdev) {
1349                                 err = __issue_discard_async(sbi, bdev,
1350                                                 start, len);
1351                                 if (err)
1352                                         return err;
1353                                 bdev = bdev2;
1354                                 start = i;
1355                                 len = 0;
1356                         }
1357                 }
1358
1359                 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1360                 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1361
1362                 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1363                         sbi->discard_blks--;
1364         }
1365
1366         if (len)
1367                 err = __issue_discard_async(sbi, bdev, start, len);
1368         return err;
1369 }
1370
1371 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1372                                                         bool check_only)
1373 {
1374         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1375         int max_blocks = sbi->blocks_per_seg;
1376         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1377         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1378         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1379         unsigned long *discard_map = (unsigned long *)se->discard_map;
1380         unsigned long *dmap = SIT_I(sbi)->tmp_map;
1381         unsigned int start = 0, end = -1;
1382         bool force = (cpc->reason & CP_DISCARD);
1383         struct discard_entry *de = NULL;
1384         struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1385         int i;
1386
1387         if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
1388                 return false;
1389
1390         if (!force) {
1391                 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
1392                         SM_I(sbi)->dcc_info->nr_discards >=
1393                                 SM_I(sbi)->dcc_info->max_discards)
1394                         return false;
1395         }
1396
1397         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1398         for (i = 0; i < entries; i++)
1399                 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1400                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1401
1402         while (force || SM_I(sbi)->dcc_info->nr_discards <=
1403                                 SM_I(sbi)->dcc_info->max_discards) {
1404                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1405                 if (start >= max_blocks)
1406                         break;
1407
1408                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1409                 if (force && start && end != max_blocks
1410                                         && (end - start) < cpc->trim_minlen)
1411                         continue;
1412
1413                 if (check_only)
1414                         return true;
1415
1416                 if (!de) {
1417                         de = f2fs_kmem_cache_alloc(discard_entry_slab,
1418                                                                 GFP_F2FS_ZERO);
1419                         de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1420                         list_add_tail(&de->list, head);
1421                 }
1422
1423                 for (i = start; i < end; i++)
1424                         __set_bit_le(i, (void *)de->discard_map);
1425
1426                 SM_I(sbi)->dcc_info->nr_discards += end - start;
1427         }
1428         return false;
1429 }
1430
1431 void release_discard_addrs(struct f2fs_sb_info *sbi)
1432 {
1433         struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1434         struct discard_entry *entry, *this;
1435
1436         /* drop caches */
1437         list_for_each_entry_safe(entry, this, head, list) {
1438                 list_del(&entry->list);
1439                 kmem_cache_free(discard_entry_slab, entry);
1440         }
1441 }
1442
1443 /*
1444  * Should call clear_prefree_segments after checkpoint is done.
1445  */
1446 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1447 {
1448         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1449         unsigned int segno;
1450
1451         mutex_lock(&dirty_i->seglist_lock);
1452         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1453                 __set_test_and_free(sbi, segno);
1454         mutex_unlock(&dirty_i->seglist_lock);
1455 }
1456
1457 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1458 {
1459         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1460         struct list_head *head = &dcc->entry_list;
1461         struct discard_entry *entry, *this;
1462         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1463         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1464         unsigned int start = 0, end = -1;
1465         unsigned int secno, start_segno;
1466         bool force = (cpc->reason & CP_DISCARD);
1467
1468         mutex_lock(&dirty_i->seglist_lock);
1469
1470         while (1) {
1471                 int i;
1472                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1473                 if (start >= MAIN_SEGS(sbi))
1474                         break;
1475                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1476                                                                 start + 1);
1477
1478                 for (i = start; i < end; i++)
1479                         clear_bit(i, prefree_map);
1480
1481                 dirty_i->nr_dirty[PRE] -= end - start;
1482
1483                 if (!test_opt(sbi, DISCARD))
1484                         continue;
1485
1486                 if (force && start >= cpc->trim_start &&
1487                                         (end - 1) <= cpc->trim_end)
1488                                 continue;
1489
1490                 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1491                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1492                                 (end - start) << sbi->log_blocks_per_seg);
1493                         continue;
1494                 }
1495 next:
1496                 secno = GET_SEC_FROM_SEG(sbi, start);
1497                 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1498                 if (!IS_CURSEC(sbi, secno) &&
1499                         !get_valid_blocks(sbi, start, true))
1500                         f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1501                                 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1502
1503                 start = start_segno + sbi->segs_per_sec;
1504                 if (start < end)
1505                         goto next;
1506                 else
1507                         end = start - 1;
1508         }
1509         mutex_unlock(&dirty_i->seglist_lock);
1510
1511         /* send small discards */
1512         list_for_each_entry_safe(entry, this, head, list) {
1513                 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1514                 bool is_valid = test_bit_le(0, entry->discard_map);
1515
1516 find_next:
1517                 if (is_valid) {
1518                         next_pos = find_next_zero_bit_le(entry->discard_map,
1519                                         sbi->blocks_per_seg, cur_pos);
1520                         len = next_pos - cur_pos;
1521
1522                         if (f2fs_sb_mounted_blkzoned(sbi->sb) ||
1523                             (force && len < cpc->trim_minlen))
1524                                 goto skip;
1525
1526                         f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1527                                                                         len);
1528                         cpc->trimmed += len;
1529                         total_len += len;
1530                 } else {
1531                         next_pos = find_next_bit_le(entry->discard_map,
1532                                         sbi->blocks_per_seg, cur_pos);
1533                 }
1534 skip:
1535                 cur_pos = next_pos;
1536                 is_valid = !is_valid;
1537
1538                 if (cur_pos < sbi->blocks_per_seg)
1539                         goto find_next;
1540
1541                 list_del(&entry->list);
1542                 dcc->nr_discards -= total_len;
1543                 kmem_cache_free(discard_entry_slab, entry);
1544         }
1545
1546         wake_up_discard_thread(sbi, false);
1547 }
1548
1549 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1550 {
1551         dev_t dev = sbi->sb->s_bdev->bd_dev;
1552         struct discard_cmd_control *dcc;
1553         int err = 0, i;
1554
1555         if (SM_I(sbi)->dcc_info) {
1556                 dcc = SM_I(sbi)->dcc_info;
1557                 goto init_thread;
1558         }
1559
1560         dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1561         if (!dcc)
1562                 return -ENOMEM;
1563
1564         dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
1565         INIT_LIST_HEAD(&dcc->entry_list);
1566         for (i = 0; i < MAX_PLIST_NUM; i++) {
1567                 INIT_LIST_HEAD(&dcc->pend_list[i]);
1568                 if (i >= dcc->discard_granularity - 1)
1569                         dcc->pend_list_tag[i] |= P_ACTIVE;
1570         }
1571         INIT_LIST_HEAD(&dcc->wait_list);
1572         mutex_init(&dcc->cmd_lock);
1573         atomic_set(&dcc->issued_discard, 0);
1574         atomic_set(&dcc->issing_discard, 0);
1575         atomic_set(&dcc->discard_cmd_cnt, 0);
1576         dcc->nr_discards = 0;
1577         dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1578         dcc->undiscard_blks = 0;
1579         dcc->root = RB_ROOT;
1580
1581         init_waitqueue_head(&dcc->discard_wait_queue);
1582         SM_I(sbi)->dcc_info = dcc;
1583 init_thread:
1584         dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1585                                 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1586         if (IS_ERR(dcc->f2fs_issue_discard)) {
1587                 err = PTR_ERR(dcc->f2fs_issue_discard);
1588                 kfree(dcc);
1589                 SM_I(sbi)->dcc_info = NULL;
1590                 return err;
1591         }
1592
1593         return err;
1594 }
1595
1596 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1597 {
1598         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1599
1600         if (!dcc)
1601                 return;
1602
1603         stop_discard_thread(sbi);
1604
1605         kfree(dcc);
1606         SM_I(sbi)->dcc_info = NULL;
1607 }
1608
1609 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1610 {
1611         struct sit_info *sit_i = SIT_I(sbi);
1612
1613         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1614                 sit_i->dirty_sentries++;
1615                 return false;
1616         }
1617
1618         return true;
1619 }
1620
1621 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1622                                         unsigned int segno, int modified)
1623 {
1624         struct seg_entry *se = get_seg_entry(sbi, segno);
1625         se->type = type;
1626         if (modified)
1627                 __mark_sit_entry_dirty(sbi, segno);
1628 }
1629
1630 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1631 {
1632         struct seg_entry *se;
1633         unsigned int segno, offset;
1634         long int new_vblocks;
1635         bool exist;
1636 #ifdef CONFIG_F2FS_CHECK_FS
1637         bool mir_exist;
1638 #endif
1639
1640         segno = GET_SEGNO(sbi, blkaddr);
1641
1642         se = get_seg_entry(sbi, segno);
1643         new_vblocks = se->valid_blocks + del;
1644         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1645
1646         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
1647                                 (new_vblocks > sbi->blocks_per_seg)));
1648
1649         se->valid_blocks = new_vblocks;
1650         se->mtime = get_mtime(sbi);
1651         SIT_I(sbi)->max_mtime = se->mtime;
1652
1653         /* Update valid block bitmap */
1654         if (del > 0) {
1655                 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
1656 #ifdef CONFIG_F2FS_CHECK_FS
1657                 mir_exist = f2fs_test_and_set_bit(offset,
1658                                                 se->cur_valid_map_mir);
1659                 if (unlikely(exist != mir_exist)) {
1660                         f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1661                                 "when setting bitmap, blk:%u, old bit:%d",
1662                                 blkaddr, exist);
1663                         f2fs_bug_on(sbi, 1);
1664                 }
1665 #endif
1666                 if (unlikely(exist)) {
1667                         f2fs_msg(sbi->sb, KERN_ERR,
1668                                 "Bitmap was wrongly set, blk:%u", blkaddr);
1669                         f2fs_bug_on(sbi, 1);
1670                         se->valid_blocks--;
1671                         del = 0;
1672                 }
1673
1674                 if (f2fs_discard_en(sbi) &&
1675                         !f2fs_test_and_set_bit(offset, se->discard_map))
1676                         sbi->discard_blks--;
1677
1678                 /* don't overwrite by SSR to keep node chain */
1679                 if (se->type == CURSEG_WARM_NODE) {
1680                         if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1681                                 se->ckpt_valid_blocks++;
1682                 }
1683         } else {
1684                 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
1685 #ifdef CONFIG_F2FS_CHECK_FS
1686                 mir_exist = f2fs_test_and_clear_bit(offset,
1687                                                 se->cur_valid_map_mir);
1688                 if (unlikely(exist != mir_exist)) {
1689                         f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1690                                 "when clearing bitmap, blk:%u, old bit:%d",
1691                                 blkaddr, exist);
1692                         f2fs_bug_on(sbi, 1);
1693                 }
1694 #endif
1695                 if (unlikely(!exist)) {
1696                         f2fs_msg(sbi->sb, KERN_ERR,
1697                                 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1698                         f2fs_bug_on(sbi, 1);
1699                         se->valid_blocks++;
1700                         del = 0;
1701                 }
1702
1703                 if (f2fs_discard_en(sbi) &&
1704                         f2fs_test_and_clear_bit(offset, se->discard_map))
1705                         sbi->discard_blks++;
1706         }
1707         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1708                 se->ckpt_valid_blocks += del;
1709
1710         __mark_sit_entry_dirty(sbi, segno);
1711
1712         /* update total number of valid blocks to be written in ckpt area */
1713         SIT_I(sbi)->written_valid_blocks += del;
1714
1715         if (sbi->segs_per_sec > 1)
1716                 get_sec_entry(sbi, segno)->valid_blocks += del;
1717 }
1718
1719 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
1720 {
1721         update_sit_entry(sbi, new, 1);
1722         if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1723                 update_sit_entry(sbi, old, -1);
1724
1725         locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1726         locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
1727 }
1728
1729 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1730 {
1731         unsigned int segno = GET_SEGNO(sbi, addr);
1732         struct sit_info *sit_i = SIT_I(sbi);
1733
1734         f2fs_bug_on(sbi, addr == NULL_ADDR);
1735         if (addr == NEW_ADDR)
1736                 return;
1737
1738         /* add it into sit main buffer */
1739         mutex_lock(&sit_i->sentry_lock);
1740
1741         update_sit_entry(sbi, addr, -1);
1742
1743         /* add it into dirty seglist */
1744         locate_dirty_segment(sbi, segno);
1745
1746         mutex_unlock(&sit_i->sentry_lock);
1747 }
1748
1749 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1750 {
1751         struct sit_info *sit_i = SIT_I(sbi);
1752         unsigned int segno, offset;
1753         struct seg_entry *se;
1754         bool is_cp = false;
1755
1756         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1757                 return true;
1758
1759         mutex_lock(&sit_i->sentry_lock);
1760
1761         segno = GET_SEGNO(sbi, blkaddr);
1762         se = get_seg_entry(sbi, segno);
1763         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1764
1765         if (f2fs_test_bit(offset, se->ckpt_valid_map))
1766                 is_cp = true;
1767
1768         mutex_unlock(&sit_i->sentry_lock);
1769
1770         return is_cp;
1771 }
1772
1773 /*
1774  * This function should be resided under the curseg_mutex lock
1775  */
1776 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1777                                         struct f2fs_summary *sum)
1778 {
1779         struct curseg_info *curseg = CURSEG_I(sbi, type);
1780         void *addr = curseg->sum_blk;
1781         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1782         memcpy(addr, sum, sizeof(struct f2fs_summary));
1783 }
1784
1785 /*
1786  * Calculate the number of current summary pages for writing
1787  */
1788 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1789 {
1790         int valid_sum_count = 0;
1791         int i, sum_in_page;
1792
1793         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1794                 if (sbi->ckpt->alloc_type[i] == SSR)
1795                         valid_sum_count += sbi->blocks_per_seg;
1796                 else {
1797                         if (for_ra)
1798                                 valid_sum_count += le16_to_cpu(
1799                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1800                         else
1801                                 valid_sum_count += curseg_blkoff(sbi, i);
1802                 }
1803         }
1804
1805         sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1806                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1807         if (valid_sum_count <= sum_in_page)
1808                 return 1;
1809         else if ((valid_sum_count - sum_in_page) <=
1810                 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1811                 return 2;
1812         return 3;
1813 }
1814
1815 /*
1816  * Caller should put this summary page
1817  */
1818 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1819 {
1820         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1821 }
1822
1823 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1824 {
1825         struct page *page = grab_meta_page(sbi, blk_addr);
1826         void *dst = page_address(page);
1827
1828         if (src)
1829                 memcpy(dst, src, PAGE_SIZE);
1830         else
1831                 memset(dst, 0, PAGE_SIZE);
1832         set_page_dirty(page);
1833         f2fs_put_page(page, 1);
1834 }
1835
1836 static void write_sum_page(struct f2fs_sb_info *sbi,
1837                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
1838 {
1839         update_meta_page(sbi, (void *)sum_blk, blk_addr);
1840 }
1841
1842 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1843                                                 int type, block_t blk_addr)
1844 {
1845         struct curseg_info *curseg = CURSEG_I(sbi, type);
1846         struct page *page = grab_meta_page(sbi, blk_addr);
1847         struct f2fs_summary_block *src = curseg->sum_blk;
1848         struct f2fs_summary_block *dst;
1849
1850         dst = (struct f2fs_summary_block *)page_address(page);
1851
1852         mutex_lock(&curseg->curseg_mutex);
1853
1854         down_read(&curseg->journal_rwsem);
1855         memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1856         up_read(&curseg->journal_rwsem);
1857
1858         memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1859         memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1860
1861         mutex_unlock(&curseg->curseg_mutex);
1862
1863         set_page_dirty(page);
1864         f2fs_put_page(page, 1);
1865 }
1866
1867 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1868 {
1869         struct curseg_info *curseg = CURSEG_I(sbi, type);
1870         unsigned int segno = curseg->segno + 1;
1871         struct free_segmap_info *free_i = FREE_I(sbi);
1872
1873         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
1874                 return !test_bit(segno, free_i->free_segmap);
1875         return 0;
1876 }
1877
1878 /*
1879  * Find a new segment from the free segments bitmap to right order
1880  * This function should be returned with success, otherwise BUG
1881  */
1882 static void get_new_segment(struct f2fs_sb_info *sbi,
1883                         unsigned int *newseg, bool new_sec, int dir)
1884 {
1885         struct free_segmap_info *free_i = FREE_I(sbi);
1886         unsigned int segno, secno, zoneno;
1887         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
1888         unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
1889         unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
1890         unsigned int left_start = hint;
1891         bool init = true;
1892         int go_left = 0;
1893         int i;
1894
1895         spin_lock(&free_i->segmap_lock);
1896
1897         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1898                 segno = find_next_zero_bit(free_i->free_segmap,
1899                         GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
1900                 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
1901                         goto got_it;
1902         }
1903 find_other_zone:
1904         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1905         if (secno >= MAIN_SECS(sbi)) {
1906                 if (dir == ALLOC_RIGHT) {
1907                         secno = find_next_zero_bit(free_i->free_secmap,
1908                                                         MAIN_SECS(sbi), 0);
1909                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1910                 } else {
1911                         go_left = 1;
1912                         left_start = hint - 1;
1913                 }
1914         }
1915         if (go_left == 0)
1916                 goto skip_left;
1917
1918         while (test_bit(left_start, free_i->free_secmap)) {
1919                 if (left_start > 0) {
1920                         left_start--;
1921                         continue;
1922                 }
1923                 left_start = find_next_zero_bit(free_i->free_secmap,
1924                                                         MAIN_SECS(sbi), 0);
1925                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1926                 break;
1927         }
1928         secno = left_start;
1929 skip_left:
1930         hint = secno;
1931         segno = GET_SEG_FROM_SEC(sbi, secno);
1932         zoneno = GET_ZONE_FROM_SEC(sbi, secno);
1933
1934         /* give up on finding another zone */
1935         if (!init)
1936                 goto got_it;
1937         if (sbi->secs_per_zone == 1)
1938                 goto got_it;
1939         if (zoneno == old_zoneno)
1940                 goto got_it;
1941         if (dir == ALLOC_LEFT) {
1942                 if (!go_left && zoneno + 1 >= total_zones)
1943                         goto got_it;
1944                 if (go_left && zoneno == 0)
1945                         goto got_it;
1946         }
1947         for (i = 0; i < NR_CURSEG_TYPE; i++)
1948                 if (CURSEG_I(sbi, i)->zone == zoneno)
1949                         break;
1950
1951         if (i < NR_CURSEG_TYPE) {
1952                 /* zone is in user, try another */
1953                 if (go_left)
1954                         hint = zoneno * sbi->secs_per_zone - 1;
1955                 else if (zoneno + 1 >= total_zones)
1956                         hint = 0;
1957                 else
1958                         hint = (zoneno + 1) * sbi->secs_per_zone;
1959                 init = false;
1960                 goto find_other_zone;
1961         }
1962 got_it:
1963         /* set it as dirty segment in free segmap */
1964         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1965         __set_inuse(sbi, segno);
1966         *newseg = segno;
1967         spin_unlock(&free_i->segmap_lock);
1968 }
1969
1970 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1971 {
1972         struct curseg_info *curseg = CURSEG_I(sbi, type);
1973         struct summary_footer *sum_footer;
1974
1975         curseg->segno = curseg->next_segno;
1976         curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
1977         curseg->next_blkoff = 0;
1978         curseg->next_segno = NULL_SEGNO;
1979
1980         sum_footer = &(curseg->sum_blk->footer);
1981         memset(sum_footer, 0, sizeof(struct summary_footer));
1982         if (IS_DATASEG(type))
1983                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1984         if (IS_NODESEG(type))
1985                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1986         __set_sit_entry_type(sbi, type, curseg->segno, modified);
1987 }
1988
1989 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
1990 {
1991         /* if segs_per_sec is large than 1, we need to keep original policy. */
1992         if (sbi->segs_per_sec != 1)
1993                 return CURSEG_I(sbi, type)->segno;
1994
1995         if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
1996                 return 0;
1997
1998         if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
1999                 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2000         return CURSEG_I(sbi, type)->segno;
2001 }
2002
2003 /*
2004  * Allocate a current working segment.
2005  * This function always allocates a free segment in LFS manner.
2006  */
2007 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2008 {
2009         struct curseg_info *curseg = CURSEG_I(sbi, type);
2010         unsigned int segno = curseg->segno;
2011         int dir = ALLOC_LEFT;
2012
2013         write_sum_page(sbi, curseg->sum_blk,
2014                                 GET_SUM_BLOCK(sbi, segno));
2015         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2016                 dir = ALLOC_RIGHT;
2017
2018         if (test_opt(sbi, NOHEAP))
2019                 dir = ALLOC_RIGHT;
2020
2021         segno = __get_next_segno(sbi, type);
2022         get_new_segment(sbi, &segno, new_sec, dir);
2023         curseg->next_segno = segno;
2024         reset_curseg(sbi, type, 1);
2025         curseg->alloc_type = LFS;
2026 }
2027
2028 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2029                         struct curseg_info *seg, block_t start)
2030 {
2031         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2032         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2033         unsigned long *target_map = SIT_I(sbi)->tmp_map;
2034         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2035         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2036         int i, pos;
2037
2038         for (i = 0; i < entries; i++)
2039                 target_map[i] = ckpt_map[i] | cur_map[i];
2040
2041         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2042
2043         seg->next_blkoff = pos;
2044 }
2045
2046 /*
2047  * If a segment is written by LFS manner, next block offset is just obtained
2048  * by increasing the current block offset. However, if a segment is written by
2049  * SSR manner, next block offset obtained by calling __next_free_blkoff
2050  */
2051 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2052                                 struct curseg_info *seg)
2053 {
2054         if (seg->alloc_type == SSR)
2055                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2056         else
2057                 seg->next_blkoff++;
2058 }
2059
2060 /*
2061  * This function always allocates a used segment(from dirty seglist) by SSR
2062  * manner, so it should recover the existing segment information of valid blocks
2063  */
2064 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2065 {
2066         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2067         struct curseg_info *curseg = CURSEG_I(sbi, type);
2068         unsigned int new_segno = curseg->next_segno;
2069         struct f2fs_summary_block *sum_node;
2070         struct page *sum_page;
2071
2072         write_sum_page(sbi, curseg->sum_blk,
2073                                 GET_SUM_BLOCK(sbi, curseg->segno));
2074         __set_test_and_inuse(sbi, new_segno);
2075
2076         mutex_lock(&dirty_i->seglist_lock);
2077         __remove_dirty_segment(sbi, new_segno, PRE);
2078         __remove_dirty_segment(sbi, new_segno, DIRTY);
2079         mutex_unlock(&dirty_i->seglist_lock);
2080
2081         reset_curseg(sbi, type, 1);
2082         curseg->alloc_type = SSR;
2083         __next_free_blkoff(sbi, curseg, 0);
2084
2085         sum_page = get_sum_page(sbi, new_segno);
2086         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2087         memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2088         f2fs_put_page(sum_page, 1);
2089 }
2090
2091 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2092 {
2093         struct curseg_info *curseg = CURSEG_I(sbi, type);
2094         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2095         unsigned segno = NULL_SEGNO;
2096         int i, cnt;
2097         bool reversed = false;
2098
2099         /* need_SSR() already forces to do this */
2100         if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2101                 curseg->next_segno = segno;
2102                 return 1;
2103         }
2104
2105         /* For node segments, let's do SSR more intensively */
2106         if (IS_NODESEG(type)) {
2107                 if (type >= CURSEG_WARM_NODE) {
2108                         reversed = true;
2109                         i = CURSEG_COLD_NODE;
2110                 } else {
2111                         i = CURSEG_HOT_NODE;
2112                 }
2113                 cnt = NR_CURSEG_NODE_TYPE;
2114         } else {
2115                 if (type >= CURSEG_WARM_DATA) {
2116                         reversed = true;
2117                         i = CURSEG_COLD_DATA;
2118                 } else {
2119                         i = CURSEG_HOT_DATA;
2120                 }
2121                 cnt = NR_CURSEG_DATA_TYPE;
2122         }
2123
2124         for (; cnt-- > 0; reversed ? i-- : i++) {
2125                 if (i == type)
2126                         continue;
2127                 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2128                         curseg->next_segno = segno;
2129                         return 1;
2130                 }
2131         }
2132         return 0;
2133 }
2134
2135 /*
2136  * flush out current segment and replace it with new segment
2137  * This function should be returned with success, otherwise BUG
2138  */
2139 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2140                                                 int type, bool force)
2141 {
2142         struct curseg_info *curseg = CURSEG_I(sbi, type);
2143
2144         if (force)
2145                 new_curseg(sbi, type, true);
2146         else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2147                                         type == CURSEG_WARM_NODE)
2148                 new_curseg(sbi, type, false);
2149         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2150                 new_curseg(sbi, type, false);
2151         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
2152                 change_curseg(sbi, type);
2153         else
2154                 new_curseg(sbi, type, false);
2155
2156         stat_inc_seg_type(sbi, curseg);
2157 }
2158
2159 void allocate_new_segments(struct f2fs_sb_info *sbi)
2160 {
2161         struct curseg_info *curseg;
2162         unsigned int old_segno;
2163         int i;
2164
2165         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2166                 curseg = CURSEG_I(sbi, i);
2167                 old_segno = curseg->segno;
2168                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2169                 locate_dirty_segment(sbi, old_segno);
2170         }
2171 }
2172
2173 static const struct segment_allocation default_salloc_ops = {
2174         .allocate_segment = allocate_segment_by_default,
2175 };
2176
2177 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2178 {
2179         __u64 trim_start = cpc->trim_start;
2180         bool has_candidate = false;
2181
2182         mutex_lock(&SIT_I(sbi)->sentry_lock);
2183         for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2184                 if (add_discard_addrs(sbi, cpc, true)) {
2185                         has_candidate = true;
2186                         break;
2187                 }
2188         }
2189         mutex_unlock(&SIT_I(sbi)->sentry_lock);
2190
2191         cpc->trim_start = trim_start;
2192         return has_candidate;
2193 }
2194
2195 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2196 {
2197         __u64 start = F2FS_BYTES_TO_BLK(range->start);
2198         __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2199         unsigned int start_segno, end_segno;
2200         struct cp_control cpc;
2201         int err = 0;
2202
2203         if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2204                 return -EINVAL;
2205
2206         cpc.trimmed = 0;
2207         if (end <= MAIN_BLKADDR(sbi))
2208                 goto out;
2209
2210         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2211                 f2fs_msg(sbi->sb, KERN_WARNING,
2212                         "Found FS corruption, run fsck to fix.");
2213                 goto out;
2214         }
2215
2216         /* start/end segment number in main_area */
2217         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2218         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2219                                                 GET_SEGNO(sbi, end);
2220         cpc.reason = CP_DISCARD;
2221         cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2222
2223         /* do checkpoint to issue discard commands safely */
2224         for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
2225                 cpc.trim_start = start_segno;
2226
2227                 if (sbi->discard_blks == 0)
2228                         break;
2229                 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
2230                         cpc.trim_end = end_segno;
2231                 else
2232                         cpc.trim_end = min_t(unsigned int,
2233                                 rounddown(start_segno +
2234                                 BATCHED_TRIM_SEGMENTS(sbi),
2235                                 sbi->segs_per_sec) - 1, end_segno);
2236
2237                 mutex_lock(&sbi->gc_mutex);
2238                 err = write_checkpoint(sbi, &cpc);
2239                 mutex_unlock(&sbi->gc_mutex);
2240                 if (err)
2241                         break;
2242
2243                 schedule();
2244         }
2245         /* It's time to issue all the filed discards */
2246         mark_discard_range_all(sbi);
2247         f2fs_wait_discard_bios(sbi, false);
2248 out:
2249         range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
2250         return err;
2251 }
2252
2253 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2254 {
2255         struct curseg_info *curseg = CURSEG_I(sbi, type);
2256         if (curseg->next_blkoff < sbi->blocks_per_seg)
2257                 return true;
2258         return false;
2259 }
2260
2261 static int __get_segment_type_2(struct f2fs_io_info *fio)
2262 {
2263         if (fio->type == DATA)
2264                 return CURSEG_HOT_DATA;
2265         else
2266                 return CURSEG_HOT_NODE;
2267 }
2268
2269 static int __get_segment_type_4(struct f2fs_io_info *fio)
2270 {
2271         if (fio->type == DATA) {
2272                 struct inode *inode = fio->page->mapping->host;
2273
2274                 if (S_ISDIR(inode->i_mode))
2275                         return CURSEG_HOT_DATA;
2276                 else
2277                         return CURSEG_COLD_DATA;
2278         } else {
2279                 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2280                         return CURSEG_WARM_NODE;
2281                 else
2282                         return CURSEG_COLD_NODE;
2283         }
2284 }
2285
2286 static int __get_segment_type_6(struct f2fs_io_info *fio)
2287 {
2288         if (fio->type == DATA) {
2289                 struct inode *inode = fio->page->mapping->host;
2290
2291                 if (is_cold_data(fio->page) || file_is_cold(inode))
2292                         return CURSEG_COLD_DATA;
2293                 if (is_inode_flag_set(inode, FI_HOT_DATA))
2294                         return CURSEG_HOT_DATA;
2295                 return CURSEG_WARM_DATA;
2296         } else {
2297                 if (IS_DNODE(fio->page))
2298                         return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2299                                                 CURSEG_HOT_NODE;
2300                 return CURSEG_COLD_NODE;
2301         }
2302 }
2303
2304 static int __get_segment_type(struct f2fs_io_info *fio)
2305 {
2306         int type = 0;
2307
2308         switch (fio->sbi->active_logs) {
2309         case 2:
2310                 type = __get_segment_type_2(fio);
2311                 break;
2312         case 4:
2313                 type = __get_segment_type_4(fio);
2314                 break;
2315         case 6:
2316                 type = __get_segment_type_6(fio);
2317                 break;
2318         default:
2319                 f2fs_bug_on(fio->sbi, true);
2320         }
2321
2322         if (IS_HOT(type))
2323                 fio->temp = HOT;
2324         else if (IS_WARM(type))
2325                 fio->temp = WARM;
2326         else
2327                 fio->temp = COLD;
2328         return type;
2329 }
2330
2331 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2332                 block_t old_blkaddr, block_t *new_blkaddr,
2333                 struct f2fs_summary *sum, int type,
2334                 struct f2fs_io_info *fio, bool add_list)
2335 {
2336         struct sit_info *sit_i = SIT_I(sbi);
2337         struct curseg_info *curseg = CURSEG_I(sbi, type);
2338
2339         mutex_lock(&curseg->curseg_mutex);
2340         mutex_lock(&sit_i->sentry_lock);
2341
2342         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2343
2344         f2fs_wait_discard_bio(sbi, *new_blkaddr);
2345
2346         /*
2347          * __add_sum_entry should be resided under the curseg_mutex
2348          * because, this function updates a summary entry in the
2349          * current summary block.
2350          */
2351         __add_sum_entry(sbi, type, sum);
2352
2353         __refresh_next_blkoff(sbi, curseg);
2354
2355         stat_inc_block_count(sbi, curseg);
2356
2357         if (!__has_curseg_space(sbi, type))
2358                 sit_i->s_ops->allocate_segment(sbi, type, false);
2359         /*
2360          * SIT information should be updated after segment allocation,
2361          * since we need to keep dirty segments precisely under SSR.
2362          */
2363         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
2364
2365         mutex_unlock(&sit_i->sentry_lock);
2366
2367         if (page && IS_NODESEG(type)) {
2368                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2369
2370                 f2fs_inode_chksum_set(sbi, page);
2371         }
2372
2373         if (add_list) {
2374                 struct f2fs_bio_info *io;
2375
2376                 INIT_LIST_HEAD(&fio->list);
2377                 fio->in_list = true;
2378                 io = sbi->write_io[fio->type] + fio->temp;
2379                 spin_lock(&io->io_lock);
2380                 list_add_tail(&fio->list, &io->io_list);
2381                 spin_unlock(&io->io_lock);
2382         }
2383
2384         mutex_unlock(&curseg->curseg_mutex);
2385 }
2386
2387 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2388 {
2389         int type = __get_segment_type(fio);
2390         int err;
2391
2392 reallocate:
2393         allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
2394                         &fio->new_blkaddr, sum, type, fio, true);
2395
2396         /* writeout dirty page into bdev */
2397         err = f2fs_submit_page_write(fio);
2398         if (err == -EAGAIN) {
2399                 fio->old_blkaddr = fio->new_blkaddr;
2400                 goto reallocate;
2401         }
2402 }
2403
2404 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2405                                         enum iostat_type io_type)
2406 {
2407         struct f2fs_io_info fio = {
2408                 .sbi = sbi,
2409                 .type = META,
2410                 .op = REQ_OP_WRITE,
2411                 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
2412                 .old_blkaddr = page->index,
2413                 .new_blkaddr = page->index,
2414                 .page = page,
2415                 .encrypted_page = NULL,
2416                 .in_list = false,
2417         };
2418
2419         if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
2420                 fio.op_flags &= ~REQ_META;
2421
2422         set_page_writeback(page);
2423         f2fs_submit_page_write(&fio);
2424
2425         f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
2426 }
2427
2428 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
2429 {
2430         struct f2fs_summary sum;
2431
2432         set_summary(&sum, nid, 0, 0);
2433         do_write_page(&sum, fio);
2434
2435         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2436 }
2437
2438 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
2439 {
2440         struct f2fs_sb_info *sbi = fio->sbi;
2441         struct f2fs_summary sum;
2442         struct node_info ni;
2443
2444         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
2445         get_node_info(sbi, dn->nid, &ni);
2446         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
2447         do_write_page(&sum, fio);
2448         f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
2449
2450         f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
2451 }
2452
2453 int rewrite_data_page(struct f2fs_io_info *fio)
2454 {
2455         int err;
2456
2457         fio->new_blkaddr = fio->old_blkaddr;
2458         stat_inc_inplace_blocks(fio->sbi);
2459
2460         err = f2fs_submit_page_bio(fio);
2461
2462         f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2463
2464         return err;
2465 }
2466
2467 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
2468                                 block_t old_blkaddr, block_t new_blkaddr,
2469                                 bool recover_curseg, bool recover_newaddr)
2470 {
2471         struct sit_info *sit_i = SIT_I(sbi);
2472         struct curseg_info *curseg;
2473         unsigned int segno, old_cursegno;
2474         struct seg_entry *se;
2475         int type;
2476         unsigned short old_blkoff;
2477
2478         segno = GET_SEGNO(sbi, new_blkaddr);
2479         se = get_seg_entry(sbi, segno);
2480         type = se->type;
2481
2482         if (!recover_curseg) {
2483                 /* for recovery flow */
2484                 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2485                         if (old_blkaddr == NULL_ADDR)
2486                                 type = CURSEG_COLD_DATA;
2487                         else
2488                                 type = CURSEG_WARM_DATA;
2489                 }
2490         } else {
2491                 if (!IS_CURSEG(sbi, segno))
2492                         type = CURSEG_WARM_DATA;
2493         }
2494
2495         curseg = CURSEG_I(sbi, type);
2496
2497         mutex_lock(&curseg->curseg_mutex);
2498         mutex_lock(&sit_i->sentry_lock);
2499
2500         old_cursegno = curseg->segno;
2501         old_blkoff = curseg->next_blkoff;
2502
2503         /* change the current segment */
2504         if (segno != curseg->segno) {
2505                 curseg->next_segno = segno;
2506                 change_curseg(sbi, type);
2507         }
2508
2509         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
2510         __add_sum_entry(sbi, type, sum);
2511
2512         if (!recover_curseg || recover_newaddr)
2513                 update_sit_entry(sbi, new_blkaddr, 1);
2514         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2515                 update_sit_entry(sbi, old_blkaddr, -1);
2516
2517         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2518         locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2519
2520         locate_dirty_segment(sbi, old_cursegno);
2521
2522         if (recover_curseg) {
2523                 if (old_cursegno != curseg->segno) {
2524                         curseg->next_segno = old_cursegno;
2525                         change_curseg(sbi, type);
2526                 }
2527                 curseg->next_blkoff = old_blkoff;
2528         }
2529
2530         mutex_unlock(&sit_i->sentry_lock);
2531         mutex_unlock(&curseg->curseg_mutex);
2532 }
2533
2534 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2535                                 block_t old_addr, block_t new_addr,
2536                                 unsigned char version, bool recover_curseg,
2537                                 bool recover_newaddr)
2538 {
2539         struct f2fs_summary sum;
2540
2541         set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2542
2543         __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2544                                         recover_curseg, recover_newaddr);
2545
2546         f2fs_update_data_blkaddr(dn, new_addr);
2547 }
2548
2549 void f2fs_wait_on_page_writeback(struct page *page,
2550                                 enum page_type type, bool ordered)
2551 {
2552         if (PageWriteback(page)) {
2553                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2554
2555                 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2556                                                 0, page->index, type);
2557                 if (ordered)
2558                         wait_on_page_writeback(page);
2559                 else
2560                         wait_for_stable_page(page);
2561         }
2562 }
2563
2564 void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
2565 {
2566         struct page *cpage;
2567
2568         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
2569                 return;
2570
2571         cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2572         if (cpage) {
2573                 f2fs_wait_on_page_writeback(cpage, DATA, true);
2574                 f2fs_put_page(cpage, 1);
2575         }
2576 }
2577
2578 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
2579 {
2580         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2581         struct curseg_info *seg_i;
2582         unsigned char *kaddr;
2583         struct page *page;
2584         block_t start;
2585         int i, j, offset;
2586
2587         start = start_sum_block(sbi);
2588
2589         page = get_meta_page(sbi, start++);
2590         kaddr = (unsigned char *)page_address(page);
2591
2592         /* Step 1: restore nat cache */
2593         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2594         memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
2595
2596         /* Step 2: restore sit cache */
2597         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2598         memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
2599         offset = 2 * SUM_JOURNAL_SIZE;
2600
2601         /* Step 3: restore summary entries */
2602         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2603                 unsigned short blk_off;
2604                 unsigned int segno;
2605
2606                 seg_i = CURSEG_I(sbi, i);
2607                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2608                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2609                 seg_i->next_segno = segno;
2610                 reset_curseg(sbi, i, 0);
2611                 seg_i->alloc_type = ckpt->alloc_type[i];
2612                 seg_i->next_blkoff = blk_off;
2613
2614                 if (seg_i->alloc_type == SSR)
2615                         blk_off = sbi->blocks_per_seg;
2616
2617                 for (j = 0; j < blk_off; j++) {
2618                         struct f2fs_summary *s;
2619                         s = (struct f2fs_summary *)(kaddr + offset);
2620                         seg_i->sum_blk->entries[j] = *s;
2621                         offset += SUMMARY_SIZE;
2622                         if (offset + SUMMARY_SIZE <= PAGE_SIZE -
2623                                                 SUM_FOOTER_SIZE)
2624                                 continue;
2625
2626                         f2fs_put_page(page, 1);
2627                         page = NULL;
2628
2629                         page = get_meta_page(sbi, start++);
2630                         kaddr = (unsigned char *)page_address(page);
2631                         offset = 0;
2632                 }
2633         }
2634         f2fs_put_page(page, 1);
2635         return 0;
2636 }
2637
2638 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2639 {
2640         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2641         struct f2fs_summary_block *sum;
2642         struct curseg_info *curseg;
2643         struct page *new;
2644         unsigned short blk_off;
2645         unsigned int segno = 0;
2646         block_t blk_addr = 0;
2647
2648         /* get segment number and block addr */
2649         if (IS_DATASEG(type)) {
2650                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2651                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2652                                                         CURSEG_HOT_DATA]);
2653                 if (__exist_node_summaries(sbi))
2654                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2655                 else
2656                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2657         } else {
2658                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2659                                                         CURSEG_HOT_NODE]);
2660                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2661                                                         CURSEG_HOT_NODE]);
2662                 if (__exist_node_summaries(sbi))
2663                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2664                                                         type - CURSEG_HOT_NODE);
2665                 else
2666                         blk_addr = GET_SUM_BLOCK(sbi, segno);
2667         }
2668
2669         new = get_meta_page(sbi, blk_addr);
2670         sum = (struct f2fs_summary_block *)page_address(new);
2671
2672         if (IS_NODESEG(type)) {
2673                 if (__exist_node_summaries(sbi)) {
2674                         struct f2fs_summary *ns = &sum->entries[0];
2675                         int i;
2676                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2677                                 ns->version = 0;
2678                                 ns->ofs_in_node = 0;
2679                         }
2680                 } else {
2681                         int err;
2682
2683                         err = restore_node_summary(sbi, segno, sum);
2684                         if (err) {
2685                                 f2fs_put_page(new, 1);
2686                                 return err;
2687                         }
2688                 }
2689         }
2690
2691         /* set uncompleted segment to curseg */
2692         curseg = CURSEG_I(sbi, type);
2693         mutex_lock(&curseg->curseg_mutex);
2694
2695         /* update journal info */
2696         down_write(&curseg->journal_rwsem);
2697         memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2698         up_write(&curseg->journal_rwsem);
2699
2700         memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2701         memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
2702         curseg->next_segno = segno;
2703         reset_curseg(sbi, type, 0);
2704         curseg->alloc_type = ckpt->alloc_type[type];
2705         curseg->next_blkoff = blk_off;
2706         mutex_unlock(&curseg->curseg_mutex);
2707         f2fs_put_page(new, 1);
2708         return 0;
2709 }
2710
2711 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2712 {
2713         struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
2714         struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
2715         int type = CURSEG_HOT_DATA;
2716         int err;
2717
2718         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
2719                 int npages = npages_for_summary_flush(sbi, true);
2720
2721                 if (npages >= 2)
2722                         ra_meta_pages(sbi, start_sum_block(sbi), npages,
2723                                                         META_CP, true);
2724
2725                 /* restore for compacted data summary */
2726                 if (read_compacted_summaries(sbi))
2727                         return -EINVAL;
2728                 type = CURSEG_HOT_NODE;
2729         }
2730
2731         if (__exist_node_summaries(sbi))
2732                 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
2733                                         NR_CURSEG_TYPE - type, META_CP, true);
2734
2735         for (; type <= CURSEG_COLD_NODE; type++) {
2736                 err = read_normal_summaries(sbi, type);
2737                 if (err)
2738                         return err;
2739         }
2740
2741         /* sanity check for summary blocks */
2742         if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
2743                         sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
2744                 return -EINVAL;
2745
2746         return 0;
2747 }
2748
2749 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2750 {
2751         struct page *page;
2752         unsigned char *kaddr;
2753         struct f2fs_summary *summary;
2754         struct curseg_info *seg_i;
2755         int written_size = 0;
2756         int i, j;
2757
2758         page = grab_meta_page(sbi, blkaddr++);
2759         kaddr = (unsigned char *)page_address(page);
2760
2761         /* Step 1: write nat cache */
2762         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2763         memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
2764         written_size += SUM_JOURNAL_SIZE;
2765
2766         /* Step 2: write sit cache */
2767         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2768         memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
2769         written_size += SUM_JOURNAL_SIZE;
2770
2771         /* Step 3: write summary entries */
2772         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2773                 unsigned short blkoff;
2774                 seg_i = CURSEG_I(sbi, i);
2775                 if (sbi->ckpt->alloc_type[i] == SSR)
2776                         blkoff = sbi->blocks_per_seg;
2777                 else
2778                         blkoff = curseg_blkoff(sbi, i);
2779
2780                 for (j = 0; j < blkoff; j++) {
2781                         if (!page) {
2782                                 page = grab_meta_page(sbi, blkaddr++);
2783                                 kaddr = (unsigned char *)page_address(page);
2784                                 written_size = 0;
2785                         }
2786                         summary = (struct f2fs_summary *)(kaddr + written_size);
2787                         *summary = seg_i->sum_blk->entries[j];
2788                         written_size += SUMMARY_SIZE;
2789
2790                         if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
2791                                                         SUM_FOOTER_SIZE)
2792                                 continue;
2793
2794                         set_page_dirty(page);
2795                         f2fs_put_page(page, 1);
2796                         page = NULL;
2797                 }
2798         }
2799         if (page) {
2800                 set_page_dirty(page);
2801                 f2fs_put_page(page, 1);
2802         }
2803 }
2804
2805 static void write_normal_summaries(struct f2fs_sb_info *sbi,
2806                                         block_t blkaddr, int type)
2807 {
2808         int i, end;
2809         if (IS_DATASEG(type))
2810                 end = type + NR_CURSEG_DATA_TYPE;
2811         else
2812                 end = type + NR_CURSEG_NODE_TYPE;
2813
2814         for (i = type; i < end; i++)
2815                 write_current_sum_page(sbi, i, blkaddr + (i - type));
2816 }
2817
2818 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2819 {
2820         if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
2821                 write_compacted_summaries(sbi, start_blk);
2822         else
2823                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2824 }
2825
2826 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2827 {
2828         write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
2829 }
2830
2831 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
2832                                         unsigned int val, int alloc)
2833 {
2834         int i;
2835
2836         if (type == NAT_JOURNAL) {
2837                 for (i = 0; i < nats_in_cursum(journal); i++) {
2838                         if (le32_to_cpu(nid_in_journal(journal, i)) == val)
2839                                 return i;
2840                 }
2841                 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2842                         return update_nats_in_cursum(journal, 1);
2843         } else if (type == SIT_JOURNAL) {
2844                 for (i = 0; i < sits_in_cursum(journal); i++)
2845                         if (le32_to_cpu(segno_in_journal(journal, i)) == val)
2846                                 return i;
2847                 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2848                         return update_sits_in_cursum(journal, 1);
2849         }
2850         return -1;
2851 }
2852
2853 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2854                                         unsigned int segno)
2855 {
2856         return get_meta_page(sbi, current_sit_addr(sbi, segno));
2857 }
2858
2859 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2860                                         unsigned int start)
2861 {
2862         struct sit_info *sit_i = SIT_I(sbi);
2863         struct page *src_page, *dst_page;
2864         pgoff_t src_off, dst_off;
2865         void *src_addr, *dst_addr;
2866
2867         src_off = current_sit_addr(sbi, start);
2868         dst_off = next_sit_addr(sbi, src_off);
2869
2870         /* get current sit block page without lock */
2871         src_page = get_meta_page(sbi, src_off);
2872         dst_page = grab_meta_page(sbi, dst_off);
2873         f2fs_bug_on(sbi, PageDirty(src_page));
2874
2875         src_addr = page_address(src_page);
2876         dst_addr = page_address(dst_page);
2877         memcpy(dst_addr, src_addr, PAGE_SIZE);
2878
2879         set_page_dirty(dst_page);
2880         f2fs_put_page(src_page, 1);
2881
2882         set_to_next_sit(sit_i, start);
2883
2884         return dst_page;
2885 }
2886
2887 static struct sit_entry_set *grab_sit_entry_set(void)
2888 {
2889         struct sit_entry_set *ses =
2890                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
2891
2892         ses->entry_cnt = 0;
2893         INIT_LIST_HEAD(&ses->set_list);
2894         return ses;
2895 }
2896
2897 static void release_sit_entry_set(struct sit_entry_set *ses)
2898 {
2899         list_del(&ses->set_list);
2900         kmem_cache_free(sit_entry_set_slab, ses);
2901 }
2902
2903 static void adjust_sit_entry_set(struct sit_entry_set *ses,
2904                                                 struct list_head *head)
2905 {
2906         struct sit_entry_set *next = ses;
2907
2908         if (list_is_last(&ses->set_list, head))
2909                 return;
2910
2911         list_for_each_entry_continue(next, head, set_list)
2912                 if (ses->entry_cnt <= next->entry_cnt)
2913                         break;
2914
2915         list_move_tail(&ses->set_list, &next->set_list);
2916 }
2917
2918 static void add_sit_entry(unsigned int segno, struct list_head *head)
2919 {
2920         struct sit_entry_set *ses;
2921         unsigned int start_segno = START_SEGNO(segno);
2922
2923         list_for_each_entry(ses, head, set_list) {
2924                 if (ses->start_segno == start_segno) {
2925                         ses->entry_cnt++;
2926                         adjust_sit_entry_set(ses, head);
2927                         return;
2928                 }
2929         }
2930
2931         ses = grab_sit_entry_set();
2932
2933         ses->start_segno = start_segno;
2934         ses->entry_cnt++;
2935         list_add(&ses->set_list, head);
2936 }
2937
2938 static void add_sits_in_set(struct f2fs_sb_info *sbi)
2939 {
2940         struct f2fs_sm_info *sm_info = SM_I(sbi);
2941         struct list_head *set_list = &sm_info->sit_entry_set;
2942         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
2943         unsigned int segno;
2944
2945         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
2946                 add_sit_entry(segno, set_list);
2947 }
2948
2949 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
2950 {
2951         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2952         struct f2fs_journal *journal = curseg->journal;
2953         int i;
2954
2955         down_write(&curseg->journal_rwsem);
2956         for (i = 0; i < sits_in_cursum(journal); i++) {
2957                 unsigned int segno;
2958                 bool dirtied;
2959
2960                 segno = le32_to_cpu(segno_in_journal(journal, i));
2961                 dirtied = __mark_sit_entry_dirty(sbi, segno);
2962
2963                 if (!dirtied)
2964                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
2965         }
2966         update_sits_in_cursum(journal, -i);
2967         up_write(&curseg->journal_rwsem);
2968 }
2969
2970 /*
2971  * CP calls this function, which flushes SIT entries including sit_journal,
2972  * and moves prefree segs to free segs.
2973  */
2974 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2975 {
2976         struct sit_info *sit_i = SIT_I(sbi);
2977         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2978         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2979         struct f2fs_journal *journal = curseg->journal;
2980         struct sit_entry_set *ses, *tmp;
2981         struct list_head *head = &SM_I(sbi)->sit_entry_set;
2982         bool to_journal = true;
2983         struct seg_entry *se;
2984
2985         mutex_lock(&sit_i->sentry_lock);
2986
2987         if (!sit_i->dirty_sentries)
2988                 goto out;
2989
2990         /*
2991          * add and account sit entries of dirty bitmap in sit entry
2992          * set temporarily
2993          */
2994         add_sits_in_set(sbi);
2995
2996         /*
2997          * if there are no enough space in journal to store dirty sit
2998          * entries, remove all entries from journal and add and account
2999          * them in sit entry set.
3000          */
3001         if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3002                 remove_sits_in_journal(sbi);
3003
3004         /*
3005          * there are two steps to flush sit entries:
3006          * #1, flush sit entries to journal in current cold data summary block.
3007          * #2, flush sit entries to sit page.
3008          */
3009         list_for_each_entry_safe(ses, tmp, head, set_list) {
3010                 struct page *page = NULL;
3011                 struct f2fs_sit_block *raw_sit = NULL;
3012                 unsigned int start_segno = ses->start_segno;
3013                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3014                                                 (unsigned long)MAIN_SEGS(sbi));
3015                 unsigned int segno = start_segno;
3016
3017                 if (to_journal &&
3018                         !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3019                         to_journal = false;
3020
3021                 if (to_journal) {
3022                         down_write(&curseg->journal_rwsem);
3023                 } else {
3024                         page = get_next_sit_page(sbi, start_segno);
3025                         raw_sit = page_address(page);
3026                 }
3027
3028                 /* flush dirty sit entries in region of current sit set */
3029                 for_each_set_bit_from(segno, bitmap, end) {
3030                         int offset, sit_offset;
3031
3032                         se = get_seg_entry(sbi, segno);
3033
3034                         /* add discard candidates */
3035                         if (!(cpc->reason & CP_DISCARD)) {
3036                                 cpc->trim_start = segno;
3037                                 add_discard_addrs(sbi, cpc, false);
3038                         }
3039
3040                         if (to_journal) {
3041                                 offset = lookup_journal_in_cursum(journal,
3042                                                         SIT_JOURNAL, segno, 1);
3043                                 f2fs_bug_on(sbi, offset < 0);
3044                                 segno_in_journal(journal, offset) =
3045                                                         cpu_to_le32(segno);
3046                                 seg_info_to_raw_sit(se,
3047                                         &sit_in_journal(journal, offset));
3048                         } else {
3049                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3050                                 seg_info_to_raw_sit(se,
3051                                                 &raw_sit->entries[sit_offset]);
3052                         }
3053
3054                         __clear_bit(segno, bitmap);
3055                         sit_i->dirty_sentries--;
3056                         ses->entry_cnt--;
3057                 }
3058
3059                 if (to_journal)
3060                         up_write(&curseg->journal_rwsem);
3061                 else
3062                         f2fs_put_page(page, 1);
3063
3064                 f2fs_bug_on(sbi, ses->entry_cnt);
3065                 release_sit_entry_set(ses);
3066         }
3067
3068         f2fs_bug_on(sbi, !list_empty(head));
3069         f2fs_bug_on(sbi, sit_i->dirty_sentries);
3070 out:
3071         if (cpc->reason & CP_DISCARD) {
3072                 __u64 trim_start = cpc->trim_start;
3073
3074                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3075                         add_discard_addrs(sbi, cpc, false);
3076
3077                 cpc->trim_start = trim_start;
3078         }
3079         mutex_unlock(&sit_i->sentry_lock);
3080
3081         set_prefree_as_free_segments(sbi);
3082 }
3083
3084 static int build_sit_info(struct f2fs_sb_info *sbi)
3085 {
3086         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3087         struct sit_info *sit_i;
3088         unsigned int sit_segs, start;
3089         char *src_bitmap;
3090         unsigned int bitmap_size;
3091
3092         /* allocate memory for SIT information */
3093         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
3094         if (!sit_i)
3095                 return -ENOMEM;
3096
3097         SM_I(sbi)->sit_info = sit_i;
3098
3099         sit_i->sentries = kvzalloc(MAIN_SEGS(sbi) *
3100                                         sizeof(struct seg_entry), GFP_KERNEL);
3101         if (!sit_i->sentries)
3102                 return -ENOMEM;
3103
3104         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3105         sit_i->dirty_sentries_bitmap = kvzalloc(bitmap_size, GFP_KERNEL);
3106         if (!sit_i->dirty_sentries_bitmap)
3107                 return -ENOMEM;
3108
3109         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3110                 sit_i->sentries[start].cur_valid_map
3111                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3112                 sit_i->sentries[start].ckpt_valid_map
3113                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3114                 if (!sit_i->sentries[start].cur_valid_map ||
3115                                 !sit_i->sentries[start].ckpt_valid_map)
3116                         return -ENOMEM;
3117
3118 #ifdef CONFIG_F2FS_CHECK_FS
3119                 sit_i->sentries[start].cur_valid_map_mir
3120                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3121                 if (!sit_i->sentries[start].cur_valid_map_mir)
3122                         return -ENOMEM;
3123 #endif
3124
3125                 if (f2fs_discard_en(sbi)) {
3126                         sit_i->sentries[start].discard_map
3127                                 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3128                         if (!sit_i->sentries[start].discard_map)
3129                                 return -ENOMEM;
3130                 }
3131         }
3132
3133         sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3134         if (!sit_i->tmp_map)
3135                 return -ENOMEM;
3136
3137         if (sbi->segs_per_sec > 1) {
3138                 sit_i->sec_entries = kvzalloc(MAIN_SECS(sbi) *
3139                                         sizeof(struct sec_entry), GFP_KERNEL);
3140                 if (!sit_i->sec_entries)
3141                         return -ENOMEM;
3142         }
3143
3144         /* get information related with SIT */
3145         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3146
3147         /* setup SIT bitmap from ckeckpoint pack */
3148         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3149         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3150
3151         sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3152         if (!sit_i->sit_bitmap)
3153                 return -ENOMEM;
3154
3155 #ifdef CONFIG_F2FS_CHECK_FS
3156         sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3157         if (!sit_i->sit_bitmap_mir)
3158                 return -ENOMEM;
3159 #endif
3160
3161         /* init SIT information */
3162         sit_i->s_ops = &default_salloc_ops;
3163
3164         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3165         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
3166         sit_i->written_valid_blocks = 0;
3167         sit_i->bitmap_size = bitmap_size;
3168         sit_i->dirty_sentries = 0;
3169         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3170         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3171         sit_i->mounted_time = ktime_get_real_seconds();
3172         mutex_init(&sit_i->sentry_lock);
3173         return 0;
3174 }
3175
3176 static int build_free_segmap(struct f2fs_sb_info *sbi)
3177 {
3178         struct free_segmap_info *free_i;
3179         unsigned int bitmap_size, sec_bitmap_size;
3180
3181         /* allocate memory for free segmap information */
3182         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
3183         if (!free_i)
3184                 return -ENOMEM;
3185
3186         SM_I(sbi)->free_info = free_i;
3187
3188         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3189         free_i->free_segmap = kvmalloc(bitmap_size, GFP_KERNEL);
3190         if (!free_i->free_segmap)
3191                 return -ENOMEM;
3192
3193         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3194         free_i->free_secmap = kvmalloc(sec_bitmap_size, GFP_KERNEL);
3195         if (!free_i->free_secmap)
3196                 return -ENOMEM;
3197
3198         /* set all segments as dirty temporarily */
3199         memset(free_i->free_segmap, 0xff, bitmap_size);
3200         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3201
3202         /* init free segmap information */
3203         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
3204         free_i->free_segments = 0;
3205         free_i->free_sections = 0;
3206         spin_lock_init(&free_i->segmap_lock);
3207         return 0;
3208 }
3209
3210 static int build_curseg(struct f2fs_sb_info *sbi)
3211 {
3212         struct curseg_info *array;
3213         int i;
3214
3215         array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
3216         if (!array)
3217                 return -ENOMEM;
3218
3219         SM_I(sbi)->curseg_array = array;
3220
3221         for (i = 0; i < NR_CURSEG_TYPE; i++) {
3222                 mutex_init(&array[i].curseg_mutex);
3223                 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
3224                 if (!array[i].sum_blk)
3225                         return -ENOMEM;
3226                 init_rwsem(&array[i].journal_rwsem);
3227                 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
3228                                                         GFP_KERNEL);
3229                 if (!array[i].journal)
3230                         return -ENOMEM;
3231                 array[i].segno = NULL_SEGNO;
3232                 array[i].next_blkoff = 0;
3233         }
3234         return restore_curseg_summaries(sbi);
3235 }
3236
3237 static void build_sit_entries(struct f2fs_sb_info *sbi)
3238 {
3239         struct sit_info *sit_i = SIT_I(sbi);
3240         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3241         struct f2fs_journal *journal = curseg->journal;
3242         struct seg_entry *se;
3243         struct f2fs_sit_entry sit;
3244         int sit_blk_cnt = SIT_BLK_CNT(sbi);
3245         unsigned int i, start, end;
3246         unsigned int readed, start_blk = 0;
3247
3248         do {
3249                 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3250                                                         META_SIT, true);
3251
3252                 start = start_blk * sit_i->sents_per_block;
3253                 end = (start_blk + readed) * sit_i->sents_per_block;
3254
3255                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
3256                         struct f2fs_sit_block *sit_blk;
3257                         struct page *page;
3258
3259                         se = &sit_i->sentries[start];
3260                         page = get_current_sit_page(sbi, start);
3261                         sit_blk = (struct f2fs_sit_block *)page_address(page);
3262                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3263                         f2fs_put_page(page, 1);
3264
3265                         check_block_count(sbi, start, &sit);
3266                         seg_info_from_raw_sit(se, &sit);
3267
3268                         /* build discard map only one time */
3269                         if (f2fs_discard_en(sbi)) {
3270                                 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3271                                         memset(se->discard_map, 0xff,
3272                                                 SIT_VBLOCK_MAP_SIZE);
3273                                 } else {
3274                                         memcpy(se->discard_map,
3275                                                 se->cur_valid_map,
3276                                                 SIT_VBLOCK_MAP_SIZE);
3277                                         sbi->discard_blks +=
3278                                                 sbi->blocks_per_seg -
3279                                                 se->valid_blocks;
3280                                 }
3281                         }
3282
3283                         if (sbi->segs_per_sec > 1)
3284                                 get_sec_entry(sbi, start)->valid_blocks +=
3285                                                         se->valid_blocks;
3286                 }
3287                 start_blk += readed;
3288         } while (start_blk < sit_blk_cnt);
3289
3290         down_read(&curseg->journal_rwsem);
3291         for (i = 0; i < sits_in_cursum(journal); i++) {
3292                 unsigned int old_valid_blocks;
3293
3294                 start = le32_to_cpu(segno_in_journal(journal, i));
3295                 se = &sit_i->sentries[start];
3296                 sit = sit_in_journal(journal, i);
3297
3298                 old_valid_blocks = se->valid_blocks;
3299
3300                 check_block_count(sbi, start, &sit);
3301                 seg_info_from_raw_sit(se, &sit);
3302
3303                 if (f2fs_discard_en(sbi)) {
3304                         if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3305                                 memset(se->discard_map, 0xff,
3306                                                         SIT_VBLOCK_MAP_SIZE);
3307                         } else {
3308                                 memcpy(se->discard_map, se->cur_valid_map,
3309                                                         SIT_VBLOCK_MAP_SIZE);
3310                                 sbi->discard_blks += old_valid_blocks -
3311                                                         se->valid_blocks;
3312                         }
3313                 }
3314
3315                 if (sbi->segs_per_sec > 1)
3316                         get_sec_entry(sbi, start)->valid_blocks +=
3317                                 se->valid_blocks - old_valid_blocks;
3318         }
3319         up_read(&curseg->journal_rwsem);
3320 }
3321
3322 static void init_free_segmap(struct f2fs_sb_info *sbi)
3323 {
3324         unsigned int start;
3325         int type;
3326
3327         for (start = 0; start < MAIN_SEGS(sbi); start++) {
3328                 struct seg_entry *sentry = get_seg_entry(sbi, start);
3329                 if (!sentry->valid_blocks)
3330                         __set_free(sbi, start);
3331                 else
3332                         SIT_I(sbi)->written_valid_blocks +=
3333                                                 sentry->valid_blocks;
3334         }
3335
3336         /* set use the current segments */
3337         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3338                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3339                 __set_test_and_inuse(sbi, curseg_t->segno);
3340         }
3341 }
3342
3343 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3344 {
3345         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3346         struct free_segmap_info *free_i = FREE_I(sbi);
3347         unsigned int segno = 0, offset = 0;
3348         unsigned short valid_blocks;
3349
3350         while (1) {
3351                 /* find dirty segment based on free segmap */
3352                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3353                 if (segno >= MAIN_SEGS(sbi))
3354                         break;
3355                 offset = segno + 1;
3356                 valid_blocks = get_valid_blocks(sbi, segno, false);
3357                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
3358                         continue;
3359                 if (valid_blocks > sbi->blocks_per_seg) {
3360                         f2fs_bug_on(sbi, 1);
3361                         continue;
3362                 }
3363                 mutex_lock(&dirty_i->seglist_lock);
3364                 __locate_dirty_segment(sbi, segno, DIRTY);
3365                 mutex_unlock(&dirty_i->seglist_lock);
3366         }
3367 }
3368
3369 static int init_victim_secmap(struct f2fs_sb_info *sbi)
3370 {
3371         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3372         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3373
3374         dirty_i->victim_secmap = kvzalloc(bitmap_size, GFP_KERNEL);
3375         if (!dirty_i->victim_secmap)
3376                 return -ENOMEM;
3377         return 0;
3378 }
3379
3380 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3381 {
3382         struct dirty_seglist_info *dirty_i;
3383         unsigned int bitmap_size, i;
3384
3385         /* allocate memory for dirty segments list information */
3386         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
3387         if (!dirty_i)
3388                 return -ENOMEM;
3389
3390         SM_I(sbi)->dirty_info = dirty_i;
3391         mutex_init(&dirty_i->seglist_lock);
3392
3393         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3394
3395         for (i = 0; i < NR_DIRTY_TYPE; i++) {
3396                 dirty_i->dirty_segmap[i] = kvzalloc(bitmap_size, GFP_KERNEL);
3397                 if (!dirty_i->dirty_segmap[i])
3398                         return -ENOMEM;
3399         }
3400
3401         init_dirty_segmap(sbi);
3402         return init_victim_secmap(sbi);
3403 }
3404
3405 /*
3406  * Update min, max modified time for cost-benefit GC algorithm
3407  */
3408 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3409 {
3410         struct sit_info *sit_i = SIT_I(sbi);
3411         unsigned int segno;
3412
3413         mutex_lock(&sit_i->sentry_lock);
3414
3415         sit_i->min_mtime = LLONG_MAX;
3416
3417         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
3418                 unsigned int i;
3419                 unsigned long long mtime = 0;
3420
3421                 for (i = 0; i < sbi->segs_per_sec; i++)
3422                         mtime += get_seg_entry(sbi, segno + i)->mtime;
3423
3424                 mtime = div_u64(mtime, sbi->segs_per_sec);
3425
3426                 if (sit_i->min_mtime > mtime)
3427                         sit_i->min_mtime = mtime;
3428         }
3429         sit_i->max_mtime = get_mtime(sbi);
3430         mutex_unlock(&sit_i->sentry_lock);
3431 }
3432
3433 int build_segment_manager(struct f2fs_sb_info *sbi)
3434 {
3435         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3436         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3437         struct f2fs_sm_info *sm_info;
3438         int err;
3439
3440         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
3441         if (!sm_info)
3442                 return -ENOMEM;
3443
3444         /* init sm info */
3445         sbi->sm_info = sm_info;
3446         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3447         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3448         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3449         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3450         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3451         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3452         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
3453         sm_info->rec_prefree_segments = sm_info->main_segments *
3454                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
3455         if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3456                 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3457
3458         if (!test_opt(sbi, LFS))
3459                 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
3460         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
3461         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
3462         sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
3463
3464         sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
3465
3466         INIT_LIST_HEAD(&sm_info->sit_entry_set);
3467
3468         if (!f2fs_readonly(sbi->sb)) {
3469                 err = create_flush_cmd_control(sbi);
3470                 if (err)
3471                         return err;
3472         }
3473
3474         err = create_discard_cmd_control(sbi);
3475         if (err)
3476                 return err;
3477
3478         err = build_sit_info(sbi);
3479         if (err)
3480                 return err;
3481         err = build_free_segmap(sbi);
3482         if (err)
3483                 return err;
3484         err = build_curseg(sbi);
3485         if (err)
3486                 return err;
3487
3488         /* reinit free segmap based on SIT */
3489         build_sit_entries(sbi);
3490
3491         init_free_segmap(sbi);
3492         err = build_dirty_segmap(sbi);
3493         if (err)
3494                 return err;
3495
3496         init_min_max_mtime(sbi);
3497         return 0;
3498 }
3499
3500 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3501                 enum dirty_type dirty_type)
3502 {
3503         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3504
3505         mutex_lock(&dirty_i->seglist_lock);
3506         kvfree(dirty_i->dirty_segmap[dirty_type]);
3507         dirty_i->nr_dirty[dirty_type] = 0;
3508         mutex_unlock(&dirty_i->seglist_lock);
3509 }
3510
3511 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
3512 {
3513         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3514         kvfree(dirty_i->victim_secmap);
3515 }
3516
3517 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3518 {
3519         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3520         int i;
3521
3522         if (!dirty_i)
3523                 return;
3524
3525         /* discard pre-free/dirty segments list */
3526         for (i = 0; i < NR_DIRTY_TYPE; i++)
3527                 discard_dirty_segmap(sbi, i);
3528
3529         destroy_victim_secmap(sbi);
3530         SM_I(sbi)->dirty_info = NULL;
3531         kfree(dirty_i);
3532 }
3533
3534 static void destroy_curseg(struct f2fs_sb_info *sbi)
3535 {
3536         struct curseg_info *array = SM_I(sbi)->curseg_array;
3537         int i;
3538
3539         if (!array)
3540                 return;
3541         SM_I(sbi)->curseg_array = NULL;
3542         for (i = 0; i < NR_CURSEG_TYPE; i++) {
3543                 kfree(array[i].sum_blk);
3544                 kfree(array[i].journal);
3545         }
3546         kfree(array);
3547 }
3548
3549 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3550 {
3551         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3552         if (!free_i)
3553                 return;
3554         SM_I(sbi)->free_info = NULL;
3555         kvfree(free_i->free_segmap);
3556         kvfree(free_i->free_secmap);
3557         kfree(free_i);
3558 }
3559
3560 static void destroy_sit_info(struct f2fs_sb_info *sbi)
3561 {
3562         struct sit_info *sit_i = SIT_I(sbi);
3563         unsigned int start;
3564
3565         if (!sit_i)
3566                 return;
3567
3568         if (sit_i->sentries) {
3569                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3570                         kfree(sit_i->sentries[start].cur_valid_map);
3571 #ifdef CONFIG_F2FS_CHECK_FS
3572                         kfree(sit_i->sentries[start].cur_valid_map_mir);
3573 #endif
3574                         kfree(sit_i->sentries[start].ckpt_valid_map);
3575                         kfree(sit_i->sentries[start].discard_map);
3576                 }
3577         }
3578         kfree(sit_i->tmp_map);
3579
3580         kvfree(sit_i->sentries);
3581         kvfree(sit_i->sec_entries);
3582         kvfree(sit_i->dirty_sentries_bitmap);
3583
3584         SM_I(sbi)->sit_info = NULL;
3585         kfree(sit_i->sit_bitmap);
3586 #ifdef CONFIG_F2FS_CHECK_FS
3587         kfree(sit_i->sit_bitmap_mir);
3588 #endif
3589         kfree(sit_i);
3590 }
3591
3592 void destroy_segment_manager(struct f2fs_sb_info *sbi)
3593 {
3594         struct f2fs_sm_info *sm_info = SM_I(sbi);
3595
3596         if (!sm_info)
3597                 return;
3598         destroy_flush_cmd_control(sbi, true);
3599         destroy_discard_cmd_control(sbi);
3600         destroy_dirty_segmap(sbi);
3601         destroy_curseg(sbi);
3602         destroy_free_segmap(sbi);
3603         destroy_sit_info(sbi);
3604         sbi->sm_info = NULL;
3605         kfree(sm_info);
3606 }
3607
3608 int __init create_segment_manager_caches(void)
3609 {
3610         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
3611                         sizeof(struct discard_entry));
3612         if (!discard_entry_slab)
3613                 goto fail;
3614
3615         discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3616                         sizeof(struct discard_cmd));
3617         if (!discard_cmd_slab)
3618                 goto destroy_discard_entry;
3619
3620         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
3621                         sizeof(struct sit_entry_set));
3622         if (!sit_entry_set_slab)
3623                 goto destroy_discard_cmd;
3624
3625         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3626                         sizeof(struct inmem_pages));
3627         if (!inmem_entry_slab)
3628                 goto destroy_sit_entry_set;
3629         return 0;
3630
3631 destroy_sit_entry_set:
3632         kmem_cache_destroy(sit_entry_set_slab);
3633 destroy_discard_cmd:
3634         kmem_cache_destroy(discard_cmd_slab);
3635 destroy_discard_entry:
3636         kmem_cache_destroy(discard_entry_slab);
3637 fail:
3638         return -ENOMEM;
3639 }
3640
3641 void destroy_segment_manager_caches(void)
3642 {
3643         kmem_cache_destroy(sit_entry_set_slab);
3644         kmem_cache_destroy(discard_cmd_slab);
3645         kmem_cache_destroy(discard_entry_slab);
3646         kmem_cache_destroy(inmem_entry_slab);
3647 }