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
  31static struct kmem_cache *discard_entry_slab;
  32static struct kmem_cache *discard_cmd_slab;
  33static struct kmem_cache *sit_entry_set_slab;
  34static struct kmem_cache *inmem_entry_slab;
  35
  36static 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 */
  55static 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 */
  99static 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;
 123pass:
 124		if (size <= BITS_PER_LONG)
 125			break;
 126		size -= BITS_PER_LONG;
 127		offset = 0;
 128		p++;
 129	}
 130	return result;
 131found:
 132	return result - size + __reverse_ffs(tmp);
 133}
 134
 135static 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;
 160pass:
 161		if (size <= BITS_PER_LONG)
 162			break;
 163		size -= BITS_PER_LONG;
 164		offset = 0;
 165		p++;
 166	}
 167	return result;
 168found:
 169	return result - size + __reverse_ffz(tmp);
 170}
 171
 172bool 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			SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
 185}
 186
 187void register_inmem_page(struct inode *inode, struct page *page)
 188{
 189	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 190	struct f2fs_inode_info *fi = F2FS_I(inode);
 191	struct inmem_pages *new;
 192
 193	f2fs_trace_pid(page);
 194
 195	set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
 196	SetPagePrivate(page);
 197
 198	new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 199
 200	/* add atomic page indices to the list */
 201	new->page = page;
 202	INIT_LIST_HEAD(&new->list);
 203
 204	/* increase reference count with clean state */
 205	mutex_lock(&fi->inmem_lock);
 206	get_page(page);
 207	list_add_tail(&new->list, &fi->inmem_pages);
 208	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 209	if (list_empty(&fi->inmem_ilist))
 210		list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
 211	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 212	inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 213	mutex_unlock(&fi->inmem_lock);
 214
 215	trace_f2fs_register_inmem_page(page, INMEM);
 216}
 217
 218static int __revoke_inmem_pages(struct inode *inode,
 219				struct list_head *head, bool drop, bool recover)
 220{
 221	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 222	struct inmem_pages *cur, *tmp;
 223	int err = 0;
 224
 225	list_for_each_entry_safe(cur, tmp, head, list) {
 226		struct page *page = cur->page;
 227
 228		if (drop)
 229			trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 230
 231		lock_page(page);
 232
 233		if (recover) {
 234			struct dnode_of_data dn;
 235			struct node_info ni;
 236
 237			trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 238retry:
 239			set_new_dnode(&dn, inode, NULL, NULL, 0);
 240			err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
 241			if (err) {
 242				if (err == -ENOMEM) {
 243					congestion_wait(BLK_RW_ASYNC, HZ/50);
 244					cond_resched();
 245					goto retry;
 246				}
 247				err = -EAGAIN;
 248				goto next;
 249			}
 250			get_node_info(sbi, dn.nid, &ni);
 251			if (cur->old_addr == NEW_ADDR) {
 252				invalidate_blocks(sbi, dn.data_blkaddr);
 253				f2fs_update_data_blkaddr(&dn, NEW_ADDR);
 254			} else
 255				f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 256					cur->old_addr, ni.version, true, true);
 257			f2fs_put_dnode(&dn);
 258		}
 259next:
 260		/* we don't need to invalidate this in the sccessful status */
 261		if (drop || recover)
 262			ClearPageUptodate(page);
 263		set_page_private(page, 0);
 264		ClearPagePrivate(page);
 265		f2fs_put_page(page, 1);
 266
 267		list_del(&cur->list);
 268		kmem_cache_free(inmem_entry_slab, cur);
 269		dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 270	}
 271	return err;
 272}
 273
 274void drop_inmem_pages_all(struct f2fs_sb_info *sbi)
 275{
 276	struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
 277	struct inode *inode;
 278	struct f2fs_inode_info *fi;
 279next:
 280	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 281	if (list_empty(head)) {
 282		spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 283		return;
 284	}
 285	fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
 286	inode = igrab(&fi->vfs_inode);
 287	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 288
 289	if (inode) {
 290		drop_inmem_pages(inode);
 291		iput(inode);
 292	}
 293	congestion_wait(BLK_RW_ASYNC, HZ/50);
 294	cond_resched();
 295	goto next;
 296}
 297
 298void drop_inmem_pages(struct inode *inode)
 299{
 300	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 301	struct f2fs_inode_info *fi = F2FS_I(inode);
 302
 303	mutex_lock(&fi->inmem_lock);
 304	__revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 305	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 306	if (!list_empty(&fi->inmem_ilist))
 307		list_del_init(&fi->inmem_ilist);
 308	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 309	mutex_unlock(&fi->inmem_lock);
 310
 311	clear_inode_flag(inode, FI_ATOMIC_FILE);
 312	clear_inode_flag(inode, FI_HOT_DATA);
 313	stat_dec_atomic_write(inode);
 314}
 315
 316void drop_inmem_page(struct inode *inode, struct page *page)
 317{
 318	struct f2fs_inode_info *fi = F2FS_I(inode);
 319	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 320	struct list_head *head = &fi->inmem_pages;
 321	struct inmem_pages *cur = NULL;
 322
 323	f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
 324
 325	mutex_lock(&fi->inmem_lock);
 326	list_for_each_entry(cur, head, list) {
 327		if (cur->page == page)
 328			break;
 329	}
 330
 331	f2fs_bug_on(sbi, !cur || cur->page != page);
 332	list_del(&cur->list);
 333	mutex_unlock(&fi->inmem_lock);
 334
 335	dec_page_count(sbi, F2FS_INMEM_PAGES);
 336	kmem_cache_free(inmem_entry_slab, cur);
 337
 338	ClearPageUptodate(page);
 339	set_page_private(page, 0);
 340	ClearPagePrivate(page);
 341	f2fs_put_page(page, 0);
 342
 343	trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
 344}
 345
 346static int __commit_inmem_pages(struct inode *inode,
 347					struct list_head *revoke_list)
 348{
 349	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 350	struct f2fs_inode_info *fi = F2FS_I(inode);
 351	struct inmem_pages *cur, *tmp;
 352	struct f2fs_io_info fio = {
 353		.sbi = sbi,
 354		.ino = inode->i_ino,
 355		.type = DATA,
 356		.op = REQ_OP_WRITE,
 357		.op_flags = REQ_SYNC | REQ_PRIO,
 358		.io_type = FS_DATA_IO,
 359	};
 360	pgoff_t last_idx = ULONG_MAX;
 361	int err = 0;
 362
 363	list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 364		struct page *page = cur->page;
 365
 366		lock_page(page);
 367		if (page->mapping == inode->i_mapping) {
 368			trace_f2fs_commit_inmem_page(page, INMEM);
 369
 370			set_page_dirty(page);
 371			f2fs_wait_on_page_writeback(page, DATA, true);
 372			if (clear_page_dirty_for_io(page)) {
 373				inode_dec_dirty_pages(inode);
 374				remove_dirty_inode(inode);
 375			}
 376retry:
 377			fio.page = page;
 378			fio.old_blkaddr = NULL_ADDR;
 379			fio.encrypted_page = NULL;
 380			fio.need_lock = LOCK_DONE;
 381			err = do_write_data_page(&fio);
 382			if (err) {
 383				if (err == -ENOMEM) {
 384					congestion_wait(BLK_RW_ASYNC, HZ/50);
 385					cond_resched();
 386					goto retry;
 387				}
 388				unlock_page(page);
 389				break;
 390			}
 391			/* record old blkaddr for revoking */
 392			cur->old_addr = fio.old_blkaddr;
 393			last_idx = page->index;
 394		}
 395		unlock_page(page);
 396		list_move_tail(&cur->list, revoke_list);
 397	}
 398
 399	if (last_idx != ULONG_MAX)
 400		f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
 401
 402	if (!err)
 403		__revoke_inmem_pages(inode, revoke_list, false, false);
 404
 405	return err;
 406}
 407
 408int commit_inmem_pages(struct inode *inode)
 409{
 410	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 411	struct f2fs_inode_info *fi = F2FS_I(inode);
 412	struct list_head revoke_list;
 413	int err;
 414
 415	INIT_LIST_HEAD(&revoke_list);
 416	f2fs_balance_fs(sbi, true);
 417	f2fs_lock_op(sbi);
 418
 419	set_inode_flag(inode, FI_ATOMIC_COMMIT);
 420
 421	mutex_lock(&fi->inmem_lock);
 422	err = __commit_inmem_pages(inode, &revoke_list);
 423	if (err) {
 424		int ret;
 425		/*
 426		 * try to revoke all committed pages, but still we could fail
 427		 * due to no memory or other reason, if that happened, EAGAIN
 428		 * will be returned, which means in such case, transaction is
 429		 * already not integrity, caller should use journal to do the
 430		 * recovery or rewrite & commit last transaction. For other
 431		 * error number, revoking was done by filesystem itself.
 432		 */
 433		ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
 434		if (ret)
 435			err = ret;
 436
 437		/* drop all uncommitted pages */
 438		__revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 439	}
 440	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 441	if (!list_empty(&fi->inmem_ilist))
 442		list_del_init(&fi->inmem_ilist);
 443	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
 444	mutex_unlock(&fi->inmem_lock);
 445
 446	clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 447
 448	f2fs_unlock_op(sbi);
 449	return err;
 450}
 451
 452/*
 453 * This function balances dirty node and dentry pages.
 454 * In addition, it controls garbage collection.
 455 */
 456void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 457{
 458#ifdef CONFIG_F2FS_FAULT_INJECTION
 459	if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
 460		f2fs_show_injection_info(FAULT_CHECKPOINT);
 461		f2fs_stop_checkpoint(sbi, false);
 462	}
 463#endif
 464
 465	/* balance_fs_bg is able to be pending */
 466	if (need && excess_cached_nats(sbi))
 467		f2fs_balance_fs_bg(sbi);
 468
 469	/*
 470	 * We should do GC or end up with checkpoint, if there are so many dirty
 471	 * dir/node pages without enough free segments.
 472	 */
 473	if (has_not_enough_free_secs(sbi, 0, 0)) {
 474		mutex_lock(&sbi->gc_mutex);
 475		f2fs_gc(sbi, false, false, NULL_SEGNO);
 476	}
 477}
 478
 479void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 480{
 481	/* try to shrink extent cache when there is no enough memory */
 482	if (!available_free_memory(sbi, EXTENT_CACHE))
 483		f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 484
 485	/* check the # of cached NAT entries */
 486	if (!available_free_memory(sbi, NAT_ENTRIES))
 487		try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 488
 489	if (!available_free_memory(sbi, FREE_NIDS))
 490		try_to_free_nids(sbi, MAX_FREE_NIDS);
 491	else
 492		build_free_nids(sbi, false, false);
 493
 494	if (!is_idle(sbi) && !excess_dirty_nats(sbi))
 495		return;
 496
 497	/* checkpoint is the only way to shrink partial cached entries */
 498	if (!available_free_memory(sbi, NAT_ENTRIES) ||
 499			!available_free_memory(sbi, INO_ENTRIES) ||
 500			excess_prefree_segs(sbi) ||
 501			excess_dirty_nats(sbi) ||
 502			f2fs_time_over(sbi, CP_TIME)) {
 503		if (test_opt(sbi, DATA_FLUSH)) {
 504			struct blk_plug plug;
 505
 506			blk_start_plug(&plug);
 507			sync_dirty_inodes(sbi, FILE_INODE);
 508			blk_finish_plug(&plug);
 509		}
 510		f2fs_sync_fs(sbi->sb, true);
 511		stat_inc_bg_cp_count(sbi->stat_info);
 512	}
 513}
 514
 515static int __submit_flush_wait(struct f2fs_sb_info *sbi,
 516				struct block_device *bdev)
 517{
 518	struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
 519	int ret;
 520
 521	bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
 522	bio_set_dev(bio, bdev);
 523	ret = submit_bio_wait(bio);
 524	bio_put(bio);
 525
 526	trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
 527				test_opt(sbi, FLUSH_MERGE), ret);
 528	return ret;
 529}
 530
 531static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
 532{
 533	int ret = 0;
 534	int i;
 535
 536	if (!sbi->s_ndevs)
 537		return __submit_flush_wait(sbi, sbi->sb->s_bdev);
 538
 539	for (i = 0; i < sbi->s_ndevs; i++) {
 540		if (!is_dirty_device(sbi, ino, i, FLUSH_INO))
 541			continue;
 542		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 543		if (ret)
 544			break;
 545	}
 546	return ret;
 547}
 548
 549static int issue_flush_thread(void *data)
 550{
 551	struct f2fs_sb_info *sbi = data;
 552	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 553	wait_queue_head_t *q = &fcc->flush_wait_queue;
 554repeat:
 555	if (kthread_should_stop())
 556		return 0;
 557
 558	sb_start_intwrite(sbi->sb);
 559
 560	if (!llist_empty(&fcc->issue_list)) {
 561		struct flush_cmd *cmd, *next;
 562		int ret;
 563
 564		fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 565		fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 566
 567		cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
 568
 569		ret = submit_flush_wait(sbi, cmd->ino);
 570		atomic_inc(&fcc->issued_flush);
 571
 572		llist_for_each_entry_safe(cmd, next,
 573					  fcc->dispatch_list, llnode) {
 574			cmd->ret = ret;
 575			complete(&cmd->wait);
 576		}
 577		fcc->dispatch_list = NULL;
 578	}
 579
 580	sb_end_intwrite(sbi->sb);
 581
 582	wait_event_interruptible(*q,
 583		kthread_should_stop() || !llist_empty(&fcc->issue_list));
 584	goto repeat;
 585}
 586
 587int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
 588{
 589	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 590	struct flush_cmd cmd;
 591	int ret;
 592
 593	if (test_opt(sbi, NOBARRIER))
 594		return 0;
 595
 596	if (!test_opt(sbi, FLUSH_MERGE)) {
 597		ret = submit_flush_wait(sbi, ino);
 598		atomic_inc(&fcc->issued_flush);
 599		return ret;
 600	}
 601
 602	if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
 603		ret = submit_flush_wait(sbi, ino);
 604		atomic_dec(&fcc->issing_flush);
 605
 606		atomic_inc(&fcc->issued_flush);
 607		return ret;
 608	}
 609
 610	cmd.ino = ino;
 611	init_completion(&cmd.wait);
 612
 613	llist_add(&cmd.llnode, &fcc->issue_list);
 614
 615	/* update issue_list before we wake up issue_flush thread */
 616	smp_mb();
 617
 618	if (waitqueue_active(&fcc->flush_wait_queue))
 619		wake_up(&fcc->flush_wait_queue);
 620
 621	if (fcc->f2fs_issue_flush) {
 622		wait_for_completion(&cmd.wait);
 623		atomic_dec(&fcc->issing_flush);
 624	} else {
 625		struct llist_node *list;
 626
 627		list = llist_del_all(&fcc->issue_list);
 628		if (!list) {
 629			wait_for_completion(&cmd.wait);
 630			atomic_dec(&fcc->issing_flush);
 631		} else {
 632			struct flush_cmd *tmp, *next;
 633
 634			ret = submit_flush_wait(sbi, ino);
 635
 636			llist_for_each_entry_safe(tmp, next, list, llnode) {
 637				if (tmp == &cmd) {
 638					cmd.ret = ret;
 639					atomic_dec(&fcc->issing_flush);
 640					continue;
 641				}
 642				tmp->ret = ret;
 643				complete(&tmp->wait);
 644			}
 645		}
 646	}
 647
 648	return cmd.ret;
 649}
 650
 651int create_flush_cmd_control(struct f2fs_sb_info *sbi)
 652{
 653	dev_t dev = sbi->sb->s_bdev->bd_dev;
 654	struct flush_cmd_control *fcc;
 655	int err = 0;
 656
 657	if (SM_I(sbi)->fcc_info) {
 658		fcc = SM_I(sbi)->fcc_info;
 659		if (fcc->f2fs_issue_flush)
 660			return err;
 661		goto init_thread;
 662	}
 663
 664	fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
 665	if (!fcc)
 666		return -ENOMEM;
 667	atomic_set(&fcc->issued_flush, 0);
 668	atomic_set(&fcc->issing_flush, 0);
 669	init_waitqueue_head(&fcc->flush_wait_queue);
 670	init_llist_head(&fcc->issue_list);
 671	SM_I(sbi)->fcc_info = fcc;
 672	if (!test_opt(sbi, FLUSH_MERGE))
 673		return err;
 674
 675init_thread:
 676	fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 677				"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 678	if (IS_ERR(fcc->f2fs_issue_flush)) {
 679		err = PTR_ERR(fcc->f2fs_issue_flush);
 680		kfree(fcc);
 681		SM_I(sbi)->fcc_info = NULL;
 682		return err;
 683	}
 684
 685	return err;
 686}
 687
 688void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 689{
 690	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
 691
 692	if (fcc && fcc->f2fs_issue_flush) {
 693		struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 694
 695		fcc->f2fs_issue_flush = NULL;
 696		kthread_stop(flush_thread);
 697	}
 698	if (free) {
 699		kfree(fcc);
 700		SM_I(sbi)->fcc_info = NULL;
 701	}
 702}
 703
 704int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
 705{
 706	int ret = 0, i;
 707
 708	if (!sbi->s_ndevs)
 709		return 0;
 710
 711	for (i = 1; i < sbi->s_ndevs; i++) {
 712		if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
 713			continue;
 714		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
 715		if (ret)
 716			break;
 717
 718		spin_lock(&sbi->dev_lock);
 719		f2fs_clear_bit(i, (char *)&sbi->dirty_device);
 720		spin_unlock(&sbi->dev_lock);
 721	}
 722
 723	return ret;
 724}
 725
 726static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 727		enum dirty_type dirty_type)
 728{
 729	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 730
 731	/* need not be added */
 732	if (IS_CURSEG(sbi, segno))
 733		return;
 734
 735	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 736		dirty_i->nr_dirty[dirty_type]++;
 737
 738	if (dirty_type == DIRTY) {
 739		struct seg_entry *sentry = get_seg_entry(sbi, segno);
 740		enum dirty_type t = sentry->type;
 741
 742		if (unlikely(t >= DIRTY)) {
 743			f2fs_bug_on(sbi, 1);
 744			return;
 745		}
 746		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 747			dirty_i->nr_dirty[t]++;
 748	}
 749}
 750
 751static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 752		enum dirty_type dirty_type)
 753{
 754	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 755
 756	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 757		dirty_i->nr_dirty[dirty_type]--;
 758
 759	if (dirty_type == DIRTY) {
 760		struct seg_entry *sentry = get_seg_entry(sbi, segno);
 761		enum dirty_type t = sentry->type;
 762
 763		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 764			dirty_i->nr_dirty[t]--;
 765
 766		if (get_valid_blocks(sbi, segno, true) == 0)
 767			clear_bit(GET_SEC_FROM_SEG(sbi, segno),
 768						dirty_i->victim_secmap);
 769	}
 770}
 771
 772/*
 773 * Should not occur error such as -ENOMEM.
 774 * Adding dirty entry into seglist is not critical operation.
 775 * If a given segment is one of current working segments, it won't be added.
 776 */
 777static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 778{
 779	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 780	unsigned short valid_blocks;
 781
 782	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 783		return;
 784
 785	mutex_lock(&dirty_i->seglist_lock);
 786
 787	valid_blocks = get_valid_blocks(sbi, segno, false);
 788
 789	if (valid_blocks == 0) {
 790		__locate_dirty_segment(sbi, segno, PRE);
 791		__remove_dirty_segment(sbi, segno, DIRTY);
 792	} else if (valid_blocks < sbi->blocks_per_seg) {
 793		__locate_dirty_segment(sbi, segno, DIRTY);
 794	} else {
 795		/* Recovery routine with SSR needs this */
 796		__remove_dirty_segment(sbi, segno, DIRTY);
 797	}
 798
 799	mutex_unlock(&dirty_i->seglist_lock);
 800}
 801
 802static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
 803		struct block_device *bdev, block_t lstart,
 804		block_t start, block_t len)
 805{
 806	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 807	struct list_head *pend_list;
 808	struct discard_cmd *dc;
 809
 810	f2fs_bug_on(sbi, !len);
 811
 812	pend_list = &dcc->pend_list[plist_idx(len)];
 813
 814	dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
 815	INIT_LIST_HEAD(&dc->list);
 816	dc->bdev = bdev;
 817	dc->lstart = lstart;
 818	dc->start = start;
 819	dc->len = len;
 820	dc->ref = 0;
 821	dc->state = D_PREP;
 822	dc->error = 0;
 823	init_completion(&dc->wait);
 824	list_add_tail(&dc->list, pend_list);
 825	atomic_inc(&dcc->discard_cmd_cnt);
 826	dcc->undiscard_blks += len;
 827
 828	return dc;
 829}
 830
 831static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
 832				struct block_device *bdev, block_t lstart,
 833				block_t start, block_t len,
 834				struct rb_node *parent, struct rb_node **p)
 835{
 836	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 837	struct discard_cmd *dc;
 838
 839	dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
 840
 841	rb_link_node(&dc->rb_node, parent, p);
 842	rb_insert_color(&dc->rb_node, &dcc->root);
 843
 844	return dc;
 845}
 846
 847static void __detach_discard_cmd(struct discard_cmd_control *dcc,
 848							struct discard_cmd *dc)
 849{
 850	if (dc->state == D_DONE)
 851		atomic_dec(&dcc->issing_discard);
 852
 853	list_del(&dc->list);
 854	rb_erase(&dc->rb_node, &dcc->root);
 855	dcc->undiscard_blks -= dc->len;
 856
 857	kmem_cache_free(discard_cmd_slab, dc);
 858
 859	atomic_dec(&dcc->discard_cmd_cnt);
 860}
 861
 862static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
 863							struct discard_cmd *dc)
 864{
 865	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 866
 867	trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
 868
 869	f2fs_bug_on(sbi, dc->ref);
 870
 871	if (dc->error == -EOPNOTSUPP)
 872		dc->error = 0;
 873
 874	if (dc->error)
 875		f2fs_msg(sbi->sb, KERN_INFO,
 876			"Issue discard(%u, %u, %u) failed, ret: %d",
 877			dc->lstart, dc->start, dc->len, dc->error);
 878	__detach_discard_cmd(dcc, dc);
 879}
 880
 881static void f2fs_submit_discard_endio(struct bio *bio)
 882{
 883	struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
 884
 885	dc->error = blk_status_to_errno(bio->bi_status);
 886	dc->state = D_DONE;
 887	complete_all(&dc->wait);
 888	bio_put(bio);
 889}
 890
 891static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
 892				block_t start, block_t end)
 893{
 894#ifdef CONFIG_F2FS_CHECK_FS
 895	struct seg_entry *sentry;
 896	unsigned int segno;
 897	block_t blk = start;
 898	unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
 899	unsigned long *map;
 900
 901	while (blk < end) {
 902		segno = GET_SEGNO(sbi, blk);
 903		sentry = get_seg_entry(sbi, segno);
 904		offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
 905
 906		if (end < START_BLOCK(sbi, segno + 1))
 907			size = GET_BLKOFF_FROM_SEG0(sbi, end);
 908		else
 909			size = max_blocks;
 910		map = (unsigned long *)(sentry->cur_valid_map);
 911		offset = __find_rev_next_bit(map, size, offset);
 912		f2fs_bug_on(sbi, offset != size);
 913		blk = START_BLOCK(sbi, segno + 1);
 914	}
 915#endif
 916}
 917
 918/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
 919static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
 920						struct discard_policy *dpolicy,
 921						struct discard_cmd *dc)
 922{
 923	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 924	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
 925					&(dcc->fstrim_list) : &(dcc->wait_list);
 926	struct bio *bio = NULL;
 927	int flag = dpolicy->sync ? REQ_SYNC : 0;
 928
 929	if (dc->state != D_PREP)
 930		return;
 931
 932	trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
 933
 934	dc->error = __blkdev_issue_discard(dc->bdev,
 935				SECTOR_FROM_BLOCK(dc->start),
 936				SECTOR_FROM_BLOCK(dc->len),
 937				GFP_NOFS, 0, &bio);
 938	if (!dc->error) {
 939		/* should keep before submission to avoid D_DONE right away */
 940		dc->state = D_SUBMIT;
 941		atomic_inc(&dcc->issued_discard);
 942		atomic_inc(&dcc->issing_discard);
 943		if (bio) {
 944			bio->bi_private = dc;
 945			bio->bi_end_io = f2fs_submit_discard_endio;
 946			bio->bi_opf |= flag;
 947			submit_bio(bio);
 948			list_move_tail(&dc->list, wait_list);
 949			__check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
 950
 951			f2fs_update_iostat(sbi, FS_DISCARD, 1);
 952		}
 953	} else {
 954		__remove_discard_cmd(sbi, dc);
 955	}
 956}
 957
 958static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
 959				struct block_device *bdev, block_t lstart,
 960				block_t start, block_t len,
 961				struct rb_node **insert_p,
 962				struct rb_node *insert_parent)
 963{
 964	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 965	struct rb_node **p;
 966	struct rb_node *parent = NULL;
 967	struct discard_cmd *dc = NULL;
 968
 969	if (insert_p && insert_parent) {
 970		parent = insert_parent;
 971		p = insert_p;
 972		goto do_insert;
 973	}
 974
 975	p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
 976do_insert:
 977	dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
 978	if (!dc)
 979		return NULL;
 980
 981	return dc;
 982}
 983
 984static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
 985						struct discard_cmd *dc)
 986{
 987	list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
 988}
 989
 990static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
 991				struct discard_cmd *dc, block_t blkaddr)
 992{
 993	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
 994	struct discard_info di = dc->di;
 995	bool modified = false;
 996
 997	if (dc->state == D_DONE || dc->len == 1) {
 998		__remove_discard_cmd(sbi, dc);
 999		return;
1000	}
1001
1002	dcc->undiscard_blks -= di.len;
1003
1004	if (blkaddr > di.lstart) {
1005		dc->len = blkaddr - dc->lstart;
1006		dcc->undiscard_blks += dc->len;
1007		__relocate_discard_cmd(dcc, dc);
1008		modified = true;
1009	}
1010
1011	if (blkaddr < di.lstart + di.len - 1) {
1012		if (modified) {
1013			__insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1014					di.start + blkaddr + 1 - di.lstart,
1015					di.lstart + di.len - 1 - blkaddr,
1016					NULL, NULL);
1017		} else {
1018			dc->lstart++;
1019			dc->len--;
1020			dc->start++;
1021			dcc->undiscard_blks += dc->len;
1022			__relocate_discard_cmd(dcc, dc);
1023		}
1024	}
1025}
1026
1027static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1028				struct block_device *bdev, block_t lstart,
1029				block_t start, block_t len)
1030{
1031	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1032	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1033	struct discard_cmd *dc;
1034	struct discard_info di = {0};
1035	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1036	block_t end = lstart + len;
1037
1038	mutex_lock(&dcc->cmd_lock);
1039
1040	dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1041					NULL, lstart,
1042					(struct rb_entry **)&prev_dc,
1043					(struct rb_entry **)&next_dc,
1044					&insert_p, &insert_parent, true);
1045	if (dc)
1046		prev_dc = dc;
1047
1048	if (!prev_dc) {
1049		di.lstart = lstart;
1050		di.len = next_dc ? next_dc->lstart - lstart : len;
1051		di.len = min(di.len, len);
1052		di.start = start;
1053	}
1054
1055	while (1) {
1056		struct rb_node *node;
1057		bool merged = false;
1058		struct discard_cmd *tdc = NULL;
1059
1060		if (prev_dc) {
1061			di.lstart = prev_dc->lstart + prev_dc->len;
1062			if (di.lstart < lstart)
1063				di.lstart = lstart;
1064			if (di.lstart >= end)
1065				break;
1066
1067			if (!next_dc || next_dc->lstart > end)
1068				di.len = end - di.lstart;
1069			else
1070				di.len = next_dc->lstart - di.lstart;
1071			di.start = start + di.lstart - lstart;
1072		}
1073
1074		if (!di.len)
1075			goto next;
1076
1077		if (prev_dc && prev_dc->state == D_PREP &&
1078			prev_dc->bdev == bdev &&
1079			__is_discard_back_mergeable(&di, &prev_dc->di)) {
1080			prev_dc->di.len += di.len;
1081			dcc->undiscard_blks += di.len;
1082			__relocate_discard_cmd(dcc, prev_dc);
1083			di = prev_dc->di;
1084			tdc = prev_dc;
1085			merged = true;
1086		}
1087
1088		if (next_dc && next_dc->state == D_PREP &&
1089			next_dc->bdev == bdev &&
1090			__is_discard_front_mergeable(&di, &next_dc->di)) {
1091			next_dc->di.lstart = di.lstart;
1092			next_dc->di.len += di.len;
1093			next_dc->di.start = di.start;
1094			dcc->undiscard_blks += di.len;
1095			__relocate_discard_cmd(dcc, next_dc);
1096			if (tdc)
1097				__remove_discard_cmd(sbi, tdc);
1098			merged = true;
1099		}
1100
1101		if (!merged) {
1102			__insert_discard_tree(sbi, bdev, di.lstart, di.start,
1103							di.len, NULL, NULL);
1104		}
1105 next:
1106		prev_dc = next_dc;
1107		if (!prev_dc)
1108			break;
1109
1110		node = rb_next(&prev_dc->rb_node);
1111		next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1112	}
1113
1114	mutex_unlock(&dcc->cmd_lock);
1115}
1116
1117static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1118		struct block_device *bdev, block_t blkstart, block_t blklen)
1119{
1120	block_t lblkstart = blkstart;
1121
1122	trace_f2fs_queue_discard(bdev, blkstart, blklen);
1123
1124	if (sbi->s_ndevs) {
1125		int devi = f2fs_target_device_index(sbi, blkstart);
1126
1127		blkstart -= FDEV(devi).start_blk;
1128	}
1129	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1130	return 0;
1131}
1132
1133static void __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
1134					struct discard_policy *dpolicy,
1135					unsigned int start, unsigned int end)
1136{
1137	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1138	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1139	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1140	struct discard_cmd *dc;
1141	struct blk_plug plug;
1142	int issued;
1143
1144next:
1145	issued = 0;
1146
1147	mutex_lock(&dcc->cmd_lock);
1148	f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1149
1150	dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1151					NULL, start,
1152					(struct rb_entry **)&prev_dc,
1153					(struct rb_entry **)&next_dc,
1154					&insert_p, &insert_parent, true);
1155	if (!dc)
1156		dc = next_dc;
1157
1158	blk_start_plug(&plug);
1159
1160	while (dc && dc->lstart <= end) {
1161		struct rb_node *node;
1162
1163		if (dc->len < dpolicy->granularity)
1164			goto skip;
1165
1166		if (dc->state != D_PREP) {
1167			list_move_tail(&dc->list, &dcc->fstrim_list);
1168			goto skip;
1169		}
1170
1171		__submit_discard_cmd(sbi, dpolicy, dc);
1172
1173		if (++issued >= dpolicy->max_requests) {
1174			start = dc->lstart + dc->len;
1175
1176			blk_finish_plug(&plug);
1177			mutex_unlock(&dcc->cmd_lock);
1178
1179			schedule();
1180
1181			goto next;
1182		}
1183skip:
1184		node = rb_next(&dc->rb_node);
1185		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1186
1187		if (fatal_signal_pending(current))
1188			break;
1189	}
1190
1191	blk_finish_plug(&plug);
1192	mutex_unlock(&dcc->cmd_lock);
1193}
1194
1195static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1196					struct discard_policy *dpolicy)
1197{
1198	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1199	struct list_head *pend_list;
1200	struct discard_cmd *dc, *tmp;
1201	struct blk_plug plug;
1202	int i, iter = 0, issued = 0;
1203	bool io_interrupted = false;
1204
1205	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1206		if (i + 1 < dpolicy->granularity)
1207			break;
1208		pend_list = &dcc->pend_list[i];
1209
1210		mutex_lock(&dcc->cmd_lock);
1211		if (list_empty(pend_list))
1212			goto next;
1213		f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1214		blk_start_plug(&plug);
1215		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1216			f2fs_bug_on(sbi, dc->state != D_PREP);
1217
1218			if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1219								!is_idle(sbi)) {
1220				io_interrupted = true;
1221				goto skip;
1222			}
1223
1224			__submit_discard_cmd(sbi, dpolicy, dc);
1225			issued++;
1226skip:
1227			if (++iter >= dpolicy->max_requests)
1228				break;
1229		}
1230		blk_finish_plug(&plug);
1231next:
1232		mutex_unlock(&dcc->cmd_lock);
1233
1234		if (iter >= dpolicy->max_requests)
1235			break;
1236	}
1237
1238	if (!issued && io_interrupted)
1239		issued = -1;
1240
1241	return issued;
1242}
1243
1244static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1245{
1246	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1247	struct list_head *pend_list;
1248	struct discard_cmd *dc, *tmp;
1249	int i;
1250	bool dropped = false;
1251
1252	mutex_lock(&dcc->cmd_lock);
1253	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1254		pend_list = &dcc->pend_list[i];
1255		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1256			f2fs_bug_on(sbi, dc->state != D_PREP);
1257			__remove_discard_cmd(sbi, dc);
1258			dropped = true;
1259		}
1260	}
1261	mutex_unlock(&dcc->cmd_lock);
1262
1263	return dropped;
1264}
1265
1266void drop_discard_cmd(struct f2fs_sb_info *sbi)
1267{
1268	__drop_discard_cmd(sbi);
1269}
1270
1271static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1272							struct discard_cmd *dc)
1273{
1274	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1275	unsigned int len = 0;
1276
1277	wait_for_completion_io(&dc->wait);
1278	mutex_lock(&dcc->cmd_lock);
1279	f2fs_bug_on(sbi, dc->state != D_DONE);
1280	dc->ref--;
1281	if (!dc->ref) {
1282		if (!dc->error)
1283			len = dc->len;
1284		__remove_discard_cmd(sbi, dc);
1285	}
1286	mutex_unlock(&dcc->cmd_lock);
1287
1288	return len;
1289}
1290
1291static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1292						struct discard_policy *dpolicy,
1293						block_t start, block_t end)
1294{
1295	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1296	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1297					&(dcc->fstrim_list) : &(dcc->wait_list);
1298	struct discard_cmd *dc, *tmp;
1299	bool need_wait;
1300	unsigned int trimmed = 0;
1301
1302next:
1303	need_wait = false;
1304
1305	mutex_lock(&dcc->cmd_lock);
1306	list_for_each_entry_safe(dc, tmp, wait_list, list) {
1307		if (dc->lstart + dc->len <= start || end <= dc->lstart)
1308			continue;
1309		if (dc->len < dpolicy->granularity)
1310			continue;
1311		if (dc->state == D_DONE && !dc->ref) {
1312			wait_for_completion_io(&dc->wait);
1313			if (!dc->error)
1314				trimmed += dc->len;
1315			__remove_discard_cmd(sbi, dc);
1316		} else {
1317			dc->ref++;
1318			need_wait = true;
1319			break;
1320		}
1321	}
1322	mutex_unlock(&dcc->cmd_lock);
1323
1324	if (need_wait) {
1325		trimmed += __wait_one_discard_bio(sbi, dc);
1326		goto next;
1327	}
1328
1329	return trimmed;
1330}
1331
1332static void __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1333						struct discard_policy *dpolicy)
1334{
1335	__wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1336}
1337
1338/* This should be covered by global mutex, &sit_i->sentry_lock */
1339static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1340{
1341	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1342	struct discard_cmd *dc;
1343	bool need_wait = false;
1344
1345	mutex_lock(&dcc->cmd_lock);
1346	dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1347	if (dc) {
1348		if (dc->state == D_PREP) {
1349			__punch_discard_cmd(sbi, dc, blkaddr);
1350		} else {
1351			dc->ref++;
1352			need_wait = true;
1353		}
1354	}
1355	mutex_unlock(&dcc->cmd_lock);
1356
1357	if (need_wait)
1358		__wait_one_discard_bio(sbi, dc);
1359}
1360
1361void stop_discard_thread(struct f2fs_sb_info *sbi)
1362{
1363	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1364
1365	if (dcc && dcc->f2fs_issue_discard) {
1366		struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1367
1368		dcc->f2fs_issue_discard = NULL;
1369		kthread_stop(discard_thread);
1370	}
1371}
1372
1373/* This comes from f2fs_put_super */
1374bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1375{
1376	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1377	struct discard_policy dpolicy;
1378	bool dropped;
1379
1380	init_discard_policy(&dpolicy, DPOLICY_UMOUNT, dcc->discard_granularity);
1381	__issue_discard_cmd(sbi, &dpolicy);
1382	dropped = __drop_discard_cmd(sbi);
1383	__wait_all_discard_cmd(sbi, &dpolicy);
1384
1385	return dropped;
1386}
1387
1388static int issue_discard_thread(void *data)
1389{
1390	struct f2fs_sb_info *sbi = data;
1391	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1392	wait_queue_head_t *q = &dcc->discard_wait_queue;
1393	struct discard_policy dpolicy;
1394	unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1395	int issued;
1396
1397	set_freezable();
1398
1399	do {
1400		init_discard_policy(&dpolicy, DPOLICY_BG,
1401					dcc->discard_granularity);
1402
1403		wait_event_interruptible_timeout(*q,
1404				kthread_should_stop() || freezing(current) ||
1405				dcc->discard_wake,
1406				msecs_to_jiffies(wait_ms));
1407		if (try_to_freeze())
1408			continue;
1409		if (f2fs_readonly(sbi->sb))
1410			continue;
1411		if (kthread_should_stop())
1412			return 0;
1413
1414		if (dcc->discard_wake)
1415			dcc->discard_wake = 0;
1416
1417		if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
1418			init_discard_policy(&dpolicy, DPOLICY_FORCE, 1);
1419
1420		sb_start_intwrite(sbi->sb);
1421
1422		issued = __issue_discard_cmd(sbi, &dpolicy);
1423		if (issued) {
1424			__wait_all_discard_cmd(sbi, &dpolicy);
1425			wait_ms = dpolicy.min_interval;
1426		} else {
1427			wait_ms = dpolicy.max_interval;
1428		}
1429
1430		sb_end_intwrite(sbi->sb);
1431
1432	} while (!kthread_should_stop());
1433	return 0;
1434}
1435
1436#ifdef CONFIG_BLK_DEV_ZONED
1437static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1438		struct block_device *bdev, block_t blkstart, block_t blklen)
1439{
1440	sector_t sector, nr_sects;
1441	block_t lblkstart = blkstart;
1442	int devi = 0;
1443
1444	if (sbi->s_ndevs) {
1445		devi = f2fs_target_device_index(sbi, blkstart);
1446		blkstart -= FDEV(devi).start_blk;
1447	}
1448
1449	/*
1450	 * We need to know the type of the zone: for conventional zones,
1451	 * use regular discard if the drive supports it. For sequential
1452	 * zones, reset the zone write pointer.
1453	 */
1454	switch (get_blkz_type(sbi, bdev, blkstart)) {
1455
1456	case BLK_ZONE_TYPE_CONVENTIONAL:
1457		if (!blk_queue_discard(bdev_get_queue(bdev)))
1458			return 0;
1459		return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1460	case BLK_ZONE_TYPE_SEQWRITE_REQ:
1461	case BLK_ZONE_TYPE_SEQWRITE_PREF:
1462		sector = SECTOR_FROM_BLOCK(blkstart);
1463		nr_sects = SECTOR_FROM_BLOCK(blklen);
1464
1465		if (sector & (bdev_zone_sectors(bdev) - 1) ||
1466				nr_sects != bdev_zone_sectors(bdev)) {
1467			f2fs_msg(sbi->sb, KERN_INFO,
1468				"(%d) %s: Unaligned discard attempted (block %x + %x)",
1469				devi, sbi->s_ndevs ? FDEV(devi).path: "",
1470				blkstart, blklen);
1471			return -EIO;
1472		}
1473		trace_f2fs_issue_reset_zone(bdev, blkstart);
1474		return blkdev_reset_zones(bdev, sector,
1475					  nr_sects, GFP_NOFS);
1476	default:
1477		/* Unknown zone type: broken device ? */
1478		return -EIO;
1479	}
1480}
1481#endif
1482
1483static int __issue_discard_async(struct f2fs_sb_info *sbi,
1484		struct block_device *bdev, block_t blkstart, block_t blklen)
1485{
1486#ifdef CONFIG_BLK_DEV_ZONED
1487	if (f2fs_sb_has_blkzoned(sbi->sb) &&
1488				bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1489		return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1490#endif
1491	return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1492}
1493
1494static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1495				block_t blkstart, block_t blklen)
1496{
1497	sector_t start = blkstart, len = 0;
1498	struct block_device *bdev;
1499	struct seg_entry *se;
1500	unsigned int offset;
1501	block_t i;
1502	int err = 0;
1503
1504	bdev = f2fs_target_device(sbi, blkstart, NULL);
1505
1506	for (i = blkstart; i < blkstart + blklen; i++, len++) {
1507		if (i != start) {
1508			struct block_device *bdev2 =
1509				f2fs_target_device(sbi, i, NULL);
1510
1511			if (bdev2 != bdev) {
1512				err = __issue_discard_async(sbi, bdev,
1513						start, len);
1514				if (err)
1515					return err;
1516				bdev = bdev2;
1517				start = i;
1518				len = 0;
1519			}
1520		}
1521
1522		se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1523		offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1524
1525		if (!f2fs_test_and_set_bit(offset, se->discard_map))
1526			sbi->discard_blks--;
1527	}
1528
1529	if (len)
1530		err = __issue_discard_async(sbi, bdev, start, len);
1531	return err;
1532}
1533
1534static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1535							bool check_only)
1536{
1537	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1538	int max_blocks = sbi->blocks_per_seg;
1539	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1540	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1541	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1542	unsigned long *discard_map = (unsigned long *)se->discard_map;
1543	unsigned long *dmap = SIT_I(sbi)->tmp_map;
1544	unsigned int start = 0, end = -1;
1545	bool force = (cpc->reason & CP_DISCARD);
1546	struct discard_entry *de = NULL;
1547	struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1548	int i;
1549
1550	if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
1551		return false;
1552
1553	if (!force) {
1554		if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
1555			SM_I(sbi)->dcc_info->nr_discards >=
1556				SM_I(sbi)->dcc_info->max_discards)
1557			return false;
1558	}
1559
1560	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1561	for (i = 0; i < entries; i++)
1562		dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1563				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1564
1565	while (force || SM_I(sbi)->dcc_info->nr_discards <=
1566				SM_I(sbi)->dcc_info->max_discards) {
1567		start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1568		if (start >= max_blocks)
1569			break;
1570
1571		end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1572		if (force && start && end != max_blocks
1573					&& (end - start) < cpc->trim_minlen)
1574			continue;
1575
1576		if (check_only)
1577			return true;
1578
1579		if (!de) {
1580			de = f2fs_kmem_cache_alloc(discard_entry_slab,
1581								GFP_F2FS_ZERO);
1582			de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1583			list_add_tail(&de->list, head);
1584		}
1585
1586		for (i = start; i < end; i++)
1587			__set_bit_le(i, (void *)de->discard_map);
1588
1589		SM_I(sbi)->dcc_info->nr_discards += end - start;
1590	}
1591	return false;
1592}
1593
1594void release_discard_addrs(struct f2fs_sb_info *sbi)
1595{
1596	struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1597	struct discard_entry *entry, *this;
1598
1599	/* drop caches */
1600	list_for_each_entry_safe(entry, this, head, list) {
1601		list_del(&entry->list);
1602		kmem_cache_free(discard_entry_slab, entry);
1603	}
1604}
1605
1606/*
1607 * Should call clear_prefree_segments after checkpoint is done.
1608 */
1609static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1610{
1611	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1612	unsigned int segno;
1613
1614	mutex_lock(&dirty_i->seglist_lock);
1615	for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1616		__set_test_and_free(sbi, segno);
1617	mutex_unlock(&dirty_i->seglist_lock);
1618}
1619
1620void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1621{
1622	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1623	struct list_head *head = &dcc->entry_list;
1624	struct discard_entry *entry, *this;
1625	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1626	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1627	unsigned int start = 0, end = -1;
1628	unsigned int secno, start_segno;
1629	bool force = (cpc->reason & CP_DISCARD);
1630
1631	mutex_lock(&dirty_i->seglist_lock);
1632
1633	while (1) {
1634		int i;
1635		start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1636		if (start >= MAIN_SEGS(sbi))
1637			break;
1638		end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1639								start + 1);
1640
1641		for (i = start; i < end; i++)
1642			clear_bit(i, prefree_map);
1643
1644		dirty_i->nr_dirty[PRE] -= end - start;
1645
1646		if (!test_opt(sbi, DISCARD))
1647			continue;
1648
1649		if (force && start >= cpc->trim_start &&
1650					(end - 1) <= cpc->trim_end)
1651				continue;
1652
1653		if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1654			f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1655				(end - start) << sbi->log_blocks_per_seg);
1656			continue;
1657		}
1658next:
1659		secno = GET_SEC_FROM_SEG(sbi, start);
1660		start_segno = GET_SEG_FROM_SEC(sbi, secno);
1661		if (!IS_CURSEC(sbi, secno) &&
1662			!get_valid_blocks(sbi, start, true))
1663			f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1664				sbi->segs_per_sec << sbi->log_blocks_per_seg);
1665
1666		start = start_segno + sbi->segs_per_sec;
1667		if (start < end)
1668			goto next;
1669		else
1670			end = start - 1;
1671	}
1672	mutex_unlock(&dirty_i->seglist_lock);
1673
1674	/* send small discards */
1675	list_for_each_entry_safe(entry, this, head, list) {
1676		unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1677		bool is_valid = test_bit_le(0, entry->discard_map);
1678
1679find_next:
1680		if (is_valid) {
1681			next_pos = find_next_zero_bit_le(entry->discard_map,
1682					sbi->blocks_per_seg, cur_pos);
1683			len = next_pos - cur_pos;
1684
1685			if (f2fs_sb_has_blkzoned(sbi->sb) ||
1686			    (force && len < cpc->trim_minlen))
1687				goto skip;
1688
1689			f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1690									len);
1691			total_len += len;
1692		} else {
1693			next_pos = find_next_bit_le(entry->discard_map,
1694					sbi->blocks_per_seg, cur_pos);
1695		}
1696skip:
1697		cur_pos = next_pos;
1698		is_valid = !is_valid;
1699
1700		if (cur_pos < sbi->blocks_per_seg)
1701			goto find_next;
1702
1703		list_del(&entry->list);
1704		dcc->nr_discards -= total_len;
1705		kmem_cache_free(discard_entry_slab, entry);
1706	}
1707
1708	wake_up_discard_thread(sbi, false);
1709}
1710
1711void init_discard_policy(struct discard_policy *dpolicy,
1712				int discard_type, unsigned int granularity)
1713{
1714	/* common policy */
1715	dpolicy->type = discard_type;
1716	dpolicy->sync = true;
1717	dpolicy->granularity = granularity;
1718
1719	dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1720	dpolicy->io_aware_gran = MAX_PLIST_NUM;
1721
1722	if (discard_type == DPOLICY_BG) {
1723		dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1724		dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1725		dpolicy->io_aware = true;
1726	} else if (discard_type == DPOLICY_FORCE) {
1727		dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1728		dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1729		dpolicy->io_aware = false;
1730	} else if (discard_type == DPOLICY_FSTRIM) {
1731		dpolicy->io_aware = false;
1732	} else if (discard_type == DPOLICY_UMOUNT) {
1733		dpolicy->io_aware = false;
1734	}
1735}
1736
1737static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1738{
1739	dev_t dev = sbi->sb->s_bdev->bd_dev;
1740	struct discard_cmd_control *dcc;
1741	int err = 0, i;
1742
1743	if (SM_I(sbi)->dcc_info) {
1744		dcc = SM_I(sbi)->dcc_info;
1745		goto init_thread;
1746	}
1747
1748	dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
1749	if (!dcc)
1750		return -ENOMEM;
1751
1752	dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
1753	INIT_LIST_HEAD(&dcc->entry_list);
1754	for (i = 0; i < MAX_PLIST_NUM; i++)
1755		INIT_LIST_HEAD(&dcc->pend_list[i]);
1756	INIT_LIST_HEAD(&dcc->wait_list);
1757	INIT_LIST_HEAD(&dcc->fstrim_list);
1758	mutex_init(&dcc->cmd_lock);
1759	atomic_set(&dcc->issued_discard, 0);
1760	atomic_set(&dcc->issing_discard, 0);
1761	atomic_set(&dcc->discard_cmd_cnt, 0);
1762	dcc->nr_discards = 0;
1763	dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1764	dcc->undiscard_blks = 0;
1765	dcc->root = RB_ROOT;
1766
1767	init_waitqueue_head(&dcc->discard_wait_queue);
1768	SM_I(sbi)->dcc_info = dcc;
1769init_thread:
1770	dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1771				"f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1772	if (IS_ERR(dcc->f2fs_issue_discard)) {
1773		err = PTR_ERR(dcc->f2fs_issue_discard);
1774		kfree(dcc);
1775		SM_I(sbi)->dcc_info = NULL;
1776		return err;
1777	}
1778
1779	return err;
1780}
1781
1782static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1783{
1784	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1785
1786	if (!dcc)
1787		return;
1788
1789	stop_discard_thread(sbi);
1790
1791	kfree(dcc);
1792	SM_I(sbi)->dcc_info = NULL;
1793}
1794
1795static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1796{
1797	struct sit_info *sit_i = SIT_I(sbi);
1798
1799	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1800		sit_i->dirty_sentries++;
1801		return false;
1802	}
1803
1804	return true;
1805}
1806
1807static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1808					unsigned int segno, int modified)
1809{
1810	struct seg_entry *se = get_seg_entry(sbi, segno);
1811	se->type = type;
1812	if (modified)
1813		__mark_sit_entry_dirty(sbi, segno);
1814}
1815
1816static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1817{
1818	struct seg_entry *se;
1819	unsigned int segno, offset;
1820	long int new_vblocks;
1821	bool exist;
1822#ifdef CONFIG_F2FS_CHECK_FS
1823	bool mir_exist;
1824#endif
1825
1826	segno = GET_SEGNO(sbi, blkaddr);
1827
1828	se = get_seg_entry(sbi, segno);
1829	new_vblocks = se->valid_blocks + del;
1830	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1831
1832	f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
1833				(new_vblocks > sbi->blocks_per_seg)));
1834
1835	se->valid_blocks = new_vblocks;
1836	se->mtime = get_mtime(sbi);
1837	SIT_I(sbi)->max_mtime = se->mtime;
1838
1839	/* Update valid block bitmap */
1840	if (del > 0) {
1841		exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
1842#ifdef CONFIG_F2FS_CHECK_FS
1843		mir_exist = f2fs_test_and_set_bit(offset,
1844						se->cur_valid_map_mir);
1845		if (unlikely(exist != mir_exist)) {
1846			f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1847				"when setting bitmap, blk:%u, old bit:%d",
1848				blkaddr, exist);
1849			f2fs_bug_on(sbi, 1);
1850		}
1851#endif
1852		if (unlikely(exist)) {
1853			f2fs_msg(sbi->sb, KERN_ERR,
1854				"Bitmap was wrongly set, blk:%u", blkaddr);
1855			f2fs_bug_on(sbi, 1);
1856			se->valid_blocks--;
1857			del = 0;
1858		}
1859
1860		if (f2fs_discard_en(sbi) &&
1861			!f2fs_test_and_set_bit(offset, se->discard_map))
1862			sbi->discard_blks--;
1863
1864		/* don't overwrite by SSR to keep node chain */
1865		if (IS_NODESEG(se->type)) {
1866			if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1867				se->ckpt_valid_blocks++;
1868		}
1869	} else {
1870		exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
1871#ifdef CONFIG_F2FS_CHECK_FS
1872		mir_exist = f2fs_test_and_clear_bit(offset,
1873						se->cur_valid_map_mir);
1874		if (unlikely(exist != mir_exist)) {
1875			f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1876				"when clearing bitmap, blk:%u, old bit:%d",
1877				blkaddr, exist);
1878			f2fs_bug_on(sbi, 1);
1879		}
1880#endif
1881		if (unlikely(!exist)) {
1882			f2fs_msg(sbi->sb, KERN_ERR,
1883				"Bitmap was wrongly cleared, blk:%u", blkaddr);
1884			f2fs_bug_on(sbi, 1);
1885			se->valid_blocks++;
1886			del = 0;
1887		}
1888
1889		if (f2fs_discard_en(sbi) &&
1890			f2fs_test_and_clear_bit(offset, se->discard_map))
1891			sbi->discard_blks++;
1892	}
1893	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1894		se->ckpt_valid_blocks += del;
1895
1896	__mark_sit_entry_dirty(sbi, segno);
1897
1898	/* update total number of valid blocks to be written in ckpt area */
1899	SIT_I(sbi)->written_valid_blocks += del;
1900
1901	if (sbi->segs_per_sec > 1)
1902		get_sec_entry(sbi, segno)->valid_blocks += del;
1903}
1904
1905void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1906{
1907	unsigned int segno = GET_SEGNO(sbi, addr);
1908	struct sit_info *sit_i = SIT_I(sbi);
1909
1910	f2fs_bug_on(sbi, addr == NULL_ADDR);
1911	if (addr == NEW_ADDR)
1912		return;
1913
1914	/* add it into sit main buffer */
1915	down_write(&sit_i->sentry_lock);
1916
1917	update_sit_entry(sbi, addr, -1);
1918
1919	/* add it into dirty seglist */
1920	locate_dirty_segment(sbi, segno);
1921
1922	up_write(&sit_i->sentry_lock);
1923}
1924
1925bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1926{
1927	struct sit_info *sit_i = SIT_I(sbi);
1928	unsigned int segno, offset;
1929	struct seg_entry *se;
1930	bool is_cp = false;
1931
1932	if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1933		return true;
1934
1935	down_read(&sit_i->sentry_lock);
1936
1937	segno = GET_SEGNO(sbi, blkaddr);
1938	se = get_seg_entry(sbi, segno);
1939	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1940
1941	if (f2fs_test_bit(offset, se->ckpt_valid_map))
1942		is_cp = true;
1943
1944	up_read(&sit_i->sentry_lock);
1945
1946	return is_cp;
1947}
1948
1949/*
1950 * This function should be resided under the curseg_mutex lock
1951 */
1952static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1953					struct f2fs_summary *sum)
1954{
1955	struct curseg_info *curseg = CURSEG_I(sbi, type);
1956	void *addr = curseg->sum_blk;
1957	addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1958	memcpy(addr, sum, sizeof(struct f2fs_summary));
1959}
1960
1961/*
1962 * Calculate the number of current summary pages for writing
1963 */
1964int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1965{
1966	int valid_sum_count = 0;
1967	int i, sum_in_page;
1968
1969	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1970		if (sbi->ckpt->alloc_type[i] == SSR)
1971			valid_sum_count += sbi->blocks_per_seg;
1972		else {
1973			if (for_ra)
1974				valid_sum_count += le16_to_cpu(
1975					F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1976			else
1977				valid_sum_count += curseg_blkoff(sbi, i);
1978		}
1979	}
1980
1981	sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1982			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1983	if (valid_sum_count <= sum_in_page)
1984		return 1;
1985	else if ((valid_sum_count - sum_in_page) <=
1986		(PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1987		return 2;
1988	return 3;
1989}
1990
1991/*
1992 * Caller should put this summary page
1993 */
1994struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1995{
1996	return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1997}
1998
1999void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
2000{
2001	struct page *page = grab_meta_page(sbi, blk_addr);
2002
2003	memcpy(page_address(page), src, PAGE_SIZE);
2004	set_page_dirty(page);
2005	f2fs_put_page(page, 1);
2006}
2007
2008static void write_sum_page(struct f2fs_sb_info *sbi,
2009			struct f2fs_summary_block *sum_blk, block_t blk_addr)
2010{
2011	update_meta_page(sbi, (void *)sum_blk, blk_addr);
2012}
2013
2014static void write_current_sum_page(struct f2fs_sb_info *sbi,
2015						int type, block_t blk_addr)
2016{
2017	struct curseg_info *curseg = CURSEG_I(sbi, type);
2018	struct page *page = grab_meta_page(sbi, blk_addr);
2019	struct f2fs_summary_block *src = curseg->sum_blk;
2020	struct f2fs_summary_block *dst;
2021
2022	dst = (struct f2fs_summary_block *)page_address(page);
2023
2024	mutex_lock(&curseg->curseg_mutex);
2025
2026	down_read(&curseg->journal_rwsem);
2027	memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2028	up_read(&curseg->journal_rwsem);
2029
2030	memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2031	memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2032
2033	mutex_unlock(&curseg->curseg_mutex);
2034
2035	set_page_dirty(page);
2036	f2fs_put_page(page, 1);
2037}
2038
2039static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2040{
2041	struct curseg_info *curseg = CURSEG_I(sbi, type);
2042	unsigned int segno = curseg->segno + 1;
2043	struct free_segmap_info *free_i = FREE_I(sbi);
2044
2045	if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2046		return !test_bit(segno, free_i->free_segmap);
2047	return 0;
2048}
2049
2050/*
2051 * Find a new segment from the free segments bitmap to right order
2052 * This function should be returned with success, otherwise BUG
2053 */
2054static void get_new_segment(struct f2fs_sb_info *sbi,
2055			unsigned int *newseg, bool new_sec, int dir)
2056{
2057	struct free_segmap_info *free_i = FREE_I(sbi);
2058	unsigned int segno, secno, zoneno;
2059	unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2060	unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2061	unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2062	unsigned int left_start = hint;
2063	bool init = true;
2064	int go_left = 0;
2065	int i;
2066
2067	spin_lock(&free_i->segmap_lock);
2068
2069	if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2070		segno = find_next_zero_bit(free_i->free_segmap,
2071			GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2072		if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2073			goto got_it;
2074	}
2075find_other_zone:
2076	secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2077	if (secno >= MAIN_SECS(sbi)) {
2078		if (dir == ALLOC_RIGHT) {
2079			secno = find_next_zero_bit(free_i->free_secmap,
2080							MAIN_SECS(sbi), 0);
2081			f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2082		} else {
2083			go_left = 1;
2084			left_start = hint - 1;
2085		}
2086	}
2087	if (go_left == 0)
2088		goto skip_left;
2089
2090	while (test_bit(left_start, free_i->free_secmap)) {
2091		if (left_start > 0) {
2092			left_start--;
2093			continue;
2094		}
2095		left_start = find_next_zero_bit(free_i->free_secmap,
2096							MAIN_SECS(sbi), 0);
2097		f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2098		break;
2099	}
2100	secno = left_start;
2101skip_left:
2102	segno = GET_SEG_FROM_SEC(sbi, secno);
2103	zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2104
2105	/* give up on finding another zone */
2106	if (!init)
2107		goto got_it;
2108	if (sbi->secs_per_zone == 1)
2109		goto got_it;
2110	if (zoneno == old_zoneno)
2111		goto got_it;
2112	if (dir == ALLOC_LEFT) {
2113		if (!go_left && zoneno + 1 >= total_zones)
2114			goto got_it;
2115		if (go_left && zoneno == 0)
2116			goto got_it;
2117	}
2118	for (i = 0; i < NR_CURSEG_TYPE; i++)
2119		if (CURSEG_I(sbi, i)->zone == zoneno)
2120			break;
2121
2122	if (i < NR_CURSEG_TYPE) {
2123		/* zone is in user, try another */
2124		if (go_left)
2125			hint = zoneno * sbi->secs_per_zone - 1;
2126		else if (zoneno + 1 >= total_zones)
2127			hint = 0;
2128		else
2129			hint = (zoneno + 1) * sbi->secs_per_zone;
2130		init = false;
2131		goto find_other_zone;
2132	}
2133got_it:
2134	/* set it as dirty segment in free segmap */
2135	f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2136	__set_inuse(sbi, segno);
2137	*newseg = segno;
2138	spin_unlock(&free_i->segmap_lock);
2139}
2140
2141static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2142{
2143	struct curseg_info *curseg = CURSEG_I(sbi, type);
2144	struct summary_footer *sum_footer;
2145
2146	curseg->segno = curseg->next_segno;
2147	curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2148	curseg->next_blkoff = 0;
2149	curseg->next_segno = NULL_SEGNO;
2150
2151	sum_footer = &(curseg->sum_blk->footer);
2152	memset(sum_footer, 0, sizeof(struct summary_footer));
2153	if (IS_DATASEG(type))
2154		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2155	if (IS_NODESEG(type))
2156		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2157	__set_sit_entry_type(sbi, type, curseg->segno, modified);
2158}
2159
2160static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2161{
2162	/* if segs_per_sec is large than 1, we need to keep original policy. */
2163	if (sbi->segs_per_sec != 1)
2164		return CURSEG_I(sbi, type)->segno;
2165
2166	if (test_opt(sbi, NOHEAP) &&
2167		(type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2168		return 0;
2169
2170	if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2171		return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2172
2173	/* find segments from 0 to reuse freed segments */
2174	if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2175		return 0;
2176
2177	return CURSEG_I(sbi, type)->segno;
2178}
2179
2180/*
2181 * Allocate a current working segment.
2182 * This function always allocates a free segment in LFS manner.
2183 */
2184static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2185{
2186	struct curseg_info *curseg = CURSEG_I(sbi, type);
2187	unsigned int segno = curseg->segno;
2188	int dir = ALLOC_LEFT;
2189
2190	write_sum_page(sbi, curseg->sum_blk,
2191				GET_SUM_BLOCK(sbi, segno));
2192	if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2193		dir = ALLOC_RIGHT;
2194
2195	if (test_opt(sbi, NOHEAP))
2196		dir = ALLOC_RIGHT;
2197
2198	segno = __get_next_segno(sbi, type);
2199	get_new_segment(sbi, &segno, new_sec, dir);
2200	curseg->next_segno = segno;
2201	reset_curseg(sbi, type, 1);
2202	curseg->alloc_type = LFS;
2203}
2204
2205static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2206			struct curseg_info *seg, block_t start)
2207{
2208	struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2209	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2210	unsigned long *target_map = SIT_I(sbi)->tmp_map;
2211	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2212	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2213	int i, pos;
2214
2215	for (i = 0; i < entries; i++)
2216		target_map[i] = ckpt_map[i] | cur_map[i];
2217
2218	pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2219
2220	seg->next_blkoff = pos;
2221}
2222
2223/*
2224 * If a segment is written by LFS manner, next block offset is just obtained
2225 * by increasing the current block offset. However, if a segment is written by
2226 * SSR manner, next block offset obtained by calling __next_free_blkoff
2227 */
2228static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2229				struct curseg_info *seg)
2230{
2231	if (seg->alloc_type == SSR)
2232		__next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2233	else
2234		seg->next_blkoff++;
2235}
2236
2237/*
2238 * This function always allocates a used segment(from dirty seglist) by SSR
2239 * manner, so it should recover the existing segment information of valid blocks
2240 */
2241static void change_curseg(struct f2fs_sb_info *sbi, int type)
2242{
2243	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2244	struct curseg_info *curseg = CURSEG_I(sbi, type);
2245	unsigned int new_segno = curseg->next_segno;
2246	struct f2fs_summary_block *sum_node;
2247	struct page *sum_page;
2248
2249	write_sum_page(sbi, curseg->sum_blk,
2250				GET_SUM_BLOCK(sbi, curseg->segno));
2251	__set_test_and_inuse(sbi, new_segno);
2252
2253	mutex_lock(&dirty_i->seglist_lock);
2254	__remove_dirty_segment(sbi, new_segno, PRE);
2255	__remove_dirty_segment(sbi, new_segno, DIRTY);
2256	mutex_unlock(&dirty_i->seglist_lock);
2257
2258	reset_curseg(sbi, type, 1);
2259	curseg->alloc_type = SSR;
2260	__next_free_blkoff(sbi, curseg, 0);
2261
2262	sum_page = get_sum_page(sbi, new_segno);
2263	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2264	memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2265	f2fs_put_page(sum_page, 1);
2266}
2267
2268static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2269{
2270	struct curseg_info *curseg = CURSEG_I(sbi, type);
2271	const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2272	unsigned segno = NULL_SEGNO;
2273	int i, cnt;
2274	bool reversed = false;
2275
2276	/* need_SSR() already forces to do this */
2277	if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2278		curseg->next_segno = segno;
2279		return 1;
2280	}
2281
2282	/* For node segments, let's do SSR more intensively */
2283	if (IS_NODESEG(type)) {
2284		if (type >= CURSEG_WARM_NODE) {
2285			reversed = true;
2286			i = CURSEG_COLD_NODE;
2287		} else {
2288			i = CURSEG_HOT_NODE;
2289		}
2290		cnt = NR_CURSEG_NODE_TYPE;
2291	} else {
2292		if (type >= CURSEG_WARM_DATA) {
2293			reversed = true;
2294			i = CURSEG_COLD_DATA;
2295		} else {
2296			i = CURSEG_HOT_DATA;
2297		}
2298		cnt = NR_CURSEG_DATA_TYPE;
2299	}
2300
2301	for (; cnt-- > 0; reversed ? i-- : i++) {
2302		if (i == type)
2303			continue;
2304		if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2305			curseg->next_segno = segno;
2306			return 1;
2307		}
2308	}
2309	return 0;
2310}
2311
2312/*
2313 * flush out current segment and replace it with new segment
2314 * This function should be returned with success, otherwise BUG
2315 */
2316static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2317						int type, bool force)
2318{
2319	struct curseg_info *curseg = CURSEG_I(sbi, type);
2320
2321	if (force)
2322		new_curseg(sbi, type, true);
2323	else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2324					type == CURSEG_WARM_NODE)
2325		new_curseg(sbi, type, false);
2326	else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2327		new_curseg(sbi, type, false);
2328	else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
2329		change_curseg(sbi, type);
2330	else
2331		new_curseg(sbi, type, false);
2332
2333	stat_inc_seg_type(sbi, curseg);
2334}
2335
2336void allocate_new_segments(struct f2fs_sb_info *sbi)
2337{
2338	struct curseg_info *curseg;
2339	unsigned int old_segno;
2340	int i;
2341
2342	down_write(&SIT_I(sbi)->sentry_lock);
2343
2344	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2345		curseg = CURSEG_I(sbi, i);
2346		old_segno = curseg->segno;
2347		SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2348		locate_dirty_segment(sbi, old_segno);
2349	}
2350
2351	up_write(&SIT_I(sbi)->sentry_lock);
2352}
2353
2354static const struct segment_allocation default_salloc_ops = {
2355	.allocate_segment = allocate_segment_by_default,
2356};
2357
2358bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2359{
2360	__u64 trim_start = cpc->trim_start;
2361	bool has_candidate = false;
2362
2363	down_write(&SIT_I(sbi)->sentry_lock);
2364	for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2365		if (add_discard_addrs(sbi, cpc, true)) {
2366			has_candidate = true;
2367			break;
2368		}
2369	}
2370	up_write(&SIT_I(sbi)->sentry_lock);
2371
2372	cpc->trim_start = trim_start;
2373	return has_candidate;
2374}
2375
2376int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2377{
2378	__u64 start = F2FS_BYTES_TO_BLK(range->start);
2379	__u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2380	unsigned int start_segno, end_segno, cur_segno;
2381	block_t start_block, end_block;
2382	struct cp_control cpc;
2383	struct discard_policy dpolicy;
2384	unsigned long long trimmed = 0;
2385	int err = 0;
2386
2387	if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2388		return -EINVAL;
2389
2390	if (end <= MAIN_BLKADDR(sbi))
2391		goto out;
2392
2393	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2394		f2fs_msg(sbi->sb, KERN_WARNING,
2395			"Found FS corruption, run fsck to fix.");
2396		goto out;
2397	}
2398
2399	/* start/end segment number in main_area */
2400	start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2401	end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2402						GET_SEGNO(sbi, end);
2403
2404	cpc.reason = CP_DISCARD;
2405	cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2406
2407	/* do checkpoint to issue discard commands safely */
2408	for (cur_segno = start_segno; cur_segno <= end_segno;
2409					cur_segno = cpc.trim_end + 1) {
2410		cpc.trim_start = cur_segno;
2411
2412		if (sbi->discard_blks == 0)
2413			break;
2414		else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
2415			cpc.trim_end = end_segno;
2416		else
2417			cpc.trim_end = min_t(unsigned int,
2418				rounddown(cur_segno +
2419				BATCHED_TRIM_SEGMENTS(sbi),
2420				sbi->segs_per_sec) - 1, end_segno);
2421
2422		mutex_lock(&sbi->gc_mutex);
2423		err = write_checkpoint(sbi, &cpc);
2424		mutex_unlock(&sbi->gc_mutex);
2425		if (err)
2426			break;
2427
2428		schedule();
2429	}
2430
2431	start_block = START_BLOCK(sbi, start_segno);
2432	end_block = START_BLOCK(sbi, min(cur_segno, end_segno) + 1);
2433
2434	init_discard_policy(&dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2435	__issue_discard_cmd_range(sbi, &dpolicy, start_block, end_block);
2436	trimmed = __wait_discard_cmd_range(sbi, &dpolicy,
2437					start_block, end_block);
2438out:
2439	range->len = F2FS_BLK_TO_BYTES(trimmed);
2440	return err;
2441}
2442
2443static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2444{
2445	struct curseg_info *curseg = CURSEG_I(sbi, type);
2446	if (curseg->next_blkoff < sbi->blocks_per_seg)
2447		return true;
2448	return false;
2449}
2450
2451int rw_hint_to_seg_type(enum rw_hint hint)
2452{
2453	switch (hint) {
2454	case WRITE_LIFE_SHORT:
2455		return CURSEG_HOT_DATA;
2456	case WRITE_LIFE_EXTREME:
2457		return CURSEG_COLD_DATA;
2458	default:
2459		return CURSEG_WARM_DATA;
2460	}
2461}
2462
2463/* This returns write hints for each segment type. This hints will be
2464 * passed down to block layer. There are mapping tables which depend on
2465 * the mount option 'whint_mode'.
2466 *
2467 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2468 *
2469 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2470 *
2471 * User                  F2FS                     Block
2472 * ----                  ----                     -----
2473 *                       META                     WRITE_LIFE_NOT_SET
2474 *                       HOT_NODE                 "
2475 *                       WARM_NODE                "
2476 *                       COLD_NODE                "
2477 * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2478 * extension list        "                        "
2479 *
2480 * -- buffered io
2481 * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2482 * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2483 * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2484 * WRITE_LIFE_NONE       "                        "
2485 * WRITE_LIFE_MEDIUM     "                        "
2486 * WRITE_LIFE_LONG       "                        "
2487 *
2488 * -- direct io
2489 * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2490 * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2491 * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2492 * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2493 * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2494 * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2495 *
2496 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2497 *
2498 * User                  F2FS                     Block
2499 * ----                  ----                     -----
2500 *                       META                     WRITE_LIFE_MEDIUM;
2501 *                       HOT_NODE                 WRITE_LIFE_NOT_SET
2502 *                       WARM_NODE                "
2503 *                       COLD_NODE                WRITE_LIFE_NONE
2504 * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2505 * extension list        "                        "
2506 *
2507 * -- buffered io
2508 * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2509 * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2510 * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
2511 * WRITE_LIFE_NONE       "                        "
2512 * WRITE_LIFE_MEDIUM     "                        "
2513 * WRITE_LIFE_LONG       "                        "
2514 *
2515 * -- direct io
2516 * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2517 * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2518 * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2519 * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2520 * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2521 * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2522 */
2523
2524enum rw_hint io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2525				enum page_type type, enum temp_type temp)
2526{
2527	if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2528		if (type == DATA) {
2529			if (temp == WARM)
2530				return WRITE_LIFE_NOT_SET;
2531			else if (temp == HOT)
2532				return WRITE_LIFE_SHORT;
2533			else if (temp == COLD)
2534				return WRITE_LIFE_EXTREME;
2535		} else {
2536			return WRITE_LIFE_NOT_SET;
2537		}
2538	} else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2539		if (type == DATA) {
2540			if (temp == WARM)
2541				return WRITE_LIFE_LONG;
2542			else if (temp == HOT)
2543				return WRITE_LIFE_SHORT;
2544			else if (temp == COLD)
2545				return WRITE_LIFE_EXTREME;
2546		} else if (type == NODE) {
2547			if (temp == WARM || temp == HOT)
2548				return WRITE_LIFE_NOT_SET;
2549			else if (temp == COLD)
2550				return WRITE_LIFE_NONE;
2551		} else if (type == META) {
2552			return WRITE_LIFE_MEDIUM;
2553		}
2554	}
2555	return WRITE_LIFE_NOT_SET;
2556}
2557
2558static int __get_segment_type_2(struct f2fs_io_info *fio)
2559{
2560	if (fio->type == DATA)
2561		return CURSEG_HOT_DATA;
2562	else
2563		return CURSEG_HOT_NODE;
2564}
2565
2566static int __get_segment_type_4(struct f2fs_io_info *fio)
2567{
2568	if (fio->type == DATA) {
2569		struct inode *inode = fio->page->mapping->host;
2570
2571		if (S_ISDIR(inode->i_mode))
2572			return CURSEG_HOT_DATA;
2573		else
2574			return CURSEG_COLD_DATA;
2575	} else {
2576		if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2577			return CURSEG_WARM_NODE;
2578		else
2579			return CURSEG_COLD_NODE;
2580	}
2581}
2582
2583static int __get_segment_type_6(struct f2fs_io_info *fio)
2584{
2585	if (fio->type == DATA) {
2586		struct inode *inode = fio->page->mapping->host;
2587
2588		if (is_cold_data(fio->page) || file_is_cold(inode))
2589			return CURSEG_COLD_DATA;
2590		if (file_is_hot(inode) ||
2591				is_inode_flag_set(inode, FI_HOT_DATA))
2592			return CURSEG_HOT_DATA;
2593		return rw_hint_to_seg_type(inode->i_write_hint);
2594	} else {
2595		if (IS_DNODE(fio->page))
2596			return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2597						CURSEG_HOT_NODE;
2598		return CURSEG_COLD_NODE;
2599	}
2600}
2601
2602static int __get_segment_type(struct f2fs_io_info *fio)
2603{
2604	int type = 0;
2605
2606	switch (F2FS_OPTION(fio->sbi).active_logs) {
2607	case 2:
2608		type = __get_segment_type_2(fio);
2609		break;
2610	case 4:
2611		type = __get_segment_type_4(fio);
2612		break;
2613	case 6:
2614		type = __get_segment_type_6(fio);
2615		break;
2616	default:
2617		f2fs_bug_on(fio->sbi, true);
2618	}
2619
2620	if (IS_HOT(type))
2621		fio->temp = HOT;
2622	else if (IS_WARM(type))
2623		fio->temp = WARM;
2624	else
2625		fio->temp = COLD;
2626	return type;
2627}
2628
2629void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2630		block_t old_blkaddr, block_t *new_blkaddr,
2631		struct f2fs_summary *sum, int type,
2632		struct f2fs_io_info *fio, bool add_list)
2633{
2634	struct sit_info *sit_i = SIT_I(sbi);
2635	struct curseg_info *curseg = CURSEG_I(sbi, type);
2636
2637	down_read(&SM_I(sbi)->curseg_lock);
2638
2639	mutex_lock(&curseg->curseg_mutex);
2640	down_write(&sit_i->sentry_lock);
2641
2642	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2643
2644	f2fs_wait_discard_bio(sbi, *new_blkaddr);
2645
2646	/*
2647	 * __add_sum_entry should be resided under the curseg_mutex
2648	 * because, this function updates a summary entry in the
2649	 * current summary block.
2650	 */
2651	__add_sum_entry(sbi, type, sum);
2652
2653	__refresh_next_blkoff(sbi, curseg);
2654
2655	stat_inc_block_count(sbi, curseg);
2656
2657	/*
2658	 * SIT information should be updated before segment allocation,
2659	 * since SSR needs latest valid block information.
2660	 */
2661	update_sit_entry(sbi, *new_blkaddr, 1);
2662	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2663		update_sit_entry(sbi, old_blkaddr, -1);
2664
2665	if (!__has_curseg_space(sbi, type))
2666		sit_i->s_ops->allocate_segment(sbi, type, false);
2667
2668	/*
2669	 * segment dirty status should be updated after segment allocation,
2670	 * so we just need to update status only one time after previous
2671	 * segment being closed.
2672	 */
2673	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2674	locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
2675
2676	up_write(&sit_i->sentry_lock);
2677
2678	if (page && IS_NODESEG(type)) {
2679		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2680
2681		f2fs_inode_chksum_set(sbi, page);
2682	}
2683
2684	if (add_list) {
2685		struct f2fs_bio_info *io;
2686
2687		INIT_LIST_HEAD(&fio->list);
2688		fio->in_list = true;
2689		io = sbi->write_io[fio->type] + fio->temp;
2690		spin_lock(&io->io_lock);
2691		list_add_tail(&fio->list, &io->io_list);
2692		spin_unlock(&io->io_lock);
2693	}
2694
2695	mutex_unlock(&curseg->curseg_mutex);
2696
2697	up_read(&SM_I(sbi)->curseg_lock);
2698}
2699
2700static void update_device_state(struct f2fs_io_info *fio)
2701{
2702	struct f2fs_sb_info *sbi = fio->sbi;
2703	unsigned int devidx;
2704
2705	if (!sbi->s_ndevs)
2706		return;
2707
2708	devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2709
2710	/* update device state for fsync */
2711	set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
2712
2713	/* update device state for checkpoint */
2714	if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2715		spin_lock(&sbi->dev_lock);
2716		f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2717		spin_unlock(&sbi->dev_lock);
2718	}
2719}
2720
2721static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2722{
2723	int type = __get_segment_type(fio);
2724	int err;
2725
2726reallocate:
2727	allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
2728			&fio->new_blkaddr, sum, type, fio, true);
2729
2730	/* writeout dirty page into bdev */
2731	err = f2fs_submit_page_write(fio);
2732	if (err == -EAGAIN) {
2733		fio->old_blkaddr = fio->new_blkaddr;
2734		goto reallocate;
2735	} else if (!err) {
2736		update_device_state(fio);
2737	}
2738}
2739
2740void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2741					enum iostat_type io_type)
2742{
2743	struct f2fs_io_info fio = {
2744		.sbi = sbi,
2745		.type = META,
2746		.temp = HOT,
2747		.op = REQ_OP_WRITE,
2748		.op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
2749		.old_blkaddr = page->index,
2750		.new_blkaddr = page->index,
2751		.page = page,
2752		.encrypted_page = NULL,
2753		.in_list = false,
2754	};
2755
2756	if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
2757		fio.op_flags &= ~REQ_META;
2758
2759	set_page_writeback(page);
2760	f2fs_submit_page_write(&fio);
2761
2762	f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
2763}
2764
2765void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
2766{
2767	struct f2fs_summary sum;
2768
2769	set_summary(&sum, nid, 0, 0);
2770	do_write_page(&sum, fio);
2771
2772	f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2773}
2774
2775void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
2776{
2777	struct f2fs_sb_info *sbi = fio->sbi;
2778	struct f2fs_summary sum;
2779	struct node_info ni;
2780
2781	f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
2782	get_node_info(sbi, dn->nid, &ni);
2783	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
2784	do_write_page(&sum, fio);
2785	f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
2786
2787	f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
2788}
2789
2790int rewrite_data_page(struct f2fs_io_info *fio)
2791{
2792	int err;
2793	struct f2fs_sb_info *sbi = fio->sbi;
2794
2795	fio->new_blkaddr = fio->old_blkaddr;
2796	/* i/o temperature is needed for passing down write hints */
2797	__get_segment_type(fio);
2798
2799	f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
2800			GET_SEGNO(sbi, fio->new_blkaddr))->type));
2801
2802	stat_inc_inplace_blocks(fio->sbi);
2803
2804	err = f2fs_submit_page_bio(fio);
2805	if (!err)
2806		update_device_state(fio);
2807
2808	f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2809
2810	return err;
2811}
2812
2813static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
2814						unsigned int segno)
2815{
2816	int i;
2817
2818	for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
2819		if (CURSEG_I(sbi, i)->segno == segno)
2820			break;
2821	}
2822	return i;
2823}
2824
2825void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
2826				block_t old_blkaddr, block_t new_blkaddr,
2827				bool recover_curseg, bool recover_newaddr)
2828{
2829	struct sit_info *sit_i = SIT_I(sbi);
2830	struct curseg_info *curseg;
2831	unsigned int segno, old_cursegno;
2832	struct seg_entry *se;
2833	int type;
2834	unsigned short old_blkoff;
2835
2836	segno = GET_SEGNO(sbi, new_blkaddr);
2837	se = get_seg_entry(sbi, segno);
2838	type = se->type;
2839
2840	down_write(&SM_I(sbi)->curseg_lock);
2841
2842	if (!recover_curseg) {
2843		/* for recovery flow */
2844		if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2845			if (old_blkaddr == NULL_ADDR)
2846				type = CURSEG_COLD_DATA;
2847			else
2848				type = CURSEG_WARM_DATA;
2849		}
2850	} else {
2851		if (IS_CURSEG(sbi, segno)) {
2852			/* se->type is volatile as SSR allocation */
2853			type = __f2fs_get_curseg(sbi, segno);
2854			f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
2855		} else {
2856			type = CURSEG_WARM_DATA;
2857		}
2858	}
2859
2860	f2fs_bug_on(sbi, !IS_DATASEG(type));
2861	curseg = CURSEG_I(sbi, type);
2862
2863	mutex_lock(&curseg->curseg_mutex);
2864	down_write(&sit_i->sentry_lock);
2865
2866	old_cursegno = curseg->segno;
2867	old_blkoff = curseg->next_blkoff;
2868
2869	/* change the current segment */
2870	if (segno != curseg->segno) {
2871		curseg->next_segno = segno;
2872		change_curseg(sbi, type);
2873	}
2874
2875	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
2876	__add_sum_entry(sbi, type, sum);
2877
2878	if (!recover_curseg || recover_newaddr)
2879		update_sit_entry(sbi, new_blkaddr, 1);
2880	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2881		update_sit_entry(sbi, old_blkaddr, -1);
2882
2883	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2884	locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2885
2886	locate_dirty_segment(sbi, old_cursegno);
2887
2888	if (recover_curseg) {
2889		if (old_cursegno != curseg->segno) {
2890			curseg->next_segno = old_cursegno;
2891			change_curseg(sbi, type);
2892		}
2893		curseg->next_blkoff = old_blkoff;
2894	}
2895
2896	up_write(&sit_i->sentry_lock);
2897	mutex_unlock(&curseg->curseg_mutex);
2898	up_write(&SM_I(sbi)->curseg_lock);
2899}
2900
2901void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2902				block_t old_addr, block_t new_addr,
2903				unsigned char version, bool recover_curseg,
2904				bool recover_newaddr)
2905{
2906	struct f2fs_summary sum;
2907
2908	set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2909
2910	__f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2911					recover_curseg, recover_newaddr);
2912
2913	f2fs_update_data_blkaddr(dn, new_addr);
2914}
2915
2916void f2fs_wait_on_page_writeback(struct page *page,
2917				enum page_type type, bool ordered)
2918{
2919	if (PageWriteback(page)) {
2920		struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2921
2922		f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2923						0, page->index, type);
2924		if (ordered)
2925			wait_on_page_writeback(page);
2926		else
2927			wait_for_stable_page(page);
2928	}
2929}
2930
2931void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
2932{
2933	struct page *cpage;
2934
2935	if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
2936		return;
2937
2938	cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2939	if (cpage) {
2940		f2fs_wait_on_page_writeback(cpage, DATA, true);
2941		f2fs_put_page(cpage, 1);
2942	}
2943}
2944
2945static void read_compacted_summaries(struct f2fs_sb_info *sbi)
2946{
2947	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2948	struct curseg_info *seg_i;
2949	unsigned char *kaddr;
2950	struct page *page;
2951	block_t start;
2952	int i, j, offset;
2953
2954	start = start_sum_block(sbi);
2955
2956	page = get_meta_page(sbi, start++);
2957	kaddr = (unsigned char *)page_address(page);
2958
2959	/* Step 1: restore nat cache */
2960	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2961	memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
2962
2963	/* Step 2: restore sit cache */
2964	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2965	memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
2966	offset = 2 * SUM_JOURNAL_SIZE;
2967
2968	/* Step 3: restore summary entries */
2969	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2970		unsigned short blk_off;
2971		unsigned int segno;
2972
2973		seg_i = CURSEG_I(sbi, i);
2974		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2975		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2976		seg_i->next_segno = segno;
2977		reset_curseg(sbi, i, 0);
2978		seg_i->alloc_type = ckpt->alloc_type[i];
2979		seg_i->next_blkoff = blk_off;
2980
2981		if (seg_i->alloc_type == SSR)
2982			blk_off = sbi->blocks_per_seg;
2983
2984		for (j = 0; j < blk_off; j++) {
2985			struct f2fs_summary *s;
2986			s = (struct f2fs_summary *)(kaddr + offset);
2987			seg_i->sum_blk->entries[j] = *s;
2988			offset += SUMMARY_SIZE;
2989			if (offset + SUMMARY_SIZE <= PAGE_SIZE -
2990						SUM_FOOTER_SIZE)
2991				continue;
2992
2993			f2fs_put_page(page, 1);
2994			page = NULL;
2995
2996			page = get_meta_page(sbi, start++);
2997			kaddr = (unsigned char *)page_address(page);
2998			offset = 0;
2999		}
3000	}
3001	f2fs_put_page(page, 1);
3002}
3003
3004static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3005{
3006	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3007	struct f2fs_summary_block *sum;
3008	struct curseg_info *curseg;
3009	struct page *new;
3010	unsigned short blk_off;
3011	unsigned int segno = 0;
3012	block_t blk_addr = 0;
3013
3014	/* get segment number and block addr */
3015	if (IS_DATASEG(type)) {
3016		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3017		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3018							CURSEG_HOT_DATA]);
3019		if (__exist_node_summaries(sbi))
3020			blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3021		else
3022			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3023	} else {
3024		segno = le32_to_cpu(ckpt->cur_node_segno[type -
3025							CURSEG_HOT_NODE]);
3026		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3027							CURSEG_HOT_NODE]);
3028		if (__exist_node_summaries(sbi))
3029			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3030							type - CURSEG_HOT_NODE);
3031		else
3032			blk_addr = GET_SUM_BLOCK(sbi, segno);
3033	}
3034
3035	new = get_meta_page(sbi, blk_addr);
3036	sum = (struct f2fs_summary_block *)page_address(new);
3037
3038	if (IS_NODESEG(type)) {
3039		if (__exist_node_summaries(sbi)) {
3040			struct f2fs_summary *ns = &sum->entries[0];
3041			int i;
3042			for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3043				ns->version = 0;
3044				ns->ofs_in_node = 0;
3045			}
3046		} else {
3047			restore_node_summary(sbi, segno, sum);
3048		}
3049	}
3050
3051	/* set uncompleted segment to curseg */
3052	curseg = CURSEG_I(sbi, type);
3053	mutex_lock(&curseg->curseg_mutex);
3054
3055	/* update journal info */
3056	down_write(&curseg->journal_rwsem);
3057	memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3058	up_write(&curseg->journal_rwsem);
3059
3060	memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3061	memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3062	curseg->next_segno = segno;
3063	reset_curseg(sbi, type, 0);
3064	curseg->alloc_type = ckpt->alloc_type[type];
3065	curseg->next_blkoff = blk_off;
3066	mutex_unlock(&curseg->curseg_mutex);
3067	f2fs_put_page(new, 1);
3068	return 0;
3069}
3070
3071static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3072{
3073	struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3074	struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3075	int type = CURSEG_HOT_DATA;
3076	int err;
3077
3078	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3079		int npages = npages_for_summary_flush(sbi, true);
3080
3081		if (npages >= 2)
3082			ra_meta_pages(sbi, start_sum_block(sbi), npages,
3083							META_CP, true);
3084
3085		/* restore for compacted data summary */
3086		read_compacted_summaries(sbi);
3087		type = CURSEG_HOT_NODE;
3088	}
3089
3090	if (__exist_node_summaries(sbi))
3091		ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3092					NR_CURSEG_TYPE - type, META_CP, true);
3093
3094	for (; type <= CURSEG_COLD_NODE; type++) {
3095		err = read_normal_summaries(sbi, type);
3096		if (err)
3097			return err;
3098	}
3099
3100	/* sanity check for summary blocks */
3101	if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3102			sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3103		return -EINVAL;
3104
3105	return 0;
3106}
3107
3108static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3109{
3110	struct page *page;
3111	unsigned char *kaddr;
3112	struct f2fs_summary *summary;
3113	struct curseg_info *seg_i;
3114	int written_size = 0;
3115	int i, j;
3116
3117	page = grab_meta_page(sbi, blkaddr++);
3118	kaddr = (unsigned char *)page_address(page);
3119
3120	/* Step 1: write nat cache */
3121	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3122	memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3123	written_size += SUM_JOURNAL_SIZE;
3124
3125	/* Step 2: write sit cache */
3126	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3127	memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3128	written_size += SUM_JOURNAL_SIZE;
3129
3130	/* Step 3: write summary entries */
3131	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3132		unsigned short blkoff;
3133		seg_i = CURSEG_I(sbi, i);
3134		if (sbi->ckpt->alloc_type[i] == SSR)
3135			blkoff = sbi->blocks_per_seg;
3136		else
3137			blkoff = curseg_blkoff(sbi, i);
3138
3139		for (j = 0; j < blkoff; j++) {
3140			if (!page) {
3141				page = grab_meta_page(sbi, blkaddr++);
3142				kaddr = (unsigned char *)page_address(page);
3143				written_size = 0;
3144			}
3145			summary = (struct f2fs_summary *)(kaddr + written_size);
3146			*summary = seg_i->sum_blk->entries[j];
3147			written_size += SUMMARY_SIZE;
3148
3149			if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3150							SUM_FOOTER_SIZE)
3151				continue;
3152
3153			set_page_dirty(page);
3154			f2fs_put_page(page, 1);
3155			page = NULL;
3156		}
3157	}
3158	if (page) {
3159		set_page_dirty(page);
3160		f2fs_put_page(page, 1);
3161	}
3162}
3163
3164static void write_normal_summaries(struct f2fs_sb_info *sbi,
3165					block_t blkaddr, int type)
3166{
3167	int i, end;
3168	if (IS_DATASEG(type))
3169		end = type + NR_CURSEG_DATA_TYPE;
3170	else
3171		end = type + NR_CURSEG_NODE_TYPE;
3172
3173	for (i = type; i < end; i++)
3174		write_current_sum_page(sbi, i, blkaddr + (i - type));
3175}
3176
3177void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3178{
3179	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3180		write_compacted_summaries(sbi, start_blk);
3181	else
3182		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3183}
3184
3185void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3186{
3187	write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3188}
3189
3190int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3191					unsigned int val, int alloc)
3192{
3193	int i;
3194
3195	if (type == NAT_JOURNAL) {
3196		for (i = 0; i < nats_in_cursum(journal); i++) {
3197			if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3198				return i;
3199		}
3200		if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3201			return update_nats_in_cursum(journal, 1);
3202	} else if (type == SIT_JOURNAL) {
3203		for (i = 0; i < sits_in_cursum(journal); i++)
3204			if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3205				return i;
3206		if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3207			return update_sits_in_cursum(journal, 1);
3208	}
3209	return -1;
3210}
3211
3212static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3213					unsigned int segno)
3214{
3215	return get_meta_page(sbi, current_sit_addr(sbi, segno));
3216}
3217
3218static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3219					unsigned int start)
3220{
3221	struct sit_info *sit_i = SIT_I(sbi);
3222	struct page *page;
3223	pgoff_t src_off, dst_off;
3224
3225	src_off = current_sit_addr(sbi, start);
3226	dst_off = next_sit_addr(sbi, src_off);
3227
3228	page = grab_meta_page(sbi, dst_off);
3229	seg_info_to_sit_page(sbi, page, start);
3230
3231	set_page_dirty(page);
3232	set_to_next_sit(sit_i, start);
3233
3234	return page;
3235}
3236
3237static struct sit_entry_set *grab_sit_entry_set(void)
3238{
3239	struct sit_entry_set *ses =
3240			f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3241
3242	ses->entry_cnt = 0;
3243	INIT_LIST_HEAD(&ses->set_list);
3244	return ses;
3245}
3246
3247static void release_sit_entry_set(struct sit_entry_set *ses)
3248{
3249	list_del(&ses->set_list);
3250	kmem_cache_free(sit_entry_set_slab, ses);
3251}
3252
3253static void adjust_sit_entry_set(struct sit_entry_set *ses,
3254						struct list_head *head)
3255{
3256	struct sit_entry_set *next = ses;
3257
3258	if (list_is_last(&ses->set_list, head))
3259		return;
3260
3261	list_for_each_entry_continue(next, head, set_list)
3262		if (ses->entry_cnt <= next->entry_cnt)
3263			break;
3264
3265	list_move_tail(&ses->set_list, &next->set_list);
3266}
3267
3268static void add_sit_entry(unsigned int segno, struct list_head *head)
3269{
3270	struct sit_entry_set *ses;
3271	unsigned int start_segno = START_SEGNO(segno);
3272
3273	list_for_each_entry(ses, head, set_list) {
3274		if (ses->start_segno == start_segno) {
3275			ses->entry_cnt++;
3276			adjust_sit_entry_set(ses, head);
3277			return;
3278		}
3279	}
3280
3281	ses = grab_sit_entry_set();
3282
3283	ses->start_segno = start_segno;
3284	ses->entry_cnt++;
3285	list_add(&ses->set_list, head);
3286}
3287
3288static void add_sits_in_set(struct f2fs_sb_info *sbi)
3289{
3290	struct f2fs_sm_info *sm_info = SM_I(sbi);
3291	struct list_head *set_list = &sm_info->sit_entry_set;
3292	unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3293	unsigned int segno;
3294
3295	for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3296		add_sit_entry(segno, set_list);
3297}
3298
3299static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3300{
3301	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3302	struct f2fs_journal *journal = curseg->journal;
3303	int i;
3304
3305	down_write(&curseg->journal_rwsem);
3306	for (i = 0; i < sits_in_cursum(journal); i++) {
3307		unsigned int segno;
3308		bool dirtied;
3309
3310		segno = le32_to_cpu(segno_in_journal(journal, i));
3311		dirtied = __mark_sit_entry_dirty(sbi, segno);
3312
3313		if (!dirtied)
3314			add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3315	}
3316	update_sits_in_cursum(journal, -i);
3317	up_write(&curseg->journal_rwsem);
3318}
3319
3320/*
3321 * CP calls this function, which flushes SIT entries including sit_journal,
3322 * and moves prefree segs to free segs.
3323 */
3324void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3325{
3326	struct sit_info *sit_i = SIT_I(sbi);
3327	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3328	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3329	struct f2fs_journal *journal = curseg->journal;
3330	struct sit_entry_set *ses, *tmp;
3331	struct list_head *head = &SM_I(sbi)->sit_entry_set;
3332	bool to_journal = true;
3333	struct seg_entry *se;
3334
3335	down_write(&sit_i->sentry_lock);
3336
3337	if (!sit_i->dirty_sentries)
3338		goto out;
3339
3340	/*
3341	 * add and account sit entries of dirty bitmap in sit entry
3342	 * set temporarily
3343	 */
3344	add_sits_in_set(sbi);
3345
3346	/*
3347	 * if there are no enough space in journal to store dirty sit
3348	 * entries, remove all entries from journal and add and account
3349	 * them in sit entry set.
3350	 */
3351	if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3352		remove_sits_in_journal(sbi);
3353
3354	/*
3355	 * there are two steps to flush sit entries:
3356	 * #1, flush sit entries to journal in current cold data summary block.
3357	 * #2, flush sit entries to sit page.
3358	 */
3359	list_for_each_entry_safe(ses, tmp, head, set_list) {
3360		struct page *page = NULL;
3361		struct f2fs_sit_block *raw_sit = NULL;
3362		unsigned int start_segno = ses->start_segno;
3363		unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3364						(unsigned long)MAIN_SEGS(sbi));
3365		unsigned int segno = start_segno;
3366
3367		if (to_journal &&
3368			!__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3369			to_journal = false;
3370
3371		if (to_journal) {
3372			down_write(&curseg->journal_rwsem);
3373		} else {
3374			page = get_next_sit_page(sbi, start_segno);
3375			raw_sit = page_address(page);
3376		}
3377
3378		/* flush dirty sit entries in region of current sit set */
3379		for_each_set_bit_from(segno, bitmap, end) {
3380			int offset, sit_offset;
3381
3382			se = get_seg_entry(sbi, segno);
3383
3384			/* add discard candidates */
3385			if (!(cpc->reason & CP_DISCARD)) {
3386				cpc->trim_start = segno;
3387				add_discard_addrs(sbi, cpc, false);
3388			}
3389
3390			if (to_journal) {
3391				offset = lookup_journal_in_cursum(journal,
3392							SIT_JOURNAL, segno, 1);
3393				f2fs_bug_on(sbi, offset < 0);
3394				segno_in_journal(journal, offset) =
3395							cpu_to_le32(segno);
3396				seg_info_to_raw_sit(se,
3397					&sit_in_journal(journal, offset));
3398			} else {
3399				sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3400				seg_info_to_raw_sit(se,
3401						&raw_sit->entries[sit_offset]);
3402			}
3403
3404			__clear_bit(segno, bitmap);
3405			sit_i->dirty_sentries--;
3406			ses->entry_cnt--;
3407		}
3408
3409		if (to_journal)
3410			up_write(&curseg->journal_rwsem);
3411		else
3412			f2fs_put_page(page, 1);
3413
3414		f2fs_bug_on(sbi, ses->entry_cnt);
3415		release_sit_entry_set(ses);
3416	}
3417
3418	f2fs_bug_on(sbi, !list_empty(head));
3419	f2fs_bug_on(sbi, sit_i->dirty_sentries);
3420out:
3421	if (cpc->reason & CP_DISCARD) {
3422		__u64 trim_start = cpc->trim_start;
3423
3424		for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3425			add_discard_addrs(sbi, cpc, false);
3426
3427		cpc->trim_start = trim_start;
3428	}
3429	up_write(&sit_i->sentry_lock);
3430
3431	set_prefree_as_free_segments(sbi);
3432}
3433
3434static int build_sit_info(struct f2fs_sb_info *sbi)
3435{
3436	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3437	struct sit_info *sit_i;
3438	unsigned int sit_segs, start;
3439	char *src_bitmap;
3440	unsigned int bitmap_size;
3441
3442	/* allocate memory for SIT information */
3443	sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3444	if (!sit_i)
3445		return -ENOMEM;
3446
3447	SM_I(sbi)->sit_info = sit_i;
3448
3449	sit_i->sentries = f2fs_kvzalloc(sbi, MAIN_SEGS(sbi) *
3450					sizeof(struct seg_entry), GFP_KERNEL);
3451	if (!sit_i->sentries)
3452		return -ENOMEM;
3453
3454	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3455	sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3456								GFP_KERNEL);
3457	if (!sit_i->dirty_sentries_bitmap)
3458		return -ENOMEM;
3459
3460	for (start = 0; start < MAIN_SEGS(sbi); start++) {
3461		sit_i->sentries[start].cur_valid_map
3462			= f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3463		sit_i->sentries[start].ckpt_valid_map
3464			= f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3465		if (!sit_i->sentries[start].cur_valid_map ||
3466				!sit_i->sentries[start].ckpt_valid_map)
3467			return -ENOMEM;
3468
3469#ifdef CONFIG_F2FS_CHECK_FS
3470		sit_i->sentries[start].cur_valid_map_mir
3471			= f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3472		if (!sit_i->sentries[start].cur_valid_map_mir)
3473			return -ENOMEM;
3474#endif
3475
3476		if (f2fs_discard_en(sbi)) {
3477			sit_i->sentries[start].discard_map
3478				= f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3479								GFP_KERNEL);
3480			if (!sit_i->sentries[start].discard_map)
3481				return -ENOMEM;
3482		}
3483	}
3484
3485	sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3486	if (!sit_i->tmp_map)
3487		return -ENOMEM;
3488
3489	if (sbi->segs_per_sec > 1) {
3490		sit_i->sec_entries = f2fs_kvzalloc(sbi, MAIN_SECS(sbi) *
3491					sizeof(struct sec_entry), GFP_KERNEL);
3492		if (!sit_i->sec_entries)
3493			return -ENOMEM;
3494	}
3495
3496	/* get information related with SIT */
3497	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3498
3499	/* setup SIT bitmap from ckeckpoint pack */
3500	bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3501	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3502
3503	sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3504	if (!sit_i->sit_bitmap)
3505		return -ENOMEM;
3506
3507#ifdef CONFIG_F2FS_CHECK_FS
3508	sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3509	if (!sit_i->sit_bitmap_mir)
3510		return -ENOMEM;
3511#endif
3512
3513	/* init SIT information */
3514	sit_i->s_ops = &default_salloc_ops;
3515
3516	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3517	sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
3518	sit_i->written_valid_blocks = 0;
3519	sit_i->bitmap_size = bitmap_size;
3520	sit_i->dirty_sentries = 0;
3521	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3522	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3523	sit_i->mounted_time = ktime_get_real_seconds();
3524	init_rwsem(&sit_i->sentry_lock);
3525	return 0;
3526}
3527
3528static int build_free_segmap(struct f2fs_sb_info *sbi)
3529{
3530	struct free_segmap_info *free_i;
3531	unsigned int bitmap_size, sec_bitmap_size;
3532
3533	/* allocate memory for free segmap information */
3534	free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
3535	if (!free_i)
3536		return -ENOMEM;
3537
3538	SM_I(sbi)->free_info = free_i;
3539
3540	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3541	free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
3542	if (!free_i->free_segmap)
3543		return -ENOMEM;
3544
3545	sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3546	free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
3547	if (!free_i->free_secmap)
3548		return -ENOMEM;
3549
3550	/* set all segments as dirty temporarily */
3551	memset(free_i->free_segmap, 0xff, bitmap_size);
3552	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3553
3554	/* init free segmap information */
3555	free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
3556	free_i->free_segments = 0;
3557	free_i->free_sections = 0;
3558	spin_lock_init(&free_i->segmap_lock);
3559	return 0;
3560}
3561
3562static int build_curseg(struct f2fs_sb_info *sbi)
3563{
3564	struct curseg_info *array;
3565	int i;
3566
3567	array = f2fs_kzalloc(sbi, sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
3568	if (!array)
3569		return -ENOMEM;
3570
3571	SM_I(sbi)->curseg_array = array;
3572
3573	for (i = 0; i < NR_CURSEG_TYPE; i++) {
3574		mutex_init(&array[i].curseg_mutex);
3575		array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
3576		if (!array[i].sum_blk)
3577			return -ENOMEM;
3578		init_rwsem(&array[i].journal_rwsem);
3579		array[i].journal = f2fs_kzalloc(sbi,
3580				sizeof(struct f2fs_journal), GFP_KERNEL);
3581		if (!array[i].journal)
3582			return -ENOMEM;
3583		array[i].segno = NULL_SEGNO;
3584		array[i].next_blkoff = 0;
3585	}
3586	return restore_curseg_summaries(sbi);
3587}
3588
3589static int build_sit_entries(struct f2fs_sb_info *sbi)
3590{
3591	struct sit_info *sit_i = SIT_I(sbi);
3592	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3593	struct f2fs_journal *journal = curseg->journal;
3594	struct seg_entry *se;
3595	struct f2fs_sit_entry sit;
3596	int sit_blk_cnt = SIT_BLK_CNT(sbi);
3597	unsigned int i, start, end;
3598	unsigned int readed, start_blk = 0;
3599	int err = 0;
3600
3601	do {
3602		readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3603							META_SIT, true);
3604
3605		start = start_blk * sit_i->sents_per_block;
3606		end = (start_blk + readed) * sit_i->sents_per_block;
3607
3608		for (; start < end && start < MAIN_SEGS(sbi); start++) {
3609			struct f2fs_sit_block *sit_blk;
3610			struct page *page;
3611
3612			se = &sit_i->sentries[start];
3613			page = get_current_sit_page(sbi, start);
3614			sit_blk = (struct f2fs_sit_block *)page_address(page);
3615			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3616			f2fs_put_page(page, 1);
3617
3618			err = check_block_count(sbi, start, &sit);
3619			if (err)
3620				return err;
3621			seg_info_from_raw_sit(se, &sit);
3622
3623			/* build discard map only one time */
3624			if (f2fs_discard_en(sbi)) {
3625				if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3626					memset(se->discard_map, 0xff,
3627						SIT_VBLOCK_MAP_SIZE);
3628				} else {
3629					memcpy(se->discard_map,
3630						se->cur_valid_map,
3631						SIT_VBLOCK_MAP_SIZE);
3632					sbi->discard_blks +=
3633						sbi->blocks_per_seg -
3634						se->valid_blocks;
3635				}
3636			}
3637
3638			if (sbi->segs_per_sec > 1)
3639				get_sec_entry(sbi, start)->valid_blocks +=
3640							se->valid_blocks;
3641		}
3642		start_blk += readed;
3643	} while (start_blk < sit_blk_cnt);
3644
3645	down_read(&curseg->journal_rwsem);
3646	for (i = 0; i < sits_in_cursum(journal); i++) {
3647		unsigned int old_valid_blocks;
3648
3649		start = le32_to_cpu(segno_in_journal(journal, i));
3650		se = &sit_i->sentries[start];
3651		sit = sit_in_journal(journal, i);
3652
3653		old_valid_blocks = se->valid_blocks;
3654
3655		err = check_block_count(sbi, start, &sit);
3656		if (err)
3657			break;
3658		seg_info_from_raw_sit(se, &sit);
3659
3660		if (f2fs_discard_en(sbi)) {
3661			if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3662				memset(se->discard_map, 0xff,
3663							SIT_VBLOCK_MAP_SIZE);
3664			} else {
3665				memcpy(se->discard_map, se->cur_valid_map,
3666							SIT_VBLOCK_MAP_SIZE);
3667				sbi->discard_blks += old_valid_blocks -
3668							se->valid_blocks;
3669			}
3670		}
3671
3672		if (sbi->segs_per_sec > 1)
3673			get_sec_entry(sbi, start)->valid_blocks +=
3674				se->valid_blocks - old_valid_blocks;
3675	}
3676	up_read(&curseg->journal_rwsem);
3677	return err;
3678}
3679
3680static void init_free_segmap(struct f2fs_sb_info *sbi)
3681{
3682	unsigned int start;
3683	int type;
3684
3685	for (start = 0; start < MAIN_SEGS(sbi); start++) {
3686		struct seg_entry *sentry = get_seg_entry(sbi, start);
3687		if (!sentry->valid_blocks)
3688			__set_free(sbi, start);
3689		else
3690			SIT_I(sbi)->written_valid_blocks +=
3691						sentry->valid_blocks;
3692	}
3693
3694	/* set use the current segments */
3695	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3696		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3697		__set_test_and_inuse(sbi, curseg_t->segno);
3698	}
3699}
3700
3701static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3702{
3703	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3704	struct free_segmap_info *free_i = FREE_I(sbi);
3705	unsigned int segno = 0, offset = 0;
3706	unsigned short valid_blocks;
3707
3708	while (1) {
3709		/* find dirty segment based on free segmap */
3710		segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3711		if (segno >= MAIN_SEGS(sbi))
3712			break;
3713		offset = segno + 1;
3714		valid_blocks = get_valid_blocks(sbi, segno, false);
3715		if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
3716			continue;
3717		if (valid_blocks > sbi->blocks_per_seg) {
3718			f2fs_bug_on(sbi, 1);
3719			continue;
3720		}
3721		mutex_lock(&dirty_i->seglist_lock);
3722		__locate_dirty_segment(sbi, segno, DIRTY);
3723		mutex_unlock(&dirty_i->seglist_lock);
3724	}
3725}
3726
3727static int init_victim_secmap(struct f2fs_sb_info *sbi)
3728{
3729	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3730	unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3731
3732	dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
3733	if (!dirty_i->victim_secmap)
3734		return -ENOMEM;
3735	return 0;
3736}
3737
3738static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3739{
3740	struct dirty_seglist_info *dirty_i;
3741	unsigned int bitmap_size, i;
3742
3743	/* allocate memory for dirty segments list information */
3744	dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
3745								GFP_KERNEL);
3746	if (!dirty_i)
3747		return -ENOMEM;
3748
3749	SM_I(sbi)->dirty_info = dirty_i;
3750	mutex_init(&dirty_i->seglist_lock);
3751
3752	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3753
3754	for (i = 0; i < NR_DIRTY_TYPE; i++) {
3755		dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
3756								GFP_KERNEL);
3757		if (!dirty_i->dirty_segmap[i])
3758			return -ENOMEM;
3759	}
3760
3761	init_dirty_segmap(sbi);
3762	return init_victim_secmap(sbi);
3763}
3764
3765/*
3766 * Update min, max modified time for cost-benefit GC algorithm
3767 */
3768static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3769{
3770	struct sit_info *sit_i = SIT_I(sbi);
3771	unsigned int segno;
3772
3773	down_write(&sit_i->sentry_lock);
3774
3775	sit_i->min_mtime = LLONG_MAX;
3776
3777	for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
3778		unsigned int i;
3779		unsigned long long mtime = 0;
3780
3781		for (i = 0; i < sbi->segs_per_sec; i++)
3782			mtime += get_seg_entry(sbi, segno + i)->mtime;
3783
3784		mtime = div_u64(mtime, sbi->segs_per_sec);
3785
3786		if (sit_i->min_mtime > mtime)
3787			sit_i->min_mtime = mtime;
3788	}
3789	sit_i->max_mtime = get_mtime(sbi);
3790	up_write(&sit_i->sentry_lock);
3791}
3792
3793int build_segment_manager(struct f2fs_sb_info *sbi)
3794{
3795	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3796	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3797	struct f2fs_sm_info *sm_info;
3798	int err;
3799
3800	sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
3801	if (!sm_info)
3802		return -ENOMEM;
3803
3804	/* init sm info */
3805	sbi->sm_info = sm_info;
3806	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3807	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3808	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3809	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3810	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3811	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3812	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
3813	sm_info->rec_prefree_segments = sm_info->main_segments *
3814					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
3815	if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3816		sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3817
3818	if (!test_opt(sbi, LFS))
3819		sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
3820	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
3821	sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
3822	sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
3823	sm_info->min_ssr_sections = reserved_sections(sbi);
3824
3825	sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
3826
3827	INIT_LIST_HEAD(&sm_info->sit_entry_set);
3828
3829	init_rwsem(&sm_info->curseg_lock);
3830
3831	if (!f2fs_readonly(sbi->sb)) {
3832		err = create_flush_cmd_control(sbi);
3833		if (err)
3834			return err;
3835	}
3836
3837	err = create_discard_cmd_control(sbi);
3838	if (err)
3839		return err;
3840
3841	err = build_sit_info(sbi);
3842	if (err)
3843		return err;
3844	err = build_free_segmap(sbi);
3845	if (err)
3846		return err;
3847	err = build_curseg(sbi);
3848	if (err)
3849		return err;
3850
3851	/* reinit free segmap based on SIT */
3852	err = build_sit_entries(sbi);
3853	if (err)
3854		return err;
3855
3856	init_free_segmap(sbi);
3857	err = build_dirty_segmap(sbi);
3858	if (err)
3859		return err;
3860
3861	init_min_max_mtime(sbi);
3862	return 0;
3863}
3864
3865static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3866		enum dirty_type dirty_type)
3867{
3868	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3869
3870	mutex_lock(&dirty_i->seglist_lock);
3871	kvfree(dirty_i->dirty_segmap[dirty_type]);
3872	dirty_i->nr_dirty[dirty_type] = 0;
3873	mutex_unlock(&dirty_i->seglist_lock);
3874}
3875
3876static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
3877{
3878	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3879	kvfree(dirty_i->victim_secmap);
3880}
3881
3882static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3883{
3884	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3885	int i;
3886
3887	if (!dirty_i)
3888		return;
3889
3890	/* discard pre-free/dirty segments list */
3891	for (i = 0; i < NR_DIRTY_TYPE; i++)
3892		discard_dirty_segmap(sbi, i);
3893
3894	destroy_victim_secmap(sbi);
3895	SM_I(sbi)->dirty_info = NULL;
3896	kfree(dirty_i);
3897}
3898
3899static void destroy_curseg(struct f2fs_sb_info *sbi)
3900{
3901	struct curseg_info *array = SM_I(sbi)->curseg_array;
3902	int i;
3903
3904	if (!array)
3905		return;
3906	SM_I(sbi)->curseg_array = NULL;
3907	for (i = 0; i < NR_CURSEG_TYPE; i++) {
3908		kfree(array[i].sum_blk);
3909		kfree(array[i].journal);
3910	}
3911	kfree(array);
3912}
3913
3914static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3915{
3916	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3917	if (!free_i)
3918		return;
3919	SM_I(sbi)->free_info = NULL;
3920	kvfree(free_i->free_segmap);
3921	kvfree(free_i->free_secmap);
3922	kfree(free_i);
3923}
3924
3925static void destroy_sit_info(struct f2fs_sb_info *sbi)
3926{
3927	struct sit_info *sit_i = SIT_I(sbi);
3928	unsigned int start;
3929
3930	if (!sit_i)
3931		return;
3932
3933	if (sit_i->sentries) {
3934		for (start = 0; start < MAIN_SEGS(sbi); start++) {
3935			kfree(sit_i->sentries[start].cur_valid_map);
3936#ifdef CONFIG_F2FS_CHECK_FS
3937			kfree(sit_i->sentries[start].cur_valid_map_mir);
3938#endif
3939			kfree(sit_i->sentries[start].ckpt_valid_map);
3940			kfree(sit_i->sentries[start].discard_map);
3941		}
3942	}
3943	kfree(sit_i->tmp_map);
3944
3945	kvfree(sit_i->sentries);
3946	kvfree(sit_i->sec_entries);
3947	kvfree(sit_i->dirty_sentries_bitmap);
3948
3949	SM_I(sbi)->sit_info = NULL;
3950	kfree(sit_i->sit_bitmap);
3951#ifdef CONFIG_F2FS_CHECK_FS
3952	kfree(sit_i->sit_bitmap_mir);
3953#endif
3954	kfree(sit_i);
3955}
3956
3957void destroy_segment_manager(struct f2fs_sb_info *sbi)
3958{
3959	struct f2fs_sm_info *sm_info = SM_I(sbi);
3960
3961	if (!sm_info)
3962		return;
3963	destroy_flush_cmd_control(sbi, true);
3964	destroy_discard_cmd_control(sbi);
3965	destroy_dirty_segmap(sbi);
3966	destroy_curseg(sbi);
3967	destroy_free_segmap(sbi);
3968	destroy_sit_info(sbi);
3969	sbi->sm_info = NULL;
3970	kfree(sm_info);
3971}
3972
3973int __init create_segment_manager_caches(void)
3974{
3975	discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
3976			sizeof(struct discard_entry));
3977	if (!discard_entry_slab)
3978		goto fail;
3979
3980	discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3981			sizeof(struct discard_cmd));
3982	if (!discard_cmd_slab)
3983		goto destroy_discard_entry;
3984
3985	sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
3986			sizeof(struct sit_entry_set));
3987	if (!sit_entry_set_slab)
3988		goto destroy_discard_cmd;
3989
3990	inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3991			sizeof(struct inmem_pages));
3992	if (!inmem_entry_slab)
3993		goto destroy_sit_entry_set;
3994	return 0;
3995
3996destroy_sit_entry_set:
3997	kmem_cache_destroy(sit_entry_set_slab);
3998destroy_discard_cmd:
3999	kmem_cache_destroy(discard_cmd_slab);
4000destroy_discard_entry:
4001	kmem_cache_destroy(discard_entry_slab);
4002fail:
4003	return -ENOMEM;
4004}
4005
4006void destroy_segment_manager_caches(void)
4007{
4008	kmem_cache_destroy(sit_entry_set_slab);
4009	kmem_cache_destroy(discard_cmd_slab);
4010	kmem_cache_destroy(discard_entry_slab);
4011	kmem_cache_destroy(inmem_entry_slab);
4012}