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
  20#include "f2fs.h"
  21#include "segment.h"
  22#include "node.h"
  23#include "trace.h"
  24#include <trace/events/f2fs.h>
  25
  26#define __reverse_ffz(x) __reverse_ffs(~(x))
  27
  28static struct kmem_cache *discard_entry_slab;
  29static struct kmem_cache *bio_entry_slab;
  30static struct kmem_cache *sit_entry_set_slab;
  31static struct kmem_cache *inmem_entry_slab;
  32
  33static unsigned long __reverse_ulong(unsigned char *str)
  34{
  35	unsigned long tmp = 0;
  36	int shift = 24, idx = 0;
  37
  38#if BITS_PER_LONG == 64
  39	shift = 56;
  40#endif
  41	while (shift >= 0) {
  42		tmp |= (unsigned long)str[idx++] << shift;
  43		shift -= BITS_PER_BYTE;
  44	}
  45	return tmp;
  46}
  47
  48/*
  49 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
  50 * MSB and LSB are reversed in a byte by f2fs_set_bit.
  51 */
  52static inline unsigned long __reverse_ffs(unsigned long word)
  53{
  54	int num = 0;
  55
  56#if BITS_PER_LONG == 64
  57	if ((word & 0xffffffff00000000UL) == 0)
  58		num += 32;
  59	else
  60		word >>= 32;
  61#endif
  62	if ((word & 0xffff0000) == 0)
  63		num += 16;
  64	else
  65		word >>= 16;
  66
  67	if ((word & 0xff00) == 0)
  68		num += 8;
  69	else
  70		word >>= 8;
  71
  72	if ((word & 0xf0) == 0)
  73		num += 4;
  74	else
  75		word >>= 4;
  76
  77	if ((word & 0xc) == 0)
  78		num += 2;
  79	else
  80		word >>= 2;
  81
  82	if ((word & 0x2) == 0)
  83		num += 1;
  84	return num;
  85}
  86
  87/*
  88 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
  89 * f2fs_set_bit makes MSB and LSB reversed in a byte.
  90 * @size must be integral times of unsigned long.
  91 * Example:
  92 *                             MSB <--> LSB
  93 *   f2fs_set_bit(0, bitmap) => 1000 0000
  94 *   f2fs_set_bit(7, bitmap) => 0000 0001
  95 */
  96static unsigned long __find_rev_next_bit(const unsigned long *addr,
  97			unsigned long size, unsigned long offset)
  98{
  99	const unsigned long *p = addr + BIT_WORD(offset);
 100	unsigned long result = size;
 101	unsigned long tmp;
 102
 103	if (offset >= size)
 104		return size;
 105
 106	size -= (offset & ~(BITS_PER_LONG - 1));
 107	offset %= BITS_PER_LONG;
 108
 109	while (1) {
 110		if (*p == 0)
 111			goto pass;
 112
 113		tmp = __reverse_ulong((unsigned char *)p);
 114
 115		tmp &= ~0UL >> offset;
 116		if (size < BITS_PER_LONG)
 117			tmp &= (~0UL << (BITS_PER_LONG - size));
 118		if (tmp)
 119			goto found;
 120pass:
 121		if (size <= BITS_PER_LONG)
 122			break;
 123		size -= BITS_PER_LONG;
 124		offset = 0;
 125		p++;
 126	}
 127	return result;
 128found:
 129	return result - size + __reverse_ffs(tmp);
 130}
 131
 132static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
 133			unsigned long size, unsigned long offset)
 134{
 135	const unsigned long *p = addr + BIT_WORD(offset);
 136	unsigned long result = size;
 137	unsigned long tmp;
 138
 139	if (offset >= size)
 140		return size;
 141
 142	size -= (offset & ~(BITS_PER_LONG - 1));
 143	offset %= BITS_PER_LONG;
 144
 145	while (1) {
 146		if (*p == ~0UL)
 147			goto pass;
 148
 149		tmp = __reverse_ulong((unsigned char *)p);
 150
 151		if (offset)
 152			tmp |= ~0UL << (BITS_PER_LONG - offset);
 153		if (size < BITS_PER_LONG)
 154			tmp |= ~0UL >> size;
 155		if (tmp != ~0UL)
 156			goto found;
 157pass:
 158		if (size <= BITS_PER_LONG)
 159			break;
 160		size -= BITS_PER_LONG;
 161		offset = 0;
 162		p++;
 163	}
 164	return result;
 165found:
 166	return result - size + __reverse_ffz(tmp);
 167}
 168
 169void register_inmem_page(struct inode *inode, struct page *page)
 170{
 171	struct f2fs_inode_info *fi = F2FS_I(inode);
 172	struct inmem_pages *new;
 173
 174	f2fs_trace_pid(page);
 175
 176	set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
 177	SetPagePrivate(page);
 178
 179	new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
 180
 181	/* add atomic page indices to the list */
 182	new->page = page;
 183	INIT_LIST_HEAD(&new->list);
 184
 185	/* increase reference count with clean state */
 186	mutex_lock(&fi->inmem_lock);
 187	get_page(page);
 188	list_add_tail(&new->list, &fi->inmem_pages);
 189	inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 190	mutex_unlock(&fi->inmem_lock);
 191
 192	trace_f2fs_register_inmem_page(page, INMEM);
 193}
 194
 195static int __revoke_inmem_pages(struct inode *inode,
 196				struct list_head *head, bool drop, bool recover)
 197{
 198	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 199	struct inmem_pages *cur, *tmp;
 200	int err = 0;
 201
 202	list_for_each_entry_safe(cur, tmp, head, list) {
 203		struct page *page = cur->page;
 204
 205		if (drop)
 206			trace_f2fs_commit_inmem_page(page, INMEM_DROP);
 207
 208		lock_page(page);
 209
 210		if (recover) {
 211			struct dnode_of_data dn;
 212			struct node_info ni;
 213
 214			trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
 215
 216			set_new_dnode(&dn, inode, NULL, NULL, 0);
 217			if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) {
 218				err = -EAGAIN;
 219				goto next;
 220			}
 221			get_node_info(sbi, dn.nid, &ni);
 222			f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
 223					cur->old_addr, ni.version, true, true);
 224			f2fs_put_dnode(&dn);
 225		}
 226next:
 227		/* we don't need to invalidate this in the sccessful status */
 228		if (drop || recover)
 229			ClearPageUptodate(page);
 230		set_page_private(page, 0);
 231		ClearPagePrivate(page);
 232		f2fs_put_page(page, 1);
 233
 234		list_del(&cur->list);
 235		kmem_cache_free(inmem_entry_slab, cur);
 236		dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
 237	}
 238	return err;
 239}
 240
 241void drop_inmem_pages(struct inode *inode)
 242{
 243	struct f2fs_inode_info *fi = F2FS_I(inode);
 244
 245	clear_inode_flag(inode, FI_ATOMIC_FILE);
 246
 247	mutex_lock(&fi->inmem_lock);
 248	__revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 249	mutex_unlock(&fi->inmem_lock);
 250}
 251
 252static int __commit_inmem_pages(struct inode *inode,
 253					struct list_head *revoke_list)
 254{
 255	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 256	struct f2fs_inode_info *fi = F2FS_I(inode);
 257	struct inmem_pages *cur, *tmp;
 258	struct f2fs_io_info fio = {
 259		.sbi = sbi,
 260		.type = DATA,
 261		.op = REQ_OP_WRITE,
 262		.op_flags = REQ_SYNC | REQ_PRIO,
 263		.encrypted_page = NULL,
 264	};
 265	bool submit_bio = false;
 266	int err = 0;
 267
 268	list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
 269		struct page *page = cur->page;
 270
 271		lock_page(page);
 272		if (page->mapping == inode->i_mapping) {
 273			trace_f2fs_commit_inmem_page(page, INMEM);
 274
 275			set_page_dirty(page);
 276			f2fs_wait_on_page_writeback(page, DATA, true);
 277			if (clear_page_dirty_for_io(page)) {
 278				inode_dec_dirty_pages(inode);
 279				remove_dirty_inode(inode);
 280			}
 281
 282			fio.page = page;
 283			err = do_write_data_page(&fio);
 284			if (err) {
 285				unlock_page(page);
 286				break;
 287			}
 288
 289			/* record old blkaddr for revoking */
 290			cur->old_addr = fio.old_blkaddr;
 291
 292			submit_bio = true;
 293		}
 294		unlock_page(page);
 295		list_move_tail(&cur->list, revoke_list);
 296	}
 297
 298	if (submit_bio)
 299		f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
 300
 301	if (!err)
 302		__revoke_inmem_pages(inode, revoke_list, false, false);
 303
 304	return err;
 305}
 306
 307int commit_inmem_pages(struct inode *inode)
 308{
 309	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 310	struct f2fs_inode_info *fi = F2FS_I(inode);
 311	struct list_head revoke_list;
 312	int err;
 313
 314	INIT_LIST_HEAD(&revoke_list);
 315	f2fs_balance_fs(sbi, true);
 316	f2fs_lock_op(sbi);
 317
 318	mutex_lock(&fi->inmem_lock);
 319	err = __commit_inmem_pages(inode, &revoke_list);
 320	if (err) {
 321		int ret;
 322		/*
 323		 * try to revoke all committed pages, but still we could fail
 324		 * due to no memory or other reason, if that happened, EAGAIN
 325		 * will be returned, which means in such case, transaction is
 326		 * already not integrity, caller should use journal to do the
 327		 * recovery or rewrite & commit last transaction. For other
 328		 * error number, revoking was done by filesystem itself.
 329		 */
 330		ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
 331		if (ret)
 332			err = ret;
 333
 334		/* drop all uncommitted pages */
 335		__revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
 336	}
 337	mutex_unlock(&fi->inmem_lock);
 338
 339	f2fs_unlock_op(sbi);
 340	return err;
 341}
 342
 343/*
 344 * This function balances dirty node and dentry pages.
 345 * In addition, it controls garbage collection.
 346 */
 347void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 348{
 349#ifdef CONFIG_F2FS_FAULT_INJECTION
 350	if (time_to_inject(sbi, FAULT_CHECKPOINT))
 351		f2fs_stop_checkpoint(sbi, false);
 352#endif
 353
 354	if (!need)
 355		return;
 356
 357	/* balance_fs_bg is able to be pending */
 358	if (excess_cached_nats(sbi))
 359		f2fs_balance_fs_bg(sbi);
 360
 361	/*
 362	 * We should do GC or end up with checkpoint, if there are so many dirty
 363	 * dir/node pages without enough free segments.
 364	 */
 365	if (has_not_enough_free_secs(sbi, 0, 0)) {
 366		mutex_lock(&sbi->gc_mutex);
 367		f2fs_gc(sbi, false, false);
 368	}
 369}
 370
 371void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
 372{
 373	/* try to shrink extent cache when there is no enough memory */
 374	if (!available_free_memory(sbi, EXTENT_CACHE))
 375		f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
 376
 377	/* check the # of cached NAT entries */
 378	if (!available_free_memory(sbi, NAT_ENTRIES))
 379		try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
 380
 381	if (!available_free_memory(sbi, FREE_NIDS))
 382		try_to_free_nids(sbi, MAX_FREE_NIDS);
 383	else
 384		build_free_nids(sbi, false);
 385
 386	if (!is_idle(sbi))
 387		return;
 388
 389	/* checkpoint is the only way to shrink partial cached entries */
 390	if (!available_free_memory(sbi, NAT_ENTRIES) ||
 391			!available_free_memory(sbi, INO_ENTRIES) ||
 392			excess_prefree_segs(sbi) ||
 393			excess_dirty_nats(sbi) ||
 394			f2fs_time_over(sbi, CP_TIME)) {
 395		if (test_opt(sbi, DATA_FLUSH)) {
 396			struct blk_plug plug;
 397
 398			blk_start_plug(&plug);
 399			sync_dirty_inodes(sbi, FILE_INODE);
 400			blk_finish_plug(&plug);
 401		}
 402		f2fs_sync_fs(sbi->sb, true);
 403		stat_inc_bg_cp_count(sbi->stat_info);
 404	}
 405}
 406
 407static int __submit_flush_wait(struct block_device *bdev)
 408{
 409	struct bio *bio = f2fs_bio_alloc(0);
 410	int ret;
 411
 412	bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
 413	bio->bi_bdev = bdev;
 414	ret = submit_bio_wait(bio);
 415	bio_put(bio);
 416	return ret;
 417}
 418
 419static int submit_flush_wait(struct f2fs_sb_info *sbi)
 420{
 421	int ret = __submit_flush_wait(sbi->sb->s_bdev);
 422	int i;
 423
 424	if (sbi->s_ndevs && !ret) {
 425		for (i = 1; i < sbi->s_ndevs; i++) {
 426			ret = __submit_flush_wait(FDEV(i).bdev);
 427			if (ret)
 428				break;
 429		}
 430	}
 431	return ret;
 432}
 433
 434static int issue_flush_thread(void *data)
 435{
 436	struct f2fs_sb_info *sbi = data;
 437	struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 438	wait_queue_head_t *q = &fcc->flush_wait_queue;
 439repeat:
 440	if (kthread_should_stop())
 441		return 0;
 442
 443	if (!llist_empty(&fcc->issue_list)) {
 444		struct flush_cmd *cmd, *next;
 445		int ret;
 446
 447		fcc->dispatch_list = llist_del_all(&fcc->issue_list);
 448		fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
 449
 450		ret = submit_flush_wait(sbi);
 451		llist_for_each_entry_safe(cmd, next,
 452					  fcc->dispatch_list, llnode) {
 453			cmd->ret = ret;
 454			complete(&cmd->wait);
 455		}
 456		fcc->dispatch_list = NULL;
 457	}
 458
 459	wait_event_interruptible(*q,
 460		kthread_should_stop() || !llist_empty(&fcc->issue_list));
 461	goto repeat;
 462}
 463
 464int f2fs_issue_flush(struct f2fs_sb_info *sbi)
 465{
 466	struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 467	struct flush_cmd cmd;
 468
 469	trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
 470					test_opt(sbi, FLUSH_MERGE));
 471
 472	if (test_opt(sbi, NOBARRIER))
 473		return 0;
 474
 475	if (!test_opt(sbi, FLUSH_MERGE) || !atomic_read(&fcc->submit_flush)) {
 476		int ret;
 477
 478		atomic_inc(&fcc->submit_flush);
 479		ret = submit_flush_wait(sbi);
 480		atomic_dec(&fcc->submit_flush);
 481		return ret;
 482	}
 483
 484	init_completion(&cmd.wait);
 485
 486	atomic_inc(&fcc->submit_flush);
 487	llist_add(&cmd.llnode, &fcc->issue_list);
 488
 489	if (!fcc->dispatch_list)
 490		wake_up(&fcc->flush_wait_queue);
 491
 492	if (fcc->f2fs_issue_flush) {
 493		wait_for_completion(&cmd.wait);
 494		atomic_dec(&fcc->submit_flush);
 495	} else {
 496		llist_del_all(&fcc->issue_list);
 497		atomic_set(&fcc->submit_flush, 0);
 498	}
 499
 500	return cmd.ret;
 501}
 502
 503int create_flush_cmd_control(struct f2fs_sb_info *sbi)
 504{
 505	dev_t dev = sbi->sb->s_bdev->bd_dev;
 506	struct flush_cmd_control *fcc;
 507	int err = 0;
 508
 509	if (SM_I(sbi)->cmd_control_info) {
 510		fcc = SM_I(sbi)->cmd_control_info;
 511		goto init_thread;
 512	}
 513
 514	fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
 515	if (!fcc)
 516		return -ENOMEM;
 517	atomic_set(&fcc->submit_flush, 0);
 518	init_waitqueue_head(&fcc->flush_wait_queue);
 519	init_llist_head(&fcc->issue_list);
 520	SM_I(sbi)->cmd_control_info = fcc;
 521init_thread:
 522	fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
 523				"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
 524	if (IS_ERR(fcc->f2fs_issue_flush)) {
 525		err = PTR_ERR(fcc->f2fs_issue_flush);
 526		kfree(fcc);
 527		SM_I(sbi)->cmd_control_info = NULL;
 528		return err;
 529	}
 530
 531	return err;
 532}
 533
 534void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
 535{
 536	struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
 537
 538	if (fcc && fcc->f2fs_issue_flush) {
 539		struct task_struct *flush_thread = fcc->f2fs_issue_flush;
 540
 541		fcc->f2fs_issue_flush = NULL;
 542		kthread_stop(flush_thread);
 543	}
 544	if (free) {
 545		kfree(fcc);
 546		SM_I(sbi)->cmd_control_info = NULL;
 547	}
 548}
 549
 550static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 551		enum dirty_type dirty_type)
 552{
 553	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 554
 555	/* need not be added */
 556	if (IS_CURSEG(sbi, segno))
 557		return;
 558
 559	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 560		dirty_i->nr_dirty[dirty_type]++;
 561
 562	if (dirty_type == DIRTY) {
 563		struct seg_entry *sentry = get_seg_entry(sbi, segno);
 564		enum dirty_type t = sentry->type;
 565
 566		if (unlikely(t >= DIRTY)) {
 567			f2fs_bug_on(sbi, 1);
 568			return;
 569		}
 570		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
 571			dirty_i->nr_dirty[t]++;
 572	}
 573}
 574
 575static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
 576		enum dirty_type dirty_type)
 577{
 578	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 579
 580	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
 581		dirty_i->nr_dirty[dirty_type]--;
 582
 583	if (dirty_type == DIRTY) {
 584		struct seg_entry *sentry = get_seg_entry(sbi, segno);
 585		enum dirty_type t = sentry->type;
 586
 587		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
 588			dirty_i->nr_dirty[t]--;
 589
 590		if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
 591			clear_bit(GET_SECNO(sbi, segno),
 592						dirty_i->victim_secmap);
 593	}
 594}
 595
 596/*
 597 * Should not occur error such as -ENOMEM.
 598 * Adding dirty entry into seglist is not critical operation.
 599 * If a given segment is one of current working segments, it won't be added.
 600 */
 601static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
 602{
 603	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 604	unsigned short valid_blocks;
 605
 606	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 607		return;
 608
 609	mutex_lock(&dirty_i->seglist_lock);
 610
 611	valid_blocks = get_valid_blocks(sbi, segno, 0);
 612
 613	if (valid_blocks == 0) {
 614		__locate_dirty_segment(sbi, segno, PRE);
 615		__remove_dirty_segment(sbi, segno, DIRTY);
 616	} else if (valid_blocks < sbi->blocks_per_seg) {
 617		__locate_dirty_segment(sbi, segno, DIRTY);
 618	} else {
 619		/* Recovery routine with SSR needs this */
 620		__remove_dirty_segment(sbi, segno, DIRTY);
 621	}
 622
 623	mutex_unlock(&dirty_i->seglist_lock);
 624}
 625
 626static struct bio_entry *__add_bio_entry(struct f2fs_sb_info *sbi,
 627							struct bio *bio)
 628{
 629	struct list_head *wait_list = &(SM_I(sbi)->wait_list);
 630	struct bio_entry *be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS);
 631
 632	INIT_LIST_HEAD(&be->list);
 633	be->bio = bio;
 634	init_completion(&be->event);
 635	list_add_tail(&be->list, wait_list);
 636
 637	return be;
 638}
 639
 640void f2fs_wait_all_discard_bio(struct f2fs_sb_info *sbi)
 641{
 642	struct list_head *wait_list = &(SM_I(sbi)->wait_list);
 643	struct bio_entry *be, *tmp;
 644
 645	list_for_each_entry_safe(be, tmp, wait_list, list) {
 646		struct bio *bio = be->bio;
 647		int err;
 648
 649		wait_for_completion_io(&be->event);
 650		err = be->error;
 651		if (err == -EOPNOTSUPP)
 652			err = 0;
 653
 654		if (err)
 655			f2fs_msg(sbi->sb, KERN_INFO,
 656				"Issue discard failed, ret: %d", err);
 657
 658		bio_put(bio);
 659		list_del(&be->list);
 660		kmem_cache_free(bio_entry_slab, be);
 661	}
 662}
 663
 664static void f2fs_submit_bio_wait_endio(struct bio *bio)
 665{
 666	struct bio_entry *be = (struct bio_entry *)bio->bi_private;
 667
 668	be->error = bio->bi_error;
 669	complete(&be->event);
 670}
 671
 672/* this function is copied from blkdev_issue_discard from block/blk-lib.c */
 673static int __f2fs_issue_discard_async(struct f2fs_sb_info *sbi,
 674		struct block_device *bdev, block_t blkstart, block_t blklen)
 675{
 676	struct bio *bio = NULL;
 677	int err;
 678
 679	trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
 680
 681	if (sbi->s_ndevs) {
 682		int devi = f2fs_target_device_index(sbi, blkstart);
 683
 684		blkstart -= FDEV(devi).start_blk;
 685	}
 686	err = __blkdev_issue_discard(bdev,
 687				SECTOR_FROM_BLOCK(blkstart),
 688				SECTOR_FROM_BLOCK(blklen),
 689				GFP_NOFS, 0, &bio);
 690	if (!err && bio) {
 691		struct bio_entry *be = __add_bio_entry(sbi, bio);
 692
 693		bio->bi_private = be;
 694		bio->bi_end_io = f2fs_submit_bio_wait_endio;
 695		bio->bi_opf |= REQ_SYNC;
 696		submit_bio(bio);
 697	}
 698
 699	return err;
 700}
 701
 702#ifdef CONFIG_BLK_DEV_ZONED
 703static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
 704		struct block_device *bdev, block_t blkstart, block_t blklen)
 705{
 706	sector_t nr_sects = SECTOR_FROM_BLOCK(blklen);
 707	sector_t sector;
 708	int devi = 0;
 709
 710	if (sbi->s_ndevs) {
 711		devi = f2fs_target_device_index(sbi, blkstart);
 712		blkstart -= FDEV(devi).start_blk;
 713	}
 714	sector = SECTOR_FROM_BLOCK(blkstart);
 715
 716	if (sector & (bdev_zone_sectors(bdev) - 1) ||
 717	    nr_sects != bdev_zone_sectors(bdev)) {
 718		f2fs_msg(sbi->sb, KERN_INFO,
 719			"(%d) %s: Unaligned discard attempted (block %x + %x)",
 720			devi, sbi->s_ndevs ? FDEV(devi).path: "",
 721			blkstart, blklen);
 722		return -EIO;
 723	}
 724
 725	/*
 726	 * We need to know the type of the zone: for conventional zones,
 727	 * use regular discard if the drive supports it. For sequential
 728	 * zones, reset the zone write pointer.
 729	 */
 730	switch (get_blkz_type(sbi, bdev, blkstart)) {
 731
 732	case BLK_ZONE_TYPE_CONVENTIONAL:
 733		if (!blk_queue_discard(bdev_get_queue(bdev)))
 734			return 0;
 735		return __f2fs_issue_discard_async(sbi, bdev, blkstart, blklen);
 736	case BLK_ZONE_TYPE_SEQWRITE_REQ:
 737	case BLK_ZONE_TYPE_SEQWRITE_PREF:
 738		trace_f2fs_issue_reset_zone(sbi->sb, blkstart);
 739		return blkdev_reset_zones(bdev, sector,
 740					  nr_sects, GFP_NOFS);
 741	default:
 742		/* Unknown zone type: broken device ? */
 743		return -EIO;
 744	}
 745}
 746#endif
 747
 748static int __issue_discard_async(struct f2fs_sb_info *sbi,
 749		struct block_device *bdev, block_t blkstart, block_t blklen)
 750{
 751#ifdef CONFIG_BLK_DEV_ZONED
 752	if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
 753				bdev_zoned_model(bdev) != BLK_ZONED_NONE)
 754		return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
 755#endif
 756	return __f2fs_issue_discard_async(sbi, bdev, blkstart, blklen);
 757}
 758
 759static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
 760				block_t blkstart, block_t blklen)
 761{
 762	sector_t start = blkstart, len = 0;
 763	struct block_device *bdev;
 764	struct seg_entry *se;
 765	unsigned int offset;
 766	block_t i;
 767	int err = 0;
 768
 769	bdev = f2fs_target_device(sbi, blkstart, NULL);
 770
 771	for (i = blkstart; i < blkstart + blklen; i++, len++) {
 772		if (i != start) {
 773			struct block_device *bdev2 =
 774				f2fs_target_device(sbi, i, NULL);
 775
 776			if (bdev2 != bdev) {
 777				err = __issue_discard_async(sbi, bdev,
 778						start, len);
 779				if (err)
 780					return err;
 781				bdev = bdev2;
 782				start = i;
 783				len = 0;
 784			}
 785		}
 786
 787		se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
 788		offset = GET_BLKOFF_FROM_SEG0(sbi, i);
 789
 790		if (!f2fs_test_and_set_bit(offset, se->discard_map))
 791			sbi->discard_blks--;
 792	}
 793
 794	if (len)
 795		err = __issue_discard_async(sbi, bdev, start, len);
 796	return err;
 797}
 798
 799static void __add_discard_entry(struct f2fs_sb_info *sbi,
 800		struct cp_control *cpc, struct seg_entry *se,
 801		unsigned int start, unsigned int end)
 802{
 803	struct list_head *head = &SM_I(sbi)->discard_list;
 804	struct discard_entry *new, *last;
 805
 806	if (!list_empty(head)) {
 807		last = list_last_entry(head, struct discard_entry, list);
 808		if (START_BLOCK(sbi, cpc->trim_start) + start ==
 809						last->blkaddr + last->len) {
 810			last->len += end - start;
 811			goto done;
 812		}
 813	}
 814
 815	new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
 816	INIT_LIST_HEAD(&new->list);
 817	new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
 818	new->len = end - start;
 819	list_add_tail(&new->list, head);
 820done:
 821	SM_I(sbi)->nr_discards += end - start;
 822}
 823
 824static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 825{
 826	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
 827	int max_blocks = sbi->blocks_per_seg;
 828	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
 829	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
 830	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
 831	unsigned long *discard_map = (unsigned long *)se->discard_map;
 832	unsigned long *dmap = SIT_I(sbi)->tmp_map;
 833	unsigned int start = 0, end = -1;
 834	bool force = (cpc->reason == CP_DISCARD);
 835	int i;
 836
 837	if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
 838		return;
 839
 840	if (!force) {
 841		if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
 842		    SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
 843			return;
 844	}
 845
 846	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
 847	for (i = 0; i < entries; i++)
 848		dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
 849				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
 850
 851	while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
 852		start = __find_rev_next_bit(dmap, max_blocks, end + 1);
 853		if (start >= max_blocks)
 854			break;
 855
 856		end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
 857		if (force && start && end != max_blocks
 858					&& (end - start) < cpc->trim_minlen)
 859			continue;
 860
 861		__add_discard_entry(sbi, cpc, se, start, end);
 862	}
 863}
 864
 865void release_discard_addrs(struct f2fs_sb_info *sbi)
 866{
 867	struct list_head *head = &(SM_I(sbi)->discard_list);
 868	struct discard_entry *entry, *this;
 869
 870	/* drop caches */
 871	list_for_each_entry_safe(entry, this, head, list) {
 872		list_del(&entry->list);
 873		kmem_cache_free(discard_entry_slab, entry);
 874	}
 875}
 876
 877/*
 878 * Should call clear_prefree_segments after checkpoint is done.
 879 */
 880static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
 881{
 882	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 883	unsigned int segno;
 884
 885	mutex_lock(&dirty_i->seglist_lock);
 886	for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
 887		__set_test_and_free(sbi, segno);
 888	mutex_unlock(&dirty_i->seglist_lock);
 889}
 890
 891void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 892{
 893	struct list_head *head = &(SM_I(sbi)->discard_list);
 894	struct discard_entry *entry, *this;
 895	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 896	struct blk_plug plug;
 897	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
 898	unsigned int start = 0, end = -1;
 899	unsigned int secno, start_segno;
 900	bool force = (cpc->reason == CP_DISCARD);
 901
 902	blk_start_plug(&plug);
 903
 904	mutex_lock(&dirty_i->seglist_lock);
 905
 906	while (1) {
 907		int i;
 908		start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
 909		if (start >= MAIN_SEGS(sbi))
 910			break;
 911		end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
 912								start + 1);
 913
 914		for (i = start; i < end; i++)
 915			clear_bit(i, prefree_map);
 916
 917		dirty_i->nr_dirty[PRE] -= end - start;
 918
 919		if (force || !test_opt(sbi, DISCARD))
 920			continue;
 921
 922		if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
 923			f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
 924				(end - start) << sbi->log_blocks_per_seg);
 925			continue;
 926		}
 927next:
 928		secno = GET_SECNO(sbi, start);
 929		start_segno = secno * sbi->segs_per_sec;
 930		if (!IS_CURSEC(sbi, secno) &&
 931			!get_valid_blocks(sbi, start, sbi->segs_per_sec))
 932			f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
 933				sbi->segs_per_sec << sbi->log_blocks_per_seg);
 934
 935		start = start_segno + sbi->segs_per_sec;
 936		if (start < end)
 937			goto next;
 938		else
 939			end = start - 1;
 940	}
 941	mutex_unlock(&dirty_i->seglist_lock);
 942
 943	/* send small discards */
 944	list_for_each_entry_safe(entry, this, head, list) {
 945		if (force && entry->len < cpc->trim_minlen)
 946			goto skip;
 947		f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
 948		cpc->trimmed += entry->len;
 949skip:
 950		list_del(&entry->list);
 951		SM_I(sbi)->nr_discards -= entry->len;
 952		kmem_cache_free(discard_entry_slab, entry);
 953	}
 954
 955	blk_finish_plug(&plug);
 956}
 957
 958static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
 959{
 960	struct sit_info *sit_i = SIT_I(sbi);
 961
 962	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
 963		sit_i->dirty_sentries++;
 964		return false;
 965	}
 966
 967	return true;
 968}
 969
 970static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
 971					unsigned int segno, int modified)
 972{
 973	struct seg_entry *se = get_seg_entry(sbi, segno);
 974	se->type = type;
 975	if (modified)
 976		__mark_sit_entry_dirty(sbi, segno);
 977}
 978
 979static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
 980{
 981	struct seg_entry *se;
 982	unsigned int segno, offset;
 983	long int new_vblocks;
 984
 985	segno = GET_SEGNO(sbi, blkaddr);
 986
 987	se = get_seg_entry(sbi, segno);
 988	new_vblocks = se->valid_blocks + del;
 989	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 990
 991	f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
 992				(new_vblocks > sbi->blocks_per_seg)));
 993
 994	se->valid_blocks = new_vblocks;
 995	se->mtime = get_mtime(sbi);
 996	SIT_I(sbi)->max_mtime = se->mtime;
 997
 998	/* Update valid block bitmap */
 999	if (del > 0) {
1000		if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
1001			f2fs_bug_on(sbi, 1);
1002		if (f2fs_discard_en(sbi) &&
1003			!f2fs_test_and_set_bit(offset, se->discard_map))
1004			sbi->discard_blks--;
1005	} else {
1006		if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
1007			f2fs_bug_on(sbi, 1);
1008		if (f2fs_discard_en(sbi) &&
1009			f2fs_test_and_clear_bit(offset, se->discard_map))
1010			sbi->discard_blks++;
1011	}
1012	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1013		se->ckpt_valid_blocks += del;
1014
1015	__mark_sit_entry_dirty(sbi, segno);
1016
1017	/* update total number of valid blocks to be written in ckpt area */
1018	SIT_I(sbi)->written_valid_blocks += del;
1019
1020	if (sbi->segs_per_sec > 1)
1021		get_sec_entry(sbi, segno)->valid_blocks += del;
1022}
1023
1024void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
1025{
1026	update_sit_entry(sbi, new, 1);
1027	if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1028		update_sit_entry(sbi, old, -1);
1029
1030	locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1031	locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
1032}
1033
1034void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1035{
1036	unsigned int segno = GET_SEGNO(sbi, addr);
1037	struct sit_info *sit_i = SIT_I(sbi);
1038
1039	f2fs_bug_on(sbi, addr == NULL_ADDR);
1040	if (addr == NEW_ADDR)
1041		return;
1042
1043	/* add it into sit main buffer */
1044	mutex_lock(&sit_i->sentry_lock);
1045
1046	update_sit_entry(sbi, addr, -1);
1047
1048	/* add it into dirty seglist */
1049	locate_dirty_segment(sbi, segno);
1050
1051	mutex_unlock(&sit_i->sentry_lock);
1052}
1053
1054bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1055{
1056	struct sit_info *sit_i = SIT_I(sbi);
1057	unsigned int segno, offset;
1058	struct seg_entry *se;
1059	bool is_cp = false;
1060
1061	if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1062		return true;
1063
1064	mutex_lock(&sit_i->sentry_lock);
1065
1066	segno = GET_SEGNO(sbi, blkaddr);
1067	se = get_seg_entry(sbi, segno);
1068	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1069
1070	if (f2fs_test_bit(offset, se->ckpt_valid_map))
1071		is_cp = true;
1072
1073	mutex_unlock(&sit_i->sentry_lock);
1074
1075	return is_cp;
1076}
1077
1078/*
1079 * This function should be resided under the curseg_mutex lock
1080 */
1081static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1082					struct f2fs_summary *sum)
1083{
1084	struct curseg_info *curseg = CURSEG_I(sbi, type);
1085	void *addr = curseg->sum_blk;
1086	addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1087	memcpy(addr, sum, sizeof(struct f2fs_summary));
1088}
1089
1090/*
1091 * Calculate the number of current summary pages for writing
1092 */
1093int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1094{
1095	int valid_sum_count = 0;
1096	int i, sum_in_page;
1097
1098	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1099		if (sbi->ckpt->alloc_type[i] == SSR)
1100			valid_sum_count += sbi->blocks_per_seg;
1101		else {
1102			if (for_ra)
1103				valid_sum_count += le16_to_cpu(
1104					F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1105			else
1106				valid_sum_count += curseg_blkoff(sbi, i);
1107		}
1108	}
1109
1110	sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1111			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1112	if (valid_sum_count <= sum_in_page)
1113		return 1;
1114	else if ((valid_sum_count - sum_in_page) <=
1115		(PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1116		return 2;
1117	return 3;
1118}
1119
1120/*
1121 * Caller should put this summary page
1122 */
1123struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1124{
1125	return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1126}
1127
1128void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1129{
1130	struct page *page = grab_meta_page(sbi, blk_addr);
1131	void *dst = page_address(page);
1132
1133	if (src)
1134		memcpy(dst, src, PAGE_SIZE);
1135	else
1136		memset(dst, 0, PAGE_SIZE);
1137	set_page_dirty(page);
1138	f2fs_put_page(page, 1);
1139}
1140
1141static void write_sum_page(struct f2fs_sb_info *sbi,
1142			struct f2fs_summary_block *sum_blk, block_t blk_addr)
1143{
1144	update_meta_page(sbi, (void *)sum_blk, blk_addr);
1145}
1146
1147static void write_current_sum_page(struct f2fs_sb_info *sbi,
1148						int type, block_t blk_addr)
1149{
1150	struct curseg_info *curseg = CURSEG_I(sbi, type);
1151	struct page *page = grab_meta_page(sbi, blk_addr);
1152	struct f2fs_summary_block *src = curseg->sum_blk;
1153	struct f2fs_summary_block *dst;
1154
1155	dst = (struct f2fs_summary_block *)page_address(page);
1156
1157	mutex_lock(&curseg->curseg_mutex);
1158
1159	down_read(&curseg->journal_rwsem);
1160	memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1161	up_read(&curseg->journal_rwsem);
1162
1163	memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1164	memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1165
1166	mutex_unlock(&curseg->curseg_mutex);
1167
1168	set_page_dirty(page);
1169	f2fs_put_page(page, 1);
1170}
1171
1172static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1173{
1174	struct curseg_info *curseg = CURSEG_I(sbi, type);
1175	unsigned int segno = curseg->segno + 1;
1176	struct free_segmap_info *free_i = FREE_I(sbi);
1177
1178	if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
1179		return !test_bit(segno, free_i->free_segmap);
1180	return 0;
1181}
1182
1183/*
1184 * Find a new segment from the free segments bitmap to right order
1185 * This function should be returned with success, otherwise BUG
1186 */
1187static void get_new_segment(struct f2fs_sb_info *sbi,
1188			unsigned int *newseg, bool new_sec, int dir)
1189{
1190	struct free_segmap_info *free_i = FREE_I(sbi);
1191	unsigned int segno, secno, zoneno;
1192	unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
1193	unsigned int hint = *newseg / sbi->segs_per_sec;
1194	unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1195	unsigned int left_start = hint;
1196	bool init = true;
1197	int go_left = 0;
1198	int i;
1199
1200	spin_lock(&free_i->segmap_lock);
1201
1202	if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1203		segno = find_next_zero_bit(free_i->free_segmap,
1204				(hint + 1) * sbi->segs_per_sec, *newseg + 1);
1205		if (segno < (hint + 1) * sbi->segs_per_sec)
1206			goto got_it;
1207	}
1208find_other_zone:
1209	secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1210	if (secno >= MAIN_SECS(sbi)) {
1211		if (dir == ALLOC_RIGHT) {
1212			secno = find_next_zero_bit(free_i->free_secmap,
1213							MAIN_SECS(sbi), 0);
1214			f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1215		} else {
1216			go_left = 1;
1217			left_start = hint - 1;
1218		}
1219	}
1220	if (go_left == 0)
1221		goto skip_left;
1222
1223	while (test_bit(left_start, free_i->free_secmap)) {
1224		if (left_start > 0) {
1225			left_start--;
1226			continue;
1227		}
1228		left_start = find_next_zero_bit(free_i->free_secmap,
1229							MAIN_SECS(sbi), 0);
1230		f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1231		break;
1232	}
1233	secno = left_start;
1234skip_left:
1235	hint = secno;
1236	segno = secno * sbi->segs_per_sec;
1237	zoneno = secno / sbi->secs_per_zone;
1238
1239	/* give up on finding another zone */
1240	if (!init)
1241		goto got_it;
1242	if (sbi->secs_per_zone == 1)
1243		goto got_it;
1244	if (zoneno == old_zoneno)
1245		goto got_it;
1246	if (dir == ALLOC_LEFT) {
1247		if (!go_left && zoneno + 1 >= total_zones)
1248			goto got_it;
1249		if (go_left && zoneno == 0)
1250			goto got_it;
1251	}
1252	for (i = 0; i < NR_CURSEG_TYPE; i++)
1253		if (CURSEG_I(sbi, i)->zone == zoneno)
1254			break;
1255
1256	if (i < NR_CURSEG_TYPE) {
1257		/* zone is in user, try another */
1258		if (go_left)
1259			hint = zoneno * sbi->secs_per_zone - 1;
1260		else if (zoneno + 1 >= total_zones)
1261			hint = 0;
1262		else
1263			hint = (zoneno + 1) * sbi->secs_per_zone;
1264		init = false;
1265		goto find_other_zone;
1266	}
1267got_it:
1268	/* set it as dirty segment in free segmap */
1269	f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1270	__set_inuse(sbi, segno);
1271	*newseg = segno;
1272	spin_unlock(&free_i->segmap_lock);
1273}
1274
1275static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1276{
1277	struct curseg_info *curseg = CURSEG_I(sbi, type);
1278	struct summary_footer *sum_footer;
1279
1280	curseg->segno = curseg->next_segno;
1281	curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1282	curseg->next_blkoff = 0;
1283	curseg->next_segno = NULL_SEGNO;
1284
1285	sum_footer = &(curseg->sum_blk->footer);
1286	memset(sum_footer, 0, sizeof(struct summary_footer));
1287	if (IS_DATASEG(type))
1288		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1289	if (IS_NODESEG(type))
1290		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1291	__set_sit_entry_type(sbi, type, curseg->segno, modified);
1292}
1293
1294/*
1295 * Allocate a current working segment.
1296 * This function always allocates a free segment in LFS manner.
1297 */
1298static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1299{
1300	struct curseg_info *curseg = CURSEG_I(sbi, type);
1301	unsigned int segno = curseg->segno;
1302	int dir = ALLOC_LEFT;
1303
1304	write_sum_page(sbi, curseg->sum_blk,
1305				GET_SUM_BLOCK(sbi, segno));
1306	if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1307		dir = ALLOC_RIGHT;
1308
1309	if (test_opt(sbi, NOHEAP))
1310		dir = ALLOC_RIGHT;
1311
1312	get_new_segment(sbi, &segno, new_sec, dir);
1313	curseg->next_segno = segno;
1314	reset_curseg(sbi, type, 1);
1315	curseg->alloc_type = LFS;
1316}
1317
1318static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1319			struct curseg_info *seg, block_t start)
1320{
1321	struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1322	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1323	unsigned long *target_map = SIT_I(sbi)->tmp_map;
1324	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1325	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1326	int i, pos;
1327
1328	for (i = 0; i < entries; i++)
1329		target_map[i] = ckpt_map[i] | cur_map[i];
1330
1331	pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1332
1333	seg->next_blkoff = pos;
1334}
1335
1336/*
1337 * If a segment is written by LFS manner, next block offset is just obtained
1338 * by increasing the current block offset. However, if a segment is written by
1339 * SSR manner, next block offset obtained by calling __next_free_blkoff
1340 */
1341static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1342				struct curseg_info *seg)
1343{
1344	if (seg->alloc_type == SSR)
1345		__next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1346	else
1347		seg->next_blkoff++;
1348}
1349
1350/*
1351 * This function always allocates a used segment(from dirty seglist) by SSR
1352 * manner, so it should recover the existing segment information of valid blocks
1353 */
1354static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1355{
1356	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1357	struct curseg_info *curseg = CURSEG_I(sbi, type);
1358	unsigned int new_segno = curseg->next_segno;
1359	struct f2fs_summary_block *sum_node;
1360	struct page *sum_page;
1361
1362	write_sum_page(sbi, curseg->sum_blk,
1363				GET_SUM_BLOCK(sbi, curseg->segno));
1364	__set_test_and_inuse(sbi, new_segno);
1365
1366	mutex_lock(&dirty_i->seglist_lock);
1367	__remove_dirty_segment(sbi, new_segno, PRE);
1368	__remove_dirty_segment(sbi, new_segno, DIRTY);
1369	mutex_unlock(&dirty_i->seglist_lock);
1370
1371	reset_curseg(sbi, type, 1);
1372	curseg->alloc_type = SSR;
1373	__next_free_blkoff(sbi, curseg, 0);
1374
1375	if (reuse) {
1376		sum_page = get_sum_page(sbi, new_segno);
1377		sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1378		memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1379		f2fs_put_page(sum_page, 1);
1380	}
1381}
1382
1383static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1384{
1385	struct curseg_info *curseg = CURSEG_I(sbi, type);
1386	const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1387
1388	if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0, 0))
1389		return v_ops->get_victim(sbi,
1390				&(curseg)->next_segno, BG_GC, type, SSR);
1391
1392	/* For data segments, let's do SSR more intensively */
1393	for (; type >= CURSEG_HOT_DATA; type--)
1394		if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1395						BG_GC, type, SSR))
1396			return 1;
1397	return 0;
1398}
1399
1400/*
1401 * flush out current segment and replace it with new segment
1402 * This function should be returned with success, otherwise BUG
1403 */
1404static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1405						int type, bool force)
1406{
1407	struct curseg_info *curseg = CURSEG_I(sbi, type);
1408
1409	if (force)
1410		new_curseg(sbi, type, true);
1411	else if (type == CURSEG_WARM_NODE)
1412		new_curseg(sbi, type, false);
1413	else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1414		new_curseg(sbi, type, false);
1415	else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1416		change_curseg(sbi, type, true);
1417	else
1418		new_curseg(sbi, type, false);
1419
1420	stat_inc_seg_type(sbi, curseg);
1421}
1422
1423void allocate_new_segments(struct f2fs_sb_info *sbi)
1424{
1425	struct curseg_info *curseg;
1426	unsigned int old_segno;
1427	int i;
1428
1429	if (test_opt(sbi, LFS))
1430		return;
1431
1432	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1433		curseg = CURSEG_I(sbi, i);
1434		old_segno = curseg->segno;
1435		SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
1436		locate_dirty_segment(sbi, old_segno);
1437	}
1438}
1439
1440static const struct segment_allocation default_salloc_ops = {
1441	.allocate_segment = allocate_segment_by_default,
1442};
1443
1444int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1445{
1446	__u64 start = F2FS_BYTES_TO_BLK(range->start);
1447	__u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1448	unsigned int start_segno, end_segno;
1449	struct cp_control cpc;
1450	int err = 0;
1451
1452	if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1453		return -EINVAL;
1454
1455	cpc.trimmed = 0;
1456	if (end <= MAIN_BLKADDR(sbi))
1457		goto out;
1458
1459	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1460		f2fs_msg(sbi->sb, KERN_WARNING,
1461			"Found FS corruption, run fsck to fix.");
1462		goto out;
1463	}
1464
1465	/* start/end segment number in main_area */
1466	start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1467	end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1468						GET_SEGNO(sbi, end);
1469	cpc.reason = CP_DISCARD;
1470	cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1471
1472	/* do checkpoint to issue discard commands safely */
1473	for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1474		cpc.trim_start = start_segno;
1475
1476		if (sbi->discard_blks == 0)
1477			break;
1478		else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1479			cpc.trim_end = end_segno;
1480		else
1481			cpc.trim_end = min_t(unsigned int,
1482				rounddown(start_segno +
1483				BATCHED_TRIM_SEGMENTS(sbi),
1484				sbi->segs_per_sec) - 1, end_segno);
1485
1486		mutex_lock(&sbi->gc_mutex);
1487		err = write_checkpoint(sbi, &cpc);
1488		mutex_unlock(&sbi->gc_mutex);
1489		if (err)
1490			break;
1491
1492		schedule();
1493	}
1494out:
1495	range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1496	return err;
1497}
1498
1499static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1500{
1501	struct curseg_info *curseg = CURSEG_I(sbi, type);
1502	if (curseg->next_blkoff < sbi->blocks_per_seg)
1503		return true;
1504	return false;
1505}
1506
1507static int __get_segment_type_2(struct page *page, enum page_type p_type)
1508{
1509	if (p_type == DATA)
1510		return CURSEG_HOT_DATA;
1511	else
1512		return CURSEG_HOT_NODE;
1513}
1514
1515static int __get_segment_type_4(struct page *page, enum page_type p_type)
1516{
1517	if (p_type == DATA) {
1518		struct inode *inode = page->mapping->host;
1519
1520		if (S_ISDIR(inode->i_mode))
1521			return CURSEG_HOT_DATA;
1522		else
1523			return CURSEG_COLD_DATA;
1524	} else {
1525		if (IS_DNODE(page) && is_cold_node(page))
1526			return CURSEG_WARM_NODE;
1527		else
1528			return CURSEG_COLD_NODE;
1529	}
1530}
1531
1532static int __get_segment_type_6(struct page *page, enum page_type p_type)
1533{
1534	if (p_type == DATA) {
1535		struct inode *inode = page->mapping->host;
1536
1537		if (S_ISDIR(inode->i_mode))
1538			return CURSEG_HOT_DATA;
1539		else if (is_cold_data(page) || file_is_cold(inode))
1540			return CURSEG_COLD_DATA;
1541		else
1542			return CURSEG_WARM_DATA;
1543	} else {
1544		if (IS_DNODE(page))
1545			return is_cold_node(page) ? CURSEG_WARM_NODE :
1546						CURSEG_HOT_NODE;
1547		else
1548			return CURSEG_COLD_NODE;
1549	}
1550}
1551
1552static int __get_segment_type(struct page *page, enum page_type p_type)
1553{
1554	switch (F2FS_P_SB(page)->active_logs) {
1555	case 2:
1556		return __get_segment_type_2(page, p_type);
1557	case 4:
1558		return __get_segment_type_4(page, p_type);
1559	}
1560	/* NR_CURSEG_TYPE(6) logs by default */
1561	f2fs_bug_on(F2FS_P_SB(page),
1562		F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1563	return __get_segment_type_6(page, p_type);
1564}
1565
1566void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1567		block_t old_blkaddr, block_t *new_blkaddr,
1568		struct f2fs_summary *sum, int type)
1569{
1570	struct sit_info *sit_i = SIT_I(sbi);
1571	struct curseg_info *curseg = CURSEG_I(sbi, type);
1572
1573	mutex_lock(&curseg->curseg_mutex);
1574	mutex_lock(&sit_i->sentry_lock);
1575
1576	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1577
1578	/*
1579	 * __add_sum_entry should be resided under the curseg_mutex
1580	 * because, this function updates a summary entry in the
1581	 * current summary block.
1582	 */
1583	__add_sum_entry(sbi, type, sum);
1584
1585	__refresh_next_blkoff(sbi, curseg);
1586
1587	stat_inc_block_count(sbi, curseg);
1588
1589	if (!__has_curseg_space(sbi, type))
1590		sit_i->s_ops->allocate_segment(sbi, type, false);
1591	/*
1592	 * SIT information should be updated before segment allocation,
1593	 * since SSR needs latest valid block information.
1594	 */
1595	refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1596
1597	mutex_unlock(&sit_i->sentry_lock);
1598
1599	if (page && IS_NODESEG(type))
1600		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1601
1602	mutex_unlock(&curseg->curseg_mutex);
1603}
1604
1605static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1606{
1607	int type = __get_segment_type(fio->page, fio->type);
1608
1609	if (fio->type == NODE || fio->type == DATA)
1610		mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1611
1612	allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1613					&fio->new_blkaddr, sum, type);
1614
1615	/* writeout dirty page into bdev */
1616	f2fs_submit_page_mbio(fio);
1617
1618	if (fio->type == NODE || fio->type == DATA)
1619		mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
1620}
1621
1622void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1623{
1624	struct f2fs_io_info fio = {
1625		.sbi = sbi,
1626		.type = META,
1627		.op = REQ_OP_WRITE,
1628		.op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
1629		.old_blkaddr = page->index,
1630		.new_blkaddr = page->index,
1631		.page = page,
1632		.encrypted_page = NULL,
1633	};
1634
1635	if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1636		fio.op_flags &= ~REQ_META;
1637
1638	set_page_writeback(page);
1639	f2fs_submit_page_mbio(&fio);
1640}
1641
1642void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1643{
1644	struct f2fs_summary sum;
1645
1646	set_summary(&sum, nid, 0, 0);
1647	do_write_page(&sum, fio);
1648}
1649
1650void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1651{
1652	struct f2fs_sb_info *sbi = fio->sbi;
1653	struct f2fs_summary sum;
1654	struct node_info ni;
1655
1656	f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1657	get_node_info(sbi, dn->nid, &ni);
1658	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1659	do_write_page(&sum, fio);
1660	f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
1661}
1662
1663void rewrite_data_page(struct f2fs_io_info *fio)
1664{
1665	fio->new_blkaddr = fio->old_blkaddr;
1666	stat_inc_inplace_blocks(fio->sbi);
1667	f2fs_submit_page_mbio(fio);
1668}
1669
1670void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1671				block_t old_blkaddr, block_t new_blkaddr,
1672				bool recover_curseg, bool recover_newaddr)
1673{
1674	struct sit_info *sit_i = SIT_I(sbi);
1675	struct curseg_info *curseg;
1676	unsigned int segno, old_cursegno;
1677	struct seg_entry *se;
1678	int type;
1679	unsigned short old_blkoff;
1680
1681	segno = GET_SEGNO(sbi, new_blkaddr);
1682	se = get_seg_entry(sbi, segno);
1683	type = se->type;
1684
1685	if (!recover_curseg) {
1686		/* for recovery flow */
1687		if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1688			if (old_blkaddr == NULL_ADDR)
1689				type = CURSEG_COLD_DATA;
1690			else
1691				type = CURSEG_WARM_DATA;
1692		}
1693	} else {
1694		if (!IS_CURSEG(sbi, segno))
1695			type = CURSEG_WARM_DATA;
1696	}
1697
1698	curseg = CURSEG_I(sbi, type);
1699
1700	mutex_lock(&curseg->curseg_mutex);
1701	mutex_lock(&sit_i->sentry_lock);
1702
1703	old_cursegno = curseg->segno;
1704	old_blkoff = curseg->next_blkoff;
1705
1706	/* change the current segment */
1707	if (segno != curseg->segno) {
1708		curseg->next_segno = segno;
1709		change_curseg(sbi, type, true);
1710	}
1711
1712	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1713	__add_sum_entry(sbi, type, sum);
1714
1715	if (!recover_curseg || recover_newaddr)
1716		update_sit_entry(sbi, new_blkaddr, 1);
1717	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1718		update_sit_entry(sbi, old_blkaddr, -1);
1719
1720	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1721	locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1722
1723	locate_dirty_segment(sbi, old_cursegno);
1724
1725	if (recover_curseg) {
1726		if (old_cursegno != curseg->segno) {
1727			curseg->next_segno = old_cursegno;
1728			change_curseg(sbi, type, true);
1729		}
1730		curseg->next_blkoff = old_blkoff;
1731	}
1732
1733	mutex_unlock(&sit_i->sentry_lock);
1734	mutex_unlock(&curseg->curseg_mutex);
1735}
1736
1737void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1738				block_t old_addr, block_t new_addr,
1739				unsigned char version, bool recover_curseg,
1740				bool recover_newaddr)
1741{
1742	struct f2fs_summary sum;
1743
1744	set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1745
1746	__f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1747					recover_curseg, recover_newaddr);
1748
1749	f2fs_update_data_blkaddr(dn, new_addr);
1750}
1751
1752void f2fs_wait_on_page_writeback(struct page *page,
1753				enum page_type type, bool ordered)
1754{
1755	if (PageWriteback(page)) {
1756		struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1757
1758		f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
1759		if (ordered)
1760			wait_on_page_writeback(page);
1761		else
1762			wait_for_stable_page(page);
1763	}
1764}
1765
1766void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1767							block_t blkaddr)
1768{
1769	struct page *cpage;
1770
1771	if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1772		return;
1773
1774	cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1775	if (cpage) {
1776		f2fs_wait_on_page_writeback(cpage, DATA, true);
1777		f2fs_put_page(cpage, 1);
1778	}
1779}
1780
1781static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1782{
1783	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1784	struct curseg_info *seg_i;
1785	unsigned char *kaddr;
1786	struct page *page;
1787	block_t start;
1788	int i, j, offset;
1789
1790	start = start_sum_block(sbi);
1791
1792	page = get_meta_page(sbi, start++);
1793	kaddr = (unsigned char *)page_address(page);
1794
1795	/* Step 1: restore nat cache */
1796	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1797	memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
1798
1799	/* Step 2: restore sit cache */
1800	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1801	memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
1802	offset = 2 * SUM_JOURNAL_SIZE;
1803
1804	/* Step 3: restore summary entries */
1805	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1806		unsigned short blk_off;
1807		unsigned int segno;
1808
1809		seg_i = CURSEG_I(sbi, i);
1810		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1811		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1812		seg_i->next_segno = segno;
1813		reset_curseg(sbi, i, 0);
1814		seg_i->alloc_type = ckpt->alloc_type[i];
1815		seg_i->next_blkoff = blk_off;
1816
1817		if (seg_i->alloc_type == SSR)
1818			blk_off = sbi->blocks_per_seg;
1819
1820		for (j = 0; j < blk_off; j++) {
1821			struct f2fs_summary *s;
1822			s = (struct f2fs_summary *)(kaddr + offset);
1823			seg_i->sum_blk->entries[j] = *s;
1824			offset += SUMMARY_SIZE;
1825			if (offset + SUMMARY_SIZE <= PAGE_SIZE -
1826						SUM_FOOTER_SIZE)
1827				continue;
1828
1829			f2fs_put_page(page, 1);
1830			page = NULL;
1831
1832			page = get_meta_page(sbi, start++);
1833			kaddr = (unsigned char *)page_address(page);
1834			offset = 0;
1835		}
1836	}
1837	f2fs_put_page(page, 1);
1838	return 0;
1839}
1840
1841static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1842{
1843	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1844	struct f2fs_summary_block *sum;
1845	struct curseg_info *curseg;
1846	struct page *new;
1847	unsigned short blk_off;
1848	unsigned int segno = 0;
1849	block_t blk_addr = 0;
1850
1851	/* get segment number and block addr */
1852	if (IS_DATASEG(type)) {
1853		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1854		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1855							CURSEG_HOT_DATA]);
1856		if (__exist_node_summaries(sbi))
1857			blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1858		else
1859			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1860	} else {
1861		segno = le32_to_cpu(ckpt->cur_node_segno[type -
1862							CURSEG_HOT_NODE]);
1863		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1864							CURSEG_HOT_NODE]);
1865		if (__exist_node_summaries(sbi))
1866			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1867							type - CURSEG_HOT_NODE);
1868		else
1869			blk_addr = GET_SUM_BLOCK(sbi, segno);
1870	}
1871
1872	new = get_meta_page(sbi, blk_addr);
1873	sum = (struct f2fs_summary_block *)page_address(new);
1874
1875	if (IS_NODESEG(type)) {
1876		if (__exist_node_summaries(sbi)) {
1877			struct f2fs_summary *ns = &sum->entries[0];
1878			int i;
1879			for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1880				ns->version = 0;
1881				ns->ofs_in_node = 0;
1882			}
1883		} else {
1884			int err;
1885
1886			err = restore_node_summary(sbi, segno, sum);
1887			if (err) {
1888				f2fs_put_page(new, 1);
1889				return err;
1890			}
1891		}
1892	}
1893
1894	/* set uncompleted segment to curseg */
1895	curseg = CURSEG_I(sbi, type);
1896	mutex_lock(&curseg->curseg_mutex);
1897
1898	/* update journal info */
1899	down_write(&curseg->journal_rwsem);
1900	memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1901	up_write(&curseg->journal_rwsem);
1902
1903	memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1904	memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
1905	curseg->next_segno = segno;
1906	reset_curseg(sbi, type, 0);
1907	curseg->alloc_type = ckpt->alloc_type[type];
1908	curseg->next_blkoff = blk_off;
1909	mutex_unlock(&curseg->curseg_mutex);
1910	f2fs_put_page(new, 1);
1911	return 0;
1912}
1913
1914static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1915{
1916	int type = CURSEG_HOT_DATA;
1917	int err;
1918
1919	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
1920		int npages = npages_for_summary_flush(sbi, true);
1921
1922		if (npages >= 2)
1923			ra_meta_pages(sbi, start_sum_block(sbi), npages,
1924							META_CP, true);
1925
1926		/* restore for compacted data summary */
1927		if (read_compacted_summaries(sbi))
1928			return -EINVAL;
1929		type = CURSEG_HOT_NODE;
1930	}
1931
1932	if (__exist_node_summaries(sbi))
1933		ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1934					NR_CURSEG_TYPE - type, META_CP, true);
1935
1936	for (; type <= CURSEG_COLD_NODE; type++) {
1937		err = read_normal_summaries(sbi, type);
1938		if (err)
1939			return err;
1940	}
1941
1942	return 0;
1943}
1944
1945static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1946{
1947	struct page *page;
1948	unsigned char *kaddr;
1949	struct f2fs_summary *summary;
1950	struct curseg_info *seg_i;
1951	int written_size = 0;
1952	int i, j;
1953
1954	page = grab_meta_page(sbi, blkaddr++);
1955	kaddr = (unsigned char *)page_address(page);
1956
1957	/* Step 1: write nat cache */
1958	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1959	memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
1960	written_size += SUM_JOURNAL_SIZE;
1961
1962	/* Step 2: write sit cache */
1963	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1964	memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
1965	written_size += SUM_JOURNAL_SIZE;
1966
1967	/* Step 3: write summary entries */
1968	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1969		unsigned short blkoff;
1970		seg_i = CURSEG_I(sbi, i);
1971		if (sbi->ckpt->alloc_type[i] == SSR)
1972			blkoff = sbi->blocks_per_seg;
1973		else
1974			blkoff = curseg_blkoff(sbi, i);
1975
1976		for (j = 0; j < blkoff; j++) {
1977			if (!page) {
1978				page = grab_meta_page(sbi, blkaddr++);
1979				kaddr = (unsigned char *)page_address(page);
1980				written_size = 0;
1981			}
1982			summary = (struct f2fs_summary *)(kaddr + written_size);
1983			*summary = seg_i->sum_blk->entries[j];
1984			written_size += SUMMARY_SIZE;
1985
1986			if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
1987							SUM_FOOTER_SIZE)
1988				continue;
1989
1990			set_page_dirty(page);
1991			f2fs_put_page(page, 1);
1992			page = NULL;
1993		}
1994	}
1995	if (page) {
1996		set_page_dirty(page);
1997		f2fs_put_page(page, 1);
1998	}
1999}
2000
2001static void write_normal_summaries(struct f2fs_sb_info *sbi,
2002					block_t blkaddr, int type)
2003{
2004	int i, end;
2005	if (IS_DATASEG(type))
2006		end = type + NR_CURSEG_DATA_TYPE;
2007	else
2008		end = type + NR_CURSEG_NODE_TYPE;
2009
2010	for (i = type; i < end; i++)
2011		write_current_sum_page(sbi, i, blkaddr + (i - type));
2012}
2013
2014void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2015{
2016	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
2017		write_compacted_summaries(sbi, start_blk);
2018	else
2019		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2020}
2021
2022void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2023{
2024	write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
2025}
2026
2027int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
2028					unsigned int val, int alloc)
2029{
2030	int i;
2031
2032	if (type == NAT_JOURNAL) {
2033		for (i = 0; i < nats_in_cursum(journal); i++) {
2034			if (le32_to_cpu(nid_in_journal(journal, i)) == val)
2035				return i;
2036		}
2037		if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2038			return update_nats_in_cursum(journal, 1);
2039	} else if (type == SIT_JOURNAL) {
2040		for (i = 0; i < sits_in_cursum(journal); i++)
2041			if (le32_to_cpu(segno_in_journal(journal, i)) == val)
2042				return i;
2043		if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2044			return update_sits_in_cursum(journal, 1);
2045	}
2046	return -1;
2047}
2048
2049static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2050					unsigned int segno)
2051{
2052	return get_meta_page(sbi, current_sit_addr(sbi, segno));
2053}
2054
2055static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2056					unsigned int start)
2057{
2058	struct sit_info *sit_i = SIT_I(sbi);
2059	struct page *src_page, *dst_page;
2060	pgoff_t src_off, dst_off;
2061	void *src_addr, *dst_addr;
2062
2063	src_off = current_sit_addr(sbi, start);
2064	dst_off = next_sit_addr(sbi, src_off);
2065
2066	/* get current sit block page without lock */
2067	src_page = get_meta_page(sbi, src_off);
2068	dst_page = grab_meta_page(sbi, dst_off);
2069	f2fs_bug_on(sbi, PageDirty(src_page));
2070
2071	src_addr = page_address(src_page);
2072	dst_addr = page_address(dst_page);
2073	memcpy(dst_addr, src_addr, PAGE_SIZE);
2074
2075	set_page_dirty(dst_page);
2076	f2fs_put_page(src_page, 1);
2077
2078	set_to_next_sit(sit_i, start);
2079
2080	return dst_page;
2081}
2082
2083static struct sit_entry_set *grab_sit_entry_set(void)
2084{
2085	struct sit_entry_set *ses =
2086			f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
2087
2088	ses->entry_cnt = 0;
2089	INIT_LIST_HEAD(&ses->set_list);
2090	return ses;
2091}
2092
2093static void release_sit_entry_set(struct sit_entry_set *ses)
2094{
2095	list_del(&ses->set_list);
2096	kmem_cache_free(sit_entry_set_slab, ses);
2097}
2098
2099static void adjust_sit_entry_set(struct sit_entry_set *ses,
2100						struct list_head *head)
2101{
2102	struct sit_entry_set *next = ses;
2103
2104	if (list_is_last(&ses->set_list, head))
2105		return;
2106
2107	list_for_each_entry_continue(next, head, set_list)
2108		if (ses->entry_cnt <= next->entry_cnt)
2109			break;
2110
2111	list_move_tail(&ses->set_list, &next->set_list);
2112}
2113
2114static void add_sit_entry(unsigned int segno, struct list_head *head)
2115{
2116	struct sit_entry_set *ses;
2117	unsigned int start_segno = START_SEGNO(segno);
2118
2119	list_for_each_entry(ses, head, set_list) {
2120		if (ses->start_segno == start_segno) {
2121			ses->entry_cnt++;
2122			adjust_sit_entry_set(ses, head);
2123			return;
2124		}
2125	}
2126
2127	ses = grab_sit_entry_set();
2128
2129	ses->start_segno = start_segno;
2130	ses->entry_cnt++;
2131	list_add(&ses->set_list, head);
2132}
2133
2134static void add_sits_in_set(struct f2fs_sb_info *sbi)
2135{
2136	struct f2fs_sm_info *sm_info = SM_I(sbi);
2137	struct list_head *set_list = &sm_info->sit_entry_set;
2138	unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
2139	unsigned int segno;
2140
2141	for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
2142		add_sit_entry(segno, set_list);
2143}
2144
2145static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
2146{
2147	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2148	struct f2fs_journal *journal = curseg->journal;
2149	int i;
2150
2151	down_write(&curseg->journal_rwsem);
2152	for (i = 0; i < sits_in_cursum(journal); i++) {
2153		unsigned int segno;
2154		bool dirtied;
2155
2156		segno = le32_to_cpu(segno_in_journal(journal, i));
2157		dirtied = __mark_sit_entry_dirty(sbi, segno);
2158
2159		if (!dirtied)
2160			add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
2161	}
2162	update_sits_in_cursum(journal, -i);
2163	up_write(&curseg->journal_rwsem);
2164}
2165
2166/*
2167 * CP calls this function, which flushes SIT entries including sit_journal,
2168 * and moves prefree segs to free segs.
2169 */
2170void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2171{
2172	struct sit_info *sit_i = SIT_I(sbi);
2173	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2174	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2175	struct f2fs_journal *journal = curseg->journal;
2176	struct sit_entry_set *ses, *tmp;
2177	struct list_head *head = &SM_I(sbi)->sit_entry_set;
2178	bool to_journal = true;
2179	struct seg_entry *se;
2180
2181	mutex_lock(&sit_i->sentry_lock);
2182
2183	if (!sit_i->dirty_sentries)
2184		goto out;
2185
2186	/*
2187	 * add and account sit entries of dirty bitmap in sit entry
2188	 * set temporarily
2189	 */
2190	add_sits_in_set(sbi);
2191
2192	/*
2193	 * if there are no enough space in journal to store dirty sit
2194	 * entries, remove all entries from journal and add and account
2195	 * them in sit entry set.
2196	 */
2197	if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
2198		remove_sits_in_journal(sbi);
2199
2200	/*
2201	 * there are two steps to flush sit entries:
2202	 * #1, flush sit entries to journal in current cold data summary block.
2203	 * #2, flush sit entries to sit page.
2204	 */
2205	list_for_each_entry_safe(ses, tmp, head, set_list) {
2206		struct page *page = NULL;
2207		struct f2fs_sit_block *raw_sit = NULL;
2208		unsigned int start_segno = ses->start_segno;
2209		unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
2210						(unsigned long)MAIN_SEGS(sbi));
2211		unsigned int segno = start_segno;
2212
2213		if (to_journal &&
2214			!__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
2215			to_journal = false;
2216
2217		if (to_journal) {
2218			down_write(&curseg->journal_rwsem);
2219		} else {
2220			page = get_next_sit_page(sbi, start_segno);
2221			raw_sit = page_address(page);
2222		}
2223
2224		/* flush dirty sit entries in region of current sit set */
2225		for_each_set_bit_from(segno, bitmap, end) {
2226			int offset, sit_offset;
2227
2228			se = get_seg_entry(sbi, segno);
2229
2230			/* add discard candidates */
2231			if (cpc->reason != CP_DISCARD) {
2232				cpc->trim_start = segno;
2233				add_discard_addrs(sbi, cpc);
2234			}
2235
2236			if (to_journal) {
2237				offset = lookup_journal_in_cursum(journal,
2238							SIT_JOURNAL, segno, 1);
2239				f2fs_bug_on(sbi, offset < 0);
2240				segno_in_journal(journal, offset) =
2241							cpu_to_le32(segno);
2242				seg_info_to_raw_sit(se,
2243					&sit_in_journal(journal, offset));
2244			} else {
2245				sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2246				seg_info_to_raw_sit(se,
2247						&raw_sit->entries[sit_offset]);
2248			}
2249
2250			__clear_bit(segno, bitmap);
2251			sit_i->dirty_sentries--;
2252			ses->entry_cnt--;
2253		}
2254
2255		if (to_journal)
2256			up_write(&curseg->journal_rwsem);
2257		else
2258			f2fs_put_page(page, 1);
2259
2260		f2fs_bug_on(sbi, ses->entry_cnt);
2261		release_sit_entry_set(ses);
2262	}
2263
2264	f2fs_bug_on(sbi, !list_empty(head));
2265	f2fs_bug_on(sbi, sit_i->dirty_sentries);
2266out:
2267	if (cpc->reason == CP_DISCARD) {
2268		for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2269			add_discard_addrs(sbi, cpc);
2270	}
2271	mutex_unlock(&sit_i->sentry_lock);
2272
2273	set_prefree_as_free_segments(sbi);
2274}
2275
2276static int build_sit_info(struct f2fs_sb_info *sbi)
2277{
2278	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2279	struct sit_info *sit_i;
2280	unsigned int sit_segs, start;
2281	char *src_bitmap, *dst_bitmap;
2282	unsigned int bitmap_size;
2283
2284	/* allocate memory for SIT information */
2285	sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2286	if (!sit_i)
2287		return -ENOMEM;
2288
2289	SM_I(sbi)->sit_info = sit_i;
2290
2291	sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2292					sizeof(struct seg_entry), GFP_KERNEL);
2293	if (!sit_i->sentries)
2294		return -ENOMEM;
2295
2296	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2297	sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2298	if (!sit_i->dirty_sentries_bitmap)
2299		return -ENOMEM;
2300
2301	for (start = 0; start < MAIN_SEGS(sbi); start++) {
2302		sit_i->sentries[start].cur_valid_map
2303			= kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2304		sit_i->sentries[start].ckpt_valid_map
2305			= kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2306		if (!sit_i->sentries[start].cur_valid_map ||
2307				!sit_i->sentries[start].ckpt_valid_map)
2308			return -ENOMEM;
2309
2310		if (f2fs_discard_en(sbi)) {
2311			sit_i->sentries[start].discard_map
2312				= kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2313			if (!sit_i->sentries[start].discard_map)
2314				return -ENOMEM;
2315		}
2316	}
2317
2318	sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2319	if (!sit_i->tmp_map)
2320		return -ENOMEM;
2321
2322	if (sbi->segs_per_sec > 1) {
2323		sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2324					sizeof(struct sec_entry), GFP_KERNEL);
2325		if (!sit_i->sec_entries)
2326			return -ENOMEM;
2327	}
2328
2329	/* get information related with SIT */
2330	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2331
2332	/* setup SIT bitmap from ckeckpoint pack */
2333	bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2334	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2335
2336	dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2337	if (!dst_bitmap)
2338		return -ENOMEM;
2339
2340	/* init SIT information */
2341	sit_i->s_ops = &default_salloc_ops;
2342
2343	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2344	sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2345	sit_i->written_valid_blocks = 0;
2346	sit_i->sit_bitmap = dst_bitmap;
2347	sit_i->bitmap_size = bitmap_size;
2348	sit_i->dirty_sentries = 0;
2349	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2350	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2351	sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2352	mutex_init(&sit_i->sentry_lock);
2353	return 0;
2354}
2355
2356static int build_free_segmap(struct f2fs_sb_info *sbi)
2357{
2358	struct free_segmap_info *free_i;
2359	unsigned int bitmap_size, sec_bitmap_size;
2360
2361	/* allocate memory for free segmap information */
2362	free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2363	if (!free_i)
2364		return -ENOMEM;
2365
2366	SM_I(sbi)->free_info = free_i;
2367
2368	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2369	free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2370	if (!free_i->free_segmap)
2371		return -ENOMEM;
2372
2373	sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2374	free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2375	if (!free_i->free_secmap)
2376		return -ENOMEM;
2377
2378	/* set all segments as dirty temporarily */
2379	memset(free_i->free_segmap, 0xff, bitmap_size);
2380	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2381
2382	/* init free segmap information */
2383	free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2384	free_i->free_segments = 0;
2385	free_i->free_sections = 0;
2386	spin_lock_init(&free_i->segmap_lock);
2387	return 0;
2388}
2389
2390static int build_curseg(struct f2fs_sb_info *sbi)
2391{
2392	struct curseg_info *array;
2393	int i;
2394
2395	array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2396	if (!array)
2397		return -ENOMEM;
2398
2399	SM_I(sbi)->curseg_array = array;
2400
2401	for (i = 0; i < NR_CURSEG_TYPE; i++) {
2402		mutex_init(&array[i].curseg_mutex);
2403		array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2404		if (!array[i].sum_blk)
2405			return -ENOMEM;
2406		init_rwsem(&array[i].journal_rwsem);
2407		array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2408							GFP_KERNEL);
2409		if (!array[i].journal)
2410			return -ENOMEM;
2411		array[i].segno = NULL_SEGNO;
2412		array[i].next_blkoff = 0;
2413	}
2414	return restore_curseg_summaries(sbi);
2415}
2416
2417static void build_sit_entries(struct f2fs_sb_info *sbi)
2418{
2419	struct sit_info *sit_i = SIT_I(sbi);
2420	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2421	struct f2fs_journal *journal = curseg->journal;
2422	struct seg_entry *se;
2423	struct f2fs_sit_entry sit;
2424	int sit_blk_cnt = SIT_BLK_CNT(sbi);
2425	unsigned int i, start, end;
2426	unsigned int readed, start_blk = 0;
2427
2428	do {
2429		readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
2430							META_SIT, true);
2431
2432		start = start_blk * sit_i->sents_per_block;
2433		end = (start_blk + readed) * sit_i->sents_per_block;
2434
2435		for (; start < end && start < MAIN_SEGS(sbi); start++) {
2436			struct f2fs_sit_block *sit_blk;
2437			struct page *page;
2438
2439			se = &sit_i->sentries[start];
2440			page = get_current_sit_page(sbi, start);
2441			sit_blk = (struct f2fs_sit_block *)page_address(page);
2442			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2443			f2fs_put_page(page, 1);
2444
2445			check_block_count(sbi, start, &sit);
2446			seg_info_from_raw_sit(se, &sit);
2447
2448			/* build discard map only one time */
2449			if (f2fs_discard_en(sbi)) {
2450				memcpy(se->discard_map, se->cur_valid_map,
2451							SIT_VBLOCK_MAP_SIZE);
2452				sbi->discard_blks += sbi->blocks_per_seg -
2453							se->valid_blocks;
2454			}
2455
2456			if (sbi->segs_per_sec > 1)
2457				get_sec_entry(sbi, start)->valid_blocks +=
2458							se->valid_blocks;
2459		}
2460		start_blk += readed;
2461	} while (start_blk < sit_blk_cnt);
2462
2463	down_read(&curseg->journal_rwsem);
2464	for (i = 0; i < sits_in_cursum(journal); i++) {
2465		unsigned int old_valid_blocks;
2466
2467		start = le32_to_cpu(segno_in_journal(journal, i));
2468		se = &sit_i->sentries[start];
2469		sit = sit_in_journal(journal, i);
2470
2471		old_valid_blocks = se->valid_blocks;
2472
2473		check_block_count(sbi, start, &sit);
2474		seg_info_from_raw_sit(se, &sit);
2475
2476		if (f2fs_discard_en(sbi)) {
2477			memcpy(se->discard_map, se->cur_valid_map,
2478						SIT_VBLOCK_MAP_SIZE);
2479			sbi->discard_blks += old_valid_blocks -
2480						se->valid_blocks;
2481		}
2482
2483		if (sbi->segs_per_sec > 1)
2484			get_sec_entry(sbi, start)->valid_blocks +=
2485				se->valid_blocks - old_valid_blocks;
2486	}
2487	up_read(&curseg->journal_rwsem);
2488}
2489
2490static void init_free_segmap(struct f2fs_sb_info *sbi)
2491{
2492	unsigned int start;
2493	int type;
2494
2495	for (start = 0; start < MAIN_SEGS(sbi); start++) {
2496		struct seg_entry *sentry = get_seg_entry(sbi, start);
2497		if (!sentry->valid_blocks)
2498			__set_free(sbi, start);
2499		else
2500			SIT_I(sbi)->written_valid_blocks +=
2501						sentry->valid_blocks;
2502	}
2503
2504	/* set use the current segments */
2505	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2506		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2507		__set_test_and_inuse(sbi, curseg_t->segno);
2508	}
2509}
2510
2511static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2512{
2513	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2514	struct free_segmap_info *free_i = FREE_I(sbi);
2515	unsigned int segno = 0, offset = 0;
2516	unsigned short valid_blocks;
2517
2518	while (1) {
2519		/* find dirty segment based on free segmap */
2520		segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2521		if (segno >= MAIN_SEGS(sbi))
2522			break;
2523		offset = segno + 1;
2524		valid_blocks = get_valid_blocks(sbi, segno, 0);
2525		if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2526			continue;
2527		if (valid_blocks > sbi->blocks_per_seg) {
2528			f2fs_bug_on(sbi, 1);
2529			continue;
2530		}
2531		mutex_lock(&dirty_i->seglist_lock);
2532		__locate_dirty_segment(sbi, segno, DIRTY);
2533		mutex_unlock(&dirty_i->seglist_lock);
2534	}
2535}
2536
2537static int init_victim_secmap(struct f2fs_sb_info *sbi)
2538{
2539	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2540	unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2541
2542	dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2543	if (!dirty_i->victim_secmap)
2544		return -ENOMEM;
2545	return 0;
2546}
2547
2548static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2549{
2550	struct dirty_seglist_info *dirty_i;
2551	unsigned int bitmap_size, i;
2552
2553	/* allocate memory for dirty segments list information */
2554	dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2555	if (!dirty_i)
2556		return -ENOMEM;
2557
2558	SM_I(sbi)->dirty_info = dirty_i;
2559	mutex_init(&dirty_i->seglist_lock);
2560
2561	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2562
2563	for (i = 0; i < NR_DIRTY_TYPE; i++) {
2564		dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2565		if (!dirty_i->dirty_segmap[i])
2566			return -ENOMEM;
2567	}
2568
2569	init_dirty_segmap(sbi);
2570	return init_victim_secmap(sbi);
2571}
2572
2573/*
2574 * Update min, max modified time for cost-benefit GC algorithm
2575 */
2576static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2577{
2578	struct sit_info *sit_i = SIT_I(sbi);
2579	unsigned int segno;
2580
2581	mutex_lock(&sit_i->sentry_lock);
2582
2583	sit_i->min_mtime = LLONG_MAX;
2584
2585	for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2586		unsigned int i;
2587		unsigned long long mtime = 0;
2588
2589		for (i = 0; i < sbi->segs_per_sec; i++)
2590			mtime += get_seg_entry(sbi, segno + i)->mtime;
2591
2592		mtime = div_u64(mtime, sbi->segs_per_sec);
2593
2594		if (sit_i->min_mtime > mtime)
2595			sit_i->min_mtime = mtime;
2596	}
2597	sit_i->max_mtime = get_mtime(sbi);
2598	mutex_unlock(&sit_i->sentry_lock);
2599}
2600
2601int build_segment_manager(struct f2fs_sb_info *sbi)
2602{
2603	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2604	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2605	struct f2fs_sm_info *sm_info;
2606	int err;
2607
2608	sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2609	if (!sm_info)
2610		return -ENOMEM;
2611
2612	/* init sm info */
2613	sbi->sm_info = sm_info;
2614	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2615	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2616	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2617	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2618	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2619	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2620	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2621	sm_info->rec_prefree_segments = sm_info->main_segments *
2622					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2623	if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2624		sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2625
2626	if (!test_opt(sbi, LFS))
2627		sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2628	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2629	sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2630
2631	INIT_LIST_HEAD(&sm_info->discard_list);
2632	INIT_LIST_HEAD(&sm_info->wait_list);
2633	sm_info->nr_discards = 0;
2634	sm_info->max_discards = 0;
2635
2636	sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2637
2638	INIT_LIST_HEAD(&sm_info->sit_entry_set);
2639
2640	if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2641		err = create_flush_cmd_control(sbi);
2642		if (err)
2643			return err;
2644	}
2645
2646	err = build_sit_info(sbi);
2647	if (err)
2648		return err;
2649	err = build_free_segmap(sbi);
2650	if (err)
2651		return err;
2652	err = build_curseg(sbi);
2653	if (err)
2654		return err;
2655
2656	/* reinit free segmap based on SIT */
2657	build_sit_entries(sbi);
2658
2659	init_free_segmap(sbi);
2660	err = build_dirty_segmap(sbi);
2661	if (err)
2662		return err;
2663
2664	init_min_max_mtime(sbi);
2665	return 0;
2666}
2667
2668static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2669		enum dirty_type dirty_type)
2670{
2671	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2672
2673	mutex_lock(&dirty_i->seglist_lock);
2674	kvfree(dirty_i->dirty_segmap[dirty_type]);
2675	dirty_i->nr_dirty[dirty_type] = 0;
2676	mutex_unlock(&dirty_i->seglist_lock);
2677}
2678
2679static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2680{
2681	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2682	kvfree(dirty_i->victim_secmap);
2683}
2684
2685static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2686{
2687	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2688	int i;
2689
2690	if (!dirty_i)
2691		return;
2692
2693	/* discard pre-free/dirty segments list */
2694	for (i = 0; i < NR_DIRTY_TYPE; i++)
2695		discard_dirty_segmap(sbi, i);
2696
2697	destroy_victim_secmap(sbi);
2698	SM_I(sbi)->dirty_info = NULL;
2699	kfree(dirty_i);
2700}
2701
2702static void destroy_curseg(struct f2fs_sb_info *sbi)
2703{
2704	struct curseg_info *array = SM_I(sbi)->curseg_array;
2705	int i;
2706
2707	if (!array)
2708		return;
2709	SM_I(sbi)->curseg_array = NULL;
2710	for (i = 0; i < NR_CURSEG_TYPE; i++) {
2711		kfree(array[i].sum_blk);
2712		kfree(array[i].journal);
2713	}
2714	kfree(array);
2715}
2716
2717static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2718{
2719	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2720	if (!free_i)
2721		return;
2722	SM_I(sbi)->free_info = NULL;
2723	kvfree(free_i->free_segmap);
2724	kvfree(free_i->free_secmap);
2725	kfree(free_i);
2726}
2727
2728static void destroy_sit_info(struct f2fs_sb_info *sbi)
2729{
2730	struct sit_info *sit_i = SIT_I(sbi);
2731	unsigned int start;
2732
2733	if (!sit_i)
2734		return;
2735
2736	if (sit_i->sentries) {
2737		for (start = 0; start < MAIN_SEGS(sbi); start++) {
2738			kfree(sit_i->sentries[start].cur_valid_map);
2739			kfree(sit_i->sentries[start].ckpt_valid_map);
2740			kfree(sit_i->sentries[start].discard_map);
2741		}
2742	}
2743	kfree(sit_i->tmp_map);
2744
2745	kvfree(sit_i->sentries);
2746	kvfree(sit_i->sec_entries);
2747	kvfree(sit_i->dirty_sentries_bitmap);
2748
2749	SM_I(sbi)->sit_info = NULL;
2750	kfree(sit_i->sit_bitmap);
2751	kfree(sit_i);
2752}
2753
2754void destroy_segment_manager(struct f2fs_sb_info *sbi)
2755{
2756	struct f2fs_sm_info *sm_info = SM_I(sbi);
2757
2758	if (!sm_info)
2759		return;
2760	destroy_flush_cmd_control(sbi, true);
2761	destroy_dirty_segmap(sbi);
2762	destroy_curseg(sbi);
2763	destroy_free_segmap(sbi);
2764	destroy_sit_info(sbi);
2765	sbi->sm_info = NULL;
2766	kfree(sm_info);
2767}
2768
2769int __init create_segment_manager_caches(void)
2770{
2771	discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2772			sizeof(struct discard_entry));
2773	if (!discard_entry_slab)
2774		goto fail;
2775
2776	bio_entry_slab = f2fs_kmem_cache_create("bio_entry",
2777			sizeof(struct bio_entry));
2778	if (!bio_entry_slab)
2779		goto destroy_discard_entry;
2780
2781	sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2782			sizeof(struct sit_entry_set));
2783	if (!sit_entry_set_slab)
2784		goto destroy_bio_entry;
2785
2786	inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2787			sizeof(struct inmem_pages));
2788	if (!inmem_entry_slab)
2789		goto destroy_sit_entry_set;
2790	return 0;
2791
2792destroy_sit_entry_set:
2793	kmem_cache_destroy(sit_entry_set_slab);
2794destroy_bio_entry:
2795	kmem_cache_destroy(bio_entry_slab);
2796destroy_discard_entry:
2797	kmem_cache_destroy(discard_entry_slab);
2798fail:
2799	return -ENOMEM;
2800}
2801
2802void destroy_segment_manager_caches(void)
2803{
2804	kmem_cache_destroy(sit_entry_set_slab);
2805	kmem_cache_destroy(bio_entry_slab);
2806	kmem_cache_destroy(discard_entry_slab);
2807	kmem_cache_destroy(inmem_entry_slab);
2808}