1/*
   2 * fs/f2fs/gc.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/module.h>
  13#include <linux/backing-dev.h>
  14#include <linux/init.h>
  15#include <linux/f2fs_fs.h>
  16#include <linux/kthread.h>
  17#include <linux/delay.h>
  18#include <linux/freezer.h>
 
 
 
  19
  20#include "f2fs.h"
  21#include "node.h"
  22#include "segment.h"
  23#include "gc.h"
 
  24#include <trace/events/f2fs.h>
  25
 
 
 
 
 
  26static int gc_thread_func(void *data)
  27{
  28	struct f2fs_sb_info *sbi = data;
  29	struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
  30	wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
  31	long wait_ms;
 
 
 
 
 
  32
  33	wait_ms = gc_th->min_sleep_time;
  34
 
  35	do {
  36		if (try_to_freeze())
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  37			continue;
  38		else
  39			wait_event_interruptible_timeout(*wq,
  40						kthread_should_stop(),
  41						msecs_to_jiffies(wait_ms));
  42		if (kthread_should_stop())
  43			break;
  44
  45		if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
  46			increase_sleep_time(gc_th, &wait_ms);
 
  47			continue;
  48		}
  49
  50#ifdef CONFIG_F2FS_FAULT_INJECTION
  51		if (time_to_inject(sbi, FAULT_CHECKPOINT))
  52			f2fs_stop_checkpoint(sbi, false);
  53#endif
 
 
 
 
 
 
  54
  55		/*
  56		 * [GC triggering condition]
  57		 * 0. GC is not conducted currently.
  58		 * 1. There are enough dirty segments.
  59		 * 2. IO subsystem is idle by checking the # of writeback pages.
  60		 * 3. IO subsystem is idle by checking the # of requests in
  61		 *    bdev's request list.
  62		 *
  63		 * Note) We have to avoid triggering GCs frequently.
  64		 * Because it is possible that some segments can be
  65		 * invalidated soon after by user update or deletion.
  66		 * So, I'd like to wait some time to collect dirty segments.
  67		 */
  68		if (!mutex_trylock(&sbi->gc_mutex))
  69			continue;
 
 
 
 
  70
  71		if (!is_idle(sbi)) {
 
 
 
 
 
 
 
 
  72			increase_sleep_time(gc_th, &wait_ms);
  73			mutex_unlock(&sbi->gc_mutex);
  74			continue;
 
  75		}
  76
  77		if (has_enough_invalid_blocks(sbi))
  78			decrease_sleep_time(gc_th, &wait_ms);
  79		else
  80			increase_sleep_time(gc_th, &wait_ms);
  81
  82		stat_inc_bggc_count(sbi);
 
 
 
 
 
 
 
 
 
 
 
  83
  84		/* if return value is not zero, no victim was selected */
  85		if (f2fs_gc(sbi, test_opt(sbi, FORCE_FG_GC), true))
  86			wait_ms = gc_th->no_gc_sleep_time;
 
 
 
 
 
 
 
 
 
 
  87
  88		trace_f2fs_background_gc(sbi->sb, wait_ms,
  89				prefree_segments(sbi), free_segments(sbi));
  90
  91		/* balancing f2fs's metadata periodically */
  92		f2fs_balance_fs_bg(sbi);
 
 
 
 
 
 
 
 
 
 
 
  93
  94	} while (!kthread_should_stop());
  95	return 0;
  96}
  97
  98int start_gc_thread(struct f2fs_sb_info *sbi)
  99{
 100	struct f2fs_gc_kthread *gc_th;
 101	dev_t dev = sbi->sb->s_bdev->bd_dev;
 102	int err = 0;
 103
 104	gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
 105	if (!gc_th) {
 106		err = -ENOMEM;
 107		goto out;
 108	}
 109
 
 110	gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
 111	gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
 112	gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
 113
 114	gc_th->gc_idle = 0;
 115
 116	sbi->gc_thread = gc_th;
 117	init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
 
 118	sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
 119			"f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
 120	if (IS_ERR(gc_th->f2fs_gc_task)) {
 121		err = PTR_ERR(gc_th->f2fs_gc_task);
 
 122		kfree(gc_th);
 123		sbi->gc_thread = NULL;
 
 124	}
 125out:
 126	return err;
 127}
 128
 129void stop_gc_thread(struct f2fs_sb_info *sbi)
 130{
 131	struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
 
 132	if (!gc_th)
 133		return;
 134	kthread_stop(gc_th->f2fs_gc_task);
 
 135	kfree(gc_th);
 136	sbi->gc_thread = NULL;
 137}
 138
 139static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type)
 140{
 141	int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
 142
 143	if (gc_th && gc_th->gc_idle) {
 144		if (gc_th->gc_idle == 1)
 
 
 145			gc_mode = GC_CB;
 146		else if (gc_th->gc_idle == 2)
 147			gc_mode = GC_GREEDY;
 
 
 
 
 
 
 
 
 
 
 
 
 
 148	}
 
 149	return gc_mode;
 150}
 151
 152static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
 153			int type, struct victim_sel_policy *p)
 154{
 155	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 156
 157	if (p->alloc_mode == SSR) {
 158		p->gc_mode = GC_GREEDY;
 159		p->dirty_segmap = dirty_i->dirty_segmap[type];
 
 
 
 
 
 160		p->max_search = dirty_i->nr_dirty[type];
 161		p->ofs_unit = 1;
 162	} else {
 163		p->gc_mode = select_gc_type(sbi->gc_thread, gc_type);
 164		p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
 165		p->max_search = dirty_i->nr_dirty[DIRTY];
 166		p->ofs_unit = sbi->segs_per_sec;
 
 
 
 
 
 
 
 
 167	}
 168
 169	/* we need to check every dirty segments in the FG_GC case */
 170	if (gc_type != FG_GC && p->max_search > sbi->max_victim_search)
 
 
 
 
 
 
 171		p->max_search = sbi->max_victim_search;
 172
 173	p->offset = sbi->last_victim[p->gc_mode];
 
 
 
 
 
 
 
 174}
 175
 176static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
 177				struct victim_sel_policy *p)
 178{
 179	/* SSR allocates in a segment unit */
 180	if (p->alloc_mode == SSR)
 181		return sbi->blocks_per_seg;
 
 
 
 
 182	if (p->gc_mode == GC_GREEDY)
 183		return sbi->blocks_per_seg * p->ofs_unit;
 184	else if (p->gc_mode == GC_CB)
 185		return UINT_MAX;
 
 
 186	else /* No other gc_mode */
 187		return 0;
 188}
 189
 190static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
 191{
 192	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 193	unsigned int secno;
 194
 195	/*
 196	 * If the gc_type is FG_GC, we can select victim segments
 197	 * selected by background GC before.
 198	 * Those segments guarantee they have small valid blocks.
 199	 */
 200	for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
 201		if (sec_usage_check(sbi, secno))
 202			continue;
 203
 204		if (no_fggc_candidate(sbi, secno))
 205			continue;
 206
 207		clear_bit(secno, dirty_i->victim_secmap);
 208		return secno * sbi->segs_per_sec;
 209	}
 210	return NULL_SEGNO;
 211}
 212
 213static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
 214{
 215	struct sit_info *sit_i = SIT_I(sbi);
 216	unsigned int secno = GET_SECNO(sbi, segno);
 217	unsigned int start = secno * sbi->segs_per_sec;
 218	unsigned long long mtime = 0;
 219	unsigned int vblocks;
 220	unsigned char age = 0;
 221	unsigned char u;
 222	unsigned int i;
 
 223
 224	for (i = 0; i < sbi->segs_per_sec; i++)
 225		mtime += get_seg_entry(sbi, start + i)->mtime;
 226	vblocks = get_valid_blocks(sbi, segno, sbi->segs_per_sec);
 227
 228	mtime = div_u64(mtime, sbi->segs_per_sec);
 229	vblocks = div_u64(vblocks, sbi->segs_per_sec);
 230
 231	u = (vblocks * 100) >> sbi->log_blocks_per_seg;
 232
 233	/* Handle if the system time has changed by the user */
 234	if (mtime < sit_i->min_mtime)
 235		sit_i->min_mtime = mtime;
 236	if (mtime > sit_i->max_mtime)
 237		sit_i->max_mtime = mtime;
 238	if (sit_i->max_mtime != sit_i->min_mtime)
 239		age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
 240				sit_i->max_mtime - sit_i->min_mtime);
 241
 242	return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
 243}
 244
 245static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
 246			unsigned int segno, struct victim_sel_policy *p)
 247{
 248	if (p->alloc_mode == SSR)
 249		return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
 250
 251	/* alloc_mode == LFS */
 252	if (p->gc_mode == GC_GREEDY)
 253		return get_valid_blocks(sbi, segno, sbi->segs_per_sec);
 254	else
 255		return get_cb_cost(sbi, segno);
 
 
 
 256}
 257
 258static unsigned int count_bits(const unsigned long *addr,
 259				unsigned int offset, unsigned int len)
 260{
 261	unsigned int end = offset + len, sum = 0;
 262
 263	while (offset < end) {
 264		if (test_bit(offset++, addr))
 265			++sum;
 266	}
 267	return sum;
 268}
 269
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 270/*
 271 * This function is called from two paths.
 272 * One is garbage collection and the other is SSR segment selection.
 273 * When it is called during GC, it just gets a victim segment
 274 * and it does not remove it from dirty seglist.
 275 * When it is called from SSR segment selection, it finds a segment
 276 * which has minimum valid blocks and removes it from dirty seglist.
 277 */
 278static int get_victim_by_default(struct f2fs_sb_info *sbi,
 279		unsigned int *result, int gc_type, int type, char alloc_mode)
 
 280{
 281	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 
 282	struct victim_sel_policy p;
 283	unsigned int secno, last_victim;
 284	unsigned int last_segment = MAIN_SEGS(sbi);
 285	unsigned int nsearched = 0;
 
 
 286
 287	mutex_lock(&dirty_i->seglist_lock);
 
 288
 289	p.alloc_mode = alloc_mode;
 290	select_policy(sbi, gc_type, type, &p);
 
 291
 
 
 292	p.min_segno = NULL_SEGNO;
 
 293	p.min_cost = get_max_cost(sbi, &p);
 294
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 295	if (p.max_search == 0)
 296		goto out;
 297
 298	last_victim = sbi->last_victim[p.gc_mode];
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 299	if (p.alloc_mode == LFS && gc_type == FG_GC) {
 300		p.min_segno = check_bg_victims(sbi);
 301		if (p.min_segno != NULL_SEGNO)
 302			goto got_it;
 303	}
 304
 305	while (1) {
 306		unsigned long cost;
 307		unsigned int segno;
 308
 309		segno = find_next_bit(p.dirty_segmap, last_segment, p.offset);
 
 
 
 
 310		if (segno >= last_segment) {
 311			if (sbi->last_victim[p.gc_mode]) {
 312				last_segment = sbi->last_victim[p.gc_mode];
 313				sbi->last_victim[p.gc_mode] = 0;
 
 314				p.offset = 0;
 315				continue;
 316			}
 317			break;
 318		}
 319
 320		p.offset = segno + p.ofs_unit;
 321		if (p.ofs_unit > 1) {
 322			p.offset -= segno % p.ofs_unit;
 323			nsearched += count_bits(p.dirty_segmap,
 324						p.offset - p.ofs_unit,
 325						p.ofs_unit);
 326		} else {
 327			nsearched++;
 328		}
 
 
 
 329
 330		secno = GET_SECNO(sbi, segno);
 331
 332		if (sec_usage_check(sbi, secno))
 333			goto next;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 334		if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
 335			goto next;
 336		if (gc_type == FG_GC && p.alloc_mode == LFS &&
 337					no_fggc_candidate(sbi, secno))
 338			goto next;
 339
 
 
 
 
 
 340		cost = get_gc_cost(sbi, segno, &p);
 341
 342		if (p.min_cost > cost) {
 343			p.min_segno = segno;
 344			p.min_cost = cost;
 345		}
 346next:
 347		if (nsearched >= p.max_search) {
 348			if (!sbi->last_victim[p.gc_mode] && segno <= last_victim)
 349				sbi->last_victim[p.gc_mode] = last_victim + 1;
 
 350			else
 351				sbi->last_victim[p.gc_mode] = segno + 1;
 
 
 352			break;
 353		}
 354	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 355	if (p.min_segno != NULL_SEGNO) {
 356got_it:
 
 
 357		if (p.alloc_mode == LFS) {
 358			secno = GET_SECNO(sbi, p.min_segno);
 359			if (gc_type == FG_GC)
 360				sbi->cur_victim_sec = secno;
 361			else
 362				set_bit(secno, dirty_i->victim_secmap);
 363		}
 364		*result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
 365
 
 
 
 366		trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
 367				sbi->cur_victim_sec,
 368				prefree_segments(sbi), free_segments(sbi));
 369	}
 370out:
 371	mutex_unlock(&dirty_i->seglist_lock);
 372
 373	return (p.min_segno == NULL_SEGNO) ? 0 : 1;
 374}
 375
 376static const struct victim_selection default_v_ops = {
 377	.get_victim = get_victim_by_default,
 378};
 379
 380static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
 381{
 382	struct inode_entry *ie;
 383
 384	ie = radix_tree_lookup(&gc_list->iroot, ino);
 385	if (ie)
 386		return ie->inode;
 387	return NULL;
 388}
 389
 390static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
 391{
 392	struct inode_entry *new_ie;
 393
 394	if (inode == find_gc_inode(gc_list, inode->i_ino)) {
 395		iput(inode);
 396		return;
 397	}
 398	new_ie = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 
 399	new_ie->inode = inode;
 400
 401	f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
 402	list_add_tail(&new_ie->list, &gc_list->ilist);
 403}
 404
 405static void put_gc_inode(struct gc_inode_list *gc_list)
 406{
 407	struct inode_entry *ie, *next_ie;
 
 408	list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
 409		radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
 410		iput(ie->inode);
 411		list_del(&ie->list);
 412		kmem_cache_free(inode_entry_slab, ie);
 413	}
 414}
 415
 416static int check_valid_map(struct f2fs_sb_info *sbi,
 417				unsigned int segno, int offset)
 418{
 419	struct sit_info *sit_i = SIT_I(sbi);
 420	struct seg_entry *sentry;
 421	int ret;
 422
 423	mutex_lock(&sit_i->sentry_lock);
 424	sentry = get_seg_entry(sbi, segno);
 425	ret = f2fs_test_bit(offset, sentry->cur_valid_map);
 426	mutex_unlock(&sit_i->sentry_lock);
 427	return ret;
 428}
 429
 430/*
 431 * This function compares node address got in summary with that in NAT.
 432 * On validity, copy that node with cold status, otherwise (invalid node)
 433 * ignore that.
 434 */
 435static void gc_node_segment(struct f2fs_sb_info *sbi,
 436		struct f2fs_summary *sum, unsigned int segno, int gc_type)
 437{
 438	struct f2fs_summary *entry;
 439	block_t start_addr;
 440	int off;
 441	int phase = 0;
 
 
 
 442
 443	start_addr = START_BLOCK(sbi, segno);
 444
 445next_step:
 446	entry = sum;
 447
 448	for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
 
 
 
 449		nid_t nid = le32_to_cpu(entry->nid);
 450		struct page *node_page;
 451		struct node_info ni;
 
 452
 453		/* stop BG_GC if there is not enough free sections. */
 454		if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
 455			return;
 456
 457		if (check_valid_map(sbi, segno, off) == 0)
 458			continue;
 459
 460		if (phase == 0) {
 461			ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
 462							META_NAT, true);
 463			continue;
 464		}
 465
 466		if (phase == 1) {
 467			ra_node_page(sbi, nid);
 468			continue;
 469		}
 470
 471		/* phase == 2 */
 472		node_page = get_node_page(sbi, nid);
 473		if (IS_ERR(node_page))
 474			continue;
 475
 476		/* block may become invalid during get_node_page */
 477		if (check_valid_map(sbi, segno, off) == 0) {
 478			f2fs_put_page(node_page, 1);
 479			continue;
 480		}
 481
 482		get_node_info(sbi, nid, &ni);
 
 
 
 
 483		if (ni.blk_addr != start_addr + off) {
 484			f2fs_put_page(node_page, 1);
 485			continue;
 486		}
 487
 488		move_node_page(node_page, gc_type);
 
 
 489		stat_inc_node_blk_count(sbi, 1, gc_type);
 490	}
 491
 492	if (++phase < 3)
 493		goto next_step;
 
 
 
 
 494}
 495
 496/*
 497 * Calculate start block index indicating the given node offset.
 498 * Be careful, caller should give this node offset only indicating direct node
 499 * blocks. If any node offsets, which point the other types of node blocks such
 500 * as indirect or double indirect node blocks, are given, it must be a caller's
 501 * bug.
 502 */
 503block_t start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
 504{
 505	unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
 506	unsigned int bidx;
 507
 508	if (node_ofs == 0)
 509		return 0;
 510
 511	if (node_ofs <= 2) {
 512		bidx = node_ofs - 1;
 513	} else if (node_ofs <= indirect_blks) {
 514		int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
 
 515		bidx = node_ofs - 2 - dec;
 516	} else {
 517		int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
 
 518		bidx = node_ofs - 5 - dec;
 519	}
 520	return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(inode);
 521}
 522
 523static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 524		struct node_info *dni, block_t blkaddr, unsigned int *nofs)
 525{
 526	struct page *node_page;
 527	nid_t nid;
 528	unsigned int ofs_in_node;
 529	block_t source_blkaddr;
 530
 531	nid = le32_to_cpu(sum->nid);
 532	ofs_in_node = le16_to_cpu(sum->ofs_in_node);
 533
 534	node_page = get_node_page(sbi, nid);
 535	if (IS_ERR(node_page))
 536		return false;
 537
 538	get_node_info(sbi, nid, dni);
 
 
 
 539
 540	if (sum->version != dni->version) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 541		f2fs_put_page(node_page, 1);
 542		return false;
 543	}
 544
 545	*nofs = ofs_of_node(node_page);
 546	source_blkaddr = datablock_addr(node_page, ofs_in_node);
 547	f2fs_put_page(node_page, 1);
 548
 549	if (source_blkaddr != blkaddr)
 
 
 
 
 
 
 
 
 
 
 
 
 550		return false;
 
 551	return true;
 552}
 553
 554static void move_encrypted_block(struct inode *inode, block_t bidx,
 555							unsigned int segno, int off)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 556{
 557	struct f2fs_io_info fio = {
 558		.sbi = F2FS_I_SB(inode),
 
 559		.type = DATA,
 
 560		.op = REQ_OP_READ,
 561		.op_flags = 0,
 562		.encrypted_page = NULL,
 
 
 563	};
 564	struct dnode_of_data dn;
 565	struct f2fs_summary sum;
 566	struct node_info ni;
 567	struct page *page;
 568	block_t newaddr;
 569	int err;
 
 
 
 
 570
 571	/* do not read out */
 572	page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
 573	if (!page)
 574		return;
 
 
 
 
 
 575
 576	if (!check_valid_map(F2FS_I_SB(inode), segno, off))
 
 577		goto out;
 578
 579	set_new_dnode(&dn, inode, NULL, NULL, 0);
 580	err = get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
 581	if (err)
 582		goto out;
 583
 584	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 585		ClearPageUptodate(page);
 
 586		goto put_out;
 587	}
 588
 589	/*
 590	 * don't cache encrypted data into meta inode until previous dirty
 591	 * data were writebacked to avoid racing between GC and flush.
 592	 */
 593	f2fs_wait_on_page_writeback(page, DATA, true);
 594
 595	get_node_info(fio.sbi, dn.nid, &ni);
 596	set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
 
 
 
 597
 598	/* read page */
 599	fio.page = page;
 600	fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
 601
 602	allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
 603							&sum, CURSEG_COLD_DATA);
 604
 605	fio.encrypted_page = pagecache_get_page(META_MAPPING(fio.sbi), newaddr,
 606					FGP_LOCK | FGP_CREAT, GFP_NOFS);
 607	if (!fio.encrypted_page) {
 608		err = -ENOMEM;
 609		goto recover_block;
 610	}
 611
 612	err = f2fs_submit_page_bio(&fio);
 613	if (err)
 614		goto put_page_out;
 615
 616	/* write page */
 617	lock_page(fio.encrypted_page);
 
 
 
 
 
 618
 619	if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) {
 620		err = -EIO;
 621		goto put_page_out;
 
 
 
 
 
 
 
 
 
 622	}
 623	if (unlikely(!PageUptodate(fio.encrypted_page))) {
 624		err = -EIO;
 625		goto put_page_out;
 
 
 
 
 
 
 
 
 
 
 626	}
 627
 
 
 
 
 
 
 
 
 
 628	set_page_dirty(fio.encrypted_page);
 629	f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true);
 630	if (clear_page_dirty_for_io(fio.encrypted_page))
 631		dec_page_count(fio.sbi, F2FS_DIRTY_META);
 632
 633	set_page_writeback(fio.encrypted_page);
 634
 635	/* allocate block address */
 636	f2fs_wait_on_page_writeback(dn.node_page, NODE, true);
 637
 638	fio.op = REQ_OP_WRITE;
 639	fio.op_flags = REQ_SYNC;
 640	fio.new_blkaddr = newaddr;
 641	f2fs_submit_page_mbio(&fio);
 
 
 
 
 
 
 
 
 642
 643	f2fs_update_data_blkaddr(&dn, newaddr);
 644	set_inode_flag(inode, FI_APPEND_WRITE);
 645	if (page->index == 0)
 646		set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
 647put_page_out:
 648	f2fs_put_page(fio.encrypted_page, 1);
 649recover_block:
 650	if (err)
 651		__f2fs_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
 652								true, true);
 
 
 
 653put_out:
 654	f2fs_put_dnode(&dn);
 655out:
 656	f2fs_put_page(page, 1);
 
 657}
 658
 659static void move_data_page(struct inode *inode, block_t bidx, int gc_type,
 660							unsigned int segno, int off)
 661{
 662	struct page *page;
 
 663
 664	page = get_lock_data_page(inode, bidx, true);
 665	if (IS_ERR(page))
 666		return;
 667
 668	if (!check_valid_map(F2FS_I_SB(inode), segno, off))
 
 
 
 
 
 
 669		goto out;
 670
 671	if (gc_type == BG_GC) {
 672		if (PageWriteback(page))
 
 673			goto out;
 
 674		set_page_dirty(page);
 675		set_cold_data(page);
 676	} else {
 677		struct f2fs_io_info fio = {
 678			.sbi = F2FS_I_SB(inode),
 
 679			.type = DATA,
 
 680			.op = REQ_OP_WRITE,
 681			.op_flags = REQ_SYNC,
 
 682			.page = page,
 683			.encrypted_page = NULL,
 
 
 684		};
 685		bool is_dirty = PageDirty(page);
 686		int err;
 687
 688retry:
 
 
 689		set_page_dirty(page);
 690		f2fs_wait_on_page_writeback(page, DATA, true);
 691		if (clear_page_dirty_for_io(page)) {
 692			inode_dec_dirty_pages(inode);
 693			remove_dirty_inode(inode);
 694		}
 695
 696		set_cold_data(page);
 697
 698		err = do_write_data_page(&fio);
 699		if (err == -ENOMEM && is_dirty) {
 700			congestion_wait(BLK_RW_ASYNC, HZ/50);
 701			goto retry;
 
 
 
 
 
 702		}
 703	}
 704out:
 705	f2fs_put_page(page, 1);
 
 706}
 707
 708/*
 709 * This function tries to get parent node of victim data block, and identifies
 710 * data block validity. If the block is valid, copy that with cold status and
 711 * modify parent node.
 712 * If the parent node is not valid or the data block address is different,
 713 * the victim data block is ignored.
 714 */
 715static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 716		struct gc_inode_list *gc_list, unsigned int segno, int gc_type)
 
 717{
 718	struct super_block *sb = sbi->sb;
 719	struct f2fs_summary *entry;
 720	block_t start_addr;
 721	int off;
 722	int phase = 0;
 
 
 723
 724	start_addr = START_BLOCK(sbi, segno);
 725
 726next_step:
 727	entry = sum;
 728
 729	for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
 730		struct page *data_page;
 731		struct inode *inode;
 732		struct node_info dni; /* dnode info for the data */
 733		unsigned int ofs_in_node, nofs;
 734		block_t start_bidx;
 735		nid_t nid = le32_to_cpu(entry->nid);
 736
 737		/* stop BG_GC if there is not enough free sections. */
 738		if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
 739			return;
 
 
 
 
 
 
 740
 741		if (check_valid_map(sbi, segno, off) == 0)
 742			continue;
 743
 744		if (phase == 0) {
 745			ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
 746							META_NAT, true);
 747			continue;
 748		}
 749
 750		if (phase == 1) {
 751			ra_node_page(sbi, nid);
 752			continue;
 753		}
 754
 755		/* Get an inode by ino with checking validity */
 756		if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
 757			continue;
 758
 759		if (phase == 2) {
 760			ra_node_page(sbi, dni.ino);
 761			continue;
 762		}
 763
 764		ofs_in_node = le16_to_cpu(entry->ofs_in_node);
 765
 766		if (phase == 3) {
 
 
 767			inode = f2fs_iget(sb, dni.ino);
 768			if (IS_ERR(inode) || is_bad_inode(inode))
 
 769				continue;
 770
 771			/* if encrypted inode, let's go phase 3 */
 772			if (f2fs_encrypted_inode(inode) &&
 773						S_ISREG(inode->i_mode)) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 774				add_gc_inode(gc_list, inode);
 775				continue;
 776			}
 777
 778			start_bidx = start_bidx_of_node(nofs, inode);
 779			data_page = get_read_data_page(inode,
 780					start_bidx + ofs_in_node, REQ_RAHEAD,
 781					true);
 782			if (IS_ERR(data_page)) {
 783				iput(inode);
 784				continue;
 785			}
 786
 787			f2fs_put_page(data_page, 0);
 788			add_gc_inode(gc_list, inode);
 789			continue;
 790		}
 791
 792		/* phase 4 */
 793		inode = find_gc_inode(gc_list, dni.ino);
 794		if (inode) {
 795			struct f2fs_inode_info *fi = F2FS_I(inode);
 796			bool locked = false;
 
 797
 798			if (S_ISREG(inode->i_mode)) {
 799				if (!down_write_trylock(&fi->dio_rwsem[READ]))
 
 800					continue;
 801				if (!down_write_trylock(
 802						&fi->dio_rwsem[WRITE])) {
 803					up_write(&fi->dio_rwsem[READ]);
 
 
 804					continue;
 805				}
 806				locked = true;
 
 
 
 807			}
 808
 809			start_bidx = start_bidx_of_node(nofs, inode)
 810								+ ofs_in_node;
 811			if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
 812				move_encrypted_block(inode, start_bidx, segno, off);
 
 813			else
 814				move_data_page(inode, start_bidx, gc_type, segno, off);
 
 
 
 
 
 815
 816			if (locked) {
 817				up_write(&fi->dio_rwsem[WRITE]);
 818				up_write(&fi->dio_rwsem[READ]);
 819			}
 820
 821			stat_inc_data_blk_count(sbi, 1, gc_type);
 822		}
 823	}
 824
 825	if (++phase < 5)
 826		goto next_step;
 
 
 827}
 828
 829static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
 830			int gc_type)
 831{
 832	struct sit_info *sit_i = SIT_I(sbi);
 833	int ret;
 834
 835	mutex_lock(&sit_i->sentry_lock);
 836	ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
 837					      NO_CHECK_TYPE, LFS);
 838	mutex_unlock(&sit_i->sentry_lock);
 839	return ret;
 840}
 841
 842static int do_garbage_collect(struct f2fs_sb_info *sbi,
 843				unsigned int start_segno,
 844				struct gc_inode_list *gc_list, int gc_type)
 
 845{
 846	struct page *sum_page;
 847	struct f2fs_summary_block *sum;
 848	struct blk_plug plug;
 849	unsigned int segno = start_segno;
 850	unsigned int end_segno = start_segno + sbi->segs_per_sec;
 851	int sec_freed = 0;
 852	unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
 853						SUM_TYPE_DATA : SUM_TYPE_NODE;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 854
 855	/* readahead multi ssa blocks those have contiguous address */
 856	if (sbi->segs_per_sec > 1)
 857		ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
 858					sbi->segs_per_sec, META_SSA, true);
 859
 860	/* reference all summary page */
 861	while (segno < end_segno) {
 862		sum_page = get_sum_page(sbi, segno++);
 
 
 
 
 
 
 
 
 
 
 
 
 863		unlock_page(sum_page);
 864	}
 865
 866	blk_start_plug(&plug);
 867
 868	for (segno = start_segno; segno < end_segno; segno++) {
 869
 870		/* find segment summary of victim */
 871		sum_page = find_get_page(META_MAPPING(sbi),
 872					GET_SUM_BLOCK(sbi, segno));
 873		f2fs_put_page(sum_page, 0);
 874
 875		if (get_valid_blocks(sbi, segno, 1) == 0 ||
 876				!PageUptodate(sum_page) ||
 877				unlikely(f2fs_cp_error(sbi)))
 878			goto next;
 
 
 
 879
 880		sum = page_address(sum_page);
 881		f2fs_bug_on(sbi, type != GET_SUM_TYPE((&sum->footer)));
 
 
 
 
 
 
 
 882
 883		/*
 884		 * this is to avoid deadlock:
 885		 * - lock_page(sum_page)         - f2fs_replace_block
 886		 *  - check_valid_map()            - mutex_lock(sentry_lock)
 887		 *   - mutex_lock(sentry_lock)     - change_curseg()
 888		 *                                  - lock_page(sum_page)
 889		 */
 890
 891		if (type == SUM_TYPE_NODE)
 892			gc_node_segment(sbi, sum->entries, segno, gc_type);
 893		else
 894			gc_data_segment(sbi, sum->entries, gc_list, segno,
 895								gc_type);
 
 
 
 
 896
 897		stat_inc_seg_count(sbi, type, gc_type);
 898next:
 
 
 
 
 
 
 
 
 
 
 
 899		f2fs_put_page(sum_page, 0);
 900	}
 901
 902	if (gc_type == FG_GC)
 903		f2fs_submit_merged_bio(sbi,
 904				(type == SUM_TYPE_NODE) ? NODE : DATA, WRITE);
 905
 906	blk_finish_plug(&plug);
 907
 908	if (gc_type == FG_GC &&
 909		get_valid_blocks(sbi, start_segno, sbi->segs_per_sec) == 0)
 910		sec_freed = 1;
 911
 912	stat_inc_call_count(sbi->stat_info);
 913
 914	return sec_freed;
 915}
 916
 917int f2fs_gc(struct f2fs_sb_info *sbi, bool sync, bool background)
 918{
 919	unsigned int segno;
 920	int gc_type = sync ? FG_GC : BG_GC;
 921	int sec_freed = 0;
 922	int ret = -EINVAL;
 923	struct cp_control cpc;
 924	struct gc_inode_list gc_list = {
 925		.ilist = LIST_HEAD_INIT(gc_list.ilist),
 926		.iroot = RADIX_TREE_INIT(GFP_NOFS),
 927	};
 
 
 
 
 
 
 
 
 
 
 
 928
 929	cpc.reason = __get_cp_reason(sbi);
 
 930gc_more:
 931	segno = NULL_SEGNO;
 932
 933	if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE)))
 934		goto stop;
 
 935	if (unlikely(f2fs_cp_error(sbi))) {
 936		ret = -EIO;
 937		goto stop;
 938	}
 939
 940	if (gc_type == BG_GC && has_not_enough_free_secs(sbi, sec_freed, 0)) {
 941		gc_type = FG_GC;
 942		/*
 943		 * If there is no victim and no prefree segment but still not
 944		 * enough free sections, we should flush dent/node blocks and do
 945		 * garbage collections.
 946		 */
 947		if (__get_victim(sbi, &segno, gc_type) ||
 948						prefree_segments(sbi)) {
 949			ret = write_checkpoint(sbi, &cpc);
 950			if (ret)
 951				goto stop;
 952			segno = NULL_SEGNO;
 953		} else if (has_not_enough_free_secs(sbi, 0, 0)) {
 954			ret = write_checkpoint(sbi, &cpc);
 955			if (ret)
 956				goto stop;
 957		}
 958	} else if (gc_type == BG_GC && !background) {
 959		/* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
 960		goto stop;
 961	}
 962
 963	if (segno == NULL_SEGNO && !__get_victim(sbi, &segno, gc_type))
 
 
 
 
 
 
 
 
 
 
 
 
 
 964		goto stop;
 965	ret = 0;
 966
 967	if (do_garbage_collect(sbi, segno, &gc_list, gc_type) &&
 968			gc_type == FG_GC)
 
 
 
 969		sec_freed++;
 970
 971	if (gc_type == FG_GC)
 972		sbi->cur_victim_sec = NULL_SEGNO;
 973
 974	if (!sync) {
 975		if (has_not_enough_free_secs(sbi, sec_freed, 0))
 976			goto gc_more;
 
 
 
 
 977
 978		if (gc_type == FG_GC)
 979			ret = write_checkpoint(sbi, &cpc);
 
 
 
 
 
 
 
 
 980	}
 
 
 
 
 
 
 
 
 
 
 
 
 981stop:
 982	mutex_unlock(&sbi->gc_mutex);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 983
 984	put_gc_inode(&gc_list);
 985
 986	if (sync)
 987		ret = sec_freed ? 0 : -EAGAIN;
 988	return ret;
 989}
 990
 991void build_gc_manager(struct f2fs_sb_info *sbi)
 
 
 
 
 
 
 
 992{
 993	u64 main_count, resv_count, ovp_count, blocks_per_sec;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 994
 
 
 995	DIRTY_I(sbi)->v_ops = &default_v_ops;
 996
 997	/* threshold of # of valid blocks in a section for victims of FG_GC */
 998	main_count = SM_I(sbi)->main_segments << sbi->log_blocks_per_seg;
 999	resv_count = SM_I(sbi)->reserved_segments << sbi->log_blocks_per_seg;
1000	ovp_count = SM_I(sbi)->ovp_segments << sbi->log_blocks_per_seg;
1001	blocks_per_sec = sbi->blocks_per_seg * sbi->segs_per_sec;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1002
1003	sbi->fggc_threshold = div_u64((main_count - ovp_count) * blocks_per_sec,
1004					(main_count - resv_count));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1005}