Re: [PATCH] f2fs: let __get_victim successfully get a segno in corner case

From: Yunlong Song <yunlong.song@huawei.com>
To: Jaegeuk Kim <jaegeuk@kernel.org>
Cc: <chao@kernel.org>, <yuchao0@huawei.com>, <sylinux@163.com>,
	<miaoxie@huawei.com>, <bintian.wang@huawei.com>,
	<linux-fsdevel@vger.kernel.org>,
	<linux-f2fs-devel@lists.sourceforge.net>,
	<linux-kernel@vger.kernel.org>
Subject: Re: [PATCH] f2fs: let __get_victim successfully get a segno in corner case
Date: Thu, 27 Jul 2017 10:56:33 +0800	[thread overview]
Message-ID: <7a3e7408-b142-45e2-c331-3c71515f7d33@huawei.com> (raw)
In-Reply-To: <20170725185532.GA36902@jaegeuk-macbookpro.roam.corp.google.com>

Hi, Jay,

Sorry for the mistake in last mail, the ovp is 462, and the reserved is 235.
I check the code and have not found problems with p.max_search yet.

Just forget the this patch, since there is still 870 segments below, so 
it should
not be the assumed case of this patch.

By the way, I have sent another patch to provide f2fs_balance_fs to 
__write_node_page,
similar as that in __write_data_page, to avoid missing f2fs_balance_fs 
in node page writeback.
Please check that.

On 2017/7/26 2:55, Jaegeuk Kim wrote:
> On 07/14, Yunlong Song wrote:
>> Suppose that the valid blocks of each section are all over sbi->fggc_threshold,
>> and even has_not_enough_free_secs is true, f2fs_gc cannot do its job since the
>> no_fggc_candidate always returns true. As a result, the reserved segments can be
>> used up, and finally there is no free segment at all, and get_new_segment cannot
>> get a free segment, filesystem will trap into a wrong status.
>>
>> To fix this, we record the segno which has a rough minimum cost and return it to
>> __get_victim to continue f2fs_gc's job.
>>
>> Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
>> ---
>>   fs/f2fs/gc.c      | 19 ++++++++++++++-----
>>   fs/f2fs/segment.h | 17 ++++++++++++++---
>>   2 files changed, 28 insertions(+), 8 deletions(-)
>>
>> diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
>> index fa3d2e2..965e783 100644
>> --- a/fs/f2fs/gc.c
>> +++ b/fs/f2fs/gc.c
>> @@ -178,6 +178,8 @@ static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
>>   		p->offset = 0;
>>   	else
>>   		p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
>> +
>> +	p->min_cost_r = UINT_MAX;
>>   }
>>   
>>   static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
>> @@ -194,7 +196,7 @@ static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
>>   		return 0;
>>   }
>>   
>> -static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
>> +static unsigned int check_bg_victims(struct f2fs_sb_info *sbi, struct victim_sel_policy *p)
>>   {
>>   	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
>>   	unsigned int secno;
>> @@ -208,11 +210,12 @@ static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
>>   		if (sec_usage_check(sbi, secno))
>>   			continue;
>>   
>> -		if (no_fggc_candidate(sbi, secno))
>> +		p->cur_segno_r = GET_SEG_FROM_SEC(sbi, secno);
>> +		if (no_fggc_candidate(sbi, secno, p))
>>   			continue;
>>   
>>   		clear_bit(secno, dirty_i->victim_secmap);
>> -		return GET_SEG_FROM_SEC(sbi, secno);
>> +		return p->cur_segno_r;
>>   	}
>>   	return NULL_SEGNO;
>>   }
>> @@ -332,7 +335,7 @@ static int get_victim_by_default(struct f2fs_sb_info *sbi,
>>   
>>   	last_victim = sm->last_victim[p.gc_mode];
>>   	if (p.alloc_mode == LFS && gc_type == FG_GC) {
>> -		p.min_segno = check_bg_victims(sbi);
>> +		p.min_segno = check_bg_victims(sbi, &p);
>>   		if (p.min_segno != NULL_SEGNO)
>>   			goto got_it;
>>   	}
>> @@ -369,8 +372,9 @@ static int get_victim_by_default(struct f2fs_sb_info *sbi,
>>   			goto next;
>>   		if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
>>   			goto next;
>> +		p.cur_segno_r = segno;
>>   		if (gc_type == FG_GC && p.alloc_mode == LFS &&
>> -					no_fggc_candidate(sbi, secno))
>> +					no_fggc_candidate(sbi, secno, &p))
>>   			goto next;
>>   
>>   		cost = get_gc_cost(sbi, segno, &p);
>> @@ -403,6 +407,11 @@ static int get_victim_by_default(struct f2fs_sb_info *sbi,
>>   		trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
>>   				sbi->cur_victim_sec,
>>   				prefree_segments(sbi), free_segments(sbi));
>> +	} else if (has_not_enough_free_secs(sbi, 0, 0)) {
>> +		p.min_segno = p.min_segno_r;
>> +		if (p.alloc_mode == LFS && gc_type == FG_GC)
>> +			clear_bit(GET_SEC_FROM_SEG(sbi, p.min_segno), dirty_i->victim_secmap);
> It seems you want to give a victim segment which has valid blocks larger than
> fggc_threshold.
>
> fggc: 507
> reserve: 462
> cnt_full: 25182
> over: 25912 (>= fggc)
> less: 870 (< fggc)
>
> What is ovp segments?
> It looks like there are 870 candidates for FG_GC, and GC with 103 segments must
> give 104 free segments in the worst case. Is there a problem on p.max_search?
>
> Thanks,
>
>> +		goto got_it;
>>   	}
>>   out:
>>   	mutex_unlock(&dirty_i->seglist_lock);
>> diff --git a/fs/f2fs/segment.h b/fs/f2fs/segment.h
>> index 6b871b4..7d2d0f3 100644
>> --- a/fs/f2fs/segment.h
>> +++ b/fs/f2fs/segment.h
>> @@ -169,6 +169,9 @@ struct victim_sel_policy {
>>   	unsigned int ofs_unit;		/* bitmap search unit */
>>   	unsigned int min_cost;		/* minimum cost */
>>   	unsigned int min_segno;		/* segment # having min. cost */
>> +	unsigned int min_cost_r;		/* rough minimum cost */
>> +	unsigned int min_segno_r;		/* segment # having rough min. cost */
>> +	unsigned int cur_segno_r;		/* segment # rough process is handling */
>>   };
>>   
>>   struct seg_entry {
>> @@ -743,11 +746,19 @@ static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
>>   }
>>   
>>   static inline bool no_fggc_candidate(struct f2fs_sb_info *sbi,
>> -						unsigned int secno)
>> +						unsigned int secno, struct victim_sel_policy *p)
>>   {
>> -	if (get_valid_blocks(sbi, GET_SEG_FROM_SEC(sbi, secno), true) >=
>> -						sbi->fggc_threshold)
>> +	unsigned int cur_cost;
>> +
>> +	cur_cost = get_valid_blocks(sbi, GET_SEG_FROM_SEC(sbi, secno), true);
>> +	if (cur_cost >= sbi->fggc_threshold) {
>> +		if (p->min_cost_r > cur_cost) {
>> +			p->min_cost_r = cur_cost;
>> +			p->min_segno_r = p->cur_segno_r;
>> +		}
>>   		return true;
>> +	}
>> +
>>   	return false;
>>   }
>>   
>> -- 
>> 1.8.5.2
> .
>

-- 
Thanks,
Yunlong Song