Message-ID: <53b12ae4-092f-eae0-676b-b35eeb5a5a0c@huawei.com>
Date:   Tue, 5 Nov 2019 11:39:52 +0800
From:   Chao Yu <yuchao0@...wei.com>
To:     Jaegeuk Kim <jaegeuk@...nel.org>
CC:     <linux-kernel@...r.kernel.org>,
        <linux-f2fs-devel@...ts.sourceforge.net>
Subject: Re: [f2fs-dev] [PATCH 1/2] f2fs: support aligned pinned file

On 2019/10/31 23:29, Jaegeuk Kim wrote:
> On 10/31, Chao Yu wrote:
>> On 2019/10/31 0:09, Jaegeuk Kim wrote:
>>> On 10/26, Chao Yu wrote:
>>>> On 2019/10/26 2:18, Jaegeuk Kim wrote:
>>>>> On 10/24, Chao Yu wrote:
>>>>>> Hi Jaegeuk,
>>>>>>
>>>>>> On 2019/10/23 1:16, Jaegeuk Kim wrote:
>>>>>>> This patch supports 2MB-aligned pinned file, which can guarantee no GC at all
>>>>>>> by allocating fully valid 2MB segment.
>>>>>>>
>>>>>>> Signed-off-by: Jaegeuk Kim <jaegeuk@...nel.org>
>>>>>>> ---
>>>>>>>  fs/f2fs/f2fs.h     |  4 +++-
>>>>>>>  fs/f2fs/file.c     | 39 ++++++++++++++++++++++++++++++++++-----
>>>>>>>  fs/f2fs/recovery.c |  2 +-
>>>>>>>  fs/f2fs/segment.c  | 21 ++++++++++++++++++++-
>>>>>>>  fs/f2fs/segment.h  |  2 ++
>>>>>>>  fs/f2fs/super.c    |  1 +
>>>>>>>  fs/f2fs/sysfs.c    |  2 ++
>>>>>>>  7 files changed, 63 insertions(+), 8 deletions(-)
>>>>>>>
>>>>>>> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
>>>>>>> index ca342f4c7db1..c681f51e351b 100644
>>>>>>> --- a/fs/f2fs/f2fs.h
>>>>>>> +++ b/fs/f2fs/f2fs.h
>>>>>>> @@ -890,6 +890,7 @@ enum {
>>>>>>>  	CURSEG_WARM_NODE,	/* direct node blocks of normal files */
>>>>>>>  	CURSEG_COLD_NODE,	/* indirect node blocks */
>>>>>>>  	NO_CHECK_TYPE,
>>>>>>> +	CURSEG_COLD_DATA_PINNED,/* cold data for pinned file */
>>>>>>>  };
>>>>>>>  
>>>>>>>  struct flush_cmd {
>>>>>>> @@ -1301,6 +1302,7 @@ struct f2fs_sb_info {
>>>>>>>  
>>>>>>>  	/* threshold for gc trials on pinned files */
>>>>>>>  	u64 gc_pin_file_threshold;
>>>>>>> +	struct rw_semaphore pin_sem;
>>>>>>>  
>>>>>>>  	/* maximum # of trials to find a victim segment for SSR and GC */
>>>>>>>  	unsigned int max_victim_search;
>>>>>>> @@ -3116,7 +3118,7 @@ void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi);
>>>>>>>  int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra);
>>>>>>>  void allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
>>>>>>>  					unsigned int start, unsigned int end);
>>>>>>> -void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi);
>>>>>>> +void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi, int type);
>>>>>>>  int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range);
>>>>>>>  bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
>>>>>>>  					struct cp_control *cpc);
>>>>>>> diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
>>>>>>> index 29bc0a542759..f6c038e8a6a7 100644
>>>>>>> --- a/fs/f2fs/file.c
>>>>>>> +++ b/fs/f2fs/file.c
>>>>>>> @@ -1545,12 +1545,41 @@ static int expand_inode_data(struct inode *inode, loff_t offset,
>>>>>>>  	if (off_end)
>>>>>>>  		map.m_len++;
>>>>>>>  
>>>>>>> -	if (f2fs_is_pinned_file(inode))
>>>>>>> -		map.m_seg_type = CURSEG_COLD_DATA;
>>>>>>> +	if (!map.m_len)
>>>>>>> +		return 0;
>>>>>>> +
>>>>>>> +	if (f2fs_is_pinned_file(inode)) {
>>>>>>> +		block_t len = (map.m_len >> sbi->log_blocks_per_seg) <<
>>>>>>> +					sbi->log_blocks_per_seg;
>>>>>>> +		block_t done = 0;
>>>>>>> +
>>>>>>> +		if (map.m_len % sbi->blocks_per_seg)
>>>>>>> +			len += sbi->blocks_per_seg;
>>>>>>>  
>>>>>>> -	err = f2fs_map_blocks(inode, &map, 1, (f2fs_is_pinned_file(inode) ?
>>>>>>> -						F2FS_GET_BLOCK_PRE_DIO :
>>>>>>> -						F2FS_GET_BLOCK_PRE_AIO));
>>>>>>> +		map.m_len = sbi->blocks_per_seg;
>>>>>>> +next_alloc:
>>>>>>> +		mutex_lock(&sbi->gc_mutex);
>>>>>>> +		err = f2fs_gc(sbi, true, false, NULL_SEGNO);
>>>>>>> +		if (err && err != -ENODATA && err != -EAGAIN)
>>>>>>> +			goto out_err;
>>>>>>
>>>>>> To grab enough free space?
>>>>>>
>>>>>> Shouldn't we call
>>>>>>
>>>>>> 	if (has_not_enough_free_secs(sbi, 0, 0)) {
>>>>>> 		mutex_lock(&sbi->gc_mutex);
>>>>>> 		f2fs_gc(sbi, false, false, NULL_SEGNO);
>>>>>> 	}
>>>>>
>>>>> The above calls gc all the time. Do we need this?
>>>>
>>>> Hmmm... my concern is why we need to run foreground GC even if there is enough
>>>> free space..
>>>
>>> In order to get the free segment easily?
>>
>> However, I doubt arbitrary foreground GC with greedy algorithm will ruin
>> hot/cold data separation, actually, for sufficient free segment case, it's
>> unnecessary to call FGGC.
> 
> Two things here; 1) I do worry much about when hitting boundary on
> has_not_enough_free_secs() which calculates # of free segments based on # of
> dirty pages. In this case, we just jump to allocate another free segment so
> I think it increases the possiblity of no free segment panic. 2) Even if we

Yup, I guess for other places, if there is thousand of threads allocating space
concurrently, we may have small probability to hit run out-of-free-space issue,
probability can increase if partition size decreases.

So I think the right way to fix all out-of-free-space issues is to use
reservation mechanism in our interface,  e.g.

f2fs_create()
{
	int reserved_block = 1 + 1;
	/*
	 * reserved blocks we may dirty/allocate in create flow:
	 * 1 block: new inode block;
	 * 1 block: parent's dent block;
	 */

	f2fs_get_budget(reserved_block); //sbi->reserved_block += 2;

	f2fs_balance_fs(sbi); //has_not_enough_free_secs(): reserved_secs += get_secs(,
sbi->reserved_block);

	f2fs_add_link();
	/* inode meta is dirty; dent block is dirty */

	f2fs_release_budget(reserved_block); //sbi->reserved_block -= 2;
}

expand_inode_data() can switch to use this mechanism to avoid that issue.

> do call FGGC a lot, I don't think it will *ruin* the hot/cold data separation
> a lot. Putting hot/warm blocks together into cold log will make another hot
> segment which was being used as cold log. IOWs, we don't need to keep hot data
> in hot log at all, but should be fine to split hot and cold data in different

With below codes, we are trying to mix cold data into warm data log with block
granularity, rather than segment granularity.

	if (type == CURSEG_COLD_DATA) {
		/* GC during CURSEG_COLD_DATA_PINNED allocation */
		if (down_read_trylock(&sbi->pin_sem)) {
			put_pin_sem = true;
		} else {
			type = CURSEG_WARM_DATA;
			curseg = CURSEG_I(sbi, type);
		}

This could cause more GC cycles on such mixed segments, or am I missing something?

Thanks,

> segments. So, I chose to go safer way since this is eating free segments
> directly.
> 
>>
>> Thanks,
>>
>>>
>>>>
>>>>>
>>>>>>
>>>>>>> +
>>>>>>> +		down_write(&sbi->pin_sem);
>>>>>>> +		map.m_seg_type = CURSEG_COLD_DATA_PINNED;
>>>>>>> +		f2fs_allocate_new_segments(sbi, CURSEG_COLD_DATA);
>>>>>>> +		err = f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
>>>>>>> +		up_write(&sbi->pin_sem);
>>>>>>> +
>>>>>>> +		done += map.m_len;
>>>>>>> +		len -= map.m_len;
>>>>>>> +		map.m_lblk += map.m_len;
>>>>>>> +		if (!err && len)
>>>>>>> +			goto next_alloc;
>>>>>>> +
>>>>>>> +		map.m_len = done;
>>>>>>> +	} else {
>>>>>>> +		err = f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
>>>>>>> +	}
>>>>>>> +out_err:
>>>>>>>  	if (err) {
>>>>>>>  		pgoff_t last_off;
>>>>>>>  
>>>>>>> diff --git a/fs/f2fs/recovery.c b/fs/f2fs/recovery.c
>>>>>>> index 783773e4560d..76477f71d4ee 100644
>>>>>>> --- a/fs/f2fs/recovery.c
>>>>>>> +++ b/fs/f2fs/recovery.c
>>>>>>> @@ -711,7 +711,7 @@ static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,
>>>>>>>  		f2fs_put_page(page, 1);
>>>>>>>  	}
>>>>>>>  	if (!err)
>>>>>>> -		f2fs_allocate_new_segments(sbi);
>>>>>>> +		f2fs_allocate_new_segments(sbi, NO_CHECK_TYPE);
>>>>>>>  	return err;
>>>>>>>  }
>>>>>>>  
>>>>>>> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
>>>>>>> index 25c750cd0272..253d72c2663c 100644
>>>>>>> --- a/fs/f2fs/segment.c
>>>>>>> +++ b/fs/f2fs/segment.c
>>>>>>> @@ -2690,7 +2690,7 @@ void allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
>>>>>>>  	up_read(&SM_I(sbi)->curseg_lock);
>>>>>>>  }
>>>>>>>  
>>>>>>> -void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
>>>>>>> +void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi, int type)
>>>>>>>  {
>>>>>>>  	struct curseg_info *curseg;
>>>>>>>  	unsigned int old_segno;
>>>>>>> @@ -2699,6 +2699,9 @@ void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
>>>>>>>  	down_write(&SIT_I(sbi)->sentry_lock);
>>>>>>>  
>>>>>>>  	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
>>>>>>> +		if (type != NO_CHECK_TYPE && i != type)
>>>>>>> +			continue;
>>>>>>> +
>>>>>>>  		curseg = CURSEG_I(sbi, i);
>>>>>>>  		old_segno = curseg->segno;
>>>>>>>  		SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
>>>>>>> @@ -3068,6 +3071,19 @@ void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
>>>>>>>  {
>>>>>>>  	struct sit_info *sit_i = SIT_I(sbi);
>>>>>>>  	struct curseg_info *curseg = CURSEG_I(sbi, type);
>>>>>>> +	bool put_pin_sem = false;
>>>>>>> +
>>>>>>> +	if (type == CURSEG_COLD_DATA) {
>>>>>>> +		/* GC during CURSEG_COLD_DATA_PINNED allocation */
>>>>>>> +		if (down_read_trylock(&sbi->pin_sem)) {
>>>>>>> +			put_pin_sem = true;
>>>>>>> +		} else {
>>>>>>> +			type = CURSEG_WARM_DATA;
>>>>>>> +			curseg = CURSEG_I(sbi, type);
>>>>>>
>>>>>> It will mix pending cold data into warm area... rather than recovering curseg to
>>>>>> write pointer of last cold segment?
>>>>>>
>>>>>> I know maybe that fallocate aligned address could be corner case, but I guess
>>>>>> there should be some better solutions can handle race case more effectively.
>>>>>>
>>>>>> One solution could be: allocating a virtual log header to select free segment as
>>>>>> 2m-aligned space target.
>>>>>
>>>>> I thought about that, but concluded to avoid too much changes.
>>>>
>>>> We have an unupstreamed feature which is based on virtual log header, I can
>>>> introduce that basic virtual log fwk, which can be used for aligned allocation
>>>> and later new features, would you like to check that?
>>>>
>>>> Thanks,
>>>>
>>>>>
>>>>>>
>>>>>> Thanks,
>>>>>>
>>>>>>> +		}
>>>>>>> +	} else if (type == CURSEG_COLD_DATA_PINNED) {
>>>>>>> +		type = CURSEG_COLD_DATA;
>>>>>>> +	}
>>>>>>>  
>>>>>>>  	down_read(&SM_I(sbi)->curseg_lock);
>>>>>>>  
>>>>>>> @@ -3133,6 +3149,9 @@ void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
>>>>>>>  	mutex_unlock(&curseg->curseg_mutex);
>>>>>>>  
>>>>>>>  	up_read(&SM_I(sbi)->curseg_lock);
>>>>>>> +
>>>>>>> +	if (put_pin_sem)
>>>>>>> +		up_read(&sbi->pin_sem);
>>>>>>>  }
>>>>>>>  
>>>>>>>  static void update_device_state(struct f2fs_io_info *fio)
>>>>>>> diff --git a/fs/f2fs/segment.h b/fs/f2fs/segment.h
>>>>>>> index 325781a1ae4d..a95467b202ea 100644
>>>>>>> --- a/fs/f2fs/segment.h
>>>>>>> +++ b/fs/f2fs/segment.h
>>>>>>> @@ -313,6 +313,8 @@ struct sit_entry_set {
>>>>>>>   */
>>>>>>>  static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
>>>>>>>  {
>>>>>>> +	if (type == CURSEG_COLD_DATA_PINNED)
>>>>>>> +		type = CURSEG_COLD_DATA;
>>>>>>>  	return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
>>>>>>>  }
>>>>>>>  
>>>>>>> diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
>>>>>>> index f320fd11db48..c02a47ce551b 100644
>>>>>>> --- a/fs/f2fs/super.c
>>>>>>> +++ b/fs/f2fs/super.c
>>>>>>> @@ -2853,6 +2853,7 @@ static void init_sb_info(struct f2fs_sb_info *sbi)
>>>>>>>  	spin_lock_init(&sbi->dev_lock);
>>>>>>>  
>>>>>>>  	init_rwsem(&sbi->sb_lock);
>>>>>>> +	init_rwsem(&sbi->pin_sem);
>>>>>>>  }
>>>>>>>  
>>>>>>>  static int init_percpu_info(struct f2fs_sb_info *sbi)
>>>>>>> diff --git a/fs/f2fs/sysfs.c b/fs/f2fs/sysfs.c
>>>>>>> index b558b64a4c9c..f164959e4224 100644
>>>>>>> --- a/fs/f2fs/sysfs.c
>>>>>>> +++ b/fs/f2fs/sysfs.c
>>>>>>> @@ -154,6 +154,8 @@ static ssize_t features_show(struct f2fs_attr *a,
>>>>>>>  	if (f2fs_sb_has_casefold(sbi))
>>>>>>>  		len += snprintf(buf + len, PAGE_SIZE - len, "%s%s",
>>>>>>>  				len ? ", " : "", "casefold");
>>>>>>> +	len += snprintf(buf + len, PAGE_SIZE - len, "%s%s",
>>>>>>> +				len ? ", " : "", "pin_file");
>>>>>>>  	len += snprintf(buf + len, PAGE_SIZE - len, "\n");
>>>>>>>  	return len;
>>>>>>>  }
>>>>>>>
>>>>> .
>>>>>
>>> .
>>>
> .
> 

Powered by blists - more mailing lists