linux-kernel - Re: [External Mail]Re: [PATCH v3] f2fs: avoid splitting bio when reading multiple pages

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <b76e5aaa-edb2-4a4d-a6a8-72f6e975f398@xiaomi.com>
Date: Wed, 25 Jun 2025 09:50:22 +0000
From: Huang Jianan <huangjianan@...omi.com>
To: Chao Yu <chao@...nel.org>, "linux-f2fs-devel@...ts.sourceforge.net"
	<linux-f2fs-devel@...ts.sourceforge.net>, "jaegeuk@...nel.org"
	<jaegeuk@...nel.org>
CC: 王辉 <wanghui33@...omi.com>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	盛勇 <shengyong1@...omi.com>
Subject: Re: [External Mail]Re: [PATCH v3] f2fs: avoid splitting bio when
 reading multiple pages

On 2025/6/25 17:48, Jianan Huang wrote:
> On 2025/6/25 16:45, Chao Yu wrote:
>>
>> On 6/25/25 14:49, Jianan Huang wrote:
>>> When fewer pages are read, nr_pages may be smaller than nr_cpages. Due
>>> to the nr_vecs limit, the compressed pages will be split into multiple
>>> bios and then merged at the block level. In this case, nr_cpages should
>>> be used to pre-allocate bvecs.
>>> To handle this case, align max_nr_pages to cluster_size, which should be
>>> enough for all compressed pages.
>>>
>>> Signed-off-by: Jianan Huang <huangjianan@...omi.com>
>>> Signed-off-by: Sheng Yong <shengyong1@...omi.com>
>>> ---
>>> Changes since v2:
>>> - Initialize index only for compressed files.
>>> Changes since v1:
>>> - Use aligned nr_pages instead of nr_cpages to pre-allocate bvecs.
>>>
>>>   fs/f2fs/data.c | 12 ++++++++++--
>>>   1 file changed, 10 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
>>> index 31e892842625..d071d9f6a811 100644
>>> --- a/fs/f2fs/data.c
>>> +++ b/fs/f2fs/data.c
>>> @@ -2303,7 +2303,7 @@ int f2fs_read_multi_pages(struct compress_ctx 
>>> *cc, struct bio **bio_ret,
>>>                }
>>>
>>>                if (!bio) {
>>> -                     bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
>>> +                     bio = f2fs_grab_read_bio(inode, blkaddr, 
>>> nr_pages - i,
>>
>> Jianan,
>>
>> Another case:
>>
>> read page #0,1,2,3 from block #1000,1001,1002, cluster_size=4.
>>
>> nr_pages=4
>> max_nr_pages=round_up(0+4,4)-round_down(0,4)=4
>>
>> f2fs_mpage_readpages() calls f2fs_read_multi_pages() when nr_pages=1, at
>> that time, max_nr_pages equals to 1 as well.
>>
>> f2fs_grab_read_bio(..., 1 - 0,...) allocate bio w/ 1 vec capacity, 
>> however,
>> we need at least 3 vecs to merge all cpages, right?
>>
> 
> Hi, chao,
> 
> If we don't align nr_pages, then when entering f2fs_read_multi_pages,
> we have nr_pages pages left, which belong to other clusters.
> If this is the last page, we can simply pass nr_pages = 0.
> 
> When allocating bio, we need:
> 1. The cpages remaining in the current cluster, which should be 
> (nr_capges - i).
> 2. The maximum cpages remaining in other clusters, which should be 
> max(nr_pages, cc->nr_cpages).
> 

align(nr_pages, cc->nr_cpages), sorry for this.

> So (nr_capges - i) + max(nr_pages, nr_cpages), should be enough for all 
> vecs?
> 
> Thanks,
> 
> 
>> Thanks,
>>
>>>                                        f2fs_ra_op_flags(rac),
>>>                                        folio->index, for_write);
>>>                        if (IS_ERR(bio)) {
>>> @@ -2376,6 +2376,14 @@ static int f2fs_mpage_readpages(struct inode 
>>> *inode,
>>>        unsigned max_nr_pages = nr_pages;
>>>        int ret = 0;
>>>
>>> +#ifdef CONFIG_F2FS_FS_COMPRESSION
>>> +     if (f2fs_compressed_file(inode)) {
>>> +             index = rac ? readahead_index(rac) : folio->index;
>>> +             max_nr_pages = round_up(index + nr_pages, 
>>> cc.cluster_size) -
>>> +                             round_down(index, cc.cluster_size);
>>> +     }
>>> +#endif
>>> +
>>>        map.m_pblk = 0;
>>>        map.m_lblk = 0;
>>>        map.m_len = 0;
>>> @@ -2385,7 +2393,7 @@ static int f2fs_mpage_readpages(struct inode 
>>> *inode,
>>>        map.m_seg_type = NO_CHECK_TYPE;
>>>        map.m_may_create = false;
>>>
>>> -     for (; nr_pages; nr_pages--) {
>>> +     for (; nr_pages; nr_pages--, max_nr_pages--) {
>>>                if (rac) {
>>>                        folio = readahead_folio(rac);
>>>                        prefetchw(&folio->flags);
>>
>