linux-kernel - Re: [f2fs-dev] [PATCH] f2fs: stop iterating f2fs_map

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

Message-ID: <ZSQ-2aH7qN_ILfEg@google.com>
Date:   Mon, 9 Oct 2023 10:56:41 -0700
From:   Jaegeuk Kim <jaegeuk@...nel.org>
To:     Chao Yu <chao@...nel.org>
Cc:     linux-kernel@...r.kernel.org,
        linux-f2fs-devel@...ts.sourceforge.net
Subject: Re: [f2fs-dev] [PATCH] f2fs: stop iterating f2fs_map_block if hole
 exists

On 10/07, Chao Yu wrote:
> On 2023/10/4 6:52, Jaegeuk Kim wrote:
> > Let's avoid unnecessary f2fs_map_block calls to load extents.
> > 
> >   # f2fs_io fadvise willneed 0 4096 /data/local/tmp/test
> > 
> >    f2fs_map_blocks: dev = (254,51), ino = 85845, file offset = 386, start blkaddr = 0x34ac00, len = 0x1400, flags = 2,
> >    f2fs_map_blocks: dev = (254,51), ino = 85845, file offset = 5506, start blkaddr = 0x34c200, len = 0x1000, flags = 2,
> >    f2fs_map_blocks: dev = (254,51), ino = 85845, file offset = 9602, start blkaddr = 0x34d600, len = 0x1200, flags = 2,
> >    f2fs_map_blocks: dev = (254,51), ino = 85845, file offset = 14210, start blkaddr = 0x34ec00, len = 0x400, flags = 2,
> >    f2fs_map_blocks: dev = (254,51), ino = 85845, file offset = 15235, start blkaddr = 0x34f401, len = 0xbff, flags = 2,
> >    f2fs_map_blocks: dev = (254,51), ino = 85845, file offset = 18306, start blkaddr = 0x350200, len = 0x1200, flags = 2
> >    f2fs_map_blocks: dev = (254,51), ino = 85845, file offset = 22915, start blkaddr = 0x351601, len = 0xa7d, flags = 2
> >    f2fs_map_blocks: dev = (254,51), ino = 85845, file offset = 25600, start blkaddr = 0x351601, len = 0x0, flags = 0
> >    f2fs_map_blocks: dev = (254,51), ino = 85845, file offset = 25601, start blkaddr = 0x351601, len = 0x0, flags = 0
> >    f2fs_map_blocks: dev = (254,51), ino = 85845, file offset = 25602, start blkaddr = 0x351601, len = 0x0, flags = 0
> >    ...
> >    f2fs_map_blocks: dev = (254,51), ino = 85845, file offset = 1037188, start blkaddr = 0x351601, len = 0x0, flags = 0
> >    f2fs_map_blocks: dev = (254,51), ino = 85845, file offset = 1038206, start blkaddr = 0x351601, len = 0x0, flags = 0
> >    f2fs_map_blocks: dev = (254,51), ino = 85845, file offset = 1039224, start blkaddr = 0x351601, len = 0x0, flags = 0
> >    f2fs_map_blocks: dev = (254,51), ino = 85845, file offset = 2075548, start blkaddr = 0x351601, len = 0x0, flags = 0
> 
> Jaegeuk,
> 
> Not sure, but it looks it's due to f2fs_precache_extents() will traverse file
> w/ range [0, max_file_blocks), since data which exceeds EOF will always be zero,
> so it's not necessary to precache its mapping info, so we'd better adjust upper
> boundary to i_size rather than max_file_blocks().
> 
> > 
> > Signed-off-by: Jaegeuk Kim <jaegeuk@...nel.org>
> > ---
> >   fs/f2fs/file.c | 2 +-
> >   1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
> > index 161826c6e200..2403fd1de5a0 100644
> > --- a/fs/f2fs/file.c
> > +++ b/fs/f2fs/file.c
> > @@ -3270,7 +3270,7 @@ int f2fs_precache_extents(struct inode *inode)
> >   		f2fs_down_write(&fi->i_gc_rwsem[WRITE]);
> >   		err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_PRECACHE);
> >   		f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
> > -		if (err)
> > +		if (err || !map.m_len)
> Well, if there is a hole in the head of file, it may break here rather
> than precaching following valid map info.
> 
> What about passing parameter offset|len from f2fs_file_fadvise() to
> f2fs_precache_extents(), and then precaching mapping info on demand.

I'd rather stop reading metadata if it meets a hole within i_size, since this
gives benefits when reading the entire non-hole case, FWIW. Rather than that,
I think generic readahead flow can hide the metadata loading.

> 
> Thanks,
> 
> >   			return err;
> >   		map.m_lblk = m_next_extent;