Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Wed, 19 Apr 2023 11:18:25 -0700
From:   Jaegeuk Kim <jaegeuk@kernel.org>
To:     Chao Yu <chao@kernel.org>
Cc:     Wu Bo <bo.wu@vivo.com>, linux-f2fs-devel@lists.sourceforge.net,
        linux-kernel@vger.kernel.org, wubo.oduw@gmail.com
Subject: Re: [PATCH 1/1] f2fs: allocate trace path buffer from names_cache
Message-ID: <ZEAwcaVlYq+PNwEK@google.com>
References: <20230414104308.6591-1-bo.wu@vivo.com>
 <e2ec7e19-634c-a065-fe7a-b052d02d4752@kernel.org>
 <ZD7AJT7ZYPji8TWk@google.com>
 <91a05857-c916-cae5-d61d-bbec6f1ae052@kernel.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <91a05857-c916-cae5-d61d-bbec6f1ae052@kernel.org>
Precedence: bulk

On 04/19, Chao Yu wrote:
> On 2023/4/19 0:07, Jaegeuk Kim wrote:
> > On 04/18, Chao Yu wrote:
> > > On 2023/4/14 18:43, Wu Bo wrote:
> > > > It would be better to use the dedicated slab to store path.
> > > > 
> > > > Signed-off-by: Wu Bo <bo.wu@vivo.com>
> > > > ---
> > > >    fs/f2fs/file.c | 4 ++--
> > > >    1 file changed, 2 insertions(+), 2 deletions(-)
> > > > 
> > > > diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
> > > > index 15dabeac4690..27137873958f 100644
> > > > --- a/fs/f2fs/file.c
> > > > +++ b/fs/f2fs/file.c
> > > > @@ -4361,7 +4361,7 @@ static void f2fs_trace_rw_file_path(struct kiocb *iocb, size_t count, int rw)
> > > >    	struct inode *inode = file_inode(iocb->ki_filp);
> > > >    	char *buf, *path;
> > > > -	buf = f2fs_kmalloc(F2FS_I_SB(inode), PATH_MAX, GFP_KERNEL);
> > > > +	buf = __getname();
> > > 
> > > How about:
> > > 
> > > buf = f2fs_kmem_cache_alloc(names_cachep, GFP_KERNEL, NULL, F2FS_I_SB(inode));
> > 
> > This looks like a hack using names_cachep?
> 
> names_cachep was exported in fs.h.

I think that's for __getname() in general, which doesn't indicate you can hack.
No one is using like that.

$ grep names_cachep fs/* -R
fs/dcache.c:struct kmem_cache *names_cachep __read_mostly;
fs/dcache.c:EXPORT_SYMBOL(names_cachep);
fs/dcache.c:	names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0,

$ grep __getname fs/* -R
fs/ceph/mds_client.c:	path = __getname();
fs/cifs/cifsproto.h:	return __getname();
fs/dcache.c:/* SLAB cache for __getname() consumers */
fs/d_path.c:	char *page = __getname();
fs/exfat/dir.c:	nb->lfn = __getname();
fs/f2fs/file.c:	buf = __getname();
fs/fat/dir.c:		*unicode = __getname();
fs/fat/namei_vfat.c:	uname = __getname();
fs/hostfs/hostfs_kern.c:	char *name = __getname();
fs/namei.c:	result = __getname();
fs/namei.c:	result = __getname();
fs/ntfs3/dir.c:	name = __getname();
fs/ntfs3/xattr.c:	buf = __getname();
fs/ntfs3/inode.c:	new_de = __getname();
fs/ntfs3/inode.c:	de = __getname();
fs/ntfs3/inode.c:	de = __getname();
fs/ntfs3/namei.c:	struct cpu_str *uni = __getname();
fs/ntfs3/namei.c:	de = __getname();
fs/ntfs3/namei.c:	struct cpu_str *uni = __getname();
fs/ntfs3/namei.c:	uni = __getname();
fs/ntfs3/namei.c:	uni1 = __getname();
fs/vboxsf/utils.c: * Returns a shfl_string allocated through __getname (must be freed using
fs/vboxsf/utils.c:	buf = __getname();
fs/vboxsf/utils.c:		shfl_path = __getname();

> 
> > Using f2fs_kmem_cache_alloc is able to inject malloc error.
> > But here is a trace event, is it ok to inject error in a trace path?
> 
> Yes, the fail path handling is very simple, so it's fine to leave it
> as it is.
> 
> Reviewed-by: Chao Yu <chao@kernel.org>

What is this for?

> 
> Thanks,
> 
> > 
> > > 
> > > >    	if (!buf)
> > > >    		return;
> > > >    	path = dentry_path_raw(file_dentry(iocb->ki_filp), buf, PATH_MAX);
> > > > @@ -4374,7 +4374,7 @@ static void f2fs_trace_rw_file_path(struct kiocb *iocb, size_t count, int rw)
> > > >    		trace_f2fs_dataread_start(inode, iocb->ki_pos, count,
> > > >    				current->pid, path, current->comm);
> > > >    free_buf:
> > > > -	kfree(buf);
> > > > +	__putname(buf);
> > > 
> > > kmem_cache_free(names_cachep, buf);
> > > 
> > > Thanks,
> > > 
> > > >    }
> > > >    static ssize_t f2fs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)