Re: [PATCH 1/2] sysfs: fix race between readdir and lseek

[Li Zefan <lizefan@huawei.com>]

On 2013/3/21 12:48, Ming Lei wrote:
> On Thu, Mar 21, 2013 at 11:28 AM, Li Zefan <lizefan@huawei.com> wrote:
>> On 2013/3/21 11:17, Ming Lei wrote:
>>> On Thu, Mar 21, 2013 at 10:41 AM, Li Zefan <lizefan@huawei.com> wrote:
>>>>
>>>> In fact the same race exists between readdir() and read()/write()...
>>>
>>> Fortunately, no read()/write() are implemented on sysfs directory, :-)
>>>
>>
>> That's irrelevant...
> 
> As far as sysfs is concerned, the filp->f_ops can't be changed in
> read/write path.
> 

Yes, it can...As I said, it's irrelevant, because it's vfs that changes
file->f_pos.

SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
{
        struct fd f = fdget(fd);
        ssize_t ret = -EBADF;

        if (f.file) {
                loff_t pos = file_pos_read(f.file);		<--- read f_pos
                ret = vfs_read(f.file, buf, count, &pos);	<--- return -EISDIR
                file_pos_write(f.file, pos);			<--- write f_pos
                fdput(f);
        }
        return ret;
}

>>
>> See my report:
>>
>> https://patchwork.kernel.org/patch/2160771/
> 
> Yes, I know there might be some mess after the commit ef3d0fd2
> (vfs: do (nearly) lockless generic_file_llseek).
> 
> Also looks it has been stated in Documentation/filesystems/Locking:
> 
> ->llseek() locking has moved from llseek to the individual llseek
> implementations.  If your fs is not using generic_file_llseek, you
> need to acquire and release the appropriate locks in your ->llseek().
> For many filesystems, it is probably safe to acquire the inode
> mutex or just to use i_size_read() instead.
> Note: this does not protect the file->f_pos against concurrent modifications
> since this is something the userspace has to take care about.
>