Re: [f2fs-dev] [PATCH 2/2] f2fs: add F2FS_IOC_SET_COMPRESS_OPTION ioctl

[Daeho Jeong <daeho43@gmail.com>]

> f2fs_readonly() is redundant with mnt_want_write_file().
>
> Also, shouldn't this require a writable file descriptor?  As-is, this ioctl can
> be called on a file owned by another user, as long as the caller has read
> access.
>
> Note: if you change this to require a writable file descriptor, then
> f2fs_readonly(), mnt_want_write_file(), and IS_IMMUTABLE() all would no longer
> be needed.

I agree that f2fs_readonly() is redundant.
But, sorry, I don't get the rest. I thought mnt_want_write_file() is a
way to check whether the caller has a proper write permission or not.
I think just using mnt_want_write_file() is enough for this ioctl. Am
I missing something?

> What if f2fs_cops[options.algorithm] == NULL, e.g. COMPRESS_LZ4 without
> CONFIG_F2FS_FS_LZ4?  Shouldn't the caller get an error then?

Good point!

> I don't think the check for i_writecount == 1 accomplishes anything because it
> just means there are no *other* writable file descriptors.  It doesn't mean that
> some other thread isn't concurrently trying to write to this same file
> descriptor.  So the lock needs to be enough.  Is it?

This is to detect any possibility of other threads mmap-ing and
writing the file.
Using only inode lock is not enough to prevent them from making dirty pages.



2020년 10월 13일 (화) 오후 3:11, Eric Biggers <ebiggers@kernel.org>님이 작성:
>
> On Tue, Oct 13, 2020 at 11:24:29AM +0900, Daeho Jeong wrote:
> > +static int f2fs_ioc_set_compress_option(struct file *filp, unsigned long arg)
> > +{
> > +     struct inode *inode = file_inode(filp);
> > +     struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
> > +     struct f2fs_comp_option option;
> > +     int ret;
> > +     int writecount;
> > +
> > +     if (!f2fs_sb_has_compression(sbi))
> > +             return -EOPNOTSUPP;
> > +
> > +     if (!f2fs_compressed_file(inode) || IS_IMMUTABLE(inode))
> > +             return -EINVAL;
> > +
> > +     if (f2fs_readonly(sbi->sb))
> > +             return -EROFS;
>

>
> > +
> > +     if (copy_from_user(&option, (struct f2fs_comp_option __user *)arg,
> > +                             sizeof(option)))
> > +             return -EFAULT;
> > +
> > +     if (option.log_cluster_size < MIN_COMPRESS_LOG_SIZE ||
> > +                     option.log_cluster_size > MAX_COMPRESS_LOG_SIZE ||
> > +                     option.algorithm >= COMPRESS_MAX)
> > +             return -EINVAL;
>
> What if f2fs_cops[options.algorithm] == NULL, e.g. COMPRESS_LZ4 without
> CONFIG_F2FS_FS_LZ4?  Shouldn't the caller get an error then?
>
> > +
> > +     ret = mnt_want_write_file(filp);
> > +     if (ret)
> > +             return ret;
> > +
> > +     inode_lock(inode);
> > +
> > +     writecount = atomic_read(&inode->i_writecount);
> > +     if ((filp->f_mode & FMODE_WRITE && writecount != 1) ||
> > +                     (!(filp->f_mode & FMODE_WRITE) && writecount)) {
> > +             ret = -EBUSY;
> > +             goto out;
> > +     }
>
> I don't think the check for i_writecount == 1 accomplishes anything because it
> just means there are no *other* writable file descriptors.  It doesn't mean that
> some other thread isn't concurrently trying to write to this same file
> descriptor.  So the lock needs to be enough.  Is it?
>
> - Eric