Re: [PATCH 1/2] splice: don't merge into linked buffers

[Jann Horn]

On Mon, Oct 15, 2018 at 11:00 PM Eric Biggers <ebiggers@xxxxxxxxxx> wrote:
> On Mon, Oct 15, 2018 at 05:04:18PM +0200, Jann Horn wrote:
> > Before this patch, it was possible for two pipes to affect each other after
> > data had been transferred between them with tee():
> >
> > ============
> > $ cat tee_test.c
> >
> > int main(void) {
> >   int pipe_a[2];
> >   if (pipe(pipe_a)) err(1, "pipe");
> >   int pipe_b[2];
> >   if (pipe(pipe_b)) err(1, "pipe");
> >   if (write(pipe_a[1], "abcd", 4) != 4) err(1, "write");
> >   if (tee(pipe_a[0], pipe_b[1], 2, 0) != 2) err(1, "tee");
> >   if (write(pipe_b[1], "xx", 2) != 2) err(1, "write");
> >
> >   char buf[5];
> >   if (read(pipe_a[0], buf, 4) != 4) err(1, "read");
> >   buf[4] = 0;
> >   printf("got back: '%s'\n", buf);
> > }
> > $ gcc -o tee_test tee_test.c
> > $ ./tee_test
> > got back: 'abxx'
> > $
> > ============
> >
> > Fix it by explicitly marking buffers as mergeable and clearing that flag in
> > splice_pipe_to_pipe() and link_pipe().
> >
> > Cc: <stable@xxxxxxxxxxxxxxx>
> > Fixes: 7c77f0b3f920 ("splice: implement pipe to pipe splicing")
> > Fixes: 70524490ee2e ("[PATCH] splice: add support for sys_tee()")
> > Signed-off-by: Jann Horn <jannh@xxxxxxxxxx>
> > ---
> > Cleanup in the next patch, to simplify backporting.
> >
> >  fs/pipe.c                 | 5 +++--
> >  fs/splice.c               | 6 ++++++
> >  include/linux/pipe_fs_i.h | 8 ++++++++
> >  3 files changed, 17 insertions(+), 2 deletions(-)
> >
> > diff --git a/fs/pipe.c b/fs/pipe.c
> > index bdc5d3c0977d..4e2eee77f855 100644
> > --- a/fs/pipe.c
> > +++ b/fs/pipe.c
> > @@ -379,7 +379,8 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from)
> >               struct pipe_buffer *buf = pipe->bufs + lastbuf;
> >               int offset = buf->offset + buf->len;
> >
> > -             if (buf->ops->can_merge && offset + chars <= PAGE_SIZE) {
> > +             if (buf->ops->can_merge && offset + chars <= PAGE_SIZE &&
> > +                 (buf->flags & PIPE_BUF_FLAG_MAYMERGE)) {
> >                       ret = pipe_buf_confirm(pipe, buf);
> >                       if (ret)
> >                               goto out;
> > @@ -439,7 +440,7 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from)
> >                       buf->ops = &anon_pipe_buf_ops;
> >                       buf->offset = 0;
> >                       buf->len = copied;
> > -                     buf->flags = 0;
> > +                     buf->flags = PIPE_BUF_FLAG_MAYMERGE;
> >                       if (is_packetized(filp)) {
> >                               buf->ops = &packet_pipe_buf_ops;
> >                               buf->flags = PIPE_BUF_FLAG_PACKET;
> > diff --git a/fs/splice.c b/fs/splice.c
> > index b3daa971f597..111977c80dfd 100644
> > --- a/fs/splice.c
> > +++ b/fs/splice.c
> > @@ -1593,6 +1593,9 @@ static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
> >                        */
> >                       obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
> >
> > +                     /* We can't merge data into a buffer we don't own. */
> > +                     obuf->flags &= ~PIPE_BUF_FLAG_MAYMERGE;
> > +
> >                       obuf->len = len;
> >                       opipe->nrbufs++;
> >                       ibuf->offset += obuf->len;
> > @@ -1667,6 +1670,9 @@ static int link_pipe(struct pipe_inode_info *ipipe,
> >                */
> >               obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
> >
> > +             /* We can't merge data into a buffer we don't own. */
> > +             obuf->flags &= ~PIPE_BUF_FLAG_MAYMERGE;
> > +
> >               if (obuf->len > len)
> >                       obuf->len = len;
> >
> > diff --git a/include/linux/pipe_fs_i.h b/include/linux/pipe_fs_i.h
> > index 5a3bb3b7c9ad..8893711f9171 100644
> > --- a/include/linux/pipe_fs_i.h
> > +++ b/include/linux/pipe_fs_i.h
> > @@ -8,6 +8,14 @@
> >  #define PIPE_BUF_FLAG_ATOMIC 0x02    /* was atomically mapped */
> >  #define PIPE_BUF_FLAG_GIFT   0x04    /* page is a gift */
> >  #define PIPE_BUF_FLAG_PACKET 0x08    /* read() as a packet */
> > +/*
> > + * Set this flag if the generic pipe read/write may coalesce data into an
> > + * existing buffer. If this is not set, a new pipe page segment is always used
> > + * for new data.
> > + * When pipe data is copied by reference (as in the tee() syscall), this flag
> > + * must be cleared on the copy.
> > + */
> > +#define PIPE_BUF_FLAG_MAYMERGE       0x10
> >
> >  /**
> >   *   struct pipe_buffer - a linux kernel pipe buffer
> > --
> > 2.19.0.605.g01d371f741-goog
> >
>
> Deja vu... https://marc.info/?l=linux-fsdevel&m=149003133809192

Oh, hah, even the reproducer looks almost the same. I hadn't seen that...

> Thanks for fixing this; I think this is the right fix.
> I verified it works for my reproducer too.
>
> Did you check whether fuse_dev_splice_write() needs to clear
> PIPE_BUF_FLAG_MAYMERGE?

No, I hadn't looked at that. But from what I can tell,
fuse_dev_splice_write() shouldn't write into the provided buffers,
only read from them.