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
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?
- Eric
