[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-block] [PULL 01/12] util/iov: introduce qemu_iovec_init_extend
From: |
Peter Maydell |
Subject: |
Re: [Qemu-block] [PULL 01/12] util/iov: introduce qemu_iovec_init_extended |
Date: |
Mon, 9 Sep 2019 18:39:24 +0100 |
On Tue, 27 Aug 2019 at 21:16, Stefan Hajnoczi <address@hidden> wrote:
>
> From: Vladimir Sementsov-Ogievskiy <address@hidden>
>
> Introduce new initialization API, to create requests with padding. Will
> be used in the following patch. New API uses qemu_iovec_init_buf if
> resulting io vector has only one element, to avoid extra allocations.
> So, we need to update qemu_iovec_destroy to support destroying such
> QIOVs.
>
> Signed-off-by: Vladimir Sementsov-Ogievskiy <address@hidden>
> Acked-by: Stefan Hajnoczi <address@hidden>
> Message-id: address@hidden
> Message-Id: <address@hidden>
> Signed-off-by: Stefan Hajnoczi <address@hidden>
Hi -- Coverity thinks this new function could have an
out-of-bounds read (CID 1405302):
> +/*
> + * Compile new iovec, combining @head_buf buffer, sub-qiov of @mid_qiov,
> + * and @tail_buf buffer into new qiov.
> + */
> +void qemu_iovec_init_extended(
> + QEMUIOVector *qiov,
> + void *head_buf, size_t head_len,
> + QEMUIOVector *mid_qiov, size_t mid_offset, size_t mid_len,
> + void *tail_buf, size_t tail_len)
> +{
> + size_t mid_head, mid_tail;
> + int total_niov, mid_niov = 0;
> + struct iovec *p, *mid_iov;
> +
> + if (mid_len) {
> + mid_iov = qiov_slice(mid_qiov, mid_offset, mid_len,
> + &mid_head, &mid_tail, &mid_niov);
> + }
> +
> + total_niov = !!head_len + mid_niov + !!tail_len;
> + if (total_niov == 1) {
> + qemu_iovec_init_buf(qiov, NULL, 0);
> + p = &qiov->local_iov;
> + } else {
> + qiov->niov = qiov->nalloc = total_niov;
> + qiov->size = head_len + mid_len + tail_len;
> + p = qiov->iov = g_new(struct iovec, qiov->niov);
> + }
> +
> + if (head_len) {
> + p->iov_base = head_buf;
> + p->iov_len = head_len;
> + p++;
> + }
> +
> + if (mid_len) {
> + memcpy(p, mid_iov, mid_niov * sizeof(*p));
> + p[0].iov_base = (uint8_t *)p[0].iov_base + mid_head;
> + p[0].iov_len -= mid_head;
> + p[mid_niov - 1].iov_len -= mid_tail;
> + p += mid_niov;
> + }
> +
> + if (tail_len) {
> + p->iov_base = tail_buf;
> + p->iov_len = tail_len;
> + }
> +}
but I'm not familiar enough with the code to be able to tell
if it's correct or if it's just getting confused. Could
somebody have a look? (It's possible it's getting confused
because the calculation of 'total_niov' uses 'mid_niov',
but the condition guarding the code that fills in that part
of the vector is 'mid_len', so it thinks it can take the
"total_niov == 1" codepath and also the "head_len == true"
and "mid_len != 0" paths; in which case using "if (mid_niov)"
instead might make it happier.)
thanks
-- PMM
- Re: [Qemu-block] [PULL 01/12] util/iov: introduce qemu_iovec_init_extended,
Peter Maydell <=