| Message ID | 20120305030824.8515.53602.stgit@jason-ThinkPad-T400 |
|---|---|
| State | New |
| Headers | show |
On Mon, Mar 05, 2012 at 11:08:24AM +0800, Jason Wang wrote: > The tx buffer would be re-allocated for tx descriptor with big size > and without LS bit set, this would make guest driver could easily let > qemu to allocate unlimited. > > In linux host, a glib failure were easy to be triggered: > > GLib-ERROR **: gmem.c:176: failed to allocate 18446744071562067968 bytes > > This patch fix this by adding a limit. As the spec didn't tell the maximum size > of buffer allowed, stick it to current CP_TX_BUFFER_SIZE (65536). > > Signed-off-by: Jason Wang <jasowang@redhat.com> > --- > hw/rtl8139.c | 9 ++++----- > 1 files changed, 4 insertions(+), 5 deletions(-) > > diff --git a/hw/rtl8139.c b/hw/rtl8139.c > index 05b8e1e..d9e742c 100644 > --- a/hw/rtl8139.c > +++ b/hw/rtl8139.c > @@ -2063,11 +2063,10 @@ static int rtl8139_cplus_transmit_one(RTL8139State *s) > > while (s->cplus_txbuffer && s->cplus_txbuffer_offset + txsize >= s->cplus_txbuffer_len) > { > - s->cplus_txbuffer_len += CP_TX_BUFFER_SIZE; > - s->cplus_txbuffer = g_realloc(s->cplus_txbuffer, s->cplus_txbuffer_len); > - > - DPRINTF("+++ C+ mode transmission buffer space changed to %d\n", > - s->cplus_txbuffer_len); > + /* The spec didn't tell the maximum size, stick to CP_TX_BUFFER_SIZE */ > + txsize = s->cplus_txbuffer_len - s->cplus_txbuffer_offset; > + DPRINTF("+++ C+ mode transmission buffer overrun, truncated descriptor" > + "length to %d\n", txsize); This makes sense to me since a 64 KB MTU is not possible - it's a safe upper limit that will not impact guests. Please change this while statement to an if statement and drop the s->cplus_txbuffer check (which is always true since we allocate immediately prior, if it is not already allocated). Stefan
diff --git a/hw/rtl8139.c b/hw/rtl8139.c index 05b8e1e..d9e742c 100644 --- a/hw/rtl8139.c +++ b/hw/rtl8139.c @@ -2063,11 +2063,10 @@ static int rtl8139_cplus_transmit_one(RTL8139State *s) while (s->cplus_txbuffer && s->cplus_txbuffer_offset + txsize >= s->cplus_txbuffer_len) { - s->cplus_txbuffer_len += CP_TX_BUFFER_SIZE; - s->cplus_txbuffer = g_realloc(s->cplus_txbuffer, s->cplus_txbuffer_len); - - DPRINTF("+++ C+ mode transmission buffer space changed to %d\n", - s->cplus_txbuffer_len); + /* The spec didn't tell the maximum size, stick to CP_TX_BUFFER_SIZE */ + txsize = s->cplus_txbuffer_len - s->cplus_txbuffer_offset; + DPRINTF("+++ C+ mode transmission buffer overrun, truncated descriptor" + "length to %d\n", txsize); } if (!s->cplus_txbuffer)
The tx buffer would be re-allocated for tx descriptor with big size and without LS bit set, this would make guest driver could easily let qemu to allocate unlimited. In linux host, a glib failure were easy to be triggered: GLib-ERROR **: gmem.c:176: failed to allocate 18446744071562067968 bytes This patch fix this by adding a limit. As the spec didn't tell the maximum size of buffer allowed, stick it to current CP_TX_BUFFER_SIZE (65536). Signed-off-by: Jason Wang <jasowang@redhat.com> --- hw/rtl8139.c | 9 ++++----- 1 files changed, 4 insertions(+), 5 deletions(-)