------------------------------------------------------------------------ *From*: Gerd Hoffmann *Subject*: [Qemu-devel] [PATCH 3/4] vmsvga: shadow fifo registers *Date*: Mon, 30 May 2016 09:09:20 +0200 ------------------------------------------------------------------------ The fifo is normal ram. So kvm vcpu threads and qemu iothread can access the fifo in parallel without syncronization. Which in turn implies we can't use the fifo pointers in-place because the guest can try changing them underneath us. So add shadows for them, to make sure the guest can't modify them after we've applied sanity checks. Fixes: CVE-2016-4454 Cc: P J P Signed-off-by: Gerd Hoffmann --- hw/display/vmware_vga.c | 57 ++++++++++++++++++++++++------------------------- 1 file changed, 28 insertions(+), 29 deletions(-) diff --git a/tools/qemu-xen/hw/display/vmware_vga.c b/hw/display/vmware_vga.c index a26e62e..de2567b 100644 --- a/tools/qemu-xen/hw/display/vmware_vga.c +++ b/tools/qemu-xen/hw/display/vmware_vga.c @@ -66,17 +66,11 @@ struct vmsvga_state_s { uint8_t *fifo_ptr; unsigned int fifo_size; - union { - uint32_t *fifo; - struct QEMU_PACKED { - uint32_t min; - uint32_t max; - uint32_t next_cmd; - uint32_t stop; - /* Add registers here when adding capabilities. */ - uint32_t fifo[0]; - } *cmd; - }; + uint32_t *fifo; + uint32_t fifo_min; + uint32_t fifo_max; + uint32_t fifo_next; + uint32_t fifo_stop; #define REDRAW_FIFO_LEN 512 struct vmsvga_rect_s { @@ -198,7 +192,7 @@ enum { */ SVGA_FIFO_MIN = 0, SVGA_FIFO_MAX, /* The distance from MIN to MAX must be at least 10K */ - SVGA_FIFO_NEXT_CMD, + SVGA_FIFO_NEXT, SVGA_FIFO_STOP, /* @@ -546,8 +540,6 @@ static inline void vmsvga_cursor_define(struct vmsvga_state_s *s, } #endif -#define CMD(f) le32_to_cpu(s->cmd->f) - static inline int vmsvga_fifo_length(struct vmsvga_state_s *s) { int num; @@ -556,38 +548,44 @@ static inline int vmsvga_fifo_length(struct vmsvga_state_s *s) return 0; } + s->fifo_min = le32_to_cpu(s->fifo[SVGA_FIFO_MIN]); + s->fifo_max = le32_to_cpu(s->fifo[SVGA_FIFO_MAX]); + s->fifo_next = le32_to_cpu(s->fifo[SVGA_FIFO_NEXT]); + s->fifo_stop = le32_to_cpu(s->fifo[SVGA_FIFO_STOP]); + /* Check range and alignment. */ - if ((CMD(min) | CMD(max) | CMD(next_cmd) | CMD(stop)) & 3) { + if ((s->fifo_min | s->fifo_max | s->fifo_next | s->fifo_stop) & 3) { return 0; } - if (CMD(min) < (uint8_t *) s->cmd->fifo - (uint8_t *) s->fifo) { + if (s->fifo_min < sizeof(uint32_t) * 4) { return 0; } - if (CMD(max) > SVGA_FIFO_SIZE || - CMD(min) >= SVGA_FIFO_SIZE || - CMD(stop) >= SVGA_FIFO_SIZE || - CMD(next_cmd) >= SVGA_FIFO_SIZE) { + if (s->fifo_max > SVGA_FIFO_SIZE || + s->fifo_min >= SVGA_FIFO_SIZE || + s->fifo_stop >= SVGA_FIFO_SIZE || + s->fifo_next >= SVGA_FIFO_SIZE) { return 0; } - if (CMD(max) < CMD(min) + 10 * 1024) { + if (s->fifo_max < s->fifo_min + 10 * 1024) { return 0; } - num = CMD(next_cmd) - CMD(stop); + num = s->fifo_next - s->fifo_stop; if (num < 0) { - num += CMD(max) - CMD(min); + num += s->fifo_max - s->fifo_min; } return num >> 2; } static inline uint32_t vmsvga_fifo_read_raw(struct vmsvga_state_s *s) { - uint32_t cmd = s->fifo[CMD(stop) >> 2]; + uint32_t cmd = s->fifo[s->fifo_stop >> 2]; - s->cmd->stop = cpu_to_le32(CMD(stop) + 4); - if (CMD(stop) >= CMD(max)) { - s->cmd->stop = s->cmd->min; + s->fifo_stop += 4; + if (s->fifo_stop >= s->fifo_max) { + s->fifo_stop = s->fifo_min; } + s->fifo[SVGA_FIFO_STOP] = cpu_to_le32(s->fifo_stop); return cmd; } @@ -607,7 +605,7 @@ static void vmsvga_fifo_run(struct vmsvga_state_s *s) len = vmsvga_fifo_length(s); while (len > 0) { /* May need to go back to the start of the command if incomplete */ - cmd_start = s->cmd->stop; + cmd_start = s->fifo_stop; switch (cmd = vmsvga_fifo_read(s)) { case SVGA_CMD_UPDATE: @@ -766,7 +764,8 @@ static void vmsvga_fifo_run(struct vmsvga_state_s *s) break; rewind: - s->cmd->stop = cmd_start; + s->fifo_stop = cmd_start; + s->fifo[SVGA_FIFO_STOP] = cpu_to_le32(s->fifo_stop); break; } } -- 1.8.3.1